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/mount.h>
24 #include <linux/time.h>
25 #include <linux/highuid.h>
26 #include <linux/pagemap.h>
27 #include <linux/dax.h>
28 #include <linux/quotaops.h>
29 #include <linux/string.h>
30 #include <linux/buffer_head.h>
31 #include <linux/writeback.h>
32 #include <linux/pagevec.h>
33 #include <linux/mpage.h>
34 #include <linux/namei.h>
35 #include <linux/uio.h>
36 #include <linux/bio.h>
37 #include <linux/workqueue.h>
38 #include <linux/kernel.h>
39 #include <linux/printk.h>
40 #include <linux/slab.h>
41 #include <linux/bitops.h>
42 #include <linux/iomap.h>
43 #include <linux/iversion.h>
44 #include <linux/dax.h>
46 #include "ext4_jbd2.h"
51 #include <trace/events/ext4.h>
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 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_meta_trans_blocks(struct inode
*inode
, int lblocks
,
147 * Test whether an inode is a fast symlink.
148 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
150 int ext4_inode_is_fast_symlink(struct inode
*inode
)
152 if (!(EXT4_I(inode
)->i_flags
& EXT4_EA_INODE_FL
)) {
153 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
154 EXT4_CLUSTER_SIZE(inode
->i_sb
) >> 9 : 0;
156 if (ext4_has_inline_data(inode
))
159 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
161 return S_ISLNK(inode
->i_mode
) && inode
->i_size
&&
162 (inode
->i_size
< EXT4_N_BLOCKS
* 4);
166 * Called at the last iput() if i_nlink is zero.
168 void ext4_evict_inode(struct inode
*inode
)
173 * Credits for final inode cleanup and freeing:
174 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
175 * (xattr block freeing), bitmap, group descriptor (inode freeing)
177 int extra_credits
= 6;
178 struct ext4_xattr_inode_array
*ea_inode_array
= NULL
;
179 bool freeze_protected
= false;
181 trace_ext4_evict_inode(inode
);
183 if (inode
->i_nlink
) {
185 * When journalling data dirty buffers are tracked only in the
186 * journal. So although mm thinks everything is clean and
187 * ready for reaping the inode might still have some pages to
188 * write in the running transaction or waiting to be
189 * checkpointed. Thus calling jbd2_journal_invalidatepage()
190 * (via truncate_inode_pages()) to discard these buffers can
191 * cause data loss. Also even if we did not discard these
192 * buffers, we would have no way to find them after the inode
193 * is reaped and thus user could see stale data if he tries to
194 * read them before the transaction is checkpointed. So be
195 * careful and force everything to disk here... We use
196 * ei->i_datasync_tid to store the newest transaction
197 * containing inode's data.
199 * Note that directories do not have this problem because they
200 * don't use page cache.
202 if (inode
->i_ino
!= EXT4_JOURNAL_INO
&&
203 ext4_should_journal_data(inode
) &&
204 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
)) &&
205 inode
->i_data
.nrpages
) {
206 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
207 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
209 jbd2_complete_transaction(journal
, commit_tid
);
210 filemap_write_and_wait(&inode
->i_data
);
212 truncate_inode_pages_final(&inode
->i_data
);
217 if (is_bad_inode(inode
))
219 dquot_initialize(inode
);
221 if (ext4_should_order_data(inode
))
222 ext4_begin_ordered_truncate(inode
, 0);
223 truncate_inode_pages_final(&inode
->i_data
);
226 * For inodes with journalled data, transaction commit could have
227 * dirtied the inode. Flush worker is ignoring it because of I_FREEING
228 * flag but we still need to remove the inode from the writeback lists.
230 if (!list_empty_careful(&inode
->i_io_list
)) {
231 WARN_ON_ONCE(!ext4_should_journal_data(inode
));
232 inode_io_list_del(inode
);
236 * Protect us against freezing - iput() caller didn't have to have any
237 * protection against it. When we are in a running transaction though,
238 * we are already protected against freezing and we cannot grab further
239 * protection due to lock ordering constraints.
241 if (!ext4_journal_current_handle()) {
242 sb_start_intwrite(inode
->i_sb
);
243 freeze_protected
= true;
246 if (!IS_NOQUOTA(inode
))
247 extra_credits
+= EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
);
250 * Block bitmap, group descriptor, and inode are accounted in both
251 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
253 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
254 ext4_blocks_for_truncate(inode
) + extra_credits
- 3);
255 if (IS_ERR(handle
)) {
256 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
258 * If we're going to skip the normal cleanup, we still need to
259 * make sure that the in-core orphan linked list is properly
262 ext4_orphan_del(NULL
, inode
);
263 if (freeze_protected
)
264 sb_end_intwrite(inode
->i_sb
);
269 ext4_handle_sync(handle
);
272 * Set inode->i_size to 0 before calling ext4_truncate(). We need
273 * special handling of symlinks here because i_size is used to
274 * determine whether ext4_inode_info->i_data contains symlink data or
275 * block mappings. Setting i_size to 0 will remove its fast symlink
276 * status. Erase i_data so that it becomes a valid empty block map.
278 if (ext4_inode_is_fast_symlink(inode
))
279 memset(EXT4_I(inode
)->i_data
, 0, sizeof(EXT4_I(inode
)->i_data
));
281 err
= ext4_mark_inode_dirty(handle
, inode
);
283 ext4_warning(inode
->i_sb
,
284 "couldn't mark inode dirty (err %d)", err
);
287 if (inode
->i_blocks
) {
288 err
= ext4_truncate(inode
);
290 ext4_error_err(inode
->i_sb
, -err
,
291 "couldn't truncate inode %lu (err %d)",
297 /* Remove xattr references. */
298 err
= ext4_xattr_delete_inode(handle
, inode
, &ea_inode_array
,
301 ext4_warning(inode
->i_sb
, "xattr delete (err %d)", err
);
303 ext4_journal_stop(handle
);
304 ext4_orphan_del(NULL
, inode
);
305 if (freeze_protected
)
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 if (freeze_protected
)
336 sb_end_intwrite(inode
->i_sb
);
337 ext4_xattr_inode_array_free(ea_inode_array
);
340 if (!list_empty(&EXT4_I(inode
)->i_fc_list
))
341 ext4_fc_mark_ineligible(inode
->i_sb
, EXT4_FC_REASON_NOMEM
, NULL
);
342 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
346 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
348 return &EXT4_I(inode
)->i_reserved_quota
;
353 * Called with i_data_sem down, which is important since we can call
354 * ext4_discard_preallocations() from here.
356 void ext4_da_update_reserve_space(struct inode
*inode
,
357 int used
, int quota_claim
)
359 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
360 struct ext4_inode_info
*ei
= EXT4_I(inode
);
362 spin_lock(&ei
->i_block_reservation_lock
);
363 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
364 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
365 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
366 "with only %d reserved data blocks",
367 __func__
, inode
->i_ino
, used
,
368 ei
->i_reserved_data_blocks
);
370 used
= ei
->i_reserved_data_blocks
;
373 /* Update per-inode reservations */
374 ei
->i_reserved_data_blocks
-= used
;
375 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
377 spin_unlock(&ei
->i_block_reservation_lock
);
379 /* Update quota subsystem for data blocks */
381 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
384 * We did fallocate with an offset that is already delayed
385 * allocated. So on delayed allocated writeback we should
386 * not re-claim the quota for fallocated blocks.
388 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
392 * If we have done all the pending block allocations and if
393 * there aren't any writers on the inode, we can discard the
394 * inode's preallocations.
396 if ((ei
->i_reserved_data_blocks
== 0) &&
397 !inode_is_open_for_write(inode
))
398 ext4_discard_preallocations(inode
, 0);
401 static int __check_block_validity(struct inode
*inode
, const char *func
,
403 struct ext4_map_blocks
*map
)
405 if (ext4_has_feature_journal(inode
->i_sb
) &&
407 le32_to_cpu(EXT4_SB(inode
->i_sb
)->s_es
->s_journal_inum
)))
409 if (!ext4_inode_block_valid(inode
, map
->m_pblk
, map
->m_len
)) {
410 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
411 "lblock %lu mapped to illegal pblock %llu "
412 "(length %d)", (unsigned long) map
->m_lblk
,
413 map
->m_pblk
, map
->m_len
);
414 return -EFSCORRUPTED
;
419 int ext4_issue_zeroout(struct inode
*inode
, ext4_lblk_t lblk
, ext4_fsblk_t pblk
,
424 if (IS_ENCRYPTED(inode
) && S_ISREG(inode
->i_mode
))
425 return fscrypt_zeroout_range(inode
, lblk
, pblk
, len
);
427 ret
= sb_issue_zeroout(inode
->i_sb
, pblk
, len
, GFP_NOFS
);
434 #define check_block_validity(inode, map) \
435 __check_block_validity((inode), __func__, __LINE__, (map))
437 #ifdef ES_AGGRESSIVE_TEST
438 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
440 struct ext4_map_blocks
*es_map
,
441 struct ext4_map_blocks
*map
,
448 * There is a race window that the result is not the same.
449 * e.g. xfstests #223 when dioread_nolock enables. The reason
450 * is that we lookup a block mapping in extent status tree with
451 * out taking i_data_sem. So at the time the unwritten extent
452 * could be converted.
454 down_read(&EXT4_I(inode
)->i_data_sem
);
455 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
456 retval
= ext4_ext_map_blocks(handle
, inode
, map
, 0);
458 retval
= ext4_ind_map_blocks(handle
, inode
, map
, 0);
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(inode
, "flag 0x%x, max_blocks %u, logical block %lu\n",
516 flags
, map
->m_len
, (unsigned long) map
->m_lblk
);
519 * ext4_map_blocks returns an int, and m_len is an unsigned int
521 if (unlikely(map
->m_len
> INT_MAX
))
522 map
->m_len
= INT_MAX
;
524 /* We can handle the block number less than EXT_MAX_BLOCKS */
525 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
526 return -EFSCORRUPTED
;
528 /* Lookup extent status tree firstly */
529 if (!(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_FC_REPLAY
) &&
530 ext4_es_lookup_extent(inode
, map
->m_lblk
, NULL
, &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
, 0);
565 retval
= ext4_ind_map_blocks(handle
, inode
, map
, 0);
570 if (unlikely(retval
!= map
->m_len
)) {
571 ext4_warning(inode
->i_sb
,
572 "ES len assertion failed for inode "
573 "%lu: retval %d != map->m_len %d",
574 inode
->i_ino
, retval
, map
->m_len
);
578 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
579 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
580 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
581 !(status
& EXTENT_STATUS_WRITTEN
) &&
582 ext4_es_scan_range(inode
, &ext4_es_is_delayed
, map
->m_lblk
,
583 map
->m_lblk
+ map
->m_len
- 1))
584 status
|= EXTENT_STATUS_DELAYED
;
585 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
586 map
->m_len
, map
->m_pblk
, status
);
590 up_read((&EXT4_I(inode
)->i_data_sem
));
593 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
594 ret
= check_block_validity(inode
, map
);
599 /* If it is only a block(s) look up */
600 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
604 * Returns if the blocks have already allocated
606 * Note that if blocks have been preallocated
607 * ext4_ext_get_block() returns the create = 0
608 * with buffer head unmapped.
610 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
612 * If we need to convert extent to unwritten
613 * we continue and do the actual work in
614 * ext4_ext_map_blocks()
616 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
620 * Here we clear m_flags because after allocating an new extent,
621 * it will be set again.
623 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
626 * New blocks allocate and/or writing to unwritten extent
627 * will possibly result in updating i_data, so we take
628 * the write lock of i_data_sem, and call get_block()
629 * with create == 1 flag.
631 down_write(&EXT4_I(inode
)->i_data_sem
);
634 * We need to check for EXT4 here because migrate
635 * could have changed the inode type in between
637 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
638 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
640 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
642 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
644 * We allocated new blocks which will result in
645 * i_data's format changing. Force the migrate
646 * to fail by clearing migrate flags
648 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
652 * Update reserved blocks/metadata blocks after successful
653 * block allocation which had been deferred till now. We don't
654 * support fallocate for non extent files. So we can update
655 * reserve space here.
658 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
659 ext4_da_update_reserve_space(inode
, retval
, 1);
665 if (unlikely(retval
!= map
->m_len
)) {
666 ext4_warning(inode
->i_sb
,
667 "ES len assertion failed for inode "
668 "%lu: retval %d != map->m_len %d",
669 inode
->i_ino
, retval
, map
->m_len
);
674 * We have to zeroout blocks before inserting them into extent
675 * status tree. Otherwise someone could look them up there and
676 * use them before they are really zeroed. We also have to
677 * unmap metadata before zeroing as otherwise writeback can
678 * overwrite zeros with stale data from block device.
680 if (flags
& EXT4_GET_BLOCKS_ZERO
&&
681 map
->m_flags
& EXT4_MAP_MAPPED
&&
682 map
->m_flags
& EXT4_MAP_NEW
) {
683 ret
= ext4_issue_zeroout(inode
, map
->m_lblk
,
684 map
->m_pblk
, map
->m_len
);
692 * If the extent has been zeroed out, we don't need to update
693 * extent status tree.
695 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
696 ext4_es_lookup_extent(inode
, map
->m_lblk
, NULL
, &es
)) {
697 if (ext4_es_is_written(&es
))
700 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
701 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
702 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
703 !(status
& EXTENT_STATUS_WRITTEN
) &&
704 ext4_es_scan_range(inode
, &ext4_es_is_delayed
, map
->m_lblk
,
705 map
->m_lblk
+ map
->m_len
- 1))
706 status
|= EXTENT_STATUS_DELAYED
;
707 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
708 map
->m_pblk
, status
);
716 up_write((&EXT4_I(inode
)->i_data_sem
));
717 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
718 ret
= check_block_validity(inode
, map
);
723 * Inodes with freshly allocated blocks where contents will be
724 * visible after transaction commit must be on transaction's
727 if (map
->m_flags
& EXT4_MAP_NEW
&&
728 !(map
->m_flags
& EXT4_MAP_UNWRITTEN
) &&
729 !(flags
& EXT4_GET_BLOCKS_ZERO
) &&
730 !ext4_is_quota_file(inode
) &&
731 ext4_should_order_data(inode
)) {
733 (loff_t
)map
->m_lblk
<< inode
->i_blkbits
;
734 loff_t length
= (loff_t
)map
->m_len
<< inode
->i_blkbits
;
736 if (flags
& EXT4_GET_BLOCKS_IO_SUBMIT
)
737 ret
= ext4_jbd2_inode_add_wait(handle
, inode
,
740 ret
= ext4_jbd2_inode_add_write(handle
, inode
,
746 if (retval
> 0 && (map
->m_flags
& EXT4_MAP_UNWRITTEN
||
747 map
->m_flags
& EXT4_MAP_MAPPED
))
748 ext4_fc_track_range(handle
, inode
, map
->m_lblk
,
749 map
->m_lblk
+ map
->m_len
- 1);
751 ext_debug(inode
, "failed with err %d\n", retval
);
756 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
757 * we have to be careful as someone else may be manipulating b_state as well.
759 static void ext4_update_bh_state(struct buffer_head
*bh
, unsigned long flags
)
761 unsigned long old_state
;
762 unsigned long new_state
;
764 flags
&= EXT4_MAP_FLAGS
;
766 /* Dummy buffer_head? Set non-atomically. */
768 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | flags
;
772 * Someone else may be modifying b_state. Be careful! This is ugly but
773 * once we get rid of using bh as a container for mapping information
774 * to pass to / from get_block functions, this can go away.
777 old_state
= READ_ONCE(bh
->b_state
);
778 new_state
= (old_state
& ~EXT4_MAP_FLAGS
) | flags
;
780 cmpxchg(&bh
->b_state
, old_state
, new_state
) != old_state
));
783 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
784 struct buffer_head
*bh
, int flags
)
786 struct ext4_map_blocks map
;
789 if (ext4_has_inline_data(inode
))
793 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
795 ret
= ext4_map_blocks(ext4_journal_current_handle(), inode
, &map
,
798 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
799 ext4_update_bh_state(bh
, map
.m_flags
);
800 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
802 } else if (ret
== 0) {
803 /* hole case, need to fill in bh->b_size */
804 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
809 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
810 struct buffer_head
*bh
, int create
)
812 return _ext4_get_block(inode
, iblock
, bh
,
813 create
? EXT4_GET_BLOCKS_CREATE
: 0);
817 * Get block function used when preparing for buffered write if we require
818 * creating an unwritten extent if blocks haven't been allocated. The extent
819 * will be converted to written after the IO is complete.
821 int ext4_get_block_unwritten(struct inode
*inode
, sector_t iblock
,
822 struct buffer_head
*bh_result
, int create
)
824 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
825 inode
->i_ino
, create
);
826 return _ext4_get_block(inode
, iblock
, bh_result
,
827 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
830 /* Maximum number of blocks we map for direct IO at once. */
831 #define DIO_MAX_BLOCKS 4096
834 * `handle' can be NULL if create is zero
836 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
837 ext4_lblk_t block
, int map_flags
)
839 struct ext4_map_blocks map
;
840 struct buffer_head
*bh
;
841 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
844 ASSERT((EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_FC_REPLAY
)
845 || handle
!= NULL
|| create
== 0);
849 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
852 return create
? ERR_PTR(-ENOSPC
) : NULL
;
856 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
858 return ERR_PTR(-ENOMEM
);
859 if (map
.m_flags
& EXT4_MAP_NEW
) {
861 ASSERT((EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_FC_REPLAY
)
862 || (handle
!= NULL
));
865 * Now that we do not always journal data, we should
866 * keep in mind whether this should always journal the
867 * new buffer as metadata. For now, regular file
868 * writes use ext4_get_block instead, so it's not a
872 BUFFER_TRACE(bh
, "call get_create_access");
873 err
= ext4_journal_get_create_access(handle
, inode
->i_sb
, bh
,
879 if (!buffer_uptodate(bh
)) {
880 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
881 set_buffer_uptodate(bh
);
884 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
885 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
889 BUFFER_TRACE(bh
, "not a new buffer");
896 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
897 ext4_lblk_t block
, int map_flags
)
899 struct buffer_head
*bh
;
902 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
905 if (!bh
|| ext4_buffer_uptodate(bh
))
908 ret
= ext4_read_bh_lock(bh
, REQ_META
| REQ_PRIO
, true);
916 /* Read a contiguous batch of blocks. */
917 int ext4_bread_batch(struct inode
*inode
, ext4_lblk_t block
, int bh_count
,
918 bool wait
, struct buffer_head
**bhs
)
922 for (i
= 0; i
< bh_count
; i
++) {
923 bhs
[i
] = ext4_getblk(NULL
, inode
, block
+ i
, 0 /* map_flags */);
924 if (IS_ERR(bhs
[i
])) {
925 err
= PTR_ERR(bhs
[i
]);
931 for (i
= 0; i
< bh_count
; i
++)
932 /* Note that NULL bhs[i] is valid because of holes. */
933 if (bhs
[i
] && !ext4_buffer_uptodate(bhs
[i
]))
934 ext4_read_bh_lock(bhs
[i
], REQ_META
| REQ_PRIO
, false);
939 for (i
= 0; i
< bh_count
; i
++)
941 wait_on_buffer(bhs
[i
]);
943 for (i
= 0; i
< bh_count
; i
++) {
944 if (bhs
[i
] && !buffer_uptodate(bhs
[i
])) {
952 for (i
= 0; i
< bh_count
; i
++) {
959 int ext4_walk_page_buffers(handle_t
*handle
, struct inode
*inode
,
960 struct buffer_head
*head
,
964 int (*fn
)(handle_t
*handle
, struct inode
*inode
,
965 struct buffer_head
*bh
))
967 struct buffer_head
*bh
;
968 unsigned block_start
, block_end
;
969 unsigned blocksize
= head
->b_size
;
971 struct buffer_head
*next
;
973 for (bh
= head
, block_start
= 0;
974 ret
== 0 && (bh
!= head
|| !block_start
);
975 block_start
= block_end
, bh
= next
) {
976 next
= bh
->b_this_page
;
977 block_end
= block_start
+ blocksize
;
978 if (block_end
<= from
|| block_start
>= to
) {
979 if (partial
&& !buffer_uptodate(bh
))
983 err
= (*fn
)(handle
, inode
, bh
);
991 * To preserve ordering, it is essential that the hole instantiation and
992 * the data write be encapsulated in a single transaction. We cannot
993 * close off a transaction and start a new one between the ext4_get_block()
994 * and the commit_write(). So doing the jbd2_journal_start at the start of
995 * prepare_write() is the right place.
997 * Also, this function can nest inside ext4_writepage(). In that case, we
998 * *know* that ext4_writepage() has generated enough buffer credits to do the
999 * whole page. So we won't block on the journal in that case, which is good,
1000 * because the caller may be PF_MEMALLOC.
1002 * By accident, ext4 can be reentered when a transaction is open via
1003 * quota file writes. If we were to commit the transaction while thus
1004 * reentered, there can be a deadlock - we would be holding a quota
1005 * lock, and the commit would never complete if another thread had a
1006 * transaction open and was blocking on the quota lock - a ranking
1009 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1010 * will _not_ run commit under these circumstances because handle->h_ref
1011 * is elevated. We'll still have enough credits for the tiny quotafile
1014 int do_journal_get_write_access(handle_t
*handle
, struct inode
*inode
,
1015 struct buffer_head
*bh
)
1017 int dirty
= buffer_dirty(bh
);
1020 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1023 * __block_write_begin() could have dirtied some buffers. Clean
1024 * the dirty bit as jbd2_journal_get_write_access() could complain
1025 * otherwise about fs integrity issues. Setting of the dirty bit
1026 * by __block_write_begin() isn't a real problem here as we clear
1027 * the bit before releasing a page lock and thus writeback cannot
1028 * ever write the buffer.
1031 clear_buffer_dirty(bh
);
1032 BUFFER_TRACE(bh
, "get write access");
1033 ret
= ext4_journal_get_write_access(handle
, inode
->i_sb
, bh
,
1036 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1040 #ifdef CONFIG_FS_ENCRYPTION
1041 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1042 get_block_t
*get_block
)
1044 unsigned from
= pos
& (PAGE_SIZE
- 1);
1045 unsigned to
= from
+ len
;
1046 struct inode
*inode
= page
->mapping
->host
;
1047 unsigned block_start
, block_end
;
1050 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1052 struct buffer_head
*bh
, *head
, *wait
[2];
1056 BUG_ON(!PageLocked(page
));
1057 BUG_ON(from
> PAGE_SIZE
);
1058 BUG_ON(to
> PAGE_SIZE
);
1061 if (!page_has_buffers(page
))
1062 create_empty_buffers(page
, blocksize
, 0);
1063 head
= page_buffers(page
);
1064 bbits
= ilog2(blocksize
);
1065 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1067 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1068 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
1069 block_end
= block_start
+ blocksize
;
1070 if (block_end
<= from
|| block_start
>= to
) {
1071 if (PageUptodate(page
)) {
1072 set_buffer_uptodate(bh
);
1077 clear_buffer_new(bh
);
1078 if (!buffer_mapped(bh
)) {
1079 WARN_ON(bh
->b_size
!= blocksize
);
1080 err
= get_block(inode
, block
, bh
, 1);
1083 if (buffer_new(bh
)) {
1084 if (PageUptodate(page
)) {
1085 clear_buffer_new(bh
);
1086 set_buffer_uptodate(bh
);
1087 mark_buffer_dirty(bh
);
1090 if (block_end
> to
|| block_start
< from
)
1091 zero_user_segments(page
, to
, block_end
,
1096 if (PageUptodate(page
)) {
1097 set_buffer_uptodate(bh
);
1100 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1101 !buffer_unwritten(bh
) &&
1102 (block_start
< from
|| block_end
> to
)) {
1103 ext4_read_bh_lock(bh
, 0, false);
1104 wait
[nr_wait
++] = bh
;
1108 * If we issued read requests, let them complete.
1110 for (i
= 0; i
< nr_wait
; i
++) {
1111 wait_on_buffer(wait
[i
]);
1112 if (!buffer_uptodate(wait
[i
]))
1115 if (unlikely(err
)) {
1116 page_zero_new_buffers(page
, from
, to
);
1117 } else if (fscrypt_inode_uses_fs_layer_crypto(inode
)) {
1118 for (i
= 0; i
< nr_wait
; i
++) {
1121 err2
= fscrypt_decrypt_pagecache_blocks(page
, blocksize
,
1122 bh_offset(wait
[i
]));
1124 clear_buffer_uptodate(wait
[i
]);
1134 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1135 loff_t pos
, unsigned len
, unsigned flags
,
1136 struct page
**pagep
, void **fsdata
)
1138 struct inode
*inode
= mapping
->host
;
1139 int ret
, needed_blocks
;
1146 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
1149 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1151 * Reserve one block more for addition to orphan list in case
1152 * we allocate blocks but write fails for some reason
1154 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1155 index
= pos
>> PAGE_SHIFT
;
1156 from
= pos
& (PAGE_SIZE
- 1);
1159 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1160 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1169 * grab_cache_page_write_begin() can take a long time if the
1170 * system is thrashing due to memory pressure, or if the page
1171 * is being written back. So grab it first before we start
1172 * the transaction handle. This also allows us to allocate
1173 * the page (if needed) without using GFP_NOFS.
1176 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1182 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1183 if (IS_ERR(handle
)) {
1185 return PTR_ERR(handle
);
1189 if (page
->mapping
!= mapping
) {
1190 /* The page got truncated from under us */
1193 ext4_journal_stop(handle
);
1196 /* In case writeback began while the page was unlocked */
1197 wait_for_stable_page(page
);
1199 #ifdef CONFIG_FS_ENCRYPTION
1200 if (ext4_should_dioread_nolock(inode
))
1201 ret
= ext4_block_write_begin(page
, pos
, len
,
1202 ext4_get_block_unwritten
);
1204 ret
= ext4_block_write_begin(page
, pos
, len
,
1207 if (ext4_should_dioread_nolock(inode
))
1208 ret
= __block_write_begin(page
, pos
, len
,
1209 ext4_get_block_unwritten
);
1211 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1213 if (!ret
&& ext4_should_journal_data(inode
)) {
1214 ret
= ext4_walk_page_buffers(handle
, inode
,
1215 page_buffers(page
), from
, to
, NULL
,
1216 do_journal_get_write_access
);
1220 bool extended
= (pos
+ len
> inode
->i_size
) &&
1221 !ext4_verity_in_progress(inode
);
1225 * __block_write_begin may have instantiated a few blocks
1226 * outside i_size. Trim these off again. Don't need
1227 * i_size_read because we hold i_rwsem.
1229 * Add inode to orphan list in case we crash before
1232 if (extended
&& ext4_can_truncate(inode
))
1233 ext4_orphan_add(handle
, inode
);
1235 ext4_journal_stop(handle
);
1237 ext4_truncate_failed_write(inode
);
1239 * If truncate failed early the inode might
1240 * still be on the orphan list; we need to
1241 * make sure the inode is removed from the
1242 * orphan list in that case.
1245 ext4_orphan_del(NULL
, inode
);
1248 if (ret
== -ENOSPC
&&
1249 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1258 /* For write_end() in data=journal mode */
1259 static int write_end_fn(handle_t
*handle
, struct inode
*inode
,
1260 struct buffer_head
*bh
)
1263 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1265 set_buffer_uptodate(bh
);
1266 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1267 clear_buffer_meta(bh
);
1268 clear_buffer_prio(bh
);
1273 * We need to pick up the new inode size which generic_commit_write gave us
1274 * `file' can be NULL - eg, when called from page_symlink().
1276 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1277 * buffers are managed internally.
1279 static int ext4_write_end(struct file
*file
,
1280 struct address_space
*mapping
,
1281 loff_t pos
, unsigned len
, unsigned copied
,
1282 struct page
*page
, void *fsdata
)
1284 handle_t
*handle
= ext4_journal_current_handle();
1285 struct inode
*inode
= mapping
->host
;
1286 loff_t old_size
= inode
->i_size
;
1288 int i_size_changed
= 0;
1289 bool verity
= ext4_verity_in_progress(inode
);
1291 trace_ext4_write_end(inode
, pos
, len
, copied
);
1293 if (ext4_has_inline_data(inode
))
1294 return ext4_write_inline_data_end(inode
, pos
, len
, copied
, page
);
1296 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
1298 * it's important to update i_size while still holding page lock:
1299 * page writeout could otherwise come in and zero beyond i_size.
1301 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1302 * blocks are being written past EOF, so skip the i_size update.
1305 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1309 if (old_size
< pos
&& !verity
)
1310 pagecache_isize_extended(inode
, old_size
, pos
);
1312 * Don't mark the inode dirty under page lock. First, it unnecessarily
1313 * makes the holding time of page lock longer. Second, it forces lock
1314 * ordering of page lock and transaction start for journaling
1318 ret
= ext4_mark_inode_dirty(handle
, inode
);
1320 if (pos
+ len
> inode
->i_size
&& !verity
&& ext4_can_truncate(inode
))
1321 /* if we have allocated more blocks and copied
1322 * less. We will have blocks allocated outside
1323 * inode->i_size. So truncate them
1325 ext4_orphan_add(handle
, inode
);
1327 ret2
= ext4_journal_stop(handle
);
1331 if (pos
+ len
> inode
->i_size
&& !verity
) {
1332 ext4_truncate_failed_write(inode
);
1334 * If truncate failed early the inode might still be
1335 * on the orphan list; we need to make sure the inode
1336 * is removed from the orphan list in that case.
1339 ext4_orphan_del(NULL
, inode
);
1342 return ret
? ret
: copied
;
1346 * This is a private version of page_zero_new_buffers() which doesn't
1347 * set the buffer to be dirty, since in data=journalled mode we need
1348 * to call ext4_handle_dirty_metadata() instead.
1350 static void ext4_journalled_zero_new_buffers(handle_t
*handle
,
1351 struct inode
*inode
,
1353 unsigned from
, unsigned to
)
1355 unsigned int block_start
= 0, block_end
;
1356 struct buffer_head
*head
, *bh
;
1358 bh
= head
= page_buffers(page
);
1360 block_end
= block_start
+ bh
->b_size
;
1361 if (buffer_new(bh
)) {
1362 if (block_end
> from
&& block_start
< to
) {
1363 if (!PageUptodate(page
)) {
1364 unsigned start
, size
;
1366 start
= max(from
, block_start
);
1367 size
= min(to
, block_end
) - start
;
1369 zero_user(page
, start
, size
);
1370 write_end_fn(handle
, inode
, bh
);
1372 clear_buffer_new(bh
);
1375 block_start
= block_end
;
1376 bh
= bh
->b_this_page
;
1377 } while (bh
!= head
);
1380 static int ext4_journalled_write_end(struct file
*file
,
1381 struct address_space
*mapping
,
1382 loff_t pos
, unsigned len
, unsigned copied
,
1383 struct page
*page
, void *fsdata
)
1385 handle_t
*handle
= ext4_journal_current_handle();
1386 struct inode
*inode
= mapping
->host
;
1387 loff_t old_size
= inode
->i_size
;
1391 int size_changed
= 0;
1392 bool verity
= ext4_verity_in_progress(inode
);
1394 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1395 from
= pos
& (PAGE_SIZE
- 1);
1398 BUG_ON(!ext4_handle_valid(handle
));
1400 if (ext4_has_inline_data(inode
))
1401 return ext4_write_inline_data_end(inode
, pos
, len
, copied
, page
);
1403 if (unlikely(copied
< len
) && !PageUptodate(page
)) {
1405 ext4_journalled_zero_new_buffers(handle
, inode
, page
, from
, to
);
1407 if (unlikely(copied
< len
))
1408 ext4_journalled_zero_new_buffers(handle
, inode
, page
,
1410 ret
= ext4_walk_page_buffers(handle
, inode
, page_buffers(page
),
1411 from
, from
+ copied
, &partial
,
1414 SetPageUptodate(page
);
1417 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1418 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1419 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1423 if (old_size
< pos
&& !verity
)
1424 pagecache_isize_extended(inode
, old_size
, pos
);
1427 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1432 if (pos
+ len
> inode
->i_size
&& !verity
&& ext4_can_truncate(inode
))
1433 /* if we have allocated more blocks and copied
1434 * less. We will have blocks allocated outside
1435 * inode->i_size. So truncate them
1437 ext4_orphan_add(handle
, inode
);
1439 ret2
= ext4_journal_stop(handle
);
1442 if (pos
+ len
> inode
->i_size
&& !verity
) {
1443 ext4_truncate_failed_write(inode
);
1445 * If truncate failed early the inode might still be
1446 * on the orphan list; we need to make sure the inode
1447 * is removed from the orphan list in that case.
1450 ext4_orphan_del(NULL
, inode
);
1453 return ret
? ret
: copied
;
1457 * Reserve space for a single cluster
1459 static int ext4_da_reserve_space(struct inode
*inode
)
1461 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1462 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1466 * We will charge metadata quota at writeout time; this saves
1467 * us from metadata over-estimation, though we may go over by
1468 * a small amount in the end. Here we just reserve for data.
1470 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1474 spin_lock(&ei
->i_block_reservation_lock
);
1475 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1476 spin_unlock(&ei
->i_block_reservation_lock
);
1477 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1480 ei
->i_reserved_data_blocks
++;
1481 trace_ext4_da_reserve_space(inode
);
1482 spin_unlock(&ei
->i_block_reservation_lock
);
1484 return 0; /* success */
1487 void ext4_da_release_space(struct inode
*inode
, int to_free
)
1489 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1490 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1493 return; /* Nothing to release, exit */
1495 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1497 trace_ext4_da_release_space(inode
, to_free
);
1498 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1500 * if there aren't enough reserved blocks, then the
1501 * counter is messed up somewhere. Since this
1502 * function is called from invalidate page, it's
1503 * harmless to return without any action.
1505 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1506 "ino %lu, to_free %d with only %d reserved "
1507 "data blocks", inode
->i_ino
, to_free
,
1508 ei
->i_reserved_data_blocks
);
1510 to_free
= ei
->i_reserved_data_blocks
;
1512 ei
->i_reserved_data_blocks
-= to_free
;
1514 /* update fs dirty data blocks counter */
1515 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1517 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1519 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1523 * Delayed allocation stuff
1526 struct mpage_da_data
{
1527 struct inode
*inode
;
1528 struct writeback_control
*wbc
;
1530 pgoff_t first_page
; /* The first page to write */
1531 pgoff_t next_page
; /* Current page to examine */
1532 pgoff_t last_page
; /* Last page to examine */
1534 * Extent to map - this can be after first_page because that can be
1535 * fully mapped. We somewhat abuse m_flags to store whether the extent
1536 * is delalloc or unwritten.
1538 struct ext4_map_blocks map
;
1539 struct ext4_io_submit io_submit
; /* IO submission data */
1540 unsigned int do_map
:1;
1541 unsigned int scanned_until_end
:1;
1544 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1549 struct pagevec pvec
;
1550 struct inode
*inode
= mpd
->inode
;
1551 struct address_space
*mapping
= inode
->i_mapping
;
1553 /* This is necessary when next_page == 0. */
1554 if (mpd
->first_page
>= mpd
->next_page
)
1557 mpd
->scanned_until_end
= 0;
1558 index
= mpd
->first_page
;
1559 end
= mpd
->next_page
- 1;
1561 ext4_lblk_t start
, last
;
1562 start
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1563 last
= end
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1564 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1567 pagevec_init(&pvec
);
1568 while (index
<= end
) {
1569 nr_pages
= pagevec_lookup_range(&pvec
, mapping
, &index
, end
);
1572 for (i
= 0; i
< nr_pages
; i
++) {
1573 struct page
*page
= pvec
.pages
[i
];
1575 BUG_ON(!PageLocked(page
));
1576 BUG_ON(PageWriteback(page
));
1578 if (page_mapped(page
))
1579 clear_page_dirty_for_io(page
);
1580 block_invalidatepage(page
, 0, PAGE_SIZE
);
1581 ClearPageUptodate(page
);
1585 pagevec_release(&pvec
);
1589 static void ext4_print_free_blocks(struct inode
*inode
)
1591 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1592 struct super_block
*sb
= inode
->i_sb
;
1593 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1595 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1596 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1597 ext4_count_free_clusters(sb
)));
1598 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1599 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1600 (long long) EXT4_C2B(EXT4_SB(sb
),
1601 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1602 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1603 (long long) EXT4_C2B(EXT4_SB(sb
),
1604 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1605 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1606 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1607 ei
->i_reserved_data_blocks
);
1611 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct inode
*inode
,
1612 struct buffer_head
*bh
)
1614 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1618 * ext4_insert_delayed_block - adds a delayed block to the extents status
1619 * tree, incrementing the reserved cluster/block
1620 * count or making a pending reservation
1623 * @inode - file containing the newly added block
1624 * @lblk - logical block to be added
1626 * Returns 0 on success, negative error code on failure.
1628 static int ext4_insert_delayed_block(struct inode
*inode
, ext4_lblk_t lblk
)
1630 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1632 bool allocated
= false;
1633 bool reserved
= false;
1636 * If the cluster containing lblk is shared with a delayed,
1637 * written, or unwritten extent in a bigalloc file system, it's
1638 * already been accounted for and does not need to be reserved.
1639 * A pending reservation must be made for the cluster if it's
1640 * shared with a written or unwritten extent and doesn't already
1641 * have one. Written and unwritten extents can be purged from the
1642 * extents status tree if the system is under memory pressure, so
1643 * it's necessary to examine the extent tree if a search of the
1644 * extents status tree doesn't get a match.
1646 if (sbi
->s_cluster_ratio
== 1) {
1647 ret
= ext4_da_reserve_space(inode
);
1648 if (ret
!= 0) /* ENOSPC */
1651 } else { /* bigalloc */
1652 if (!ext4_es_scan_clu(inode
, &ext4_es_is_delonly
, lblk
)) {
1653 if (!ext4_es_scan_clu(inode
,
1654 &ext4_es_is_mapped
, lblk
)) {
1655 ret
= ext4_clu_mapped(inode
,
1656 EXT4_B2C(sbi
, lblk
));
1660 ret
= ext4_da_reserve_space(inode
);
1661 if (ret
!= 0) /* ENOSPC */
1673 ret
= ext4_es_insert_delayed_block(inode
, lblk
, allocated
);
1674 if (ret
&& reserved
)
1675 ext4_da_release_space(inode
, 1);
1682 * This function is grabs code from the very beginning of
1683 * ext4_map_blocks, but assumes that the caller is from delayed write
1684 * time. This function looks up the requested blocks and sets the
1685 * buffer delay bit under the protection of i_data_sem.
1687 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1688 struct ext4_map_blocks
*map
,
1689 struct buffer_head
*bh
)
1691 struct extent_status es
;
1693 sector_t invalid_block
= ~((sector_t
) 0xffff);
1694 #ifdef ES_AGGRESSIVE_TEST
1695 struct ext4_map_blocks orig_map
;
1697 memcpy(&orig_map
, map
, sizeof(*map
));
1700 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1704 ext_debug(inode
, "max_blocks %u, logical block %lu\n", map
->m_len
,
1705 (unsigned long) map
->m_lblk
);
1707 /* Lookup extent status tree firstly */
1708 if (ext4_es_lookup_extent(inode
, iblock
, NULL
, &es
)) {
1709 if (ext4_es_is_hole(&es
)) {
1711 down_read(&EXT4_I(inode
)->i_data_sem
);
1716 * Delayed extent could be allocated by fallocate.
1717 * So we need to check it.
1719 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1720 map_bh(bh
, inode
->i_sb
, invalid_block
);
1722 set_buffer_delay(bh
);
1726 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1727 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1728 if (retval
> map
->m_len
)
1729 retval
= map
->m_len
;
1730 map
->m_len
= retval
;
1731 if (ext4_es_is_written(&es
))
1732 map
->m_flags
|= EXT4_MAP_MAPPED
;
1733 else if (ext4_es_is_unwritten(&es
))
1734 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1738 #ifdef ES_AGGRESSIVE_TEST
1739 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1745 * Try to see if we can get the block without requesting a new
1746 * file system block.
1748 down_read(&EXT4_I(inode
)->i_data_sem
);
1749 if (ext4_has_inline_data(inode
))
1751 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1752 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1754 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1761 * XXX: __block_prepare_write() unmaps passed block,
1765 ret
= ext4_insert_delayed_block(inode
, map
->m_lblk
);
1771 map_bh(bh
, inode
->i_sb
, invalid_block
);
1773 set_buffer_delay(bh
);
1774 } else if (retval
> 0) {
1776 unsigned int status
;
1778 if (unlikely(retval
!= map
->m_len
)) {
1779 ext4_warning(inode
->i_sb
,
1780 "ES len assertion failed for inode "
1781 "%lu: retval %d != map->m_len %d",
1782 inode
->i_ino
, retval
, map
->m_len
);
1786 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1787 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1788 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1789 map
->m_pblk
, status
);
1795 up_read((&EXT4_I(inode
)->i_data_sem
));
1801 * This is a special get_block_t callback which is used by
1802 * ext4_da_write_begin(). It will either return mapped block or
1803 * reserve space for a single block.
1805 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1806 * We also have b_blocknr = -1 and b_bdev initialized properly
1808 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1809 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1810 * initialized properly.
1812 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1813 struct buffer_head
*bh
, int create
)
1815 struct ext4_map_blocks map
;
1818 BUG_ON(create
== 0);
1819 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1821 map
.m_lblk
= iblock
;
1825 * first, we need to know whether the block is allocated already
1826 * preallocated blocks are unmapped but should treated
1827 * the same as allocated blocks.
1829 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1833 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1834 ext4_update_bh_state(bh
, map
.m_flags
);
1836 if (buffer_unwritten(bh
)) {
1837 /* A delayed write to unwritten bh should be marked
1838 * new and mapped. Mapped ensures that we don't do
1839 * get_block multiple times when we write to the same
1840 * offset and new ensures that we do proper zero out
1841 * for partial write.
1844 set_buffer_mapped(bh
);
1849 static int __ext4_journalled_writepage(struct page
*page
,
1852 struct address_space
*mapping
= page
->mapping
;
1853 struct inode
*inode
= mapping
->host
;
1854 handle_t
*handle
= NULL
;
1855 int ret
= 0, err
= 0;
1856 int inline_data
= ext4_has_inline_data(inode
);
1857 struct buffer_head
*inode_bh
= NULL
;
1860 ClearPageChecked(page
);
1863 BUG_ON(page
->index
!= 0);
1864 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1865 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1866 if (inode_bh
== NULL
)
1870 * We need to release the page lock before we start the
1871 * journal, so grab a reference so the page won't disappear
1872 * out from under us.
1877 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1878 ext4_writepage_trans_blocks(inode
));
1879 if (IS_ERR(handle
)) {
1880 ret
= PTR_ERR(handle
);
1882 goto out_no_pagelock
;
1884 BUG_ON(!ext4_handle_valid(handle
));
1888 size
= i_size_read(inode
);
1889 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
1890 /* The page got truncated from under us */
1891 ext4_journal_stop(handle
);
1897 ret
= ext4_mark_inode_dirty(handle
, inode
);
1899 struct buffer_head
*page_bufs
= page_buffers(page
);
1901 if (page
->index
== size
>> PAGE_SHIFT
)
1902 len
= size
& ~PAGE_MASK
;
1906 ret
= ext4_walk_page_buffers(handle
, inode
, page_bufs
, 0, len
,
1907 NULL
, do_journal_get_write_access
);
1909 err
= ext4_walk_page_buffers(handle
, inode
, page_bufs
, 0, len
,
1910 NULL
, write_end_fn
);
1914 err
= ext4_jbd2_inode_add_write(handle
, inode
, page_offset(page
), len
);
1917 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1918 err
= ext4_journal_stop(handle
);
1922 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1931 * Note that we don't need to start a transaction unless we're journaling data
1932 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1933 * need to file the inode to the transaction's list in ordered mode because if
1934 * we are writing back data added by write(), the inode is already there and if
1935 * we are writing back data modified via mmap(), no one guarantees in which
1936 * transaction the data will hit the disk. In case we are journaling data, we
1937 * cannot start transaction directly because transaction start ranks above page
1938 * lock so we have to do some magic.
1940 * This function can get called via...
1941 * - ext4_writepages after taking page lock (have journal handle)
1942 * - journal_submit_inode_data_buffers (no journal handle)
1943 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1944 * - grab_page_cache when doing write_begin (have journal handle)
1946 * We don't do any block allocation in this function. If we have page with
1947 * multiple blocks we need to write those buffer_heads that are mapped. This
1948 * is important for mmaped based write. So if we do with blocksize 1K
1949 * truncate(f, 1024);
1950 * a = mmap(f, 0, 4096);
1952 * truncate(f, 4096);
1953 * we have in the page first buffer_head mapped via page_mkwrite call back
1954 * but other buffer_heads would be unmapped but dirty (dirty done via the
1955 * do_wp_page). So writepage should write the first block. If we modify
1956 * the mmap area beyond 1024 we will again get a page_fault and the
1957 * page_mkwrite callback will do the block allocation and mark the
1958 * buffer_heads mapped.
1960 * We redirty the page if we have any buffer_heads that is either delay or
1961 * unwritten in the page.
1963 * We can get recursively called as show below.
1965 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1968 * But since we don't do any block allocation we should not deadlock.
1969 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1971 static int ext4_writepage(struct page
*page
,
1972 struct writeback_control
*wbc
)
1977 struct buffer_head
*page_bufs
= NULL
;
1978 struct inode
*inode
= page
->mapping
->host
;
1979 struct ext4_io_submit io_submit
;
1980 bool keep_towrite
= false;
1982 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
)))) {
1983 inode
->i_mapping
->a_ops
->invalidatepage(page
, 0, PAGE_SIZE
);
1988 trace_ext4_writepage(page
);
1989 size
= i_size_read(inode
);
1990 if (page
->index
== size
>> PAGE_SHIFT
&&
1991 !ext4_verity_in_progress(inode
))
1992 len
= size
& ~PAGE_MASK
;
1996 /* Should never happen but for bugs in other kernel subsystems */
1997 if (!page_has_buffers(page
)) {
1998 ext4_warning_inode(inode
,
1999 "page %lu does not have buffers attached", page
->index
);
2000 ClearPageDirty(page
);
2005 page_bufs
= page_buffers(page
);
2007 * We cannot do block allocation or other extent handling in this
2008 * function. If there are buffers needing that, we have to redirty
2009 * the page. But we may reach here when we do a journal commit via
2010 * journal_submit_inode_data_buffers() and in that case we must write
2011 * allocated buffers to achieve data=ordered mode guarantees.
2013 * Also, if there is only one buffer per page (the fs block
2014 * size == the page size), if one buffer needs block
2015 * allocation or needs to modify the extent tree to clear the
2016 * unwritten flag, we know that the page can't be written at
2017 * all, so we might as well refuse the write immediately.
2018 * Unfortunately if the block size != page size, we can't as
2019 * easily detect this case using ext4_walk_page_buffers(), but
2020 * for the extremely common case, this is an optimization that
2021 * skips a useless round trip through ext4_bio_write_page().
2023 if (ext4_walk_page_buffers(NULL
, inode
, page_bufs
, 0, len
, NULL
,
2024 ext4_bh_delay_or_unwritten
)) {
2025 redirty_page_for_writepage(wbc
, page
);
2026 if ((current
->flags
& PF_MEMALLOC
) ||
2027 (inode
->i_sb
->s_blocksize
== PAGE_SIZE
)) {
2029 * For memory cleaning there's no point in writing only
2030 * some buffers. So just bail out. Warn if we came here
2031 * from direct reclaim.
2033 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2038 keep_towrite
= true;
2041 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2043 * It's mmapped pagecache. Add buffers and journal it. There
2044 * doesn't seem much point in redirtying the page here.
2046 return __ext4_journalled_writepage(page
, len
);
2048 ext4_io_submit_init(&io_submit
, wbc
);
2049 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2050 if (!io_submit
.io_end
) {
2051 redirty_page_for_writepage(wbc
, page
);
2055 ret
= ext4_bio_write_page(&io_submit
, page
, len
, keep_towrite
);
2056 ext4_io_submit(&io_submit
);
2057 /* Drop io_end reference we got from init */
2058 ext4_put_io_end_defer(io_submit
.io_end
);
2062 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
2068 BUG_ON(page
->index
!= mpd
->first_page
);
2069 clear_page_dirty_for_io(page
);
2071 * We have to be very careful here! Nothing protects writeback path
2072 * against i_size changes and the page can be writeably mapped into
2073 * page tables. So an application can be growing i_size and writing
2074 * data through mmap while writeback runs. clear_page_dirty_for_io()
2075 * write-protects our page in page tables and the page cannot get
2076 * written to again until we release page lock. So only after
2077 * clear_page_dirty_for_io() we are safe to sample i_size for
2078 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2079 * on the barrier provided by TestClearPageDirty in
2080 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2081 * after page tables are updated.
2083 size
= i_size_read(mpd
->inode
);
2084 if (page
->index
== size
>> PAGE_SHIFT
&&
2085 !ext4_verity_in_progress(mpd
->inode
))
2086 len
= size
& ~PAGE_MASK
;
2089 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, false);
2091 mpd
->wbc
->nr_to_write
--;
2097 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2100 * mballoc gives us at most this number of blocks...
2101 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2102 * The rest of mballoc seems to handle chunks up to full group size.
2104 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2107 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2109 * @mpd - extent of blocks
2110 * @lblk - logical number of the block in the file
2111 * @bh - buffer head we want to add to the extent
2113 * The function is used to collect contig. blocks in the same state. If the
2114 * buffer doesn't require mapping for writeback and we haven't started the
2115 * extent of buffers to map yet, the function returns 'true' immediately - the
2116 * caller can write the buffer right away. Otherwise the function returns true
2117 * if the block has been added to the extent, false if the block couldn't be
2120 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
2121 struct buffer_head
*bh
)
2123 struct ext4_map_blocks
*map
= &mpd
->map
;
2125 /* Buffer that doesn't need mapping for writeback? */
2126 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
2127 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
2128 /* So far no extent to map => we write the buffer right away */
2129 if (map
->m_len
== 0)
2134 /* First block in the extent? */
2135 if (map
->m_len
== 0) {
2136 /* We cannot map unless handle is started... */
2141 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2145 /* Don't go larger than mballoc is willing to allocate */
2146 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2149 /* Can we merge the block to our big extent? */
2150 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2151 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2159 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2161 * @mpd - extent of blocks for mapping
2162 * @head - the first buffer in the page
2163 * @bh - buffer we should start processing from
2164 * @lblk - logical number of the block in the file corresponding to @bh
2166 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2167 * the page for IO if all buffers in this page were mapped and there's no
2168 * accumulated extent of buffers to map or add buffers in the page to the
2169 * extent of buffers to map. The function returns 1 if the caller can continue
2170 * by processing the next page, 0 if it should stop adding buffers to the
2171 * extent to map because we cannot extend it anymore. It can also return value
2172 * < 0 in case of error during IO submission.
2174 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2175 struct buffer_head
*head
,
2176 struct buffer_head
*bh
,
2179 struct inode
*inode
= mpd
->inode
;
2181 ext4_lblk_t blocks
= (i_size_read(inode
) + i_blocksize(inode
) - 1)
2182 >> inode
->i_blkbits
;
2184 if (ext4_verity_in_progress(inode
))
2185 blocks
= EXT_MAX_BLOCKS
;
2188 BUG_ON(buffer_locked(bh
));
2190 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2191 /* Found extent to map? */
2194 /* Buffer needs mapping and handle is not started? */
2197 /* Everything mapped so far and we hit EOF */
2200 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2201 /* So far everything mapped? Submit the page for IO. */
2202 if (mpd
->map
.m_len
== 0) {
2203 err
= mpage_submit_page(mpd
, head
->b_page
);
2207 if (lblk
>= blocks
) {
2208 mpd
->scanned_until_end
= 1;
2215 * mpage_process_page - update page buffers corresponding to changed extent and
2216 * may submit fully mapped page for IO
2218 * @mpd - description of extent to map, on return next extent to map
2219 * @m_lblk - logical block mapping.
2220 * @m_pblk - corresponding physical mapping.
2221 * @map_bh - determines on return whether this page requires any further
2223 * Scan given page buffers corresponding to changed extent and update buffer
2224 * state according to new extent state.
2225 * We map delalloc buffers to their physical location, clear unwritten bits.
2226 * If the given page is not fully mapped, we update @map to the next extent in
2227 * the given page that needs mapping & return @map_bh as true.
2229 static int mpage_process_page(struct mpage_da_data
*mpd
, struct page
*page
,
2230 ext4_lblk_t
*m_lblk
, ext4_fsblk_t
*m_pblk
,
2233 struct buffer_head
*head
, *bh
;
2234 ext4_io_end_t
*io_end
= mpd
->io_submit
.io_end
;
2235 ext4_lblk_t lblk
= *m_lblk
;
2236 ext4_fsblk_t pblock
= *m_pblk
;
2238 int blkbits
= mpd
->inode
->i_blkbits
;
2239 ssize_t io_end_size
= 0;
2240 struct ext4_io_end_vec
*io_end_vec
= ext4_last_io_end_vec(io_end
);
2242 bh
= head
= page_buffers(page
);
2244 if (lblk
< mpd
->map
.m_lblk
)
2246 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2248 * Buffer after end of mapped extent.
2249 * Find next buffer in the page to map.
2252 mpd
->map
.m_flags
= 0;
2253 io_end_vec
->size
+= io_end_size
;
2255 err
= mpage_process_page_bufs(mpd
, head
, bh
, lblk
);
2258 if (!err
&& mpd
->map
.m_len
&& mpd
->map
.m_lblk
> lblk
) {
2259 io_end_vec
= ext4_alloc_io_end_vec(io_end
);
2260 if (IS_ERR(io_end_vec
)) {
2261 err
= PTR_ERR(io_end_vec
);
2264 io_end_vec
->offset
= (loff_t
)mpd
->map
.m_lblk
<< blkbits
;
2269 if (buffer_delay(bh
)) {
2270 clear_buffer_delay(bh
);
2271 bh
->b_blocknr
= pblock
++;
2273 clear_buffer_unwritten(bh
);
2274 io_end_size
+= (1 << blkbits
);
2275 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2277 io_end_vec
->size
+= io_end_size
;
2286 * mpage_map_buffers - update buffers corresponding to changed extent and
2287 * submit fully mapped pages for IO
2289 * @mpd - description of extent to map, on return next extent to map
2291 * Scan buffers corresponding to changed extent (we expect corresponding pages
2292 * to be already locked) and update buffer state according to new extent state.
2293 * We map delalloc buffers to their physical location, clear unwritten bits,
2294 * and mark buffers as uninit when we perform writes to unwritten extents
2295 * and do extent conversion after IO is finished. If the last page is not fully
2296 * mapped, we update @map to the next extent in the last page that needs
2297 * mapping. Otherwise we submit the page for IO.
2299 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2301 struct pagevec pvec
;
2303 struct inode
*inode
= mpd
->inode
;
2304 int bpp_bits
= PAGE_SHIFT
- inode
->i_blkbits
;
2307 ext4_fsblk_t pblock
;
2309 bool map_bh
= false;
2311 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2312 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2313 lblk
= start
<< bpp_bits
;
2314 pblock
= mpd
->map
.m_pblk
;
2316 pagevec_init(&pvec
);
2317 while (start
<= end
) {
2318 nr_pages
= pagevec_lookup_range(&pvec
, inode
->i_mapping
,
2322 for (i
= 0; i
< nr_pages
; i
++) {
2323 struct page
*page
= pvec
.pages
[i
];
2325 err
= mpage_process_page(mpd
, page
, &lblk
, &pblock
,
2328 * If map_bh is true, means page may require further bh
2329 * mapping, or maybe the page was submitted for IO.
2330 * So we return to call further extent mapping.
2332 if (err
< 0 || map_bh
)
2334 /* Page fully mapped - let IO run! */
2335 err
= mpage_submit_page(mpd
, page
);
2339 pagevec_release(&pvec
);
2341 /* Extent fully mapped and matches with page boundary. We are done. */
2343 mpd
->map
.m_flags
= 0;
2346 pagevec_release(&pvec
);
2350 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2352 struct inode
*inode
= mpd
->inode
;
2353 struct ext4_map_blocks
*map
= &mpd
->map
;
2354 int get_blocks_flags
;
2355 int err
, dioread_nolock
;
2357 trace_ext4_da_write_pages_extent(inode
, map
);
2359 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2360 * to convert an unwritten extent to be initialized (in the case
2361 * where we have written into one or more preallocated blocks). It is
2362 * possible that we're going to need more metadata blocks than
2363 * previously reserved. However we must not fail because we're in
2364 * writeback and there is nothing we can do about it so it might result
2365 * in data loss. So use reserved blocks to allocate metadata if
2368 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2369 * the blocks in question are delalloc blocks. This indicates
2370 * that the blocks and quotas has already been checked when
2371 * the data was copied into the page cache.
2373 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2374 EXT4_GET_BLOCKS_METADATA_NOFAIL
|
2375 EXT4_GET_BLOCKS_IO_SUBMIT
;
2376 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2378 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2379 if (map
->m_flags
& BIT(BH_Delay
))
2380 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2382 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2385 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2386 if (!mpd
->io_submit
.io_end
->handle
&&
2387 ext4_handle_valid(handle
)) {
2388 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2389 handle
->h_rsv_handle
= NULL
;
2391 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2394 BUG_ON(map
->m_len
== 0);
2399 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2400 * mpd->len and submit pages underlying it for IO
2402 * @handle - handle for journal operations
2403 * @mpd - extent to map
2404 * @give_up_on_write - we set this to true iff there is a fatal error and there
2405 * is no hope of writing the data. The caller should discard
2406 * dirty pages to avoid infinite loops.
2408 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2409 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2410 * them to initialized or split the described range from larger unwritten
2411 * extent. Note that we need not map all the described range since allocation
2412 * can return less blocks or the range is covered by more unwritten extents. We
2413 * cannot map more because we are limited by reserved transaction credits. On
2414 * the other hand we always make sure that the last touched page is fully
2415 * mapped so that it can be written out (and thus forward progress is
2416 * guaranteed). After mapping we submit all mapped pages for IO.
2418 static int mpage_map_and_submit_extent(handle_t
*handle
,
2419 struct mpage_da_data
*mpd
,
2420 bool *give_up_on_write
)
2422 struct inode
*inode
= mpd
->inode
;
2423 struct ext4_map_blocks
*map
= &mpd
->map
;
2427 ext4_io_end_t
*io_end
= mpd
->io_submit
.io_end
;
2428 struct ext4_io_end_vec
*io_end_vec
;
2430 io_end_vec
= ext4_alloc_io_end_vec(io_end
);
2431 if (IS_ERR(io_end_vec
))
2432 return PTR_ERR(io_end_vec
);
2433 io_end_vec
->offset
= ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2435 err
= mpage_map_one_extent(handle
, mpd
);
2437 struct super_block
*sb
= inode
->i_sb
;
2439 if (ext4_forced_shutdown(EXT4_SB(sb
)) ||
2440 ext4_test_mount_flag(sb
, EXT4_MF_FS_ABORTED
))
2441 goto invalidate_dirty_pages
;
2443 * Let the uper layers retry transient errors.
2444 * In the case of ENOSPC, if ext4_count_free_blocks()
2445 * is non-zero, a commit should free up blocks.
2447 if ((err
== -ENOMEM
) ||
2448 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2450 goto update_disksize
;
2453 ext4_msg(sb
, KERN_CRIT
,
2454 "Delayed block allocation failed for "
2455 "inode %lu at logical offset %llu with"
2456 " max blocks %u with error %d",
2458 (unsigned long long)map
->m_lblk
,
2459 (unsigned)map
->m_len
, -err
);
2460 ext4_msg(sb
, KERN_CRIT
,
2461 "This should not happen!! Data will "
2464 ext4_print_free_blocks(inode
);
2465 invalidate_dirty_pages
:
2466 *give_up_on_write
= true;
2471 * Update buffer state, submit mapped pages, and get us new
2474 err
= mpage_map_and_submit_buffers(mpd
);
2476 goto update_disksize
;
2477 } while (map
->m_len
);
2481 * Update on-disk size after IO is submitted. Races with
2482 * truncate are avoided by checking i_size under i_data_sem.
2484 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_SHIFT
;
2485 if (disksize
> READ_ONCE(EXT4_I(inode
)->i_disksize
)) {
2489 down_write(&EXT4_I(inode
)->i_data_sem
);
2490 i_size
= i_size_read(inode
);
2491 if (disksize
> i_size
)
2493 if (disksize
> EXT4_I(inode
)->i_disksize
)
2494 EXT4_I(inode
)->i_disksize
= disksize
;
2495 up_write(&EXT4_I(inode
)->i_data_sem
);
2496 err2
= ext4_mark_inode_dirty(handle
, inode
);
2498 ext4_error_err(inode
->i_sb
, -err2
,
2499 "Failed to mark inode %lu dirty",
2509 * Calculate the total number of credits to reserve for one writepages
2510 * iteration. This is called from ext4_writepages(). We map an extent of
2511 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2512 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2513 * bpp - 1 blocks in bpp different extents.
2515 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2517 int bpp
= ext4_journal_blocks_per_page(inode
);
2519 return ext4_meta_trans_blocks(inode
,
2520 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2524 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2525 * and underlying extent to map
2527 * @mpd - where to look for pages
2529 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2530 * IO immediately. When we find a page which isn't mapped we start accumulating
2531 * extent of buffers underlying these pages that needs mapping (formed by
2532 * either delayed or unwritten buffers). We also lock the pages containing
2533 * these buffers. The extent found is returned in @mpd structure (starting at
2534 * mpd->lblk with length mpd->len blocks).
2536 * Note that this function can attach bios to one io_end structure which are
2537 * neither logically nor physically contiguous. Although it may seem as an
2538 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2539 * case as we need to track IO to all buffers underlying a page in one io_end.
2541 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2543 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2544 struct pagevec pvec
;
2545 unsigned int nr_pages
;
2546 long left
= mpd
->wbc
->nr_to_write
;
2547 pgoff_t index
= mpd
->first_page
;
2548 pgoff_t end
= mpd
->last_page
;
2551 int blkbits
= mpd
->inode
->i_blkbits
;
2553 struct buffer_head
*head
;
2555 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2556 tag
= PAGECACHE_TAG_TOWRITE
;
2558 tag
= PAGECACHE_TAG_DIRTY
;
2560 pagevec_init(&pvec
);
2562 mpd
->next_page
= index
;
2563 while (index
<= end
) {
2564 nr_pages
= pagevec_lookup_range_tag(&pvec
, mapping
, &index
, end
,
2569 for (i
= 0; i
< nr_pages
; i
++) {
2570 struct page
*page
= pvec
.pages
[i
];
2573 * Accumulated enough dirty pages? This doesn't apply
2574 * to WB_SYNC_ALL mode. For integrity sync we have to
2575 * keep going because someone may be concurrently
2576 * dirtying pages, and we might have synced a lot of
2577 * newly appeared dirty pages, but have not synced all
2578 * of the old dirty pages.
2580 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2583 /* If we can't merge this page, we are done. */
2584 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2589 * If the page is no longer dirty, or its mapping no
2590 * longer corresponds to inode we are writing (which
2591 * means it has been truncated or invalidated), or the
2592 * page is already under writeback and we are not doing
2593 * a data integrity writeback, skip the page
2595 if (!PageDirty(page
) ||
2596 (PageWriteback(page
) &&
2597 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2598 unlikely(page
->mapping
!= mapping
)) {
2603 wait_on_page_writeback(page
);
2604 BUG_ON(PageWriteback(page
));
2607 * Should never happen but for buggy code in
2608 * other subsystems that call
2609 * set_page_dirty() without properly warning
2610 * the file system first. See [1] for more
2613 * [1] https://lore.kernel.org/linux-mm/20180103100430.GE4911@quack2.suse.cz
2615 if (!page_has_buffers(page
)) {
2616 ext4_warning_inode(mpd
->inode
, "page %lu does not have buffers attached", page
->index
);
2617 ClearPageDirty(page
);
2622 if (mpd
->map
.m_len
== 0)
2623 mpd
->first_page
= page
->index
;
2624 mpd
->next_page
= page
->index
+ 1;
2625 /* Add all dirty buffers to mpd */
2626 lblk
= ((ext4_lblk_t
)page
->index
) <<
2627 (PAGE_SHIFT
- blkbits
);
2628 head
= page_buffers(page
);
2629 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2635 pagevec_release(&pvec
);
2638 mpd
->scanned_until_end
= 1;
2641 pagevec_release(&pvec
);
2645 static int ext4_writepages(struct address_space
*mapping
,
2646 struct writeback_control
*wbc
)
2648 pgoff_t writeback_index
= 0;
2649 long nr_to_write
= wbc
->nr_to_write
;
2650 int range_whole
= 0;
2652 handle_t
*handle
= NULL
;
2653 struct mpage_da_data mpd
;
2654 struct inode
*inode
= mapping
->host
;
2655 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2656 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2657 struct blk_plug plug
;
2658 bool give_up_on_write
= false;
2660 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
2663 percpu_down_read(&sbi
->s_writepages_rwsem
);
2664 trace_ext4_writepages(inode
, wbc
);
2667 * No pages to write? This is mainly a kludge to avoid starting
2668 * a transaction for special inodes like journal inode on last iput()
2669 * because that could violate lock ordering on umount
2671 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2672 goto out_writepages
;
2674 if (ext4_should_journal_data(inode
)) {
2675 ret
= generic_writepages(mapping
, wbc
);
2676 goto out_writepages
;
2680 * If the filesystem has aborted, it is read-only, so return
2681 * right away instead of dumping stack traces later on that
2682 * will obscure the real source of the problem. We test
2683 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2684 * the latter could be true if the filesystem is mounted
2685 * read-only, and in that case, ext4_writepages should
2686 * *never* be called, so if that ever happens, we would want
2689 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping
->host
->i_sb
)) ||
2690 ext4_test_mount_flag(inode
->i_sb
, EXT4_MF_FS_ABORTED
))) {
2692 goto out_writepages
;
2696 * If we have inline data and arrive here, it means that
2697 * we will soon create the block for the 1st page, so
2698 * we'd better clear the inline data here.
2700 if (ext4_has_inline_data(inode
)) {
2701 /* Just inode will be modified... */
2702 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2703 if (IS_ERR(handle
)) {
2704 ret
= PTR_ERR(handle
);
2705 goto out_writepages
;
2707 BUG_ON(ext4_test_inode_state(inode
,
2708 EXT4_STATE_MAY_INLINE_DATA
));
2709 ext4_destroy_inline_data(handle
, inode
);
2710 ext4_journal_stop(handle
);
2713 if (ext4_should_dioread_nolock(inode
)) {
2715 * We may need to convert up to one extent per block in
2716 * the page and we may dirty the inode.
2718 rsv_blocks
= 1 + ext4_chunk_trans_blocks(inode
,
2719 PAGE_SIZE
>> inode
->i_blkbits
);
2722 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2725 if (wbc
->range_cyclic
) {
2726 writeback_index
= mapping
->writeback_index
;
2727 if (writeback_index
)
2729 mpd
.first_page
= writeback_index
;
2732 mpd
.first_page
= wbc
->range_start
>> PAGE_SHIFT
;
2733 mpd
.last_page
= wbc
->range_end
>> PAGE_SHIFT
;
2738 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2740 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2741 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2742 blk_start_plug(&plug
);
2745 * First writeback pages that don't need mapping - we can avoid
2746 * starting a transaction unnecessarily and also avoid being blocked
2747 * in the block layer on device congestion while having transaction
2751 mpd
.scanned_until_end
= 0;
2752 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2753 if (!mpd
.io_submit
.io_end
) {
2757 ret
= mpage_prepare_extent_to_map(&mpd
);
2758 /* Unlock pages we didn't use */
2759 mpage_release_unused_pages(&mpd
, false);
2760 /* Submit prepared bio */
2761 ext4_io_submit(&mpd
.io_submit
);
2762 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2763 mpd
.io_submit
.io_end
= NULL
;
2767 while (!mpd
.scanned_until_end
&& wbc
->nr_to_write
> 0) {
2768 /* For each extent of pages we use new io_end */
2769 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2770 if (!mpd
.io_submit
.io_end
) {
2776 * We have two constraints: We find one extent to map and we
2777 * must always write out whole page (makes a difference when
2778 * blocksize < pagesize) so that we don't block on IO when we
2779 * try to write out the rest of the page. Journalled mode is
2780 * not supported by delalloc.
2782 BUG_ON(ext4_should_journal_data(inode
));
2783 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2785 /* start a new transaction */
2786 handle
= ext4_journal_start_with_reserve(inode
,
2787 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2788 if (IS_ERR(handle
)) {
2789 ret
= PTR_ERR(handle
);
2790 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2791 "%ld pages, ino %lu; err %d", __func__
,
2792 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2793 /* Release allocated io_end */
2794 ext4_put_io_end(mpd
.io_submit
.io_end
);
2795 mpd
.io_submit
.io_end
= NULL
;
2800 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2801 ret
= mpage_prepare_extent_to_map(&mpd
);
2802 if (!ret
&& mpd
.map
.m_len
)
2803 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2806 * Caution: If the handle is synchronous,
2807 * ext4_journal_stop() can wait for transaction commit
2808 * to finish which may depend on writeback of pages to
2809 * complete or on page lock to be released. In that
2810 * case, we have to wait until after we have
2811 * submitted all the IO, released page locks we hold,
2812 * and dropped io_end reference (for extent conversion
2813 * to be able to complete) before stopping the handle.
2815 if (!ext4_handle_valid(handle
) || handle
->h_sync
== 0) {
2816 ext4_journal_stop(handle
);
2820 /* Unlock pages we didn't use */
2821 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2822 /* Submit prepared bio */
2823 ext4_io_submit(&mpd
.io_submit
);
2826 * Drop our io_end reference we got from init. We have
2827 * to be careful and use deferred io_end finishing if
2828 * we are still holding the transaction as we can
2829 * release the last reference to io_end which may end
2830 * up doing unwritten extent conversion.
2833 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2834 ext4_journal_stop(handle
);
2836 ext4_put_io_end(mpd
.io_submit
.io_end
);
2837 mpd
.io_submit
.io_end
= NULL
;
2839 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2841 * Commit the transaction which would
2842 * free blocks released in the transaction
2845 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2849 /* Fatal error - ENOMEM, EIO... */
2854 blk_finish_plug(&plug
);
2855 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2857 mpd
.last_page
= writeback_index
- 1;
2863 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2865 * Set the writeback_index so that range_cyclic
2866 * mode will write it back later
2868 mapping
->writeback_index
= mpd
.first_page
;
2871 trace_ext4_writepages_result(inode
, wbc
, ret
,
2872 nr_to_write
- wbc
->nr_to_write
);
2873 percpu_up_read(&sbi
->s_writepages_rwsem
);
2877 static int ext4_dax_writepages(struct address_space
*mapping
,
2878 struct writeback_control
*wbc
)
2881 long nr_to_write
= wbc
->nr_to_write
;
2882 struct inode
*inode
= mapping
->host
;
2883 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2885 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
2888 percpu_down_read(&sbi
->s_writepages_rwsem
);
2889 trace_ext4_writepages(inode
, wbc
);
2891 ret
= dax_writeback_mapping_range(mapping
, sbi
->s_daxdev
, wbc
);
2892 trace_ext4_writepages_result(inode
, wbc
, ret
,
2893 nr_to_write
- wbc
->nr_to_write
);
2894 percpu_up_read(&sbi
->s_writepages_rwsem
);
2898 static int ext4_nonda_switch(struct super_block
*sb
)
2900 s64 free_clusters
, dirty_clusters
;
2901 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2904 * switch to non delalloc mode if we are running low
2905 * on free block. The free block accounting via percpu
2906 * counters can get slightly wrong with percpu_counter_batch getting
2907 * accumulated on each CPU without updating global counters
2908 * Delalloc need an accurate free block accounting. So switch
2909 * to non delalloc when we are near to error range.
2912 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2914 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2916 * Start pushing delalloc when 1/2 of free blocks are dirty.
2918 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2919 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2921 if (2 * free_clusters
< 3 * dirty_clusters
||
2922 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2924 * free block count is less than 150% of dirty blocks
2925 * or free blocks is less than watermark
2932 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2933 loff_t pos
, unsigned len
, unsigned flags
,
2934 struct page
**pagep
, void **fsdata
)
2936 int ret
, retries
= 0;
2939 struct inode
*inode
= mapping
->host
;
2941 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
2944 index
= pos
>> PAGE_SHIFT
;
2946 if (ext4_nonda_switch(inode
->i_sb
) || S_ISLNK(inode
->i_mode
) ||
2947 ext4_verity_in_progress(inode
)) {
2948 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2949 return ext4_write_begin(file
, mapping
, pos
,
2950 len
, flags
, pagep
, fsdata
);
2952 *fsdata
= (void *)0;
2953 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2955 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2956 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2966 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2970 /* In case writeback began while the page was unlocked */
2971 wait_for_stable_page(page
);
2973 #ifdef CONFIG_FS_ENCRYPTION
2974 ret
= ext4_block_write_begin(page
, pos
, len
,
2975 ext4_da_get_block_prep
);
2977 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2983 * block_write_begin may have instantiated a few blocks
2984 * outside i_size. Trim these off again. Don't need
2985 * i_size_read because we hold inode lock.
2987 if (pos
+ len
> inode
->i_size
)
2988 ext4_truncate_failed_write(inode
);
2990 if (ret
== -ENOSPC
&&
2991 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3001 * Check if we should update i_disksize
3002 * when write to the end of file but not require block allocation
3004 static int ext4_da_should_update_i_disksize(struct page
*page
,
3005 unsigned long offset
)
3007 struct buffer_head
*bh
;
3008 struct inode
*inode
= page
->mapping
->host
;
3012 bh
= page_buffers(page
);
3013 idx
= offset
>> inode
->i_blkbits
;
3015 for (i
= 0; i
< idx
; i
++)
3016 bh
= bh
->b_this_page
;
3018 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3023 static int ext4_da_write_end(struct file
*file
,
3024 struct address_space
*mapping
,
3025 loff_t pos
, unsigned len
, unsigned copied
,
3026 struct page
*page
, void *fsdata
)
3028 struct inode
*inode
= mapping
->host
;
3030 unsigned long start
, end
;
3031 int write_mode
= (int)(unsigned long)fsdata
;
3033 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
3034 return ext4_write_end(file
, mapping
, pos
,
3035 len
, copied
, page
, fsdata
);
3037 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3039 if (write_mode
!= CONVERT_INLINE_DATA
&&
3040 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
3041 ext4_has_inline_data(inode
))
3042 return ext4_write_inline_data_end(inode
, pos
, len
, copied
, page
);
3044 start
= pos
& (PAGE_SIZE
- 1);
3045 end
= start
+ copied
- 1;
3048 * Since we are holding inode lock, we are sure i_disksize <=
3049 * i_size. We also know that if i_disksize < i_size, there are
3050 * delalloc writes pending in the range upto i_size. If the end of
3051 * the current write is <= i_size, there's no need to touch
3052 * i_disksize since writeback will push i_disksize upto i_size
3053 * eventually. If the end of the current write is > i_size and
3054 * inside an allocated block (ext4_da_should_update_i_disksize()
3055 * check), we need to update i_disksize here as neither
3056 * ext4_writepage() nor certain ext4_writepages() paths not
3057 * allocating blocks update i_disksize.
3059 * Note that we defer inode dirtying to generic_write_end() /
3060 * ext4_da_write_inline_data_end().
3062 new_i_size
= pos
+ copied
;
3063 if (copied
&& new_i_size
> inode
->i_size
&&
3064 ext4_da_should_update_i_disksize(page
, end
))
3065 ext4_update_i_disksize(inode
, new_i_size
);
3067 return generic_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
3071 * Force all delayed allocation blocks to be allocated for a given inode.
3073 int ext4_alloc_da_blocks(struct inode
*inode
)
3075 trace_ext4_alloc_da_blocks(inode
);
3077 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
3081 * We do something simple for now. The filemap_flush() will
3082 * also start triggering a write of the data blocks, which is
3083 * not strictly speaking necessary (and for users of
3084 * laptop_mode, not even desirable). However, to do otherwise
3085 * would require replicating code paths in:
3087 * ext4_writepages() ->
3088 * write_cache_pages() ---> (via passed in callback function)
3089 * __mpage_da_writepage() -->
3090 * mpage_add_bh_to_extent()
3091 * mpage_da_map_blocks()
3093 * The problem is that write_cache_pages(), located in
3094 * mm/page-writeback.c, marks pages clean in preparation for
3095 * doing I/O, which is not desirable if we're not planning on
3098 * We could call write_cache_pages(), and then redirty all of
3099 * the pages by calling redirty_page_for_writepage() but that
3100 * would be ugly in the extreme. So instead we would need to
3101 * replicate parts of the code in the above functions,
3102 * simplifying them because we wouldn't actually intend to
3103 * write out the pages, but rather only collect contiguous
3104 * logical block extents, call the multi-block allocator, and
3105 * then update the buffer heads with the block allocations.
3107 * For now, though, we'll cheat by calling filemap_flush(),
3108 * which will map the blocks, and start the I/O, but not
3109 * actually wait for the I/O to complete.
3111 return filemap_flush(inode
->i_mapping
);
3115 * bmap() is special. It gets used by applications such as lilo and by
3116 * the swapper to find the on-disk block of a specific piece of data.
3118 * Naturally, this is dangerous if the block concerned is still in the
3119 * journal. If somebody makes a swapfile on an ext4 data-journaling
3120 * filesystem and enables swap, then they may get a nasty shock when the
3121 * data getting swapped to that swapfile suddenly gets overwritten by
3122 * the original zero's written out previously to the journal and
3123 * awaiting writeback in the kernel's buffer cache.
3125 * So, if we see any bmap calls here on a modified, data-journaled file,
3126 * take extra steps to flush any blocks which might be in the cache.
3128 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3130 struct inode
*inode
= mapping
->host
;
3135 * We can get here for an inline file via the FIBMAP ioctl
3137 if (ext4_has_inline_data(inode
))
3140 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3141 test_opt(inode
->i_sb
, DELALLOC
)) {
3143 * With delalloc we want to sync the file
3144 * so that we can make sure we allocate
3147 filemap_write_and_wait(mapping
);
3150 if (EXT4_JOURNAL(inode
) &&
3151 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3153 * This is a REALLY heavyweight approach, but the use of
3154 * bmap on dirty files is expected to be extremely rare:
3155 * only if we run lilo or swapon on a freshly made file
3156 * do we expect this to happen.
3158 * (bmap requires CAP_SYS_RAWIO so this does not
3159 * represent an unprivileged user DOS attack --- we'd be
3160 * in trouble if mortal users could trigger this path at
3163 * NB. EXT4_STATE_JDATA is not set on files other than
3164 * regular files. If somebody wants to bmap a directory
3165 * or symlink and gets confused because the buffer
3166 * hasn't yet been flushed to disk, they deserve
3167 * everything they get.
3170 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3171 journal
= EXT4_JOURNAL(inode
);
3172 jbd2_journal_lock_updates(journal
);
3173 err
= jbd2_journal_flush(journal
, 0);
3174 jbd2_journal_unlock_updates(journal
);
3180 return iomap_bmap(mapping
, block
, &ext4_iomap_ops
);
3183 static int ext4_readpage(struct file
*file
, struct page
*page
)
3186 struct inode
*inode
= page
->mapping
->host
;
3188 trace_ext4_readpage(page
);
3190 if (ext4_has_inline_data(inode
))
3191 ret
= ext4_readpage_inline(inode
, page
);
3194 return ext4_mpage_readpages(inode
, NULL
, page
);
3199 static void ext4_readahead(struct readahead_control
*rac
)
3201 struct inode
*inode
= rac
->mapping
->host
;
3203 /* If the file has inline data, no need to do readahead. */
3204 if (ext4_has_inline_data(inode
))
3207 ext4_mpage_readpages(inode
, rac
, NULL
);
3210 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3211 unsigned int length
)
3213 trace_ext4_invalidatepage(page
, offset
, length
);
3215 /* No journalling happens on data buffers when this function is used */
3216 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3218 block_invalidatepage(page
, offset
, length
);
3221 static int __ext4_journalled_invalidatepage(struct page
*page
,
3222 unsigned int offset
,
3223 unsigned int length
)
3225 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3227 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3230 * If it's a full truncate we just forget about the pending dirtying
3232 if (offset
== 0 && length
== PAGE_SIZE
)
3233 ClearPageChecked(page
);
3235 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3238 /* Wrapper for aops... */
3239 static void ext4_journalled_invalidatepage(struct page
*page
,
3240 unsigned int offset
,
3241 unsigned int length
)
3243 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3246 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3248 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3250 trace_ext4_releasepage(page
);
3252 /* Page has dirty journalled data -> cannot release */
3253 if (PageChecked(page
))
3256 return jbd2_journal_try_to_free_buffers(journal
, page
);
3258 return try_to_free_buffers(page
);
3261 static bool ext4_inode_datasync_dirty(struct inode
*inode
)
3263 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
3266 if (jbd2_transaction_committed(journal
,
3267 EXT4_I(inode
)->i_datasync_tid
))
3269 if (test_opt2(inode
->i_sb
, JOURNAL_FAST_COMMIT
))
3270 return !list_empty(&EXT4_I(inode
)->i_fc_list
);
3274 /* Any metadata buffers to write? */
3275 if (!list_empty(&inode
->i_mapping
->private_list
))
3277 return inode
->i_state
& I_DIRTY_DATASYNC
;
3280 static void ext4_set_iomap(struct inode
*inode
, struct iomap
*iomap
,
3281 struct ext4_map_blocks
*map
, loff_t offset
,
3282 loff_t length
, unsigned int flags
)
3284 u8 blkbits
= inode
->i_blkbits
;
3287 * Writes that span EOF might trigger an I/O size update on completion,
3288 * so consider them to be dirty for the purpose of O_DSYNC, even if
3289 * there is no other metadata changes being made or are pending.
3292 if (ext4_inode_datasync_dirty(inode
) ||
3293 offset
+ length
> i_size_read(inode
))
3294 iomap
->flags
|= IOMAP_F_DIRTY
;
3296 if (map
->m_flags
& EXT4_MAP_NEW
)
3297 iomap
->flags
|= IOMAP_F_NEW
;
3299 if (flags
& IOMAP_DAX
)
3300 iomap
->dax_dev
= EXT4_SB(inode
->i_sb
)->s_daxdev
;
3302 iomap
->bdev
= inode
->i_sb
->s_bdev
;
3303 iomap
->offset
= (u64
) map
->m_lblk
<< blkbits
;
3304 iomap
->length
= (u64
) map
->m_len
<< blkbits
;
3306 if ((map
->m_flags
& EXT4_MAP_MAPPED
) &&
3307 !ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3308 iomap
->flags
|= IOMAP_F_MERGED
;
3311 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3312 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3313 * set. In order for any allocated unwritten extents to be converted
3314 * into written extents correctly within the ->end_io() handler, we
3315 * need to ensure that the iomap->type is set appropriately. Hence, the
3316 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3319 if (map
->m_flags
& EXT4_MAP_UNWRITTEN
) {
3320 iomap
->type
= IOMAP_UNWRITTEN
;
3321 iomap
->addr
= (u64
) map
->m_pblk
<< blkbits
;
3322 if (flags
& IOMAP_DAX
)
3323 iomap
->addr
+= EXT4_SB(inode
->i_sb
)->s_dax_part_off
;
3324 } else if (map
->m_flags
& EXT4_MAP_MAPPED
) {
3325 iomap
->type
= IOMAP_MAPPED
;
3326 iomap
->addr
= (u64
) map
->m_pblk
<< blkbits
;
3327 if (flags
& IOMAP_DAX
)
3328 iomap
->addr
+= EXT4_SB(inode
->i_sb
)->s_dax_part_off
;
3330 iomap
->type
= IOMAP_HOLE
;
3331 iomap
->addr
= IOMAP_NULL_ADDR
;
3335 static int ext4_iomap_alloc(struct inode
*inode
, struct ext4_map_blocks
*map
,
3339 u8 blkbits
= inode
->i_blkbits
;
3340 int ret
, dio_credits
, m_flags
= 0, retries
= 0;
3343 * Trim the mapping request to the maximum value that we can map at
3344 * once for direct I/O.
3346 if (map
->m_len
> DIO_MAX_BLOCKS
)
3347 map
->m_len
= DIO_MAX_BLOCKS
;
3348 dio_credits
= ext4_chunk_trans_blocks(inode
, map
->m_len
);
3352 * Either we allocate blocks and then don't get an unwritten extent, so
3353 * in that case we have reserved enough credits. Or, the blocks are
3354 * already allocated and unwritten. In that case, the extent conversion
3355 * fits into the credits as well.
3357 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, dio_credits
);
3359 return PTR_ERR(handle
);
3362 * DAX and direct I/O are the only two operations that are currently
3363 * supported with IOMAP_WRITE.
3365 WARN_ON(!(flags
& (IOMAP_DAX
| IOMAP_DIRECT
)));
3366 if (flags
& IOMAP_DAX
)
3367 m_flags
= EXT4_GET_BLOCKS_CREATE_ZERO
;
3369 * We use i_size instead of i_disksize here because delalloc writeback
3370 * can complete at any point during the I/O and subsequently push the
3371 * i_disksize out to i_size. This could be beyond where direct I/O is
3372 * happening and thus expose allocated blocks to direct I/O reads.
3374 else if (((loff_t
)map
->m_lblk
<< blkbits
) >= i_size_read(inode
))
3375 m_flags
= EXT4_GET_BLOCKS_CREATE
;
3376 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3377 m_flags
= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
3379 ret
= ext4_map_blocks(handle
, inode
, map
, m_flags
);
3382 * We cannot fill holes in indirect tree based inodes as that could
3383 * expose stale data in the case of a crash. Use the magic error code
3384 * to fallback to buffered I/O.
3386 if (!m_flags
&& !ret
)
3389 ext4_journal_stop(handle
);
3390 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3397 static int ext4_iomap_begin(struct inode
*inode
, loff_t offset
, loff_t length
,
3398 unsigned flags
, struct iomap
*iomap
, struct iomap
*srcmap
)
3401 struct ext4_map_blocks map
;
3402 u8 blkbits
= inode
->i_blkbits
;
3404 if ((offset
>> blkbits
) > EXT4_MAX_LOGICAL_BLOCK
)
3407 if (WARN_ON_ONCE(ext4_has_inline_data(inode
)))
3411 * Calculate the first and last logical blocks respectively.
3413 map
.m_lblk
= offset
>> blkbits
;
3414 map
.m_len
= min_t(loff_t
, (offset
+ length
- 1) >> blkbits
,
3415 EXT4_MAX_LOGICAL_BLOCK
) - map
.m_lblk
+ 1;
3417 if (flags
& IOMAP_WRITE
) {
3419 * We check here if the blocks are already allocated, then we
3420 * don't need to start a journal txn and we can directly return
3421 * the mapping information. This could boost performance
3422 * especially in multi-threaded overwrite requests.
3424 if (offset
+ length
<= i_size_read(inode
)) {
3425 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3426 if (ret
> 0 && (map
.m_flags
& EXT4_MAP_MAPPED
))
3429 ret
= ext4_iomap_alloc(inode
, &map
, flags
);
3431 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3437 ext4_set_iomap(inode
, iomap
, &map
, offset
, length
, flags
);
3442 static int ext4_iomap_overwrite_begin(struct inode
*inode
, loff_t offset
,
3443 loff_t length
, unsigned flags
, struct iomap
*iomap
,
3444 struct iomap
*srcmap
)
3449 * Even for writes we don't need to allocate blocks, so just pretend
3450 * we are reading to save overhead of starting a transaction.
3452 flags
&= ~IOMAP_WRITE
;
3453 ret
= ext4_iomap_begin(inode
, offset
, length
, flags
, iomap
, srcmap
);
3454 WARN_ON_ONCE(iomap
->type
!= IOMAP_MAPPED
);
3458 static int ext4_iomap_end(struct inode
*inode
, loff_t offset
, loff_t length
,
3459 ssize_t written
, unsigned flags
, struct iomap
*iomap
)
3462 * Check to see whether an error occurred while writing out the data to
3463 * the allocated blocks. If so, return the magic error code so that we
3464 * fallback to buffered I/O and attempt to complete the remainder of
3465 * the I/O. Any blocks that may have been allocated in preparation for
3466 * the direct I/O will be reused during buffered I/O.
3468 if (flags
& (IOMAP_WRITE
| IOMAP_DIRECT
) && written
== 0)
3474 const struct iomap_ops ext4_iomap_ops
= {
3475 .iomap_begin
= ext4_iomap_begin
,
3476 .iomap_end
= ext4_iomap_end
,
3479 const struct iomap_ops ext4_iomap_overwrite_ops
= {
3480 .iomap_begin
= ext4_iomap_overwrite_begin
,
3481 .iomap_end
= ext4_iomap_end
,
3484 static bool ext4_iomap_is_delalloc(struct inode
*inode
,
3485 struct ext4_map_blocks
*map
)
3487 struct extent_status es
;
3488 ext4_lblk_t offset
= 0, end
= map
->m_lblk
+ map
->m_len
- 1;
3490 ext4_es_find_extent_range(inode
, &ext4_es_is_delayed
,
3491 map
->m_lblk
, end
, &es
);
3493 if (!es
.es_len
|| es
.es_lblk
> end
)
3496 if (es
.es_lblk
> map
->m_lblk
) {
3497 map
->m_len
= es
.es_lblk
- map
->m_lblk
;
3501 offset
= map
->m_lblk
- es
.es_lblk
;
3502 map
->m_len
= es
.es_len
- offset
;
3507 static int ext4_iomap_begin_report(struct inode
*inode
, loff_t offset
,
3508 loff_t length
, unsigned int flags
,
3509 struct iomap
*iomap
, struct iomap
*srcmap
)
3512 bool delalloc
= false;
3513 struct ext4_map_blocks map
;
3514 u8 blkbits
= inode
->i_blkbits
;
3516 if ((offset
>> blkbits
) > EXT4_MAX_LOGICAL_BLOCK
)
3519 if (ext4_has_inline_data(inode
)) {
3520 ret
= ext4_inline_data_iomap(inode
, iomap
);
3521 if (ret
!= -EAGAIN
) {
3522 if (ret
== 0 && offset
>= iomap
->length
)
3529 * Calculate the first and last logical block respectively.
3531 map
.m_lblk
= offset
>> blkbits
;
3532 map
.m_len
= min_t(loff_t
, (offset
+ length
- 1) >> blkbits
,
3533 EXT4_MAX_LOGICAL_BLOCK
) - map
.m_lblk
+ 1;
3536 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3537 * So handle it here itself instead of querying ext4_map_blocks().
3538 * Since ext4_map_blocks() will warn about it and will return
3541 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
3542 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3544 if (offset
>= sbi
->s_bitmap_maxbytes
) {
3550 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3554 delalloc
= ext4_iomap_is_delalloc(inode
, &map
);
3557 ext4_set_iomap(inode
, iomap
, &map
, offset
, length
, flags
);
3558 if (delalloc
&& iomap
->type
== IOMAP_HOLE
)
3559 iomap
->type
= IOMAP_DELALLOC
;
3564 const struct iomap_ops ext4_iomap_report_ops
= {
3565 .iomap_begin
= ext4_iomap_begin_report
,
3569 * Whenever the page is being dirtied, corresponding buffers should already be
3570 * attached to the transaction (we take care of this in ext4_page_mkwrite() and
3571 * ext4_write_begin()). However we cannot move buffers to dirty transaction
3572 * lists here because ->set_page_dirty is called under VFS locks and the page
3573 * is not necessarily locked.
3575 * We cannot just dirty the page and leave attached buffers clean, because the
3576 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3577 * or jbddirty because all the journalling code will explode.
3579 * So what we do is to mark the page "pending dirty" and next time writepage
3580 * is called, propagate that into the buffers appropriately.
3582 static int ext4_journalled_set_page_dirty(struct page
*page
)
3584 WARN_ON_ONCE(!page_has_buffers(page
));
3585 SetPageChecked(page
);
3586 return __set_page_dirty_nobuffers(page
);
3589 static int ext4_set_page_dirty(struct page
*page
)
3591 WARN_ON_ONCE(!PageLocked(page
) && !PageDirty(page
));
3592 WARN_ON_ONCE(!page_has_buffers(page
));
3593 return __set_page_dirty_buffers(page
);
3596 static int ext4_iomap_swap_activate(struct swap_info_struct
*sis
,
3597 struct file
*file
, sector_t
*span
)
3599 return iomap_swapfile_activate(sis
, file
, span
,
3600 &ext4_iomap_report_ops
);
3603 static const struct address_space_operations ext4_aops
= {
3604 .readpage
= ext4_readpage
,
3605 .readahead
= ext4_readahead
,
3606 .writepage
= ext4_writepage
,
3607 .writepages
= ext4_writepages
,
3608 .write_begin
= ext4_write_begin
,
3609 .write_end
= ext4_write_end
,
3610 .set_page_dirty
= ext4_set_page_dirty
,
3612 .invalidatepage
= ext4_invalidatepage
,
3613 .releasepage
= ext4_releasepage
,
3614 .direct_IO
= noop_direct_IO
,
3615 .migratepage
= buffer_migrate_page
,
3616 .is_partially_uptodate
= block_is_partially_uptodate
,
3617 .error_remove_page
= generic_error_remove_page
,
3618 .swap_activate
= ext4_iomap_swap_activate
,
3621 static const struct address_space_operations ext4_journalled_aops
= {
3622 .readpage
= ext4_readpage
,
3623 .readahead
= ext4_readahead
,
3624 .writepage
= ext4_writepage
,
3625 .writepages
= ext4_writepages
,
3626 .write_begin
= ext4_write_begin
,
3627 .write_end
= ext4_journalled_write_end
,
3628 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3630 .invalidatepage
= ext4_journalled_invalidatepage
,
3631 .releasepage
= ext4_releasepage
,
3632 .direct_IO
= noop_direct_IO
,
3633 .is_partially_uptodate
= block_is_partially_uptodate
,
3634 .error_remove_page
= generic_error_remove_page
,
3635 .swap_activate
= ext4_iomap_swap_activate
,
3638 static const struct address_space_operations ext4_da_aops
= {
3639 .readpage
= ext4_readpage
,
3640 .readahead
= ext4_readahead
,
3641 .writepage
= ext4_writepage
,
3642 .writepages
= ext4_writepages
,
3643 .write_begin
= ext4_da_write_begin
,
3644 .write_end
= ext4_da_write_end
,
3645 .set_page_dirty
= ext4_set_page_dirty
,
3647 .invalidatepage
= ext4_invalidatepage
,
3648 .releasepage
= ext4_releasepage
,
3649 .direct_IO
= noop_direct_IO
,
3650 .migratepage
= buffer_migrate_page
,
3651 .is_partially_uptodate
= block_is_partially_uptodate
,
3652 .error_remove_page
= generic_error_remove_page
,
3653 .swap_activate
= ext4_iomap_swap_activate
,
3656 static const struct address_space_operations ext4_dax_aops
= {
3657 .writepages
= ext4_dax_writepages
,
3658 .direct_IO
= noop_direct_IO
,
3659 .set_page_dirty
= __set_page_dirty_no_writeback
,
3661 .invalidatepage
= noop_invalidatepage
,
3662 .swap_activate
= ext4_iomap_swap_activate
,
3665 void ext4_set_aops(struct inode
*inode
)
3667 switch (ext4_inode_journal_mode(inode
)) {
3668 case EXT4_INODE_ORDERED_DATA_MODE
:
3669 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3671 case EXT4_INODE_JOURNAL_DATA_MODE
:
3672 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3678 inode
->i_mapping
->a_ops
= &ext4_dax_aops
;
3679 else if (test_opt(inode
->i_sb
, DELALLOC
))
3680 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3682 inode
->i_mapping
->a_ops
= &ext4_aops
;
3685 static int __ext4_block_zero_page_range(handle_t
*handle
,
3686 struct address_space
*mapping
, loff_t from
, loff_t length
)
3688 ext4_fsblk_t index
= from
>> PAGE_SHIFT
;
3689 unsigned offset
= from
& (PAGE_SIZE
-1);
3690 unsigned blocksize
, pos
;
3692 struct inode
*inode
= mapping
->host
;
3693 struct buffer_head
*bh
;
3697 page
= find_or_create_page(mapping
, from
>> PAGE_SHIFT
,
3698 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
3702 blocksize
= inode
->i_sb
->s_blocksize
;
3704 iblock
= index
<< (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3706 if (!page_has_buffers(page
))
3707 create_empty_buffers(page
, blocksize
, 0);
3709 /* Find the buffer that contains "offset" */
3710 bh
= page_buffers(page
);
3712 while (offset
>= pos
) {
3713 bh
= bh
->b_this_page
;
3717 if (buffer_freed(bh
)) {
3718 BUFFER_TRACE(bh
, "freed: skip");
3721 if (!buffer_mapped(bh
)) {
3722 BUFFER_TRACE(bh
, "unmapped");
3723 ext4_get_block(inode
, iblock
, bh
, 0);
3724 /* unmapped? It's a hole - nothing to do */
3725 if (!buffer_mapped(bh
)) {
3726 BUFFER_TRACE(bh
, "still unmapped");
3731 /* Ok, it's mapped. Make sure it's up-to-date */
3732 if (PageUptodate(page
))
3733 set_buffer_uptodate(bh
);
3735 if (!buffer_uptodate(bh
)) {
3736 err
= ext4_read_bh_lock(bh
, 0, true);
3739 if (fscrypt_inode_uses_fs_layer_crypto(inode
)) {
3740 /* We expect the key to be set. */
3741 BUG_ON(!fscrypt_has_encryption_key(inode
));
3742 err
= fscrypt_decrypt_pagecache_blocks(page
, blocksize
,
3745 clear_buffer_uptodate(bh
);
3750 if (ext4_should_journal_data(inode
)) {
3751 BUFFER_TRACE(bh
, "get write access");
3752 err
= ext4_journal_get_write_access(handle
, inode
->i_sb
, bh
,
3757 zero_user(page
, offset
, length
);
3758 BUFFER_TRACE(bh
, "zeroed end of block");
3760 if (ext4_should_journal_data(inode
)) {
3761 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3764 mark_buffer_dirty(bh
);
3765 if (ext4_should_order_data(inode
))
3766 err
= ext4_jbd2_inode_add_write(handle
, inode
, from
,
3777 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3778 * starting from file offset 'from'. The range to be zero'd must
3779 * be contained with in one block. If the specified range exceeds
3780 * the end of the block it will be shortened to end of the block
3781 * that corresponds to 'from'
3783 static int ext4_block_zero_page_range(handle_t
*handle
,
3784 struct address_space
*mapping
, loff_t from
, loff_t length
)
3786 struct inode
*inode
= mapping
->host
;
3787 unsigned offset
= from
& (PAGE_SIZE
-1);
3788 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3789 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
3792 * correct length if it does not fall between
3793 * 'from' and the end of the block
3795 if (length
> max
|| length
< 0)
3798 if (IS_DAX(inode
)) {
3799 return dax_zero_range(inode
, from
, length
, NULL
,
3802 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3806 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3807 * up to the end of the block which corresponds to `from'.
3808 * This required during truncate. We need to physically zero the tail end
3809 * of that block so it doesn't yield old data if the file is later grown.
3811 static int ext4_block_truncate_page(handle_t
*handle
,
3812 struct address_space
*mapping
, loff_t from
)
3814 unsigned offset
= from
& (PAGE_SIZE
-1);
3817 struct inode
*inode
= mapping
->host
;
3819 /* If we are processing an encrypted inode during orphan list handling */
3820 if (IS_ENCRYPTED(inode
) && !fscrypt_has_encryption_key(inode
))
3823 blocksize
= inode
->i_sb
->s_blocksize
;
3824 length
= blocksize
- (offset
& (blocksize
- 1));
3826 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3829 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
3830 loff_t lstart
, loff_t length
)
3832 struct super_block
*sb
= inode
->i_sb
;
3833 struct address_space
*mapping
= inode
->i_mapping
;
3834 unsigned partial_start
, partial_end
;
3835 ext4_fsblk_t start
, end
;
3836 loff_t byte_end
= (lstart
+ length
- 1);
3839 partial_start
= lstart
& (sb
->s_blocksize
- 1);
3840 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
3842 start
= lstart
>> sb
->s_blocksize_bits
;
3843 end
= byte_end
>> sb
->s_blocksize_bits
;
3845 /* Handle partial zero within the single block */
3847 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
3848 err
= ext4_block_zero_page_range(handle
, mapping
,
3852 /* Handle partial zero out on the start of the range */
3853 if (partial_start
) {
3854 err
= ext4_block_zero_page_range(handle
, mapping
,
3855 lstart
, sb
->s_blocksize
);
3859 /* Handle partial zero out on the end of the range */
3860 if (partial_end
!= sb
->s_blocksize
- 1)
3861 err
= ext4_block_zero_page_range(handle
, mapping
,
3862 byte_end
- partial_end
,
3867 int ext4_can_truncate(struct inode
*inode
)
3869 if (S_ISREG(inode
->i_mode
))
3871 if (S_ISDIR(inode
->i_mode
))
3873 if (S_ISLNK(inode
->i_mode
))
3874 return !ext4_inode_is_fast_symlink(inode
);
3879 * We have to make sure i_disksize gets properly updated before we truncate
3880 * page cache due to hole punching or zero range. Otherwise i_disksize update
3881 * can get lost as it may have been postponed to submission of writeback but
3882 * that will never happen after we truncate page cache.
3884 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
3890 loff_t size
= i_size_read(inode
);
3892 WARN_ON(!inode_is_locked(inode
));
3893 if (offset
> size
|| offset
+ len
< size
)
3896 if (EXT4_I(inode
)->i_disksize
>= size
)
3899 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
3901 return PTR_ERR(handle
);
3902 ext4_update_i_disksize(inode
, size
);
3903 ret
= ext4_mark_inode_dirty(handle
, inode
);
3904 ext4_journal_stop(handle
);
3909 static void ext4_wait_dax_page(struct inode
*inode
)
3911 filemap_invalidate_unlock(inode
->i_mapping
);
3913 filemap_invalidate_lock(inode
->i_mapping
);
3916 int ext4_break_layouts(struct inode
*inode
)
3921 if (WARN_ON_ONCE(!rwsem_is_locked(&inode
->i_mapping
->invalidate_lock
)))
3925 page
= dax_layout_busy_page(inode
->i_mapping
);
3929 error
= ___wait_var_event(&page
->_refcount
,
3930 atomic_read(&page
->_refcount
) == 1,
3931 TASK_INTERRUPTIBLE
, 0, 0,
3932 ext4_wait_dax_page(inode
));
3933 } while (error
== 0);
3939 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3940 * associated with the given offset and length
3942 * @inode: File inode
3943 * @offset: The offset where the hole will begin
3944 * @len: The length of the hole
3946 * Returns: 0 on success or negative on failure
3949 int ext4_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
3951 struct inode
*inode
= file_inode(file
);
3952 struct super_block
*sb
= inode
->i_sb
;
3953 ext4_lblk_t first_block
, stop_block
;
3954 struct address_space
*mapping
= inode
->i_mapping
;
3955 loff_t first_block_offset
, last_block_offset
, max_length
;
3956 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3958 unsigned int credits
;
3959 int ret
= 0, ret2
= 0;
3961 trace_ext4_punch_hole(inode
, offset
, length
, 0);
3963 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
3964 if (ext4_has_inline_data(inode
)) {
3965 filemap_invalidate_lock(mapping
);
3966 ret
= ext4_convert_inline_data(inode
);
3967 filemap_invalidate_unlock(mapping
);
3973 * Write out all dirty pages to avoid race conditions
3974 * Then release them.
3976 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3977 ret
= filemap_write_and_wait_range(mapping
, offset
,
3978 offset
+ length
- 1);
3985 /* No need to punch hole beyond i_size */
3986 if (offset
>= inode
->i_size
)
3990 * If the hole extends beyond i_size, set the hole
3991 * to end after the page that contains i_size
3993 if (offset
+ length
> inode
->i_size
) {
3994 length
= inode
->i_size
+
3995 PAGE_SIZE
- (inode
->i_size
& (PAGE_SIZE
- 1)) -
4000 * For punch hole the length + offset needs to be within one block
4001 * before last range. Adjust the length if it goes beyond that limit.
4003 max_length
= sbi
->s_bitmap_maxbytes
- inode
->i_sb
->s_blocksize
;
4004 if (offset
+ length
> max_length
)
4005 length
= max_length
- offset
;
4007 if (offset
& (sb
->s_blocksize
- 1) ||
4008 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
4010 * Attach jinode to inode for jbd2 if we do any zeroing of
4013 ret
= ext4_inode_attach_jinode(inode
);
4019 /* Wait all existing dio workers, newcomers will block on i_rwsem */
4020 inode_dio_wait(inode
);
4022 ret
= file_modified(file
);
4027 * Prevent page faults from reinstantiating pages we have released from
4030 filemap_invalidate_lock(mapping
);
4032 ret
= ext4_break_layouts(inode
);
4036 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
4037 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
4039 /* Now release the pages and zero block aligned part of pages*/
4040 if (last_block_offset
> first_block_offset
) {
4041 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
4044 truncate_pagecache_range(inode
, first_block_offset
,
4048 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4049 credits
= ext4_writepage_trans_blocks(inode
);
4051 credits
= ext4_blocks_for_truncate(inode
);
4052 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4053 if (IS_ERR(handle
)) {
4054 ret
= PTR_ERR(handle
);
4055 ext4_std_error(sb
, ret
);
4059 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
4064 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4065 EXT4_BLOCK_SIZE_BITS(sb
);
4066 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4068 /* If there are blocks to remove, do it */
4069 if (stop_block
> first_block
) {
4071 down_write(&EXT4_I(inode
)->i_data_sem
);
4072 ext4_discard_preallocations(inode
, 0);
4074 ret
= ext4_es_remove_extent(inode
, first_block
,
4075 stop_block
- first_block
);
4077 up_write(&EXT4_I(inode
)->i_data_sem
);
4081 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4082 ret
= ext4_ext_remove_space(inode
, first_block
,
4085 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
4088 up_write(&EXT4_I(inode
)->i_data_sem
);
4090 ext4_fc_track_range(handle
, inode
, first_block
, stop_block
);
4092 ext4_handle_sync(handle
);
4094 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4095 ret2
= ext4_mark_inode_dirty(handle
, inode
);
4099 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4101 ext4_journal_stop(handle
);
4103 filemap_invalidate_unlock(mapping
);
4105 inode_unlock(inode
);
4109 int ext4_inode_attach_jinode(struct inode
*inode
)
4111 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4112 struct jbd2_inode
*jinode
;
4114 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
4117 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
4118 spin_lock(&inode
->i_lock
);
4121 spin_unlock(&inode
->i_lock
);
4124 ei
->jinode
= jinode
;
4125 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
4128 spin_unlock(&inode
->i_lock
);
4129 if (unlikely(jinode
!= NULL
))
4130 jbd2_free_inode(jinode
);
4137 * We block out ext4_get_block() block instantiations across the entire
4138 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4139 * simultaneously on behalf of the same inode.
4141 * As we work through the truncate and commit bits of it to the journal there
4142 * is one core, guiding principle: the file's tree must always be consistent on
4143 * disk. We must be able to restart the truncate after a crash.
4145 * The file's tree may be transiently inconsistent in memory (although it
4146 * probably isn't), but whenever we close off and commit a journal transaction,
4147 * the contents of (the filesystem + the journal) must be consistent and
4148 * restartable. It's pretty simple, really: bottom up, right to left (although
4149 * left-to-right works OK too).
4151 * Note that at recovery time, journal replay occurs *before* the restart of
4152 * truncate against the orphan inode list.
4154 * The committed inode has the new, desired i_size (which is the same as
4155 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4156 * that this inode's truncate did not complete and it will again call
4157 * ext4_truncate() to have another go. So there will be instantiated blocks
4158 * to the right of the truncation point in a crashed ext4 filesystem. But
4159 * that's fine - as long as they are linked from the inode, the post-crash
4160 * ext4_truncate() run will find them and release them.
4162 int ext4_truncate(struct inode
*inode
)
4164 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4165 unsigned int credits
;
4168 struct address_space
*mapping
= inode
->i_mapping
;
4171 * There is a possibility that we're either freeing the inode
4172 * or it's a completely new inode. In those cases we might not
4173 * have i_rwsem locked because it's not necessary.
4175 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
4176 WARN_ON(!inode_is_locked(inode
));
4177 trace_ext4_truncate_enter(inode
);
4179 if (!ext4_can_truncate(inode
))
4182 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4183 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4185 if (ext4_has_inline_data(inode
)) {
4188 err
= ext4_inline_data_truncate(inode
, &has_inline
);
4189 if (err
|| has_inline
)
4193 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4194 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
4195 if (ext4_inode_attach_jinode(inode
) < 0)
4199 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4200 credits
= ext4_writepage_trans_blocks(inode
);
4202 credits
= ext4_blocks_for_truncate(inode
);
4204 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4205 if (IS_ERR(handle
)) {
4206 err
= PTR_ERR(handle
);
4210 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
4211 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
4214 * We add the inode to the orphan list, so that if this
4215 * truncate spans multiple transactions, and we crash, we will
4216 * resume the truncate when the filesystem recovers. It also
4217 * marks the inode dirty, to catch the new size.
4219 * Implication: the file must always be in a sane, consistent
4220 * truncatable state while each transaction commits.
4222 err
= ext4_orphan_add(handle
, inode
);
4226 down_write(&EXT4_I(inode
)->i_data_sem
);
4228 ext4_discard_preallocations(inode
, 0);
4230 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4231 err
= ext4_ext_truncate(handle
, inode
);
4233 ext4_ind_truncate(handle
, inode
);
4235 up_write(&ei
->i_data_sem
);
4240 ext4_handle_sync(handle
);
4244 * If this was a simple ftruncate() and the file will remain alive,
4245 * then we need to clear up the orphan record which we created above.
4246 * However, if this was a real unlink then we were called by
4247 * ext4_evict_inode(), and we allow that function to clean up the
4248 * orphan info for us.
4251 ext4_orphan_del(handle
, inode
);
4253 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4254 err2
= ext4_mark_inode_dirty(handle
, inode
);
4255 if (unlikely(err2
&& !err
))
4257 ext4_journal_stop(handle
);
4260 trace_ext4_truncate_exit(inode
);
4264 static inline u64
ext4_inode_peek_iversion(const struct inode
*inode
)
4266 if (unlikely(EXT4_I(inode
)->i_flags
& EXT4_EA_INODE_FL
))
4267 return inode_peek_iversion_raw(inode
);
4269 return inode_peek_iversion(inode
);
4272 static int ext4_inode_blocks_set(struct ext4_inode
*raw_inode
,
4273 struct ext4_inode_info
*ei
)
4275 struct inode
*inode
= &(ei
->vfs_inode
);
4276 u64 i_blocks
= READ_ONCE(inode
->i_blocks
);
4277 struct super_block
*sb
= inode
->i_sb
;
4279 if (i_blocks
<= ~0U) {
4281 * i_blocks can be represented in a 32 bit variable
4282 * as multiple of 512 bytes
4284 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4285 raw_inode
->i_blocks_high
= 0;
4286 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4291 * This should never happen since sb->s_maxbytes should not have
4292 * allowed this, sb->s_maxbytes was set according to the huge_file
4293 * feature in ext4_fill_super().
4295 if (!ext4_has_feature_huge_file(sb
))
4296 return -EFSCORRUPTED
;
4298 if (i_blocks
<= 0xffffffffffffULL
) {
4300 * i_blocks can be represented in a 48 bit variable
4301 * as multiple of 512 bytes
4303 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4304 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4305 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4307 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4308 /* i_block is stored in file system block size */
4309 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4310 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4311 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4316 static int ext4_fill_raw_inode(struct inode
*inode
, struct ext4_inode
*raw_inode
)
4318 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4325 err
= ext4_inode_blocks_set(raw_inode
, ei
);
4327 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4328 i_uid
= i_uid_read(inode
);
4329 i_gid
= i_gid_read(inode
);
4330 i_projid
= from_kprojid(&init_user_ns
, ei
->i_projid
);
4331 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4332 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4333 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4335 * Fix up interoperability with old kernels. Otherwise,
4336 * old inodes get re-used with the upper 16 bits of the
4339 if (ei
->i_dtime
&& list_empty(&ei
->i_orphan
)) {
4340 raw_inode
->i_uid_high
= 0;
4341 raw_inode
->i_gid_high
= 0;
4343 raw_inode
->i_uid_high
=
4344 cpu_to_le16(high_16_bits(i_uid
));
4345 raw_inode
->i_gid_high
=
4346 cpu_to_le16(high_16_bits(i_gid
));
4349 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4350 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4351 raw_inode
->i_uid_high
= 0;
4352 raw_inode
->i_gid_high
= 0;
4354 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4356 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4357 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4358 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4359 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4361 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4362 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4363 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
4364 raw_inode
->i_file_acl_high
=
4365 cpu_to_le16(ei
->i_file_acl
>> 32);
4366 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4367 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4369 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4370 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4371 if (old_valid_dev(inode
->i_rdev
)) {
4372 raw_inode
->i_block
[0] =
4373 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4374 raw_inode
->i_block
[1] = 0;
4376 raw_inode
->i_block
[0] = 0;
4377 raw_inode
->i_block
[1] =
4378 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4379 raw_inode
->i_block
[2] = 0;
4381 } else if (!ext4_has_inline_data(inode
)) {
4382 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4383 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4386 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4387 u64 ivers
= ext4_inode_peek_iversion(inode
);
4389 raw_inode
->i_disk_version
= cpu_to_le32(ivers
);
4390 if (ei
->i_extra_isize
) {
4391 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4392 raw_inode
->i_version_hi
=
4393 cpu_to_le32(ivers
>> 32);
4394 raw_inode
->i_extra_isize
=
4395 cpu_to_le16(ei
->i_extra_isize
);
4399 if (i_projid
!= EXT4_DEF_PROJID
&&
4400 !ext4_has_feature_project(inode
->i_sb
))
4401 err
= err
?: -EFSCORRUPTED
;
4403 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4404 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4405 raw_inode
->i_projid
= cpu_to_le32(i_projid
);
4407 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4412 * ext4_get_inode_loc returns with an extra refcount against the inode's
4413 * underlying buffer_head on success. If we pass 'inode' and it does not
4414 * have in-inode xattr, we have all inode data in memory that is needed
4415 * to recreate the on-disk version of this inode.
4417 static int __ext4_get_inode_loc(struct super_block
*sb
, unsigned long ino
,
4418 struct inode
*inode
, struct ext4_iloc
*iloc
,
4419 ext4_fsblk_t
*ret_block
)
4421 struct ext4_group_desc
*gdp
;
4422 struct buffer_head
*bh
;
4424 struct blk_plug plug
;
4425 int inodes_per_block
, inode_offset
;
4428 if (ino
< EXT4_ROOT_INO
||
4429 ino
> le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
))
4430 return -EFSCORRUPTED
;
4432 iloc
->block_group
= (ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4433 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4438 * Figure out the offset within the block group inode table
4440 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4441 inode_offset
= ((ino
- 1) %
4442 EXT4_INODES_PER_GROUP(sb
));
4443 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4444 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4446 bh
= sb_getblk(sb
, block
);
4449 if (ext4_buffer_uptodate(bh
))
4453 if (ext4_buffer_uptodate(bh
)) {
4454 /* Someone brought it uptodate while we waited */
4460 * If we have all information of the inode in memory and this
4461 * is the only valid inode in the block, we need not read the
4464 if (inode
&& !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
)) {
4465 struct buffer_head
*bitmap_bh
;
4468 start
= inode_offset
& ~(inodes_per_block
- 1);
4470 /* Is the inode bitmap in cache? */
4471 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4472 if (unlikely(!bitmap_bh
))
4476 * If the inode bitmap isn't in cache then the
4477 * optimisation may end up performing two reads instead
4478 * of one, so skip it.
4480 if (!buffer_uptodate(bitmap_bh
)) {
4484 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4485 if (i
== inode_offset
)
4487 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4491 if (i
== start
+ inodes_per_block
) {
4492 struct ext4_inode
*raw_inode
=
4493 (struct ext4_inode
*) (bh
->b_data
+ iloc
->offset
);
4495 /* all other inodes are free, so skip I/O */
4496 memset(bh
->b_data
, 0, bh
->b_size
);
4497 if (!ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4498 ext4_fill_raw_inode(inode
, raw_inode
);
4499 set_buffer_uptodate(bh
);
4507 * If we need to do any I/O, try to pre-readahead extra
4508 * blocks from the inode table.
4510 blk_start_plug(&plug
);
4511 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4512 ext4_fsblk_t b
, end
, table
;
4514 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4516 table
= ext4_inode_table(sb
, gdp
);
4517 /* s_inode_readahead_blks is always a power of 2 */
4518 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4522 num
= EXT4_INODES_PER_GROUP(sb
);
4523 if (ext4_has_group_desc_csum(sb
))
4524 num
-= ext4_itable_unused_count(sb
, gdp
);
4525 table
+= num
/ inodes_per_block
;
4529 ext4_sb_breadahead_unmovable(sb
, b
++);
4533 * There are other valid inodes in the buffer, this inode
4534 * has in-inode xattrs, or we don't have this inode in memory.
4535 * Read the block from disk.
4537 trace_ext4_load_inode(sb
, ino
);
4538 ext4_read_bh_nowait(bh
, REQ_META
| REQ_PRIO
, NULL
);
4539 blk_finish_plug(&plug
);
4541 ext4_simulate_fail_bh(sb
, bh
, EXT4_SIM_INODE_EIO
);
4542 if (!buffer_uptodate(bh
)) {
4553 static int __ext4_get_inode_loc_noinmem(struct inode
*inode
,
4554 struct ext4_iloc
*iloc
)
4556 ext4_fsblk_t err_blk
= 0;
4559 ret
= __ext4_get_inode_loc(inode
->i_sb
, inode
->i_ino
, NULL
, iloc
,
4563 ext4_error_inode_block(inode
, err_blk
, EIO
,
4564 "unable to read itable block");
4569 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4571 ext4_fsblk_t err_blk
= 0;
4574 ret
= __ext4_get_inode_loc(inode
->i_sb
, inode
->i_ino
, inode
, iloc
,
4578 ext4_error_inode_block(inode
, err_blk
, EIO
,
4579 "unable to read itable block");
4585 int ext4_get_fc_inode_loc(struct super_block
*sb
, unsigned long ino
,
4586 struct ext4_iloc
*iloc
)
4588 return __ext4_get_inode_loc(sb
, ino
, NULL
, iloc
, NULL
);
4591 static bool ext4_should_enable_dax(struct inode
*inode
)
4593 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4595 if (test_opt2(inode
->i_sb
, DAX_NEVER
))
4597 if (!S_ISREG(inode
->i_mode
))
4599 if (ext4_should_journal_data(inode
))
4601 if (ext4_has_inline_data(inode
))
4603 if (ext4_test_inode_flag(inode
, EXT4_INODE_ENCRYPT
))
4605 if (ext4_test_inode_flag(inode
, EXT4_INODE_VERITY
))
4607 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX
, &sbi
->s_ext4_flags
))
4609 if (test_opt(inode
->i_sb
, DAX_ALWAYS
))
4612 return ext4_test_inode_flag(inode
, EXT4_INODE_DAX
);
4615 void ext4_set_inode_flags(struct inode
*inode
, bool init
)
4617 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4618 unsigned int new_fl
= 0;
4620 WARN_ON_ONCE(IS_DAX(inode
) && init
);
4622 if (flags
& EXT4_SYNC_FL
)
4624 if (flags
& EXT4_APPEND_FL
)
4626 if (flags
& EXT4_IMMUTABLE_FL
)
4627 new_fl
|= S_IMMUTABLE
;
4628 if (flags
& EXT4_NOATIME_FL
)
4629 new_fl
|= S_NOATIME
;
4630 if (flags
& EXT4_DIRSYNC_FL
)
4631 new_fl
|= S_DIRSYNC
;
4633 /* Because of the way inode_set_flags() works we must preserve S_DAX
4634 * here if already set. */
4635 new_fl
|= (inode
->i_flags
& S_DAX
);
4636 if (init
&& ext4_should_enable_dax(inode
))
4639 if (flags
& EXT4_ENCRYPT_FL
)
4640 new_fl
|= S_ENCRYPTED
;
4641 if (flags
& EXT4_CASEFOLD_FL
)
4642 new_fl
|= S_CASEFOLD
;
4643 if (flags
& EXT4_VERITY_FL
)
4645 inode_set_flags(inode
, new_fl
,
4646 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
|
4647 S_ENCRYPTED
|S_CASEFOLD
|S_VERITY
);
4650 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4651 struct ext4_inode_info
*ei
)
4654 struct inode
*inode
= &(ei
->vfs_inode
);
4655 struct super_block
*sb
= inode
->i_sb
;
4657 if (ext4_has_feature_huge_file(sb
)) {
4658 /* we are using combined 48 bit field */
4659 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4660 le32_to_cpu(raw_inode
->i_blocks_lo
);
4661 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4662 /* i_blocks represent file system block size */
4663 return i_blocks
<< (inode
->i_blkbits
- 9);
4668 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4672 static inline int ext4_iget_extra_inode(struct inode
*inode
,
4673 struct ext4_inode
*raw_inode
,
4674 struct ext4_inode_info
*ei
)
4676 __le32
*magic
= (void *)raw_inode
+
4677 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4679 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
+ sizeof(__le32
) <=
4680 EXT4_INODE_SIZE(inode
->i_sb
) &&
4681 *magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4682 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4683 return ext4_find_inline_data_nolock(inode
);
4685 EXT4_I(inode
)->i_inline_off
= 0;
4689 int ext4_get_projid(struct inode
*inode
, kprojid_t
*projid
)
4691 if (!ext4_has_feature_project(inode
->i_sb
))
4693 *projid
= EXT4_I(inode
)->i_projid
;
4698 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4699 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4702 static inline void ext4_inode_set_iversion_queried(struct inode
*inode
, u64 val
)
4704 if (unlikely(EXT4_I(inode
)->i_flags
& EXT4_EA_INODE_FL
))
4705 inode_set_iversion_raw(inode
, val
);
4707 inode_set_iversion_queried(inode
, val
);
4710 struct inode
*__ext4_iget(struct super_block
*sb
, unsigned long ino
,
4711 ext4_iget_flags flags
, const char *function
,
4714 struct ext4_iloc iloc
;
4715 struct ext4_inode
*raw_inode
;
4716 struct ext4_inode_info
*ei
;
4717 struct ext4_super_block
*es
= EXT4_SB(sb
)->s_es
;
4718 struct inode
*inode
;
4719 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4727 if ((!(flags
& EXT4_IGET_SPECIAL
) &&
4728 ((ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
) ||
4729 ino
== le32_to_cpu(es
->s_usr_quota_inum
) ||
4730 ino
== le32_to_cpu(es
->s_grp_quota_inum
) ||
4731 ino
== le32_to_cpu(es
->s_prj_quota_inum
) ||
4732 ino
== le32_to_cpu(es
->s_orphan_file_inum
))) ||
4733 (ino
< EXT4_ROOT_INO
) ||
4734 (ino
> le32_to_cpu(es
->s_inodes_count
))) {
4735 if (flags
& EXT4_IGET_HANDLE
)
4736 return ERR_PTR(-ESTALE
);
4737 __ext4_error(sb
, function
, line
, false, EFSCORRUPTED
, 0,
4738 "inode #%lu: comm %s: iget: illegal inode #",
4739 ino
, current
->comm
);
4740 return ERR_PTR(-EFSCORRUPTED
);
4743 inode
= iget_locked(sb
, ino
);
4745 return ERR_PTR(-ENOMEM
);
4746 if (!(inode
->i_state
& I_NEW
))
4752 ret
= __ext4_get_inode_loc_noinmem(inode
, &iloc
);
4755 raw_inode
= ext4_raw_inode(&iloc
);
4757 if ((ino
== EXT4_ROOT_INO
) && (raw_inode
->i_links_count
== 0)) {
4758 ext4_error_inode(inode
, function
, line
, 0,
4759 "iget: root inode unallocated");
4760 ret
= -EFSCORRUPTED
;
4764 if ((flags
& EXT4_IGET_HANDLE
) &&
4765 (raw_inode
->i_links_count
== 0) && (raw_inode
->i_mode
== 0)) {
4770 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4771 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4772 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4773 EXT4_INODE_SIZE(inode
->i_sb
) ||
4774 (ei
->i_extra_isize
& 3)) {
4775 ext4_error_inode(inode
, function
, line
, 0,
4776 "iget: bad extra_isize %u "
4779 EXT4_INODE_SIZE(inode
->i_sb
));
4780 ret
= -EFSCORRUPTED
;
4784 ei
->i_extra_isize
= 0;
4786 /* Precompute checksum seed for inode metadata */
4787 if (ext4_has_metadata_csum(sb
)) {
4788 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4790 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4791 __le32 gen
= raw_inode
->i_generation
;
4792 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4794 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4798 if ((!ext4_inode_csum_verify(inode
, raw_inode
, ei
) ||
4799 ext4_simulate_fail(sb
, EXT4_SIM_INODE_CRC
)) &&
4800 (!(EXT4_SB(sb
)->s_mount_state
& EXT4_FC_REPLAY
))) {
4801 ext4_error_inode_err(inode
, function
, line
, 0,
4802 EFSBADCRC
, "iget: checksum invalid");
4807 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4808 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4809 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4810 if (ext4_has_feature_project(sb
) &&
4811 EXT4_INODE_SIZE(sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4812 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4813 i_projid
= (projid_t
)le32_to_cpu(raw_inode
->i_projid
);
4815 i_projid
= EXT4_DEF_PROJID
;
4817 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4818 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4819 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4821 i_uid_write(inode
, i_uid
);
4822 i_gid_write(inode
, i_gid
);
4823 ei
->i_projid
= make_kprojid(&init_user_ns
, i_projid
);
4824 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4826 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4827 ei
->i_inline_off
= 0;
4828 ei
->i_dir_start_lookup
= 0;
4829 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4830 /* We now have enough fields to check if the inode was active or not.
4831 * This is needed because nfsd might try to access dead inodes
4832 * the test is that same one that e2fsck uses
4833 * NeilBrown 1999oct15
4835 if (inode
->i_nlink
== 0) {
4836 if ((inode
->i_mode
== 0 ||
4837 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4838 ino
!= EXT4_BOOT_LOADER_INO
) {
4839 /* this inode is deleted */
4843 /* The only unlinked inodes we let through here have
4844 * valid i_mode and are being read by the orphan
4845 * recovery code: that's fine, we're about to complete
4846 * the process of deleting those.
4847 * OR it is the EXT4_BOOT_LOADER_INO which is
4848 * not initialized on a new filesystem. */
4850 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4851 ext4_set_inode_flags(inode
, true);
4852 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4853 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4854 if (ext4_has_feature_64bit(sb
))
4856 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4857 inode
->i_size
= ext4_isize(sb
, raw_inode
);
4858 if ((size
= i_size_read(inode
)) < 0) {
4859 ext4_error_inode(inode
, function
, line
, 0,
4860 "iget: bad i_size value: %lld", size
);
4861 ret
= -EFSCORRUPTED
;
4865 * If dir_index is not enabled but there's dir with INDEX flag set,
4866 * we'd normally treat htree data as empty space. But with metadata
4867 * checksumming that corrupts checksums so forbid that.
4869 if (!ext4_has_feature_dir_index(sb
) && ext4_has_metadata_csum(sb
) &&
4870 ext4_test_inode_flag(inode
, EXT4_INODE_INDEX
)) {
4871 ext4_error_inode(inode
, function
, line
, 0,
4872 "iget: Dir with htree data on filesystem without dir_index feature.");
4873 ret
= -EFSCORRUPTED
;
4876 ei
->i_disksize
= inode
->i_size
;
4878 ei
->i_reserved_quota
= 0;
4880 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4881 ei
->i_block_group
= iloc
.block_group
;
4882 ei
->i_last_alloc_group
= ~0;
4884 * NOTE! The in-memory inode i_data array is in little-endian order
4885 * even on big-endian machines: we do NOT byteswap the block numbers!
4887 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4888 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4889 INIT_LIST_HEAD(&ei
->i_orphan
);
4890 ext4_fc_init_inode(&ei
->vfs_inode
);
4893 * Set transaction id's of transactions that have to be committed
4894 * to finish f[data]sync. We set them to currently running transaction
4895 * as we cannot be sure that the inode or some of its metadata isn't
4896 * part of the transaction - the inode could have been reclaimed and
4897 * now it is reread from disk.
4900 transaction_t
*transaction
;
4903 read_lock(&journal
->j_state_lock
);
4904 if (journal
->j_running_transaction
)
4905 transaction
= journal
->j_running_transaction
;
4907 transaction
= journal
->j_committing_transaction
;
4909 tid
= transaction
->t_tid
;
4911 tid
= journal
->j_commit_sequence
;
4912 read_unlock(&journal
->j_state_lock
);
4913 ei
->i_sync_tid
= tid
;
4914 ei
->i_datasync_tid
= tid
;
4917 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4918 if (ei
->i_extra_isize
== 0) {
4919 /* The extra space is currently unused. Use it. */
4920 BUILD_BUG_ON(sizeof(struct ext4_inode
) & 3);
4921 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4922 EXT4_GOOD_OLD_INODE_SIZE
;
4924 ret
= ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4930 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4931 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4932 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4933 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4935 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4936 u64 ivers
= le32_to_cpu(raw_inode
->i_disk_version
);
4938 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4939 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4941 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4943 ext4_inode_set_iversion_queried(inode
, ivers
);
4947 if (ei
->i_file_acl
&&
4948 !ext4_inode_block_valid(inode
, ei
->i_file_acl
, 1)) {
4949 ext4_error_inode(inode
, function
, line
, 0,
4950 "iget: bad extended attribute block %llu",
4952 ret
= -EFSCORRUPTED
;
4954 } else if (!ext4_has_inline_data(inode
)) {
4955 /* validate the block references in the inode */
4956 if (!(EXT4_SB(sb
)->s_mount_state
& EXT4_FC_REPLAY
) &&
4957 (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4958 (S_ISLNK(inode
->i_mode
) &&
4959 !ext4_inode_is_fast_symlink(inode
)))) {
4960 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4961 ret
= ext4_ext_check_inode(inode
);
4963 ret
= ext4_ind_check_inode(inode
);
4969 if (S_ISREG(inode
->i_mode
)) {
4970 inode
->i_op
= &ext4_file_inode_operations
;
4971 inode
->i_fop
= &ext4_file_operations
;
4972 ext4_set_aops(inode
);
4973 } else if (S_ISDIR(inode
->i_mode
)) {
4974 inode
->i_op
= &ext4_dir_inode_operations
;
4975 inode
->i_fop
= &ext4_dir_operations
;
4976 } else if (S_ISLNK(inode
->i_mode
)) {
4977 /* VFS does not allow setting these so must be corruption */
4978 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
)) {
4979 ext4_error_inode(inode
, function
, line
, 0,
4980 "iget: immutable or append flags "
4981 "not allowed on symlinks");
4982 ret
= -EFSCORRUPTED
;
4985 if (IS_ENCRYPTED(inode
)) {
4986 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
4987 ext4_set_aops(inode
);
4988 } else if (ext4_inode_is_fast_symlink(inode
)) {
4989 inode
->i_link
= (char *)ei
->i_data
;
4990 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4991 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4992 sizeof(ei
->i_data
) - 1);
4994 inode
->i_op
= &ext4_symlink_inode_operations
;
4995 ext4_set_aops(inode
);
4997 inode_nohighmem(inode
);
4998 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4999 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
5000 inode
->i_op
= &ext4_special_inode_operations
;
5001 if (raw_inode
->i_block
[0])
5002 init_special_inode(inode
, inode
->i_mode
,
5003 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
5005 init_special_inode(inode
, inode
->i_mode
,
5006 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
5007 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
5008 make_bad_inode(inode
);
5010 ret
= -EFSCORRUPTED
;
5011 ext4_error_inode(inode
, function
, line
, 0,
5012 "iget: bogus i_mode (%o)", inode
->i_mode
);
5015 if (IS_CASEFOLDED(inode
) && !ext4_has_feature_casefold(inode
->i_sb
))
5016 ext4_error_inode(inode
, function
, line
, 0,
5017 "casefold flag without casefold feature");
5020 unlock_new_inode(inode
);
5026 return ERR_PTR(ret
);
5029 static void __ext4_update_other_inode_time(struct super_block
*sb
,
5030 unsigned long orig_ino
,
5032 struct ext4_inode
*raw_inode
)
5034 struct inode
*inode
;
5036 inode
= find_inode_by_ino_rcu(sb
, ino
);
5040 if (!inode_is_dirtytime_only(inode
))
5043 spin_lock(&inode
->i_lock
);
5044 if (inode_is_dirtytime_only(inode
)) {
5045 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5047 inode
->i_state
&= ~I_DIRTY_TIME
;
5048 spin_unlock(&inode
->i_lock
);
5050 spin_lock(&ei
->i_raw_lock
);
5051 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5052 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5053 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5054 ext4_inode_csum_set(inode
, raw_inode
, ei
);
5055 spin_unlock(&ei
->i_raw_lock
);
5056 trace_ext4_other_inode_update_time(inode
, orig_ino
);
5059 spin_unlock(&inode
->i_lock
);
5063 * Opportunistically update the other time fields for other inodes in
5064 * the same inode table block.
5066 static void ext4_update_other_inodes_time(struct super_block
*sb
,
5067 unsigned long orig_ino
, char *buf
)
5070 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
5071 int inode_size
= EXT4_INODE_SIZE(sb
);
5074 * Calculate the first inode in the inode table block. Inode
5075 * numbers are one-based. That is, the first inode in a block
5076 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5078 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
5080 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
5081 if (ino
== orig_ino
)
5083 __ext4_update_other_inode_time(sb
, orig_ino
, ino
,
5084 (struct ext4_inode
*)buf
);
5090 * Post the struct inode info into an on-disk inode location in the
5091 * buffer-cache. This gobbles the caller's reference to the
5092 * buffer_head in the inode location struct.
5094 * The caller must have write access to iloc->bh.
5096 static int ext4_do_update_inode(handle_t
*handle
,
5097 struct inode
*inode
,
5098 struct ext4_iloc
*iloc
)
5100 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
5101 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5102 struct buffer_head
*bh
= iloc
->bh
;
5103 struct super_block
*sb
= inode
->i_sb
;
5105 int need_datasync
= 0, set_large_file
= 0;
5107 spin_lock(&ei
->i_raw_lock
);
5110 * For fields not tracked in the in-memory inode, initialise them
5111 * to zero for new inodes.
5113 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
5114 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5116 if (READ_ONCE(ei
->i_disksize
) != ext4_isize(inode
->i_sb
, raw_inode
))
5118 if (ei
->i_disksize
> 0x7fffffffULL
) {
5119 if (!ext4_has_feature_large_file(sb
) ||
5120 EXT4_SB(sb
)->s_es
->s_rev_level
== cpu_to_le32(EXT4_GOOD_OLD_REV
))
5124 err
= ext4_fill_raw_inode(inode
, raw_inode
);
5125 spin_unlock(&ei
->i_raw_lock
);
5127 EXT4_ERROR_INODE(inode
, "corrupted inode contents");
5131 if (inode
->i_sb
->s_flags
& SB_LAZYTIME
)
5132 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
5135 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5136 err
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5139 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5140 if (set_large_file
) {
5141 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
5142 err
= ext4_journal_get_write_access(handle
, sb
,
5147 lock_buffer(EXT4_SB(sb
)->s_sbh
);
5148 ext4_set_feature_large_file(sb
);
5149 ext4_superblock_csum_set(sb
);
5150 unlock_buffer(EXT4_SB(sb
)->s_sbh
);
5151 ext4_handle_sync(handle
);
5152 err
= ext4_handle_dirty_metadata(handle
, NULL
,
5153 EXT4_SB(sb
)->s_sbh
);
5155 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
5157 ext4_std_error(inode
->i_sb
, err
);
5164 * ext4_write_inode()
5166 * We are called from a few places:
5168 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5169 * Here, there will be no transaction running. We wait for any running
5170 * transaction to commit.
5172 * - Within flush work (sys_sync(), kupdate and such).
5173 * We wait on commit, if told to.
5175 * - Within iput_final() -> write_inode_now()
5176 * We wait on commit, if told to.
5178 * In all cases it is actually safe for us to return without doing anything,
5179 * because the inode has been copied into a raw inode buffer in
5180 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5183 * Note that we are absolutely dependent upon all inode dirtiers doing the
5184 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5185 * which we are interested.
5187 * It would be a bug for them to not do this. The code:
5189 * mark_inode_dirty(inode)
5191 * inode->i_size = expr;
5193 * is in error because write_inode() could occur while `stuff()' is running,
5194 * and the new i_size will be lost. Plus the inode will no longer be on the
5195 * superblock's dirty inode list.
5197 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5201 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
) ||
5202 sb_rdonly(inode
->i_sb
))
5205 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5208 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5209 if (ext4_journal_current_handle()) {
5210 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5216 * No need to force transaction in WB_SYNC_NONE mode. Also
5217 * ext4_sync_fs() will force the commit after everything is
5220 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
5223 err
= ext4_fc_commit(EXT4_SB(inode
->i_sb
)->s_journal
,
5224 EXT4_I(inode
)->i_sync_tid
);
5226 struct ext4_iloc iloc
;
5228 err
= __ext4_get_inode_loc_noinmem(inode
, &iloc
);
5232 * sync(2) will flush the whole buffer cache. No need to do
5233 * it here separately for each inode.
5235 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
5236 sync_dirty_buffer(iloc
.bh
);
5237 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5238 ext4_error_inode_block(inode
, iloc
.bh
->b_blocknr
, EIO
,
5239 "IO error syncing inode");
5248 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5249 * buffers that are attached to a page stradding i_size and are undergoing
5250 * commit. In that case we have to wait for commit to finish and try again.
5252 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
5256 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
5257 tid_t commit_tid
= 0;
5260 offset
= inode
->i_size
& (PAGE_SIZE
- 1);
5262 * If the page is fully truncated, we don't need to wait for any commit
5263 * (and we even should not as __ext4_journalled_invalidatepage() may
5264 * strip all buffers from the page but keep the page dirty which can then
5265 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5266 * buffers). Also we don't need to wait for any commit if all buffers in
5267 * the page remain valid. This is most beneficial for the common case of
5268 * blocksize == PAGESIZE.
5270 if (!offset
|| offset
> (PAGE_SIZE
- i_blocksize(inode
)))
5273 page
= find_lock_page(inode
->i_mapping
,
5274 inode
->i_size
>> PAGE_SHIFT
);
5277 ret
= __ext4_journalled_invalidatepage(page
, offset
,
5278 PAGE_SIZE
- offset
);
5284 read_lock(&journal
->j_state_lock
);
5285 if (journal
->j_committing_transaction
)
5286 commit_tid
= journal
->j_committing_transaction
->t_tid
;
5287 read_unlock(&journal
->j_state_lock
);
5289 jbd2_log_wait_commit(journal
, commit_tid
);
5296 * Called from notify_change.
5298 * We want to trap VFS attempts to truncate the file as soon as
5299 * possible. In particular, we want to make sure that when the VFS
5300 * shrinks i_size, we put the inode on the orphan list and modify
5301 * i_disksize immediately, so that during the subsequent flushing of
5302 * dirty pages and freeing of disk blocks, we can guarantee that any
5303 * commit will leave the blocks being flushed in an unused state on
5304 * disk. (On recovery, the inode will get truncated and the blocks will
5305 * be freed, so we have a strong guarantee that no future commit will
5306 * leave these blocks visible to the user.)
5308 * Another thing we have to assure is that if we are in ordered mode
5309 * and inode is still attached to the committing transaction, we must
5310 * we start writeout of all the dirty pages which are being truncated.
5311 * This way we are sure that all the data written in the previous
5312 * transaction are already on disk (truncate waits for pages under
5315 * Called with inode->i_rwsem down.
5317 int ext4_setattr(struct user_namespace
*mnt_userns
, struct dentry
*dentry
,
5320 struct inode
*inode
= d_inode(dentry
);
5323 const unsigned int ia_valid
= attr
->ia_valid
;
5325 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5328 if (unlikely(IS_IMMUTABLE(inode
)))
5331 if (unlikely(IS_APPEND(inode
) &&
5332 (ia_valid
& (ATTR_MODE
| ATTR_UID
|
5333 ATTR_GID
| ATTR_TIMES_SET
))))
5336 error
= setattr_prepare(mnt_userns
, dentry
, attr
);
5340 error
= fscrypt_prepare_setattr(dentry
, attr
);
5344 error
= fsverity_prepare_setattr(dentry
, attr
);
5348 if (is_quota_modification(inode
, attr
)) {
5349 error
= dquot_initialize(inode
);
5354 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
5355 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
5358 /* (user+group)*(old+new) structure, inode write (sb,
5359 * inode block, ? - but truncate inode update has it) */
5360 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
5361 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
5362 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
5363 if (IS_ERR(handle
)) {
5364 error
= PTR_ERR(handle
);
5368 /* dquot_transfer() calls back ext4_get_inode_usage() which
5369 * counts xattr inode references.
5371 down_read(&EXT4_I(inode
)->xattr_sem
);
5372 error
= dquot_transfer(inode
, attr
);
5373 up_read(&EXT4_I(inode
)->xattr_sem
);
5376 ext4_journal_stop(handle
);
5379 /* Update corresponding info in inode so that everything is in
5380 * one transaction */
5381 if (attr
->ia_valid
& ATTR_UID
)
5382 inode
->i_uid
= attr
->ia_uid
;
5383 if (attr
->ia_valid
& ATTR_GID
)
5384 inode
->i_gid
= attr
->ia_gid
;
5385 error
= ext4_mark_inode_dirty(handle
, inode
);
5386 ext4_journal_stop(handle
);
5387 if (unlikely(error
)) {
5392 if (attr
->ia_valid
& ATTR_SIZE
) {
5394 loff_t oldsize
= inode
->i_size
;
5395 int shrink
= (attr
->ia_size
< inode
->i_size
);
5397 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5398 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5400 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
) {
5404 if (!S_ISREG(inode
->i_mode
)) {
5408 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
5409 inode_inc_iversion(inode
);
5412 if (ext4_should_order_data(inode
)) {
5413 error
= ext4_begin_ordered_truncate(inode
,
5419 * Blocks are going to be removed from the inode. Wait
5420 * for dio in flight.
5422 inode_dio_wait(inode
);
5425 filemap_invalidate_lock(inode
->i_mapping
);
5427 rc
= ext4_break_layouts(inode
);
5429 filemap_invalidate_unlock(inode
->i_mapping
);
5433 if (attr
->ia_size
!= inode
->i_size
) {
5434 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
5435 if (IS_ERR(handle
)) {
5436 error
= PTR_ERR(handle
);
5439 if (ext4_handle_valid(handle
) && shrink
) {
5440 error
= ext4_orphan_add(handle
, inode
);
5444 * Update c/mtime on truncate up, ext4_truncate() will
5445 * update c/mtime in shrink case below
5448 inode
->i_mtime
= current_time(inode
);
5449 inode
->i_ctime
= inode
->i_mtime
;
5453 ext4_fc_track_range(handle
, inode
,
5454 (attr
->ia_size
> 0 ? attr
->ia_size
- 1 : 0) >>
5455 inode
->i_sb
->s_blocksize_bits
,
5456 EXT_MAX_BLOCKS
- 1);
5458 ext4_fc_track_range(
5460 (oldsize
> 0 ? oldsize
- 1 : oldsize
) >>
5461 inode
->i_sb
->s_blocksize_bits
,
5462 (attr
->ia_size
> 0 ? attr
->ia_size
- 1 : 0) >>
5463 inode
->i_sb
->s_blocksize_bits
);
5465 down_write(&EXT4_I(inode
)->i_data_sem
);
5466 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5467 rc
= ext4_mark_inode_dirty(handle
, inode
);
5471 * We have to update i_size under i_data_sem together
5472 * with i_disksize to avoid races with writeback code
5473 * running ext4_wb_update_i_disksize().
5476 i_size_write(inode
, attr
->ia_size
);
5477 up_write(&EXT4_I(inode
)->i_data_sem
);
5478 ext4_journal_stop(handle
);
5482 pagecache_isize_extended(inode
, oldsize
,
5484 } else if (ext4_should_journal_data(inode
)) {
5485 ext4_wait_for_tail_page_commit(inode
);
5490 * Truncate pagecache after we've waited for commit
5491 * in data=journal mode to make pages freeable.
5493 truncate_pagecache(inode
, inode
->i_size
);
5495 * Call ext4_truncate() even if i_size didn't change to
5496 * truncate possible preallocated blocks.
5498 if (attr
->ia_size
<= oldsize
) {
5499 rc
= ext4_truncate(inode
);
5504 filemap_invalidate_unlock(inode
->i_mapping
);
5508 setattr_copy(mnt_userns
, inode
, attr
);
5509 mark_inode_dirty(inode
);
5513 * If the call to ext4_truncate failed to get a transaction handle at
5514 * all, we need to clean up the in-core orphan list manually.
5516 if (orphan
&& inode
->i_nlink
)
5517 ext4_orphan_del(NULL
, inode
);
5519 if (!error
&& (ia_valid
& ATTR_MODE
))
5520 rc
= posix_acl_chmod(mnt_userns
, inode
, inode
->i_mode
);
5524 ext4_std_error(inode
->i_sb
, error
);
5530 int ext4_getattr(struct user_namespace
*mnt_userns
, const struct path
*path
,
5531 struct kstat
*stat
, u32 request_mask
, unsigned int query_flags
)
5533 struct inode
*inode
= d_inode(path
->dentry
);
5534 struct ext4_inode
*raw_inode
;
5535 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5538 if ((request_mask
& STATX_BTIME
) &&
5539 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_crtime
)) {
5540 stat
->result_mask
|= STATX_BTIME
;
5541 stat
->btime
.tv_sec
= ei
->i_crtime
.tv_sec
;
5542 stat
->btime
.tv_nsec
= ei
->i_crtime
.tv_nsec
;
5545 flags
= ei
->i_flags
& EXT4_FL_USER_VISIBLE
;
5546 if (flags
& EXT4_APPEND_FL
)
5547 stat
->attributes
|= STATX_ATTR_APPEND
;
5548 if (flags
& EXT4_COMPR_FL
)
5549 stat
->attributes
|= STATX_ATTR_COMPRESSED
;
5550 if (flags
& EXT4_ENCRYPT_FL
)
5551 stat
->attributes
|= STATX_ATTR_ENCRYPTED
;
5552 if (flags
& EXT4_IMMUTABLE_FL
)
5553 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
5554 if (flags
& EXT4_NODUMP_FL
)
5555 stat
->attributes
|= STATX_ATTR_NODUMP
;
5556 if (flags
& EXT4_VERITY_FL
)
5557 stat
->attributes
|= STATX_ATTR_VERITY
;
5559 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
5560 STATX_ATTR_COMPRESSED
|
5561 STATX_ATTR_ENCRYPTED
|
5562 STATX_ATTR_IMMUTABLE
|
5566 generic_fillattr(mnt_userns
, inode
, stat
);
5570 int ext4_file_getattr(struct user_namespace
*mnt_userns
,
5571 const struct path
*path
, struct kstat
*stat
,
5572 u32 request_mask
, unsigned int query_flags
)
5574 struct inode
*inode
= d_inode(path
->dentry
);
5575 u64 delalloc_blocks
;
5577 ext4_getattr(mnt_userns
, path
, stat
, request_mask
, query_flags
);
5580 * If there is inline data in the inode, the inode will normally not
5581 * have data blocks allocated (it may have an external xattr block).
5582 * Report at least one sector for such files, so tools like tar, rsync,
5583 * others don't incorrectly think the file is completely sparse.
5585 if (unlikely(ext4_has_inline_data(inode
)))
5586 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5589 * We can't update i_blocks if the block allocation is delayed
5590 * otherwise in the case of system crash before the real block
5591 * allocation is done, we will have i_blocks inconsistent with
5592 * on-disk file blocks.
5593 * We always keep i_blocks updated together with real
5594 * allocation. But to not confuse with user, stat
5595 * will return the blocks that include the delayed allocation
5596 * blocks for this file.
5598 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5599 EXT4_I(inode
)->i_reserved_data_blocks
);
5600 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5604 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5607 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5608 return ext4_ind_trans_blocks(inode
, lblocks
);
5609 return ext4_ext_index_trans_blocks(inode
, pextents
);
5613 * Account for index blocks, block groups bitmaps and block group
5614 * descriptor blocks if modify datablocks and index blocks
5615 * worse case, the indexs blocks spread over different block groups
5617 * If datablocks are discontiguous, they are possible to spread over
5618 * different block groups too. If they are contiguous, with flexbg,
5619 * they could still across block group boundary.
5621 * Also account for superblock, inode, quota and xattr blocks
5623 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5626 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5632 * How many index blocks need to touch to map @lblocks logical blocks
5633 * to @pextents physical extents?
5635 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5640 * Now let's see how many group bitmaps and group descriptors need
5643 groups
= idxblocks
+ pextents
;
5645 if (groups
> ngroups
)
5647 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5648 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5650 /* bitmaps and block group descriptor blocks */
5651 ret
+= groups
+ gdpblocks
;
5653 /* Blocks for super block, inode, quota and xattr blocks */
5654 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5660 * Calculate the total number of credits to reserve to fit
5661 * the modification of a single pages into a single transaction,
5662 * which may include multiple chunks of block allocations.
5664 * This could be called via ext4_write_begin()
5666 * We need to consider the worse case, when
5667 * one new block per extent.
5669 int ext4_writepage_trans_blocks(struct inode
*inode
)
5671 int bpp
= ext4_journal_blocks_per_page(inode
);
5674 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5676 /* Account for data blocks for journalled mode */
5677 if (ext4_should_journal_data(inode
))
5683 * Calculate the journal credits for a chunk of data modification.
5685 * This is called from DIO, fallocate or whoever calling
5686 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5688 * journal buffers for data blocks are not included here, as DIO
5689 * and fallocate do no need to journal data buffers.
5691 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5693 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5697 * The caller must have previously called ext4_reserve_inode_write().
5698 * Give this, we know that the caller already has write access to iloc->bh.
5700 int ext4_mark_iloc_dirty(handle_t
*handle
,
5701 struct inode
*inode
, struct ext4_iloc
*iloc
)
5705 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
)))) {
5709 ext4_fc_track_inode(handle
, inode
);
5711 if (IS_I_VERSION(inode
))
5712 inode_inc_iversion(inode
);
5714 /* the do_update_inode consumes one bh->b_count */
5717 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5718 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5724 * On success, We end up with an outstanding reference count against
5725 * iloc->bh. This _must_ be cleaned up later.
5729 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5730 struct ext4_iloc
*iloc
)
5734 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5737 err
= ext4_get_inode_loc(inode
, iloc
);
5739 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5740 err
= ext4_journal_get_write_access(handle
, inode
->i_sb
,
5741 iloc
->bh
, EXT4_JTR_NONE
);
5747 ext4_std_error(inode
->i_sb
, err
);
5751 static int __ext4_expand_extra_isize(struct inode
*inode
,
5752 unsigned int new_extra_isize
,
5753 struct ext4_iloc
*iloc
,
5754 handle_t
*handle
, int *no_expand
)
5756 struct ext4_inode
*raw_inode
;
5757 struct ext4_xattr_ibody_header
*header
;
5758 unsigned int inode_size
= EXT4_INODE_SIZE(inode
->i_sb
);
5759 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5762 /* this was checked at iget time, but double check for good measure */
5763 if ((EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
> inode_size
) ||
5764 (ei
->i_extra_isize
& 3)) {
5765 EXT4_ERROR_INODE(inode
, "bad extra_isize %u (inode size %u)",
5767 EXT4_INODE_SIZE(inode
->i_sb
));
5768 return -EFSCORRUPTED
;
5770 if ((new_extra_isize
< ei
->i_extra_isize
) ||
5771 (new_extra_isize
< 4) ||
5772 (new_extra_isize
> inode_size
- EXT4_GOOD_OLD_INODE_SIZE
))
5773 return -EINVAL
; /* Should never happen */
5775 raw_inode
= ext4_raw_inode(iloc
);
5777 header
= IHDR(inode
, raw_inode
);
5779 /* No extended attributes present */
5780 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5781 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5782 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
+
5783 EXT4_I(inode
)->i_extra_isize
, 0,
5784 new_extra_isize
- EXT4_I(inode
)->i_extra_isize
);
5785 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5789 /* try to expand with EAs present */
5790 error
= ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5794 * Inode size expansion failed; don't try again
5803 * Expand an inode by new_extra_isize bytes.
5804 * Returns 0 on success or negative error number on failure.
5806 static int ext4_try_to_expand_extra_isize(struct inode
*inode
,
5807 unsigned int new_extra_isize
,
5808 struct ext4_iloc iloc
,
5814 if (ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
))
5818 * In nojournal mode, we can immediately attempt to expand
5819 * the inode. When journaled, we first need to obtain extra
5820 * buffer credits since we may write into the EA block
5821 * with this same handle. If journal_extend fails, then it will
5822 * only result in a minor loss of functionality for that inode.
5823 * If this is felt to be critical, then e2fsck should be run to
5824 * force a large enough s_min_extra_isize.
5826 if (ext4_journal_extend(handle
,
5827 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
), 0) != 0)
5830 if (ext4_write_trylock_xattr(inode
, &no_expand
) == 0)
5833 error
= __ext4_expand_extra_isize(inode
, new_extra_isize
, &iloc
,
5834 handle
, &no_expand
);
5835 ext4_write_unlock_xattr(inode
, &no_expand
);
5840 int ext4_expand_extra_isize(struct inode
*inode
,
5841 unsigned int new_extra_isize
,
5842 struct ext4_iloc
*iloc
)
5848 if (ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5853 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
,
5854 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
));
5855 if (IS_ERR(handle
)) {
5856 error
= PTR_ERR(handle
);
5861 ext4_write_lock_xattr(inode
, &no_expand
);
5863 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5864 error
= ext4_journal_get_write_access(handle
, inode
->i_sb
, iloc
->bh
,
5871 error
= __ext4_expand_extra_isize(inode
, new_extra_isize
, iloc
,
5872 handle
, &no_expand
);
5874 rc
= ext4_mark_iloc_dirty(handle
, inode
, iloc
);
5879 ext4_write_unlock_xattr(inode
, &no_expand
);
5880 ext4_journal_stop(handle
);
5885 * What we do here is to mark the in-core inode as clean with respect to inode
5886 * dirtiness (it may still be data-dirty).
5887 * This means that the in-core inode may be reaped by prune_icache
5888 * without having to perform any I/O. This is a very good thing,
5889 * because *any* task may call prune_icache - even ones which
5890 * have a transaction open against a different journal.
5892 * Is this cheating? Not really. Sure, we haven't written the
5893 * inode out, but prune_icache isn't a user-visible syncing function.
5894 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5895 * we start and wait on commits.
5897 int __ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
,
5898 const char *func
, unsigned int line
)
5900 struct ext4_iloc iloc
;
5901 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5905 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5906 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5910 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
)
5911 ext4_try_to_expand_extra_isize(inode
, sbi
->s_want_extra_isize
,
5914 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5917 ext4_error_inode_err(inode
, func
, line
, 0, err
,
5918 "mark_inode_dirty error");
5923 * ext4_dirty_inode() is called from __mark_inode_dirty()
5925 * We're really interested in the case where a file is being extended.
5926 * i_size has been changed by generic_commit_write() and we thus need
5927 * to include the updated inode in the current transaction.
5929 * Also, dquot_alloc_block() will always dirty the inode when blocks
5930 * are allocated to the file.
5932 * If the inode is marked synchronous, we don't honour that here - doing
5933 * so would cause a commit on atime updates, which we don't bother doing.
5934 * We handle synchronous inodes at the highest possible level.
5936 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5940 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5943 ext4_mark_inode_dirty(handle
, inode
);
5944 ext4_journal_stop(handle
);
5947 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5952 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5955 * We have to be very careful here: changing a data block's
5956 * journaling status dynamically is dangerous. If we write a
5957 * data block to the journal, change the status and then delete
5958 * that block, we risk forgetting to revoke the old log record
5959 * from the journal and so a subsequent replay can corrupt data.
5960 * So, first we make sure that the journal is empty and that
5961 * nobody is changing anything.
5964 journal
= EXT4_JOURNAL(inode
);
5967 if (is_journal_aborted(journal
))
5970 /* Wait for all existing dio workers */
5971 inode_dio_wait(inode
);
5974 * Before flushing the journal and switching inode's aops, we have
5975 * to flush all dirty data the inode has. There can be outstanding
5976 * delayed allocations, there can be unwritten extents created by
5977 * fallocate or buffered writes in dioread_nolock mode covered by
5978 * dirty data which can be converted only after flushing the dirty
5979 * data (and journalled aops don't know how to handle these cases).
5982 filemap_invalidate_lock(inode
->i_mapping
);
5983 err
= filemap_write_and_wait(inode
->i_mapping
);
5985 filemap_invalidate_unlock(inode
->i_mapping
);
5990 percpu_down_write(&sbi
->s_writepages_rwsem
);
5991 jbd2_journal_lock_updates(journal
);
5994 * OK, there are no updates running now, and all cached data is
5995 * synced to disk. We are now in a completely consistent state
5996 * which doesn't have anything in the journal, and we know that
5997 * no filesystem updates are running, so it is safe to modify
5998 * the inode's in-core data-journaling state flag now.
6002 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
6004 err
= jbd2_journal_flush(journal
, 0);
6006 jbd2_journal_unlock_updates(journal
);
6007 percpu_up_write(&sbi
->s_writepages_rwsem
);
6010 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
6012 ext4_set_aops(inode
);
6014 jbd2_journal_unlock_updates(journal
);
6015 percpu_up_write(&sbi
->s_writepages_rwsem
);
6018 filemap_invalidate_unlock(inode
->i_mapping
);
6020 /* Finally we can mark the inode as dirty. */
6022 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
6024 return PTR_ERR(handle
);
6026 ext4_fc_mark_ineligible(inode
->i_sb
,
6027 EXT4_FC_REASON_JOURNAL_FLAG_CHANGE
, handle
);
6028 err
= ext4_mark_inode_dirty(handle
, inode
);
6029 ext4_handle_sync(handle
);
6030 ext4_journal_stop(handle
);
6031 ext4_std_error(inode
->i_sb
, err
);
6036 static int ext4_bh_unmapped(handle_t
*handle
, struct inode
*inode
,
6037 struct buffer_head
*bh
)
6039 return !buffer_mapped(bh
);
6042 vm_fault_t
ext4_page_mkwrite(struct vm_fault
*vmf
)
6044 struct vm_area_struct
*vma
= vmf
->vma
;
6045 struct page
*page
= vmf
->page
;
6050 struct file
*file
= vma
->vm_file
;
6051 struct inode
*inode
= file_inode(file
);
6052 struct address_space
*mapping
= inode
->i_mapping
;
6054 get_block_t
*get_block
;
6057 if (unlikely(IS_IMMUTABLE(inode
)))
6058 return VM_FAULT_SIGBUS
;
6060 sb_start_pagefault(inode
->i_sb
);
6061 file_update_time(vma
->vm_file
);
6063 filemap_invalidate_lock_shared(mapping
);
6065 err
= ext4_convert_inline_data(inode
);
6070 * On data journalling we skip straight to the transaction handle:
6071 * there's no delalloc; page truncated will be checked later; the
6072 * early return w/ all buffers mapped (calculates size/len) can't
6073 * be used; and there's no dioread_nolock, so only ext4_get_block.
6075 if (ext4_should_journal_data(inode
))
6078 /* Delalloc case is easy... */
6079 if (test_opt(inode
->i_sb
, DELALLOC
) &&
6080 !ext4_nonda_switch(inode
->i_sb
)) {
6082 err
= block_page_mkwrite(vma
, vmf
,
6083 ext4_da_get_block_prep
);
6084 } while (err
== -ENOSPC
&&
6085 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
6090 size
= i_size_read(inode
);
6091 /* Page got truncated from under us? */
6092 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
6094 ret
= VM_FAULT_NOPAGE
;
6098 if (page
->index
== size
>> PAGE_SHIFT
)
6099 len
= size
& ~PAGE_MASK
;
6103 * Return if we have all the buffers mapped. This avoids the need to do
6104 * journal_start/journal_stop which can block and take a long time
6106 * This cannot be done for data journalling, as we have to add the
6107 * inode to the transaction's list to writeprotect pages on commit.
6109 if (page_has_buffers(page
)) {
6110 if (!ext4_walk_page_buffers(NULL
, inode
, page_buffers(page
),
6112 ext4_bh_unmapped
)) {
6113 /* Wait so that we don't change page under IO */
6114 wait_for_stable_page(page
);
6115 ret
= VM_FAULT_LOCKED
;
6120 /* OK, we need to fill the hole... */
6121 if (ext4_should_dioread_nolock(inode
))
6122 get_block
= ext4_get_block_unwritten
;
6124 get_block
= ext4_get_block
;
6126 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
6127 ext4_writepage_trans_blocks(inode
));
6128 if (IS_ERR(handle
)) {
6129 ret
= VM_FAULT_SIGBUS
;
6133 * Data journalling can't use block_page_mkwrite() because it
6134 * will set_buffer_dirty() before do_journal_get_write_access()
6135 * thus might hit warning messages for dirty metadata buffers.
6137 if (!ext4_should_journal_data(inode
)) {
6138 err
= block_page_mkwrite(vma
, vmf
, get_block
);
6141 size
= i_size_read(inode
);
6142 /* Page got truncated from under us? */
6143 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
6144 ret
= VM_FAULT_NOPAGE
;
6148 if (page
->index
== size
>> PAGE_SHIFT
)
6149 len
= size
& ~PAGE_MASK
;
6153 err
= __block_write_begin(page
, 0, len
, ext4_get_block
);
6155 ret
= VM_FAULT_SIGBUS
;
6156 if (ext4_walk_page_buffers(handle
, inode
,
6157 page_buffers(page
), 0, len
, NULL
,
6158 do_journal_get_write_access
))
6160 if (ext4_walk_page_buffers(handle
, inode
,
6161 page_buffers(page
), 0, len
, NULL
,
6164 if (ext4_jbd2_inode_add_write(handle
, inode
,
6165 page_offset(page
), len
))
6167 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
6172 ext4_journal_stop(handle
);
6173 if (err
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
6176 ret
= block_page_mkwrite_return(err
);
6178 filemap_invalidate_unlock_shared(mapping
);
6179 sb_end_pagefault(inode
->i_sb
);
6183 ext4_journal_stop(handle
);