1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/inode.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
23 #include <linux/time.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/dax.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/bitops.h>
41 #include <linux/iomap.h>
43 #include "ext4_jbd2.h"
48 #include <trace/events/ext4.h>
50 #define MPAGE_DA_EXTENT_TAIL 0x01
52 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
53 struct ext4_inode_info
*ei
)
55 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
58 int offset
= offsetof(struct ext4_inode
, i_checksum_lo
);
59 unsigned int csum_size
= sizeof(dummy_csum
);
61 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
, offset
);
62 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
, csum_size
);
64 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
65 EXT4_GOOD_OLD_INODE_SIZE
- offset
);
67 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
68 offset
= offsetof(struct ext4_inode
, i_checksum_hi
);
69 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+
70 EXT4_GOOD_OLD_INODE_SIZE
,
71 offset
- EXT4_GOOD_OLD_INODE_SIZE
);
72 if (EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
73 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
,
77 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
78 EXT4_INODE_SIZE(inode
->i_sb
) - offset
);
84 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
85 struct ext4_inode_info
*ei
)
87 __u32 provided
, calculated
;
89 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
90 cpu_to_le32(EXT4_OS_LINUX
) ||
91 !ext4_has_metadata_csum(inode
->i_sb
))
94 provided
= le16_to_cpu(raw
->i_checksum_lo
);
95 calculated
= ext4_inode_csum(inode
, raw
, ei
);
96 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
97 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
98 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
100 calculated
&= 0xFFFF;
102 return provided
== calculated
;
105 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
106 struct ext4_inode_info
*ei
)
110 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
111 cpu_to_le32(EXT4_OS_LINUX
) ||
112 !ext4_has_metadata_csum(inode
->i_sb
))
115 csum
= ext4_inode_csum(inode
, raw
, ei
);
116 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
117 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
118 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
119 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
122 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
125 trace_ext4_begin_ordered_truncate(inode
, new_size
);
127 * If jinode is zero, then we never opened the file for
128 * writing, so there's no need to call
129 * jbd2_journal_begin_ordered_truncate() since there's no
130 * outstanding writes we need to flush.
132 if (!EXT4_I(inode
)->jinode
)
134 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
135 EXT4_I(inode
)->jinode
,
139 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
140 unsigned int length
);
141 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
142 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
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 * Restart the transaction associated with *handle. This does a commit,
167 * so before we call here everything must be consistently dirtied against
170 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
176 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
177 * moment, get_block can be called only for blocks inside i_size since
178 * page cache has been already dropped and writes are blocked by
179 * i_mutex. So we can safely drop the i_data_sem here.
181 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
182 jbd_debug(2, "restarting handle %p\n", handle
);
183 up_write(&EXT4_I(inode
)->i_data_sem
);
184 ret
= ext4_journal_restart(handle
, nblocks
);
185 down_write(&EXT4_I(inode
)->i_data_sem
);
186 ext4_discard_preallocations(inode
);
192 * Called at the last iput() if i_nlink is zero.
194 void ext4_evict_inode(struct inode
*inode
)
198 int extra_credits
= 3;
199 struct ext4_xattr_inode_array
*ea_inode_array
= NULL
;
201 trace_ext4_evict_inode(inode
);
203 if (inode
->i_nlink
) {
205 * When journalling data dirty buffers are tracked only in the
206 * journal. So although mm thinks everything is clean and
207 * ready for reaping the inode might still have some pages to
208 * write in the running transaction or waiting to be
209 * checkpointed. Thus calling jbd2_journal_invalidatepage()
210 * (via truncate_inode_pages()) to discard these buffers can
211 * cause data loss. Also even if we did not discard these
212 * buffers, we would have no way to find them after the inode
213 * is reaped and thus user could see stale data if he tries to
214 * read them before the transaction is checkpointed. So be
215 * careful and force everything to disk here... We use
216 * ei->i_datasync_tid to store the newest transaction
217 * containing inode's data.
219 * Note that directories do not have this problem because they
220 * don't use page cache.
222 if (inode
->i_ino
!= EXT4_JOURNAL_INO
&&
223 ext4_should_journal_data(inode
) &&
224 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
)) &&
225 inode
->i_data
.nrpages
) {
226 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
227 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
229 jbd2_complete_transaction(journal
, commit_tid
);
230 filemap_write_and_wait(&inode
->i_data
);
232 truncate_inode_pages_final(&inode
->i_data
);
237 if (is_bad_inode(inode
))
239 dquot_initialize(inode
);
241 if (ext4_should_order_data(inode
))
242 ext4_begin_ordered_truncate(inode
, 0);
243 truncate_inode_pages_final(&inode
->i_data
);
246 * Protect us against freezing - iput() caller didn't have to have any
247 * protection against it
249 sb_start_intwrite(inode
->i_sb
);
251 if (!IS_NOQUOTA(inode
))
252 extra_credits
+= EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
);
254 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
255 ext4_blocks_for_truncate(inode
)+extra_credits
);
256 if (IS_ERR(handle
)) {
257 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
259 * If we're going to skip the normal cleanup, we still need to
260 * make sure that the in-core orphan linked list is properly
263 ext4_orphan_del(NULL
, inode
);
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(inode
->i_sb
,
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 sb_end_intwrite(inode
->i_sb
);
306 ext4_xattr_inode_array_free(ea_inode_array
);
311 * Kill off the orphan record which ext4_truncate created.
312 * AKPM: I think this can be inside the above `if'.
313 * Note that ext4_orphan_del() has to be able to cope with the
314 * deletion of a non-existent orphan - this is because we don't
315 * know if ext4_truncate() actually created an orphan record.
316 * (Well, we could do this if we need to, but heck - it works)
318 ext4_orphan_del(handle
, inode
);
319 EXT4_I(inode
)->i_dtime
= get_seconds();
322 * One subtle ordering requirement: if anything has gone wrong
323 * (transaction abort, IO errors, whatever), then we can still
324 * do these next steps (the fs will already have been marked as
325 * having errors), but we can't free the inode if the mark_dirty
328 if (ext4_mark_inode_dirty(handle
, inode
))
329 /* If that failed, just do the required in-core inode clear. */
330 ext4_clear_inode(inode
);
332 ext4_free_inode(handle
, inode
);
333 ext4_journal_stop(handle
);
334 sb_end_intwrite(inode
->i_sb
);
335 ext4_xattr_inode_array_free(ea_inode_array
);
338 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
342 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
344 return &EXT4_I(inode
)->i_reserved_quota
;
349 * Called with i_data_sem down, which is important since we can call
350 * ext4_discard_preallocations() from here.
352 void ext4_da_update_reserve_space(struct inode
*inode
,
353 int used
, int quota_claim
)
355 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
356 struct ext4_inode_info
*ei
= EXT4_I(inode
);
358 spin_lock(&ei
->i_block_reservation_lock
);
359 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
360 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
361 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
362 "with only %d reserved data blocks",
363 __func__
, inode
->i_ino
, used
,
364 ei
->i_reserved_data_blocks
);
366 used
= ei
->i_reserved_data_blocks
;
369 /* Update per-inode reservations */
370 ei
->i_reserved_data_blocks
-= used
;
371 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
373 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
375 /* Update quota subsystem for data blocks */
377 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
380 * We did fallocate with an offset that is already delayed
381 * allocated. So on delayed allocated writeback we should
382 * not re-claim the quota for fallocated blocks.
384 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
388 * If we have done all the pending block allocations and if
389 * there aren't any writers on the inode, we can discard the
390 * inode's preallocations.
392 if ((ei
->i_reserved_data_blocks
== 0) &&
393 (atomic_read(&inode
->i_writecount
) == 0))
394 ext4_discard_preallocations(inode
);
397 static int __check_block_validity(struct inode
*inode
, const char *func
,
399 struct ext4_map_blocks
*map
)
401 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
403 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
404 "lblock %lu mapped to illegal pblock %llu "
405 "(length %d)", (unsigned long) map
->m_lblk
,
406 map
->m_pblk
, map
->m_len
);
407 return -EFSCORRUPTED
;
412 int ext4_issue_zeroout(struct inode
*inode
, ext4_lblk_t lblk
, ext4_fsblk_t pblk
,
417 if (ext4_encrypted_inode(inode
))
418 return fscrypt_zeroout_range(inode
, lblk
, pblk
, len
);
420 ret
= sb_issue_zeroout(inode
->i_sb
, pblk
, len
, GFP_NOFS
);
427 #define check_block_validity(inode, map) \
428 __check_block_validity((inode), __func__, __LINE__, (map))
430 #ifdef ES_AGGRESSIVE_TEST
431 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
433 struct ext4_map_blocks
*es_map
,
434 struct ext4_map_blocks
*map
,
441 * There is a race window that the result is not the same.
442 * e.g. xfstests #223 when dioread_nolock enables. The reason
443 * is that we lookup a block mapping in extent status tree with
444 * out taking i_data_sem. So at the time the unwritten extent
445 * could be converted.
447 down_read(&EXT4_I(inode
)->i_data_sem
);
448 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
449 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
450 EXT4_GET_BLOCKS_KEEP_SIZE
);
452 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
453 EXT4_GET_BLOCKS_KEEP_SIZE
);
455 up_read((&EXT4_I(inode
)->i_data_sem
));
458 * We don't check m_len because extent will be collpased in status
459 * tree. So the m_len might not equal.
461 if (es_map
->m_lblk
!= map
->m_lblk
||
462 es_map
->m_flags
!= map
->m_flags
||
463 es_map
->m_pblk
!= map
->m_pblk
) {
464 printk("ES cache assertion failed for inode: %lu "
465 "es_cached ex [%d/%d/%llu/%x] != "
466 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
467 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
468 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
469 map
->m_len
, map
->m_pblk
, map
->m_flags
,
473 #endif /* ES_AGGRESSIVE_TEST */
476 * The ext4_map_blocks() function tries to look up the requested blocks,
477 * and returns if the blocks are already mapped.
479 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
480 * and store the allocated blocks in the result buffer head and mark it
483 * If file type is extents based, it will call ext4_ext_map_blocks(),
484 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
487 * On success, it returns the number of blocks being mapped or allocated. if
488 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
489 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
491 * It returns 0 if plain look up failed (blocks have not been allocated), in
492 * that case, @map is returned as unmapped but we still do fill map->m_len to
493 * indicate the length of a hole starting at map->m_lblk.
495 * It returns the error in case of allocation failure.
497 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
498 struct ext4_map_blocks
*map
, int flags
)
500 struct extent_status es
;
503 #ifdef ES_AGGRESSIVE_TEST
504 struct ext4_map_blocks orig_map
;
506 memcpy(&orig_map
, map
, sizeof(*map
));
510 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
511 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
512 (unsigned long) map
->m_lblk
);
515 * ext4_map_blocks returns an int, and m_len is an unsigned int
517 if (unlikely(map
->m_len
> INT_MAX
))
518 map
->m_len
= INT_MAX
;
520 /* We can handle the block number less than EXT_MAX_BLOCKS */
521 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
522 return -EFSCORRUPTED
;
524 /* Lookup extent status tree firstly */
525 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
526 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
527 map
->m_pblk
= ext4_es_pblock(&es
) +
528 map
->m_lblk
- es
.es_lblk
;
529 map
->m_flags
|= ext4_es_is_written(&es
) ?
530 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
531 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
532 if (retval
> map
->m_len
)
535 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
537 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
538 if (retval
> map
->m_len
)
545 #ifdef ES_AGGRESSIVE_TEST
546 ext4_map_blocks_es_recheck(handle
, inode
, map
,
553 * Try to see if we can get the block without requesting a new
556 down_read(&EXT4_I(inode
)->i_data_sem
);
557 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
558 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
559 EXT4_GET_BLOCKS_KEEP_SIZE
);
561 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
562 EXT4_GET_BLOCKS_KEEP_SIZE
);
567 if (unlikely(retval
!= map
->m_len
)) {
568 ext4_warning(inode
->i_sb
,
569 "ES len assertion failed for inode "
570 "%lu: retval %d != map->m_len %d",
571 inode
->i_ino
, retval
, map
->m_len
);
575 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
576 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
577 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
578 !(status
& EXTENT_STATUS_WRITTEN
) &&
579 ext4_find_delalloc_range(inode
, map
->m_lblk
,
580 map
->m_lblk
+ map
->m_len
- 1))
581 status
|= EXTENT_STATUS_DELAYED
;
582 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
583 map
->m_len
, map
->m_pblk
, status
);
587 up_read((&EXT4_I(inode
)->i_data_sem
));
590 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
591 ret
= check_block_validity(inode
, map
);
596 /* If it is only a block(s) look up */
597 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
601 * Returns if the blocks have already allocated
603 * Note that if blocks have been preallocated
604 * ext4_ext_get_block() returns the create = 0
605 * with buffer head unmapped.
607 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
609 * If we need to convert extent to unwritten
610 * we continue and do the actual work in
611 * ext4_ext_map_blocks()
613 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
617 * Here we clear m_flags because after allocating an new extent,
618 * it will be set again.
620 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
623 * New blocks allocate and/or writing to unwritten extent
624 * will possibly result in updating i_data, so we take
625 * the write lock of i_data_sem, and call get_block()
626 * with create == 1 flag.
628 down_write(&EXT4_I(inode
)->i_data_sem
);
631 * We need to check for EXT4 here because migrate
632 * could have changed the inode type in between
634 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
635 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
637 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
639 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
641 * We allocated new blocks which will result in
642 * i_data's format changing. Force the migrate
643 * to fail by clearing migrate flags
645 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
649 * Update reserved blocks/metadata blocks after successful
650 * block allocation which had been deferred till now. We don't
651 * support fallocate for non extent files. So we can update
652 * reserve space here.
655 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
656 ext4_da_update_reserve_space(inode
, retval
, 1);
662 if (unlikely(retval
!= map
->m_len
)) {
663 ext4_warning(inode
->i_sb
,
664 "ES len assertion failed for inode "
665 "%lu: retval %d != map->m_len %d",
666 inode
->i_ino
, retval
, map
->m_len
);
671 * We have to zeroout blocks before inserting them into extent
672 * status tree. Otherwise someone could look them up there and
673 * use them before they are really zeroed. We also have to
674 * unmap metadata before zeroing as otherwise writeback can
675 * overwrite zeros with stale data from block device.
677 if (flags
& EXT4_GET_BLOCKS_ZERO
&&
678 map
->m_flags
& EXT4_MAP_MAPPED
&&
679 map
->m_flags
& EXT4_MAP_NEW
) {
680 clean_bdev_aliases(inode
->i_sb
->s_bdev
, map
->m_pblk
,
682 ret
= ext4_issue_zeroout(inode
, map
->m_lblk
,
683 map
->m_pblk
, map
->m_len
);
691 * If the extent has been zeroed out, we don't need to update
692 * extent status tree.
694 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
695 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
696 if (ext4_es_is_written(&es
))
699 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
700 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
701 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
702 !(status
& EXTENT_STATUS_WRITTEN
) &&
703 ext4_find_delalloc_range(inode
, map
->m_lblk
,
704 map
->m_lblk
+ map
->m_len
- 1))
705 status
|= EXTENT_STATUS_DELAYED
;
706 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
707 map
->m_pblk
, status
);
715 up_write((&EXT4_I(inode
)->i_data_sem
));
716 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
717 ret
= check_block_validity(inode
, map
);
722 * Inodes with freshly allocated blocks where contents will be
723 * visible after transaction commit must be on transaction's
726 if (map
->m_flags
& EXT4_MAP_NEW
&&
727 !(map
->m_flags
& EXT4_MAP_UNWRITTEN
) &&
728 !(flags
& EXT4_GET_BLOCKS_ZERO
) &&
729 !ext4_is_quota_file(inode
) &&
730 ext4_should_order_data(inode
)) {
731 if (flags
& EXT4_GET_BLOCKS_IO_SUBMIT
)
732 ret
= ext4_jbd2_inode_add_wait(handle
, inode
);
734 ret
= ext4_jbd2_inode_add_write(handle
, inode
);
743 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
744 * we have to be careful as someone else may be manipulating b_state as well.
746 static void ext4_update_bh_state(struct buffer_head
*bh
, unsigned long flags
)
748 unsigned long old_state
;
749 unsigned long new_state
;
751 flags
&= EXT4_MAP_FLAGS
;
753 /* Dummy buffer_head? Set non-atomically. */
755 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | flags
;
759 * Someone else may be modifying b_state. Be careful! This is ugly but
760 * once we get rid of using bh as a container for mapping information
761 * to pass to / from get_block functions, this can go away.
764 old_state
= READ_ONCE(bh
->b_state
);
765 new_state
= (old_state
& ~EXT4_MAP_FLAGS
) | flags
;
767 cmpxchg(&bh
->b_state
, old_state
, new_state
) != old_state
));
770 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
771 struct buffer_head
*bh
, int flags
)
773 struct ext4_map_blocks map
;
776 if (ext4_has_inline_data(inode
))
780 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
782 ret
= ext4_map_blocks(ext4_journal_current_handle(), inode
, &map
,
785 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
786 ext4_update_bh_state(bh
, map
.m_flags
);
787 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
789 } else if (ret
== 0) {
790 /* hole case, need to fill in bh->b_size */
791 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
796 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
797 struct buffer_head
*bh
, int create
)
799 return _ext4_get_block(inode
, iblock
, bh
,
800 create
? EXT4_GET_BLOCKS_CREATE
: 0);
804 * Get block function used when preparing for buffered write if we require
805 * creating an unwritten extent if blocks haven't been allocated. The extent
806 * will be converted to written after the IO is complete.
808 int ext4_get_block_unwritten(struct inode
*inode
, sector_t iblock
,
809 struct buffer_head
*bh_result
, int create
)
811 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
812 inode
->i_ino
, create
);
813 return _ext4_get_block(inode
, iblock
, bh_result
,
814 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
817 /* Maximum number of blocks we map for direct IO at once. */
818 #define DIO_MAX_BLOCKS 4096
821 * Get blocks function for the cases that need to start a transaction -
822 * generally difference cases of direct IO and DAX IO. It also handles retries
825 static int ext4_get_block_trans(struct inode
*inode
, sector_t iblock
,
826 struct buffer_head
*bh_result
, int flags
)
833 /* Trim mapping request to maximum we can map at once for DIO */
834 if (bh_result
->b_size
>> inode
->i_blkbits
> DIO_MAX_BLOCKS
)
835 bh_result
->b_size
= DIO_MAX_BLOCKS
<< inode
->i_blkbits
;
836 dio_credits
= ext4_chunk_trans_blocks(inode
,
837 bh_result
->b_size
>> inode
->i_blkbits
);
839 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, dio_credits
);
841 return PTR_ERR(handle
);
843 ret
= _ext4_get_block(inode
, iblock
, bh_result
, flags
);
844 ext4_journal_stop(handle
);
846 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
851 /* Get block function for DIO reads and writes to inodes without extents */
852 int ext4_dio_get_block(struct inode
*inode
, sector_t iblock
,
853 struct buffer_head
*bh
, int create
)
855 /* We don't expect handle for direct IO */
856 WARN_ON_ONCE(ext4_journal_current_handle());
859 return _ext4_get_block(inode
, iblock
, bh
, 0);
860 return ext4_get_block_trans(inode
, iblock
, bh
, EXT4_GET_BLOCKS_CREATE
);
864 * Get block function for AIO DIO writes when we create unwritten extent if
865 * blocks are not allocated yet. The extent will be converted to written
866 * after IO is complete.
868 static int ext4_dio_get_block_unwritten_async(struct inode
*inode
,
869 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
873 /* We don't expect handle for direct IO */
874 WARN_ON_ONCE(ext4_journal_current_handle());
876 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
877 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
880 * When doing DIO using unwritten extents, we need io_end to convert
881 * unwritten extents to written on IO completion. We allocate io_end
882 * once we spot unwritten extent and store it in b_private. Generic
883 * DIO code keeps b_private set and furthermore passes the value to
884 * our completion callback in 'private' argument.
886 if (!ret
&& buffer_unwritten(bh_result
)) {
887 if (!bh_result
->b_private
) {
888 ext4_io_end_t
*io_end
;
890 io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
893 bh_result
->b_private
= io_end
;
894 ext4_set_io_unwritten_flag(inode
, io_end
);
896 set_buffer_defer_completion(bh_result
);
903 * Get block function for non-AIO DIO writes when we create unwritten extent if
904 * blocks are not allocated yet. The extent will be converted to written
905 * after IO is complete by ext4_direct_IO_write().
907 static int ext4_dio_get_block_unwritten_sync(struct inode
*inode
,
908 sector_t iblock
, struct buffer_head
*bh_result
, int create
)
912 /* We don't expect handle for direct IO */
913 WARN_ON_ONCE(ext4_journal_current_handle());
915 ret
= ext4_get_block_trans(inode
, iblock
, bh_result
,
916 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
919 * Mark inode as having pending DIO writes to unwritten extents.
920 * ext4_direct_IO_write() checks this flag and converts extents to
923 if (!ret
&& buffer_unwritten(bh_result
))
924 ext4_set_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
929 static int ext4_dio_get_block_overwrite(struct inode
*inode
, sector_t iblock
,
930 struct buffer_head
*bh_result
, int create
)
934 ext4_debug("ext4_dio_get_block_overwrite: inode %lu, create flag %d\n",
935 inode
->i_ino
, create
);
936 /* We don't expect handle for direct IO */
937 WARN_ON_ONCE(ext4_journal_current_handle());
939 ret
= _ext4_get_block(inode
, iblock
, bh_result
, 0);
941 * Blocks should have been preallocated! ext4_file_write_iter() checks
944 WARN_ON_ONCE(!buffer_mapped(bh_result
) || buffer_unwritten(bh_result
));
951 * `handle' can be NULL if create is zero
953 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
954 ext4_lblk_t block
, int map_flags
)
956 struct ext4_map_blocks map
;
957 struct buffer_head
*bh
;
958 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
961 J_ASSERT(handle
!= NULL
|| create
== 0);
965 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
968 return create
? ERR_PTR(-ENOSPC
) : NULL
;
972 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
974 return ERR_PTR(-ENOMEM
);
975 if (map
.m_flags
& EXT4_MAP_NEW
) {
976 J_ASSERT(create
!= 0);
977 J_ASSERT(handle
!= NULL
);
980 * Now that we do not always journal data, we should
981 * keep in mind whether this should always journal the
982 * new buffer as metadata. For now, regular file
983 * writes use ext4_get_block instead, so it's not a
987 BUFFER_TRACE(bh
, "call get_create_access");
988 err
= ext4_journal_get_create_access(handle
, bh
);
993 if (!buffer_uptodate(bh
)) {
994 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
995 set_buffer_uptodate(bh
);
998 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
999 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
1003 BUFFER_TRACE(bh
, "not a new buffer");
1007 return ERR_PTR(err
);
1010 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
1011 ext4_lblk_t block
, int map_flags
)
1013 struct buffer_head
*bh
;
1015 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
1018 if (!bh
|| buffer_uptodate(bh
))
1020 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1, &bh
);
1022 if (buffer_uptodate(bh
))
1025 return ERR_PTR(-EIO
);
1028 /* Read a contiguous batch of blocks. */
1029 int ext4_bread_batch(struct inode
*inode
, ext4_lblk_t block
, int bh_count
,
1030 bool wait
, struct buffer_head
**bhs
)
1034 for (i
= 0; i
< bh_count
; i
++) {
1035 bhs
[i
] = ext4_getblk(NULL
, inode
, block
+ i
, 0 /* map_flags */);
1036 if (IS_ERR(bhs
[i
])) {
1037 err
= PTR_ERR(bhs
[i
]);
1043 for (i
= 0; i
< bh_count
; i
++)
1044 /* Note that NULL bhs[i] is valid because of holes. */
1045 if (bhs
[i
] && !buffer_uptodate(bhs
[i
]))
1046 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1,
1052 for (i
= 0; i
< bh_count
; i
++)
1054 wait_on_buffer(bhs
[i
]);
1056 for (i
= 0; i
< bh_count
; i
++) {
1057 if (bhs
[i
] && !buffer_uptodate(bhs
[i
])) {
1065 for (i
= 0; i
< bh_count
; i
++) {
1072 int ext4_walk_page_buffers(handle_t
*handle
,
1073 struct buffer_head
*head
,
1077 int (*fn
)(handle_t
*handle
,
1078 struct buffer_head
*bh
))
1080 struct buffer_head
*bh
;
1081 unsigned block_start
, block_end
;
1082 unsigned blocksize
= head
->b_size
;
1084 struct buffer_head
*next
;
1086 for (bh
= head
, block_start
= 0;
1087 ret
== 0 && (bh
!= head
|| !block_start
);
1088 block_start
= block_end
, bh
= next
) {
1089 next
= bh
->b_this_page
;
1090 block_end
= block_start
+ blocksize
;
1091 if (block_end
<= from
|| block_start
>= to
) {
1092 if (partial
&& !buffer_uptodate(bh
))
1096 err
= (*fn
)(handle
, bh
);
1104 * To preserve ordering, it is essential that the hole instantiation and
1105 * the data write be encapsulated in a single transaction. We cannot
1106 * close off a transaction and start a new one between the ext4_get_block()
1107 * and the commit_write(). So doing the jbd2_journal_start at the start of
1108 * prepare_write() is the right place.
1110 * Also, this function can nest inside ext4_writepage(). In that case, we
1111 * *know* that ext4_writepage() has generated enough buffer credits to do the
1112 * whole page. So we won't block on the journal in that case, which is good,
1113 * because the caller may be PF_MEMALLOC.
1115 * By accident, ext4 can be reentered when a transaction is open via
1116 * quota file writes. If we were to commit the transaction while thus
1117 * reentered, there can be a deadlock - we would be holding a quota
1118 * lock, and the commit would never complete if another thread had a
1119 * transaction open and was blocking on the quota lock - a ranking
1122 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
1123 * will _not_ run commit under these circumstances because handle->h_ref
1124 * is elevated. We'll still have enough credits for the tiny quotafile
1127 int do_journal_get_write_access(handle_t
*handle
,
1128 struct buffer_head
*bh
)
1130 int dirty
= buffer_dirty(bh
);
1133 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1136 * __block_write_begin() could have dirtied some buffers. Clean
1137 * the dirty bit as jbd2_journal_get_write_access() could complain
1138 * otherwise about fs integrity issues. Setting of the dirty bit
1139 * by __block_write_begin() isn't a real problem here as we clear
1140 * the bit before releasing a page lock and thus writeback cannot
1141 * ever write the buffer.
1144 clear_buffer_dirty(bh
);
1145 BUFFER_TRACE(bh
, "get write access");
1146 ret
= ext4_journal_get_write_access(handle
, bh
);
1148 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1152 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1153 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1154 get_block_t
*get_block
)
1156 unsigned from
= pos
& (PAGE_SIZE
- 1);
1157 unsigned to
= from
+ len
;
1158 struct inode
*inode
= page
->mapping
->host
;
1159 unsigned block_start
, block_end
;
1162 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1164 struct buffer_head
*bh
, *head
, *wait
[2], **wait_bh
= wait
;
1165 bool decrypt
= false;
1167 BUG_ON(!PageLocked(page
));
1168 BUG_ON(from
> PAGE_SIZE
);
1169 BUG_ON(to
> PAGE_SIZE
);
1172 if (!page_has_buffers(page
))
1173 create_empty_buffers(page
, blocksize
, 0);
1174 head
= page_buffers(page
);
1175 bbits
= ilog2(blocksize
);
1176 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1178 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1179 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
1180 block_end
= block_start
+ blocksize
;
1181 if (block_end
<= from
|| block_start
>= to
) {
1182 if (PageUptodate(page
)) {
1183 if (!buffer_uptodate(bh
))
1184 set_buffer_uptodate(bh
);
1189 clear_buffer_new(bh
);
1190 if (!buffer_mapped(bh
)) {
1191 WARN_ON(bh
->b_size
!= blocksize
);
1192 err
= get_block(inode
, block
, bh
, 1);
1195 if (buffer_new(bh
)) {
1196 clean_bdev_bh_alias(bh
);
1197 if (PageUptodate(page
)) {
1198 clear_buffer_new(bh
);
1199 set_buffer_uptodate(bh
);
1200 mark_buffer_dirty(bh
);
1203 if (block_end
> to
|| block_start
< from
)
1204 zero_user_segments(page
, to
, block_end
,
1209 if (PageUptodate(page
)) {
1210 if (!buffer_uptodate(bh
))
1211 set_buffer_uptodate(bh
);
1214 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1215 !buffer_unwritten(bh
) &&
1216 (block_start
< from
|| block_end
> to
)) {
1217 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
1219 decrypt
= ext4_encrypted_inode(inode
) &&
1220 S_ISREG(inode
->i_mode
);
1224 * If we issued read requests, let them complete.
1226 while (wait_bh
> wait
) {
1227 wait_on_buffer(*--wait_bh
);
1228 if (!buffer_uptodate(*wait_bh
))
1232 page_zero_new_buffers(page
, from
, to
);
1234 err
= fscrypt_decrypt_page(page
->mapping
->host
, page
,
1235 PAGE_SIZE
, 0, page
->index
);
1240 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1241 loff_t pos
, unsigned len
, unsigned flags
,
1242 struct page
**pagep
, void **fsdata
)
1244 struct inode
*inode
= mapping
->host
;
1245 int ret
, needed_blocks
;
1252 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
1255 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1257 * Reserve one block more for addition to orphan list in case
1258 * we allocate blocks but write fails for some reason
1260 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1261 index
= pos
>> PAGE_SHIFT
;
1262 from
= pos
& (PAGE_SIZE
- 1);
1265 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1266 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1275 * grab_cache_page_write_begin() can take a long time if the
1276 * system is thrashing due to memory pressure, or if the page
1277 * is being written back. So grab it first before we start
1278 * the transaction handle. This also allows us to allocate
1279 * the page (if needed) without using GFP_NOFS.
1282 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1288 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1289 if (IS_ERR(handle
)) {
1291 return PTR_ERR(handle
);
1295 if (page
->mapping
!= mapping
) {
1296 /* The page got truncated from under us */
1299 ext4_journal_stop(handle
);
1302 /* In case writeback began while the page was unlocked */
1303 wait_for_stable_page(page
);
1305 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1306 if (ext4_should_dioread_nolock(inode
))
1307 ret
= ext4_block_write_begin(page
, pos
, len
,
1308 ext4_get_block_unwritten
);
1310 ret
= ext4_block_write_begin(page
, pos
, len
,
1313 if (ext4_should_dioread_nolock(inode
))
1314 ret
= __block_write_begin(page
, pos
, len
,
1315 ext4_get_block_unwritten
);
1317 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1319 if (!ret
&& ext4_should_journal_data(inode
)) {
1320 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1322 do_journal_get_write_access
);
1328 * __block_write_begin may have instantiated a few blocks
1329 * outside i_size. Trim these off again. Don't need
1330 * i_size_read because we hold i_mutex.
1332 * Add inode to orphan list in case we crash before
1335 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1336 ext4_orphan_add(handle
, inode
);
1338 ext4_journal_stop(handle
);
1339 if (pos
+ len
> inode
->i_size
) {
1340 ext4_truncate_failed_write(inode
);
1342 * If truncate failed early the inode might
1343 * still be on the orphan list; we need to
1344 * make sure the inode is removed from the
1345 * orphan list in that case.
1348 ext4_orphan_del(NULL
, inode
);
1351 if (ret
== -ENOSPC
&&
1352 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1361 /* For write_end() in data=journal mode */
1362 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1365 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1367 set_buffer_uptodate(bh
);
1368 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1369 clear_buffer_meta(bh
);
1370 clear_buffer_prio(bh
);
1375 * We need to pick up the new inode size which generic_commit_write gave us
1376 * `file' can be NULL - eg, when called from page_symlink().
1378 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1379 * buffers are managed internally.
1381 static int ext4_write_end(struct file
*file
,
1382 struct address_space
*mapping
,
1383 loff_t pos
, unsigned len
, unsigned copied
,
1384 struct page
*page
, void *fsdata
)
1386 handle_t
*handle
= ext4_journal_current_handle();
1387 struct inode
*inode
= mapping
->host
;
1388 loff_t old_size
= inode
->i_size
;
1390 int i_size_changed
= 0;
1391 int inline_data
= ext4_has_inline_data(inode
);
1393 trace_ext4_write_end(inode
, pos
, len
, copied
);
1395 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1404 copied
= block_write_end(file
, mapping
, pos
,
1405 len
, copied
, page
, fsdata
);
1407 * it's important to update i_size while still holding page lock:
1408 * page writeout could otherwise come in and zero beyond i_size.
1410 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1415 pagecache_isize_extended(inode
, old_size
, pos
);
1417 * Don't mark the inode dirty under page lock. First, it unnecessarily
1418 * makes the holding time of page lock longer. Second, it forces lock
1419 * ordering of page lock and transaction start for journaling
1422 if (i_size_changed
|| inline_data
)
1423 ext4_mark_inode_dirty(handle
, inode
);
1425 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1426 /* if we have allocated more blocks and copied
1427 * less. We will have blocks allocated outside
1428 * inode->i_size. So truncate them
1430 ext4_orphan_add(handle
, inode
);
1432 ret2
= ext4_journal_stop(handle
);
1436 if (pos
+ len
> inode
->i_size
) {
1437 ext4_truncate_failed_write(inode
);
1439 * If truncate failed early the inode might still be
1440 * on the orphan list; we need to make sure the inode
1441 * is removed from the orphan list in that case.
1444 ext4_orphan_del(NULL
, inode
);
1447 return ret
? ret
: copied
;
1451 * This is a private version of page_zero_new_buffers() which doesn't
1452 * set the buffer to be dirty, since in data=journalled mode we need
1453 * to call ext4_handle_dirty_metadata() instead.
1455 static void ext4_journalled_zero_new_buffers(handle_t
*handle
,
1457 unsigned from
, unsigned to
)
1459 unsigned int block_start
= 0, block_end
;
1460 struct buffer_head
*head
, *bh
;
1462 bh
= head
= page_buffers(page
);
1464 block_end
= block_start
+ bh
->b_size
;
1465 if (buffer_new(bh
)) {
1466 if (block_end
> from
&& block_start
< to
) {
1467 if (!PageUptodate(page
)) {
1468 unsigned start
, size
;
1470 start
= max(from
, block_start
);
1471 size
= min(to
, block_end
) - start
;
1473 zero_user(page
, start
, size
);
1474 write_end_fn(handle
, bh
);
1476 clear_buffer_new(bh
);
1479 block_start
= block_end
;
1480 bh
= bh
->b_this_page
;
1481 } while (bh
!= head
);
1484 static int ext4_journalled_write_end(struct file
*file
,
1485 struct address_space
*mapping
,
1486 loff_t pos
, unsigned len
, unsigned copied
,
1487 struct page
*page
, void *fsdata
)
1489 handle_t
*handle
= ext4_journal_current_handle();
1490 struct inode
*inode
= mapping
->host
;
1491 loff_t old_size
= inode
->i_size
;
1495 int size_changed
= 0;
1496 int inline_data
= ext4_has_inline_data(inode
);
1498 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1499 from
= pos
& (PAGE_SIZE
- 1);
1502 BUG_ON(!ext4_handle_valid(handle
));
1505 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1513 } else if (unlikely(copied
< len
) && !PageUptodate(page
)) {
1515 ext4_journalled_zero_new_buffers(handle
, page
, from
, to
);
1517 if (unlikely(copied
< len
))
1518 ext4_journalled_zero_new_buffers(handle
, page
,
1520 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1521 from
+ copied
, &partial
,
1524 SetPageUptodate(page
);
1526 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1527 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1528 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1533 pagecache_isize_extended(inode
, old_size
, pos
);
1535 if (size_changed
|| inline_data
) {
1536 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1541 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1542 /* if we have allocated more blocks and copied
1543 * less. We will have blocks allocated outside
1544 * inode->i_size. So truncate them
1546 ext4_orphan_add(handle
, inode
);
1549 ret2
= ext4_journal_stop(handle
);
1552 if (pos
+ len
> inode
->i_size
) {
1553 ext4_truncate_failed_write(inode
);
1555 * If truncate failed early the inode might still be
1556 * on the orphan list; we need to make sure the inode
1557 * is removed from the orphan list in that case.
1560 ext4_orphan_del(NULL
, inode
);
1563 return ret
? ret
: copied
;
1567 * Reserve space for a single cluster
1569 static int ext4_da_reserve_space(struct inode
*inode
)
1571 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1572 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1576 * We will charge metadata quota at writeout time; this saves
1577 * us from metadata over-estimation, though we may go over by
1578 * a small amount in the end. Here we just reserve for data.
1580 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1584 spin_lock(&ei
->i_block_reservation_lock
);
1585 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1586 spin_unlock(&ei
->i_block_reservation_lock
);
1587 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1590 ei
->i_reserved_data_blocks
++;
1591 trace_ext4_da_reserve_space(inode
);
1592 spin_unlock(&ei
->i_block_reservation_lock
);
1594 return 0; /* success */
1597 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1599 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1600 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1603 return; /* Nothing to release, exit */
1605 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1607 trace_ext4_da_release_space(inode
, to_free
);
1608 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1610 * if there aren't enough reserved blocks, then the
1611 * counter is messed up somewhere. Since this
1612 * function is called from invalidate page, it's
1613 * harmless to return without any action.
1615 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1616 "ino %lu, to_free %d with only %d reserved "
1617 "data blocks", inode
->i_ino
, to_free
,
1618 ei
->i_reserved_data_blocks
);
1620 to_free
= ei
->i_reserved_data_blocks
;
1622 ei
->i_reserved_data_blocks
-= to_free
;
1624 /* update fs dirty data blocks counter */
1625 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1627 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1629 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1632 static void ext4_da_page_release_reservation(struct page
*page
,
1633 unsigned int offset
,
1634 unsigned int length
)
1636 int to_release
= 0, contiguous_blks
= 0;
1637 struct buffer_head
*head
, *bh
;
1638 unsigned int curr_off
= 0;
1639 struct inode
*inode
= page
->mapping
->host
;
1640 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1641 unsigned int stop
= offset
+ length
;
1645 BUG_ON(stop
> PAGE_SIZE
|| stop
< length
);
1647 head
= page_buffers(page
);
1650 unsigned int next_off
= curr_off
+ bh
->b_size
;
1652 if (next_off
> stop
)
1655 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1658 clear_buffer_delay(bh
);
1659 } else if (contiguous_blks
) {
1660 lblk
= page
->index
<<
1661 (PAGE_SHIFT
- inode
->i_blkbits
);
1662 lblk
+= (curr_off
>> inode
->i_blkbits
) -
1664 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1665 contiguous_blks
= 0;
1667 curr_off
= next_off
;
1668 } while ((bh
= bh
->b_this_page
) != head
);
1670 if (contiguous_blks
) {
1671 lblk
= page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1672 lblk
+= (curr_off
>> inode
->i_blkbits
) - contiguous_blks
;
1673 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1676 /* If we have released all the blocks belonging to a cluster, then we
1677 * need to release the reserved space for that cluster. */
1678 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1679 while (num_clusters
> 0) {
1680 lblk
= (page
->index
<< (PAGE_SHIFT
- inode
->i_blkbits
)) +
1681 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1682 if (sbi
->s_cluster_ratio
== 1 ||
1683 !ext4_find_delalloc_cluster(inode
, lblk
))
1684 ext4_da_release_space(inode
, 1);
1691 * Delayed allocation stuff
1694 struct mpage_da_data
{
1695 struct inode
*inode
;
1696 struct writeback_control
*wbc
;
1698 pgoff_t first_page
; /* The first page to write */
1699 pgoff_t next_page
; /* Current page to examine */
1700 pgoff_t last_page
; /* Last page to examine */
1702 * Extent to map - this can be after first_page because that can be
1703 * fully mapped. We somewhat abuse m_flags to store whether the extent
1704 * is delalloc or unwritten.
1706 struct ext4_map_blocks map
;
1707 struct ext4_io_submit io_submit
; /* IO submission data */
1708 unsigned int do_map
:1;
1711 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1716 struct pagevec pvec
;
1717 struct inode
*inode
= mpd
->inode
;
1718 struct address_space
*mapping
= inode
->i_mapping
;
1720 /* This is necessary when next_page == 0. */
1721 if (mpd
->first_page
>= mpd
->next_page
)
1724 index
= mpd
->first_page
;
1725 end
= mpd
->next_page
- 1;
1727 ext4_lblk_t start
, last
;
1728 start
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1729 last
= end
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1730 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1733 pagevec_init(&pvec
, 0);
1734 while (index
<= end
) {
1735 nr_pages
= pagevec_lookup_range(&pvec
, mapping
, &index
, end
);
1738 for (i
= 0; i
< nr_pages
; i
++) {
1739 struct page
*page
= pvec
.pages
[i
];
1741 BUG_ON(!PageLocked(page
));
1742 BUG_ON(PageWriteback(page
));
1744 if (page_mapped(page
))
1745 clear_page_dirty_for_io(page
);
1746 block_invalidatepage(page
, 0, PAGE_SIZE
);
1747 ClearPageUptodate(page
);
1751 pagevec_release(&pvec
);
1755 static void ext4_print_free_blocks(struct inode
*inode
)
1757 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1758 struct super_block
*sb
= inode
->i_sb
;
1759 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1761 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1762 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1763 ext4_count_free_clusters(sb
)));
1764 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1765 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1766 (long long) EXT4_C2B(EXT4_SB(sb
),
1767 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1768 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1769 (long long) EXT4_C2B(EXT4_SB(sb
),
1770 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1771 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1772 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1773 ei
->i_reserved_data_blocks
);
1777 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1779 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1783 * This function is grabs code from the very beginning of
1784 * ext4_map_blocks, but assumes that the caller is from delayed write
1785 * time. This function looks up the requested blocks and sets the
1786 * buffer delay bit under the protection of i_data_sem.
1788 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1789 struct ext4_map_blocks
*map
,
1790 struct buffer_head
*bh
)
1792 struct extent_status es
;
1794 sector_t invalid_block
= ~((sector_t
) 0xffff);
1795 #ifdef ES_AGGRESSIVE_TEST
1796 struct ext4_map_blocks orig_map
;
1798 memcpy(&orig_map
, map
, sizeof(*map
));
1801 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1805 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1806 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1807 (unsigned long) map
->m_lblk
);
1809 /* Lookup extent status tree firstly */
1810 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1811 if (ext4_es_is_hole(&es
)) {
1813 down_read(&EXT4_I(inode
)->i_data_sem
);
1818 * Delayed extent could be allocated by fallocate.
1819 * So we need to check it.
1821 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1822 map_bh(bh
, inode
->i_sb
, invalid_block
);
1824 set_buffer_delay(bh
);
1828 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1829 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1830 if (retval
> map
->m_len
)
1831 retval
= map
->m_len
;
1832 map
->m_len
= retval
;
1833 if (ext4_es_is_written(&es
))
1834 map
->m_flags
|= EXT4_MAP_MAPPED
;
1835 else if (ext4_es_is_unwritten(&es
))
1836 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1840 #ifdef ES_AGGRESSIVE_TEST
1841 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1847 * Try to see if we can get the block without requesting a new
1848 * file system block.
1850 down_read(&EXT4_I(inode
)->i_data_sem
);
1851 if (ext4_has_inline_data(inode
))
1853 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1854 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1856 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1862 * XXX: __block_prepare_write() unmaps passed block,
1866 * If the block was allocated from previously allocated cluster,
1867 * then we don't need to reserve it again. However we still need
1868 * to reserve metadata for every block we're going to write.
1870 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
== 1 ||
1871 !ext4_find_delalloc_cluster(inode
, map
->m_lblk
)) {
1872 ret
= ext4_da_reserve_space(inode
);
1874 /* not enough space to reserve */
1880 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1881 ~0, EXTENT_STATUS_DELAYED
);
1887 map_bh(bh
, inode
->i_sb
, invalid_block
);
1889 set_buffer_delay(bh
);
1890 } else if (retval
> 0) {
1892 unsigned int status
;
1894 if (unlikely(retval
!= map
->m_len
)) {
1895 ext4_warning(inode
->i_sb
,
1896 "ES len assertion failed for inode "
1897 "%lu: retval %d != map->m_len %d",
1898 inode
->i_ino
, retval
, map
->m_len
);
1902 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1903 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1904 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1905 map
->m_pblk
, status
);
1911 up_read((&EXT4_I(inode
)->i_data_sem
));
1917 * This is a special get_block_t callback which is used by
1918 * ext4_da_write_begin(). It will either return mapped block or
1919 * reserve space for a single block.
1921 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1922 * We also have b_blocknr = -1 and b_bdev initialized properly
1924 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1925 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1926 * initialized properly.
1928 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1929 struct buffer_head
*bh
, int create
)
1931 struct ext4_map_blocks map
;
1934 BUG_ON(create
== 0);
1935 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1937 map
.m_lblk
= iblock
;
1941 * first, we need to know whether the block is allocated already
1942 * preallocated blocks are unmapped but should treated
1943 * the same as allocated blocks.
1945 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1949 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1950 ext4_update_bh_state(bh
, map
.m_flags
);
1952 if (buffer_unwritten(bh
)) {
1953 /* A delayed write to unwritten bh should be marked
1954 * new and mapped. Mapped ensures that we don't do
1955 * get_block multiple times when we write to the same
1956 * offset and new ensures that we do proper zero out
1957 * for partial write.
1960 set_buffer_mapped(bh
);
1965 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1971 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1977 static int __ext4_journalled_writepage(struct page
*page
,
1980 struct address_space
*mapping
= page
->mapping
;
1981 struct inode
*inode
= mapping
->host
;
1982 struct buffer_head
*page_bufs
= NULL
;
1983 handle_t
*handle
= NULL
;
1984 int ret
= 0, err
= 0;
1985 int inline_data
= ext4_has_inline_data(inode
);
1986 struct buffer_head
*inode_bh
= NULL
;
1988 ClearPageChecked(page
);
1991 BUG_ON(page
->index
!= 0);
1992 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1993 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1994 if (inode_bh
== NULL
)
1997 page_bufs
= page_buffers(page
);
2002 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
2006 * We need to release the page lock before we start the
2007 * journal, so grab a reference so the page won't disappear
2008 * out from under us.
2013 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2014 ext4_writepage_trans_blocks(inode
));
2015 if (IS_ERR(handle
)) {
2016 ret
= PTR_ERR(handle
);
2018 goto out_no_pagelock
;
2020 BUG_ON(!ext4_handle_valid(handle
));
2024 if (page
->mapping
!= mapping
) {
2025 /* The page got truncated from under us */
2026 ext4_journal_stop(handle
);
2032 ret
= ext4_mark_inode_dirty(handle
, inode
);
2034 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2035 do_journal_get_write_access
);
2037 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2042 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
2043 err
= ext4_journal_stop(handle
);
2047 if (!ext4_has_inline_data(inode
))
2048 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
2050 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2059 * Note that we don't need to start a transaction unless we're journaling data
2060 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2061 * need to file the inode to the transaction's list in ordered mode because if
2062 * we are writing back data added by write(), the inode is already there and if
2063 * we are writing back data modified via mmap(), no one guarantees in which
2064 * transaction the data will hit the disk. In case we are journaling data, we
2065 * cannot start transaction directly because transaction start ranks above page
2066 * lock so we have to do some magic.
2068 * This function can get called via...
2069 * - ext4_writepages after taking page lock (have journal handle)
2070 * - journal_submit_inode_data_buffers (no journal handle)
2071 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2072 * - grab_page_cache when doing write_begin (have journal handle)
2074 * We don't do any block allocation in this function. If we have page with
2075 * multiple blocks we need to write those buffer_heads that are mapped. This
2076 * is important for mmaped based write. So if we do with blocksize 1K
2077 * truncate(f, 1024);
2078 * a = mmap(f, 0, 4096);
2080 * truncate(f, 4096);
2081 * we have in the page first buffer_head mapped via page_mkwrite call back
2082 * but other buffer_heads would be unmapped but dirty (dirty done via the
2083 * do_wp_page). So writepage should write the first block. If we modify
2084 * the mmap area beyond 1024 we will again get a page_fault and the
2085 * page_mkwrite callback will do the block allocation and mark the
2086 * buffer_heads mapped.
2088 * We redirty the page if we have any buffer_heads that is either delay or
2089 * unwritten in the page.
2091 * We can get recursively called as show below.
2093 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2096 * But since we don't do any block allocation we should not deadlock.
2097 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2099 static int ext4_writepage(struct page
*page
,
2100 struct writeback_control
*wbc
)
2105 struct buffer_head
*page_bufs
= NULL
;
2106 struct inode
*inode
= page
->mapping
->host
;
2107 struct ext4_io_submit io_submit
;
2108 bool keep_towrite
= false;
2110 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
)))) {
2111 ext4_invalidatepage(page
, 0, PAGE_SIZE
);
2116 trace_ext4_writepage(page
);
2117 size
= i_size_read(inode
);
2118 if (page
->index
== size
>> PAGE_SHIFT
)
2119 len
= size
& ~PAGE_MASK
;
2123 page_bufs
= page_buffers(page
);
2125 * We cannot do block allocation or other extent handling in this
2126 * function. If there are buffers needing that, we have to redirty
2127 * the page. But we may reach here when we do a journal commit via
2128 * journal_submit_inode_data_buffers() and in that case we must write
2129 * allocated buffers to achieve data=ordered mode guarantees.
2131 * Also, if there is only one buffer per page (the fs block
2132 * size == the page size), if one buffer needs block
2133 * allocation or needs to modify the extent tree to clear the
2134 * unwritten flag, we know that the page can't be written at
2135 * all, so we might as well refuse the write immediately.
2136 * Unfortunately if the block size != page size, we can't as
2137 * easily detect this case using ext4_walk_page_buffers(), but
2138 * for the extremely common case, this is an optimization that
2139 * skips a useless round trip through ext4_bio_write_page().
2141 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2142 ext4_bh_delay_or_unwritten
)) {
2143 redirty_page_for_writepage(wbc
, page
);
2144 if ((current
->flags
& PF_MEMALLOC
) ||
2145 (inode
->i_sb
->s_blocksize
== PAGE_SIZE
)) {
2147 * For memory cleaning there's no point in writing only
2148 * some buffers. So just bail out. Warn if we came here
2149 * from direct reclaim.
2151 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2156 keep_towrite
= true;
2159 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2161 * It's mmapped pagecache. Add buffers and journal it. There
2162 * doesn't seem much point in redirtying the page here.
2164 return __ext4_journalled_writepage(page
, len
);
2166 ext4_io_submit_init(&io_submit
, wbc
);
2167 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2168 if (!io_submit
.io_end
) {
2169 redirty_page_for_writepage(wbc
, page
);
2173 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
2174 ext4_io_submit(&io_submit
);
2175 /* Drop io_end reference we got from init */
2176 ext4_put_io_end_defer(io_submit
.io_end
);
2180 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
2186 BUG_ON(page
->index
!= mpd
->first_page
);
2187 clear_page_dirty_for_io(page
);
2189 * We have to be very careful here! Nothing protects writeback path
2190 * against i_size changes and the page can be writeably mapped into
2191 * page tables. So an application can be growing i_size and writing
2192 * data through mmap while writeback runs. clear_page_dirty_for_io()
2193 * write-protects our page in page tables and the page cannot get
2194 * written to again until we release page lock. So only after
2195 * clear_page_dirty_for_io() we are safe to sample i_size for
2196 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2197 * on the barrier provided by TestClearPageDirty in
2198 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2199 * after page tables are updated.
2201 size
= i_size_read(mpd
->inode
);
2202 if (page
->index
== size
>> PAGE_SHIFT
)
2203 len
= size
& ~PAGE_MASK
;
2206 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
2208 mpd
->wbc
->nr_to_write
--;
2214 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
2217 * mballoc gives us at most this number of blocks...
2218 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2219 * The rest of mballoc seems to handle chunks up to full group size.
2221 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2224 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2226 * @mpd - extent of blocks
2227 * @lblk - logical number of the block in the file
2228 * @bh - buffer head we want to add to the extent
2230 * The function is used to collect contig. blocks in the same state. If the
2231 * buffer doesn't require mapping for writeback and we haven't started the
2232 * extent of buffers to map yet, the function returns 'true' immediately - the
2233 * caller can write the buffer right away. Otherwise the function returns true
2234 * if the block has been added to the extent, false if the block couldn't be
2237 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
2238 struct buffer_head
*bh
)
2240 struct ext4_map_blocks
*map
= &mpd
->map
;
2242 /* Buffer that doesn't need mapping for writeback? */
2243 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
2244 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
2245 /* So far no extent to map => we write the buffer right away */
2246 if (map
->m_len
== 0)
2251 /* First block in the extent? */
2252 if (map
->m_len
== 0) {
2253 /* We cannot map unless handle is started... */
2258 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2262 /* Don't go larger than mballoc is willing to allocate */
2263 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2266 /* Can we merge the block to our big extent? */
2267 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2268 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2276 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2278 * @mpd - extent of blocks for mapping
2279 * @head - the first buffer in the page
2280 * @bh - buffer we should start processing from
2281 * @lblk - logical number of the block in the file corresponding to @bh
2283 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2284 * the page for IO if all buffers in this page were mapped and there's no
2285 * accumulated extent of buffers to map or add buffers in the page to the
2286 * extent of buffers to map. The function returns 1 if the caller can continue
2287 * by processing the next page, 0 if it should stop adding buffers to the
2288 * extent to map because we cannot extend it anymore. It can also return value
2289 * < 0 in case of error during IO submission.
2291 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2292 struct buffer_head
*head
,
2293 struct buffer_head
*bh
,
2296 struct inode
*inode
= mpd
->inode
;
2298 ext4_lblk_t blocks
= (i_size_read(inode
) + i_blocksize(inode
) - 1)
2299 >> inode
->i_blkbits
;
2302 BUG_ON(buffer_locked(bh
));
2304 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2305 /* Found extent to map? */
2308 /* Buffer needs mapping and handle is not started? */
2311 /* Everything mapped so far and we hit EOF */
2314 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2315 /* So far everything mapped? Submit the page for IO. */
2316 if (mpd
->map
.m_len
== 0) {
2317 err
= mpage_submit_page(mpd
, head
->b_page
);
2321 return lblk
< blocks
;
2325 * mpage_map_buffers - update buffers corresponding to changed extent and
2326 * submit fully mapped pages for IO
2328 * @mpd - description of extent to map, on return next extent to map
2330 * Scan buffers corresponding to changed extent (we expect corresponding pages
2331 * to be already locked) and update buffer state according to new extent state.
2332 * We map delalloc buffers to their physical location, clear unwritten bits,
2333 * and mark buffers as uninit when we perform writes to unwritten extents
2334 * and do extent conversion after IO is finished. If the last page is not fully
2335 * mapped, we update @map to the next extent in the last page that needs
2336 * mapping. Otherwise we submit the page for IO.
2338 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2340 struct pagevec pvec
;
2342 struct inode
*inode
= mpd
->inode
;
2343 struct buffer_head
*head
, *bh
;
2344 int bpp_bits
= PAGE_SHIFT
- inode
->i_blkbits
;
2350 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2351 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2352 lblk
= start
<< bpp_bits
;
2353 pblock
= mpd
->map
.m_pblk
;
2355 pagevec_init(&pvec
, 0);
2356 while (start
<= end
) {
2357 nr_pages
= pagevec_lookup_range(&pvec
, inode
->i_mapping
,
2361 for (i
= 0; i
< nr_pages
; i
++) {
2362 struct page
*page
= pvec
.pages
[i
];
2364 bh
= head
= page_buffers(page
);
2366 if (lblk
< mpd
->map
.m_lblk
)
2368 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2370 * Buffer after end of mapped extent.
2371 * Find next buffer in the page to map.
2374 mpd
->map
.m_flags
= 0;
2376 * FIXME: If dioread_nolock supports
2377 * blocksize < pagesize, we need to make
2378 * sure we add size mapped so far to
2379 * io_end->size as the following call
2380 * can submit the page for IO.
2382 err
= mpage_process_page_bufs(mpd
, head
,
2384 pagevec_release(&pvec
);
2389 if (buffer_delay(bh
)) {
2390 clear_buffer_delay(bh
);
2391 bh
->b_blocknr
= pblock
++;
2393 clear_buffer_unwritten(bh
);
2394 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2397 * FIXME: This is going to break if dioread_nolock
2398 * supports blocksize < pagesize as we will try to
2399 * convert potentially unmapped parts of inode.
2401 mpd
->io_submit
.io_end
->size
+= PAGE_SIZE
;
2402 /* Page fully mapped - let IO run! */
2403 err
= mpage_submit_page(mpd
, page
);
2405 pagevec_release(&pvec
);
2409 pagevec_release(&pvec
);
2411 /* Extent fully mapped and matches with page boundary. We are done. */
2413 mpd
->map
.m_flags
= 0;
2417 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2419 struct inode
*inode
= mpd
->inode
;
2420 struct ext4_map_blocks
*map
= &mpd
->map
;
2421 int get_blocks_flags
;
2422 int err
, dioread_nolock
;
2424 trace_ext4_da_write_pages_extent(inode
, map
);
2426 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2427 * to convert an unwritten extent to be initialized (in the case
2428 * where we have written into one or more preallocated blocks). It is
2429 * possible that we're going to need more metadata blocks than
2430 * previously reserved. However we must not fail because we're in
2431 * writeback and there is nothing we can do about it so it might result
2432 * in data loss. So use reserved blocks to allocate metadata if
2435 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2436 * the blocks in question are delalloc blocks. This indicates
2437 * that the blocks and quotas has already been checked when
2438 * the data was copied into the page cache.
2440 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2441 EXT4_GET_BLOCKS_METADATA_NOFAIL
|
2442 EXT4_GET_BLOCKS_IO_SUBMIT
;
2443 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2445 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2446 if (map
->m_flags
& (1 << BH_Delay
))
2447 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2449 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2452 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2453 if (!mpd
->io_submit
.io_end
->handle
&&
2454 ext4_handle_valid(handle
)) {
2455 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2456 handle
->h_rsv_handle
= NULL
;
2458 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2461 BUG_ON(map
->m_len
== 0);
2462 if (map
->m_flags
& EXT4_MAP_NEW
) {
2463 clean_bdev_aliases(inode
->i_sb
->s_bdev
, map
->m_pblk
,
2470 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2471 * mpd->len and submit pages underlying it for IO
2473 * @handle - handle for journal operations
2474 * @mpd - extent to map
2475 * @give_up_on_write - we set this to true iff there is a fatal error and there
2476 * is no hope of writing the data. The caller should discard
2477 * dirty pages to avoid infinite loops.
2479 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2480 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2481 * them to initialized or split the described range from larger unwritten
2482 * extent. Note that we need not map all the described range since allocation
2483 * can return less blocks or the range is covered by more unwritten extents. We
2484 * cannot map more because we are limited by reserved transaction credits. On
2485 * the other hand we always make sure that the last touched page is fully
2486 * mapped so that it can be written out (and thus forward progress is
2487 * guaranteed). After mapping we submit all mapped pages for IO.
2489 static int mpage_map_and_submit_extent(handle_t
*handle
,
2490 struct mpage_da_data
*mpd
,
2491 bool *give_up_on_write
)
2493 struct inode
*inode
= mpd
->inode
;
2494 struct ext4_map_blocks
*map
= &mpd
->map
;
2499 mpd
->io_submit
.io_end
->offset
=
2500 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2502 err
= mpage_map_one_extent(handle
, mpd
);
2504 struct super_block
*sb
= inode
->i_sb
;
2506 if (ext4_forced_shutdown(EXT4_SB(sb
)) ||
2507 EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2508 goto invalidate_dirty_pages
;
2510 * Let the uper layers retry transient errors.
2511 * In the case of ENOSPC, if ext4_count_free_blocks()
2512 * is non-zero, a commit should free up blocks.
2514 if ((err
== -ENOMEM
) ||
2515 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2517 goto update_disksize
;
2520 ext4_msg(sb
, KERN_CRIT
,
2521 "Delayed block allocation failed for "
2522 "inode %lu at logical offset %llu with"
2523 " max blocks %u with error %d",
2525 (unsigned long long)map
->m_lblk
,
2526 (unsigned)map
->m_len
, -err
);
2527 ext4_msg(sb
, KERN_CRIT
,
2528 "This should not happen!! Data will "
2531 ext4_print_free_blocks(inode
);
2532 invalidate_dirty_pages
:
2533 *give_up_on_write
= true;
2538 * Update buffer state, submit mapped pages, and get us new
2541 err
= mpage_map_and_submit_buffers(mpd
);
2543 goto update_disksize
;
2544 } while (map
->m_len
);
2548 * Update on-disk size after IO is submitted. Races with
2549 * truncate are avoided by checking i_size under i_data_sem.
2551 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_SHIFT
;
2552 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2556 down_write(&EXT4_I(inode
)->i_data_sem
);
2557 i_size
= i_size_read(inode
);
2558 if (disksize
> i_size
)
2560 if (disksize
> EXT4_I(inode
)->i_disksize
)
2561 EXT4_I(inode
)->i_disksize
= disksize
;
2562 up_write(&EXT4_I(inode
)->i_data_sem
);
2563 err2
= ext4_mark_inode_dirty(handle
, inode
);
2565 ext4_error(inode
->i_sb
,
2566 "Failed to mark inode %lu dirty",
2575 * Calculate the total number of credits to reserve for one writepages
2576 * iteration. This is called from ext4_writepages(). We map an extent of
2577 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2578 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2579 * bpp - 1 blocks in bpp different extents.
2581 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2583 int bpp
= ext4_journal_blocks_per_page(inode
);
2585 return ext4_meta_trans_blocks(inode
,
2586 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2590 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2591 * and underlying extent to map
2593 * @mpd - where to look for pages
2595 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2596 * IO immediately. When we find a page which isn't mapped we start accumulating
2597 * extent of buffers underlying these pages that needs mapping (formed by
2598 * either delayed or unwritten buffers). We also lock the pages containing
2599 * these buffers. The extent found is returned in @mpd structure (starting at
2600 * mpd->lblk with length mpd->len blocks).
2602 * Note that this function can attach bios to one io_end structure which are
2603 * neither logically nor physically contiguous. Although it may seem as an
2604 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2605 * case as we need to track IO to all buffers underlying a page in one io_end.
2607 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2609 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2610 struct pagevec pvec
;
2611 unsigned int nr_pages
;
2612 long left
= mpd
->wbc
->nr_to_write
;
2613 pgoff_t index
= mpd
->first_page
;
2614 pgoff_t end
= mpd
->last_page
;
2617 int blkbits
= mpd
->inode
->i_blkbits
;
2619 struct buffer_head
*head
;
2621 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2622 tag
= PAGECACHE_TAG_TOWRITE
;
2624 tag
= PAGECACHE_TAG_DIRTY
;
2626 pagevec_init(&pvec
, 0);
2628 mpd
->next_page
= index
;
2629 while (index
<= end
) {
2630 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2631 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2635 for (i
= 0; i
< nr_pages
; i
++) {
2636 struct page
*page
= pvec
.pages
[i
];
2639 * At this point, the page may be truncated or
2640 * invalidated (changing page->mapping to NULL), or
2641 * even swizzled back from swapper_space to tmpfs file
2642 * mapping. However, page->index will not change
2643 * because we have a reference on the page.
2645 if (page
->index
> end
)
2649 * Accumulated enough dirty pages? This doesn't apply
2650 * to WB_SYNC_ALL mode. For integrity sync we have to
2651 * keep going because someone may be concurrently
2652 * dirtying pages, and we might have synced a lot of
2653 * newly appeared dirty pages, but have not synced all
2654 * of the old dirty pages.
2656 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2659 /* If we can't merge this page, we are done. */
2660 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2665 * If the page is no longer dirty, or its mapping no
2666 * longer corresponds to inode we are writing (which
2667 * means it has been truncated or invalidated), or the
2668 * page is already under writeback and we are not doing
2669 * a data integrity writeback, skip the page
2671 if (!PageDirty(page
) ||
2672 (PageWriteback(page
) &&
2673 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2674 unlikely(page
->mapping
!= mapping
)) {
2679 wait_on_page_writeback(page
);
2680 BUG_ON(PageWriteback(page
));
2682 if (mpd
->map
.m_len
== 0)
2683 mpd
->first_page
= page
->index
;
2684 mpd
->next_page
= page
->index
+ 1;
2685 /* Add all dirty buffers to mpd */
2686 lblk
= ((ext4_lblk_t
)page
->index
) <<
2687 (PAGE_SHIFT
- blkbits
);
2688 head
= page_buffers(page
);
2689 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2695 pagevec_release(&pvec
);
2700 pagevec_release(&pvec
);
2704 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2707 struct address_space
*mapping
= data
;
2708 int ret
= ext4_writepage(page
, wbc
);
2709 mapping_set_error(mapping
, ret
);
2713 static int ext4_writepages(struct address_space
*mapping
,
2714 struct writeback_control
*wbc
)
2716 pgoff_t writeback_index
= 0;
2717 long nr_to_write
= wbc
->nr_to_write
;
2718 int range_whole
= 0;
2720 handle_t
*handle
= NULL
;
2721 struct mpage_da_data mpd
;
2722 struct inode
*inode
= mapping
->host
;
2723 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2724 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2726 struct blk_plug plug
;
2727 bool give_up_on_write
= false;
2729 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
2732 percpu_down_read(&sbi
->s_journal_flag_rwsem
);
2733 trace_ext4_writepages(inode
, wbc
);
2735 if (dax_mapping(mapping
)) {
2736 ret
= dax_writeback_mapping_range(mapping
, inode
->i_sb
->s_bdev
,
2738 goto out_writepages
;
2742 * No pages to write? This is mainly a kludge to avoid starting
2743 * a transaction for special inodes like journal inode on last iput()
2744 * because that could violate lock ordering on umount
2746 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2747 goto out_writepages
;
2749 if (ext4_should_journal_data(inode
)) {
2750 struct blk_plug plug
;
2752 blk_start_plug(&plug
);
2753 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2754 blk_finish_plug(&plug
);
2755 goto out_writepages
;
2759 * If the filesystem has aborted, it is read-only, so return
2760 * right away instead of dumping stack traces later on that
2761 * will obscure the real source of the problem. We test
2762 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2763 * the latter could be true if the filesystem is mounted
2764 * read-only, and in that case, ext4_writepages should
2765 * *never* be called, so if that ever happens, we would want
2768 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping
->host
->i_sb
)) ||
2769 sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2771 goto out_writepages
;
2774 if (ext4_should_dioread_nolock(inode
)) {
2776 * We may need to convert up to one extent per block in
2777 * the page and we may dirty the inode.
2779 rsv_blocks
= 1 + ext4_chunk_trans_blocks(inode
,
2780 PAGE_SIZE
>> inode
->i_blkbits
);
2784 * If we have inline data and arrive here, it means that
2785 * we will soon create the block for the 1st page, so
2786 * we'd better clear the inline data here.
2788 if (ext4_has_inline_data(inode
)) {
2789 /* Just inode will be modified... */
2790 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2791 if (IS_ERR(handle
)) {
2792 ret
= PTR_ERR(handle
);
2793 goto out_writepages
;
2795 BUG_ON(ext4_test_inode_state(inode
,
2796 EXT4_STATE_MAY_INLINE_DATA
));
2797 ext4_destroy_inline_data(handle
, inode
);
2798 ext4_journal_stop(handle
);
2801 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2804 if (wbc
->range_cyclic
) {
2805 writeback_index
= mapping
->writeback_index
;
2806 if (writeback_index
)
2808 mpd
.first_page
= writeback_index
;
2811 mpd
.first_page
= wbc
->range_start
>> PAGE_SHIFT
;
2812 mpd
.last_page
= wbc
->range_end
>> PAGE_SHIFT
;
2817 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2819 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2820 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2822 blk_start_plug(&plug
);
2825 * First writeback pages that don't need mapping - we can avoid
2826 * starting a transaction unnecessarily and also avoid being blocked
2827 * in the block layer on device congestion while having transaction
2831 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2832 if (!mpd
.io_submit
.io_end
) {
2836 ret
= mpage_prepare_extent_to_map(&mpd
);
2837 /* Submit prepared bio */
2838 ext4_io_submit(&mpd
.io_submit
);
2839 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2840 mpd
.io_submit
.io_end
= NULL
;
2841 /* Unlock pages we didn't use */
2842 mpage_release_unused_pages(&mpd
, false);
2846 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2847 /* For each extent of pages we use new io_end */
2848 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2849 if (!mpd
.io_submit
.io_end
) {
2855 * We have two constraints: We find one extent to map and we
2856 * must always write out whole page (makes a difference when
2857 * blocksize < pagesize) so that we don't block on IO when we
2858 * try to write out the rest of the page. Journalled mode is
2859 * not supported by delalloc.
2861 BUG_ON(ext4_should_journal_data(inode
));
2862 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2864 /* start a new transaction */
2865 handle
= ext4_journal_start_with_reserve(inode
,
2866 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2867 if (IS_ERR(handle
)) {
2868 ret
= PTR_ERR(handle
);
2869 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2870 "%ld pages, ino %lu; err %d", __func__
,
2871 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2872 /* Release allocated io_end */
2873 ext4_put_io_end(mpd
.io_submit
.io_end
);
2874 mpd
.io_submit
.io_end
= NULL
;
2879 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2880 ret
= mpage_prepare_extent_to_map(&mpd
);
2883 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2887 * We scanned the whole range (or exhausted
2888 * nr_to_write), submitted what was mapped and
2889 * didn't find anything needing mapping. We are
2896 * Caution: If the handle is synchronous,
2897 * ext4_journal_stop() can wait for transaction commit
2898 * to finish which may depend on writeback of pages to
2899 * complete or on page lock to be released. In that
2900 * case, we have to wait until after after we have
2901 * submitted all the IO, released page locks we hold,
2902 * and dropped io_end reference (for extent conversion
2903 * to be able to complete) before stopping the handle.
2905 if (!ext4_handle_valid(handle
) || handle
->h_sync
== 0) {
2906 ext4_journal_stop(handle
);
2910 /* Submit prepared bio */
2911 ext4_io_submit(&mpd
.io_submit
);
2912 /* Unlock pages we didn't use */
2913 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2915 * Drop our io_end reference we got from init. We have
2916 * to be careful and use deferred io_end finishing if
2917 * we are still holding the transaction as we can
2918 * release the last reference to io_end which may end
2919 * up doing unwritten extent conversion.
2922 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2923 ext4_journal_stop(handle
);
2925 ext4_put_io_end(mpd
.io_submit
.io_end
);
2926 mpd
.io_submit
.io_end
= NULL
;
2928 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2930 * Commit the transaction which would
2931 * free blocks released in the transaction
2934 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2938 /* Fatal error - ENOMEM, EIO... */
2943 blk_finish_plug(&plug
);
2944 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2946 mpd
.last_page
= writeback_index
- 1;
2952 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2954 * Set the writeback_index so that range_cyclic
2955 * mode will write it back later
2957 mapping
->writeback_index
= mpd
.first_page
;
2960 trace_ext4_writepages_result(inode
, wbc
, ret
,
2961 nr_to_write
- wbc
->nr_to_write
);
2962 percpu_up_read(&sbi
->s_journal_flag_rwsem
);
2966 static int ext4_nonda_switch(struct super_block
*sb
)
2968 s64 free_clusters
, dirty_clusters
;
2969 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2972 * switch to non delalloc mode if we are running low
2973 * on free block. The free block accounting via percpu
2974 * counters can get slightly wrong with percpu_counter_batch getting
2975 * accumulated on each CPU without updating global counters
2976 * Delalloc need an accurate free block accounting. So switch
2977 * to non delalloc when we are near to error range.
2980 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2982 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2984 * Start pushing delalloc when 1/2 of free blocks are dirty.
2986 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2987 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2989 if (2 * free_clusters
< 3 * dirty_clusters
||
2990 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2992 * free block count is less than 150% of dirty blocks
2993 * or free blocks is less than watermark
3000 /* We always reserve for an inode update; the superblock could be there too */
3001 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
3003 if (likely(ext4_has_feature_large_file(inode
->i_sb
)))
3006 if (pos
+ len
<= 0x7fffffffULL
)
3009 /* We might need to update the superblock to set LARGE_FILE */
3013 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
3014 loff_t pos
, unsigned len
, unsigned flags
,
3015 struct page
**pagep
, void **fsdata
)
3017 int ret
, retries
= 0;
3020 struct inode
*inode
= mapping
->host
;
3023 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
3026 index
= pos
>> PAGE_SHIFT
;
3028 if (ext4_nonda_switch(inode
->i_sb
) ||
3029 S_ISLNK(inode
->i_mode
)) {
3030 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
3031 return ext4_write_begin(file
, mapping
, pos
,
3032 len
, flags
, pagep
, fsdata
);
3034 *fsdata
= (void *)0;
3035 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
3037 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
3038 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
3048 * grab_cache_page_write_begin() can take a long time if the
3049 * system is thrashing due to memory pressure, or if the page
3050 * is being written back. So grab it first before we start
3051 * the transaction handle. This also allows us to allocate
3052 * the page (if needed) without using GFP_NOFS.
3055 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
3061 * With delayed allocation, we don't log the i_disksize update
3062 * if there is delayed block allocation. But we still need
3063 * to journalling the i_disksize update if writes to the end
3064 * of file which has an already mapped buffer.
3067 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
3068 ext4_da_write_credits(inode
, pos
, len
));
3069 if (IS_ERR(handle
)) {
3071 return PTR_ERR(handle
);
3075 if (page
->mapping
!= mapping
) {
3076 /* The page got truncated from under us */
3079 ext4_journal_stop(handle
);
3082 /* In case writeback began while the page was unlocked */
3083 wait_for_stable_page(page
);
3085 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3086 ret
= ext4_block_write_begin(page
, pos
, len
,
3087 ext4_da_get_block_prep
);
3089 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
3093 ext4_journal_stop(handle
);
3095 * block_write_begin may have instantiated a few blocks
3096 * outside i_size. Trim these off again. Don't need
3097 * i_size_read because we hold i_mutex.
3099 if (pos
+ len
> inode
->i_size
)
3100 ext4_truncate_failed_write(inode
);
3102 if (ret
== -ENOSPC
&&
3103 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3115 * Check if we should update i_disksize
3116 * when write to the end of file but not require block allocation
3118 static int ext4_da_should_update_i_disksize(struct page
*page
,
3119 unsigned long offset
)
3121 struct buffer_head
*bh
;
3122 struct inode
*inode
= page
->mapping
->host
;
3126 bh
= page_buffers(page
);
3127 idx
= offset
>> inode
->i_blkbits
;
3129 for (i
= 0; i
< idx
; i
++)
3130 bh
= bh
->b_this_page
;
3132 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3137 static int ext4_da_write_end(struct file
*file
,
3138 struct address_space
*mapping
,
3139 loff_t pos
, unsigned len
, unsigned copied
,
3140 struct page
*page
, void *fsdata
)
3142 struct inode
*inode
= mapping
->host
;
3144 handle_t
*handle
= ext4_journal_current_handle();
3146 unsigned long start
, end
;
3147 int write_mode
= (int)(unsigned long)fsdata
;
3149 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
3150 return ext4_write_end(file
, mapping
, pos
,
3151 len
, copied
, page
, fsdata
);
3153 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3154 start
= pos
& (PAGE_SIZE
- 1);
3155 end
= start
+ copied
- 1;
3158 * generic_write_end() will run mark_inode_dirty() if i_size
3159 * changes. So let's piggyback the i_disksize mark_inode_dirty
3162 new_i_size
= pos
+ copied
;
3163 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
3164 if (ext4_has_inline_data(inode
) ||
3165 ext4_da_should_update_i_disksize(page
, end
)) {
3166 ext4_update_i_disksize(inode
, new_i_size
);
3167 /* We need to mark inode dirty even if
3168 * new_i_size is less that inode->i_size
3169 * bu greater than i_disksize.(hint delalloc)
3171 ext4_mark_inode_dirty(handle
, inode
);
3175 if (write_mode
!= CONVERT_INLINE_DATA
&&
3176 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
3177 ext4_has_inline_data(inode
))
3178 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
3181 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3187 ret2
= ext4_journal_stop(handle
);
3191 return ret
? ret
: copied
;
3194 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
3195 unsigned int length
)
3198 * Drop reserved blocks
3200 BUG_ON(!PageLocked(page
));
3201 if (!page_has_buffers(page
))
3204 ext4_da_page_release_reservation(page
, offset
, length
);
3207 ext4_invalidatepage(page
, offset
, length
);
3213 * Force all delayed allocation blocks to be allocated for a given inode.
3215 int ext4_alloc_da_blocks(struct inode
*inode
)
3217 trace_ext4_alloc_da_blocks(inode
);
3219 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
3223 * We do something simple for now. The filemap_flush() will
3224 * also start triggering a write of the data blocks, which is
3225 * not strictly speaking necessary (and for users of
3226 * laptop_mode, not even desirable). However, to do otherwise
3227 * would require replicating code paths in:
3229 * ext4_writepages() ->
3230 * write_cache_pages() ---> (via passed in callback function)
3231 * __mpage_da_writepage() -->
3232 * mpage_add_bh_to_extent()
3233 * mpage_da_map_blocks()
3235 * The problem is that write_cache_pages(), located in
3236 * mm/page-writeback.c, marks pages clean in preparation for
3237 * doing I/O, which is not desirable if we're not planning on
3240 * We could call write_cache_pages(), and then redirty all of
3241 * the pages by calling redirty_page_for_writepage() but that
3242 * would be ugly in the extreme. So instead we would need to
3243 * replicate parts of the code in the above functions,
3244 * simplifying them because we wouldn't actually intend to
3245 * write out the pages, but rather only collect contiguous
3246 * logical block extents, call the multi-block allocator, and
3247 * then update the buffer heads with the block allocations.
3249 * For now, though, we'll cheat by calling filemap_flush(),
3250 * which will map the blocks, and start the I/O, but not
3251 * actually wait for the I/O to complete.
3253 return filemap_flush(inode
->i_mapping
);
3257 * bmap() is special. It gets used by applications such as lilo and by
3258 * the swapper to find the on-disk block of a specific piece of data.
3260 * Naturally, this is dangerous if the block concerned is still in the
3261 * journal. If somebody makes a swapfile on an ext4 data-journaling
3262 * filesystem and enables swap, then they may get a nasty shock when the
3263 * data getting swapped to that swapfile suddenly gets overwritten by
3264 * the original zero's written out previously to the journal and
3265 * awaiting writeback in the kernel's buffer cache.
3267 * So, if we see any bmap calls here on a modified, data-journaled file,
3268 * take extra steps to flush any blocks which might be in the cache.
3270 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3272 struct inode
*inode
= mapping
->host
;
3277 * We can get here for an inline file via the FIBMAP ioctl
3279 if (ext4_has_inline_data(inode
))
3282 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3283 test_opt(inode
->i_sb
, DELALLOC
)) {
3285 * With delalloc we want to sync the file
3286 * so that we can make sure we allocate
3289 filemap_write_and_wait(mapping
);
3292 if (EXT4_JOURNAL(inode
) &&
3293 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3295 * This is a REALLY heavyweight approach, but the use of
3296 * bmap on dirty files is expected to be extremely rare:
3297 * only if we run lilo or swapon on a freshly made file
3298 * do we expect this to happen.
3300 * (bmap requires CAP_SYS_RAWIO so this does not
3301 * represent an unprivileged user DOS attack --- we'd be
3302 * in trouble if mortal users could trigger this path at
3305 * NB. EXT4_STATE_JDATA is not set on files other than
3306 * regular files. If somebody wants to bmap a directory
3307 * or symlink and gets confused because the buffer
3308 * hasn't yet been flushed to disk, they deserve
3309 * everything they get.
3312 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3313 journal
= EXT4_JOURNAL(inode
);
3314 jbd2_journal_lock_updates(journal
);
3315 err
= jbd2_journal_flush(journal
);
3316 jbd2_journal_unlock_updates(journal
);
3322 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3325 static int ext4_readpage(struct file
*file
, struct page
*page
)
3328 struct inode
*inode
= page
->mapping
->host
;
3330 trace_ext4_readpage(page
);
3332 if (ext4_has_inline_data(inode
))
3333 ret
= ext4_readpage_inline(inode
, page
);
3336 return ext4_mpage_readpages(page
->mapping
, NULL
, page
, 1);
3342 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3343 struct list_head
*pages
, unsigned nr_pages
)
3345 struct inode
*inode
= mapping
->host
;
3347 /* If the file has inline data, no need to do readpages. */
3348 if (ext4_has_inline_data(inode
))
3351 return ext4_mpage_readpages(mapping
, pages
, NULL
, nr_pages
);
3354 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3355 unsigned int length
)
3357 trace_ext4_invalidatepage(page
, offset
, length
);
3359 /* No journalling happens on data buffers when this function is used */
3360 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3362 block_invalidatepage(page
, offset
, length
);
3365 static int __ext4_journalled_invalidatepage(struct page
*page
,
3366 unsigned int offset
,
3367 unsigned int length
)
3369 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3371 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3374 * If it's a full truncate we just forget about the pending dirtying
3376 if (offset
== 0 && length
== PAGE_SIZE
)
3377 ClearPageChecked(page
);
3379 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3382 /* Wrapper for aops... */
3383 static void ext4_journalled_invalidatepage(struct page
*page
,
3384 unsigned int offset
,
3385 unsigned int length
)
3387 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3390 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3392 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3394 trace_ext4_releasepage(page
);
3396 /* Page has dirty journalled data -> cannot release */
3397 if (PageChecked(page
))
3400 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3402 return try_to_free_buffers(page
);
3405 #ifdef CONFIG_FS_DAX
3406 static int ext4_iomap_begin(struct inode
*inode
, loff_t offset
, loff_t length
,
3407 unsigned flags
, struct iomap
*iomap
)
3409 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3410 unsigned int blkbits
= inode
->i_blkbits
;
3411 unsigned long first_block
, last_block
;
3412 struct ext4_map_blocks map
;
3415 if ((offset
>> blkbits
) > EXT4_MAX_LOGICAL_BLOCK
)
3417 first_block
= offset
>> blkbits
;
3418 last_block
= min_t(loff_t
, (offset
+ length
- 1) >> blkbits
,
3419 EXT4_MAX_LOGICAL_BLOCK
);
3421 if (WARN_ON_ONCE(ext4_has_inline_data(inode
)))
3424 map
.m_lblk
= first_block
;
3425 map
.m_len
= last_block
- first_block
+ 1;
3427 if (!(flags
& IOMAP_WRITE
)) {
3428 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3434 /* Trim mapping request to maximum we can map at once for DIO */
3435 if (map
.m_len
> DIO_MAX_BLOCKS
)
3436 map
.m_len
= DIO_MAX_BLOCKS
;
3437 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
3440 * Either we allocate blocks and then we don't get unwritten
3441 * extent so we have reserved enough credits, or the blocks
3442 * are already allocated and unwritten and in that case
3443 * extent conversion fits in the credits as well.
3445 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
3448 return PTR_ERR(handle
);
3450 ret
= ext4_map_blocks(handle
, inode
, &map
,
3451 EXT4_GET_BLOCKS_CREATE_ZERO
);
3453 ext4_journal_stop(handle
);
3454 if (ret
== -ENOSPC
&&
3455 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3461 * If we added blocks beyond i_size, we need to make sure they
3462 * will get truncated if we crash before updating i_size in
3463 * ext4_iomap_end(). For faults we don't need to do that (and
3464 * even cannot because for orphan list operations inode_lock is
3465 * required) - if we happen to instantiate block beyond i_size,
3466 * it is because we race with truncate which has already added
3467 * the inode to the orphan list.
3469 if (!(flags
& IOMAP_FAULT
) && first_block
+ map
.m_len
>
3470 (i_size_read(inode
) + (1 << blkbits
) - 1) >> blkbits
) {
3473 err
= ext4_orphan_add(handle
, inode
);
3475 ext4_journal_stop(handle
);
3479 ext4_journal_stop(handle
);
3483 iomap
->bdev
= inode
->i_sb
->s_bdev
;
3484 iomap
->dax_dev
= sbi
->s_daxdev
;
3485 iomap
->offset
= first_block
<< blkbits
;
3488 iomap
->type
= IOMAP_HOLE
;
3489 iomap
->blkno
= IOMAP_NULL_BLOCK
;
3490 iomap
->length
= (u64
)map
.m_len
<< blkbits
;
3492 if (map
.m_flags
& EXT4_MAP_MAPPED
) {
3493 iomap
->type
= IOMAP_MAPPED
;
3494 } else if (map
.m_flags
& EXT4_MAP_UNWRITTEN
) {
3495 iomap
->type
= IOMAP_UNWRITTEN
;
3500 iomap
->blkno
= (sector_t
)map
.m_pblk
<< (blkbits
- 9);
3501 iomap
->length
= (u64
)map
.m_len
<< blkbits
;
3504 if (map
.m_flags
& EXT4_MAP_NEW
)
3505 iomap
->flags
|= IOMAP_F_NEW
;
3509 static int ext4_iomap_end(struct inode
*inode
, loff_t offset
, loff_t length
,
3510 ssize_t written
, unsigned flags
, struct iomap
*iomap
)
3514 int blkbits
= inode
->i_blkbits
;
3515 bool truncate
= false;
3517 if (!(flags
& IOMAP_WRITE
) || (flags
& IOMAP_FAULT
))
3520 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3521 if (IS_ERR(handle
)) {
3522 ret
= PTR_ERR(handle
);
3525 if (ext4_update_inode_size(inode
, offset
+ written
))
3526 ext4_mark_inode_dirty(handle
, inode
);
3528 * We may need to truncate allocated but not written blocks beyond EOF.
3530 if (iomap
->offset
+ iomap
->length
>
3531 ALIGN(inode
->i_size
, 1 << blkbits
)) {
3532 ext4_lblk_t written_blk
, end_blk
;
3534 written_blk
= (offset
+ written
) >> blkbits
;
3535 end_blk
= (offset
+ length
) >> blkbits
;
3536 if (written_blk
< end_blk
&& ext4_can_truncate(inode
))
3540 * Remove inode from orphan list if we were extending a inode and
3541 * everything went fine.
3543 if (!truncate
&& inode
->i_nlink
&&
3544 !list_empty(&EXT4_I(inode
)->i_orphan
))
3545 ext4_orphan_del(handle
, inode
);
3546 ext4_journal_stop(handle
);
3548 ext4_truncate_failed_write(inode
);
3551 * If truncate failed early the inode might still be on the
3552 * orphan list; we need to make sure the inode is removed from
3553 * the orphan list in that case.
3556 ext4_orphan_del(NULL
, inode
);
3561 const struct iomap_ops ext4_iomap_ops
= {
3562 .iomap_begin
= ext4_iomap_begin
,
3563 .iomap_end
= ext4_iomap_end
,
3568 static int ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3569 ssize_t size
, void *private)
3571 ext4_io_end_t
*io_end
= private;
3573 /* if not async direct IO just return */
3577 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3578 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3579 io_end
, io_end
->inode
->i_ino
, iocb
, offset
, size
);
3582 * Error during AIO DIO. We cannot convert unwritten extents as the
3583 * data was not written. Just clear the unwritten flag and drop io_end.
3586 ext4_clear_io_unwritten_flag(io_end
);
3589 io_end
->offset
= offset
;
3590 io_end
->size
= size
;
3591 ext4_put_io_end(io_end
);
3597 * Handling of direct IO writes.
3599 * For ext4 extent files, ext4 will do direct-io write even to holes,
3600 * preallocated extents, and those write extend the file, no need to
3601 * fall back to buffered IO.
3603 * For holes, we fallocate those blocks, mark them as unwritten
3604 * If those blocks were preallocated, we mark sure they are split, but
3605 * still keep the range to write as unwritten.
3607 * The unwritten extents will be converted to written when DIO is completed.
3608 * For async direct IO, since the IO may still pending when return, we
3609 * set up an end_io call back function, which will do the conversion
3610 * when async direct IO completed.
3612 * If the O_DIRECT write will extend the file then add this inode to the
3613 * orphan list. So recovery will truncate it back to the original size
3614 * if the machine crashes during the write.
3617 static ssize_t
ext4_direct_IO_write(struct kiocb
*iocb
, struct iov_iter
*iter
)
3619 struct file
*file
= iocb
->ki_filp
;
3620 struct inode
*inode
= file
->f_mapping
->host
;
3622 loff_t offset
= iocb
->ki_pos
;
3623 size_t count
= iov_iter_count(iter
);
3625 get_block_t
*get_block_func
= NULL
;
3627 loff_t final_size
= offset
+ count
;
3631 if (final_size
> inode
->i_size
) {
3632 /* Credits for sb + inode write */
3633 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3634 if (IS_ERR(handle
)) {
3635 ret
= PTR_ERR(handle
);
3638 ret
= ext4_orphan_add(handle
, inode
);
3640 ext4_journal_stop(handle
);
3644 ext4_update_i_disksize(inode
, inode
->i_size
);
3645 ext4_journal_stop(handle
);
3648 BUG_ON(iocb
->private == NULL
);
3651 * Make all waiters for direct IO properly wait also for extent
3652 * conversion. This also disallows race between truncate() and
3653 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3655 inode_dio_begin(inode
);
3657 /* If we do a overwrite dio, i_mutex locking can be released */
3658 overwrite
= *((int *)iocb
->private);
3661 inode_unlock(inode
);
3664 * For extent mapped files we could direct write to holes and fallocate.
3666 * Allocated blocks to fill the hole are marked as unwritten to prevent
3667 * parallel buffered read to expose the stale data before DIO complete
3670 * As to previously fallocated extents, ext4 get_block will just simply
3671 * mark the buffer mapped but still keep the extents unwritten.
3673 * For non AIO case, we will convert those unwritten extents to written
3674 * after return back from blockdev_direct_IO. That way we save us from
3675 * allocating io_end structure and also the overhead of offloading
3676 * the extent convertion to a workqueue.
3678 * For async DIO, the conversion needs to be deferred when the
3679 * IO is completed. The ext4 end_io callback function will be
3680 * called to take care of the conversion work. Here for async
3681 * case, we allocate an io_end structure to hook to the iocb.
3683 iocb
->private = NULL
;
3685 get_block_func
= ext4_dio_get_block_overwrite
;
3686 else if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
) ||
3687 round_down(offset
, i_blocksize(inode
)) >= inode
->i_size
) {
3688 get_block_func
= ext4_dio_get_block
;
3689 dio_flags
= DIO_LOCKING
| DIO_SKIP_HOLES
;
3690 } else if (is_sync_kiocb(iocb
)) {
3691 get_block_func
= ext4_dio_get_block_unwritten_sync
;
3692 dio_flags
= DIO_LOCKING
;
3694 get_block_func
= ext4_dio_get_block_unwritten_async
;
3695 dio_flags
= DIO_LOCKING
;
3697 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
, iter
,
3698 get_block_func
, ext4_end_io_dio
, NULL
,
3701 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3702 EXT4_STATE_DIO_UNWRITTEN
)) {
3705 * for non AIO case, since the IO is already
3706 * completed, we could do the conversion right here
3708 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3712 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3715 inode_dio_end(inode
);
3716 /* take i_mutex locking again if we do a ovewrite dio */
3720 if (ret
< 0 && final_size
> inode
->i_size
)
3721 ext4_truncate_failed_write(inode
);
3723 /* Handle extending of i_size after direct IO write */
3727 /* Credits for sb + inode write */
3728 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
3729 if (IS_ERR(handle
)) {
3731 * We wrote the data but cannot extend
3732 * i_size. Bail out. In async io case, we do
3733 * not return error here because we have
3734 * already submmitted the corresponding
3735 * bio. Returning error here makes the caller
3736 * think that this IO is done and failed
3737 * resulting in race with bio's completion
3741 ret
= PTR_ERR(handle
);
3743 ext4_orphan_del(NULL
, inode
);
3748 ext4_orphan_del(handle
, inode
);
3750 loff_t end
= offset
+ ret
;
3751 if (end
> inode
->i_size
) {
3752 ext4_update_i_disksize(inode
, end
);
3753 i_size_write(inode
, end
);
3755 * We're going to return a positive `ret'
3756 * here due to non-zero-length I/O, so there's
3757 * no way of reporting error returns from
3758 * ext4_mark_inode_dirty() to userspace. So
3761 ext4_mark_inode_dirty(handle
, inode
);
3764 err
= ext4_journal_stop(handle
);
3772 static ssize_t
ext4_direct_IO_read(struct kiocb
*iocb
, struct iov_iter
*iter
)
3774 struct address_space
*mapping
= iocb
->ki_filp
->f_mapping
;
3775 struct inode
*inode
= mapping
->host
;
3776 size_t count
= iov_iter_count(iter
);
3780 * Shared inode_lock is enough for us - it protects against concurrent
3781 * writes & truncates and since we take care of writing back page cache,
3782 * we are protected against page writeback as well.
3784 inode_lock_shared(inode
);
3785 ret
= filemap_write_and_wait_range(mapping
, iocb
->ki_pos
,
3786 iocb
->ki_pos
+ count
- 1);
3789 ret
= __blockdev_direct_IO(iocb
, inode
, inode
->i_sb
->s_bdev
,
3790 iter
, ext4_dio_get_block
, NULL
, NULL
, 0);
3792 inode_unlock_shared(inode
);
3796 static ssize_t
ext4_direct_IO(struct kiocb
*iocb
, struct iov_iter
*iter
)
3798 struct file
*file
= iocb
->ki_filp
;
3799 struct inode
*inode
= file
->f_mapping
->host
;
3800 size_t count
= iov_iter_count(iter
);
3801 loff_t offset
= iocb
->ki_pos
;
3804 #ifdef CONFIG_EXT4_FS_ENCRYPTION
3805 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
))
3810 * If we are doing data journalling we don't support O_DIRECT
3812 if (ext4_should_journal_data(inode
))
3815 /* Let buffer I/O handle the inline data case. */
3816 if (ext4_has_inline_data(inode
))
3819 /* DAX uses iomap path now */
3820 if (WARN_ON_ONCE(IS_DAX(inode
)))
3823 trace_ext4_direct_IO_enter(inode
, offset
, count
, iov_iter_rw(iter
));
3824 if (iov_iter_rw(iter
) == READ
)
3825 ret
= ext4_direct_IO_read(iocb
, iter
);
3827 ret
= ext4_direct_IO_write(iocb
, iter
);
3828 trace_ext4_direct_IO_exit(inode
, offset
, count
, iov_iter_rw(iter
), ret
);
3833 * Pages can be marked dirty completely asynchronously from ext4's journalling
3834 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3835 * much here because ->set_page_dirty is called under VFS locks. The page is
3836 * not necessarily locked.
3838 * We cannot just dirty the page and leave attached buffers clean, because the
3839 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3840 * or jbddirty because all the journalling code will explode.
3842 * So what we do is to mark the page "pending dirty" and next time writepage
3843 * is called, propagate that into the buffers appropriately.
3845 static int ext4_journalled_set_page_dirty(struct page
*page
)
3847 SetPageChecked(page
);
3848 return __set_page_dirty_nobuffers(page
);
3851 static int ext4_set_page_dirty(struct page
*page
)
3853 WARN_ON_ONCE(!PageLocked(page
) && !PageDirty(page
));
3854 WARN_ON_ONCE(!page_has_buffers(page
));
3855 return __set_page_dirty_buffers(page
);
3858 static const struct address_space_operations ext4_aops
= {
3859 .readpage
= ext4_readpage
,
3860 .readpages
= ext4_readpages
,
3861 .writepage
= ext4_writepage
,
3862 .writepages
= ext4_writepages
,
3863 .write_begin
= ext4_write_begin
,
3864 .write_end
= ext4_write_end
,
3865 .set_page_dirty
= ext4_set_page_dirty
,
3867 .invalidatepage
= ext4_invalidatepage
,
3868 .releasepage
= ext4_releasepage
,
3869 .direct_IO
= ext4_direct_IO
,
3870 .migratepage
= buffer_migrate_page
,
3871 .is_partially_uptodate
= block_is_partially_uptodate
,
3872 .error_remove_page
= generic_error_remove_page
,
3875 static const struct address_space_operations ext4_journalled_aops
= {
3876 .readpage
= ext4_readpage
,
3877 .readpages
= ext4_readpages
,
3878 .writepage
= ext4_writepage
,
3879 .writepages
= ext4_writepages
,
3880 .write_begin
= ext4_write_begin
,
3881 .write_end
= ext4_journalled_write_end
,
3882 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3884 .invalidatepage
= ext4_journalled_invalidatepage
,
3885 .releasepage
= ext4_releasepage
,
3886 .direct_IO
= ext4_direct_IO
,
3887 .is_partially_uptodate
= block_is_partially_uptodate
,
3888 .error_remove_page
= generic_error_remove_page
,
3891 static const struct address_space_operations ext4_da_aops
= {
3892 .readpage
= ext4_readpage
,
3893 .readpages
= ext4_readpages
,
3894 .writepage
= ext4_writepage
,
3895 .writepages
= ext4_writepages
,
3896 .write_begin
= ext4_da_write_begin
,
3897 .write_end
= ext4_da_write_end
,
3898 .set_page_dirty
= ext4_set_page_dirty
,
3900 .invalidatepage
= ext4_da_invalidatepage
,
3901 .releasepage
= ext4_releasepage
,
3902 .direct_IO
= ext4_direct_IO
,
3903 .migratepage
= buffer_migrate_page
,
3904 .is_partially_uptodate
= block_is_partially_uptodate
,
3905 .error_remove_page
= generic_error_remove_page
,
3908 void ext4_set_aops(struct inode
*inode
)
3910 switch (ext4_inode_journal_mode(inode
)) {
3911 case EXT4_INODE_ORDERED_DATA_MODE
:
3912 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3914 case EXT4_INODE_JOURNAL_DATA_MODE
:
3915 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3920 if (test_opt(inode
->i_sb
, DELALLOC
))
3921 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3923 inode
->i_mapping
->a_ops
= &ext4_aops
;
3926 static int __ext4_block_zero_page_range(handle_t
*handle
,
3927 struct address_space
*mapping
, loff_t from
, loff_t length
)
3929 ext4_fsblk_t index
= from
>> PAGE_SHIFT
;
3930 unsigned offset
= from
& (PAGE_SIZE
-1);
3931 unsigned blocksize
, pos
;
3933 struct inode
*inode
= mapping
->host
;
3934 struct buffer_head
*bh
;
3938 page
= find_or_create_page(mapping
, from
>> PAGE_SHIFT
,
3939 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
3943 blocksize
= inode
->i_sb
->s_blocksize
;
3945 iblock
= index
<< (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3947 if (!page_has_buffers(page
))
3948 create_empty_buffers(page
, blocksize
, 0);
3950 /* Find the buffer that contains "offset" */
3951 bh
= page_buffers(page
);
3953 while (offset
>= pos
) {
3954 bh
= bh
->b_this_page
;
3958 if (buffer_freed(bh
)) {
3959 BUFFER_TRACE(bh
, "freed: skip");
3962 if (!buffer_mapped(bh
)) {
3963 BUFFER_TRACE(bh
, "unmapped");
3964 ext4_get_block(inode
, iblock
, bh
, 0);
3965 /* unmapped? It's a hole - nothing to do */
3966 if (!buffer_mapped(bh
)) {
3967 BUFFER_TRACE(bh
, "still unmapped");
3972 /* Ok, it's mapped. Make sure it's up-to-date */
3973 if (PageUptodate(page
))
3974 set_buffer_uptodate(bh
);
3976 if (!buffer_uptodate(bh
)) {
3978 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
3980 /* Uhhuh. Read error. Complain and punt. */
3981 if (!buffer_uptodate(bh
))
3983 if (S_ISREG(inode
->i_mode
) &&
3984 ext4_encrypted_inode(inode
)) {
3985 /* We expect the key to be set. */
3986 BUG_ON(!fscrypt_has_encryption_key(inode
));
3987 BUG_ON(blocksize
!= PAGE_SIZE
);
3988 WARN_ON_ONCE(fscrypt_decrypt_page(page
->mapping
->host
,
3989 page
, PAGE_SIZE
, 0, page
->index
));
3992 if (ext4_should_journal_data(inode
)) {
3993 BUFFER_TRACE(bh
, "get write access");
3994 err
= ext4_journal_get_write_access(handle
, bh
);
3998 zero_user(page
, offset
, length
);
3999 BUFFER_TRACE(bh
, "zeroed end of block");
4001 if (ext4_should_journal_data(inode
)) {
4002 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
4005 mark_buffer_dirty(bh
);
4006 if (ext4_should_order_data(inode
))
4007 err
= ext4_jbd2_inode_add_write(handle
, inode
);
4017 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
4018 * starting from file offset 'from'. The range to be zero'd must
4019 * be contained with in one block. If the specified range exceeds
4020 * the end of the block it will be shortened to end of the block
4021 * that cooresponds to 'from'
4023 static int ext4_block_zero_page_range(handle_t
*handle
,
4024 struct address_space
*mapping
, loff_t from
, loff_t length
)
4026 struct inode
*inode
= mapping
->host
;
4027 unsigned offset
= from
& (PAGE_SIZE
-1);
4028 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
4029 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
4032 * correct length if it does not fall between
4033 * 'from' and the end of the block
4035 if (length
> max
|| length
< 0)
4038 if (IS_DAX(inode
)) {
4039 return iomap_zero_range(inode
, from
, length
, NULL
,
4042 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
4046 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
4047 * up to the end of the block which corresponds to `from'.
4048 * This required during truncate. We need to physically zero the tail end
4049 * of that block so it doesn't yield old data if the file is later grown.
4051 static int ext4_block_truncate_page(handle_t
*handle
,
4052 struct address_space
*mapping
, loff_t from
)
4054 unsigned offset
= from
& (PAGE_SIZE
-1);
4057 struct inode
*inode
= mapping
->host
;
4059 /* If we are processing an encrypted inode during orphan list handling */
4060 if (ext4_encrypted_inode(inode
) && !fscrypt_has_encryption_key(inode
))
4063 blocksize
= inode
->i_sb
->s_blocksize
;
4064 length
= blocksize
- (offset
& (blocksize
- 1));
4066 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
4069 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
4070 loff_t lstart
, loff_t length
)
4072 struct super_block
*sb
= inode
->i_sb
;
4073 struct address_space
*mapping
= inode
->i_mapping
;
4074 unsigned partial_start
, partial_end
;
4075 ext4_fsblk_t start
, end
;
4076 loff_t byte_end
= (lstart
+ length
- 1);
4079 partial_start
= lstart
& (sb
->s_blocksize
- 1);
4080 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
4082 start
= lstart
>> sb
->s_blocksize_bits
;
4083 end
= byte_end
>> sb
->s_blocksize_bits
;
4085 /* Handle partial zero within the single block */
4087 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
4088 err
= ext4_block_zero_page_range(handle
, mapping
,
4092 /* Handle partial zero out on the start of the range */
4093 if (partial_start
) {
4094 err
= ext4_block_zero_page_range(handle
, mapping
,
4095 lstart
, sb
->s_blocksize
);
4099 /* Handle partial zero out on the end of the range */
4100 if (partial_end
!= sb
->s_blocksize
- 1)
4101 err
= ext4_block_zero_page_range(handle
, mapping
,
4102 byte_end
- partial_end
,
4107 int ext4_can_truncate(struct inode
*inode
)
4109 if (S_ISREG(inode
->i_mode
))
4111 if (S_ISDIR(inode
->i_mode
))
4113 if (S_ISLNK(inode
->i_mode
))
4114 return !ext4_inode_is_fast_symlink(inode
);
4119 * We have to make sure i_disksize gets properly updated before we truncate
4120 * page cache due to hole punching or zero range. Otherwise i_disksize update
4121 * can get lost as it may have been postponed to submission of writeback but
4122 * that will never happen after we truncate page cache.
4124 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
4128 loff_t size
= i_size_read(inode
);
4130 WARN_ON(!inode_is_locked(inode
));
4131 if (offset
> size
|| offset
+ len
< size
)
4134 if (EXT4_I(inode
)->i_disksize
>= size
)
4137 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
4139 return PTR_ERR(handle
);
4140 ext4_update_i_disksize(inode
, size
);
4141 ext4_mark_inode_dirty(handle
, inode
);
4142 ext4_journal_stop(handle
);
4148 * ext4_punch_hole: punches a hole in a file by releasing the blocks
4149 * associated with the given offset and length
4151 * @inode: File inode
4152 * @offset: The offset where the hole will begin
4153 * @len: The length of the hole
4155 * Returns: 0 on success or negative on failure
4158 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
4160 struct super_block
*sb
= inode
->i_sb
;
4161 ext4_lblk_t first_block
, stop_block
;
4162 struct address_space
*mapping
= inode
->i_mapping
;
4163 loff_t first_block_offset
, last_block_offset
;
4165 unsigned int credits
;
4168 if (!S_ISREG(inode
->i_mode
))
4171 trace_ext4_punch_hole(inode
, offset
, length
, 0);
4174 * Write out all dirty pages to avoid race conditions
4175 * Then release them.
4177 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
4178 ret
= filemap_write_and_wait_range(mapping
, offset
,
4179 offset
+ length
- 1);
4186 /* No need to punch hole beyond i_size */
4187 if (offset
>= inode
->i_size
)
4191 * If the hole extends beyond i_size, set the hole
4192 * to end after the page that contains i_size
4194 if (offset
+ length
> inode
->i_size
) {
4195 length
= inode
->i_size
+
4196 PAGE_SIZE
- (inode
->i_size
& (PAGE_SIZE
- 1)) -
4200 if (offset
& (sb
->s_blocksize
- 1) ||
4201 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
4203 * Attach jinode to inode for jbd2 if we do any zeroing of
4206 ret
= ext4_inode_attach_jinode(inode
);
4212 /* Wait all existing dio workers, newcomers will block on i_mutex */
4213 ext4_inode_block_unlocked_dio(inode
);
4214 inode_dio_wait(inode
);
4217 * Prevent page faults from reinstantiating pages we have released from
4220 down_write(&EXT4_I(inode
)->i_mmap_sem
);
4221 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
4222 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
4224 /* Now release the pages and zero block aligned part of pages*/
4225 if (last_block_offset
> first_block_offset
) {
4226 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
4229 truncate_pagecache_range(inode
, first_block_offset
,
4233 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4234 credits
= ext4_writepage_trans_blocks(inode
);
4236 credits
= ext4_blocks_for_truncate(inode
);
4237 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4238 if (IS_ERR(handle
)) {
4239 ret
= PTR_ERR(handle
);
4240 ext4_std_error(sb
, ret
);
4244 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
4249 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4250 EXT4_BLOCK_SIZE_BITS(sb
);
4251 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4253 /* If there are blocks to remove, do it */
4254 if (stop_block
> first_block
) {
4256 down_write(&EXT4_I(inode
)->i_data_sem
);
4257 ext4_discard_preallocations(inode
);
4259 ret
= ext4_es_remove_extent(inode
, first_block
,
4260 stop_block
- first_block
);
4262 up_write(&EXT4_I(inode
)->i_data_sem
);
4266 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4267 ret
= ext4_ext_remove_space(inode
, first_block
,
4270 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
4273 up_write(&EXT4_I(inode
)->i_data_sem
);
4276 ext4_handle_sync(handle
);
4278 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4279 ext4_mark_inode_dirty(handle
, inode
);
4281 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4283 ext4_journal_stop(handle
);
4285 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4286 ext4_inode_resume_unlocked_dio(inode
);
4288 inode_unlock(inode
);
4292 int ext4_inode_attach_jinode(struct inode
*inode
)
4294 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4295 struct jbd2_inode
*jinode
;
4297 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
4300 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
4301 spin_lock(&inode
->i_lock
);
4304 spin_unlock(&inode
->i_lock
);
4307 ei
->jinode
= jinode
;
4308 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
4311 spin_unlock(&inode
->i_lock
);
4312 if (unlikely(jinode
!= NULL
))
4313 jbd2_free_inode(jinode
);
4320 * We block out ext4_get_block() block instantiations across the entire
4321 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4322 * simultaneously on behalf of the same inode.
4324 * As we work through the truncate and commit bits of it to the journal there
4325 * is one core, guiding principle: the file's tree must always be consistent on
4326 * disk. We must be able to restart the truncate after a crash.
4328 * The file's tree may be transiently inconsistent in memory (although it
4329 * probably isn't), but whenever we close off and commit a journal transaction,
4330 * the contents of (the filesystem + the journal) must be consistent and
4331 * restartable. It's pretty simple, really: bottom up, right to left (although
4332 * left-to-right works OK too).
4334 * Note that at recovery time, journal replay occurs *before* the restart of
4335 * truncate against the orphan inode list.
4337 * The committed inode has the new, desired i_size (which is the same as
4338 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4339 * that this inode's truncate did not complete and it will again call
4340 * ext4_truncate() to have another go. So there will be instantiated blocks
4341 * to the right of the truncation point in a crashed ext4 filesystem. But
4342 * that's fine - as long as they are linked from the inode, the post-crash
4343 * ext4_truncate() run will find them and release them.
4345 int ext4_truncate(struct inode
*inode
)
4347 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4348 unsigned int credits
;
4351 struct address_space
*mapping
= inode
->i_mapping
;
4354 * There is a possibility that we're either freeing the inode
4355 * or it's a completely new inode. In those cases we might not
4356 * have i_mutex locked because it's not necessary.
4358 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
4359 WARN_ON(!inode_is_locked(inode
));
4360 trace_ext4_truncate_enter(inode
);
4362 if (!ext4_can_truncate(inode
))
4365 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
4367 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4368 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4370 if (ext4_has_inline_data(inode
)) {
4373 err
= ext4_inline_data_truncate(inode
, &has_inline
);
4380 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4381 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
4382 if (ext4_inode_attach_jinode(inode
) < 0)
4386 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4387 credits
= ext4_writepage_trans_blocks(inode
);
4389 credits
= ext4_blocks_for_truncate(inode
);
4391 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4393 return PTR_ERR(handle
);
4395 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
4396 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
4399 * We add the inode to the orphan list, so that if this
4400 * truncate spans multiple transactions, and we crash, we will
4401 * resume the truncate when the filesystem recovers. It also
4402 * marks the inode dirty, to catch the new size.
4404 * Implication: the file must always be in a sane, consistent
4405 * truncatable state while each transaction commits.
4407 err
= ext4_orphan_add(handle
, inode
);
4411 down_write(&EXT4_I(inode
)->i_data_sem
);
4413 ext4_discard_preallocations(inode
);
4415 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4416 err
= ext4_ext_truncate(handle
, inode
);
4418 ext4_ind_truncate(handle
, inode
);
4420 up_write(&ei
->i_data_sem
);
4425 ext4_handle_sync(handle
);
4429 * If this was a simple ftruncate() and the file will remain alive,
4430 * then we need to clear up the orphan record which we created above.
4431 * However, if this was a real unlink then we were called by
4432 * ext4_evict_inode(), and we allow that function to clean up the
4433 * orphan info for us.
4436 ext4_orphan_del(handle
, inode
);
4438 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4439 ext4_mark_inode_dirty(handle
, inode
);
4440 ext4_journal_stop(handle
);
4442 trace_ext4_truncate_exit(inode
);
4447 * ext4_get_inode_loc returns with an extra refcount against the inode's
4448 * underlying buffer_head on success. If 'in_mem' is true, we have all
4449 * data in memory that is needed to recreate the on-disk version of this
4452 static int __ext4_get_inode_loc(struct inode
*inode
,
4453 struct ext4_iloc
*iloc
, int in_mem
)
4455 struct ext4_group_desc
*gdp
;
4456 struct buffer_head
*bh
;
4457 struct super_block
*sb
= inode
->i_sb
;
4459 int inodes_per_block
, inode_offset
;
4462 if (inode
->i_ino
< EXT4_ROOT_INO
||
4463 inode
->i_ino
> le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
))
4464 return -EFSCORRUPTED
;
4466 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4467 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4472 * Figure out the offset within the block group inode table
4474 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4475 inode_offset
= ((inode
->i_ino
- 1) %
4476 EXT4_INODES_PER_GROUP(sb
));
4477 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4478 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4480 bh
= sb_getblk(sb
, block
);
4483 if (!buffer_uptodate(bh
)) {
4487 * If the buffer has the write error flag, we have failed
4488 * to write out another inode in the same block. In this
4489 * case, we don't have to read the block because we may
4490 * read the old inode data successfully.
4492 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4493 set_buffer_uptodate(bh
);
4495 if (buffer_uptodate(bh
)) {
4496 /* someone brought it uptodate while we waited */
4502 * If we have all information of the inode in memory and this
4503 * is the only valid inode in the block, we need not read the
4507 struct buffer_head
*bitmap_bh
;
4510 start
= inode_offset
& ~(inodes_per_block
- 1);
4512 /* Is the inode bitmap in cache? */
4513 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4514 if (unlikely(!bitmap_bh
))
4518 * If the inode bitmap isn't in cache then the
4519 * optimisation may end up performing two reads instead
4520 * of one, so skip it.
4522 if (!buffer_uptodate(bitmap_bh
)) {
4526 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4527 if (i
== inode_offset
)
4529 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4533 if (i
== start
+ inodes_per_block
) {
4534 /* all other inodes are free, so skip I/O */
4535 memset(bh
->b_data
, 0, bh
->b_size
);
4536 set_buffer_uptodate(bh
);
4544 * If we need to do any I/O, try to pre-readahead extra
4545 * blocks from the inode table.
4547 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4548 ext4_fsblk_t b
, end
, table
;
4550 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4552 table
= ext4_inode_table(sb
, gdp
);
4553 /* s_inode_readahead_blks is always a power of 2 */
4554 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4558 num
= EXT4_INODES_PER_GROUP(sb
);
4559 if (ext4_has_group_desc_csum(sb
))
4560 num
-= ext4_itable_unused_count(sb
, gdp
);
4561 table
+= num
/ inodes_per_block
;
4565 sb_breadahead(sb
, b
++);
4569 * There are other valid inodes in the buffer, this inode
4570 * has in-inode xattrs, or we don't have this inode in memory.
4571 * Read the block from disk.
4573 trace_ext4_load_inode(inode
);
4575 bh
->b_end_io
= end_buffer_read_sync
;
4576 submit_bh(REQ_OP_READ
, REQ_META
| REQ_PRIO
, bh
);
4578 if (!buffer_uptodate(bh
)) {
4579 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4580 "unable to read itable block");
4590 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4592 /* We have all inode data except xattrs in memory here. */
4593 return __ext4_get_inode_loc(inode
, iloc
,
4594 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4597 void ext4_set_inode_flags(struct inode
*inode
)
4599 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4600 unsigned int new_fl
= 0;
4602 if (flags
& EXT4_SYNC_FL
)
4604 if (flags
& EXT4_APPEND_FL
)
4606 if (flags
& EXT4_IMMUTABLE_FL
)
4607 new_fl
|= S_IMMUTABLE
;
4608 if (flags
& EXT4_NOATIME_FL
)
4609 new_fl
|= S_NOATIME
;
4610 if (flags
& EXT4_DIRSYNC_FL
)
4611 new_fl
|= S_DIRSYNC
;
4612 if (test_opt(inode
->i_sb
, DAX
) && S_ISREG(inode
->i_mode
) &&
4613 !ext4_should_journal_data(inode
) && !ext4_has_inline_data(inode
) &&
4614 !ext4_encrypted_inode(inode
))
4616 inode_set_flags(inode
, new_fl
,
4617 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
);
4620 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4621 struct ext4_inode_info
*ei
)
4624 struct inode
*inode
= &(ei
->vfs_inode
);
4625 struct super_block
*sb
= inode
->i_sb
;
4627 if (ext4_has_feature_huge_file(sb
)) {
4628 /* we are using combined 48 bit field */
4629 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4630 le32_to_cpu(raw_inode
->i_blocks_lo
);
4631 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4632 /* i_blocks represent file system block size */
4633 return i_blocks
<< (inode
->i_blkbits
- 9);
4638 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4642 static inline int ext4_iget_extra_inode(struct inode
*inode
,
4643 struct ext4_inode
*raw_inode
,
4644 struct ext4_inode_info
*ei
)
4646 __le32
*magic
= (void *)raw_inode
+
4647 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4649 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
+ sizeof(__le32
) <=
4650 EXT4_INODE_SIZE(inode
->i_sb
) &&
4651 *magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4652 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4653 return ext4_find_inline_data_nolock(inode
);
4655 EXT4_I(inode
)->i_inline_off
= 0;
4659 int ext4_get_projid(struct inode
*inode
, kprojid_t
*projid
)
4661 if (!ext4_has_feature_project(inode
->i_sb
))
4663 *projid
= EXT4_I(inode
)->i_projid
;
4667 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4669 struct ext4_iloc iloc
;
4670 struct ext4_inode
*raw_inode
;
4671 struct ext4_inode_info
*ei
;
4672 struct inode
*inode
;
4673 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4681 inode
= iget_locked(sb
, ino
);
4683 return ERR_PTR(-ENOMEM
);
4684 if (!(inode
->i_state
& I_NEW
))
4690 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4693 raw_inode
= ext4_raw_inode(&iloc
);
4695 if ((ino
== EXT4_ROOT_INO
) && (raw_inode
->i_links_count
== 0)) {
4696 EXT4_ERROR_INODE(inode
, "root inode unallocated");
4697 ret
= -EFSCORRUPTED
;
4701 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4702 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4703 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4704 EXT4_INODE_SIZE(inode
->i_sb
) ||
4705 (ei
->i_extra_isize
& 3)) {
4706 EXT4_ERROR_INODE(inode
,
4707 "bad extra_isize %u (inode size %u)",
4709 EXT4_INODE_SIZE(inode
->i_sb
));
4710 ret
= -EFSCORRUPTED
;
4714 ei
->i_extra_isize
= 0;
4716 /* Precompute checksum seed for inode metadata */
4717 if (ext4_has_metadata_csum(sb
)) {
4718 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4720 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4721 __le32 gen
= raw_inode
->i_generation
;
4722 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4724 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4728 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4729 EXT4_ERROR_INODE(inode
, "checksum invalid");
4734 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4735 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4736 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4737 if (ext4_has_feature_project(sb
) &&
4738 EXT4_INODE_SIZE(sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4739 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4740 i_projid
= (projid_t
)le32_to_cpu(raw_inode
->i_projid
);
4742 i_projid
= EXT4_DEF_PROJID
;
4744 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4745 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4746 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4748 i_uid_write(inode
, i_uid
);
4749 i_gid_write(inode
, i_gid
);
4750 ei
->i_projid
= make_kprojid(&init_user_ns
, i_projid
);
4751 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4753 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4754 ei
->i_inline_off
= 0;
4755 ei
->i_dir_start_lookup
= 0;
4756 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4757 /* We now have enough fields to check if the inode was active or not.
4758 * This is needed because nfsd might try to access dead inodes
4759 * the test is that same one that e2fsck uses
4760 * NeilBrown 1999oct15
4762 if (inode
->i_nlink
== 0) {
4763 if ((inode
->i_mode
== 0 ||
4764 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4765 ino
!= EXT4_BOOT_LOADER_INO
) {
4766 /* this inode is deleted */
4770 /* The only unlinked inodes we let through here have
4771 * valid i_mode and are being read by the orphan
4772 * recovery code: that's fine, we're about to complete
4773 * the process of deleting those.
4774 * OR it is the EXT4_BOOT_LOADER_INO which is
4775 * not initialized on a new filesystem. */
4777 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4778 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4779 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4780 if (ext4_has_feature_64bit(sb
))
4782 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4783 inode
->i_size
= ext4_isize(sb
, raw_inode
);
4784 if ((size
= i_size_read(inode
)) < 0) {
4785 EXT4_ERROR_INODE(inode
, "bad i_size value: %lld", size
);
4786 ret
= -EFSCORRUPTED
;
4789 ei
->i_disksize
= inode
->i_size
;
4791 ei
->i_reserved_quota
= 0;
4793 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4794 ei
->i_block_group
= iloc
.block_group
;
4795 ei
->i_last_alloc_group
= ~0;
4797 * NOTE! The in-memory inode i_data array is in little-endian order
4798 * even on big-endian machines: we do NOT byteswap the block numbers!
4800 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4801 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4802 INIT_LIST_HEAD(&ei
->i_orphan
);
4805 * Set transaction id's of transactions that have to be committed
4806 * to finish f[data]sync. We set them to currently running transaction
4807 * as we cannot be sure that the inode or some of its metadata isn't
4808 * part of the transaction - the inode could have been reclaimed and
4809 * now it is reread from disk.
4812 transaction_t
*transaction
;
4815 read_lock(&journal
->j_state_lock
);
4816 if (journal
->j_running_transaction
)
4817 transaction
= journal
->j_running_transaction
;
4819 transaction
= journal
->j_committing_transaction
;
4821 tid
= transaction
->t_tid
;
4823 tid
= journal
->j_commit_sequence
;
4824 read_unlock(&journal
->j_state_lock
);
4825 ei
->i_sync_tid
= tid
;
4826 ei
->i_datasync_tid
= tid
;
4829 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4830 if (ei
->i_extra_isize
== 0) {
4831 /* The extra space is currently unused. Use it. */
4832 BUILD_BUG_ON(sizeof(struct ext4_inode
) & 3);
4833 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4834 EXT4_GOOD_OLD_INODE_SIZE
;
4836 ret
= ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4842 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4843 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4844 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4845 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4847 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4848 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4849 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4850 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4852 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4857 if (ei
->i_file_acl
&&
4858 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4859 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4861 ret
= -EFSCORRUPTED
;
4863 } else if (!ext4_has_inline_data(inode
)) {
4864 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4865 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4866 (S_ISLNK(inode
->i_mode
) &&
4867 !ext4_inode_is_fast_symlink(inode
))))
4868 /* Validate extent which is part of inode */
4869 ret
= ext4_ext_check_inode(inode
);
4870 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4871 (S_ISLNK(inode
->i_mode
) &&
4872 !ext4_inode_is_fast_symlink(inode
))) {
4873 /* Validate block references which are part of inode */
4874 ret
= ext4_ind_check_inode(inode
);
4880 if (S_ISREG(inode
->i_mode
)) {
4881 inode
->i_op
= &ext4_file_inode_operations
;
4882 inode
->i_fop
= &ext4_file_operations
;
4883 ext4_set_aops(inode
);
4884 } else if (S_ISDIR(inode
->i_mode
)) {
4885 inode
->i_op
= &ext4_dir_inode_operations
;
4886 inode
->i_fop
= &ext4_dir_operations
;
4887 } else if (S_ISLNK(inode
->i_mode
)) {
4888 if (ext4_encrypted_inode(inode
)) {
4889 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
4890 ext4_set_aops(inode
);
4891 } else if (ext4_inode_is_fast_symlink(inode
)) {
4892 inode
->i_link
= (char *)ei
->i_data
;
4893 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4894 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4895 sizeof(ei
->i_data
) - 1);
4897 inode
->i_op
= &ext4_symlink_inode_operations
;
4898 ext4_set_aops(inode
);
4900 inode_nohighmem(inode
);
4901 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4902 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4903 inode
->i_op
= &ext4_special_inode_operations
;
4904 if (raw_inode
->i_block
[0])
4905 init_special_inode(inode
, inode
->i_mode
,
4906 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4908 init_special_inode(inode
, inode
->i_mode
,
4909 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4910 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4911 make_bad_inode(inode
);
4913 ret
= -EFSCORRUPTED
;
4914 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4918 ext4_set_inode_flags(inode
);
4920 unlock_new_inode(inode
);
4926 return ERR_PTR(ret
);
4929 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4931 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4932 return ERR_PTR(-EFSCORRUPTED
);
4933 return ext4_iget(sb
, ino
);
4936 static int ext4_inode_blocks_set(handle_t
*handle
,
4937 struct ext4_inode
*raw_inode
,
4938 struct ext4_inode_info
*ei
)
4940 struct inode
*inode
= &(ei
->vfs_inode
);
4941 u64 i_blocks
= inode
->i_blocks
;
4942 struct super_block
*sb
= inode
->i_sb
;
4944 if (i_blocks
<= ~0U) {
4946 * i_blocks can be represented in a 32 bit variable
4947 * as multiple of 512 bytes
4949 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4950 raw_inode
->i_blocks_high
= 0;
4951 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4954 if (!ext4_has_feature_huge_file(sb
))
4957 if (i_blocks
<= 0xffffffffffffULL
) {
4959 * i_blocks can be represented in a 48 bit variable
4960 * as multiple of 512 bytes
4962 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4963 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4964 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4966 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4967 /* i_block is stored in file system block size */
4968 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4969 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4970 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4975 struct other_inode
{
4976 unsigned long orig_ino
;
4977 struct ext4_inode
*raw_inode
;
4980 static int other_inode_match(struct inode
* inode
, unsigned long ino
,
4983 struct other_inode
*oi
= (struct other_inode
*) data
;
4985 if ((inode
->i_ino
!= ino
) ||
4986 (inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4987 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
4988 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
4990 spin_lock(&inode
->i_lock
);
4991 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4992 I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) == 0) &&
4993 (inode
->i_state
& I_DIRTY_TIME
)) {
4994 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4996 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
4997 spin_unlock(&inode
->i_lock
);
4999 spin_lock(&ei
->i_raw_lock
);
5000 EXT4_INODE_SET_XTIME(i_ctime
, inode
, oi
->raw_inode
);
5001 EXT4_INODE_SET_XTIME(i_mtime
, inode
, oi
->raw_inode
);
5002 EXT4_INODE_SET_XTIME(i_atime
, inode
, oi
->raw_inode
);
5003 ext4_inode_csum_set(inode
, oi
->raw_inode
, ei
);
5004 spin_unlock(&ei
->i_raw_lock
);
5005 trace_ext4_other_inode_update_time(inode
, oi
->orig_ino
);
5008 spin_unlock(&inode
->i_lock
);
5013 * Opportunistically update the other time fields for other inodes in
5014 * the same inode table block.
5016 static void ext4_update_other_inodes_time(struct super_block
*sb
,
5017 unsigned long orig_ino
, char *buf
)
5019 struct other_inode oi
;
5021 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
5022 int inode_size
= EXT4_INODE_SIZE(sb
);
5024 oi
.orig_ino
= orig_ino
;
5026 * Calculate the first inode in the inode table block. Inode
5027 * numbers are one-based. That is, the first inode in a block
5028 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
5030 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
5031 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
5032 if (ino
== orig_ino
)
5034 oi
.raw_inode
= (struct ext4_inode
*) buf
;
5035 (void) find_inode_nowait(sb
, ino
, other_inode_match
, &oi
);
5040 * Post the struct inode info into an on-disk inode location in the
5041 * buffer-cache. This gobbles the caller's reference to the
5042 * buffer_head in the inode location struct.
5044 * The caller must have write access to iloc->bh.
5046 static int ext4_do_update_inode(handle_t
*handle
,
5047 struct inode
*inode
,
5048 struct ext4_iloc
*iloc
)
5050 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
5051 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5052 struct buffer_head
*bh
= iloc
->bh
;
5053 struct super_block
*sb
= inode
->i_sb
;
5054 int err
= 0, rc
, block
;
5055 int need_datasync
= 0, set_large_file
= 0;
5060 spin_lock(&ei
->i_raw_lock
);
5062 /* For fields not tracked in the in-memory inode,
5063 * initialise them to zero for new inodes. */
5064 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
5065 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
5067 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
5068 i_uid
= i_uid_read(inode
);
5069 i_gid
= i_gid_read(inode
);
5070 i_projid
= from_kprojid(&init_user_ns
, ei
->i_projid
);
5071 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
5072 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
5073 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
5075 * Fix up interoperability with old kernels. Otherwise, old inodes get
5076 * re-used with the upper 16 bits of the uid/gid intact
5078 if (ei
->i_dtime
&& list_empty(&ei
->i_orphan
)) {
5079 raw_inode
->i_uid_high
= 0;
5080 raw_inode
->i_gid_high
= 0;
5082 raw_inode
->i_uid_high
=
5083 cpu_to_le16(high_16_bits(i_uid
));
5084 raw_inode
->i_gid_high
=
5085 cpu_to_le16(high_16_bits(i_gid
));
5088 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
5089 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
5090 raw_inode
->i_uid_high
= 0;
5091 raw_inode
->i_gid_high
= 0;
5093 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5095 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5096 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5097 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5098 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5100 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
5102 spin_unlock(&ei
->i_raw_lock
);
5105 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5106 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
5107 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
5108 raw_inode
->i_file_acl_high
=
5109 cpu_to_le16(ei
->i_file_acl
>> 32);
5110 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5111 if (ei
->i_disksize
!= ext4_isize(inode
->i_sb
, raw_inode
)) {
5112 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5115 if (ei
->i_disksize
> 0x7fffffffULL
) {
5116 if (!ext4_has_feature_large_file(sb
) ||
5117 EXT4_SB(sb
)->s_es
->s_rev_level
==
5118 cpu_to_le32(EXT4_GOOD_OLD_REV
))
5121 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5122 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5123 if (old_valid_dev(inode
->i_rdev
)) {
5124 raw_inode
->i_block
[0] =
5125 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5126 raw_inode
->i_block
[1] = 0;
5128 raw_inode
->i_block
[0] = 0;
5129 raw_inode
->i_block
[1] =
5130 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5131 raw_inode
->i_block
[2] = 0;
5133 } else if (!ext4_has_inline_data(inode
)) {
5134 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5135 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5138 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
5139 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
5140 if (ei
->i_extra_isize
) {
5141 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5142 raw_inode
->i_version_hi
=
5143 cpu_to_le32(inode
->i_version
>> 32);
5144 raw_inode
->i_extra_isize
=
5145 cpu_to_le16(ei
->i_extra_isize
);
5149 BUG_ON(!ext4_has_feature_project(inode
->i_sb
) &&
5150 i_projid
!= EXT4_DEF_PROJID
);
5152 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
5153 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
5154 raw_inode
->i_projid
= cpu_to_le32(i_projid
);
5156 ext4_inode_csum_set(inode
, raw_inode
, ei
);
5157 spin_unlock(&ei
->i_raw_lock
);
5158 if (inode
->i_sb
->s_flags
& MS_LAZYTIME
)
5159 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
5162 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5163 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5166 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5167 if (set_large_file
) {
5168 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
5169 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
5172 ext4_update_dynamic_rev(sb
);
5173 ext4_set_feature_large_file(sb
);
5174 ext4_handle_sync(handle
);
5175 err
= ext4_handle_dirty_super(handle
, sb
);
5177 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
5180 ext4_std_error(inode
->i_sb
, err
);
5185 * ext4_write_inode()
5187 * We are called from a few places:
5189 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5190 * Here, there will be no transaction running. We wait for any running
5191 * transaction to commit.
5193 * - Within flush work (sys_sync(), kupdate and such).
5194 * We wait on commit, if told to.
5196 * - Within iput_final() -> write_inode_now()
5197 * We wait on commit, if told to.
5199 * In all cases it is actually safe for us to return without doing anything,
5200 * because the inode has been copied into a raw inode buffer in
5201 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5204 * Note that we are absolutely dependent upon all inode dirtiers doing the
5205 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5206 * which we are interested.
5208 * It would be a bug for them to not do this. The code:
5210 * mark_inode_dirty(inode)
5212 * inode->i_size = expr;
5214 * is in error because write_inode() could occur while `stuff()' is running,
5215 * and the new i_size will be lost. Plus the inode will no longer be on the
5216 * superblock's dirty inode list.
5218 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5222 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
) ||
5223 sb_rdonly(inode
->i_sb
))
5226 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5229 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5230 if (ext4_journal_current_handle()) {
5231 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5237 * No need to force transaction in WB_SYNC_NONE mode. Also
5238 * ext4_sync_fs() will force the commit after everything is
5241 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
5244 err
= jbd2_complete_transaction(EXT4_SB(inode
->i_sb
)->s_journal
,
5245 EXT4_I(inode
)->i_sync_tid
);
5247 struct ext4_iloc iloc
;
5249 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5253 * sync(2) will flush the whole buffer cache. No need to do
5254 * it here separately for each inode.
5256 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
5257 sync_dirty_buffer(iloc
.bh
);
5258 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5259 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
5260 "IO error syncing inode");
5269 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5270 * buffers that are attached to a page stradding i_size and are undergoing
5271 * commit. In that case we have to wait for commit to finish and try again.
5273 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
5277 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
5278 tid_t commit_tid
= 0;
5281 offset
= inode
->i_size
& (PAGE_SIZE
- 1);
5283 * All buffers in the last page remain valid? Then there's nothing to
5284 * do. We do the check mainly to optimize the common PAGE_SIZE ==
5287 if (offset
> PAGE_SIZE
- i_blocksize(inode
))
5290 page
= find_lock_page(inode
->i_mapping
,
5291 inode
->i_size
>> PAGE_SHIFT
);
5294 ret
= __ext4_journalled_invalidatepage(page
, offset
,
5295 PAGE_SIZE
- offset
);
5301 read_lock(&journal
->j_state_lock
);
5302 if (journal
->j_committing_transaction
)
5303 commit_tid
= journal
->j_committing_transaction
->t_tid
;
5304 read_unlock(&journal
->j_state_lock
);
5306 jbd2_log_wait_commit(journal
, commit_tid
);
5313 * Called from notify_change.
5315 * We want to trap VFS attempts to truncate the file as soon as
5316 * possible. In particular, we want to make sure that when the VFS
5317 * shrinks i_size, we put the inode on the orphan list and modify
5318 * i_disksize immediately, so that during the subsequent flushing of
5319 * dirty pages and freeing of disk blocks, we can guarantee that any
5320 * commit will leave the blocks being flushed in an unused state on
5321 * disk. (On recovery, the inode will get truncated and the blocks will
5322 * be freed, so we have a strong guarantee that no future commit will
5323 * leave these blocks visible to the user.)
5325 * Another thing we have to assure is that if we are in ordered mode
5326 * and inode is still attached to the committing transaction, we must
5327 * we start writeout of all the dirty pages which are being truncated.
5328 * This way we are sure that all the data written in the previous
5329 * transaction are already on disk (truncate waits for pages under
5332 * Called with inode->i_mutex down.
5334 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5336 struct inode
*inode
= d_inode(dentry
);
5339 const unsigned int ia_valid
= attr
->ia_valid
;
5341 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5344 error
= setattr_prepare(dentry
, attr
);
5348 if (is_quota_modification(inode
, attr
)) {
5349 error
= dquot_initialize(inode
);
5353 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
5354 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
5357 /* (user+group)*(old+new) structure, inode write (sb,
5358 * inode block, ? - but truncate inode update has it) */
5359 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
5360 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
5361 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
5362 if (IS_ERR(handle
)) {
5363 error
= PTR_ERR(handle
);
5367 /* dquot_transfer() calls back ext4_get_inode_usage() which
5368 * counts xattr inode references.
5370 down_read(&EXT4_I(inode
)->xattr_sem
);
5371 error
= dquot_transfer(inode
, attr
);
5372 up_read(&EXT4_I(inode
)->xattr_sem
);
5375 ext4_journal_stop(handle
);
5378 /* Update corresponding info in inode so that everything is in
5379 * one transaction */
5380 if (attr
->ia_valid
& ATTR_UID
)
5381 inode
->i_uid
= attr
->ia_uid
;
5382 if (attr
->ia_valid
& ATTR_GID
)
5383 inode
->i_gid
= attr
->ia_gid
;
5384 error
= ext4_mark_inode_dirty(handle
, inode
);
5385 ext4_journal_stop(handle
);
5388 if (attr
->ia_valid
& ATTR_SIZE
) {
5390 loff_t oldsize
= inode
->i_size
;
5391 int shrink
= (attr
->ia_size
<= inode
->i_size
);
5393 if (ext4_encrypted_inode(inode
)) {
5394 error
= fscrypt_get_encryption_info(inode
);
5397 if (!fscrypt_has_encryption_key(inode
))
5401 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5402 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5404 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5407 if (!S_ISREG(inode
->i_mode
))
5410 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
5411 inode_inc_iversion(inode
);
5413 if (ext4_should_order_data(inode
) &&
5414 (attr
->ia_size
< inode
->i_size
)) {
5415 error
= ext4_begin_ordered_truncate(inode
,
5420 if (attr
->ia_size
!= inode
->i_size
) {
5421 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
5422 if (IS_ERR(handle
)) {
5423 error
= PTR_ERR(handle
);
5426 if (ext4_handle_valid(handle
) && shrink
) {
5427 error
= ext4_orphan_add(handle
, inode
);
5431 * Update c/mtime on truncate up, ext4_truncate() will
5432 * update c/mtime in shrink case below
5435 inode
->i_mtime
= current_time(inode
);
5436 inode
->i_ctime
= inode
->i_mtime
;
5438 down_write(&EXT4_I(inode
)->i_data_sem
);
5439 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5440 rc
= ext4_mark_inode_dirty(handle
, inode
);
5444 * We have to update i_size under i_data_sem together
5445 * with i_disksize to avoid races with writeback code
5446 * running ext4_wb_update_i_disksize().
5449 i_size_write(inode
, attr
->ia_size
);
5450 up_write(&EXT4_I(inode
)->i_data_sem
);
5451 ext4_journal_stop(handle
);
5453 if (orphan
&& inode
->i_nlink
)
5454 ext4_orphan_del(NULL
, inode
);
5459 pagecache_isize_extended(inode
, oldsize
, inode
->i_size
);
5462 * Blocks are going to be removed from the inode. Wait
5463 * for dio in flight. Temporarily disable
5464 * dioread_nolock to prevent livelock.
5467 if (!ext4_should_journal_data(inode
)) {
5468 ext4_inode_block_unlocked_dio(inode
);
5469 inode_dio_wait(inode
);
5470 ext4_inode_resume_unlocked_dio(inode
);
5472 ext4_wait_for_tail_page_commit(inode
);
5474 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5476 * Truncate pagecache after we've waited for commit
5477 * in data=journal mode to make pages freeable.
5479 truncate_pagecache(inode
, inode
->i_size
);
5481 rc
= ext4_truncate(inode
);
5485 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5489 setattr_copy(inode
, attr
);
5490 mark_inode_dirty(inode
);
5494 * If the call to ext4_truncate failed to get a transaction handle at
5495 * all, we need to clean up the in-core orphan list manually.
5497 if (orphan
&& inode
->i_nlink
)
5498 ext4_orphan_del(NULL
, inode
);
5500 if (!error
&& (ia_valid
& ATTR_MODE
))
5501 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
5504 ext4_std_error(inode
->i_sb
, error
);
5510 int ext4_getattr(const struct path
*path
, struct kstat
*stat
,
5511 u32 request_mask
, unsigned int query_flags
)
5513 struct inode
*inode
= d_inode(path
->dentry
);
5514 struct ext4_inode
*raw_inode
;
5515 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5518 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_crtime
)) {
5519 stat
->result_mask
|= STATX_BTIME
;
5520 stat
->btime
.tv_sec
= ei
->i_crtime
.tv_sec
;
5521 stat
->btime
.tv_nsec
= ei
->i_crtime
.tv_nsec
;
5524 flags
= ei
->i_flags
& EXT4_FL_USER_VISIBLE
;
5525 if (flags
& EXT4_APPEND_FL
)
5526 stat
->attributes
|= STATX_ATTR_APPEND
;
5527 if (flags
& EXT4_COMPR_FL
)
5528 stat
->attributes
|= STATX_ATTR_COMPRESSED
;
5529 if (flags
& EXT4_ENCRYPT_FL
)
5530 stat
->attributes
|= STATX_ATTR_ENCRYPTED
;
5531 if (flags
& EXT4_IMMUTABLE_FL
)
5532 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
5533 if (flags
& EXT4_NODUMP_FL
)
5534 stat
->attributes
|= STATX_ATTR_NODUMP
;
5536 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
5537 STATX_ATTR_COMPRESSED
|
5538 STATX_ATTR_ENCRYPTED
|
5539 STATX_ATTR_IMMUTABLE
|
5542 generic_fillattr(inode
, stat
);
5546 int ext4_file_getattr(const struct path
*path
, struct kstat
*stat
,
5547 u32 request_mask
, unsigned int query_flags
)
5549 struct inode
*inode
= d_inode(path
->dentry
);
5550 u64 delalloc_blocks
;
5552 ext4_getattr(path
, stat
, request_mask
, query_flags
);
5555 * If there is inline data in the inode, the inode will normally not
5556 * have data blocks allocated (it may have an external xattr block).
5557 * Report at least one sector for such files, so tools like tar, rsync,
5558 * others don't incorrectly think the file is completely sparse.
5560 if (unlikely(ext4_has_inline_data(inode
)))
5561 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5564 * We can't update i_blocks if the block allocation is delayed
5565 * otherwise in the case of system crash before the real block
5566 * allocation is done, we will have i_blocks inconsistent with
5567 * on-disk file blocks.
5568 * We always keep i_blocks updated together with real
5569 * allocation. But to not confuse with user, stat
5570 * will return the blocks that include the delayed allocation
5571 * blocks for this file.
5573 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5574 EXT4_I(inode
)->i_reserved_data_blocks
);
5575 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5579 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5582 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5583 return ext4_ind_trans_blocks(inode
, lblocks
);
5584 return ext4_ext_index_trans_blocks(inode
, pextents
);
5588 * Account for index blocks, block groups bitmaps and block group
5589 * descriptor blocks if modify datablocks and index blocks
5590 * worse case, the indexs blocks spread over different block groups
5592 * If datablocks are discontiguous, they are possible to spread over
5593 * different block groups too. If they are contiguous, with flexbg,
5594 * they could still across block group boundary.
5596 * Also account for superblock, inode, quota and xattr blocks
5598 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5601 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5607 * How many index blocks need to touch to map @lblocks logical blocks
5608 * to @pextents physical extents?
5610 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5615 * Now let's see how many group bitmaps and group descriptors need
5618 groups
= idxblocks
+ pextents
;
5620 if (groups
> ngroups
)
5622 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5623 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5625 /* bitmaps and block group descriptor blocks */
5626 ret
+= groups
+ gdpblocks
;
5628 /* Blocks for super block, inode, quota and xattr blocks */
5629 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5635 * Calculate the total number of credits to reserve to fit
5636 * the modification of a single pages into a single transaction,
5637 * which may include multiple chunks of block allocations.
5639 * This could be called via ext4_write_begin()
5641 * We need to consider the worse case, when
5642 * one new block per extent.
5644 int ext4_writepage_trans_blocks(struct inode
*inode
)
5646 int bpp
= ext4_journal_blocks_per_page(inode
);
5649 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5651 /* Account for data blocks for journalled mode */
5652 if (ext4_should_journal_data(inode
))
5658 * Calculate the journal credits for a chunk of data modification.
5660 * This is called from DIO, fallocate or whoever calling
5661 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5663 * journal buffers for data blocks are not included here, as DIO
5664 * and fallocate do no need to journal data buffers.
5666 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5668 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5672 * The caller must have previously called ext4_reserve_inode_write().
5673 * Give this, we know that the caller already has write access to iloc->bh.
5675 int ext4_mark_iloc_dirty(handle_t
*handle
,
5676 struct inode
*inode
, struct ext4_iloc
*iloc
)
5680 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
)))) {
5684 if (IS_I_VERSION(inode
))
5685 inode_inc_iversion(inode
);
5687 /* the do_update_inode consumes one bh->b_count */
5690 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5691 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5697 * On success, We end up with an outstanding reference count against
5698 * iloc->bh. This _must_ be cleaned up later.
5702 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5703 struct ext4_iloc
*iloc
)
5707 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5710 err
= ext4_get_inode_loc(inode
, iloc
);
5712 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5713 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5719 ext4_std_error(inode
->i_sb
, err
);
5723 static int __ext4_expand_extra_isize(struct inode
*inode
,
5724 unsigned int new_extra_isize
,
5725 struct ext4_iloc
*iloc
,
5726 handle_t
*handle
, int *no_expand
)
5728 struct ext4_inode
*raw_inode
;
5729 struct ext4_xattr_ibody_header
*header
;
5732 raw_inode
= ext4_raw_inode(iloc
);
5734 header
= IHDR(inode
, raw_inode
);
5736 /* No extended attributes present */
5737 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5738 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5739 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
+
5740 EXT4_I(inode
)->i_extra_isize
, 0,
5741 new_extra_isize
- EXT4_I(inode
)->i_extra_isize
);
5742 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5746 /* try to expand with EAs present */
5747 error
= ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5751 * Inode size expansion failed; don't try again
5760 * Expand an inode by new_extra_isize bytes.
5761 * Returns 0 on success or negative error number on failure.
5763 static int ext4_try_to_expand_extra_isize(struct inode
*inode
,
5764 unsigned int new_extra_isize
,
5765 struct ext4_iloc iloc
,
5771 if (ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
))
5775 * In nojournal mode, we can immediately attempt to expand
5776 * the inode. When journaled, we first need to obtain extra
5777 * buffer credits since we may write into the EA block
5778 * with this same handle. If journal_extend fails, then it will
5779 * only result in a minor loss of functionality for that inode.
5780 * If this is felt to be critical, then e2fsck should be run to
5781 * force a large enough s_min_extra_isize.
5783 if (ext4_handle_valid(handle
) &&
5784 jbd2_journal_extend(handle
,
5785 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
)) != 0)
5788 if (ext4_write_trylock_xattr(inode
, &no_expand
) == 0)
5791 error
= __ext4_expand_extra_isize(inode
, new_extra_isize
, &iloc
,
5792 handle
, &no_expand
);
5793 ext4_write_unlock_xattr(inode
, &no_expand
);
5798 int ext4_expand_extra_isize(struct inode
*inode
,
5799 unsigned int new_extra_isize
,
5800 struct ext4_iloc
*iloc
)
5806 if (ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5811 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
,
5812 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
));
5813 if (IS_ERR(handle
)) {
5814 error
= PTR_ERR(handle
);
5819 ext4_write_lock_xattr(inode
, &no_expand
);
5821 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5822 error
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5828 error
= __ext4_expand_extra_isize(inode
, new_extra_isize
, iloc
,
5829 handle
, &no_expand
);
5831 rc
= ext4_mark_iloc_dirty(handle
, inode
, iloc
);
5835 ext4_write_unlock_xattr(inode
, &no_expand
);
5837 ext4_journal_stop(handle
);
5842 * What we do here is to mark the in-core inode as clean with respect to inode
5843 * dirtiness (it may still be data-dirty).
5844 * This means that the in-core inode may be reaped by prune_icache
5845 * without having to perform any I/O. This is a very good thing,
5846 * because *any* task may call prune_icache - even ones which
5847 * have a transaction open against a different journal.
5849 * Is this cheating? Not really. Sure, we haven't written the
5850 * inode out, but prune_icache isn't a user-visible syncing function.
5851 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5852 * we start and wait on commits.
5854 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5856 struct ext4_iloc iloc
;
5857 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5861 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5862 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5866 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
)
5867 ext4_try_to_expand_extra_isize(inode
, sbi
->s_want_extra_isize
,
5870 return ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5874 * ext4_dirty_inode() is called from __mark_inode_dirty()
5876 * We're really interested in the case where a file is being extended.
5877 * i_size has been changed by generic_commit_write() and we thus need
5878 * to include the updated inode in the current transaction.
5880 * Also, dquot_alloc_block() will always dirty the inode when blocks
5881 * are allocated to the file.
5883 * If the inode is marked synchronous, we don't honour that here - doing
5884 * so would cause a commit on atime updates, which we don't bother doing.
5885 * We handle synchronous inodes at the highest possible level.
5887 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5888 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5889 * to copy into the on-disk inode structure are the timestamp files.
5891 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5895 if (flags
== I_DIRTY_TIME
)
5897 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5901 ext4_mark_inode_dirty(handle
, inode
);
5903 ext4_journal_stop(handle
);
5910 * Bind an inode's backing buffer_head into this transaction, to prevent
5911 * it from being flushed to disk early. Unlike
5912 * ext4_reserve_inode_write, this leaves behind no bh reference and
5913 * returns no iloc structure, so the caller needs to repeat the iloc
5914 * lookup to mark the inode dirty later.
5916 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5918 struct ext4_iloc iloc
;
5922 err
= ext4_get_inode_loc(inode
, &iloc
);
5924 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5925 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5927 err
= ext4_handle_dirty_metadata(handle
,
5933 ext4_std_error(inode
->i_sb
, err
);
5938 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5943 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5946 * We have to be very careful here: changing a data block's
5947 * journaling status dynamically is dangerous. If we write a
5948 * data block to the journal, change the status and then delete
5949 * that block, we risk forgetting to revoke the old log record
5950 * from the journal and so a subsequent replay can corrupt data.
5951 * So, first we make sure that the journal is empty and that
5952 * nobody is changing anything.
5955 journal
= EXT4_JOURNAL(inode
);
5958 if (is_journal_aborted(journal
))
5961 /* Wait for all existing dio workers */
5962 ext4_inode_block_unlocked_dio(inode
);
5963 inode_dio_wait(inode
);
5966 * Before flushing the journal and switching inode's aops, we have
5967 * to flush all dirty data the inode has. There can be outstanding
5968 * delayed allocations, there can be unwritten extents created by
5969 * fallocate or buffered writes in dioread_nolock mode covered by
5970 * dirty data which can be converted only after flushing the dirty
5971 * data (and journalled aops don't know how to handle these cases).
5974 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5975 err
= filemap_write_and_wait(inode
->i_mapping
);
5977 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5978 ext4_inode_resume_unlocked_dio(inode
);
5983 percpu_down_write(&sbi
->s_journal_flag_rwsem
);
5984 jbd2_journal_lock_updates(journal
);
5987 * OK, there are no updates running now, and all cached data is
5988 * synced to disk. We are now in a completely consistent state
5989 * which doesn't have anything in the journal, and we know that
5990 * no filesystem updates are running, so it is safe to modify
5991 * the inode's in-core data-journaling state flag now.
5995 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5997 err
= jbd2_journal_flush(journal
);
5999 jbd2_journal_unlock_updates(journal
);
6000 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
6001 ext4_inode_resume_unlocked_dio(inode
);
6004 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
6006 ext4_set_aops(inode
);
6008 jbd2_journal_unlock_updates(journal
);
6009 percpu_up_write(&sbi
->s_journal_flag_rwsem
);
6012 up_write(&EXT4_I(inode
)->i_mmap_sem
);
6013 ext4_inode_resume_unlocked_dio(inode
);
6015 /* Finally we can mark the inode as dirty. */
6017 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
6019 return PTR_ERR(handle
);
6021 err
= ext4_mark_inode_dirty(handle
, inode
);
6022 ext4_handle_sync(handle
);
6023 ext4_journal_stop(handle
);
6024 ext4_std_error(inode
->i_sb
, err
);
6029 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
6031 return !buffer_mapped(bh
);
6034 int ext4_page_mkwrite(struct vm_fault
*vmf
)
6036 struct vm_area_struct
*vma
= vmf
->vma
;
6037 struct page
*page
= vmf
->page
;
6041 struct file
*file
= vma
->vm_file
;
6042 struct inode
*inode
= file_inode(file
);
6043 struct address_space
*mapping
= inode
->i_mapping
;
6045 get_block_t
*get_block
;
6048 sb_start_pagefault(inode
->i_sb
);
6049 file_update_time(vma
->vm_file
);
6051 down_read(&EXT4_I(inode
)->i_mmap_sem
);
6053 ret
= ext4_convert_inline_data(inode
);
6057 /* Delalloc case is easy... */
6058 if (test_opt(inode
->i_sb
, DELALLOC
) &&
6059 !ext4_should_journal_data(inode
) &&
6060 !ext4_nonda_switch(inode
->i_sb
)) {
6062 ret
= block_page_mkwrite(vma
, vmf
,
6063 ext4_da_get_block_prep
);
6064 } while (ret
== -ENOSPC
&&
6065 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
6070 size
= i_size_read(inode
);
6071 /* Page got truncated from under us? */
6072 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
6074 ret
= VM_FAULT_NOPAGE
;
6078 if (page
->index
== size
>> PAGE_SHIFT
)
6079 len
= size
& ~PAGE_MASK
;
6083 * Return if we have all the buffers mapped. This avoids the need to do
6084 * journal_start/journal_stop which can block and take a long time
6086 if (page_has_buffers(page
)) {
6087 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
6089 ext4_bh_unmapped
)) {
6090 /* Wait so that we don't change page under IO */
6091 wait_for_stable_page(page
);
6092 ret
= VM_FAULT_LOCKED
;
6097 /* OK, we need to fill the hole... */
6098 if (ext4_should_dioread_nolock(inode
))
6099 get_block
= ext4_get_block_unwritten
;
6101 get_block
= ext4_get_block
;
6103 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
6104 ext4_writepage_trans_blocks(inode
));
6105 if (IS_ERR(handle
)) {
6106 ret
= VM_FAULT_SIGBUS
;
6109 ret
= block_page_mkwrite(vma
, vmf
, get_block
);
6110 if (!ret
&& ext4_should_journal_data(inode
)) {
6111 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
6112 PAGE_SIZE
, NULL
, do_journal_get_write_access
)) {
6114 ret
= VM_FAULT_SIGBUS
;
6115 ext4_journal_stop(handle
);
6118 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
6120 ext4_journal_stop(handle
);
6121 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
6124 ret
= block_page_mkwrite_return(ret
);
6126 up_read(&EXT4_I(inode
)->i_mmap_sem
);
6127 sb_end_pagefault(inode
->i_sb
);
6131 int ext4_filemap_fault(struct vm_fault
*vmf
)
6133 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
6136 down_read(&EXT4_I(inode
)->i_mmap_sem
);
6137 err
= filemap_fault(vmf
);
6138 up_read(&EXT4_I(inode
)->i_mmap_sem
);
6144 * Find the first extent at or after @lblk in an inode that is not a hole.
6145 * Search for @map_len blocks at most. The extent is returned in @result.
6147 * The function returns 1 if we found an extent. The function returns 0 in
6148 * case there is no extent at or after @lblk and in that case also sets
6149 * @result->es_len to 0. In case of error, the error code is returned.
6151 int ext4_get_next_extent(struct inode
*inode
, ext4_lblk_t lblk
,
6152 unsigned int map_len
, struct extent_status
*result
)
6154 struct ext4_map_blocks map
;
6155 struct extent_status es
= {};
6159 map
.m_len
= map_len
;
6162 * For non-extent based files this loop may iterate several times since
6163 * we do not determine full hole size.
6165 while (map
.m_len
> 0) {
6166 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
6169 /* There's extent covering m_lblk? Just return it. */
6173 ext4_es_store_pblock(result
, map
.m_pblk
);
6174 result
->es_lblk
= map
.m_lblk
;
6175 result
->es_len
= map
.m_len
;
6176 if (map
.m_flags
& EXT4_MAP_UNWRITTEN
)
6177 status
= EXTENT_STATUS_UNWRITTEN
;
6179 status
= EXTENT_STATUS_WRITTEN
;
6180 ext4_es_store_status(result
, status
);
6183 ext4_es_find_delayed_extent_range(inode
, map
.m_lblk
,
6184 map
.m_lblk
+ map
.m_len
- 1,
6186 /* Is delalloc data before next block in extent tree? */
6187 if (es
.es_len
&& es
.es_lblk
< map
.m_lblk
+ map
.m_len
) {
6188 ext4_lblk_t offset
= 0;
6190 if (es
.es_lblk
< lblk
)
6191 offset
= lblk
- es
.es_lblk
;
6192 result
->es_lblk
= es
.es_lblk
+ offset
;
6193 ext4_es_store_pblock(result
,
6194 ext4_es_pblock(&es
) + offset
);
6195 result
->es_len
= es
.es_len
- offset
;
6196 ext4_es_store_status(result
, ext4_es_status(&es
));
6200 /* There's a hole at m_lblk, advance us after it */
6201 map
.m_lblk
+= map
.m_len
;
6202 map_len
-= map
.m_len
;
6203 map
.m_len
= map_len
;