1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/inode.c
5 * Copyright (C) 1992, 1993, 1994, 1995
6 * Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
12 * linux/fs/minix/inode.c
14 * Copyright (C) 1991, 1992 Linus Torvalds
16 * 64-bit file support on 64-bit platforms by Jakub Jelinek
17 * (jj@sunsite.ms.mff.cuni.cz)
19 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
23 #include <linux/time.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/dax.h>
27 #include <linux/quotaops.h>
28 #include <linux/string.h>
29 #include <linux/buffer_head.h>
30 #include <linux/writeback.h>
31 #include <linux/pagevec.h>
32 #include <linux/mpage.h>
33 #include <linux/namei.h>
34 #include <linux/uio.h>
35 #include <linux/bio.h>
36 #include <linux/workqueue.h>
37 #include <linux/kernel.h>
38 #include <linux/printk.h>
39 #include <linux/slab.h>
40 #include <linux/bitops.h>
41 #include <linux/iomap.h>
42 #include <linux/iversion.h>
44 #include "ext4_jbd2.h"
49 #include <trace/events/ext4.h>
51 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
52 struct ext4_inode_info
*ei
)
54 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
57 int offset
= offsetof(struct ext4_inode
, i_checksum_lo
);
58 unsigned int csum_size
= sizeof(dummy_csum
);
60 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
, offset
);
61 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
, csum_size
);
63 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
64 EXT4_GOOD_OLD_INODE_SIZE
- offset
);
66 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
67 offset
= offsetof(struct ext4_inode
, i_checksum_hi
);
68 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+
69 EXT4_GOOD_OLD_INODE_SIZE
,
70 offset
- EXT4_GOOD_OLD_INODE_SIZE
);
71 if (EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
72 csum
= ext4_chksum(sbi
, csum
, (__u8
*)&dummy_csum
,
76 csum
= ext4_chksum(sbi
, csum
, (__u8
*)raw
+ offset
,
77 EXT4_INODE_SIZE(inode
->i_sb
) - offset
);
83 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
84 struct ext4_inode_info
*ei
)
86 __u32 provided
, calculated
;
88 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
89 cpu_to_le32(EXT4_OS_LINUX
) ||
90 !ext4_has_metadata_csum(inode
->i_sb
))
93 provided
= le16_to_cpu(raw
->i_checksum_lo
);
94 calculated
= ext4_inode_csum(inode
, raw
, ei
);
95 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
96 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
97 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
101 return provided
== calculated
;
104 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
105 struct ext4_inode_info
*ei
)
109 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
110 cpu_to_le32(EXT4_OS_LINUX
) ||
111 !ext4_has_metadata_csum(inode
->i_sb
))
114 csum
= ext4_inode_csum(inode
, raw
, ei
);
115 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
116 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
117 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
118 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
121 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
124 trace_ext4_begin_ordered_truncate(inode
, new_size
);
126 * If jinode is zero, then we never opened the file for
127 * writing, so there's no need to call
128 * jbd2_journal_begin_ordered_truncate() since there's no
129 * outstanding writes we need to flush.
131 if (!EXT4_I(inode
)->jinode
)
133 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
134 EXT4_I(inode
)->jinode
,
138 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
139 unsigned int length
);
140 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
141 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
142 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
146 * Test whether an inode is a fast symlink.
147 * A fast symlink has its symlink data stored in ext4_inode_info->i_data.
149 int ext4_inode_is_fast_symlink(struct inode
*inode
)
151 if (!(EXT4_I(inode
)->i_flags
& EXT4_EA_INODE_FL
)) {
152 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
153 EXT4_CLUSTER_SIZE(inode
->i_sb
) >> 9 : 0;
155 if (ext4_has_inline_data(inode
))
158 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
160 return S_ISLNK(inode
->i_mode
) && inode
->i_size
&&
161 (inode
->i_size
< EXT4_N_BLOCKS
* 4);
165 * Called at the last iput() if i_nlink is zero.
167 void ext4_evict_inode(struct inode
*inode
)
172 * Credits for final inode cleanup and freeing:
173 * sb + inode (ext4_orphan_del()), block bitmap, group descriptor
174 * (xattr block freeing), bitmap, group descriptor (inode freeing)
176 int extra_credits
= 6;
177 struct ext4_xattr_inode_array
*ea_inode_array
= NULL
;
179 trace_ext4_evict_inode(inode
);
181 if (inode
->i_nlink
) {
183 * When journalling data dirty buffers are tracked only in the
184 * journal. So although mm thinks everything is clean and
185 * ready for reaping the inode might still have some pages to
186 * write in the running transaction or waiting to be
187 * checkpointed. Thus calling jbd2_journal_invalidatepage()
188 * (via truncate_inode_pages()) to discard these buffers can
189 * cause data loss. Also even if we did not discard these
190 * buffers, we would have no way to find them after the inode
191 * is reaped and thus user could see stale data if he tries to
192 * read them before the transaction is checkpointed. So be
193 * careful and force everything to disk here... We use
194 * ei->i_datasync_tid to store the newest transaction
195 * containing inode's data.
197 * Note that directories do not have this problem because they
198 * don't use page cache.
200 if (inode
->i_ino
!= EXT4_JOURNAL_INO
&&
201 ext4_should_journal_data(inode
) &&
202 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
)) &&
203 inode
->i_data
.nrpages
) {
204 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
205 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
207 jbd2_complete_transaction(journal
, commit_tid
);
208 filemap_write_and_wait(&inode
->i_data
);
210 truncate_inode_pages_final(&inode
->i_data
);
215 if (is_bad_inode(inode
))
217 dquot_initialize(inode
);
219 if (ext4_should_order_data(inode
))
220 ext4_begin_ordered_truncate(inode
, 0);
221 truncate_inode_pages_final(&inode
->i_data
);
224 * For inodes with journalled data, transaction commit could have
225 * dirtied the inode. Flush worker is ignoring it because of I_FREEING
226 * flag but we still need to remove the inode from the writeback lists.
228 if (!list_empty_careful(&inode
->i_io_list
)) {
229 WARN_ON_ONCE(!ext4_should_journal_data(inode
));
230 inode_io_list_del(inode
);
234 * Protect us against freezing - iput() caller didn't have to have any
235 * protection against it
237 sb_start_intwrite(inode
->i_sb
);
239 if (!IS_NOQUOTA(inode
))
240 extra_credits
+= EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
);
243 * Block bitmap, group descriptor, and inode are accounted in both
244 * ext4_blocks_for_truncate() and extra_credits. So subtract 3.
246 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
247 ext4_blocks_for_truncate(inode
) + extra_credits
- 3);
248 if (IS_ERR(handle
)) {
249 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
251 * If we're going to skip the normal cleanup, we still need to
252 * make sure that the in-core orphan linked list is properly
255 ext4_orphan_del(NULL
, inode
);
256 sb_end_intwrite(inode
->i_sb
);
261 ext4_handle_sync(handle
);
264 * Set inode->i_size to 0 before calling ext4_truncate(). We need
265 * special handling of symlinks here because i_size is used to
266 * determine whether ext4_inode_info->i_data contains symlink data or
267 * block mappings. Setting i_size to 0 will remove its fast symlink
268 * status. Erase i_data so that it becomes a valid empty block map.
270 if (ext4_inode_is_fast_symlink(inode
))
271 memset(EXT4_I(inode
)->i_data
, 0, sizeof(EXT4_I(inode
)->i_data
));
273 err
= ext4_mark_inode_dirty(handle
, inode
);
275 ext4_warning(inode
->i_sb
,
276 "couldn't mark inode dirty (err %d)", err
);
279 if (inode
->i_blocks
) {
280 err
= ext4_truncate(inode
);
282 ext4_error_err(inode
->i_sb
, -err
,
283 "couldn't truncate inode %lu (err %d)",
289 /* Remove xattr references. */
290 err
= ext4_xattr_delete_inode(handle
, inode
, &ea_inode_array
,
293 ext4_warning(inode
->i_sb
, "xattr delete (err %d)", err
);
295 ext4_journal_stop(handle
);
296 ext4_orphan_del(NULL
, inode
);
297 sb_end_intwrite(inode
->i_sb
);
298 ext4_xattr_inode_array_free(ea_inode_array
);
303 * Kill off the orphan record which ext4_truncate created.
304 * AKPM: I think this can be inside the above `if'.
305 * Note that ext4_orphan_del() has to be able to cope with the
306 * deletion of a non-existent orphan - this is because we don't
307 * know if ext4_truncate() actually created an orphan record.
308 * (Well, we could do this if we need to, but heck - it works)
310 ext4_orphan_del(handle
, inode
);
311 EXT4_I(inode
)->i_dtime
= (__u32
)ktime_get_real_seconds();
314 * One subtle ordering requirement: if anything has gone wrong
315 * (transaction abort, IO errors, whatever), then we can still
316 * do these next steps (the fs will already have been marked as
317 * having errors), but we can't free the inode if the mark_dirty
320 if (ext4_mark_inode_dirty(handle
, inode
))
321 /* If that failed, just do the required in-core inode clear. */
322 ext4_clear_inode(inode
);
324 ext4_free_inode(handle
, inode
);
325 ext4_journal_stop(handle
);
326 sb_end_intwrite(inode
->i_sb
);
327 ext4_xattr_inode_array_free(ea_inode_array
);
330 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
334 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
336 return &EXT4_I(inode
)->i_reserved_quota
;
341 * Called with i_data_sem down, which is important since we can call
342 * ext4_discard_preallocations() from here.
344 void ext4_da_update_reserve_space(struct inode
*inode
,
345 int used
, int quota_claim
)
347 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
348 struct ext4_inode_info
*ei
= EXT4_I(inode
);
350 spin_lock(&ei
->i_block_reservation_lock
);
351 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
352 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
353 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
354 "with only %d reserved data blocks",
355 __func__
, inode
->i_ino
, used
,
356 ei
->i_reserved_data_blocks
);
358 used
= ei
->i_reserved_data_blocks
;
361 /* Update per-inode reservations */
362 ei
->i_reserved_data_blocks
-= used
;
363 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, used
);
365 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
367 /* Update quota subsystem for data blocks */
369 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
372 * We did fallocate with an offset that is already delayed
373 * allocated. So on delayed allocated writeback we should
374 * not re-claim the quota for fallocated blocks.
376 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
380 * If we have done all the pending block allocations and if
381 * there aren't any writers on the inode, we can discard the
382 * inode's preallocations.
384 if ((ei
->i_reserved_data_blocks
== 0) &&
385 !inode_is_open_for_write(inode
))
386 ext4_discard_preallocations(inode
, 0);
389 static int __check_block_validity(struct inode
*inode
, const char *func
,
391 struct ext4_map_blocks
*map
)
393 if (ext4_has_feature_journal(inode
->i_sb
) &&
395 le32_to_cpu(EXT4_SB(inode
->i_sb
)->s_es
->s_journal_inum
)))
397 if (!ext4_inode_block_valid(inode
, map
->m_pblk
, map
->m_len
)) {
398 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
399 "lblock %lu mapped to illegal pblock %llu "
400 "(length %d)", (unsigned long) map
->m_lblk
,
401 map
->m_pblk
, map
->m_len
);
402 return -EFSCORRUPTED
;
407 int ext4_issue_zeroout(struct inode
*inode
, ext4_lblk_t lblk
, ext4_fsblk_t pblk
,
412 if (IS_ENCRYPTED(inode
) && S_ISREG(inode
->i_mode
))
413 return fscrypt_zeroout_range(inode
, lblk
, pblk
, len
);
415 ret
= sb_issue_zeroout(inode
->i_sb
, pblk
, len
, GFP_NOFS
);
422 #define check_block_validity(inode, map) \
423 __check_block_validity((inode), __func__, __LINE__, (map))
425 #ifdef ES_AGGRESSIVE_TEST
426 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
428 struct ext4_map_blocks
*es_map
,
429 struct ext4_map_blocks
*map
,
436 * There is a race window that the result is not the same.
437 * e.g. xfstests #223 when dioread_nolock enables. The reason
438 * is that we lookup a block mapping in extent status tree with
439 * out taking i_data_sem. So at the time the unwritten extent
440 * could be converted.
442 down_read(&EXT4_I(inode
)->i_data_sem
);
443 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
444 retval
= ext4_ext_map_blocks(handle
, inode
, map
, 0);
446 retval
= ext4_ind_map_blocks(handle
, inode
, map
, 0);
448 up_read((&EXT4_I(inode
)->i_data_sem
));
451 * We don't check m_len because extent will be collpased in status
452 * tree. So the m_len might not equal.
454 if (es_map
->m_lblk
!= map
->m_lblk
||
455 es_map
->m_flags
!= map
->m_flags
||
456 es_map
->m_pblk
!= map
->m_pblk
) {
457 printk("ES cache assertion failed for inode: %lu "
458 "es_cached ex [%d/%d/%llu/%x] != "
459 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
460 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
461 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
462 map
->m_len
, map
->m_pblk
, map
->m_flags
,
466 #endif /* ES_AGGRESSIVE_TEST */
469 * The ext4_map_blocks() function tries to look up the requested blocks,
470 * and returns if the blocks are already mapped.
472 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
473 * and store the allocated blocks in the result buffer head and mark it
476 * If file type is extents based, it will call ext4_ext_map_blocks(),
477 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
480 * On success, it returns the number of blocks being mapped or allocated. if
481 * create==0 and the blocks are pre-allocated and unwritten, the resulting @map
482 * is marked as unwritten. If the create == 1, it will mark @map as mapped.
484 * It returns 0 if plain look up failed (blocks have not been allocated), in
485 * that case, @map is returned as unmapped but we still do fill map->m_len to
486 * indicate the length of a hole starting at map->m_lblk.
488 * It returns the error in case of allocation failure.
490 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
491 struct ext4_map_blocks
*map
, int flags
)
493 struct extent_status es
;
496 #ifdef ES_AGGRESSIVE_TEST
497 struct ext4_map_blocks orig_map
;
499 memcpy(&orig_map
, map
, sizeof(*map
));
503 ext_debug(inode
, "flag 0x%x, max_blocks %u, logical block %lu\n",
504 flags
, map
->m_len
, (unsigned long) map
->m_lblk
);
507 * ext4_map_blocks returns an int, and m_len is an unsigned int
509 if (unlikely(map
->m_len
> INT_MAX
))
510 map
->m_len
= INT_MAX
;
512 /* We can handle the block number less than EXT_MAX_BLOCKS */
513 if (unlikely(map
->m_lblk
>= EXT_MAX_BLOCKS
))
514 return -EFSCORRUPTED
;
516 /* Lookup extent status tree firstly */
517 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, NULL
, &es
)) {
518 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
519 map
->m_pblk
= ext4_es_pblock(&es
) +
520 map
->m_lblk
- es
.es_lblk
;
521 map
->m_flags
|= ext4_es_is_written(&es
) ?
522 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
523 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
524 if (retval
> map
->m_len
)
527 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
529 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
530 if (retval
> map
->m_len
)
537 #ifdef ES_AGGRESSIVE_TEST
538 ext4_map_blocks_es_recheck(handle
, inode
, map
,
545 * Try to see if we can get the block without requesting a new
548 down_read(&EXT4_I(inode
)->i_data_sem
);
549 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
550 retval
= ext4_ext_map_blocks(handle
, inode
, map
, 0);
552 retval
= ext4_ind_map_blocks(handle
, inode
, map
, 0);
557 if (unlikely(retval
!= map
->m_len
)) {
558 ext4_warning(inode
->i_sb
,
559 "ES len assertion failed for inode "
560 "%lu: retval %d != map->m_len %d",
561 inode
->i_ino
, retval
, map
->m_len
);
565 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
566 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
567 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
568 !(status
& EXTENT_STATUS_WRITTEN
) &&
569 ext4_es_scan_range(inode
, &ext4_es_is_delayed
, map
->m_lblk
,
570 map
->m_lblk
+ map
->m_len
- 1))
571 status
|= EXTENT_STATUS_DELAYED
;
572 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
573 map
->m_len
, map
->m_pblk
, status
);
577 up_read((&EXT4_I(inode
)->i_data_sem
));
580 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
581 ret
= check_block_validity(inode
, map
);
586 /* If it is only a block(s) look up */
587 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
591 * Returns if the blocks have already allocated
593 * Note that if blocks have been preallocated
594 * ext4_ext_get_block() returns the create = 0
595 * with buffer head unmapped.
597 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
599 * If we need to convert extent to unwritten
600 * we continue and do the actual work in
601 * ext4_ext_map_blocks()
603 if (!(flags
& EXT4_GET_BLOCKS_CONVERT_UNWRITTEN
))
607 * Here we clear m_flags because after allocating an new extent,
608 * it will be set again.
610 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
613 * New blocks allocate and/or writing to unwritten extent
614 * will possibly result in updating i_data, so we take
615 * the write lock of i_data_sem, and call get_block()
616 * with create == 1 flag.
618 down_write(&EXT4_I(inode
)->i_data_sem
);
621 * We need to check for EXT4 here because migrate
622 * could have changed the inode type in between
624 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
625 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
627 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
629 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
631 * We allocated new blocks which will result in
632 * i_data's format changing. Force the migrate
633 * to fail by clearing migrate flags
635 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
639 * Update reserved blocks/metadata blocks after successful
640 * block allocation which had been deferred till now. We don't
641 * support fallocate for non extent files. So we can update
642 * reserve space here.
645 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
646 ext4_da_update_reserve_space(inode
, retval
, 1);
652 if (unlikely(retval
!= map
->m_len
)) {
653 ext4_warning(inode
->i_sb
,
654 "ES len assertion failed for inode "
655 "%lu: retval %d != map->m_len %d",
656 inode
->i_ino
, retval
, map
->m_len
);
661 * We have to zeroout blocks before inserting them into extent
662 * status tree. Otherwise someone could look them up there and
663 * use them before they are really zeroed. We also have to
664 * unmap metadata before zeroing as otherwise writeback can
665 * overwrite zeros with stale data from block device.
667 if (flags
& EXT4_GET_BLOCKS_ZERO
&&
668 map
->m_flags
& EXT4_MAP_MAPPED
&&
669 map
->m_flags
& EXT4_MAP_NEW
) {
670 ret
= ext4_issue_zeroout(inode
, map
->m_lblk
,
671 map
->m_pblk
, map
->m_len
);
679 * If the extent has been zeroed out, we don't need to update
680 * extent status tree.
682 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
683 ext4_es_lookup_extent(inode
, map
->m_lblk
, NULL
, &es
)) {
684 if (ext4_es_is_written(&es
))
687 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
688 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
689 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
690 !(status
& EXTENT_STATUS_WRITTEN
) &&
691 ext4_es_scan_range(inode
, &ext4_es_is_delayed
, map
->m_lblk
,
692 map
->m_lblk
+ map
->m_len
- 1))
693 status
|= EXTENT_STATUS_DELAYED
;
694 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
695 map
->m_pblk
, status
);
703 up_write((&EXT4_I(inode
)->i_data_sem
));
704 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
705 ret
= check_block_validity(inode
, map
);
710 * Inodes with freshly allocated blocks where contents will be
711 * visible after transaction commit must be on transaction's
714 if (map
->m_flags
& EXT4_MAP_NEW
&&
715 !(map
->m_flags
& EXT4_MAP_UNWRITTEN
) &&
716 !(flags
& EXT4_GET_BLOCKS_ZERO
) &&
717 !ext4_is_quota_file(inode
) &&
718 ext4_should_order_data(inode
)) {
720 (loff_t
)map
->m_lblk
<< inode
->i_blkbits
;
721 loff_t length
= (loff_t
)map
->m_len
<< inode
->i_blkbits
;
723 if (flags
& EXT4_GET_BLOCKS_IO_SUBMIT
)
724 ret
= ext4_jbd2_inode_add_wait(handle
, inode
,
727 ret
= ext4_jbd2_inode_add_write(handle
, inode
,
735 ext_debug(inode
, "failed with err %d\n", retval
);
740 * Update EXT4_MAP_FLAGS in bh->b_state. For buffer heads attached to pages
741 * we have to be careful as someone else may be manipulating b_state as well.
743 static void ext4_update_bh_state(struct buffer_head
*bh
, unsigned long flags
)
745 unsigned long old_state
;
746 unsigned long new_state
;
748 flags
&= EXT4_MAP_FLAGS
;
750 /* Dummy buffer_head? Set non-atomically. */
752 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | flags
;
756 * Someone else may be modifying b_state. Be careful! This is ugly but
757 * once we get rid of using bh as a container for mapping information
758 * to pass to / from get_block functions, this can go away.
761 old_state
= READ_ONCE(bh
->b_state
);
762 new_state
= (old_state
& ~EXT4_MAP_FLAGS
) | flags
;
764 cmpxchg(&bh
->b_state
, old_state
, new_state
) != old_state
));
767 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
768 struct buffer_head
*bh
, int flags
)
770 struct ext4_map_blocks map
;
773 if (ext4_has_inline_data(inode
))
777 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
779 ret
= ext4_map_blocks(ext4_journal_current_handle(), inode
, &map
,
782 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
783 ext4_update_bh_state(bh
, map
.m_flags
);
784 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
786 } else if (ret
== 0) {
787 /* hole case, need to fill in bh->b_size */
788 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
793 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
794 struct buffer_head
*bh
, int create
)
796 return _ext4_get_block(inode
, iblock
, bh
,
797 create
? EXT4_GET_BLOCKS_CREATE
: 0);
801 * Get block function used when preparing for buffered write if we require
802 * creating an unwritten extent if blocks haven't been allocated. The extent
803 * will be converted to written after the IO is complete.
805 int ext4_get_block_unwritten(struct inode
*inode
, sector_t iblock
,
806 struct buffer_head
*bh_result
, int create
)
808 ext4_debug("ext4_get_block_unwritten: inode %lu, create flag %d\n",
809 inode
->i_ino
, create
);
810 return _ext4_get_block(inode
, iblock
, bh_result
,
811 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
814 /* Maximum number of blocks we map for direct IO at once. */
815 #define DIO_MAX_BLOCKS 4096
818 * `handle' can be NULL if create is zero
820 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
821 ext4_lblk_t block
, int map_flags
)
823 struct ext4_map_blocks map
;
824 struct buffer_head
*bh
;
825 int create
= map_flags
& EXT4_GET_BLOCKS_CREATE
;
828 J_ASSERT(handle
!= NULL
|| create
== 0);
832 err
= ext4_map_blocks(handle
, inode
, &map
, map_flags
);
835 return create
? ERR_PTR(-ENOSPC
) : NULL
;
839 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
841 return ERR_PTR(-ENOMEM
);
842 if (map
.m_flags
& EXT4_MAP_NEW
) {
843 J_ASSERT(create
!= 0);
844 J_ASSERT(handle
!= NULL
);
847 * Now that we do not always journal data, we should
848 * keep in mind whether this should always journal the
849 * new buffer as metadata. For now, regular file
850 * writes use ext4_get_block instead, so it's not a
854 BUFFER_TRACE(bh
, "call get_create_access");
855 err
= ext4_journal_get_create_access(handle
, bh
);
860 if (!buffer_uptodate(bh
)) {
861 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
862 set_buffer_uptodate(bh
);
865 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
866 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
870 BUFFER_TRACE(bh
, "not a new buffer");
877 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
878 ext4_lblk_t block
, int map_flags
)
880 struct buffer_head
*bh
;
882 bh
= ext4_getblk(handle
, inode
, block
, map_flags
);
885 if (!bh
|| ext4_buffer_uptodate(bh
))
887 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1, &bh
);
889 if (buffer_uptodate(bh
))
892 return ERR_PTR(-EIO
);
895 /* Read a contiguous batch of blocks. */
896 int ext4_bread_batch(struct inode
*inode
, ext4_lblk_t block
, int bh_count
,
897 bool wait
, struct buffer_head
**bhs
)
901 for (i
= 0; i
< bh_count
; i
++) {
902 bhs
[i
] = ext4_getblk(NULL
, inode
, block
+ i
, 0 /* map_flags */);
903 if (IS_ERR(bhs
[i
])) {
904 err
= PTR_ERR(bhs
[i
]);
910 for (i
= 0; i
< bh_count
; i
++)
911 /* Note that NULL bhs[i] is valid because of holes. */
912 if (bhs
[i
] && !ext4_buffer_uptodate(bhs
[i
]))
913 ll_rw_block(REQ_OP_READ
, REQ_META
| REQ_PRIO
, 1,
919 for (i
= 0; i
< bh_count
; i
++)
921 wait_on_buffer(bhs
[i
]);
923 for (i
= 0; i
< bh_count
; i
++) {
924 if (bhs
[i
] && !buffer_uptodate(bhs
[i
])) {
932 for (i
= 0; i
< bh_count
; i
++) {
939 int ext4_walk_page_buffers(handle_t
*handle
,
940 struct buffer_head
*head
,
944 int (*fn
)(handle_t
*handle
,
945 struct buffer_head
*bh
))
947 struct buffer_head
*bh
;
948 unsigned block_start
, block_end
;
949 unsigned blocksize
= head
->b_size
;
951 struct buffer_head
*next
;
953 for (bh
= head
, block_start
= 0;
954 ret
== 0 && (bh
!= head
|| !block_start
);
955 block_start
= block_end
, bh
= next
) {
956 next
= bh
->b_this_page
;
957 block_end
= block_start
+ blocksize
;
958 if (block_end
<= from
|| block_start
>= to
) {
959 if (partial
&& !buffer_uptodate(bh
))
963 err
= (*fn
)(handle
, bh
);
971 * To preserve ordering, it is essential that the hole instantiation and
972 * the data write be encapsulated in a single transaction. We cannot
973 * close off a transaction and start a new one between the ext4_get_block()
974 * and the commit_write(). So doing the jbd2_journal_start at the start of
975 * prepare_write() is the right place.
977 * Also, this function can nest inside ext4_writepage(). In that case, we
978 * *know* that ext4_writepage() has generated enough buffer credits to do the
979 * whole page. So we won't block on the journal in that case, which is good,
980 * because the caller may be PF_MEMALLOC.
982 * By accident, ext4 can be reentered when a transaction is open via
983 * quota file writes. If we were to commit the transaction while thus
984 * reentered, there can be a deadlock - we would be holding a quota
985 * lock, and the commit would never complete if another thread had a
986 * transaction open and was blocking on the quota lock - a ranking
989 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
990 * will _not_ run commit under these circumstances because handle->h_ref
991 * is elevated. We'll still have enough credits for the tiny quotafile
994 int do_journal_get_write_access(handle_t
*handle
,
995 struct buffer_head
*bh
)
997 int dirty
= buffer_dirty(bh
);
1000 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1003 * __block_write_begin() could have dirtied some buffers. Clean
1004 * the dirty bit as jbd2_journal_get_write_access() could complain
1005 * otherwise about fs integrity issues. Setting of the dirty bit
1006 * by __block_write_begin() isn't a real problem here as we clear
1007 * the bit before releasing a page lock and thus writeback cannot
1008 * ever write the buffer.
1011 clear_buffer_dirty(bh
);
1012 BUFFER_TRACE(bh
, "get write access");
1013 ret
= ext4_journal_get_write_access(handle
, bh
);
1015 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1019 #ifdef CONFIG_FS_ENCRYPTION
1020 static int ext4_block_write_begin(struct page
*page
, loff_t pos
, unsigned len
,
1021 get_block_t
*get_block
)
1023 unsigned from
= pos
& (PAGE_SIZE
- 1);
1024 unsigned to
= from
+ len
;
1025 struct inode
*inode
= page
->mapping
->host
;
1026 unsigned block_start
, block_end
;
1029 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
1031 struct buffer_head
*bh
, *head
, *wait
[2];
1035 BUG_ON(!PageLocked(page
));
1036 BUG_ON(from
> PAGE_SIZE
);
1037 BUG_ON(to
> PAGE_SIZE
);
1040 if (!page_has_buffers(page
))
1041 create_empty_buffers(page
, blocksize
, 0);
1042 head
= page_buffers(page
);
1043 bbits
= ilog2(blocksize
);
1044 block
= (sector_t
)page
->index
<< (PAGE_SHIFT
- bbits
);
1046 for (bh
= head
, block_start
= 0; bh
!= head
|| !block_start
;
1047 block
++, block_start
= block_end
, bh
= bh
->b_this_page
) {
1048 block_end
= block_start
+ blocksize
;
1049 if (block_end
<= from
|| block_start
>= to
) {
1050 if (PageUptodate(page
)) {
1051 if (!buffer_uptodate(bh
))
1052 set_buffer_uptodate(bh
);
1057 clear_buffer_new(bh
);
1058 if (!buffer_mapped(bh
)) {
1059 WARN_ON(bh
->b_size
!= blocksize
);
1060 err
= get_block(inode
, block
, bh
, 1);
1063 if (buffer_new(bh
)) {
1064 if (PageUptodate(page
)) {
1065 clear_buffer_new(bh
);
1066 set_buffer_uptodate(bh
);
1067 mark_buffer_dirty(bh
);
1070 if (block_end
> to
|| block_start
< from
)
1071 zero_user_segments(page
, to
, block_end
,
1076 if (PageUptodate(page
)) {
1077 if (!buffer_uptodate(bh
))
1078 set_buffer_uptodate(bh
);
1081 if (!buffer_uptodate(bh
) && !buffer_delay(bh
) &&
1082 !buffer_unwritten(bh
) &&
1083 (block_start
< from
|| block_end
> to
)) {
1084 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
1085 wait
[nr_wait
++] = bh
;
1089 * If we issued read requests, let them complete.
1091 for (i
= 0; i
< nr_wait
; i
++) {
1092 wait_on_buffer(wait
[i
]);
1093 if (!buffer_uptodate(wait
[i
]))
1096 if (unlikely(err
)) {
1097 page_zero_new_buffers(page
, from
, to
);
1098 } else if (IS_ENCRYPTED(inode
) && S_ISREG(inode
->i_mode
)) {
1099 for (i
= 0; i
< nr_wait
; i
++) {
1102 err2
= fscrypt_decrypt_pagecache_blocks(page
, blocksize
,
1103 bh_offset(wait
[i
]));
1105 clear_buffer_uptodate(wait
[i
]);
1115 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
1116 loff_t pos
, unsigned len
, unsigned flags
,
1117 struct page
**pagep
, void **fsdata
)
1119 struct inode
*inode
= mapping
->host
;
1120 int ret
, needed_blocks
;
1127 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
1130 trace_ext4_write_begin(inode
, pos
, len
, flags
);
1132 * Reserve one block more for addition to orphan list in case
1133 * we allocate blocks but write fails for some reason
1135 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1136 index
= pos
>> PAGE_SHIFT
;
1137 from
= pos
& (PAGE_SIZE
- 1);
1140 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1141 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1150 * grab_cache_page_write_begin() can take a long time if the
1151 * system is thrashing due to memory pressure, or if the page
1152 * is being written back. So grab it first before we start
1153 * the transaction handle. This also allows us to allocate
1154 * the page (if needed) without using GFP_NOFS.
1157 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1163 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1164 if (IS_ERR(handle
)) {
1166 return PTR_ERR(handle
);
1170 if (page
->mapping
!= mapping
) {
1171 /* The page got truncated from under us */
1174 ext4_journal_stop(handle
);
1177 /* In case writeback began while the page was unlocked */
1178 wait_for_stable_page(page
);
1180 #ifdef CONFIG_FS_ENCRYPTION
1181 if (ext4_should_dioread_nolock(inode
))
1182 ret
= ext4_block_write_begin(page
, pos
, len
,
1183 ext4_get_block_unwritten
);
1185 ret
= ext4_block_write_begin(page
, pos
, len
,
1188 if (ext4_should_dioread_nolock(inode
))
1189 ret
= __block_write_begin(page
, pos
, len
,
1190 ext4_get_block_unwritten
);
1192 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1194 if (!ret
&& ext4_should_journal_data(inode
)) {
1195 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1197 do_journal_get_write_access
);
1201 bool extended
= (pos
+ len
> inode
->i_size
) &&
1202 !ext4_verity_in_progress(inode
);
1206 * __block_write_begin may have instantiated a few blocks
1207 * outside i_size. Trim these off again. Don't need
1208 * i_size_read because we hold i_mutex.
1210 * Add inode to orphan list in case we crash before
1213 if (extended
&& ext4_can_truncate(inode
))
1214 ext4_orphan_add(handle
, inode
);
1216 ext4_journal_stop(handle
);
1218 ext4_truncate_failed_write(inode
);
1220 * If truncate failed early the inode might
1221 * still be on the orphan list; we need to
1222 * make sure the inode is removed from the
1223 * orphan list in that case.
1226 ext4_orphan_del(NULL
, inode
);
1229 if (ret
== -ENOSPC
&&
1230 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1239 /* For write_end() in data=journal mode */
1240 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1243 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1245 set_buffer_uptodate(bh
);
1246 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1247 clear_buffer_meta(bh
);
1248 clear_buffer_prio(bh
);
1253 * We need to pick up the new inode size which generic_commit_write gave us
1254 * `file' can be NULL - eg, when called from page_symlink().
1256 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1257 * buffers are managed internally.
1259 static int ext4_write_end(struct file
*file
,
1260 struct address_space
*mapping
,
1261 loff_t pos
, unsigned len
, unsigned copied
,
1262 struct page
*page
, void *fsdata
)
1264 handle_t
*handle
= ext4_journal_current_handle();
1265 struct inode
*inode
= mapping
->host
;
1266 loff_t old_size
= inode
->i_size
;
1268 int i_size_changed
= 0;
1269 int inline_data
= ext4_has_inline_data(inode
);
1270 bool verity
= ext4_verity_in_progress(inode
);
1272 trace_ext4_write_end(inode
, pos
, len
, copied
);
1274 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1283 copied
= block_write_end(file
, mapping
, pos
,
1284 len
, copied
, page
, fsdata
);
1286 * it's important to update i_size while still holding page lock:
1287 * page writeout could otherwise come in and zero beyond i_size.
1289 * If FS_IOC_ENABLE_VERITY is running on this inode, then Merkle tree
1290 * blocks are being written past EOF, so skip the i_size update.
1293 i_size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1297 if (old_size
< pos
&& !verity
)
1298 pagecache_isize_extended(inode
, old_size
, pos
);
1300 * Don't mark the inode dirty under page lock. First, it unnecessarily
1301 * makes the holding time of page lock longer. Second, it forces lock
1302 * ordering of page lock and transaction start for journaling
1305 if (i_size_changed
|| inline_data
)
1306 ret
= ext4_mark_inode_dirty(handle
, inode
);
1308 if (pos
+ len
> inode
->i_size
&& !verity
&& ext4_can_truncate(inode
))
1309 /* if we have allocated more blocks and copied
1310 * less. We will have blocks allocated outside
1311 * inode->i_size. So truncate them
1313 ext4_orphan_add(handle
, inode
);
1315 ret2
= ext4_journal_stop(handle
);
1319 if (pos
+ len
> inode
->i_size
&& !verity
) {
1320 ext4_truncate_failed_write(inode
);
1322 * If truncate failed early the inode might still be
1323 * on the orphan list; we need to make sure the inode
1324 * is removed from the orphan list in that case.
1327 ext4_orphan_del(NULL
, inode
);
1330 return ret
? ret
: copied
;
1334 * This is a private version of page_zero_new_buffers() which doesn't
1335 * set the buffer to be dirty, since in data=journalled mode we need
1336 * to call ext4_handle_dirty_metadata() instead.
1338 static void ext4_journalled_zero_new_buffers(handle_t
*handle
,
1340 unsigned from
, unsigned to
)
1342 unsigned int block_start
= 0, block_end
;
1343 struct buffer_head
*head
, *bh
;
1345 bh
= head
= page_buffers(page
);
1347 block_end
= block_start
+ bh
->b_size
;
1348 if (buffer_new(bh
)) {
1349 if (block_end
> from
&& block_start
< to
) {
1350 if (!PageUptodate(page
)) {
1351 unsigned start
, size
;
1353 start
= max(from
, block_start
);
1354 size
= min(to
, block_end
) - start
;
1356 zero_user(page
, start
, size
);
1357 write_end_fn(handle
, bh
);
1359 clear_buffer_new(bh
);
1362 block_start
= block_end
;
1363 bh
= bh
->b_this_page
;
1364 } while (bh
!= head
);
1367 static int ext4_journalled_write_end(struct file
*file
,
1368 struct address_space
*mapping
,
1369 loff_t pos
, unsigned len
, unsigned copied
,
1370 struct page
*page
, void *fsdata
)
1372 handle_t
*handle
= ext4_journal_current_handle();
1373 struct inode
*inode
= mapping
->host
;
1374 loff_t old_size
= inode
->i_size
;
1378 int size_changed
= 0;
1379 int inline_data
= ext4_has_inline_data(inode
);
1380 bool verity
= ext4_verity_in_progress(inode
);
1382 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1383 from
= pos
& (PAGE_SIZE
- 1);
1386 BUG_ON(!ext4_handle_valid(handle
));
1389 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1397 } else if (unlikely(copied
< len
) && !PageUptodate(page
)) {
1399 ext4_journalled_zero_new_buffers(handle
, page
, from
, to
);
1401 if (unlikely(copied
< len
))
1402 ext4_journalled_zero_new_buffers(handle
, page
,
1404 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1405 from
+ copied
, &partial
,
1408 SetPageUptodate(page
);
1411 size_changed
= ext4_update_inode_size(inode
, pos
+ copied
);
1412 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1413 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1417 if (old_size
< pos
&& !verity
)
1418 pagecache_isize_extended(inode
, old_size
, pos
);
1420 if (size_changed
|| inline_data
) {
1421 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1426 if (pos
+ len
> inode
->i_size
&& !verity
&& ext4_can_truncate(inode
))
1427 /* if we have allocated more blocks and copied
1428 * less. We will have blocks allocated outside
1429 * inode->i_size. So truncate them
1431 ext4_orphan_add(handle
, inode
);
1434 ret2
= ext4_journal_stop(handle
);
1437 if (pos
+ len
> inode
->i_size
&& !verity
) {
1438 ext4_truncate_failed_write(inode
);
1440 * If truncate failed early the inode might still be
1441 * on the orphan list; we need to make sure the inode
1442 * is removed from the orphan list in that case.
1445 ext4_orphan_del(NULL
, inode
);
1448 return ret
? ret
: copied
;
1452 * Reserve space for a single cluster
1454 static int ext4_da_reserve_space(struct inode
*inode
)
1456 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1457 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1461 * We will charge metadata quota at writeout time; this saves
1462 * us from metadata over-estimation, though we may go over by
1463 * a small amount in the end. Here we just reserve for data.
1465 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1469 spin_lock(&ei
->i_block_reservation_lock
);
1470 if (ext4_claim_free_clusters(sbi
, 1, 0)) {
1471 spin_unlock(&ei
->i_block_reservation_lock
);
1472 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1475 ei
->i_reserved_data_blocks
++;
1476 trace_ext4_da_reserve_space(inode
);
1477 spin_unlock(&ei
->i_block_reservation_lock
);
1479 return 0; /* success */
1482 void ext4_da_release_space(struct inode
*inode
, int to_free
)
1484 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1485 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1488 return; /* Nothing to release, exit */
1490 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1492 trace_ext4_da_release_space(inode
, to_free
);
1493 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1495 * if there aren't enough reserved blocks, then the
1496 * counter is messed up somewhere. Since this
1497 * function is called from invalidate page, it's
1498 * harmless to return without any action.
1500 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1501 "ino %lu, to_free %d with only %d reserved "
1502 "data blocks", inode
->i_ino
, to_free
,
1503 ei
->i_reserved_data_blocks
);
1505 to_free
= ei
->i_reserved_data_blocks
;
1507 ei
->i_reserved_data_blocks
-= to_free
;
1509 /* update fs dirty data blocks counter */
1510 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1512 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1514 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1518 * Delayed allocation stuff
1521 struct mpage_da_data
{
1522 struct inode
*inode
;
1523 struct writeback_control
*wbc
;
1525 pgoff_t first_page
; /* The first page to write */
1526 pgoff_t next_page
; /* Current page to examine */
1527 pgoff_t last_page
; /* Last page to examine */
1529 * Extent to map - this can be after first_page because that can be
1530 * fully mapped. We somewhat abuse m_flags to store whether the extent
1531 * is delalloc or unwritten.
1533 struct ext4_map_blocks map
;
1534 struct ext4_io_submit io_submit
; /* IO submission data */
1535 unsigned int do_map
:1;
1536 unsigned int scanned_until_end
:1;
1539 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1544 struct pagevec pvec
;
1545 struct inode
*inode
= mpd
->inode
;
1546 struct address_space
*mapping
= inode
->i_mapping
;
1548 /* This is necessary when next_page == 0. */
1549 if (mpd
->first_page
>= mpd
->next_page
)
1552 mpd
->scanned_until_end
= 0;
1553 index
= mpd
->first_page
;
1554 end
= mpd
->next_page
- 1;
1556 ext4_lblk_t start
, last
;
1557 start
= index
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1558 last
= end
<< (PAGE_SHIFT
- inode
->i_blkbits
);
1559 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1562 pagevec_init(&pvec
);
1563 while (index
<= end
) {
1564 nr_pages
= pagevec_lookup_range(&pvec
, mapping
, &index
, end
);
1567 for (i
= 0; i
< nr_pages
; i
++) {
1568 struct page
*page
= pvec
.pages
[i
];
1570 BUG_ON(!PageLocked(page
));
1571 BUG_ON(PageWriteback(page
));
1573 if (page_mapped(page
))
1574 clear_page_dirty_for_io(page
);
1575 block_invalidatepage(page
, 0, PAGE_SIZE
);
1576 ClearPageUptodate(page
);
1580 pagevec_release(&pvec
);
1584 static void ext4_print_free_blocks(struct inode
*inode
)
1586 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1587 struct super_block
*sb
= inode
->i_sb
;
1588 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1590 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1591 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1592 ext4_count_free_clusters(sb
)));
1593 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1594 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1595 (long long) EXT4_C2B(EXT4_SB(sb
),
1596 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1597 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1598 (long long) EXT4_C2B(EXT4_SB(sb
),
1599 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1600 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1601 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1602 ei
->i_reserved_data_blocks
);
1606 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1608 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1612 * ext4_insert_delayed_block - adds a delayed block to the extents status
1613 * tree, incrementing the reserved cluster/block
1614 * count or making a pending reservation
1617 * @inode - file containing the newly added block
1618 * @lblk - logical block to be added
1620 * Returns 0 on success, negative error code on failure.
1622 static int ext4_insert_delayed_block(struct inode
*inode
, ext4_lblk_t lblk
)
1624 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1626 bool allocated
= false;
1629 * If the cluster containing lblk is shared with a delayed,
1630 * written, or unwritten extent in a bigalloc file system, it's
1631 * already been accounted for and does not need to be reserved.
1632 * A pending reservation must be made for the cluster if it's
1633 * shared with a written or unwritten extent and doesn't already
1634 * have one. Written and unwritten extents can be purged from the
1635 * extents status tree if the system is under memory pressure, so
1636 * it's necessary to examine the extent tree if a search of the
1637 * extents status tree doesn't get a match.
1639 if (sbi
->s_cluster_ratio
== 1) {
1640 ret
= ext4_da_reserve_space(inode
);
1641 if (ret
!= 0) /* ENOSPC */
1643 } else { /* bigalloc */
1644 if (!ext4_es_scan_clu(inode
, &ext4_es_is_delonly
, lblk
)) {
1645 if (!ext4_es_scan_clu(inode
,
1646 &ext4_es_is_mapped
, lblk
)) {
1647 ret
= ext4_clu_mapped(inode
,
1648 EXT4_B2C(sbi
, lblk
));
1652 ret
= ext4_da_reserve_space(inode
);
1653 if (ret
!= 0) /* ENOSPC */
1664 ret
= ext4_es_insert_delayed_block(inode
, lblk
, allocated
);
1671 * This function is grabs code from the very beginning of
1672 * ext4_map_blocks, but assumes that the caller is from delayed write
1673 * time. This function looks up the requested blocks and sets the
1674 * buffer delay bit under the protection of i_data_sem.
1676 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1677 struct ext4_map_blocks
*map
,
1678 struct buffer_head
*bh
)
1680 struct extent_status es
;
1682 sector_t invalid_block
= ~((sector_t
) 0xffff);
1683 #ifdef ES_AGGRESSIVE_TEST
1684 struct ext4_map_blocks orig_map
;
1686 memcpy(&orig_map
, map
, sizeof(*map
));
1689 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1693 ext_debug(inode
, "max_blocks %u, logical block %lu\n", map
->m_len
,
1694 (unsigned long) map
->m_lblk
);
1696 /* Lookup extent status tree firstly */
1697 if (ext4_es_lookup_extent(inode
, iblock
, NULL
, &es
)) {
1698 if (ext4_es_is_hole(&es
)) {
1700 down_read(&EXT4_I(inode
)->i_data_sem
);
1705 * Delayed extent could be allocated by fallocate.
1706 * So we need to check it.
1708 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1709 map_bh(bh
, inode
->i_sb
, invalid_block
);
1711 set_buffer_delay(bh
);
1715 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1716 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1717 if (retval
> map
->m_len
)
1718 retval
= map
->m_len
;
1719 map
->m_len
= retval
;
1720 if (ext4_es_is_written(&es
))
1721 map
->m_flags
|= EXT4_MAP_MAPPED
;
1722 else if (ext4_es_is_unwritten(&es
))
1723 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1727 #ifdef ES_AGGRESSIVE_TEST
1728 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1734 * Try to see if we can get the block without requesting a new
1735 * file system block.
1737 down_read(&EXT4_I(inode
)->i_data_sem
);
1738 if (ext4_has_inline_data(inode
))
1740 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1741 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1743 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1750 * XXX: __block_prepare_write() unmaps passed block,
1754 ret
= ext4_insert_delayed_block(inode
, map
->m_lblk
);
1760 map_bh(bh
, inode
->i_sb
, invalid_block
);
1762 set_buffer_delay(bh
);
1763 } else if (retval
> 0) {
1765 unsigned int status
;
1767 if (unlikely(retval
!= map
->m_len
)) {
1768 ext4_warning(inode
->i_sb
,
1769 "ES len assertion failed for inode "
1770 "%lu: retval %d != map->m_len %d",
1771 inode
->i_ino
, retval
, map
->m_len
);
1775 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1776 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1777 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1778 map
->m_pblk
, status
);
1784 up_read((&EXT4_I(inode
)->i_data_sem
));
1790 * This is a special get_block_t callback which is used by
1791 * ext4_da_write_begin(). It will either return mapped block or
1792 * reserve space for a single block.
1794 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1795 * We also have b_blocknr = -1 and b_bdev initialized properly
1797 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1798 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1799 * initialized properly.
1801 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1802 struct buffer_head
*bh
, int create
)
1804 struct ext4_map_blocks map
;
1807 BUG_ON(create
== 0);
1808 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1810 map
.m_lblk
= iblock
;
1814 * first, we need to know whether the block is allocated already
1815 * preallocated blocks are unmapped but should treated
1816 * the same as allocated blocks.
1818 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1822 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1823 ext4_update_bh_state(bh
, map
.m_flags
);
1825 if (buffer_unwritten(bh
)) {
1826 /* A delayed write to unwritten bh should be marked
1827 * new and mapped. Mapped ensures that we don't do
1828 * get_block multiple times when we write to the same
1829 * offset and new ensures that we do proper zero out
1830 * for partial write.
1833 set_buffer_mapped(bh
);
1838 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1844 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1850 static int __ext4_journalled_writepage(struct page
*page
,
1853 struct address_space
*mapping
= page
->mapping
;
1854 struct inode
*inode
= mapping
->host
;
1855 struct buffer_head
*page_bufs
= NULL
;
1856 handle_t
*handle
= NULL
;
1857 int ret
= 0, err
= 0;
1858 int inline_data
= ext4_has_inline_data(inode
);
1859 struct buffer_head
*inode_bh
= NULL
;
1861 ClearPageChecked(page
);
1864 BUG_ON(page
->index
!= 0);
1865 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1866 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1867 if (inode_bh
== NULL
)
1870 page_bufs
= page_buffers(page
);
1875 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1879 * We need to release the page lock before we start the
1880 * journal, so grab a reference so the page won't disappear
1881 * out from under us.
1886 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1887 ext4_writepage_trans_blocks(inode
));
1888 if (IS_ERR(handle
)) {
1889 ret
= PTR_ERR(handle
);
1891 goto out_no_pagelock
;
1893 BUG_ON(!ext4_handle_valid(handle
));
1897 if (page
->mapping
!= mapping
) {
1898 /* The page got truncated from under us */
1899 ext4_journal_stop(handle
);
1905 ret
= ext4_mark_inode_dirty(handle
, inode
);
1907 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1908 do_journal_get_write_access
);
1910 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1915 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1916 err
= ext4_journal_stop(handle
);
1920 if (!ext4_has_inline_data(inode
))
1921 ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
,
1923 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1932 * Note that we don't need to start a transaction unless we're journaling data
1933 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1934 * need to file the inode to the transaction's list in ordered mode because if
1935 * we are writing back data added by write(), the inode is already there and if
1936 * we are writing back data modified via mmap(), no one guarantees in which
1937 * transaction the data will hit the disk. In case we are journaling data, we
1938 * cannot start transaction directly because transaction start ranks above page
1939 * lock so we have to do some magic.
1941 * This function can get called via...
1942 * - ext4_writepages after taking page lock (have journal handle)
1943 * - journal_submit_inode_data_buffers (no journal handle)
1944 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1945 * - grab_page_cache when doing write_begin (have journal handle)
1947 * We don't do any block allocation in this function. If we have page with
1948 * multiple blocks we need to write those buffer_heads that are mapped. This
1949 * is important for mmaped based write. So if we do with blocksize 1K
1950 * truncate(f, 1024);
1951 * a = mmap(f, 0, 4096);
1953 * truncate(f, 4096);
1954 * we have in the page first buffer_head mapped via page_mkwrite call back
1955 * but other buffer_heads would be unmapped but dirty (dirty done via the
1956 * do_wp_page). So writepage should write the first block. If we modify
1957 * the mmap area beyond 1024 we will again get a page_fault and the
1958 * page_mkwrite callback will do the block allocation and mark the
1959 * buffer_heads mapped.
1961 * We redirty the page if we have any buffer_heads that is either delay or
1962 * unwritten in the page.
1964 * We can get recursively called as show below.
1966 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1969 * But since we don't do any block allocation we should not deadlock.
1970 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1972 static int ext4_writepage(struct page
*page
,
1973 struct writeback_control
*wbc
)
1978 struct buffer_head
*page_bufs
= NULL
;
1979 struct inode
*inode
= page
->mapping
->host
;
1980 struct ext4_io_submit io_submit
;
1981 bool keep_towrite
= false;
1983 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
)))) {
1984 inode
->i_mapping
->a_ops
->invalidatepage(page
, 0, PAGE_SIZE
);
1989 trace_ext4_writepage(page
);
1990 size
= i_size_read(inode
);
1991 if (page
->index
== size
>> PAGE_SHIFT
&&
1992 !ext4_verity_in_progress(inode
))
1993 len
= size
& ~PAGE_MASK
;
1997 page_bufs
= page_buffers(page
);
1999 * We cannot do block allocation or other extent handling in this
2000 * function. If there are buffers needing that, we have to redirty
2001 * the page. But we may reach here when we do a journal commit via
2002 * journal_submit_inode_data_buffers() and in that case we must write
2003 * allocated buffers to achieve data=ordered mode guarantees.
2005 * Also, if there is only one buffer per page (the fs block
2006 * size == the page size), if one buffer needs block
2007 * allocation or needs to modify the extent tree to clear the
2008 * unwritten flag, we know that the page can't be written at
2009 * all, so we might as well refuse the write immediately.
2010 * Unfortunately if the block size != page size, we can't as
2011 * easily detect this case using ext4_walk_page_buffers(), but
2012 * for the extremely common case, this is an optimization that
2013 * skips a useless round trip through ext4_bio_write_page().
2015 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2016 ext4_bh_delay_or_unwritten
)) {
2017 redirty_page_for_writepage(wbc
, page
);
2018 if ((current
->flags
& PF_MEMALLOC
) ||
2019 (inode
->i_sb
->s_blocksize
== PAGE_SIZE
)) {
2021 * For memory cleaning there's no point in writing only
2022 * some buffers. So just bail out. Warn if we came here
2023 * from direct reclaim.
2025 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2030 keep_towrite
= true;
2033 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2035 * It's mmapped pagecache. Add buffers and journal it. There
2036 * doesn't seem much point in redirtying the page here.
2038 return __ext4_journalled_writepage(page
, len
);
2040 ext4_io_submit_init(&io_submit
, wbc
);
2041 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
2042 if (!io_submit
.io_end
) {
2043 redirty_page_for_writepage(wbc
, page
);
2047 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
, keep_towrite
);
2048 ext4_io_submit(&io_submit
);
2049 /* Drop io_end reference we got from init */
2050 ext4_put_io_end_defer(io_submit
.io_end
);
2054 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
2060 BUG_ON(page
->index
!= mpd
->first_page
);
2061 clear_page_dirty_for_io(page
);
2063 * We have to be very careful here! Nothing protects writeback path
2064 * against i_size changes and the page can be writeably mapped into
2065 * page tables. So an application can be growing i_size and writing
2066 * data through mmap while writeback runs. clear_page_dirty_for_io()
2067 * write-protects our page in page tables and the page cannot get
2068 * written to again until we release page lock. So only after
2069 * clear_page_dirty_for_io() we are safe to sample i_size for
2070 * ext4_bio_write_page() to zero-out tail of the written page. We rely
2071 * on the barrier provided by TestClearPageDirty in
2072 * clear_page_dirty_for_io() to make sure i_size is really sampled only
2073 * after page tables are updated.
2075 size
= i_size_read(mpd
->inode
);
2076 if (page
->index
== size
>> PAGE_SHIFT
&&
2077 !ext4_verity_in_progress(mpd
->inode
))
2078 len
= size
& ~PAGE_MASK
;
2081 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
, false);
2083 mpd
->wbc
->nr_to_write
--;
2089 #define BH_FLAGS (BIT(BH_Unwritten) | BIT(BH_Delay))
2092 * mballoc gives us at most this number of blocks...
2093 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
2094 * The rest of mballoc seems to handle chunks up to full group size.
2096 #define MAX_WRITEPAGES_EXTENT_LEN 2048
2099 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
2101 * @mpd - extent of blocks
2102 * @lblk - logical number of the block in the file
2103 * @bh - buffer head we want to add to the extent
2105 * The function is used to collect contig. blocks in the same state. If the
2106 * buffer doesn't require mapping for writeback and we haven't started the
2107 * extent of buffers to map yet, the function returns 'true' immediately - the
2108 * caller can write the buffer right away. Otherwise the function returns true
2109 * if the block has been added to the extent, false if the block couldn't be
2112 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
2113 struct buffer_head
*bh
)
2115 struct ext4_map_blocks
*map
= &mpd
->map
;
2117 /* Buffer that doesn't need mapping for writeback? */
2118 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
2119 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
2120 /* So far no extent to map => we write the buffer right away */
2121 if (map
->m_len
== 0)
2126 /* First block in the extent? */
2127 if (map
->m_len
== 0) {
2128 /* We cannot map unless handle is started... */
2133 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
2137 /* Don't go larger than mballoc is willing to allocate */
2138 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
2141 /* Can we merge the block to our big extent? */
2142 if (lblk
== map
->m_lblk
+ map
->m_len
&&
2143 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
2151 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
2153 * @mpd - extent of blocks for mapping
2154 * @head - the first buffer in the page
2155 * @bh - buffer we should start processing from
2156 * @lblk - logical number of the block in the file corresponding to @bh
2158 * Walk through page buffers from @bh upto @head (exclusive) and either submit
2159 * the page for IO if all buffers in this page were mapped and there's no
2160 * accumulated extent of buffers to map or add buffers in the page to the
2161 * extent of buffers to map. The function returns 1 if the caller can continue
2162 * by processing the next page, 0 if it should stop adding buffers to the
2163 * extent to map because we cannot extend it anymore. It can also return value
2164 * < 0 in case of error during IO submission.
2166 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
2167 struct buffer_head
*head
,
2168 struct buffer_head
*bh
,
2171 struct inode
*inode
= mpd
->inode
;
2173 ext4_lblk_t blocks
= (i_size_read(inode
) + i_blocksize(inode
) - 1)
2174 >> inode
->i_blkbits
;
2176 if (ext4_verity_in_progress(inode
))
2177 blocks
= EXT_MAX_BLOCKS
;
2180 BUG_ON(buffer_locked(bh
));
2182 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2183 /* Found extent to map? */
2186 /* Buffer needs mapping and handle is not started? */
2189 /* Everything mapped so far and we hit EOF */
2192 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2193 /* So far everything mapped? Submit the page for IO. */
2194 if (mpd
->map
.m_len
== 0) {
2195 err
= mpage_submit_page(mpd
, head
->b_page
);
2199 if (lblk
>= blocks
) {
2200 mpd
->scanned_until_end
= 1;
2207 * mpage_process_page - update page buffers corresponding to changed extent and
2208 * may submit fully mapped page for IO
2210 * @mpd - description of extent to map, on return next extent to map
2211 * @m_lblk - logical block mapping.
2212 * @m_pblk - corresponding physical mapping.
2213 * @map_bh - determines on return whether this page requires any further
2215 * Scan given page buffers corresponding to changed extent and update buffer
2216 * state according to new extent state.
2217 * We map delalloc buffers to their physical location, clear unwritten bits.
2218 * If the given page is not fully mapped, we update @map to the next extent in
2219 * the given page that needs mapping & return @map_bh as true.
2221 static int mpage_process_page(struct mpage_da_data
*mpd
, struct page
*page
,
2222 ext4_lblk_t
*m_lblk
, ext4_fsblk_t
*m_pblk
,
2225 struct buffer_head
*head
, *bh
;
2226 ext4_io_end_t
*io_end
= mpd
->io_submit
.io_end
;
2227 ext4_lblk_t lblk
= *m_lblk
;
2228 ext4_fsblk_t pblock
= *m_pblk
;
2230 int blkbits
= mpd
->inode
->i_blkbits
;
2231 ssize_t io_end_size
= 0;
2232 struct ext4_io_end_vec
*io_end_vec
= ext4_last_io_end_vec(io_end
);
2234 bh
= head
= page_buffers(page
);
2236 if (lblk
< mpd
->map
.m_lblk
)
2238 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2240 * Buffer after end of mapped extent.
2241 * Find next buffer in the page to map.
2244 mpd
->map
.m_flags
= 0;
2245 io_end_vec
->size
+= io_end_size
;
2248 err
= mpage_process_page_bufs(mpd
, head
, bh
, lblk
);
2251 if (!err
&& mpd
->map
.m_len
&& mpd
->map
.m_lblk
> lblk
) {
2252 io_end_vec
= ext4_alloc_io_end_vec(io_end
);
2253 if (IS_ERR(io_end_vec
)) {
2254 err
= PTR_ERR(io_end_vec
);
2257 io_end_vec
->offset
= mpd
->map
.m_lblk
<< blkbits
;
2262 if (buffer_delay(bh
)) {
2263 clear_buffer_delay(bh
);
2264 bh
->b_blocknr
= pblock
++;
2266 clear_buffer_unwritten(bh
);
2267 io_end_size
+= (1 << blkbits
);
2268 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2270 io_end_vec
->size
+= io_end_size
;
2280 * mpage_map_buffers - update buffers corresponding to changed extent and
2281 * submit fully mapped pages for IO
2283 * @mpd - description of extent to map, on return next extent to map
2285 * Scan buffers corresponding to changed extent (we expect corresponding pages
2286 * to be already locked) and update buffer state according to new extent state.
2287 * We map delalloc buffers to their physical location, clear unwritten bits,
2288 * and mark buffers as uninit when we perform writes to unwritten extents
2289 * and do extent conversion after IO is finished. If the last page is not fully
2290 * mapped, we update @map to the next extent in the last page that needs
2291 * mapping. Otherwise we submit the page for IO.
2293 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2295 struct pagevec pvec
;
2297 struct inode
*inode
= mpd
->inode
;
2298 int bpp_bits
= PAGE_SHIFT
- inode
->i_blkbits
;
2301 ext4_fsblk_t pblock
;
2303 bool map_bh
= false;
2305 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2306 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2307 lblk
= start
<< bpp_bits
;
2308 pblock
= mpd
->map
.m_pblk
;
2310 pagevec_init(&pvec
);
2311 while (start
<= end
) {
2312 nr_pages
= pagevec_lookup_range(&pvec
, inode
->i_mapping
,
2316 for (i
= 0; i
< nr_pages
; i
++) {
2317 struct page
*page
= pvec
.pages
[i
];
2319 err
= mpage_process_page(mpd
, page
, &lblk
, &pblock
,
2322 * If map_bh is true, means page may require further bh
2323 * mapping, or maybe the page was submitted for IO.
2324 * So we return to call further extent mapping.
2326 if (err
< 0 || map_bh
)
2328 /* Page fully mapped - let IO run! */
2329 err
= mpage_submit_page(mpd
, page
);
2333 pagevec_release(&pvec
);
2335 /* Extent fully mapped and matches with page boundary. We are done. */
2337 mpd
->map
.m_flags
= 0;
2340 pagevec_release(&pvec
);
2344 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2346 struct inode
*inode
= mpd
->inode
;
2347 struct ext4_map_blocks
*map
= &mpd
->map
;
2348 int get_blocks_flags
;
2349 int err
, dioread_nolock
;
2351 trace_ext4_da_write_pages_extent(inode
, map
);
2353 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2354 * to convert an unwritten extent to be initialized (in the case
2355 * where we have written into one or more preallocated blocks). It is
2356 * possible that we're going to need more metadata blocks than
2357 * previously reserved. However we must not fail because we're in
2358 * writeback and there is nothing we can do about it so it might result
2359 * in data loss. So use reserved blocks to allocate metadata if
2362 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if
2363 * the blocks in question are delalloc blocks. This indicates
2364 * that the blocks and quotas has already been checked when
2365 * the data was copied into the page cache.
2367 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2368 EXT4_GET_BLOCKS_METADATA_NOFAIL
|
2369 EXT4_GET_BLOCKS_IO_SUBMIT
;
2370 dioread_nolock
= ext4_should_dioread_nolock(inode
);
2372 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2373 if (map
->m_flags
& BIT(BH_Delay
))
2374 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2376 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2379 if (dioread_nolock
&& (map
->m_flags
& EXT4_MAP_UNWRITTEN
)) {
2380 if (!mpd
->io_submit
.io_end
->handle
&&
2381 ext4_handle_valid(handle
)) {
2382 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2383 handle
->h_rsv_handle
= NULL
;
2385 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2388 BUG_ON(map
->m_len
== 0);
2393 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2394 * mpd->len and submit pages underlying it for IO
2396 * @handle - handle for journal operations
2397 * @mpd - extent to map
2398 * @give_up_on_write - we set this to true iff there is a fatal error and there
2399 * is no hope of writing the data. The caller should discard
2400 * dirty pages to avoid infinite loops.
2402 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2403 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2404 * them to initialized or split the described range from larger unwritten
2405 * extent. Note that we need not map all the described range since allocation
2406 * can return less blocks or the range is covered by more unwritten extents. We
2407 * cannot map more because we are limited by reserved transaction credits. On
2408 * the other hand we always make sure that the last touched page is fully
2409 * mapped so that it can be written out (and thus forward progress is
2410 * guaranteed). After mapping we submit all mapped pages for IO.
2412 static int mpage_map_and_submit_extent(handle_t
*handle
,
2413 struct mpage_da_data
*mpd
,
2414 bool *give_up_on_write
)
2416 struct inode
*inode
= mpd
->inode
;
2417 struct ext4_map_blocks
*map
= &mpd
->map
;
2421 ext4_io_end_t
*io_end
= mpd
->io_submit
.io_end
;
2422 struct ext4_io_end_vec
*io_end_vec
;
2424 io_end_vec
= ext4_alloc_io_end_vec(io_end
);
2425 if (IS_ERR(io_end_vec
))
2426 return PTR_ERR(io_end_vec
);
2427 io_end_vec
->offset
= ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2429 err
= mpage_map_one_extent(handle
, mpd
);
2431 struct super_block
*sb
= inode
->i_sb
;
2433 if (ext4_forced_shutdown(EXT4_SB(sb
)) ||
2434 EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2435 goto invalidate_dirty_pages
;
2437 * Let the uper layers retry transient errors.
2438 * In the case of ENOSPC, if ext4_count_free_blocks()
2439 * is non-zero, a commit should free up blocks.
2441 if ((err
== -ENOMEM
) ||
2442 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
))) {
2444 goto update_disksize
;
2447 ext4_msg(sb
, KERN_CRIT
,
2448 "Delayed block allocation failed for "
2449 "inode %lu at logical offset %llu with"
2450 " max blocks %u with error %d",
2452 (unsigned long long)map
->m_lblk
,
2453 (unsigned)map
->m_len
, -err
);
2454 ext4_msg(sb
, KERN_CRIT
,
2455 "This should not happen!! Data will "
2458 ext4_print_free_blocks(inode
);
2459 invalidate_dirty_pages
:
2460 *give_up_on_write
= true;
2465 * Update buffer state, submit mapped pages, and get us new
2468 err
= mpage_map_and_submit_buffers(mpd
);
2470 goto update_disksize
;
2471 } while (map
->m_len
);
2475 * Update on-disk size after IO is submitted. Races with
2476 * truncate are avoided by checking i_size under i_data_sem.
2478 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_SHIFT
;
2479 if (disksize
> READ_ONCE(EXT4_I(inode
)->i_disksize
)) {
2483 down_write(&EXT4_I(inode
)->i_data_sem
);
2484 i_size
= i_size_read(inode
);
2485 if (disksize
> i_size
)
2487 if (disksize
> EXT4_I(inode
)->i_disksize
)
2488 EXT4_I(inode
)->i_disksize
= disksize
;
2489 up_write(&EXT4_I(inode
)->i_data_sem
);
2490 err2
= ext4_mark_inode_dirty(handle
, inode
);
2492 ext4_error_err(inode
->i_sb
, -err2
,
2493 "Failed to mark inode %lu dirty",
2503 * Calculate the total number of credits to reserve for one writepages
2504 * iteration. This is called from ext4_writepages(). We map an extent of
2505 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2506 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2507 * bpp - 1 blocks in bpp different extents.
2509 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2511 int bpp
= ext4_journal_blocks_per_page(inode
);
2513 return ext4_meta_trans_blocks(inode
,
2514 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2518 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2519 * and underlying extent to map
2521 * @mpd - where to look for pages
2523 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2524 * IO immediately. When we find a page which isn't mapped we start accumulating
2525 * extent of buffers underlying these pages that needs mapping (formed by
2526 * either delayed or unwritten buffers). We also lock the pages containing
2527 * these buffers. The extent found is returned in @mpd structure (starting at
2528 * mpd->lblk with length mpd->len blocks).
2530 * Note that this function can attach bios to one io_end structure which are
2531 * neither logically nor physically contiguous. Although it may seem as an
2532 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2533 * case as we need to track IO to all buffers underlying a page in one io_end.
2535 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2537 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2538 struct pagevec pvec
;
2539 unsigned int nr_pages
;
2540 long left
= mpd
->wbc
->nr_to_write
;
2541 pgoff_t index
= mpd
->first_page
;
2542 pgoff_t end
= mpd
->last_page
;
2545 int blkbits
= mpd
->inode
->i_blkbits
;
2547 struct buffer_head
*head
;
2549 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2550 tag
= PAGECACHE_TAG_TOWRITE
;
2552 tag
= PAGECACHE_TAG_DIRTY
;
2554 pagevec_init(&pvec
);
2556 mpd
->next_page
= index
;
2557 while (index
<= end
) {
2558 nr_pages
= pagevec_lookup_range_tag(&pvec
, mapping
, &index
, end
,
2563 for (i
= 0; i
< nr_pages
; i
++) {
2564 struct page
*page
= pvec
.pages
[i
];
2567 * Accumulated enough dirty pages? This doesn't apply
2568 * to WB_SYNC_ALL mode. For integrity sync we have to
2569 * keep going because someone may be concurrently
2570 * dirtying pages, and we might have synced a lot of
2571 * newly appeared dirty pages, but have not synced all
2572 * of the old dirty pages.
2574 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&& left
<= 0)
2577 /* If we can't merge this page, we are done. */
2578 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2583 * If the page is no longer dirty, or its mapping no
2584 * longer corresponds to inode we are writing (which
2585 * means it has been truncated or invalidated), or the
2586 * page is already under writeback and we are not doing
2587 * a data integrity writeback, skip the page
2589 if (!PageDirty(page
) ||
2590 (PageWriteback(page
) &&
2591 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2592 unlikely(page
->mapping
!= mapping
)) {
2597 wait_on_page_writeback(page
);
2598 BUG_ON(PageWriteback(page
));
2600 if (mpd
->map
.m_len
== 0)
2601 mpd
->first_page
= page
->index
;
2602 mpd
->next_page
= page
->index
+ 1;
2603 /* Add all dirty buffers to mpd */
2604 lblk
= ((ext4_lblk_t
)page
->index
) <<
2605 (PAGE_SHIFT
- blkbits
);
2606 head
= page_buffers(page
);
2607 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2613 pagevec_release(&pvec
);
2616 mpd
->scanned_until_end
= 1;
2619 pagevec_release(&pvec
);
2623 static int ext4_writepages(struct address_space
*mapping
,
2624 struct writeback_control
*wbc
)
2626 pgoff_t writeback_index
= 0;
2627 long nr_to_write
= wbc
->nr_to_write
;
2628 int range_whole
= 0;
2630 handle_t
*handle
= NULL
;
2631 struct mpage_da_data mpd
;
2632 struct inode
*inode
= mapping
->host
;
2633 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2634 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2635 struct blk_plug plug
;
2636 bool give_up_on_write
= false;
2638 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
2641 percpu_down_read(&sbi
->s_writepages_rwsem
);
2642 trace_ext4_writepages(inode
, wbc
);
2645 * No pages to write? This is mainly a kludge to avoid starting
2646 * a transaction for special inodes like journal inode on last iput()
2647 * because that could violate lock ordering on umount
2649 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2650 goto out_writepages
;
2652 if (ext4_should_journal_data(inode
)) {
2653 ret
= generic_writepages(mapping
, wbc
);
2654 goto out_writepages
;
2658 * If the filesystem has aborted, it is read-only, so return
2659 * right away instead of dumping stack traces later on that
2660 * will obscure the real source of the problem. We test
2661 * EXT4_MF_FS_ABORTED instead of sb->s_flag's SB_RDONLY because
2662 * the latter could be true if the filesystem is mounted
2663 * read-only, and in that case, ext4_writepages should
2664 * *never* be called, so if that ever happens, we would want
2667 if (unlikely(ext4_forced_shutdown(EXT4_SB(mapping
->host
->i_sb
)) ||
2668 sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
2670 goto out_writepages
;
2674 * If we have inline data and arrive here, it means that
2675 * we will soon create the block for the 1st page, so
2676 * we'd better clear the inline data here.
2678 if (ext4_has_inline_data(inode
)) {
2679 /* Just inode will be modified... */
2680 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2681 if (IS_ERR(handle
)) {
2682 ret
= PTR_ERR(handle
);
2683 goto out_writepages
;
2685 BUG_ON(ext4_test_inode_state(inode
,
2686 EXT4_STATE_MAY_INLINE_DATA
));
2687 ext4_destroy_inline_data(handle
, inode
);
2688 ext4_journal_stop(handle
);
2691 if (ext4_should_dioread_nolock(inode
)) {
2693 * We may need to convert up to one extent per block in
2694 * the page and we may dirty the inode.
2696 rsv_blocks
= 1 + ext4_chunk_trans_blocks(inode
,
2697 PAGE_SIZE
>> inode
->i_blkbits
);
2700 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2703 if (wbc
->range_cyclic
) {
2704 writeback_index
= mapping
->writeback_index
;
2705 if (writeback_index
)
2707 mpd
.first_page
= writeback_index
;
2710 mpd
.first_page
= wbc
->range_start
>> PAGE_SHIFT
;
2711 mpd
.last_page
= wbc
->range_end
>> PAGE_SHIFT
;
2716 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2718 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2719 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2720 blk_start_plug(&plug
);
2723 * First writeback pages that don't need mapping - we can avoid
2724 * starting a transaction unnecessarily and also avoid being blocked
2725 * in the block layer on device congestion while having transaction
2729 mpd
.scanned_until_end
= 0;
2730 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2731 if (!mpd
.io_submit
.io_end
) {
2735 ret
= mpage_prepare_extent_to_map(&mpd
);
2736 /* Unlock pages we didn't use */
2737 mpage_release_unused_pages(&mpd
, false);
2738 /* Submit prepared bio */
2739 ext4_io_submit(&mpd
.io_submit
);
2740 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2741 mpd
.io_submit
.io_end
= NULL
;
2745 while (!mpd
.scanned_until_end
&& wbc
->nr_to_write
> 0) {
2746 /* For each extent of pages we use new io_end */
2747 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2748 if (!mpd
.io_submit
.io_end
) {
2754 * We have two constraints: We find one extent to map and we
2755 * must always write out whole page (makes a difference when
2756 * blocksize < pagesize) so that we don't block on IO when we
2757 * try to write out the rest of the page. Journalled mode is
2758 * not supported by delalloc.
2760 BUG_ON(ext4_should_journal_data(inode
));
2761 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2763 /* start a new transaction */
2764 handle
= ext4_journal_start_with_reserve(inode
,
2765 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2766 if (IS_ERR(handle
)) {
2767 ret
= PTR_ERR(handle
);
2768 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2769 "%ld pages, ino %lu; err %d", __func__
,
2770 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2771 /* Release allocated io_end */
2772 ext4_put_io_end(mpd
.io_submit
.io_end
);
2773 mpd
.io_submit
.io_end
= NULL
;
2778 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2779 ret
= mpage_prepare_extent_to_map(&mpd
);
2780 if (!ret
&& mpd
.map
.m_len
)
2781 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2784 * Caution: If the handle is synchronous,
2785 * ext4_journal_stop() can wait for transaction commit
2786 * to finish which may depend on writeback of pages to
2787 * complete or on page lock to be released. In that
2788 * case, we have to wait until after after we have
2789 * submitted all the IO, released page locks we hold,
2790 * and dropped io_end reference (for extent conversion
2791 * to be able to complete) before stopping the handle.
2793 if (!ext4_handle_valid(handle
) || handle
->h_sync
== 0) {
2794 ext4_journal_stop(handle
);
2798 /* Unlock pages we didn't use */
2799 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2800 /* Submit prepared bio */
2801 ext4_io_submit(&mpd
.io_submit
);
2804 * Drop our io_end reference we got from init. We have
2805 * to be careful and use deferred io_end finishing if
2806 * we are still holding the transaction as we can
2807 * release the last reference to io_end which may end
2808 * up doing unwritten extent conversion.
2811 ext4_put_io_end_defer(mpd
.io_submit
.io_end
);
2812 ext4_journal_stop(handle
);
2814 ext4_put_io_end(mpd
.io_submit
.io_end
);
2815 mpd
.io_submit
.io_end
= NULL
;
2817 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2819 * Commit the transaction which would
2820 * free blocks released in the transaction
2823 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2827 /* Fatal error - ENOMEM, EIO... */
2832 blk_finish_plug(&plug
);
2833 if (!ret
&& !cycled
&& wbc
->nr_to_write
> 0) {
2835 mpd
.last_page
= writeback_index
- 1;
2841 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2843 * Set the writeback_index so that range_cyclic
2844 * mode will write it back later
2846 mapping
->writeback_index
= mpd
.first_page
;
2849 trace_ext4_writepages_result(inode
, wbc
, ret
,
2850 nr_to_write
- wbc
->nr_to_write
);
2851 percpu_up_read(&sbi
->s_writepages_rwsem
);
2855 static int ext4_dax_writepages(struct address_space
*mapping
,
2856 struct writeback_control
*wbc
)
2859 long nr_to_write
= wbc
->nr_to_write
;
2860 struct inode
*inode
= mapping
->host
;
2861 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2863 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
2866 percpu_down_read(&sbi
->s_writepages_rwsem
);
2867 trace_ext4_writepages(inode
, wbc
);
2869 ret
= dax_writeback_mapping_range(mapping
, sbi
->s_daxdev
, wbc
);
2870 trace_ext4_writepages_result(inode
, wbc
, ret
,
2871 nr_to_write
- wbc
->nr_to_write
);
2872 percpu_up_read(&sbi
->s_writepages_rwsem
);
2876 static int ext4_nonda_switch(struct super_block
*sb
)
2878 s64 free_clusters
, dirty_clusters
;
2879 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2882 * switch to non delalloc mode if we are running low
2883 * on free block. The free block accounting via percpu
2884 * counters can get slightly wrong with percpu_counter_batch getting
2885 * accumulated on each CPU without updating global counters
2886 * Delalloc need an accurate free block accounting. So switch
2887 * to non delalloc when we are near to error range.
2890 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2892 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2894 * Start pushing delalloc when 1/2 of free blocks are dirty.
2896 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2897 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2899 if (2 * free_clusters
< 3 * dirty_clusters
||
2900 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2902 * free block count is less than 150% of dirty blocks
2903 * or free blocks is less than watermark
2910 /* We always reserve for an inode update; the superblock could be there too */
2911 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2913 if (likely(ext4_has_feature_large_file(inode
->i_sb
)))
2916 if (pos
+ len
<= 0x7fffffffULL
)
2919 /* We might need to update the superblock to set LARGE_FILE */
2923 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2924 loff_t pos
, unsigned len
, unsigned flags
,
2925 struct page
**pagep
, void **fsdata
)
2927 int ret
, retries
= 0;
2930 struct inode
*inode
= mapping
->host
;
2933 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
2936 index
= pos
>> PAGE_SHIFT
;
2938 if (ext4_nonda_switch(inode
->i_sb
) || S_ISLNK(inode
->i_mode
) ||
2939 ext4_verity_in_progress(inode
)) {
2940 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2941 return ext4_write_begin(file
, mapping
, pos
,
2942 len
, flags
, pagep
, fsdata
);
2944 *fsdata
= (void *)0;
2945 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2947 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2948 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2958 * grab_cache_page_write_begin() can take a long time if the
2959 * system is thrashing due to memory pressure, or if the page
2960 * is being written back. So grab it first before we start
2961 * the transaction handle. This also allows us to allocate
2962 * the page (if needed) without using GFP_NOFS.
2965 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2971 * With delayed allocation, we don't log the i_disksize update
2972 * if there is delayed block allocation. But we still need
2973 * to journalling the i_disksize update if writes to the end
2974 * of file which has an already mapped buffer.
2977 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2978 ext4_da_write_credits(inode
, pos
, len
));
2979 if (IS_ERR(handle
)) {
2981 return PTR_ERR(handle
);
2985 if (page
->mapping
!= mapping
) {
2986 /* The page got truncated from under us */
2989 ext4_journal_stop(handle
);
2992 /* In case writeback began while the page was unlocked */
2993 wait_for_stable_page(page
);
2995 #ifdef CONFIG_FS_ENCRYPTION
2996 ret
= ext4_block_write_begin(page
, pos
, len
,
2997 ext4_da_get_block_prep
);
2999 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
3003 ext4_journal_stop(handle
);
3005 * block_write_begin may have instantiated a few blocks
3006 * outside i_size. Trim these off again. Don't need
3007 * i_size_read because we hold i_mutex.
3009 if (pos
+ len
> inode
->i_size
)
3010 ext4_truncate_failed_write(inode
);
3012 if (ret
== -ENOSPC
&&
3013 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3025 * Check if we should update i_disksize
3026 * when write to the end of file but not require block allocation
3028 static int ext4_da_should_update_i_disksize(struct page
*page
,
3029 unsigned long offset
)
3031 struct buffer_head
*bh
;
3032 struct inode
*inode
= page
->mapping
->host
;
3036 bh
= page_buffers(page
);
3037 idx
= offset
>> inode
->i_blkbits
;
3039 for (i
= 0; i
< idx
; i
++)
3040 bh
= bh
->b_this_page
;
3042 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
3047 static int ext4_da_write_end(struct file
*file
,
3048 struct address_space
*mapping
,
3049 loff_t pos
, unsigned len
, unsigned copied
,
3050 struct page
*page
, void *fsdata
)
3052 struct inode
*inode
= mapping
->host
;
3054 handle_t
*handle
= ext4_journal_current_handle();
3056 unsigned long start
, end
;
3057 int write_mode
= (int)(unsigned long)fsdata
;
3059 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
3060 return ext4_write_end(file
, mapping
, pos
,
3061 len
, copied
, page
, fsdata
);
3063 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
3064 start
= pos
& (PAGE_SIZE
- 1);
3065 end
= start
+ copied
- 1;
3068 * generic_write_end() will run mark_inode_dirty() if i_size
3069 * changes. So let's piggyback the i_disksize mark_inode_dirty
3072 new_i_size
= pos
+ copied
;
3073 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
3074 if (ext4_has_inline_data(inode
) ||
3075 ext4_da_should_update_i_disksize(page
, end
)) {
3076 ext4_update_i_disksize(inode
, new_i_size
);
3077 /* We need to mark inode dirty even if
3078 * new_i_size is less that inode->i_size
3079 * bu greater than i_disksize.(hint delalloc)
3081 ret
= ext4_mark_inode_dirty(handle
, inode
);
3085 if (write_mode
!= CONVERT_INLINE_DATA
&&
3086 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
3087 ext4_has_inline_data(inode
))
3088 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
3091 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
3097 ret2
= ext4_journal_stop(handle
);
3098 if (unlikely(ret2
&& !ret
))
3101 return ret
? ret
: copied
;
3105 * Force all delayed allocation blocks to be allocated for a given inode.
3107 int ext4_alloc_da_blocks(struct inode
*inode
)
3109 trace_ext4_alloc_da_blocks(inode
);
3111 if (!EXT4_I(inode
)->i_reserved_data_blocks
)
3115 * We do something simple for now. The filemap_flush() will
3116 * also start triggering a write of the data blocks, which is
3117 * not strictly speaking necessary (and for users of
3118 * laptop_mode, not even desirable). However, to do otherwise
3119 * would require replicating code paths in:
3121 * ext4_writepages() ->
3122 * write_cache_pages() ---> (via passed in callback function)
3123 * __mpage_da_writepage() -->
3124 * mpage_add_bh_to_extent()
3125 * mpage_da_map_blocks()
3127 * The problem is that write_cache_pages(), located in
3128 * mm/page-writeback.c, marks pages clean in preparation for
3129 * doing I/O, which is not desirable if we're not planning on
3132 * We could call write_cache_pages(), and then redirty all of
3133 * the pages by calling redirty_page_for_writepage() but that
3134 * would be ugly in the extreme. So instead we would need to
3135 * replicate parts of the code in the above functions,
3136 * simplifying them because we wouldn't actually intend to
3137 * write out the pages, but rather only collect contiguous
3138 * logical block extents, call the multi-block allocator, and
3139 * then update the buffer heads with the block allocations.
3141 * For now, though, we'll cheat by calling filemap_flush(),
3142 * which will map the blocks, and start the I/O, but not
3143 * actually wait for the I/O to complete.
3145 return filemap_flush(inode
->i_mapping
);
3149 * bmap() is special. It gets used by applications such as lilo and by
3150 * the swapper to find the on-disk block of a specific piece of data.
3152 * Naturally, this is dangerous if the block concerned is still in the
3153 * journal. If somebody makes a swapfile on an ext4 data-journaling
3154 * filesystem and enables swap, then they may get a nasty shock when the
3155 * data getting swapped to that swapfile suddenly gets overwritten by
3156 * the original zero's written out previously to the journal and
3157 * awaiting writeback in the kernel's buffer cache.
3159 * So, if we see any bmap calls here on a modified, data-journaled file,
3160 * take extra steps to flush any blocks which might be in the cache.
3162 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
3164 struct inode
*inode
= mapping
->host
;
3169 * We can get here for an inline file via the FIBMAP ioctl
3171 if (ext4_has_inline_data(inode
))
3174 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
3175 test_opt(inode
->i_sb
, DELALLOC
)) {
3177 * With delalloc we want to sync the file
3178 * so that we can make sure we allocate
3181 filemap_write_and_wait(mapping
);
3184 if (EXT4_JOURNAL(inode
) &&
3185 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
3187 * This is a REALLY heavyweight approach, but the use of
3188 * bmap on dirty files is expected to be extremely rare:
3189 * only if we run lilo or swapon on a freshly made file
3190 * do we expect this to happen.
3192 * (bmap requires CAP_SYS_RAWIO so this does not
3193 * represent an unprivileged user DOS attack --- we'd be
3194 * in trouble if mortal users could trigger this path at
3197 * NB. EXT4_STATE_JDATA is not set on files other than
3198 * regular files. If somebody wants to bmap a directory
3199 * or symlink and gets confused because the buffer
3200 * hasn't yet been flushed to disk, they deserve
3201 * everything they get.
3204 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3205 journal
= EXT4_JOURNAL(inode
);
3206 jbd2_journal_lock_updates(journal
);
3207 err
= jbd2_journal_flush(journal
);
3208 jbd2_journal_unlock_updates(journal
);
3214 return iomap_bmap(mapping
, block
, &ext4_iomap_ops
);
3217 static int ext4_readpage(struct file
*file
, struct page
*page
)
3220 struct inode
*inode
= page
->mapping
->host
;
3222 trace_ext4_readpage(page
);
3224 if (ext4_has_inline_data(inode
))
3225 ret
= ext4_readpage_inline(inode
, page
);
3228 return ext4_mpage_readpages(inode
, NULL
, page
);
3233 static void ext4_readahead(struct readahead_control
*rac
)
3235 struct inode
*inode
= rac
->mapping
->host
;
3237 /* If the file has inline data, no need to do readahead. */
3238 if (ext4_has_inline_data(inode
))
3241 ext4_mpage_readpages(inode
, rac
, NULL
);
3244 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
3245 unsigned int length
)
3247 trace_ext4_invalidatepage(page
, offset
, length
);
3249 /* No journalling happens on data buffers when this function is used */
3250 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3252 block_invalidatepage(page
, offset
, length
);
3255 static int __ext4_journalled_invalidatepage(struct page
*page
,
3256 unsigned int offset
,
3257 unsigned int length
)
3259 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3261 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
3264 * If it's a full truncate we just forget about the pending dirtying
3266 if (offset
== 0 && length
== PAGE_SIZE
)
3267 ClearPageChecked(page
);
3269 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
3272 /* Wrapper for aops... */
3273 static void ext4_journalled_invalidatepage(struct page
*page
,
3274 unsigned int offset
,
3275 unsigned int length
)
3277 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
3280 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3282 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3284 trace_ext4_releasepage(page
);
3286 /* Page has dirty journalled data -> cannot release */
3287 if (PageChecked(page
))
3290 return jbd2_journal_try_to_free_buffers(journal
, page
);
3292 return try_to_free_buffers(page
);
3295 static bool ext4_inode_datasync_dirty(struct inode
*inode
)
3297 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
3300 return !jbd2_transaction_committed(journal
,
3301 EXT4_I(inode
)->i_datasync_tid
);
3302 /* Any metadata buffers to write? */
3303 if (!list_empty(&inode
->i_mapping
->private_list
))
3305 return inode
->i_state
& I_DIRTY_DATASYNC
;
3308 static void ext4_set_iomap(struct inode
*inode
, struct iomap
*iomap
,
3309 struct ext4_map_blocks
*map
, loff_t offset
,
3312 u8 blkbits
= inode
->i_blkbits
;
3315 * Writes that span EOF might trigger an I/O size update on completion,
3316 * so consider them to be dirty for the purpose of O_DSYNC, even if
3317 * there is no other metadata changes being made or are pending.
3320 if (ext4_inode_datasync_dirty(inode
) ||
3321 offset
+ length
> i_size_read(inode
))
3322 iomap
->flags
|= IOMAP_F_DIRTY
;
3324 if (map
->m_flags
& EXT4_MAP_NEW
)
3325 iomap
->flags
|= IOMAP_F_NEW
;
3327 iomap
->bdev
= inode
->i_sb
->s_bdev
;
3328 iomap
->dax_dev
= EXT4_SB(inode
->i_sb
)->s_daxdev
;
3329 iomap
->offset
= (u64
) map
->m_lblk
<< blkbits
;
3330 iomap
->length
= (u64
) map
->m_len
<< blkbits
;
3332 if ((map
->m_flags
& EXT4_MAP_MAPPED
) &&
3333 !ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3334 iomap
->flags
|= IOMAP_F_MERGED
;
3337 * Flags passed to ext4_map_blocks() for direct I/O writes can result
3338 * in m_flags having both EXT4_MAP_MAPPED and EXT4_MAP_UNWRITTEN bits
3339 * set. In order for any allocated unwritten extents to be converted
3340 * into written extents correctly within the ->end_io() handler, we
3341 * need to ensure that the iomap->type is set appropriately. Hence, the
3342 * reason why we need to check whether the EXT4_MAP_UNWRITTEN bit has
3345 if (map
->m_flags
& EXT4_MAP_UNWRITTEN
) {
3346 iomap
->type
= IOMAP_UNWRITTEN
;
3347 iomap
->addr
= (u64
) map
->m_pblk
<< blkbits
;
3348 } else if (map
->m_flags
& EXT4_MAP_MAPPED
) {
3349 iomap
->type
= IOMAP_MAPPED
;
3350 iomap
->addr
= (u64
) map
->m_pblk
<< blkbits
;
3352 iomap
->type
= IOMAP_HOLE
;
3353 iomap
->addr
= IOMAP_NULL_ADDR
;
3357 static int ext4_iomap_alloc(struct inode
*inode
, struct ext4_map_blocks
*map
,
3361 u8 blkbits
= inode
->i_blkbits
;
3362 int ret
, dio_credits
, m_flags
= 0, retries
= 0;
3365 * Trim the mapping request to the maximum value that we can map at
3366 * once for direct I/O.
3368 if (map
->m_len
> DIO_MAX_BLOCKS
)
3369 map
->m_len
= DIO_MAX_BLOCKS
;
3370 dio_credits
= ext4_chunk_trans_blocks(inode
, map
->m_len
);
3374 * Either we allocate blocks and then don't get an unwritten extent, so
3375 * in that case we have reserved enough credits. Or, the blocks are
3376 * already allocated and unwritten. In that case, the extent conversion
3377 * fits into the credits as well.
3379 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
, dio_credits
);
3381 return PTR_ERR(handle
);
3384 * DAX and direct I/O are the only two operations that are currently
3385 * supported with IOMAP_WRITE.
3387 WARN_ON(!IS_DAX(inode
) && !(flags
& IOMAP_DIRECT
));
3389 m_flags
= EXT4_GET_BLOCKS_CREATE_ZERO
;
3391 * We use i_size instead of i_disksize here because delalloc writeback
3392 * can complete at any point during the I/O and subsequently push the
3393 * i_disksize out to i_size. This could be beyond where direct I/O is
3394 * happening and thus expose allocated blocks to direct I/O reads.
3396 else if ((map
->m_lblk
* (1 << blkbits
)) >= i_size_read(inode
))
3397 m_flags
= EXT4_GET_BLOCKS_CREATE
;
3398 else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3399 m_flags
= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
3401 ret
= ext4_map_blocks(handle
, inode
, map
, m_flags
);
3404 * We cannot fill holes in indirect tree based inodes as that could
3405 * expose stale data in the case of a crash. Use the magic error code
3406 * to fallback to buffered I/O.
3408 if (!m_flags
&& !ret
)
3411 ext4_journal_stop(handle
);
3412 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
3419 static int ext4_iomap_begin(struct inode
*inode
, loff_t offset
, loff_t length
,
3420 unsigned flags
, struct iomap
*iomap
, struct iomap
*srcmap
)
3423 struct ext4_map_blocks map
;
3424 u8 blkbits
= inode
->i_blkbits
;
3426 if ((offset
>> blkbits
) > EXT4_MAX_LOGICAL_BLOCK
)
3429 if (WARN_ON_ONCE(ext4_has_inline_data(inode
)))
3433 * Calculate the first and last logical blocks respectively.
3435 map
.m_lblk
= offset
>> blkbits
;
3436 map
.m_len
= min_t(loff_t
, (offset
+ length
- 1) >> blkbits
,
3437 EXT4_MAX_LOGICAL_BLOCK
) - map
.m_lblk
+ 1;
3439 if (flags
& IOMAP_WRITE
)
3440 ret
= ext4_iomap_alloc(inode
, &map
, flags
);
3442 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3447 ext4_set_iomap(inode
, iomap
, &map
, offset
, length
);
3452 static int ext4_iomap_overwrite_begin(struct inode
*inode
, loff_t offset
,
3453 loff_t length
, unsigned flags
, struct iomap
*iomap
,
3454 struct iomap
*srcmap
)
3459 * Even for writes we don't need to allocate blocks, so just pretend
3460 * we are reading to save overhead of starting a transaction.
3462 flags
&= ~IOMAP_WRITE
;
3463 ret
= ext4_iomap_begin(inode
, offset
, length
, flags
, iomap
, srcmap
);
3464 WARN_ON_ONCE(iomap
->type
!= IOMAP_MAPPED
);
3468 static int ext4_iomap_end(struct inode
*inode
, loff_t offset
, loff_t length
,
3469 ssize_t written
, unsigned flags
, struct iomap
*iomap
)
3472 * Check to see whether an error occurred while writing out the data to
3473 * the allocated blocks. If so, return the magic error code so that we
3474 * fallback to buffered I/O and attempt to complete the remainder of
3475 * the I/O. Any blocks that may have been allocated in preparation for
3476 * the direct I/O will be reused during buffered I/O.
3478 if (flags
& (IOMAP_WRITE
| IOMAP_DIRECT
) && written
== 0)
3484 const struct iomap_ops ext4_iomap_ops
= {
3485 .iomap_begin
= ext4_iomap_begin
,
3486 .iomap_end
= ext4_iomap_end
,
3489 const struct iomap_ops ext4_iomap_overwrite_ops
= {
3490 .iomap_begin
= ext4_iomap_overwrite_begin
,
3491 .iomap_end
= ext4_iomap_end
,
3494 static bool ext4_iomap_is_delalloc(struct inode
*inode
,
3495 struct ext4_map_blocks
*map
)
3497 struct extent_status es
;
3498 ext4_lblk_t offset
= 0, end
= map
->m_lblk
+ map
->m_len
- 1;
3500 ext4_es_find_extent_range(inode
, &ext4_es_is_delayed
,
3501 map
->m_lblk
, end
, &es
);
3503 if (!es
.es_len
|| es
.es_lblk
> end
)
3506 if (es
.es_lblk
> map
->m_lblk
) {
3507 map
->m_len
= es
.es_lblk
- map
->m_lblk
;
3511 offset
= map
->m_lblk
- es
.es_lblk
;
3512 map
->m_len
= es
.es_len
- offset
;
3517 static int ext4_iomap_begin_report(struct inode
*inode
, loff_t offset
,
3518 loff_t length
, unsigned int flags
,
3519 struct iomap
*iomap
, struct iomap
*srcmap
)
3522 bool delalloc
= false;
3523 struct ext4_map_blocks map
;
3524 u8 blkbits
= inode
->i_blkbits
;
3526 if ((offset
>> blkbits
) > EXT4_MAX_LOGICAL_BLOCK
)
3529 if (ext4_has_inline_data(inode
)) {
3530 ret
= ext4_inline_data_iomap(inode
, iomap
);
3531 if (ret
!= -EAGAIN
) {
3532 if (ret
== 0 && offset
>= iomap
->length
)
3539 * Calculate the first and last logical block respectively.
3541 map
.m_lblk
= offset
>> blkbits
;
3542 map
.m_len
= min_t(loff_t
, (offset
+ length
- 1) >> blkbits
,
3543 EXT4_MAX_LOGICAL_BLOCK
) - map
.m_lblk
+ 1;
3546 * Fiemap callers may call for offset beyond s_bitmap_maxbytes.
3547 * So handle it here itself instead of querying ext4_map_blocks().
3548 * Since ext4_map_blocks() will warn about it and will return
3551 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
3552 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3554 if (offset
>= sbi
->s_bitmap_maxbytes
) {
3560 ret
= ext4_map_blocks(NULL
, inode
, &map
, 0);
3564 delalloc
= ext4_iomap_is_delalloc(inode
, &map
);
3567 ext4_set_iomap(inode
, iomap
, &map
, offset
, length
);
3568 if (delalloc
&& iomap
->type
== IOMAP_HOLE
)
3569 iomap
->type
= IOMAP_DELALLOC
;
3574 const struct iomap_ops ext4_iomap_report_ops
= {
3575 .iomap_begin
= ext4_iomap_begin_report
,
3579 * Pages can be marked dirty completely asynchronously from ext4's journalling
3580 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3581 * much here because ->set_page_dirty is called under VFS locks. The page is
3582 * not necessarily locked.
3584 * We cannot just dirty the page and leave attached buffers clean, because the
3585 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3586 * or jbddirty because all the journalling code will explode.
3588 * So what we do is to mark the page "pending dirty" and next time writepage
3589 * is called, propagate that into the buffers appropriately.
3591 static int ext4_journalled_set_page_dirty(struct page
*page
)
3593 SetPageChecked(page
);
3594 return __set_page_dirty_nobuffers(page
);
3597 static int ext4_set_page_dirty(struct page
*page
)
3599 WARN_ON_ONCE(!PageLocked(page
) && !PageDirty(page
));
3600 WARN_ON_ONCE(!page_has_buffers(page
));
3601 return __set_page_dirty_buffers(page
);
3604 static const struct address_space_operations ext4_aops
= {
3605 .readpage
= ext4_readpage
,
3606 .readahead
= ext4_readahead
,
3607 .writepage
= ext4_writepage
,
3608 .writepages
= ext4_writepages
,
3609 .write_begin
= ext4_write_begin
,
3610 .write_end
= ext4_write_end
,
3611 .set_page_dirty
= ext4_set_page_dirty
,
3613 .invalidatepage
= ext4_invalidatepage
,
3614 .releasepage
= ext4_releasepage
,
3615 .direct_IO
= noop_direct_IO
,
3616 .migratepage
= buffer_migrate_page
,
3617 .is_partially_uptodate
= block_is_partially_uptodate
,
3618 .error_remove_page
= generic_error_remove_page
,
3621 static const struct address_space_operations ext4_journalled_aops
= {
3622 .readpage
= ext4_readpage
,
3623 .readahead
= ext4_readahead
,
3624 .writepage
= ext4_writepage
,
3625 .writepages
= ext4_writepages
,
3626 .write_begin
= ext4_write_begin
,
3627 .write_end
= ext4_journalled_write_end
,
3628 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3630 .invalidatepage
= ext4_journalled_invalidatepage
,
3631 .releasepage
= ext4_releasepage
,
3632 .direct_IO
= noop_direct_IO
,
3633 .is_partially_uptodate
= block_is_partially_uptodate
,
3634 .error_remove_page
= generic_error_remove_page
,
3637 static const struct address_space_operations ext4_da_aops
= {
3638 .readpage
= ext4_readpage
,
3639 .readahead
= ext4_readahead
,
3640 .writepage
= ext4_writepage
,
3641 .writepages
= ext4_writepages
,
3642 .write_begin
= ext4_da_write_begin
,
3643 .write_end
= ext4_da_write_end
,
3644 .set_page_dirty
= ext4_set_page_dirty
,
3646 .invalidatepage
= ext4_invalidatepage
,
3647 .releasepage
= ext4_releasepage
,
3648 .direct_IO
= noop_direct_IO
,
3649 .migratepage
= buffer_migrate_page
,
3650 .is_partially_uptodate
= block_is_partially_uptodate
,
3651 .error_remove_page
= generic_error_remove_page
,
3654 static const struct address_space_operations ext4_dax_aops
= {
3655 .writepages
= ext4_dax_writepages
,
3656 .direct_IO
= noop_direct_IO
,
3657 .set_page_dirty
= noop_set_page_dirty
,
3659 .invalidatepage
= noop_invalidatepage
,
3662 void ext4_set_aops(struct inode
*inode
)
3664 switch (ext4_inode_journal_mode(inode
)) {
3665 case EXT4_INODE_ORDERED_DATA_MODE
:
3666 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3668 case EXT4_INODE_JOURNAL_DATA_MODE
:
3669 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3675 inode
->i_mapping
->a_ops
= &ext4_dax_aops
;
3676 else if (test_opt(inode
->i_sb
, DELALLOC
))
3677 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3679 inode
->i_mapping
->a_ops
= &ext4_aops
;
3682 static int __ext4_block_zero_page_range(handle_t
*handle
,
3683 struct address_space
*mapping
, loff_t from
, loff_t length
)
3685 ext4_fsblk_t index
= from
>> PAGE_SHIFT
;
3686 unsigned offset
= from
& (PAGE_SIZE
-1);
3687 unsigned blocksize
, pos
;
3689 struct inode
*inode
= mapping
->host
;
3690 struct buffer_head
*bh
;
3694 page
= find_or_create_page(mapping
, from
>> PAGE_SHIFT
,
3695 mapping_gfp_constraint(mapping
, ~__GFP_FS
));
3699 blocksize
= inode
->i_sb
->s_blocksize
;
3701 iblock
= index
<< (PAGE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3703 if (!page_has_buffers(page
))
3704 create_empty_buffers(page
, blocksize
, 0);
3706 /* Find the buffer that contains "offset" */
3707 bh
= page_buffers(page
);
3709 while (offset
>= pos
) {
3710 bh
= bh
->b_this_page
;
3714 if (buffer_freed(bh
)) {
3715 BUFFER_TRACE(bh
, "freed: skip");
3718 if (!buffer_mapped(bh
)) {
3719 BUFFER_TRACE(bh
, "unmapped");
3720 ext4_get_block(inode
, iblock
, bh
, 0);
3721 /* unmapped? It's a hole - nothing to do */
3722 if (!buffer_mapped(bh
)) {
3723 BUFFER_TRACE(bh
, "still unmapped");
3728 /* Ok, it's mapped. Make sure it's up-to-date */
3729 if (PageUptodate(page
))
3730 set_buffer_uptodate(bh
);
3732 if (!buffer_uptodate(bh
)) {
3734 ll_rw_block(REQ_OP_READ
, 0, 1, &bh
);
3736 /* Uhhuh. Read error. Complain and punt. */
3737 if (!buffer_uptodate(bh
))
3739 if (S_ISREG(inode
->i_mode
) && IS_ENCRYPTED(inode
)) {
3740 /* We expect the key to be set. */
3741 BUG_ON(!fscrypt_has_encryption_key(inode
));
3742 err
= fscrypt_decrypt_pagecache_blocks(page
, blocksize
,
3745 clear_buffer_uptodate(bh
);
3750 if (ext4_should_journal_data(inode
)) {
3751 BUFFER_TRACE(bh
, "get write access");
3752 err
= ext4_journal_get_write_access(handle
, bh
);
3756 zero_user(page
, offset
, length
);
3757 BUFFER_TRACE(bh
, "zeroed end of block");
3759 if (ext4_should_journal_data(inode
)) {
3760 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3763 mark_buffer_dirty(bh
);
3764 if (ext4_should_order_data(inode
))
3765 err
= ext4_jbd2_inode_add_write(handle
, inode
, from
,
3776 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3777 * starting from file offset 'from'. The range to be zero'd must
3778 * be contained with in one block. If the specified range exceeds
3779 * the end of the block it will be shortened to end of the block
3780 * that cooresponds to 'from'
3782 static int ext4_block_zero_page_range(handle_t
*handle
,
3783 struct address_space
*mapping
, loff_t from
, loff_t length
)
3785 struct inode
*inode
= mapping
->host
;
3786 unsigned offset
= from
& (PAGE_SIZE
-1);
3787 unsigned blocksize
= inode
->i_sb
->s_blocksize
;
3788 unsigned max
= blocksize
- (offset
& (blocksize
- 1));
3791 * correct length if it does not fall between
3792 * 'from' and the end of the block
3794 if (length
> max
|| length
< 0)
3797 if (IS_DAX(inode
)) {
3798 return iomap_zero_range(inode
, from
, length
, NULL
,
3801 return __ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3805 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3806 * up to the end of the block which corresponds to `from'.
3807 * This required during truncate. We need to physically zero the tail end
3808 * of that block so it doesn't yield old data if the file is later grown.
3810 static int ext4_block_truncate_page(handle_t
*handle
,
3811 struct address_space
*mapping
, loff_t from
)
3813 unsigned offset
= from
& (PAGE_SIZE
-1);
3816 struct inode
*inode
= mapping
->host
;
3818 /* If we are processing an encrypted inode during orphan list handling */
3819 if (IS_ENCRYPTED(inode
) && !fscrypt_has_encryption_key(inode
))
3822 blocksize
= inode
->i_sb
->s_blocksize
;
3823 length
= blocksize
- (offset
& (blocksize
- 1));
3825 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3828 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
3829 loff_t lstart
, loff_t length
)
3831 struct super_block
*sb
= inode
->i_sb
;
3832 struct address_space
*mapping
= inode
->i_mapping
;
3833 unsigned partial_start
, partial_end
;
3834 ext4_fsblk_t start
, end
;
3835 loff_t byte_end
= (lstart
+ length
- 1);
3838 partial_start
= lstart
& (sb
->s_blocksize
- 1);
3839 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
3841 start
= lstart
>> sb
->s_blocksize_bits
;
3842 end
= byte_end
>> sb
->s_blocksize_bits
;
3844 /* Handle partial zero within the single block */
3846 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
3847 err
= ext4_block_zero_page_range(handle
, mapping
,
3851 /* Handle partial zero out on the start of the range */
3852 if (partial_start
) {
3853 err
= ext4_block_zero_page_range(handle
, mapping
,
3854 lstart
, sb
->s_blocksize
);
3858 /* Handle partial zero out on the end of the range */
3859 if (partial_end
!= sb
->s_blocksize
- 1)
3860 err
= ext4_block_zero_page_range(handle
, mapping
,
3861 byte_end
- partial_end
,
3866 int ext4_can_truncate(struct inode
*inode
)
3868 if (S_ISREG(inode
->i_mode
))
3870 if (S_ISDIR(inode
->i_mode
))
3872 if (S_ISLNK(inode
->i_mode
))
3873 return !ext4_inode_is_fast_symlink(inode
);
3878 * We have to make sure i_disksize gets properly updated before we truncate
3879 * page cache due to hole punching or zero range. Otherwise i_disksize update
3880 * can get lost as it may have been postponed to submission of writeback but
3881 * that will never happen after we truncate page cache.
3883 int ext4_update_disksize_before_punch(struct inode
*inode
, loff_t offset
,
3889 loff_t size
= i_size_read(inode
);
3891 WARN_ON(!inode_is_locked(inode
));
3892 if (offset
> size
|| offset
+ len
< size
)
3895 if (EXT4_I(inode
)->i_disksize
>= size
)
3898 handle
= ext4_journal_start(inode
, EXT4_HT_MISC
, 1);
3900 return PTR_ERR(handle
);
3901 ext4_update_i_disksize(inode
, size
);
3902 ret
= ext4_mark_inode_dirty(handle
, inode
);
3903 ext4_journal_stop(handle
);
3908 static void ext4_wait_dax_page(struct ext4_inode_info
*ei
)
3910 up_write(&ei
->i_mmap_sem
);
3912 down_write(&ei
->i_mmap_sem
);
3915 int ext4_break_layouts(struct inode
*inode
)
3917 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3921 if (WARN_ON_ONCE(!rwsem_is_locked(&ei
->i_mmap_sem
)))
3925 page
= dax_layout_busy_page(inode
->i_mapping
);
3929 error
= ___wait_var_event(&page
->_refcount
,
3930 atomic_read(&page
->_refcount
) == 1,
3931 TASK_INTERRUPTIBLE
, 0, 0,
3932 ext4_wait_dax_page(ei
));
3933 } while (error
== 0);
3939 * ext4_punch_hole: punches a hole in a file by releasing the blocks
3940 * associated with the given offset and length
3942 * @inode: File inode
3943 * @offset: The offset where the hole will begin
3944 * @len: The length of the hole
3946 * Returns: 0 on success or negative on failure
3949 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
3951 struct super_block
*sb
= inode
->i_sb
;
3952 ext4_lblk_t first_block
, stop_block
;
3953 struct address_space
*mapping
= inode
->i_mapping
;
3954 loff_t first_block_offset
, last_block_offset
;
3956 unsigned int credits
;
3957 int ret
= 0, ret2
= 0;
3959 trace_ext4_punch_hole(inode
, offset
, length
, 0);
3961 ext4_clear_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
);
3962 if (ext4_has_inline_data(inode
)) {
3963 down_write(&EXT4_I(inode
)->i_mmap_sem
);
3964 ret
= ext4_convert_inline_data(inode
);
3965 up_write(&EXT4_I(inode
)->i_mmap_sem
);
3971 * Write out all dirty pages to avoid race conditions
3972 * Then release them.
3974 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3975 ret
= filemap_write_and_wait_range(mapping
, offset
,
3976 offset
+ length
- 1);
3983 /* No need to punch hole beyond i_size */
3984 if (offset
>= inode
->i_size
)
3988 * If the hole extends beyond i_size, set the hole
3989 * to end after the page that contains i_size
3991 if (offset
+ length
> inode
->i_size
) {
3992 length
= inode
->i_size
+
3993 PAGE_SIZE
- (inode
->i_size
& (PAGE_SIZE
- 1)) -
3997 if (offset
& (sb
->s_blocksize
- 1) ||
3998 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
4000 * Attach jinode to inode for jbd2 if we do any zeroing of
4003 ret
= ext4_inode_attach_jinode(inode
);
4009 /* Wait all existing dio workers, newcomers will block on i_mutex */
4010 inode_dio_wait(inode
);
4013 * Prevent page faults from reinstantiating pages we have released from
4016 down_write(&EXT4_I(inode
)->i_mmap_sem
);
4018 ret
= ext4_break_layouts(inode
);
4022 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
4023 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
4025 /* Now release the pages and zero block aligned part of pages*/
4026 if (last_block_offset
> first_block_offset
) {
4027 ret
= ext4_update_disksize_before_punch(inode
, offset
, length
);
4030 truncate_pagecache_range(inode
, first_block_offset
,
4034 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4035 credits
= ext4_writepage_trans_blocks(inode
);
4037 credits
= ext4_blocks_for_truncate(inode
);
4038 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4039 if (IS_ERR(handle
)) {
4040 ret
= PTR_ERR(handle
);
4041 ext4_std_error(sb
, ret
);
4045 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
4050 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
4051 EXT4_BLOCK_SIZE_BITS(sb
);
4052 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
4054 /* If there are blocks to remove, do it */
4055 if (stop_block
> first_block
) {
4057 down_write(&EXT4_I(inode
)->i_data_sem
);
4058 ext4_discard_preallocations(inode
, 0);
4060 ret
= ext4_es_remove_extent(inode
, first_block
,
4061 stop_block
- first_block
);
4063 up_write(&EXT4_I(inode
)->i_data_sem
);
4067 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4068 ret
= ext4_ext_remove_space(inode
, first_block
,
4071 ret
= ext4_ind_remove_space(handle
, inode
, first_block
,
4074 up_write(&EXT4_I(inode
)->i_data_sem
);
4077 ext4_handle_sync(handle
);
4079 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4080 ret2
= ext4_mark_inode_dirty(handle
, inode
);
4084 ext4_update_inode_fsync_trans(handle
, inode
, 1);
4086 ext4_journal_stop(handle
);
4088 up_write(&EXT4_I(inode
)->i_mmap_sem
);
4090 inode_unlock(inode
);
4094 int ext4_inode_attach_jinode(struct inode
*inode
)
4096 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4097 struct jbd2_inode
*jinode
;
4099 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
4102 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
4103 spin_lock(&inode
->i_lock
);
4106 spin_unlock(&inode
->i_lock
);
4109 ei
->jinode
= jinode
;
4110 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
4113 spin_unlock(&inode
->i_lock
);
4114 if (unlikely(jinode
!= NULL
))
4115 jbd2_free_inode(jinode
);
4122 * We block out ext4_get_block() block instantiations across the entire
4123 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
4124 * simultaneously on behalf of the same inode.
4126 * As we work through the truncate and commit bits of it to the journal there
4127 * is one core, guiding principle: the file's tree must always be consistent on
4128 * disk. We must be able to restart the truncate after a crash.
4130 * The file's tree may be transiently inconsistent in memory (although it
4131 * probably isn't), but whenever we close off and commit a journal transaction,
4132 * the contents of (the filesystem + the journal) must be consistent and
4133 * restartable. It's pretty simple, really: bottom up, right to left (although
4134 * left-to-right works OK too).
4136 * Note that at recovery time, journal replay occurs *before* the restart of
4137 * truncate against the orphan inode list.
4139 * The committed inode has the new, desired i_size (which is the same as
4140 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
4141 * that this inode's truncate did not complete and it will again call
4142 * ext4_truncate() to have another go. So there will be instantiated blocks
4143 * to the right of the truncation point in a crashed ext4 filesystem. But
4144 * that's fine - as long as they are linked from the inode, the post-crash
4145 * ext4_truncate() run will find them and release them.
4147 int ext4_truncate(struct inode
*inode
)
4149 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4150 unsigned int credits
;
4153 struct address_space
*mapping
= inode
->i_mapping
;
4156 * There is a possibility that we're either freeing the inode
4157 * or it's a completely new inode. In those cases we might not
4158 * have i_mutex locked because it's not necessary.
4160 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
4161 WARN_ON(!inode_is_locked(inode
));
4162 trace_ext4_truncate_enter(inode
);
4164 if (!ext4_can_truncate(inode
))
4167 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
4168 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
4170 if (ext4_has_inline_data(inode
)) {
4173 err
= ext4_inline_data_truncate(inode
, &has_inline
);
4174 if (err
|| has_inline
)
4178 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
4179 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
4180 if (ext4_inode_attach_jinode(inode
) < 0)
4184 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4185 credits
= ext4_writepage_trans_blocks(inode
);
4187 credits
= ext4_blocks_for_truncate(inode
);
4189 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
4190 if (IS_ERR(handle
)) {
4191 err
= PTR_ERR(handle
);
4195 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
4196 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
4199 * We add the inode to the orphan list, so that if this
4200 * truncate spans multiple transactions, and we crash, we will
4201 * resume the truncate when the filesystem recovers. It also
4202 * marks the inode dirty, to catch the new size.
4204 * Implication: the file must always be in a sane, consistent
4205 * truncatable state while each transaction commits.
4207 err
= ext4_orphan_add(handle
, inode
);
4211 down_write(&EXT4_I(inode
)->i_data_sem
);
4213 ext4_discard_preallocations(inode
, 0);
4215 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4216 err
= ext4_ext_truncate(handle
, inode
);
4218 ext4_ind_truncate(handle
, inode
);
4220 up_write(&ei
->i_data_sem
);
4225 ext4_handle_sync(handle
);
4229 * If this was a simple ftruncate() and the file will remain alive,
4230 * then we need to clear up the orphan record which we created above.
4231 * However, if this was a real unlink then we were called by
4232 * ext4_evict_inode(), and we allow that function to clean up the
4233 * orphan info for us.
4236 ext4_orphan_del(handle
, inode
);
4238 inode
->i_mtime
= inode
->i_ctime
= current_time(inode
);
4239 err2
= ext4_mark_inode_dirty(handle
, inode
);
4240 if (unlikely(err2
&& !err
))
4242 ext4_journal_stop(handle
);
4245 trace_ext4_truncate_exit(inode
);
4250 * ext4_get_inode_loc returns with an extra refcount against the inode's
4251 * underlying buffer_head on success. If 'in_mem' is true, we have all
4252 * data in memory that is needed to recreate the on-disk version of this
4255 static int __ext4_get_inode_loc(struct inode
*inode
,
4256 struct ext4_iloc
*iloc
, int in_mem
)
4258 struct ext4_group_desc
*gdp
;
4259 struct buffer_head
*bh
;
4260 struct super_block
*sb
= inode
->i_sb
;
4262 struct blk_plug plug
;
4263 int inodes_per_block
, inode_offset
;
4266 if (inode
->i_ino
< EXT4_ROOT_INO
||
4267 inode
->i_ino
> le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
))
4268 return -EFSCORRUPTED
;
4270 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
4271 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
4276 * Figure out the offset within the block group inode table
4278 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4279 inode_offset
= ((inode
->i_ino
- 1) %
4280 EXT4_INODES_PER_GROUP(sb
));
4281 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4282 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4284 bh
= sb_getblk(sb
, block
);
4287 if (ext4_simulate_fail(sb
, EXT4_SIM_INODE_EIO
))
4289 if (!buffer_uptodate(bh
)) {
4293 * If the buffer has the write error flag, we have failed
4294 * to write out another inode in the same block. In this
4295 * case, we don't have to read the block because we may
4296 * read the old inode data successfully.
4298 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4299 set_buffer_uptodate(bh
);
4301 if (buffer_uptodate(bh
)) {
4302 /* someone brought it uptodate while we waited */
4308 * If we have all information of the inode in memory and this
4309 * is the only valid inode in the block, we need not read the
4313 struct buffer_head
*bitmap_bh
;
4316 start
= inode_offset
& ~(inodes_per_block
- 1);
4318 /* Is the inode bitmap in cache? */
4319 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4320 if (unlikely(!bitmap_bh
))
4324 * If the inode bitmap isn't in cache then the
4325 * optimisation may end up performing two reads instead
4326 * of one, so skip it.
4328 if (!buffer_uptodate(bitmap_bh
)) {
4332 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4333 if (i
== inode_offset
)
4335 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4339 if (i
== start
+ inodes_per_block
) {
4340 /* all other inodes are free, so skip I/O */
4341 memset(bh
->b_data
, 0, bh
->b_size
);
4342 set_buffer_uptodate(bh
);
4350 * If we need to do any I/O, try to pre-readahead extra
4351 * blocks from the inode table.
4353 blk_start_plug(&plug
);
4354 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4355 ext4_fsblk_t b
, end
, table
;
4357 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4359 table
= ext4_inode_table(sb
, gdp
);
4360 /* s_inode_readahead_blks is always a power of 2 */
4361 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4365 num
= EXT4_INODES_PER_GROUP(sb
);
4366 if (ext4_has_group_desc_csum(sb
))
4367 num
-= ext4_itable_unused_count(sb
, gdp
);
4368 table
+= num
/ inodes_per_block
;
4372 sb_breadahead_unmovable(sb
, b
++);
4376 * There are other valid inodes in the buffer, this inode
4377 * has in-inode xattrs, or we don't have this inode in memory.
4378 * Read the block from disk.
4380 trace_ext4_load_inode(inode
);
4382 bh
->b_end_io
= end_buffer_read_sync
;
4383 submit_bh(REQ_OP_READ
, REQ_META
| REQ_PRIO
, bh
);
4384 blk_finish_plug(&plug
);
4386 if (!buffer_uptodate(bh
)) {
4388 ext4_error_inode_block(inode
, block
, EIO
,
4389 "unable to read itable block");
4399 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4401 /* We have all inode data except xattrs in memory here. */
4402 return __ext4_get_inode_loc(inode
, iloc
,
4403 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4406 static bool ext4_should_enable_dax(struct inode
*inode
)
4408 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4410 if (test_opt2(inode
->i_sb
, DAX_NEVER
))
4412 if (!S_ISREG(inode
->i_mode
))
4414 if (ext4_should_journal_data(inode
))
4416 if (ext4_has_inline_data(inode
))
4418 if (ext4_test_inode_flag(inode
, EXT4_INODE_ENCRYPT
))
4420 if (ext4_test_inode_flag(inode
, EXT4_INODE_VERITY
))
4422 if (!test_bit(EXT4_FLAGS_BDEV_IS_DAX
, &sbi
->s_ext4_flags
))
4424 if (test_opt(inode
->i_sb
, DAX_ALWAYS
))
4427 return ext4_test_inode_flag(inode
, EXT4_INODE_DAX
);
4430 void ext4_set_inode_flags(struct inode
*inode
, bool init
)
4432 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4433 unsigned int new_fl
= 0;
4435 WARN_ON_ONCE(IS_DAX(inode
) && init
);
4437 if (flags
& EXT4_SYNC_FL
)
4439 if (flags
& EXT4_APPEND_FL
)
4441 if (flags
& EXT4_IMMUTABLE_FL
)
4442 new_fl
|= S_IMMUTABLE
;
4443 if (flags
& EXT4_NOATIME_FL
)
4444 new_fl
|= S_NOATIME
;
4445 if (flags
& EXT4_DIRSYNC_FL
)
4446 new_fl
|= S_DIRSYNC
;
4448 /* Because of the way inode_set_flags() works we must preserve S_DAX
4449 * here if already set. */
4450 new_fl
|= (inode
->i_flags
& S_DAX
);
4451 if (init
&& ext4_should_enable_dax(inode
))
4454 if (flags
& EXT4_ENCRYPT_FL
)
4455 new_fl
|= S_ENCRYPTED
;
4456 if (flags
& EXT4_CASEFOLD_FL
)
4457 new_fl
|= S_CASEFOLD
;
4458 if (flags
& EXT4_VERITY_FL
)
4460 inode_set_flags(inode
, new_fl
,
4461 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
|S_DAX
|
4462 S_ENCRYPTED
|S_CASEFOLD
|S_VERITY
);
4465 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4466 struct ext4_inode_info
*ei
)
4469 struct inode
*inode
= &(ei
->vfs_inode
);
4470 struct super_block
*sb
= inode
->i_sb
;
4472 if (ext4_has_feature_huge_file(sb
)) {
4473 /* we are using combined 48 bit field */
4474 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4475 le32_to_cpu(raw_inode
->i_blocks_lo
);
4476 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4477 /* i_blocks represent file system block size */
4478 return i_blocks
<< (inode
->i_blkbits
- 9);
4483 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4487 static inline int ext4_iget_extra_inode(struct inode
*inode
,
4488 struct ext4_inode
*raw_inode
,
4489 struct ext4_inode_info
*ei
)
4491 __le32
*magic
= (void *)raw_inode
+
4492 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4494 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
+ sizeof(__le32
) <=
4495 EXT4_INODE_SIZE(inode
->i_sb
) &&
4496 *magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4497 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4498 return ext4_find_inline_data_nolock(inode
);
4500 EXT4_I(inode
)->i_inline_off
= 0;
4504 int ext4_get_projid(struct inode
*inode
, kprojid_t
*projid
)
4506 if (!ext4_has_feature_project(inode
->i_sb
))
4508 *projid
= EXT4_I(inode
)->i_projid
;
4513 * ext4 has self-managed i_version for ea inodes, it stores the lower 32bit of
4514 * refcount in i_version, so use raw values if inode has EXT4_EA_INODE_FL flag
4517 static inline void ext4_inode_set_iversion_queried(struct inode
*inode
, u64 val
)
4519 if (unlikely(EXT4_I(inode
)->i_flags
& EXT4_EA_INODE_FL
))
4520 inode_set_iversion_raw(inode
, val
);
4522 inode_set_iversion_queried(inode
, val
);
4524 static inline u64
ext4_inode_peek_iversion(const struct inode
*inode
)
4526 if (unlikely(EXT4_I(inode
)->i_flags
& EXT4_EA_INODE_FL
))
4527 return inode_peek_iversion_raw(inode
);
4529 return inode_peek_iversion(inode
);
4532 struct inode
*__ext4_iget(struct super_block
*sb
, unsigned long ino
,
4533 ext4_iget_flags flags
, const char *function
,
4536 struct ext4_iloc iloc
;
4537 struct ext4_inode
*raw_inode
;
4538 struct ext4_inode_info
*ei
;
4539 struct inode
*inode
;
4540 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4548 if ((!(flags
& EXT4_IGET_SPECIAL
) &&
4549 (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)) ||
4550 (ino
< EXT4_ROOT_INO
) ||
4551 (ino
> le32_to_cpu(EXT4_SB(sb
)->s_es
->s_inodes_count
))) {
4552 if (flags
& EXT4_IGET_HANDLE
)
4553 return ERR_PTR(-ESTALE
);
4554 __ext4_error(sb
, function
, line
, EFSCORRUPTED
, 0,
4555 "inode #%lu: comm %s: iget: illegal inode #",
4556 ino
, current
->comm
);
4557 return ERR_PTR(-EFSCORRUPTED
);
4560 inode
= iget_locked(sb
, ino
);
4562 return ERR_PTR(-ENOMEM
);
4563 if (!(inode
->i_state
& I_NEW
))
4569 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4572 raw_inode
= ext4_raw_inode(&iloc
);
4574 if ((ino
== EXT4_ROOT_INO
) && (raw_inode
->i_links_count
== 0)) {
4575 ext4_error_inode(inode
, function
, line
, 0,
4576 "iget: root inode unallocated");
4577 ret
= -EFSCORRUPTED
;
4581 if ((flags
& EXT4_IGET_HANDLE
) &&
4582 (raw_inode
->i_links_count
== 0) && (raw_inode
->i_mode
== 0)) {
4587 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4588 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4589 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4590 EXT4_INODE_SIZE(inode
->i_sb
) ||
4591 (ei
->i_extra_isize
& 3)) {
4592 ext4_error_inode(inode
, function
, line
, 0,
4593 "iget: bad extra_isize %u "
4596 EXT4_INODE_SIZE(inode
->i_sb
));
4597 ret
= -EFSCORRUPTED
;
4601 ei
->i_extra_isize
= 0;
4603 /* Precompute checksum seed for inode metadata */
4604 if (ext4_has_metadata_csum(sb
)) {
4605 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4607 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4608 __le32 gen
= raw_inode
->i_generation
;
4609 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4611 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4615 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
) ||
4616 ext4_simulate_fail(sb
, EXT4_SIM_INODE_CRC
)) {
4617 ext4_error_inode_err(inode
, function
, line
, 0, EFSBADCRC
,
4618 "iget: checksum invalid");
4623 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4624 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4625 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4626 if (ext4_has_feature_project(sb
) &&
4627 EXT4_INODE_SIZE(sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
4628 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
4629 i_projid
= (projid_t
)le32_to_cpu(raw_inode
->i_projid
);
4631 i_projid
= EXT4_DEF_PROJID
;
4633 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4634 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4635 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4637 i_uid_write(inode
, i_uid
);
4638 i_gid_write(inode
, i_gid
);
4639 ei
->i_projid
= make_kprojid(&init_user_ns
, i_projid
);
4640 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4642 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4643 ei
->i_inline_off
= 0;
4644 ei
->i_dir_start_lookup
= 0;
4645 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4646 /* We now have enough fields to check if the inode was active or not.
4647 * This is needed because nfsd might try to access dead inodes
4648 * the test is that same one that e2fsck uses
4649 * NeilBrown 1999oct15
4651 if (inode
->i_nlink
== 0) {
4652 if ((inode
->i_mode
== 0 ||
4653 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4654 ino
!= EXT4_BOOT_LOADER_INO
) {
4655 /* this inode is deleted */
4659 /* The only unlinked inodes we let through here have
4660 * valid i_mode and are being read by the orphan
4661 * recovery code: that's fine, we're about to complete
4662 * the process of deleting those.
4663 * OR it is the EXT4_BOOT_LOADER_INO which is
4664 * not initialized on a new filesystem. */
4666 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4667 ext4_set_inode_flags(inode
, true);
4668 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4669 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4670 if (ext4_has_feature_64bit(sb
))
4672 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4673 inode
->i_size
= ext4_isize(sb
, raw_inode
);
4674 if ((size
= i_size_read(inode
)) < 0) {
4675 ext4_error_inode(inode
, function
, line
, 0,
4676 "iget: bad i_size value: %lld", size
);
4677 ret
= -EFSCORRUPTED
;
4681 * If dir_index is not enabled but there's dir with INDEX flag set,
4682 * we'd normally treat htree data as empty space. But with metadata
4683 * checksumming that corrupts checksums so forbid that.
4685 if (!ext4_has_feature_dir_index(sb
) && ext4_has_metadata_csum(sb
) &&
4686 ext4_test_inode_flag(inode
, EXT4_INODE_INDEX
)) {
4687 ext4_error_inode(inode
, function
, line
, 0,
4688 "iget: Dir with htree data on filesystem without dir_index feature.");
4689 ret
= -EFSCORRUPTED
;
4692 ei
->i_disksize
= inode
->i_size
;
4694 ei
->i_reserved_quota
= 0;
4696 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4697 ei
->i_block_group
= iloc
.block_group
;
4698 ei
->i_last_alloc_group
= ~0;
4700 * NOTE! The in-memory inode i_data array is in little-endian order
4701 * even on big-endian machines: we do NOT byteswap the block numbers!
4703 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4704 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4705 INIT_LIST_HEAD(&ei
->i_orphan
);
4708 * Set transaction id's of transactions that have to be committed
4709 * to finish f[data]sync. We set them to currently running transaction
4710 * as we cannot be sure that the inode or some of its metadata isn't
4711 * part of the transaction - the inode could have been reclaimed and
4712 * now it is reread from disk.
4715 transaction_t
*transaction
;
4718 read_lock(&journal
->j_state_lock
);
4719 if (journal
->j_running_transaction
)
4720 transaction
= journal
->j_running_transaction
;
4722 transaction
= journal
->j_committing_transaction
;
4724 tid
= transaction
->t_tid
;
4726 tid
= journal
->j_commit_sequence
;
4727 read_unlock(&journal
->j_state_lock
);
4728 ei
->i_sync_tid
= tid
;
4729 ei
->i_datasync_tid
= tid
;
4732 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4733 if (ei
->i_extra_isize
== 0) {
4734 /* The extra space is currently unused. Use it. */
4735 BUILD_BUG_ON(sizeof(struct ext4_inode
) & 3);
4736 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4737 EXT4_GOOD_OLD_INODE_SIZE
;
4739 ret
= ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4745 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4746 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4747 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4748 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4750 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
4751 u64 ivers
= le32_to_cpu(raw_inode
->i_disk_version
);
4753 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4754 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4756 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4758 ext4_inode_set_iversion_queried(inode
, ivers
);
4762 if (ei
->i_file_acl
&&
4763 !ext4_inode_block_valid(inode
, ei
->i_file_acl
, 1)) {
4764 ext4_error_inode(inode
, function
, line
, 0,
4765 "iget: bad extended attribute block %llu",
4767 ret
= -EFSCORRUPTED
;
4769 } else if (!ext4_has_inline_data(inode
)) {
4770 /* validate the block references in the inode */
4771 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4772 (S_ISLNK(inode
->i_mode
) &&
4773 !ext4_inode_is_fast_symlink(inode
))) {
4774 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
4775 ret
= ext4_ext_check_inode(inode
);
4777 ret
= ext4_ind_check_inode(inode
);
4783 if (S_ISREG(inode
->i_mode
)) {
4784 inode
->i_op
= &ext4_file_inode_operations
;
4785 inode
->i_fop
= &ext4_file_operations
;
4786 ext4_set_aops(inode
);
4787 } else if (S_ISDIR(inode
->i_mode
)) {
4788 inode
->i_op
= &ext4_dir_inode_operations
;
4789 inode
->i_fop
= &ext4_dir_operations
;
4790 } else if (S_ISLNK(inode
->i_mode
)) {
4791 /* VFS does not allow setting these so must be corruption */
4792 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
)) {
4793 ext4_error_inode(inode
, function
, line
, 0,
4794 "iget: immutable or append flags "
4795 "not allowed on symlinks");
4796 ret
= -EFSCORRUPTED
;
4799 if (IS_ENCRYPTED(inode
)) {
4800 inode
->i_op
= &ext4_encrypted_symlink_inode_operations
;
4801 ext4_set_aops(inode
);
4802 } else if (ext4_inode_is_fast_symlink(inode
)) {
4803 inode
->i_link
= (char *)ei
->i_data
;
4804 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4805 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4806 sizeof(ei
->i_data
) - 1);
4808 inode
->i_op
= &ext4_symlink_inode_operations
;
4809 ext4_set_aops(inode
);
4811 inode_nohighmem(inode
);
4812 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4813 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4814 inode
->i_op
= &ext4_special_inode_operations
;
4815 if (raw_inode
->i_block
[0])
4816 init_special_inode(inode
, inode
->i_mode
,
4817 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4819 init_special_inode(inode
, inode
->i_mode
,
4820 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4821 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4822 make_bad_inode(inode
);
4824 ret
= -EFSCORRUPTED
;
4825 ext4_error_inode(inode
, function
, line
, 0,
4826 "iget: bogus i_mode (%o)", inode
->i_mode
);
4829 if (IS_CASEFOLDED(inode
) && !ext4_has_feature_casefold(inode
->i_sb
))
4830 ext4_error_inode(inode
, function
, line
, 0,
4831 "casefold flag without casefold feature");
4834 unlock_new_inode(inode
);
4840 return ERR_PTR(ret
);
4843 static int ext4_inode_blocks_set(handle_t
*handle
,
4844 struct ext4_inode
*raw_inode
,
4845 struct ext4_inode_info
*ei
)
4847 struct inode
*inode
= &(ei
->vfs_inode
);
4848 u64 i_blocks
= READ_ONCE(inode
->i_blocks
);
4849 struct super_block
*sb
= inode
->i_sb
;
4851 if (i_blocks
<= ~0U) {
4853 * i_blocks can be represented in a 32 bit variable
4854 * as multiple of 512 bytes
4856 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4857 raw_inode
->i_blocks_high
= 0;
4858 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4861 if (!ext4_has_feature_huge_file(sb
))
4864 if (i_blocks
<= 0xffffffffffffULL
) {
4866 * i_blocks can be represented in a 48 bit variable
4867 * as multiple of 512 bytes
4869 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4870 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4871 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4873 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4874 /* i_block is stored in file system block size */
4875 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4876 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4877 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4882 static void __ext4_update_other_inode_time(struct super_block
*sb
,
4883 unsigned long orig_ino
,
4885 struct ext4_inode
*raw_inode
)
4887 struct inode
*inode
;
4889 inode
= find_inode_by_ino_rcu(sb
, ino
);
4893 if ((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4895 ((inode
->i_state
& I_DIRTY_TIME
) == 0))
4898 spin_lock(&inode
->i_lock
);
4899 if (((inode
->i_state
& (I_FREEING
| I_WILL_FREE
| I_NEW
|
4900 I_DIRTY_INODE
)) == 0) &&
4901 (inode
->i_state
& I_DIRTY_TIME
)) {
4902 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4904 inode
->i_state
&= ~(I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
);
4905 spin_unlock(&inode
->i_lock
);
4907 spin_lock(&ei
->i_raw_lock
);
4908 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4909 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4910 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4911 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4912 spin_unlock(&ei
->i_raw_lock
);
4913 trace_ext4_other_inode_update_time(inode
, orig_ino
);
4916 spin_unlock(&inode
->i_lock
);
4920 * Opportunistically update the other time fields for other inodes in
4921 * the same inode table block.
4923 static void ext4_update_other_inodes_time(struct super_block
*sb
,
4924 unsigned long orig_ino
, char *buf
)
4927 int i
, inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
4928 int inode_size
= EXT4_INODE_SIZE(sb
);
4931 * Calculate the first inode in the inode table block. Inode
4932 * numbers are one-based. That is, the first inode in a block
4933 * (assuming 4k blocks and 256 byte inodes) is (n*16 + 1).
4935 ino
= ((orig_ino
- 1) & ~(inodes_per_block
- 1)) + 1;
4937 for (i
= 0; i
< inodes_per_block
; i
++, ino
++, buf
+= inode_size
) {
4938 if (ino
== orig_ino
)
4940 __ext4_update_other_inode_time(sb
, orig_ino
, ino
,
4941 (struct ext4_inode
*)buf
);
4947 * Post the struct inode info into an on-disk inode location in the
4948 * buffer-cache. This gobbles the caller's reference to the
4949 * buffer_head in the inode location struct.
4951 * The caller must have write access to iloc->bh.
4953 static int ext4_do_update_inode(handle_t
*handle
,
4954 struct inode
*inode
,
4955 struct ext4_iloc
*iloc
)
4957 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4958 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4959 struct buffer_head
*bh
= iloc
->bh
;
4960 struct super_block
*sb
= inode
->i_sb
;
4961 int err
= 0, rc
, block
;
4962 int need_datasync
= 0, set_large_file
= 0;
4967 spin_lock(&ei
->i_raw_lock
);
4969 /* For fields not tracked in the in-memory inode,
4970 * initialise them to zero for new inodes. */
4971 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4972 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4974 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4975 i_uid
= i_uid_read(inode
);
4976 i_gid
= i_gid_read(inode
);
4977 i_projid
= from_kprojid(&init_user_ns
, ei
->i_projid
);
4978 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4979 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4980 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4982 * Fix up interoperability with old kernels. Otherwise, old inodes get
4983 * re-used with the upper 16 bits of the uid/gid intact
4985 if (ei
->i_dtime
&& list_empty(&ei
->i_orphan
)) {
4986 raw_inode
->i_uid_high
= 0;
4987 raw_inode
->i_gid_high
= 0;
4989 raw_inode
->i_uid_high
=
4990 cpu_to_le16(high_16_bits(i_uid
));
4991 raw_inode
->i_gid_high
=
4992 cpu_to_le16(high_16_bits(i_gid
));
4995 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4996 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4997 raw_inode
->i_uid_high
= 0;
4998 raw_inode
->i_gid_high
= 0;
5000 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
5002 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
5003 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
5004 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
5005 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
5007 err
= ext4_inode_blocks_set(handle
, raw_inode
, ei
);
5009 spin_unlock(&ei
->i_raw_lock
);
5012 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
5013 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
5014 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
)))
5015 raw_inode
->i_file_acl_high
=
5016 cpu_to_le16(ei
->i_file_acl
>> 32);
5017 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
5018 if (READ_ONCE(ei
->i_disksize
) != ext4_isize(inode
->i_sb
, raw_inode
)) {
5019 ext4_isize_set(raw_inode
, ei
->i_disksize
);
5022 if (ei
->i_disksize
> 0x7fffffffULL
) {
5023 if (!ext4_has_feature_large_file(sb
) ||
5024 EXT4_SB(sb
)->s_es
->s_rev_level
==
5025 cpu_to_le32(EXT4_GOOD_OLD_REV
))
5028 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
5029 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
5030 if (old_valid_dev(inode
->i_rdev
)) {
5031 raw_inode
->i_block
[0] =
5032 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
5033 raw_inode
->i_block
[1] = 0;
5035 raw_inode
->i_block
[0] = 0;
5036 raw_inode
->i_block
[1] =
5037 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
5038 raw_inode
->i_block
[2] = 0;
5040 } else if (!ext4_has_inline_data(inode
)) {
5041 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
5042 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
5045 if (likely(!test_opt2(inode
->i_sb
, HURD_COMPAT
))) {
5046 u64 ivers
= ext4_inode_peek_iversion(inode
);
5048 raw_inode
->i_disk_version
= cpu_to_le32(ivers
);
5049 if (ei
->i_extra_isize
) {
5050 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
5051 raw_inode
->i_version_hi
=
5052 cpu_to_le32(ivers
>> 32);
5053 raw_inode
->i_extra_isize
=
5054 cpu_to_le16(ei
->i_extra_isize
);
5058 BUG_ON(!ext4_has_feature_project(inode
->i_sb
) &&
5059 i_projid
!= EXT4_DEF_PROJID
);
5061 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
5062 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_projid
))
5063 raw_inode
->i_projid
= cpu_to_le32(i_projid
);
5065 ext4_inode_csum_set(inode
, raw_inode
, ei
);
5066 spin_unlock(&ei
->i_raw_lock
);
5067 if (inode
->i_sb
->s_flags
& SB_LAZYTIME
)
5068 ext4_update_other_inodes_time(inode
->i_sb
, inode
->i_ino
,
5071 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
5072 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
5075 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
5076 if (set_large_file
) {
5077 BUFFER_TRACE(EXT4_SB(sb
)->s_sbh
, "get write access");
5078 err
= ext4_journal_get_write_access(handle
, EXT4_SB(sb
)->s_sbh
);
5081 ext4_set_feature_large_file(sb
);
5082 ext4_handle_sync(handle
);
5083 err
= ext4_handle_dirty_super(handle
, sb
);
5085 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
5088 ext4_std_error(inode
->i_sb
, err
);
5093 * ext4_write_inode()
5095 * We are called from a few places:
5097 * - Within generic_file_aio_write() -> generic_write_sync() for O_SYNC files.
5098 * Here, there will be no transaction running. We wait for any running
5099 * transaction to commit.
5101 * - Within flush work (sys_sync(), kupdate and such).
5102 * We wait on commit, if told to.
5104 * - Within iput_final() -> write_inode_now()
5105 * We wait on commit, if told to.
5107 * In all cases it is actually safe for us to return without doing anything,
5108 * because the inode has been copied into a raw inode buffer in
5109 * ext4_mark_inode_dirty(). This is a correctness thing for WB_SYNC_ALL
5112 * Note that we are absolutely dependent upon all inode dirtiers doing the
5113 * right thing: they *must* call mark_inode_dirty() after dirtying info in
5114 * which we are interested.
5116 * It would be a bug for them to not do this. The code:
5118 * mark_inode_dirty(inode)
5120 * inode->i_size = expr;
5122 * is in error because write_inode() could occur while `stuff()' is running,
5123 * and the new i_size will be lost. Plus the inode will no longer be on the
5124 * superblock's dirty inode list.
5126 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
5130 if (WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
) ||
5131 sb_rdonly(inode
->i_sb
))
5134 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5137 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
5138 if (ext4_journal_current_handle()) {
5139 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
5145 * No need to force transaction in WB_SYNC_NONE mode. Also
5146 * ext4_sync_fs() will force the commit after everything is
5149 if (wbc
->sync_mode
!= WB_SYNC_ALL
|| wbc
->for_sync
)
5152 err
= jbd2_complete_transaction(EXT4_SB(inode
->i_sb
)->s_journal
,
5153 EXT4_I(inode
)->i_sync_tid
);
5155 struct ext4_iloc iloc
;
5157 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
5161 * sync(2) will flush the whole buffer cache. No need to do
5162 * it here separately for each inode.
5164 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
)
5165 sync_dirty_buffer(iloc
.bh
);
5166 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
5167 ext4_error_inode_block(inode
, iloc
.bh
->b_blocknr
, EIO
,
5168 "IO error syncing inode");
5177 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
5178 * buffers that are attached to a page stradding i_size and are undergoing
5179 * commit. In that case we have to wait for commit to finish and try again.
5181 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
5185 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
5186 tid_t commit_tid
= 0;
5189 offset
= inode
->i_size
& (PAGE_SIZE
- 1);
5191 * If the page is fully truncated, we don't need to wait for any commit
5192 * (and we even should not as __ext4_journalled_invalidatepage() may
5193 * strip all buffers from the page but keep the page dirty which can then
5194 * confuse e.g. concurrent ext4_writepage() seeing dirty page without
5195 * buffers). Also we don't need to wait for any commit if all buffers in
5196 * the page remain valid. This is most beneficial for the common case of
5197 * blocksize == PAGESIZE.
5199 if (!offset
|| offset
> (PAGE_SIZE
- i_blocksize(inode
)))
5202 page
= find_lock_page(inode
->i_mapping
,
5203 inode
->i_size
>> PAGE_SHIFT
);
5206 ret
= __ext4_journalled_invalidatepage(page
, offset
,
5207 PAGE_SIZE
- offset
);
5213 read_lock(&journal
->j_state_lock
);
5214 if (journal
->j_committing_transaction
)
5215 commit_tid
= journal
->j_committing_transaction
->t_tid
;
5216 read_unlock(&journal
->j_state_lock
);
5218 jbd2_log_wait_commit(journal
, commit_tid
);
5225 * Called from notify_change.
5227 * We want to trap VFS attempts to truncate the file as soon as
5228 * possible. In particular, we want to make sure that when the VFS
5229 * shrinks i_size, we put the inode on the orphan list and modify
5230 * i_disksize immediately, so that during the subsequent flushing of
5231 * dirty pages and freeing of disk blocks, we can guarantee that any
5232 * commit will leave the blocks being flushed in an unused state on
5233 * disk. (On recovery, the inode will get truncated and the blocks will
5234 * be freed, so we have a strong guarantee that no future commit will
5235 * leave these blocks visible to the user.)
5237 * Another thing we have to assure is that if we are in ordered mode
5238 * and inode is still attached to the committing transaction, we must
5239 * we start writeout of all the dirty pages which are being truncated.
5240 * This way we are sure that all the data written in the previous
5241 * transaction are already on disk (truncate waits for pages under
5244 * Called with inode->i_mutex down.
5246 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
5248 struct inode
*inode
= d_inode(dentry
);
5251 const unsigned int ia_valid
= attr
->ia_valid
;
5253 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5256 if (unlikely(IS_IMMUTABLE(inode
)))
5259 if (unlikely(IS_APPEND(inode
) &&
5260 (ia_valid
& (ATTR_MODE
| ATTR_UID
|
5261 ATTR_GID
| ATTR_TIMES_SET
))))
5264 error
= setattr_prepare(dentry
, attr
);
5268 error
= fscrypt_prepare_setattr(dentry
, attr
);
5272 error
= fsverity_prepare_setattr(dentry
, attr
);
5276 if (is_quota_modification(inode
, attr
)) {
5277 error
= dquot_initialize(inode
);
5281 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
5282 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
5285 /* (user+group)*(old+new) structure, inode write (sb,
5286 * inode block, ? - but truncate inode update has it) */
5287 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
5288 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
5289 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
5290 if (IS_ERR(handle
)) {
5291 error
= PTR_ERR(handle
);
5295 /* dquot_transfer() calls back ext4_get_inode_usage() which
5296 * counts xattr inode references.
5298 down_read(&EXT4_I(inode
)->xattr_sem
);
5299 error
= dquot_transfer(inode
, attr
);
5300 up_read(&EXT4_I(inode
)->xattr_sem
);
5303 ext4_journal_stop(handle
);
5306 /* Update corresponding info in inode so that everything is in
5307 * one transaction */
5308 if (attr
->ia_valid
& ATTR_UID
)
5309 inode
->i_uid
= attr
->ia_uid
;
5310 if (attr
->ia_valid
& ATTR_GID
)
5311 inode
->i_gid
= attr
->ia_gid
;
5312 error
= ext4_mark_inode_dirty(handle
, inode
);
5313 ext4_journal_stop(handle
);
5314 if (unlikely(error
))
5318 if (attr
->ia_valid
& ATTR_SIZE
) {
5320 loff_t oldsize
= inode
->i_size
;
5321 int shrink
= (attr
->ia_size
< inode
->i_size
);
5323 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
5324 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5326 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
5329 if (!S_ISREG(inode
->i_mode
))
5332 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
5333 inode_inc_iversion(inode
);
5336 if (ext4_should_order_data(inode
)) {
5337 error
= ext4_begin_ordered_truncate(inode
,
5343 * Blocks are going to be removed from the inode. Wait
5344 * for dio in flight.
5346 inode_dio_wait(inode
);
5349 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5351 rc
= ext4_break_layouts(inode
);
5353 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5357 if (attr
->ia_size
!= inode
->i_size
) {
5358 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
5359 if (IS_ERR(handle
)) {
5360 error
= PTR_ERR(handle
);
5363 if (ext4_handle_valid(handle
) && shrink
) {
5364 error
= ext4_orphan_add(handle
, inode
);
5368 * Update c/mtime on truncate up, ext4_truncate() will
5369 * update c/mtime in shrink case below
5372 inode
->i_mtime
= current_time(inode
);
5373 inode
->i_ctime
= inode
->i_mtime
;
5375 down_write(&EXT4_I(inode
)->i_data_sem
);
5376 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
5377 rc
= ext4_mark_inode_dirty(handle
, inode
);
5381 * We have to update i_size under i_data_sem together
5382 * with i_disksize to avoid races with writeback code
5383 * running ext4_wb_update_i_disksize().
5386 i_size_write(inode
, attr
->ia_size
);
5387 up_write(&EXT4_I(inode
)->i_data_sem
);
5388 ext4_journal_stop(handle
);
5392 pagecache_isize_extended(inode
, oldsize
,
5394 } else if (ext4_should_journal_data(inode
)) {
5395 ext4_wait_for_tail_page_commit(inode
);
5400 * Truncate pagecache after we've waited for commit
5401 * in data=journal mode to make pages freeable.
5403 truncate_pagecache(inode
, inode
->i_size
);
5405 * Call ext4_truncate() even if i_size didn't change to
5406 * truncate possible preallocated blocks.
5408 if (attr
->ia_size
<= oldsize
) {
5409 rc
= ext4_truncate(inode
);
5414 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5418 setattr_copy(inode
, attr
);
5419 mark_inode_dirty(inode
);
5423 * If the call to ext4_truncate failed to get a transaction handle at
5424 * all, we need to clean up the in-core orphan list manually.
5426 if (orphan
&& inode
->i_nlink
)
5427 ext4_orphan_del(NULL
, inode
);
5429 if (!error
&& (ia_valid
& ATTR_MODE
))
5430 rc
= posix_acl_chmod(inode
, inode
->i_mode
);
5433 ext4_std_error(inode
->i_sb
, error
);
5439 int ext4_getattr(const struct path
*path
, struct kstat
*stat
,
5440 u32 request_mask
, unsigned int query_flags
)
5442 struct inode
*inode
= d_inode(path
->dentry
);
5443 struct ext4_inode
*raw_inode
;
5444 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5447 if ((request_mask
& STATX_BTIME
) &&
5448 EXT4_FITS_IN_INODE(raw_inode
, ei
, i_crtime
)) {
5449 stat
->result_mask
|= STATX_BTIME
;
5450 stat
->btime
.tv_sec
= ei
->i_crtime
.tv_sec
;
5451 stat
->btime
.tv_nsec
= ei
->i_crtime
.tv_nsec
;
5454 flags
= ei
->i_flags
& EXT4_FL_USER_VISIBLE
;
5455 if (flags
& EXT4_APPEND_FL
)
5456 stat
->attributes
|= STATX_ATTR_APPEND
;
5457 if (flags
& EXT4_COMPR_FL
)
5458 stat
->attributes
|= STATX_ATTR_COMPRESSED
;
5459 if (flags
& EXT4_ENCRYPT_FL
)
5460 stat
->attributes
|= STATX_ATTR_ENCRYPTED
;
5461 if (flags
& EXT4_IMMUTABLE_FL
)
5462 stat
->attributes
|= STATX_ATTR_IMMUTABLE
;
5463 if (flags
& EXT4_NODUMP_FL
)
5464 stat
->attributes
|= STATX_ATTR_NODUMP
;
5465 if (flags
& EXT4_VERITY_FL
)
5466 stat
->attributes
|= STATX_ATTR_VERITY
;
5468 stat
->attributes_mask
|= (STATX_ATTR_APPEND
|
5469 STATX_ATTR_COMPRESSED
|
5470 STATX_ATTR_ENCRYPTED
|
5471 STATX_ATTR_IMMUTABLE
|
5475 generic_fillattr(inode
, stat
);
5479 int ext4_file_getattr(const struct path
*path
, struct kstat
*stat
,
5480 u32 request_mask
, unsigned int query_flags
)
5482 struct inode
*inode
= d_inode(path
->dentry
);
5483 u64 delalloc_blocks
;
5485 ext4_getattr(path
, stat
, request_mask
, query_flags
);
5488 * If there is inline data in the inode, the inode will normally not
5489 * have data blocks allocated (it may have an external xattr block).
5490 * Report at least one sector for such files, so tools like tar, rsync,
5491 * others don't incorrectly think the file is completely sparse.
5493 if (unlikely(ext4_has_inline_data(inode
)))
5494 stat
->blocks
+= (stat
->size
+ 511) >> 9;
5497 * We can't update i_blocks if the block allocation is delayed
5498 * otherwise in the case of system crash before the real block
5499 * allocation is done, we will have i_blocks inconsistent with
5500 * on-disk file blocks.
5501 * We always keep i_blocks updated together with real
5502 * allocation. But to not confuse with user, stat
5503 * will return the blocks that include the delayed allocation
5504 * blocks for this file.
5506 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
5507 EXT4_I(inode
)->i_reserved_data_blocks
);
5508 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
- 9);
5512 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
5515 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
5516 return ext4_ind_trans_blocks(inode
, lblocks
);
5517 return ext4_ext_index_trans_blocks(inode
, pextents
);
5521 * Account for index blocks, block groups bitmaps and block group
5522 * descriptor blocks if modify datablocks and index blocks
5523 * worse case, the indexs blocks spread over different block groups
5525 * If datablocks are discontiguous, they are possible to spread over
5526 * different block groups too. If they are contiguous, with flexbg,
5527 * they could still across block group boundary.
5529 * Also account for superblock, inode, quota and xattr blocks
5531 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
5534 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
5540 * How many index blocks need to touch to map @lblocks logical blocks
5541 * to @pextents physical extents?
5543 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
5548 * Now let's see how many group bitmaps and group descriptors need
5551 groups
= idxblocks
+ pextents
;
5553 if (groups
> ngroups
)
5555 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
5556 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
5558 /* bitmaps and block group descriptor blocks */
5559 ret
+= groups
+ gdpblocks
;
5561 /* Blocks for super block, inode, quota and xattr blocks */
5562 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
5568 * Calculate the total number of credits to reserve to fit
5569 * the modification of a single pages into a single transaction,
5570 * which may include multiple chunks of block allocations.
5572 * This could be called via ext4_write_begin()
5574 * We need to consider the worse case, when
5575 * one new block per extent.
5577 int ext4_writepage_trans_blocks(struct inode
*inode
)
5579 int bpp
= ext4_journal_blocks_per_page(inode
);
5582 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
5584 /* Account for data blocks for journalled mode */
5585 if (ext4_should_journal_data(inode
))
5591 * Calculate the journal credits for a chunk of data modification.
5593 * This is called from DIO, fallocate or whoever calling
5594 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
5596 * journal buffers for data blocks are not included here, as DIO
5597 * and fallocate do no need to journal data buffers.
5599 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
5601 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5605 * The caller must have previously called ext4_reserve_inode_write().
5606 * Give this, we know that the caller already has write access to iloc->bh.
5608 int ext4_mark_iloc_dirty(handle_t
*handle
,
5609 struct inode
*inode
, struct ext4_iloc
*iloc
)
5613 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
)))) {
5617 if (IS_I_VERSION(inode
))
5618 inode_inc_iversion(inode
);
5620 /* the do_update_inode consumes one bh->b_count */
5623 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5624 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5630 * On success, We end up with an outstanding reference count against
5631 * iloc->bh. This _must_ be cleaned up later.
5635 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5636 struct ext4_iloc
*iloc
)
5640 if (unlikely(ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))))
5643 err
= ext4_get_inode_loc(inode
, iloc
);
5645 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5646 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5652 ext4_std_error(inode
->i_sb
, err
);
5656 static int __ext4_expand_extra_isize(struct inode
*inode
,
5657 unsigned int new_extra_isize
,
5658 struct ext4_iloc
*iloc
,
5659 handle_t
*handle
, int *no_expand
)
5661 struct ext4_inode
*raw_inode
;
5662 struct ext4_xattr_ibody_header
*header
;
5663 unsigned int inode_size
= EXT4_INODE_SIZE(inode
->i_sb
);
5664 struct ext4_inode_info
*ei
= EXT4_I(inode
);
5667 /* this was checked at iget time, but double check for good measure */
5668 if ((EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
> inode_size
) ||
5669 (ei
->i_extra_isize
& 3)) {
5670 EXT4_ERROR_INODE(inode
, "bad extra_isize %u (inode size %u)",
5672 EXT4_INODE_SIZE(inode
->i_sb
));
5673 return -EFSCORRUPTED
;
5675 if ((new_extra_isize
< ei
->i_extra_isize
) ||
5676 (new_extra_isize
< 4) ||
5677 (new_extra_isize
> inode_size
- EXT4_GOOD_OLD_INODE_SIZE
))
5678 return -EINVAL
; /* Should never happen */
5680 raw_inode
= ext4_raw_inode(iloc
);
5682 header
= IHDR(inode
, raw_inode
);
5684 /* No extended attributes present */
5685 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5686 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5687 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
+
5688 EXT4_I(inode
)->i_extra_isize
, 0,
5689 new_extra_isize
- EXT4_I(inode
)->i_extra_isize
);
5690 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5694 /* try to expand with EAs present */
5695 error
= ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5699 * Inode size expansion failed; don't try again
5708 * Expand an inode by new_extra_isize bytes.
5709 * Returns 0 on success or negative error number on failure.
5711 static int ext4_try_to_expand_extra_isize(struct inode
*inode
,
5712 unsigned int new_extra_isize
,
5713 struct ext4_iloc iloc
,
5719 if (ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
))
5723 * In nojournal mode, we can immediately attempt to expand
5724 * the inode. When journaled, we first need to obtain extra
5725 * buffer credits since we may write into the EA block
5726 * with this same handle. If journal_extend fails, then it will
5727 * only result in a minor loss of functionality for that inode.
5728 * If this is felt to be critical, then e2fsck should be run to
5729 * force a large enough s_min_extra_isize.
5731 if (ext4_journal_extend(handle
,
5732 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
), 0) != 0)
5735 if (ext4_write_trylock_xattr(inode
, &no_expand
) == 0)
5738 error
= __ext4_expand_extra_isize(inode
, new_extra_isize
, &iloc
,
5739 handle
, &no_expand
);
5740 ext4_write_unlock_xattr(inode
, &no_expand
);
5745 int ext4_expand_extra_isize(struct inode
*inode
,
5746 unsigned int new_extra_isize
,
5747 struct ext4_iloc
*iloc
)
5753 if (ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5758 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
,
5759 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
));
5760 if (IS_ERR(handle
)) {
5761 error
= PTR_ERR(handle
);
5766 ext4_write_lock_xattr(inode
, &no_expand
);
5768 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5769 error
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5775 error
= __ext4_expand_extra_isize(inode
, new_extra_isize
, iloc
,
5776 handle
, &no_expand
);
5778 rc
= ext4_mark_iloc_dirty(handle
, inode
, iloc
);
5783 ext4_write_unlock_xattr(inode
, &no_expand
);
5784 ext4_journal_stop(handle
);
5789 * What we do here is to mark the in-core inode as clean with respect to inode
5790 * dirtiness (it may still be data-dirty).
5791 * This means that the in-core inode may be reaped by prune_icache
5792 * without having to perform any I/O. This is a very good thing,
5793 * because *any* task may call prune_icache - even ones which
5794 * have a transaction open against a different journal.
5796 * Is this cheating? Not really. Sure, we haven't written the
5797 * inode out, but prune_icache isn't a user-visible syncing function.
5798 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5799 * we start and wait on commits.
5801 int __ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
,
5802 const char *func
, unsigned int line
)
5804 struct ext4_iloc iloc
;
5805 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5809 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5810 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5814 if (EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
)
5815 ext4_try_to_expand_extra_isize(inode
, sbi
->s_want_extra_isize
,
5818 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5821 ext4_error_inode_err(inode
, func
, line
, 0, err
,
5822 "mark_inode_dirty error");
5827 * ext4_dirty_inode() is called from __mark_inode_dirty()
5829 * We're really interested in the case where a file is being extended.
5830 * i_size has been changed by generic_commit_write() and we thus need
5831 * to include the updated inode in the current transaction.
5833 * Also, dquot_alloc_block() will always dirty the inode when blocks
5834 * are allocated to the file.
5836 * If the inode is marked synchronous, we don't honour that here - doing
5837 * so would cause a commit on atime updates, which we don't bother doing.
5838 * We handle synchronous inodes at the highest possible level.
5840 * If only the I_DIRTY_TIME flag is set, we can skip everything. If
5841 * I_DIRTY_TIME and I_DIRTY_SYNC is set, the only inode fields we need
5842 * to copy into the on-disk inode structure are the timestamp files.
5844 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5848 if (flags
== I_DIRTY_TIME
)
5850 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5854 ext4_mark_inode_dirty(handle
, inode
);
5856 ext4_journal_stop(handle
);
5861 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5866 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5869 * We have to be very careful here: changing a data block's
5870 * journaling status dynamically is dangerous. If we write a
5871 * data block to the journal, change the status and then delete
5872 * that block, we risk forgetting to revoke the old log record
5873 * from the journal and so a subsequent replay can corrupt data.
5874 * So, first we make sure that the journal is empty and that
5875 * nobody is changing anything.
5878 journal
= EXT4_JOURNAL(inode
);
5881 if (is_journal_aborted(journal
))
5884 /* Wait for all existing dio workers */
5885 inode_dio_wait(inode
);
5888 * Before flushing the journal and switching inode's aops, we have
5889 * to flush all dirty data the inode has. There can be outstanding
5890 * delayed allocations, there can be unwritten extents created by
5891 * fallocate or buffered writes in dioread_nolock mode covered by
5892 * dirty data which can be converted only after flushing the dirty
5893 * data (and journalled aops don't know how to handle these cases).
5896 down_write(&EXT4_I(inode
)->i_mmap_sem
);
5897 err
= filemap_write_and_wait(inode
->i_mapping
);
5899 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5904 percpu_down_write(&sbi
->s_writepages_rwsem
);
5905 jbd2_journal_lock_updates(journal
);
5908 * OK, there are no updates running now, and all cached data is
5909 * synced to disk. We are now in a completely consistent state
5910 * which doesn't have anything in the journal, and we know that
5911 * no filesystem updates are running, so it is safe to modify
5912 * the inode's in-core data-journaling state flag now.
5916 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5918 err
= jbd2_journal_flush(journal
);
5920 jbd2_journal_unlock_updates(journal
);
5921 percpu_up_write(&sbi
->s_writepages_rwsem
);
5924 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5926 ext4_set_aops(inode
);
5928 jbd2_journal_unlock_updates(journal
);
5929 percpu_up_write(&sbi
->s_writepages_rwsem
);
5932 up_write(&EXT4_I(inode
)->i_mmap_sem
);
5934 /* Finally we can mark the inode as dirty. */
5936 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5938 return PTR_ERR(handle
);
5940 err
= ext4_mark_inode_dirty(handle
, inode
);
5941 ext4_handle_sync(handle
);
5942 ext4_journal_stop(handle
);
5943 ext4_std_error(inode
->i_sb
, err
);
5948 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5950 return !buffer_mapped(bh
);
5953 vm_fault_t
ext4_page_mkwrite(struct vm_fault
*vmf
)
5955 struct vm_area_struct
*vma
= vmf
->vma
;
5956 struct page
*page
= vmf
->page
;
5961 struct file
*file
= vma
->vm_file
;
5962 struct inode
*inode
= file_inode(file
);
5963 struct address_space
*mapping
= inode
->i_mapping
;
5965 get_block_t
*get_block
;
5968 if (unlikely(IS_IMMUTABLE(inode
)))
5969 return VM_FAULT_SIGBUS
;
5971 sb_start_pagefault(inode
->i_sb
);
5972 file_update_time(vma
->vm_file
);
5974 down_read(&EXT4_I(inode
)->i_mmap_sem
);
5976 err
= ext4_convert_inline_data(inode
);
5980 /* Delalloc case is easy... */
5981 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5982 !ext4_should_journal_data(inode
) &&
5983 !ext4_nonda_switch(inode
->i_sb
)) {
5985 err
= block_page_mkwrite(vma
, vmf
,
5986 ext4_da_get_block_prep
);
5987 } while (err
== -ENOSPC
&&
5988 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5993 size
= i_size_read(inode
);
5994 /* Page got truncated from under us? */
5995 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5997 ret
= VM_FAULT_NOPAGE
;
6001 if (page
->index
== size
>> PAGE_SHIFT
)
6002 len
= size
& ~PAGE_MASK
;
6006 * Return if we have all the buffers mapped. This avoids the need to do
6007 * journal_start/journal_stop which can block and take a long time
6009 if (page_has_buffers(page
)) {
6010 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
6012 ext4_bh_unmapped
)) {
6013 /* Wait so that we don't change page under IO */
6014 wait_for_stable_page(page
);
6015 ret
= VM_FAULT_LOCKED
;
6020 /* OK, we need to fill the hole... */
6021 if (ext4_should_dioread_nolock(inode
))
6022 get_block
= ext4_get_block_unwritten
;
6024 get_block
= ext4_get_block
;
6026 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
6027 ext4_writepage_trans_blocks(inode
));
6028 if (IS_ERR(handle
)) {
6029 ret
= VM_FAULT_SIGBUS
;
6032 err
= block_page_mkwrite(vma
, vmf
, get_block
);
6033 if (!err
&& ext4_should_journal_data(inode
)) {
6034 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
6035 PAGE_SIZE
, NULL
, do_journal_get_write_access
)) {
6037 ret
= VM_FAULT_SIGBUS
;
6038 ext4_journal_stop(handle
);
6041 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
6043 ext4_journal_stop(handle
);
6044 if (err
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
6047 ret
= block_page_mkwrite_return(err
);
6049 up_read(&EXT4_I(inode
)->i_mmap_sem
);
6050 sb_end_pagefault(inode
->i_sb
);
6054 vm_fault_t
ext4_filemap_fault(struct vm_fault
*vmf
)
6056 struct inode
*inode
= file_inode(vmf
->vma
->vm_file
);
6059 down_read(&EXT4_I(inode
)->i_mmap_sem
);
6060 ret
= filemap_fault(vmf
);
6061 up_read(&EXT4_I(inode
)->i_mmap_sem
);