2 * linux/fs/ext4/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * 64-bit file support on 64-bit platforms by Jakub Jelinek
16 * (jj@sunsite.ms.mff.cuni.cz)
18 * Assorted race fixes, rewrite of ext4_get_block() by Al Viro, 2000
22 #include <linux/time.h>
23 #include <linux/jbd2.h>
24 #include <linux/highuid.h>
25 #include <linux/pagemap.h>
26 #include <linux/quotaops.h>
27 #include <linux/string.h>
28 #include <linux/buffer_head.h>
29 #include <linux/writeback.h>
30 #include <linux/pagevec.h>
31 #include <linux/mpage.h>
32 #include <linux/namei.h>
33 #include <linux/uio.h>
34 #include <linux/bio.h>
35 #include <linux/workqueue.h>
36 #include <linux/kernel.h>
37 #include <linux/printk.h>
38 #include <linux/slab.h>
39 #include <linux/ratelimit.h>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
51 struct ext4_inode_info
*ei
)
53 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
58 csum_lo
= raw
->i_checksum_lo
;
59 raw
->i_checksum_lo
= 0;
60 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
61 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
62 csum_hi
= raw
->i_checksum_hi
;
63 raw
->i_checksum_hi
= 0;
66 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
,
67 EXT4_INODE_SIZE(inode
->i_sb
));
69 raw
->i_checksum_lo
= csum_lo
;
70 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
71 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
72 raw
->i_checksum_hi
= csum_hi
;
77 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
78 struct ext4_inode_info
*ei
)
80 __u32 provided
, calculated
;
82 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
83 cpu_to_le32(EXT4_OS_LINUX
) ||
84 !EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
85 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
88 provided
= le16_to_cpu(raw
->i_checksum_lo
);
89 calculated
= ext4_inode_csum(inode
, raw
, ei
);
90 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
91 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
92 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
96 return provided
== calculated
;
99 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
100 struct ext4_inode_info
*ei
)
104 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
105 cpu_to_le32(EXT4_OS_LINUX
) ||
106 !EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
107 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
110 csum
= ext4_inode_csum(inode
, raw
, ei
);
111 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
112 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
113 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
114 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
117 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
120 trace_ext4_begin_ordered_truncate(inode
, new_size
);
122 * If jinode is zero, then we never opened the file for
123 * writing, so there's no need to call
124 * jbd2_journal_begin_ordered_truncate() since there's no
125 * outstanding writes we need to flush.
127 if (!EXT4_I(inode
)->jinode
)
129 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
130 EXT4_I(inode
)->jinode
,
134 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
135 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
136 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
137 static int ext4_discard_partial_page_buffers_no_lock(handle_t
*handle
,
138 struct inode
*inode
, struct page
*page
, loff_t from
,
139 loff_t length
, int flags
);
142 * Test whether an inode is a fast symlink.
144 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
146 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
147 (inode
->i_sb
->s_blocksize
>> 9) : 0;
149 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
153 * Restart the transaction associated with *handle. This does a commit,
154 * so before we call here everything must be consistently dirtied against
157 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
163 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
164 * moment, get_block can be called only for blocks inside i_size since
165 * page cache has been already dropped and writes are blocked by
166 * i_mutex. So we can safely drop the i_data_sem here.
168 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
169 jbd_debug(2, "restarting handle %p\n", handle
);
170 up_write(&EXT4_I(inode
)->i_data_sem
);
171 ret
= ext4_journal_restart(handle
, nblocks
);
172 down_write(&EXT4_I(inode
)->i_data_sem
);
173 ext4_discard_preallocations(inode
);
179 * Called at the last iput() if i_nlink is zero.
181 void ext4_evict_inode(struct inode
*inode
)
186 trace_ext4_evict_inode(inode
);
188 ext4_ioend_wait(inode
);
190 if (inode
->i_nlink
) {
192 * When journalling data dirty buffers are tracked only in the
193 * journal. So although mm thinks everything is clean and
194 * ready for reaping the inode might still have some pages to
195 * write in the running transaction or waiting to be
196 * checkpointed. Thus calling jbd2_journal_invalidatepage()
197 * (via truncate_inode_pages()) to discard these buffers can
198 * cause data loss. Also even if we did not discard these
199 * buffers, we would have no way to find them after the inode
200 * is reaped and thus user could see stale data if he tries to
201 * read them before the transaction is checkpointed. So be
202 * careful and force everything to disk here... We use
203 * ei->i_datasync_tid to store the newest transaction
204 * containing inode's data.
206 * Note that directories do not have this problem because they
207 * don't use page cache.
209 if (ext4_should_journal_data(inode
) &&
210 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
))) {
211 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
212 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
214 jbd2_log_start_commit(journal
, commit_tid
);
215 jbd2_log_wait_commit(journal
, commit_tid
);
216 filemap_write_and_wait(&inode
->i_data
);
218 truncate_inode_pages(&inode
->i_data
, 0);
222 if (!is_bad_inode(inode
))
223 dquot_initialize(inode
);
225 if (ext4_should_order_data(inode
))
226 ext4_begin_ordered_truncate(inode
, 0);
227 truncate_inode_pages(&inode
->i_data
, 0);
229 if (is_bad_inode(inode
))
233 * Protect us against freezing - iput() caller didn't have to have any
234 * protection against it
236 sb_start_intwrite(inode
->i_sb
);
237 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
238 ext4_blocks_for_truncate(inode
)+3);
239 if (IS_ERR(handle
)) {
240 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
242 * If we're going to skip the normal cleanup, we still need to
243 * make sure that the in-core orphan linked list is properly
246 ext4_orphan_del(NULL
, inode
);
247 sb_end_intwrite(inode
->i_sb
);
252 ext4_handle_sync(handle
);
254 err
= ext4_mark_inode_dirty(handle
, inode
);
256 ext4_warning(inode
->i_sb
,
257 "couldn't mark inode dirty (err %d)", err
);
261 ext4_truncate(inode
);
264 * ext4_ext_truncate() doesn't reserve any slop when it
265 * restarts journal transactions; therefore there may not be
266 * enough credits left in the handle to remove the inode from
267 * the orphan list and set the dtime field.
269 if (!ext4_handle_has_enough_credits(handle
, 3)) {
270 err
= ext4_journal_extend(handle
, 3);
272 err
= ext4_journal_restart(handle
, 3);
274 ext4_warning(inode
->i_sb
,
275 "couldn't extend journal (err %d)", err
);
277 ext4_journal_stop(handle
);
278 ext4_orphan_del(NULL
, inode
);
279 sb_end_intwrite(inode
->i_sb
);
285 * Kill off the orphan record which ext4_truncate created.
286 * AKPM: I think this can be inside the above `if'.
287 * Note that ext4_orphan_del() has to be able to cope with the
288 * deletion of a non-existent orphan - this is because we don't
289 * know if ext4_truncate() actually created an orphan record.
290 * (Well, we could do this if we need to, but heck - it works)
292 ext4_orphan_del(handle
, inode
);
293 EXT4_I(inode
)->i_dtime
= get_seconds();
296 * One subtle ordering requirement: if anything has gone wrong
297 * (transaction abort, IO errors, whatever), then we can still
298 * do these next steps (the fs will already have been marked as
299 * having errors), but we can't free the inode if the mark_dirty
302 if (ext4_mark_inode_dirty(handle
, inode
))
303 /* If that failed, just do the required in-core inode clear. */
304 ext4_clear_inode(inode
);
306 ext4_free_inode(handle
, inode
);
307 ext4_journal_stop(handle
);
308 sb_end_intwrite(inode
->i_sb
);
311 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
315 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
317 return &EXT4_I(inode
)->i_reserved_quota
;
322 * Calculate the number of metadata blocks need to reserve
323 * to allocate a block located at @lblock
325 static int ext4_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
327 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
328 return ext4_ext_calc_metadata_amount(inode
, lblock
);
330 return ext4_ind_calc_metadata_amount(inode
, lblock
);
334 * Called with i_data_sem down, which is important since we can call
335 * ext4_discard_preallocations() from here.
337 void ext4_da_update_reserve_space(struct inode
*inode
,
338 int used
, int quota_claim
)
340 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
341 struct ext4_inode_info
*ei
= EXT4_I(inode
);
343 spin_lock(&ei
->i_block_reservation_lock
);
344 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
345 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
346 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
347 "with only %d reserved data blocks",
348 __func__
, inode
->i_ino
, used
,
349 ei
->i_reserved_data_blocks
);
351 used
= ei
->i_reserved_data_blocks
;
354 if (unlikely(ei
->i_allocated_meta_blocks
> ei
->i_reserved_meta_blocks
)) {
355 ext4_warning(inode
->i_sb
, "ino %lu, allocated %d "
356 "with only %d reserved metadata blocks "
357 "(releasing %d blocks with reserved %d data blocks)",
358 inode
->i_ino
, ei
->i_allocated_meta_blocks
,
359 ei
->i_reserved_meta_blocks
, used
,
360 ei
->i_reserved_data_blocks
);
362 ei
->i_allocated_meta_blocks
= ei
->i_reserved_meta_blocks
;
365 /* Update per-inode reservations */
366 ei
->i_reserved_data_blocks
-= used
;
367 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
368 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
369 used
+ ei
->i_allocated_meta_blocks
);
370 ei
->i_allocated_meta_blocks
= 0;
372 if (ei
->i_reserved_data_blocks
== 0) {
374 * We can release all of the reserved metadata blocks
375 * only when we have written all of the delayed
378 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
379 ei
->i_reserved_meta_blocks
);
380 ei
->i_reserved_meta_blocks
= 0;
381 ei
->i_da_metadata_calc_len
= 0;
383 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
385 /* Update quota subsystem for data blocks */
387 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
390 * We did fallocate with an offset that is already delayed
391 * allocated. So on delayed allocated writeback we should
392 * not re-claim the quota for fallocated blocks.
394 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
398 * If we have done all the pending block allocations and if
399 * there aren't any writers on the inode, we can discard the
400 * inode's preallocations.
402 if ((ei
->i_reserved_data_blocks
== 0) &&
403 (atomic_read(&inode
->i_writecount
) == 0))
404 ext4_discard_preallocations(inode
);
407 static int __check_block_validity(struct inode
*inode
, const char *func
,
409 struct ext4_map_blocks
*map
)
411 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
413 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
414 "lblock %lu mapped to illegal pblock "
415 "(length %d)", (unsigned long) map
->m_lblk
,
422 #define check_block_validity(inode, map) \
423 __check_block_validity((inode), __func__, __LINE__, (map))
426 * Return the number of contiguous dirty pages in a given inode
427 * starting at page frame idx.
429 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
430 unsigned int max_pages
)
432 struct address_space
*mapping
= inode
->i_mapping
;
436 int i
, nr_pages
, done
= 0;
440 pagevec_init(&pvec
, 0);
443 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
445 (pgoff_t
)PAGEVEC_SIZE
);
448 for (i
= 0; i
< nr_pages
; i
++) {
449 struct page
*page
= pvec
.pages
[i
];
450 struct buffer_head
*bh
, *head
;
453 if (unlikely(page
->mapping
!= mapping
) ||
455 PageWriteback(page
) ||
456 page
->index
!= idx
) {
461 if (page_has_buffers(page
)) {
462 bh
= head
= page_buffers(page
);
464 if (!buffer_delay(bh
) &&
465 !buffer_unwritten(bh
))
467 bh
= bh
->b_this_page
;
468 } while (!done
&& (bh
!= head
));
475 if (num
>= max_pages
) {
480 pagevec_release(&pvec
);
485 #ifdef ES_AGGRESSIVE_TEST
486 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
488 struct ext4_map_blocks
*es_map
,
489 struct ext4_map_blocks
*map
,
496 * There is a race window that the result is not the same.
497 * e.g. xfstests #223 when dioread_nolock enables. The reason
498 * is that we lookup a block mapping in extent status tree with
499 * out taking i_data_sem. So at the time the unwritten extent
500 * could be converted.
502 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
503 down_read((&EXT4_I(inode
)->i_data_sem
));
504 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
505 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
506 EXT4_GET_BLOCKS_KEEP_SIZE
);
508 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
509 EXT4_GET_BLOCKS_KEEP_SIZE
);
511 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
512 up_read((&EXT4_I(inode
)->i_data_sem
));
514 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
515 * because it shouldn't be marked in es_map->m_flags.
517 map
->m_flags
&= ~(EXT4_MAP_FROM_CLUSTER
| EXT4_MAP_BOUNDARY
);
520 * We don't check m_len because extent will be collpased in status
521 * tree. So the m_len might not equal.
523 if (es_map
->m_lblk
!= map
->m_lblk
||
524 es_map
->m_flags
!= map
->m_flags
||
525 es_map
->m_pblk
!= map
->m_pblk
) {
526 printk("ES cache assertation failed for inode: %lu "
527 "es_cached ex [%d/%d/%llu/%x] != "
528 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
529 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
530 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
531 map
->m_len
, map
->m_pblk
, map
->m_flags
,
535 #endif /* ES_AGGRESSIVE_TEST */
538 * The ext4_map_blocks() function tries to look up the requested blocks,
539 * and returns if the blocks are already mapped.
541 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
542 * and store the allocated blocks in the result buffer head and mark it
545 * If file type is extents based, it will call ext4_ext_map_blocks(),
546 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
549 * On success, it returns the number of blocks being mapped or allocate.
550 * if create==0 and the blocks are pre-allocated and uninitialized block,
551 * the result buffer head is unmapped. If the create ==1, it will make sure
552 * the buffer head is mapped.
554 * It returns 0 if plain look up failed (blocks have not been allocated), in
555 * that case, buffer head is unmapped
557 * It returns the error in case of allocation failure.
559 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
560 struct ext4_map_blocks
*map
, int flags
)
562 struct extent_status es
;
564 #ifdef ES_AGGRESSIVE_TEST
565 struct ext4_map_blocks orig_map
;
567 memcpy(&orig_map
, map
, sizeof(*map
));
571 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
572 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
573 (unsigned long) map
->m_lblk
);
575 /* Lookup extent status tree firstly */
576 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
577 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
578 map
->m_pblk
= ext4_es_pblock(&es
) +
579 map
->m_lblk
- es
.es_lblk
;
580 map
->m_flags
|= ext4_es_is_written(&es
) ?
581 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
582 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
583 if (retval
> map
->m_len
)
586 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
591 #ifdef ES_AGGRESSIVE_TEST
592 ext4_map_blocks_es_recheck(handle
, inode
, map
,
599 * Try to see if we can get the block without requesting a new
602 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
603 down_read((&EXT4_I(inode
)->i_data_sem
));
604 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
605 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
606 EXT4_GET_BLOCKS_KEEP_SIZE
);
608 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
609 EXT4_GET_BLOCKS_KEEP_SIZE
);
613 unsigned long long status
;
615 #ifdef ES_AGGRESSIVE_TEST
616 if (retval
!= map
->m_len
) {
617 printk("ES len assertation failed for inode: %lu "
618 "retval %d != map->m_len %d "
619 "in %s (lookup)\n", inode
->i_ino
, retval
,
620 map
->m_len
, __func__
);
624 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
625 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
626 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
627 ext4_find_delalloc_range(inode
, map
->m_lblk
,
628 map
->m_lblk
+ map
->m_len
- 1))
629 status
|= EXTENT_STATUS_DELAYED
;
630 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
631 map
->m_len
, map
->m_pblk
, status
);
635 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
636 up_read((&EXT4_I(inode
)->i_data_sem
));
639 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
640 int ret
= check_block_validity(inode
, map
);
645 /* If it is only a block(s) look up */
646 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
650 * Returns if the blocks have already allocated
652 * Note that if blocks have been preallocated
653 * ext4_ext_get_block() returns the create = 0
654 * with buffer head unmapped.
656 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
660 * Here we clear m_flags because after allocating an new extent,
661 * it will be set again.
663 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
666 * New blocks allocate and/or writing to uninitialized extent
667 * will possibly result in updating i_data, so we take
668 * the write lock of i_data_sem, and call get_blocks()
669 * with create == 1 flag.
671 down_write((&EXT4_I(inode
)->i_data_sem
));
674 * if the caller is from delayed allocation writeout path
675 * we have already reserved fs blocks for allocation
676 * let the underlying get_block() function know to
677 * avoid double accounting
679 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
680 ext4_set_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
682 * We need to check for EXT4 here because migrate
683 * could have changed the inode type in between
685 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
686 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
688 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
690 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
692 * We allocated new blocks which will result in
693 * i_data's format changing. Force the migrate
694 * to fail by clearing migrate flags
696 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
700 * Update reserved blocks/metadata blocks after successful
701 * block allocation which had been deferred till now. We don't
702 * support fallocate for non extent files. So we can update
703 * reserve space here.
706 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
707 ext4_da_update_reserve_space(inode
, retval
, 1);
709 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
710 ext4_clear_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
714 unsigned long long status
;
716 #ifdef ES_AGGRESSIVE_TEST
717 if (retval
!= map
->m_len
) {
718 printk("ES len assertation failed for inode: %lu "
719 "retval %d != map->m_len %d "
720 "in %s (allocation)\n", inode
->i_ino
, retval
,
721 map
->m_len
, __func__
);
726 * If the extent has been zeroed out, we don't need to update
727 * extent status tree.
729 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
730 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
731 if (ext4_es_is_written(&es
))
734 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
735 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
736 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
737 ext4_find_delalloc_range(inode
, map
->m_lblk
,
738 map
->m_lblk
+ map
->m_len
- 1))
739 status
|= EXTENT_STATUS_DELAYED
;
740 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
741 map
->m_pblk
, status
);
747 up_write((&EXT4_I(inode
)->i_data_sem
));
748 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
749 int ret
= check_block_validity(inode
, map
);
756 /* Maximum number of blocks we map for direct IO at once. */
757 #define DIO_MAX_BLOCKS 4096
759 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
760 struct buffer_head
*bh
, int flags
)
762 handle_t
*handle
= ext4_journal_current_handle();
763 struct ext4_map_blocks map
;
764 int ret
= 0, started
= 0;
767 if (ext4_has_inline_data(inode
))
771 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
773 if (flags
&& !(flags
& EXT4_GET_BLOCKS_NO_LOCK
) && !handle
) {
774 /* Direct IO write... */
775 if (map
.m_len
> DIO_MAX_BLOCKS
)
776 map
.m_len
= DIO_MAX_BLOCKS
;
777 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
778 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
780 if (IS_ERR(handle
)) {
781 ret
= PTR_ERR(handle
);
787 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
789 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
790 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
791 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
795 ext4_journal_stop(handle
);
799 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
800 struct buffer_head
*bh
, int create
)
802 return _ext4_get_block(inode
, iblock
, bh
,
803 create
? EXT4_GET_BLOCKS_CREATE
: 0);
807 * `handle' can be NULL if create is zero
809 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
810 ext4_lblk_t block
, int create
, int *errp
)
812 struct ext4_map_blocks map
;
813 struct buffer_head
*bh
;
816 J_ASSERT(handle
!= NULL
|| create
== 0);
820 err
= ext4_map_blocks(handle
, inode
, &map
,
821 create
? EXT4_GET_BLOCKS_CREATE
: 0);
823 /* ensure we send some value back into *errp */
826 if (create
&& err
== 0)
827 err
= -ENOSPC
; /* should never happen */
833 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
838 if (map
.m_flags
& EXT4_MAP_NEW
) {
839 J_ASSERT(create
!= 0);
840 J_ASSERT(handle
!= NULL
);
843 * Now that we do not always journal data, we should
844 * keep in mind whether this should always journal the
845 * new buffer as metadata. For now, regular file
846 * writes use ext4_get_block instead, so it's not a
850 BUFFER_TRACE(bh
, "call get_create_access");
851 fatal
= ext4_journal_get_create_access(handle
, bh
);
852 if (!fatal
&& !buffer_uptodate(bh
)) {
853 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
854 set_buffer_uptodate(bh
);
857 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
858 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
862 BUFFER_TRACE(bh
, "not a new buffer");
872 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
873 ext4_lblk_t block
, int create
, int *err
)
875 struct buffer_head
*bh
;
877 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
880 if (buffer_uptodate(bh
))
882 ll_rw_block(READ
| REQ_META
| REQ_PRIO
, 1, &bh
);
884 if (buffer_uptodate(bh
))
891 int ext4_walk_page_buffers(handle_t
*handle
,
892 struct buffer_head
*head
,
896 int (*fn
)(handle_t
*handle
,
897 struct buffer_head
*bh
))
899 struct buffer_head
*bh
;
900 unsigned block_start
, block_end
;
901 unsigned blocksize
= head
->b_size
;
903 struct buffer_head
*next
;
905 for (bh
= head
, block_start
= 0;
906 ret
== 0 && (bh
!= head
|| !block_start
);
907 block_start
= block_end
, bh
= next
) {
908 next
= bh
->b_this_page
;
909 block_end
= block_start
+ blocksize
;
910 if (block_end
<= from
|| block_start
>= to
) {
911 if (partial
&& !buffer_uptodate(bh
))
915 err
= (*fn
)(handle
, bh
);
923 * To preserve ordering, it is essential that the hole instantiation and
924 * the data write be encapsulated in a single transaction. We cannot
925 * close off a transaction and start a new one between the ext4_get_block()
926 * and the commit_write(). So doing the jbd2_journal_start at the start of
927 * prepare_write() is the right place.
929 * Also, this function can nest inside ext4_writepage(). In that case, we
930 * *know* that ext4_writepage() has generated enough buffer credits to do the
931 * whole page. So we won't block on the journal in that case, which is good,
932 * because the caller may be PF_MEMALLOC.
934 * By accident, ext4 can be reentered when a transaction is open via
935 * quota file writes. If we were to commit the transaction while thus
936 * reentered, there can be a deadlock - we would be holding a quota
937 * lock, and the commit would never complete if another thread had a
938 * transaction open and was blocking on the quota lock - a ranking
941 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
942 * will _not_ run commit under these circumstances because handle->h_ref
943 * is elevated. We'll still have enough credits for the tiny quotafile
946 int do_journal_get_write_access(handle_t
*handle
,
947 struct buffer_head
*bh
)
949 int dirty
= buffer_dirty(bh
);
952 if (!buffer_mapped(bh
) || buffer_freed(bh
))
955 * __block_write_begin() could have dirtied some buffers. Clean
956 * the dirty bit as jbd2_journal_get_write_access() could complain
957 * otherwise about fs integrity issues. Setting of the dirty bit
958 * by __block_write_begin() isn't a real problem here as we clear
959 * the bit before releasing a page lock and thus writeback cannot
960 * ever write the buffer.
963 clear_buffer_dirty(bh
);
964 ret
= ext4_journal_get_write_access(handle
, bh
);
966 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
970 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
971 struct buffer_head
*bh_result
, int create
);
972 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
973 loff_t pos
, unsigned len
, unsigned flags
,
974 struct page
**pagep
, void **fsdata
)
976 struct inode
*inode
= mapping
->host
;
977 int ret
, needed_blocks
;
984 trace_ext4_write_begin(inode
, pos
, len
, flags
);
986 * Reserve one block more for addition to orphan list in case
987 * we allocate blocks but write fails for some reason
989 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
990 index
= pos
>> PAGE_CACHE_SHIFT
;
991 from
= pos
& (PAGE_CACHE_SIZE
- 1);
994 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
995 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1004 * grab_cache_page_write_begin() can take a long time if the
1005 * system is thrashing due to memory pressure, or if the page
1006 * is being written back. So grab it first before we start
1007 * the transaction handle. This also allows us to allocate
1008 * the page (if needed) without using GFP_NOFS.
1011 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1017 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1018 if (IS_ERR(handle
)) {
1019 page_cache_release(page
);
1020 return PTR_ERR(handle
);
1024 if (page
->mapping
!= mapping
) {
1025 /* The page got truncated from under us */
1027 page_cache_release(page
);
1028 ext4_journal_stop(handle
);
1031 wait_on_page_writeback(page
);
1033 if (ext4_should_dioread_nolock(inode
))
1034 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
1036 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1038 if (!ret
&& ext4_should_journal_data(inode
)) {
1039 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1041 do_journal_get_write_access
);
1047 * __block_write_begin may have instantiated a few blocks
1048 * outside i_size. Trim these off again. Don't need
1049 * i_size_read because we hold i_mutex.
1051 * Add inode to orphan list in case we crash before
1054 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1055 ext4_orphan_add(handle
, inode
);
1057 ext4_journal_stop(handle
);
1058 if (pos
+ len
> inode
->i_size
) {
1059 ext4_truncate_failed_write(inode
);
1061 * If truncate failed early the inode might
1062 * still be on the orphan list; we need to
1063 * make sure the inode is removed from the
1064 * orphan list in that case.
1067 ext4_orphan_del(NULL
, inode
);
1070 if (ret
== -ENOSPC
&&
1071 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1073 page_cache_release(page
);
1080 /* For write_end() in data=journal mode */
1081 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1083 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1085 set_buffer_uptodate(bh
);
1086 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1089 static int ext4_generic_write_end(struct file
*file
,
1090 struct address_space
*mapping
,
1091 loff_t pos
, unsigned len
, unsigned copied
,
1092 struct page
*page
, void *fsdata
)
1094 int i_size_changed
= 0;
1095 struct inode
*inode
= mapping
->host
;
1096 handle_t
*handle
= ext4_journal_current_handle();
1098 if (ext4_has_inline_data(inode
))
1099 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1102 copied
= block_write_end(file
, mapping
, pos
,
1103 len
, copied
, page
, fsdata
);
1106 * No need to use i_size_read() here, the i_size
1107 * cannot change under us because we hold i_mutex.
1109 * But it's important to update i_size while still holding page lock:
1110 * page writeout could otherwise come in and zero beyond i_size.
1112 if (pos
+ copied
> inode
->i_size
) {
1113 i_size_write(inode
, pos
+ copied
);
1117 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1118 /* We need to mark inode dirty even if
1119 * new_i_size is less that inode->i_size
1120 * bu greater than i_disksize.(hint delalloc)
1122 ext4_update_i_disksize(inode
, (pos
+ copied
));
1126 page_cache_release(page
);
1129 * Don't mark the inode dirty under page lock. First, it unnecessarily
1130 * makes the holding time of page lock longer. Second, it forces lock
1131 * ordering of page lock and transaction start for journaling
1135 ext4_mark_inode_dirty(handle
, inode
);
1141 * We need to pick up the new inode size which generic_commit_write gave us
1142 * `file' can be NULL - eg, when called from page_symlink().
1144 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1145 * buffers are managed internally.
1147 static int ext4_ordered_write_end(struct file
*file
,
1148 struct address_space
*mapping
,
1149 loff_t pos
, unsigned len
, unsigned copied
,
1150 struct page
*page
, void *fsdata
)
1152 handle_t
*handle
= ext4_journal_current_handle();
1153 struct inode
*inode
= mapping
->host
;
1156 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1157 ret
= ext4_jbd2_file_inode(handle
, inode
);
1160 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1163 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1164 /* if we have allocated more blocks and copied
1165 * less. We will have blocks allocated outside
1166 * inode->i_size. So truncate them
1168 ext4_orphan_add(handle
, inode
);
1173 page_cache_release(page
);
1176 ret2
= ext4_journal_stop(handle
);
1180 if (pos
+ len
> inode
->i_size
) {
1181 ext4_truncate_failed_write(inode
);
1183 * If truncate failed early the inode might still be
1184 * on the orphan list; we need to make sure the inode
1185 * is removed from the orphan list in that case.
1188 ext4_orphan_del(NULL
, inode
);
1192 return ret
? ret
: copied
;
1195 static int ext4_writeback_write_end(struct file
*file
,
1196 struct address_space
*mapping
,
1197 loff_t pos
, unsigned len
, unsigned copied
,
1198 struct page
*page
, void *fsdata
)
1200 handle_t
*handle
= ext4_journal_current_handle();
1201 struct inode
*inode
= mapping
->host
;
1204 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1205 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1208 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1209 /* if we have allocated more blocks and copied
1210 * less. We will have blocks allocated outside
1211 * inode->i_size. So truncate them
1213 ext4_orphan_add(handle
, inode
);
1218 ret2
= ext4_journal_stop(handle
);
1222 if (pos
+ len
> inode
->i_size
) {
1223 ext4_truncate_failed_write(inode
);
1225 * If truncate failed early the inode might still be
1226 * on the orphan list; we need to make sure the inode
1227 * is removed from the orphan list in that case.
1230 ext4_orphan_del(NULL
, inode
);
1233 return ret
? ret
: copied
;
1236 static int ext4_journalled_write_end(struct file
*file
,
1237 struct address_space
*mapping
,
1238 loff_t pos
, unsigned len
, unsigned copied
,
1239 struct page
*page
, void *fsdata
)
1241 handle_t
*handle
= ext4_journal_current_handle();
1242 struct inode
*inode
= mapping
->host
;
1248 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1249 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1252 BUG_ON(!ext4_handle_valid(handle
));
1254 if (ext4_has_inline_data(inode
))
1255 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1259 if (!PageUptodate(page
))
1261 page_zero_new_buffers(page
, from
+copied
, to
);
1264 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1265 to
, &partial
, write_end_fn
);
1267 SetPageUptodate(page
);
1269 new_i_size
= pos
+ copied
;
1270 if (new_i_size
> inode
->i_size
)
1271 i_size_write(inode
, pos
+copied
);
1272 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1273 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1274 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1275 ext4_update_i_disksize(inode
, new_i_size
);
1276 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1282 page_cache_release(page
);
1283 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1284 /* if we have allocated more blocks and copied
1285 * less. We will have blocks allocated outside
1286 * inode->i_size. So truncate them
1288 ext4_orphan_add(handle
, inode
);
1290 ret2
= ext4_journal_stop(handle
);
1293 if (pos
+ len
> inode
->i_size
) {
1294 ext4_truncate_failed_write(inode
);
1296 * If truncate failed early the inode might still be
1297 * on the orphan list; we need to make sure the inode
1298 * is removed from the orphan list in that case.
1301 ext4_orphan_del(NULL
, inode
);
1304 return ret
? ret
: copied
;
1308 * Reserve a single cluster located at lblock
1310 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1313 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1314 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1315 unsigned int md_needed
;
1317 ext4_lblk_t save_last_lblock
;
1321 * We will charge metadata quota at writeout time; this saves
1322 * us from metadata over-estimation, though we may go over by
1323 * a small amount in the end. Here we just reserve for data.
1325 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1330 * recalculate the amount of metadata blocks to reserve
1331 * in order to allocate nrblocks
1332 * worse case is one extent per block
1335 spin_lock(&ei
->i_block_reservation_lock
);
1337 * ext4_calc_metadata_amount() has side effects, which we have
1338 * to be prepared undo if we fail to claim space.
1340 save_len
= ei
->i_da_metadata_calc_len
;
1341 save_last_lblock
= ei
->i_da_metadata_calc_last_lblock
;
1342 md_needed
= EXT4_NUM_B2C(sbi
,
1343 ext4_calc_metadata_amount(inode
, lblock
));
1344 trace_ext4_da_reserve_space(inode
, md_needed
);
1347 * We do still charge estimated metadata to the sb though;
1348 * we cannot afford to run out of free blocks.
1350 if (ext4_claim_free_clusters(sbi
, md_needed
+ 1, 0)) {
1351 ei
->i_da_metadata_calc_len
= save_len
;
1352 ei
->i_da_metadata_calc_last_lblock
= save_last_lblock
;
1353 spin_unlock(&ei
->i_block_reservation_lock
);
1354 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1358 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1361 ei
->i_reserved_data_blocks
++;
1362 ei
->i_reserved_meta_blocks
+= md_needed
;
1363 spin_unlock(&ei
->i_block_reservation_lock
);
1365 return 0; /* success */
1368 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1370 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1371 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1374 return; /* Nothing to release, exit */
1376 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1378 trace_ext4_da_release_space(inode
, to_free
);
1379 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1381 * if there aren't enough reserved blocks, then the
1382 * counter is messed up somewhere. Since this
1383 * function is called from invalidate page, it's
1384 * harmless to return without any action.
1386 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1387 "ino %lu, to_free %d with only %d reserved "
1388 "data blocks", inode
->i_ino
, to_free
,
1389 ei
->i_reserved_data_blocks
);
1391 to_free
= ei
->i_reserved_data_blocks
;
1393 ei
->i_reserved_data_blocks
-= to_free
;
1395 if (ei
->i_reserved_data_blocks
== 0) {
1397 * We can release all of the reserved metadata blocks
1398 * only when we have written all of the delayed
1399 * allocation blocks.
1400 * Note that in case of bigalloc, i_reserved_meta_blocks,
1401 * i_reserved_data_blocks, etc. refer to number of clusters.
1403 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
1404 ei
->i_reserved_meta_blocks
);
1405 ei
->i_reserved_meta_blocks
= 0;
1406 ei
->i_da_metadata_calc_len
= 0;
1409 /* update fs dirty data blocks counter */
1410 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1412 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1414 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1417 static void ext4_da_page_release_reservation(struct page
*page
,
1418 unsigned long offset
)
1421 struct buffer_head
*head
, *bh
;
1422 unsigned int curr_off
= 0;
1423 struct inode
*inode
= page
->mapping
->host
;
1424 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1428 head
= page_buffers(page
);
1431 unsigned int next_off
= curr_off
+ bh
->b_size
;
1433 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1435 clear_buffer_delay(bh
);
1437 curr_off
= next_off
;
1438 } while ((bh
= bh
->b_this_page
) != head
);
1441 lblk
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1442 ext4_es_remove_extent(inode
, lblk
, to_release
);
1445 /* If we have released all the blocks belonging to a cluster, then we
1446 * need to release the reserved space for that cluster. */
1447 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1448 while (num_clusters
> 0) {
1449 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1450 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1451 if (sbi
->s_cluster_ratio
== 1 ||
1452 !ext4_find_delalloc_cluster(inode
, lblk
))
1453 ext4_da_release_space(inode
, 1);
1460 * Delayed allocation stuff
1464 * mpage_da_submit_io - walks through extent of pages and try to write
1465 * them with writepage() call back
1467 * @mpd->inode: inode
1468 * @mpd->first_page: first page of the extent
1469 * @mpd->next_page: page after the last page of the extent
1471 * By the time mpage_da_submit_io() is called we expect all blocks
1472 * to be allocated. this may be wrong if allocation failed.
1474 * As pages are already locked by write_cache_pages(), we can't use it
1476 static int mpage_da_submit_io(struct mpage_da_data
*mpd
,
1477 struct ext4_map_blocks
*map
)
1479 struct pagevec pvec
;
1480 unsigned long index
, end
;
1481 int ret
= 0, err
, nr_pages
, i
;
1482 struct inode
*inode
= mpd
->inode
;
1483 struct address_space
*mapping
= inode
->i_mapping
;
1484 loff_t size
= i_size_read(inode
);
1485 unsigned int len
, block_start
;
1486 struct buffer_head
*bh
, *page_bufs
= NULL
;
1487 sector_t pblock
= 0, cur_logical
= 0;
1488 struct ext4_io_submit io_submit
;
1490 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1491 memset(&io_submit
, 0, sizeof(io_submit
));
1493 * We need to start from the first_page to the next_page - 1
1494 * to make sure we also write the mapped dirty buffer_heads.
1495 * If we look at mpd->b_blocknr we would only be looking
1496 * at the currently mapped buffer_heads.
1498 index
= mpd
->first_page
;
1499 end
= mpd
->next_page
- 1;
1501 pagevec_init(&pvec
, 0);
1502 while (index
<= end
) {
1503 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1506 for (i
= 0; i
< nr_pages
; i
++) {
1508 struct page
*page
= pvec
.pages
[i
];
1510 index
= page
->index
;
1514 if (index
== size
>> PAGE_CACHE_SHIFT
)
1515 len
= size
& ~PAGE_CACHE_MASK
;
1517 len
= PAGE_CACHE_SIZE
;
1519 cur_logical
= index
<< (PAGE_CACHE_SHIFT
-
1521 pblock
= map
->m_pblk
+ (cur_logical
-
1526 BUG_ON(!PageLocked(page
));
1527 BUG_ON(PageWriteback(page
));
1529 bh
= page_bufs
= page_buffers(page
);
1532 if (map
&& (cur_logical
>= map
->m_lblk
) &&
1533 (cur_logical
<= (map
->m_lblk
+
1534 (map
->m_len
- 1)))) {
1535 if (buffer_delay(bh
)) {
1536 clear_buffer_delay(bh
);
1537 bh
->b_blocknr
= pblock
;
1539 if (buffer_unwritten(bh
) ||
1541 BUG_ON(bh
->b_blocknr
!= pblock
);
1542 if (map
->m_flags
& EXT4_MAP_UNINIT
)
1543 set_buffer_uninit(bh
);
1544 clear_buffer_unwritten(bh
);
1548 * skip page if block allocation undone and
1551 if (ext4_bh_delay_or_unwritten(NULL
, bh
))
1553 bh
= bh
->b_this_page
;
1554 block_start
+= bh
->b_size
;
1557 } while (bh
!= page_bufs
);
1564 clear_page_dirty_for_io(page
);
1565 err
= ext4_bio_write_page(&io_submit
, page
, len
,
1568 mpd
->pages_written
++;
1570 * In error case, we have to continue because
1571 * remaining pages are still locked
1576 pagevec_release(&pvec
);
1578 ext4_io_submit(&io_submit
);
1582 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
)
1586 struct pagevec pvec
;
1587 struct inode
*inode
= mpd
->inode
;
1588 struct address_space
*mapping
= inode
->i_mapping
;
1589 ext4_lblk_t start
, last
;
1591 index
= mpd
->first_page
;
1592 end
= mpd
->next_page
- 1;
1594 start
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1595 last
= end
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1596 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1598 pagevec_init(&pvec
, 0);
1599 while (index
<= end
) {
1600 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1603 for (i
= 0; i
< nr_pages
; i
++) {
1604 struct page
*page
= pvec
.pages
[i
];
1605 if (page
->index
> end
)
1607 BUG_ON(!PageLocked(page
));
1608 BUG_ON(PageWriteback(page
));
1609 block_invalidatepage(page
, 0);
1610 ClearPageUptodate(page
);
1613 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1614 pagevec_release(&pvec
);
1619 static void ext4_print_free_blocks(struct inode
*inode
)
1621 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1622 struct super_block
*sb
= inode
->i_sb
;
1624 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1625 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1626 ext4_count_free_clusters(inode
->i_sb
)));
1627 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1628 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1629 (long long) EXT4_C2B(EXT4_SB(inode
->i_sb
),
1630 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1631 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1632 (long long) EXT4_C2B(EXT4_SB(inode
->i_sb
),
1633 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1634 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1635 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1636 EXT4_I(inode
)->i_reserved_data_blocks
);
1637 ext4_msg(sb
, KERN_CRIT
, "i_reserved_meta_blocks=%u",
1638 EXT4_I(inode
)->i_reserved_meta_blocks
);
1643 * mpage_da_map_and_submit - go through given space, map them
1644 * if necessary, and then submit them for I/O
1646 * @mpd - bh describing space
1648 * The function skips space we know is already mapped to disk blocks.
1651 static void mpage_da_map_and_submit(struct mpage_da_data
*mpd
)
1653 int err
, blks
, get_blocks_flags
;
1654 struct ext4_map_blocks map
, *mapp
= NULL
;
1655 sector_t next
= mpd
->b_blocknr
;
1656 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1657 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
1658 handle_t
*handle
= NULL
;
1661 * If the blocks are mapped already, or we couldn't accumulate
1662 * any blocks, then proceed immediately to the submission stage.
1664 if ((mpd
->b_size
== 0) ||
1665 ((mpd
->b_state
& (1 << BH_Mapped
)) &&
1666 !(mpd
->b_state
& (1 << BH_Delay
)) &&
1667 !(mpd
->b_state
& (1 << BH_Unwritten
))))
1670 handle
= ext4_journal_current_handle();
1674 * Call ext4_map_blocks() to allocate any delayed allocation
1675 * blocks, or to convert an uninitialized extent to be
1676 * initialized (in the case where we have written into
1677 * one or more preallocated blocks).
1679 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1680 * indicate that we are on the delayed allocation path. This
1681 * affects functions in many different parts of the allocation
1682 * call path. This flag exists primarily because we don't
1683 * want to change *many* call functions, so ext4_map_blocks()
1684 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1685 * inode's allocation semaphore is taken.
1687 * If the blocks in questions were delalloc blocks, set
1688 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1689 * variables are updated after the blocks have been allocated.
1692 map
.m_len
= max_blocks
;
1693 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
1694 if (ext4_should_dioread_nolock(mpd
->inode
))
1695 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
1696 if (mpd
->b_state
& (1 << BH_Delay
))
1697 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
1699 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
1701 struct super_block
*sb
= mpd
->inode
->i_sb
;
1705 * If get block returns EAGAIN or ENOSPC and there
1706 * appears to be free blocks we will just let
1707 * mpage_da_submit_io() unlock all of the pages.
1712 if (err
== -ENOSPC
&& ext4_count_free_clusters(sb
)) {
1718 * get block failure will cause us to loop in
1719 * writepages, because a_ops->writepage won't be able
1720 * to make progress. The page will be redirtied by
1721 * writepage and writepages will again try to write
1724 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
1725 ext4_msg(sb
, KERN_CRIT
,
1726 "delayed block allocation failed for inode %lu "
1727 "at logical offset %llu with max blocks %zd "
1728 "with error %d", mpd
->inode
->i_ino
,
1729 (unsigned long long) next
,
1730 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1731 ext4_msg(sb
, KERN_CRIT
,
1732 "This should not happen!! Data will be lost");
1734 ext4_print_free_blocks(mpd
->inode
);
1736 /* invalidate all the pages */
1737 ext4_da_block_invalidatepages(mpd
);
1739 /* Mark this page range as having been completed */
1746 if (map
.m_flags
& EXT4_MAP_NEW
) {
1747 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
1750 for (i
= 0; i
< map
.m_len
; i
++)
1751 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
1755 * Update on-disk size along with block allocation.
1757 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
1758 if (disksize
> i_size_read(mpd
->inode
))
1759 disksize
= i_size_read(mpd
->inode
);
1760 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
1761 ext4_update_i_disksize(mpd
->inode
, disksize
);
1762 err
= ext4_mark_inode_dirty(handle
, mpd
->inode
);
1764 ext4_error(mpd
->inode
->i_sb
,
1765 "Failed to mark inode %lu dirty",
1770 mpage_da_submit_io(mpd
, mapp
);
1774 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1775 (1 << BH_Delay) | (1 << BH_Unwritten))
1778 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1780 * @mpd->lbh - extent of blocks
1781 * @logical - logical number of the block in the file
1782 * @b_state - b_state of the buffer head added
1784 * the function is used to collect contig. blocks in same state
1786 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, sector_t logical
,
1787 unsigned long b_state
)
1790 int blkbits
= mpd
->inode
->i_blkbits
;
1791 int nrblocks
= mpd
->b_size
>> blkbits
;
1794 * XXX Don't go larger than mballoc is willing to allocate
1795 * This is a stopgap solution. We eventually need to fold
1796 * mpage_da_submit_io() into this function and then call
1797 * ext4_map_blocks() multiple times in a loop
1799 if (nrblocks
>= (8*1024*1024 >> blkbits
))
1802 /* check if the reserved journal credits might overflow */
1803 if (!ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
)) {
1804 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1806 * With non-extent format we are limited by the journal
1807 * credit available. Total credit needed to insert
1808 * nrblocks contiguous blocks is dependent on the
1809 * nrblocks. So limit nrblocks.
1815 * First block in the extent
1817 if (mpd
->b_size
== 0) {
1818 mpd
->b_blocknr
= logical
;
1819 mpd
->b_size
= 1 << blkbits
;
1820 mpd
->b_state
= b_state
& BH_FLAGS
;
1824 next
= mpd
->b_blocknr
+ nrblocks
;
1826 * Can we merge the block to our big extent?
1828 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
1829 mpd
->b_size
+= 1 << blkbits
;
1835 * We couldn't merge the block to our extent, so we
1836 * need to flush current extent and start new one
1838 mpage_da_map_and_submit(mpd
);
1842 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1844 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1848 * This function is grabs code from the very beginning of
1849 * ext4_map_blocks, but assumes that the caller is from delayed write
1850 * time. This function looks up the requested blocks and sets the
1851 * buffer delay bit under the protection of i_data_sem.
1853 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1854 struct ext4_map_blocks
*map
,
1855 struct buffer_head
*bh
)
1857 struct extent_status es
;
1859 sector_t invalid_block
= ~((sector_t
) 0xffff);
1860 #ifdef ES_AGGRESSIVE_TEST
1861 struct ext4_map_blocks orig_map
;
1863 memcpy(&orig_map
, map
, sizeof(*map
));
1866 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1870 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1871 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1872 (unsigned long) map
->m_lblk
);
1874 /* Lookup extent status tree firstly */
1875 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1877 if (ext4_es_is_hole(&es
)) {
1879 down_read((&EXT4_I(inode
)->i_data_sem
));
1884 * Delayed extent could be allocated by fallocate.
1885 * So we need to check it.
1887 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1888 map_bh(bh
, inode
->i_sb
, invalid_block
);
1890 set_buffer_delay(bh
);
1894 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1895 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1896 if (retval
> map
->m_len
)
1897 retval
= map
->m_len
;
1898 map
->m_len
= retval
;
1899 if (ext4_es_is_written(&es
))
1900 map
->m_flags
|= EXT4_MAP_MAPPED
;
1901 else if (ext4_es_is_unwritten(&es
))
1902 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1906 #ifdef ES_AGGRESSIVE_TEST
1907 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1913 * Try to see if we can get the block without requesting a new
1914 * file system block.
1916 down_read((&EXT4_I(inode
)->i_data_sem
));
1917 if (ext4_has_inline_data(inode
)) {
1919 * We will soon create blocks for this page, and let
1920 * us pretend as if the blocks aren't allocated yet.
1921 * In case of clusters, we have to handle the work
1922 * of mapping from cluster so that the reserved space
1923 * is calculated properly.
1925 if ((EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) &&
1926 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
1927 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
1929 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1930 retval
= ext4_ext_map_blocks(NULL
, inode
, map
,
1931 EXT4_GET_BLOCKS_NO_PUT_HOLE
);
1933 retval
= ext4_ind_map_blocks(NULL
, inode
, map
,
1934 EXT4_GET_BLOCKS_NO_PUT_HOLE
);
1940 * XXX: __block_prepare_write() unmaps passed block,
1943 /* If the block was allocated from previously allocated cluster,
1944 * then we dont need to reserve it again. */
1945 if (!(map
->m_flags
& EXT4_MAP_FROM_CLUSTER
)) {
1946 ret
= ext4_da_reserve_space(inode
, iblock
);
1948 /* not enough space to reserve */
1954 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1955 ~0, EXTENT_STATUS_DELAYED
);
1961 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1962 * and it should not appear on the bh->b_state.
1964 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
1966 map_bh(bh
, inode
->i_sb
, invalid_block
);
1968 set_buffer_delay(bh
);
1969 } else if (retval
> 0) {
1971 unsigned long long status
;
1973 #ifdef ES_AGGRESSIVE_TEST
1974 if (retval
!= map
->m_len
) {
1975 printk("ES len assertation failed for inode: %lu "
1976 "retval %d != map->m_len %d "
1977 "in %s (lookup)\n", inode
->i_ino
, retval
,
1978 map
->m_len
, __func__
);
1982 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1983 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1984 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1985 map
->m_pblk
, status
);
1991 up_read((&EXT4_I(inode
)->i_data_sem
));
1997 * This is a special get_blocks_t callback which is used by
1998 * ext4_da_write_begin(). It will either return mapped block or
1999 * reserve space for a single block.
2001 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2002 * We also have b_blocknr = -1 and b_bdev initialized properly
2004 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2005 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2006 * initialized properly.
2008 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2009 struct buffer_head
*bh
, int create
)
2011 struct ext4_map_blocks map
;
2014 BUG_ON(create
== 0);
2015 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
2017 map
.m_lblk
= iblock
;
2021 * first, we need to know whether the block is allocated already
2022 * preallocated blocks are unmapped but should treated
2023 * the same as allocated blocks.
2025 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
2029 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
2030 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
2032 if (buffer_unwritten(bh
)) {
2033 /* A delayed write to unwritten bh should be marked
2034 * new and mapped. Mapped ensures that we don't do
2035 * get_block multiple times when we write to the same
2036 * offset and new ensures that we do proper zero out
2037 * for partial write.
2040 set_buffer_mapped(bh
);
2045 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2051 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2057 static int __ext4_journalled_writepage(struct page
*page
,
2060 struct address_space
*mapping
= page
->mapping
;
2061 struct inode
*inode
= mapping
->host
;
2062 struct buffer_head
*page_bufs
= NULL
;
2063 handle_t
*handle
= NULL
;
2064 int ret
= 0, err
= 0;
2065 int inline_data
= ext4_has_inline_data(inode
);
2066 struct buffer_head
*inode_bh
= NULL
;
2068 ClearPageChecked(page
);
2071 BUG_ON(page
->index
!= 0);
2072 BUG_ON(len
> ext4_get_max_inline_size(inode
));
2073 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
2074 if (inode_bh
== NULL
)
2077 page_bufs
= page_buffers(page
);
2082 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
2085 /* As soon as we unlock the page, it can go away, but we have
2086 * references to buffers so we are safe */
2089 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2090 ext4_writepage_trans_blocks(inode
));
2091 if (IS_ERR(handle
)) {
2092 ret
= PTR_ERR(handle
);
2096 BUG_ON(!ext4_handle_valid(handle
));
2099 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
2101 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
2104 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2105 do_journal_get_write_access
);
2107 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2112 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
2113 err
= ext4_journal_stop(handle
);
2117 if (!ext4_has_inline_data(inode
))
2118 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
2120 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2127 * Note that we don't need to start a transaction unless we're journaling data
2128 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2129 * need to file the inode to the transaction's list in ordered mode because if
2130 * we are writing back data added by write(), the inode is already there and if
2131 * we are writing back data modified via mmap(), no one guarantees in which
2132 * transaction the data will hit the disk. In case we are journaling data, we
2133 * cannot start transaction directly because transaction start ranks above page
2134 * lock so we have to do some magic.
2136 * This function can get called via...
2137 * - ext4_da_writepages after taking page lock (have journal handle)
2138 * - journal_submit_inode_data_buffers (no journal handle)
2139 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2140 * - grab_page_cache when doing write_begin (have journal handle)
2142 * We don't do any block allocation in this function. If we have page with
2143 * multiple blocks we need to write those buffer_heads that are mapped. This
2144 * is important for mmaped based write. So if we do with blocksize 1K
2145 * truncate(f, 1024);
2146 * a = mmap(f, 0, 4096);
2148 * truncate(f, 4096);
2149 * we have in the page first buffer_head mapped via page_mkwrite call back
2150 * but other buffer_heads would be unmapped but dirty (dirty done via the
2151 * do_wp_page). So writepage should write the first block. If we modify
2152 * the mmap area beyond 1024 we will again get a page_fault and the
2153 * page_mkwrite callback will do the block allocation and mark the
2154 * buffer_heads mapped.
2156 * We redirty the page if we have any buffer_heads that is either delay or
2157 * unwritten in the page.
2159 * We can get recursively called as show below.
2161 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2164 * But since we don't do any block allocation we should not deadlock.
2165 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2167 static int ext4_writepage(struct page
*page
,
2168 struct writeback_control
*wbc
)
2173 struct buffer_head
*page_bufs
= NULL
;
2174 struct inode
*inode
= page
->mapping
->host
;
2175 struct ext4_io_submit io_submit
;
2177 trace_ext4_writepage(page
);
2178 size
= i_size_read(inode
);
2179 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2180 len
= size
& ~PAGE_CACHE_MASK
;
2182 len
= PAGE_CACHE_SIZE
;
2184 page_bufs
= page_buffers(page
);
2186 * We cannot do block allocation or other extent handling in this
2187 * function. If there are buffers needing that, we have to redirty
2188 * the page. But we may reach here when we do a journal commit via
2189 * journal_submit_inode_data_buffers() and in that case we must write
2190 * allocated buffers to achieve data=ordered mode guarantees.
2192 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2193 ext4_bh_delay_or_unwritten
)) {
2194 redirty_page_for_writepage(wbc
, page
);
2195 if (current
->flags
& PF_MEMALLOC
) {
2197 * For memory cleaning there's no point in writing only
2198 * some buffers. So just bail out. Warn if we came here
2199 * from direct reclaim.
2201 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2208 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2210 * It's mmapped pagecache. Add buffers and journal it. There
2211 * doesn't seem much point in redirtying the page here.
2213 return __ext4_journalled_writepage(page
, len
);
2215 memset(&io_submit
, 0, sizeof(io_submit
));
2216 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
);
2217 ext4_io_submit(&io_submit
);
2222 * This is called via ext4_da_writepages() to
2223 * calculate the total number of credits to reserve to fit
2224 * a single extent allocation into a single transaction,
2225 * ext4_da_writpeages() will loop calling this before
2226 * the block allocation.
2229 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2231 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2234 * With non-extent format the journal credit needed to
2235 * insert nrblocks contiguous block is dependent on
2236 * number of contiguous block. So we will limit
2237 * number of contiguous block to a sane value
2239 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
2240 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2241 max_blocks
= EXT4_MAX_TRANS_DATA
;
2243 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2247 * write_cache_pages_da - walk the list of dirty pages of the given
2248 * address space and accumulate pages that need writing, and call
2249 * mpage_da_map_and_submit to map a single contiguous memory region
2250 * and then write them.
2252 static int write_cache_pages_da(handle_t
*handle
,
2253 struct address_space
*mapping
,
2254 struct writeback_control
*wbc
,
2255 struct mpage_da_data
*mpd
,
2256 pgoff_t
*done_index
)
2258 struct buffer_head
*bh
, *head
;
2259 struct inode
*inode
= mapping
->host
;
2260 struct pagevec pvec
;
2261 unsigned int nr_pages
;
2264 long nr_to_write
= wbc
->nr_to_write
;
2265 int i
, tag
, ret
= 0;
2267 memset(mpd
, 0, sizeof(struct mpage_da_data
));
2270 pagevec_init(&pvec
, 0);
2271 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2272 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2274 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2275 tag
= PAGECACHE_TAG_TOWRITE
;
2277 tag
= PAGECACHE_TAG_DIRTY
;
2279 *done_index
= index
;
2280 while (index
<= end
) {
2281 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2282 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2286 for (i
= 0; i
< nr_pages
; i
++) {
2287 struct page
*page
= pvec
.pages
[i
];
2290 * At this point, the page may be truncated or
2291 * invalidated (changing page->mapping to NULL), or
2292 * even swizzled back from swapper_space to tmpfs file
2293 * mapping. However, page->index will not change
2294 * because we have a reference on the page.
2296 if (page
->index
> end
)
2299 *done_index
= page
->index
+ 1;
2302 * If we can't merge this page, and we have
2303 * accumulated an contiguous region, write it
2305 if ((mpd
->next_page
!= page
->index
) &&
2306 (mpd
->next_page
!= mpd
->first_page
)) {
2307 mpage_da_map_and_submit(mpd
);
2308 goto ret_extent_tail
;
2314 * If the page is no longer dirty, or its
2315 * mapping no longer corresponds to inode we
2316 * are writing (which means it has been
2317 * truncated or invalidated), or the page is
2318 * already under writeback and we are not
2319 * doing a data integrity writeback, skip the page
2321 if (!PageDirty(page
) ||
2322 (PageWriteback(page
) &&
2323 (wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2324 unlikely(page
->mapping
!= mapping
)) {
2329 wait_on_page_writeback(page
);
2330 BUG_ON(PageWriteback(page
));
2333 * If we have inline data and arrive here, it means that
2334 * we will soon create the block for the 1st page, so
2335 * we'd better clear the inline data here.
2337 if (ext4_has_inline_data(inode
)) {
2338 BUG_ON(ext4_test_inode_state(inode
,
2339 EXT4_STATE_MAY_INLINE_DATA
));
2340 ext4_destroy_inline_data(handle
, inode
);
2343 if (mpd
->next_page
!= page
->index
)
2344 mpd
->first_page
= page
->index
;
2345 mpd
->next_page
= page
->index
+ 1;
2346 logical
= (sector_t
) page
->index
<<
2347 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2349 /* Add all dirty buffers to mpd */
2350 head
= page_buffers(page
);
2353 BUG_ON(buffer_locked(bh
));
2355 * We need to try to allocate unmapped blocks
2356 * in the same page. Otherwise we won't make
2357 * progress with the page in ext4_writepage
2359 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2360 mpage_add_bh_to_extent(mpd
, logical
,
2363 goto ret_extent_tail
;
2364 } else if (buffer_dirty(bh
) &&
2365 buffer_mapped(bh
)) {
2367 * mapped dirty buffer. We need to
2368 * update the b_state because we look
2369 * at b_state in mpage_da_map_blocks.
2370 * We don't update b_size because if we
2371 * find an unmapped buffer_head later
2372 * we need to use the b_state flag of
2375 if (mpd
->b_size
== 0)
2377 bh
->b_state
& BH_FLAGS
;
2380 } while ((bh
= bh
->b_this_page
) != head
);
2382 if (nr_to_write
> 0) {
2384 if (nr_to_write
== 0 &&
2385 wbc
->sync_mode
== WB_SYNC_NONE
)
2387 * We stop writing back only if we are
2388 * not doing integrity sync. In case of
2389 * integrity sync we have to keep going
2390 * because someone may be concurrently
2391 * dirtying pages, and we might have
2392 * synced a lot of newly appeared dirty
2393 * pages, but have not synced all of the
2399 pagevec_release(&pvec
);
2404 ret
= MPAGE_DA_EXTENT_TAIL
;
2406 pagevec_release(&pvec
);
2412 static int ext4_da_writepages(struct address_space
*mapping
,
2413 struct writeback_control
*wbc
)
2416 int range_whole
= 0;
2417 handle_t
*handle
= NULL
;
2418 struct mpage_da_data mpd
;
2419 struct inode
*inode
= mapping
->host
;
2420 int pages_written
= 0;
2421 unsigned int max_pages
;
2422 int range_cyclic
, cycled
= 1, io_done
= 0;
2423 int needed_blocks
, ret
= 0;
2424 long desired_nr_to_write
, nr_to_writebump
= 0;
2425 loff_t range_start
= wbc
->range_start
;
2426 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2427 pgoff_t done_index
= 0;
2429 struct blk_plug plug
;
2431 trace_ext4_da_writepages(inode
, wbc
);
2434 * No pages to write? This is mainly a kludge to avoid starting
2435 * a transaction for special inodes like journal inode on last iput()
2436 * because that could violate lock ordering on umount
2438 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2442 * If the filesystem has aborted, it is read-only, so return
2443 * right away instead of dumping stack traces later on that
2444 * will obscure the real source of the problem. We test
2445 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2446 * the latter could be true if the filesystem is mounted
2447 * read-only, and in that case, ext4_da_writepages should
2448 * *never* be called, so if that ever happens, we would want
2451 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2454 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2457 range_cyclic
= wbc
->range_cyclic
;
2458 if (wbc
->range_cyclic
) {
2459 index
= mapping
->writeback_index
;
2462 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2463 wbc
->range_end
= LLONG_MAX
;
2464 wbc
->range_cyclic
= 0;
2467 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2468 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2472 * This works around two forms of stupidity. The first is in
2473 * the writeback code, which caps the maximum number of pages
2474 * written to be 1024 pages. This is wrong on multiple
2475 * levels; different architectues have a different page size,
2476 * which changes the maximum amount of data which gets
2477 * written. Secondly, 4 megabytes is way too small. XFS
2478 * forces this value to be 16 megabytes by multiplying
2479 * nr_to_write parameter by four, and then relies on its
2480 * allocator to allocate larger extents to make them
2481 * contiguous. Unfortunately this brings us to the second
2482 * stupidity, which is that ext4's mballoc code only allocates
2483 * at most 2048 blocks. So we force contiguous writes up to
2484 * the number of dirty blocks in the inode, or
2485 * sbi->max_writeback_mb_bump whichever is smaller.
2487 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2488 if (!range_cyclic
&& range_whole
) {
2489 if (wbc
->nr_to_write
== LONG_MAX
)
2490 desired_nr_to_write
= wbc
->nr_to_write
;
2492 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2494 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2496 if (desired_nr_to_write
> max_pages
)
2497 desired_nr_to_write
= max_pages
;
2499 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2500 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2501 wbc
->nr_to_write
= desired_nr_to_write
;
2505 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2506 tag_pages_for_writeback(mapping
, index
, end
);
2508 blk_start_plug(&plug
);
2509 while (!ret
&& wbc
->nr_to_write
> 0) {
2512 * we insert one extent at a time. So we need
2513 * credit needed for single extent allocation.
2514 * journalled mode is currently not supported
2517 BUG_ON(ext4_should_journal_data(inode
));
2518 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2520 /* start a new transaction*/
2521 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2523 if (IS_ERR(handle
)) {
2524 ret
= PTR_ERR(handle
);
2525 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2526 "%ld pages, ino %lu; err %d", __func__
,
2527 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2528 blk_finish_plug(&plug
);
2529 goto out_writepages
;
2533 * Now call write_cache_pages_da() to find the next
2534 * contiguous region of logical blocks that need
2535 * blocks to be allocated by ext4 and submit them.
2537 ret
= write_cache_pages_da(handle
, mapping
,
2538 wbc
, &mpd
, &done_index
);
2540 * If we have a contiguous extent of pages and we
2541 * haven't done the I/O yet, map the blocks and submit
2544 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2545 mpage_da_map_and_submit(&mpd
);
2546 ret
= MPAGE_DA_EXTENT_TAIL
;
2548 trace_ext4_da_write_pages(inode
, &mpd
);
2549 wbc
->nr_to_write
-= mpd
.pages_written
;
2551 ext4_journal_stop(handle
);
2553 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2554 /* commit the transaction which would
2555 * free blocks released in the transaction
2558 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2560 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2562 * Got one extent now try with rest of the pages.
2563 * If mpd.retval is set -EIO, journal is aborted.
2564 * So we don't need to write any more.
2566 pages_written
+= mpd
.pages_written
;
2569 } else if (wbc
->nr_to_write
)
2571 * There is no more writeout needed
2572 * or we requested for a noblocking writeout
2573 * and we found the device congested
2577 blk_finish_plug(&plug
);
2578 if (!io_done
&& !cycled
) {
2581 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2582 wbc
->range_end
= mapping
->writeback_index
- 1;
2587 wbc
->range_cyclic
= range_cyclic
;
2588 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2590 * set the writeback_index so that range_cyclic
2591 * mode will write it back later
2593 mapping
->writeback_index
= done_index
;
2596 wbc
->nr_to_write
-= nr_to_writebump
;
2597 wbc
->range_start
= range_start
;
2598 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2602 static int ext4_nonda_switch(struct super_block
*sb
)
2604 s64 free_blocks
, dirty_blocks
;
2605 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2608 * switch to non delalloc mode if we are running low
2609 * on free block. The free block accounting via percpu
2610 * counters can get slightly wrong with percpu_counter_batch getting
2611 * accumulated on each CPU without updating global counters
2612 * Delalloc need an accurate free block accounting. So switch
2613 * to non delalloc when we are near to error range.
2615 free_blocks
= EXT4_C2B(sbi
,
2616 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
));
2617 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2619 * Start pushing delalloc when 1/2 of free blocks are dirty.
2621 if (dirty_blocks
&& (free_blocks
< 2 * dirty_blocks
) &&
2622 !writeback_in_progress(sb
->s_bdi
) &&
2623 down_read_trylock(&sb
->s_umount
)) {
2624 writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2625 up_read(&sb
->s_umount
);
2628 if (2 * free_blocks
< 3 * dirty_blocks
||
2629 free_blocks
< (dirty_blocks
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2631 * free block count is less than 150% of dirty blocks
2632 * or free blocks is less than watermark
2639 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2640 loff_t pos
, unsigned len
, unsigned flags
,
2641 struct page
**pagep
, void **fsdata
)
2643 int ret
, retries
= 0;
2646 struct inode
*inode
= mapping
->host
;
2649 index
= pos
>> PAGE_CACHE_SHIFT
;
2651 if (ext4_nonda_switch(inode
->i_sb
)) {
2652 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2653 return ext4_write_begin(file
, mapping
, pos
,
2654 len
, flags
, pagep
, fsdata
);
2656 *fsdata
= (void *)0;
2657 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2659 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2660 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2670 * grab_cache_page_write_begin() can take a long time if the
2671 * system is thrashing due to memory pressure, or if the page
2672 * is being written back. So grab it first before we start
2673 * the transaction handle. This also allows us to allocate
2674 * the page (if needed) without using GFP_NOFS.
2677 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2683 * With delayed allocation, we don't log the i_disksize update
2684 * if there is delayed block allocation. But we still need
2685 * to journalling the i_disksize update if writes to the end
2686 * of file which has an already mapped buffer.
2689 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, 1);
2690 if (IS_ERR(handle
)) {
2691 page_cache_release(page
);
2692 return PTR_ERR(handle
);
2696 if (page
->mapping
!= mapping
) {
2697 /* The page got truncated from under us */
2699 page_cache_release(page
);
2700 ext4_journal_stop(handle
);
2703 /* In case writeback began while the page was unlocked */
2704 wait_on_page_writeback(page
);
2706 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2709 ext4_journal_stop(handle
);
2711 * block_write_begin may have instantiated a few blocks
2712 * outside i_size. Trim these off again. Don't need
2713 * i_size_read because we hold i_mutex.
2715 if (pos
+ len
> inode
->i_size
)
2716 ext4_truncate_failed_write(inode
);
2718 if (ret
== -ENOSPC
&&
2719 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2722 page_cache_release(page
);
2731 * Check if we should update i_disksize
2732 * when write to the end of file but not require block allocation
2734 static int ext4_da_should_update_i_disksize(struct page
*page
,
2735 unsigned long offset
)
2737 struct buffer_head
*bh
;
2738 struct inode
*inode
= page
->mapping
->host
;
2742 bh
= page_buffers(page
);
2743 idx
= offset
>> inode
->i_blkbits
;
2745 for (i
= 0; i
< idx
; i
++)
2746 bh
= bh
->b_this_page
;
2748 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2753 static int ext4_da_write_end(struct file
*file
,
2754 struct address_space
*mapping
,
2755 loff_t pos
, unsigned len
, unsigned copied
,
2756 struct page
*page
, void *fsdata
)
2758 struct inode
*inode
= mapping
->host
;
2760 handle_t
*handle
= ext4_journal_current_handle();
2762 unsigned long start
, end
;
2763 int write_mode
= (int)(unsigned long)fsdata
;
2765 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2766 switch (ext4_inode_journal_mode(inode
)) {
2767 case EXT4_INODE_ORDERED_DATA_MODE
:
2768 return ext4_ordered_write_end(file
, mapping
, pos
,
2769 len
, copied
, page
, fsdata
);
2770 case EXT4_INODE_WRITEBACK_DATA_MODE
:
2771 return ext4_writeback_write_end(file
, mapping
, pos
,
2772 len
, copied
, page
, fsdata
);
2778 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2779 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2780 end
= start
+ copied
- 1;
2783 * generic_write_end() will run mark_inode_dirty() if i_size
2784 * changes. So let's piggyback the i_disksize mark_inode_dirty
2787 new_i_size
= pos
+ copied
;
2788 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2789 if (ext4_has_inline_data(inode
) ||
2790 ext4_da_should_update_i_disksize(page
, end
)) {
2791 down_write(&EXT4_I(inode
)->i_data_sem
);
2792 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
2793 EXT4_I(inode
)->i_disksize
= new_i_size
;
2794 up_write(&EXT4_I(inode
)->i_data_sem
);
2795 /* We need to mark inode dirty even if
2796 * new_i_size is less that inode->i_size
2797 * bu greater than i_disksize.(hint delalloc)
2799 ext4_mark_inode_dirty(handle
, inode
);
2803 if (write_mode
!= CONVERT_INLINE_DATA
&&
2804 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
2805 ext4_has_inline_data(inode
))
2806 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
2809 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2815 ret2
= ext4_journal_stop(handle
);
2819 return ret
? ret
: copied
;
2822 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2825 * Drop reserved blocks
2827 BUG_ON(!PageLocked(page
));
2828 if (!page_has_buffers(page
))
2831 ext4_da_page_release_reservation(page
, offset
);
2834 ext4_invalidatepage(page
, offset
);
2840 * Force all delayed allocation blocks to be allocated for a given inode.
2842 int ext4_alloc_da_blocks(struct inode
*inode
)
2844 trace_ext4_alloc_da_blocks(inode
);
2846 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2847 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2851 * We do something simple for now. The filemap_flush() will
2852 * also start triggering a write of the data blocks, which is
2853 * not strictly speaking necessary (and for users of
2854 * laptop_mode, not even desirable). However, to do otherwise
2855 * would require replicating code paths in:
2857 * ext4_da_writepages() ->
2858 * write_cache_pages() ---> (via passed in callback function)
2859 * __mpage_da_writepage() -->
2860 * mpage_add_bh_to_extent()
2861 * mpage_da_map_blocks()
2863 * The problem is that write_cache_pages(), located in
2864 * mm/page-writeback.c, marks pages clean in preparation for
2865 * doing I/O, which is not desirable if we're not planning on
2868 * We could call write_cache_pages(), and then redirty all of
2869 * the pages by calling redirty_page_for_writepage() but that
2870 * would be ugly in the extreme. So instead we would need to
2871 * replicate parts of the code in the above functions,
2872 * simplifying them because we wouldn't actually intend to
2873 * write out the pages, but rather only collect contiguous
2874 * logical block extents, call the multi-block allocator, and
2875 * then update the buffer heads with the block allocations.
2877 * For now, though, we'll cheat by calling filemap_flush(),
2878 * which will map the blocks, and start the I/O, but not
2879 * actually wait for the I/O to complete.
2881 return filemap_flush(inode
->i_mapping
);
2885 * bmap() is special. It gets used by applications such as lilo and by
2886 * the swapper to find the on-disk block of a specific piece of data.
2888 * Naturally, this is dangerous if the block concerned is still in the
2889 * journal. If somebody makes a swapfile on an ext4 data-journaling
2890 * filesystem and enables swap, then they may get a nasty shock when the
2891 * data getting swapped to that swapfile suddenly gets overwritten by
2892 * the original zero's written out previously to the journal and
2893 * awaiting writeback in the kernel's buffer cache.
2895 * So, if we see any bmap calls here on a modified, data-journaled file,
2896 * take extra steps to flush any blocks which might be in the cache.
2898 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2900 struct inode
*inode
= mapping
->host
;
2905 * We can get here for an inline file via the FIBMAP ioctl
2907 if (ext4_has_inline_data(inode
))
2910 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2911 test_opt(inode
->i_sb
, DELALLOC
)) {
2913 * With delalloc we want to sync the file
2914 * so that we can make sure we allocate
2917 filemap_write_and_wait(mapping
);
2920 if (EXT4_JOURNAL(inode
) &&
2921 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2923 * This is a REALLY heavyweight approach, but the use of
2924 * bmap on dirty files is expected to be extremely rare:
2925 * only if we run lilo or swapon on a freshly made file
2926 * do we expect this to happen.
2928 * (bmap requires CAP_SYS_RAWIO so this does not
2929 * represent an unprivileged user DOS attack --- we'd be
2930 * in trouble if mortal users could trigger this path at
2933 * NB. EXT4_STATE_JDATA is not set on files other than
2934 * regular files. If somebody wants to bmap a directory
2935 * or symlink and gets confused because the buffer
2936 * hasn't yet been flushed to disk, they deserve
2937 * everything they get.
2940 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2941 journal
= EXT4_JOURNAL(inode
);
2942 jbd2_journal_lock_updates(journal
);
2943 err
= jbd2_journal_flush(journal
);
2944 jbd2_journal_unlock_updates(journal
);
2950 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2953 static int ext4_readpage(struct file
*file
, struct page
*page
)
2956 struct inode
*inode
= page
->mapping
->host
;
2958 trace_ext4_readpage(page
);
2960 if (ext4_has_inline_data(inode
))
2961 ret
= ext4_readpage_inline(inode
, page
);
2964 return mpage_readpage(page
, ext4_get_block
);
2970 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2971 struct list_head
*pages
, unsigned nr_pages
)
2973 struct inode
*inode
= mapping
->host
;
2975 /* If the file has inline data, no need to do readpages. */
2976 if (ext4_has_inline_data(inode
))
2979 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2982 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2984 trace_ext4_invalidatepage(page
, offset
);
2986 /* No journalling happens on data buffers when this function is used */
2987 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
2989 block_invalidatepage(page
, offset
);
2992 static int __ext4_journalled_invalidatepage(struct page
*page
,
2993 unsigned long offset
)
2995 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2997 trace_ext4_journalled_invalidatepage(page
, offset
);
3000 * If it's a full truncate we just forget about the pending dirtying
3003 ClearPageChecked(page
);
3005 return jbd2_journal_invalidatepage(journal
, page
, offset
);
3008 /* Wrapper for aops... */
3009 static void ext4_journalled_invalidatepage(struct page
*page
,
3010 unsigned long offset
)
3012 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
) < 0);
3015 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3017 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3019 trace_ext4_releasepage(page
);
3021 WARN_ON(PageChecked(page
));
3022 if (!page_has_buffers(page
))
3025 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3027 return try_to_free_buffers(page
);
3031 * ext4_get_block used when preparing for a DIO write or buffer write.
3032 * We allocate an uinitialized extent if blocks haven't been allocated.
3033 * The extent will be converted to initialized after the IO is complete.
3035 int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
3036 struct buffer_head
*bh_result
, int create
)
3038 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3039 inode
->i_ino
, create
);
3040 return _ext4_get_block(inode
, iblock
, bh_result
,
3041 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
3044 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
3045 struct buffer_head
*bh_result
, int create
)
3047 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3048 inode
->i_ino
, create
);
3049 return _ext4_get_block(inode
, iblock
, bh_result
,
3050 EXT4_GET_BLOCKS_NO_LOCK
);
3053 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3054 ssize_t size
, void *private, int ret
,
3057 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
3058 ext4_io_end_t
*io_end
= iocb
->private;
3060 /* if not async direct IO or dio with 0 bytes write, just return */
3061 if (!io_end
|| !size
)
3064 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3065 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3066 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3069 iocb
->private = NULL
;
3071 /* if not aio dio with unwritten extents, just free io and return */
3072 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
3073 ext4_free_io_end(io_end
);
3075 inode_dio_done(inode
);
3077 aio_complete(iocb
, ret
, 0);
3081 io_end
->offset
= offset
;
3082 io_end
->size
= size
;
3084 io_end
->iocb
= iocb
;
3085 io_end
->result
= ret
;
3088 ext4_add_complete_io(io_end
);
3092 * For ext4 extent files, ext4 will do direct-io write to holes,
3093 * preallocated extents, and those write extend the file, no need to
3094 * fall back to buffered IO.
3096 * For holes, we fallocate those blocks, mark them as uninitialized
3097 * If those blocks were preallocated, we mark sure they are split, but
3098 * still keep the range to write as uninitialized.
3100 * The unwritten extents will be converted to written when DIO is completed.
3101 * For async direct IO, since the IO may still pending when return, we
3102 * set up an end_io call back function, which will do the conversion
3103 * when async direct IO completed.
3105 * If the O_DIRECT write will extend the file then add this inode to the
3106 * orphan list. So recovery will truncate it back to the original size
3107 * if the machine crashes during the write.
3110 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3111 const struct iovec
*iov
, loff_t offset
,
3112 unsigned long nr_segs
)
3114 struct file
*file
= iocb
->ki_filp
;
3115 struct inode
*inode
= file
->f_mapping
->host
;
3117 size_t count
= iov_length(iov
, nr_segs
);
3119 get_block_t
*get_block_func
= NULL
;
3121 loff_t final_size
= offset
+ count
;
3123 /* Use the old path for reads and writes beyond i_size. */
3124 if (rw
!= WRITE
|| final_size
> inode
->i_size
)
3125 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3127 BUG_ON(iocb
->private == NULL
);
3129 /* If we do a overwrite dio, i_mutex locking can be released */
3130 overwrite
= *((int *)iocb
->private);
3133 atomic_inc(&inode
->i_dio_count
);
3134 down_read(&EXT4_I(inode
)->i_data_sem
);
3135 mutex_unlock(&inode
->i_mutex
);
3139 * We could direct write to holes and fallocate.
3141 * Allocated blocks to fill the hole are marked as
3142 * uninitialized to prevent parallel buffered read to expose
3143 * the stale data before DIO complete the data IO.
3145 * As to previously fallocated extents, ext4 get_block will
3146 * just simply mark the buffer mapped but still keep the
3147 * extents uninitialized.
3149 * For non AIO case, we will convert those unwritten extents
3150 * to written after return back from blockdev_direct_IO.
3152 * For async DIO, the conversion needs to be deferred when the
3153 * IO is completed. The ext4 end_io callback function will be
3154 * called to take care of the conversion work. Here for async
3155 * case, we allocate an io_end structure to hook to the iocb.
3157 iocb
->private = NULL
;
3158 ext4_inode_aio_set(inode
, NULL
);
3159 if (!is_sync_kiocb(iocb
)) {
3160 ext4_io_end_t
*io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
3165 io_end
->flag
|= EXT4_IO_END_DIRECT
;
3166 iocb
->private = io_end
;
3168 * we save the io structure for current async direct
3169 * IO, so that later ext4_map_blocks() could flag the
3170 * io structure whether there is a unwritten extents
3171 * needs to be converted when IO is completed.
3173 ext4_inode_aio_set(inode
, io_end
);
3177 get_block_func
= ext4_get_block_write_nolock
;
3179 get_block_func
= ext4_get_block_write
;
3180 dio_flags
= DIO_LOCKING
;
3182 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3183 inode
->i_sb
->s_bdev
, iov
,
3191 ext4_inode_aio_set(inode
, NULL
);
3193 * The io_end structure takes a reference to the inode, that
3194 * structure needs to be destroyed and the reference to the
3195 * inode need to be dropped, when IO is complete, even with 0
3196 * byte write, or failed.
3198 * In the successful AIO DIO case, the io_end structure will
3199 * be destroyed and the reference to the inode will be dropped
3200 * after the end_io call back function is called.
3202 * In the case there is 0 byte write, or error case, since VFS
3203 * direct IO won't invoke the end_io call back function, we
3204 * need to free the end_io structure here.
3206 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3207 ext4_free_io_end(iocb
->private);
3208 iocb
->private = NULL
;
3209 } else if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3210 EXT4_STATE_DIO_UNWRITTEN
)) {
3213 * for non AIO case, since the IO is already
3214 * completed, we could do the conversion right here
3216 err
= ext4_convert_unwritten_extents(inode
,
3220 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3224 /* take i_mutex locking again if we do a ovewrite dio */
3226 inode_dio_done(inode
);
3227 up_read(&EXT4_I(inode
)->i_data_sem
);
3228 mutex_lock(&inode
->i_mutex
);
3234 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3235 const struct iovec
*iov
, loff_t offset
,
3236 unsigned long nr_segs
)
3238 struct file
*file
= iocb
->ki_filp
;
3239 struct inode
*inode
= file
->f_mapping
->host
;
3243 * If we are doing data journalling we don't support O_DIRECT
3245 if (ext4_should_journal_data(inode
))
3248 /* Let buffer I/O handle the inline data case. */
3249 if (ext4_has_inline_data(inode
))
3252 trace_ext4_direct_IO_enter(inode
, offset
, iov_length(iov
, nr_segs
), rw
);
3253 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3254 ret
= ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3256 ret
= ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3257 trace_ext4_direct_IO_exit(inode
, offset
,
3258 iov_length(iov
, nr_segs
), rw
, ret
);
3263 * Pages can be marked dirty completely asynchronously from ext4's journalling
3264 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3265 * much here because ->set_page_dirty is called under VFS locks. The page is
3266 * not necessarily locked.
3268 * We cannot just dirty the page and leave attached buffers clean, because the
3269 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3270 * or jbddirty because all the journalling code will explode.
3272 * So what we do is to mark the page "pending dirty" and next time writepage
3273 * is called, propagate that into the buffers appropriately.
3275 static int ext4_journalled_set_page_dirty(struct page
*page
)
3277 SetPageChecked(page
);
3278 return __set_page_dirty_nobuffers(page
);
3281 static const struct address_space_operations ext4_ordered_aops
= {
3282 .readpage
= ext4_readpage
,
3283 .readpages
= ext4_readpages
,
3284 .writepage
= ext4_writepage
,
3285 .write_begin
= ext4_write_begin
,
3286 .write_end
= ext4_ordered_write_end
,
3288 .invalidatepage
= ext4_invalidatepage
,
3289 .releasepage
= ext4_releasepage
,
3290 .direct_IO
= ext4_direct_IO
,
3291 .migratepage
= buffer_migrate_page
,
3292 .is_partially_uptodate
= block_is_partially_uptodate
,
3293 .error_remove_page
= generic_error_remove_page
,
3296 static const struct address_space_operations ext4_writeback_aops
= {
3297 .readpage
= ext4_readpage
,
3298 .readpages
= ext4_readpages
,
3299 .writepage
= ext4_writepage
,
3300 .write_begin
= ext4_write_begin
,
3301 .write_end
= ext4_writeback_write_end
,
3303 .invalidatepage
= ext4_invalidatepage
,
3304 .releasepage
= ext4_releasepage
,
3305 .direct_IO
= ext4_direct_IO
,
3306 .migratepage
= buffer_migrate_page
,
3307 .is_partially_uptodate
= block_is_partially_uptodate
,
3308 .error_remove_page
= generic_error_remove_page
,
3311 static const struct address_space_operations ext4_journalled_aops
= {
3312 .readpage
= ext4_readpage
,
3313 .readpages
= ext4_readpages
,
3314 .writepage
= ext4_writepage
,
3315 .write_begin
= ext4_write_begin
,
3316 .write_end
= ext4_journalled_write_end
,
3317 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3319 .invalidatepage
= ext4_journalled_invalidatepage
,
3320 .releasepage
= ext4_releasepage
,
3321 .direct_IO
= ext4_direct_IO
,
3322 .is_partially_uptodate
= block_is_partially_uptodate
,
3323 .error_remove_page
= generic_error_remove_page
,
3326 static const struct address_space_operations ext4_da_aops
= {
3327 .readpage
= ext4_readpage
,
3328 .readpages
= ext4_readpages
,
3329 .writepage
= ext4_writepage
,
3330 .writepages
= ext4_da_writepages
,
3331 .write_begin
= ext4_da_write_begin
,
3332 .write_end
= ext4_da_write_end
,
3334 .invalidatepage
= ext4_da_invalidatepage
,
3335 .releasepage
= ext4_releasepage
,
3336 .direct_IO
= ext4_direct_IO
,
3337 .migratepage
= buffer_migrate_page
,
3338 .is_partially_uptodate
= block_is_partially_uptodate
,
3339 .error_remove_page
= generic_error_remove_page
,
3342 void ext4_set_aops(struct inode
*inode
)
3344 switch (ext4_inode_journal_mode(inode
)) {
3345 case EXT4_INODE_ORDERED_DATA_MODE
:
3346 if (test_opt(inode
->i_sb
, DELALLOC
))
3347 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3349 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3351 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3352 if (test_opt(inode
->i_sb
, DELALLOC
))
3353 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3355 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3357 case EXT4_INODE_JOURNAL_DATA_MODE
:
3358 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3367 * ext4_discard_partial_page_buffers()
3368 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3369 * This function finds and locks the page containing the offset
3370 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3371 * Calling functions that already have the page locked should call
3372 * ext4_discard_partial_page_buffers_no_lock directly.
3374 int ext4_discard_partial_page_buffers(handle_t
*handle
,
3375 struct address_space
*mapping
, loff_t from
,
3376 loff_t length
, int flags
)
3378 struct inode
*inode
= mapping
->host
;
3382 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3383 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3387 err
= ext4_discard_partial_page_buffers_no_lock(handle
, inode
, page
,
3388 from
, length
, flags
);
3391 page_cache_release(page
);
3396 * ext4_discard_partial_page_buffers_no_lock()
3397 * Zeros a page range of length 'length' starting from offset 'from'.
3398 * Buffer heads that correspond to the block aligned regions of the
3399 * zeroed range will be unmapped. Unblock aligned regions
3400 * will have the corresponding buffer head mapped if needed so that
3401 * that region of the page can be updated with the partial zero out.
3403 * This function assumes that the page has already been locked. The
3404 * The range to be discarded must be contained with in the given page.
3405 * If the specified range exceeds the end of the page it will be shortened
3406 * to the end of the page that corresponds to 'from'. This function is
3407 * appropriate for updating a page and it buffer heads to be unmapped and
3408 * zeroed for blocks that have been either released, or are going to be
3411 * handle: The journal handle
3412 * inode: The files inode
3413 * page: A locked page that contains the offset "from"
3414 * from: The starting byte offset (from the beginning of the file)
3415 * to begin discarding
3416 * len: The length of bytes to discard
3417 * flags: Optional flags that may be used:
3419 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3420 * Only zero the regions of the page whose buffer heads
3421 * have already been unmapped. This flag is appropriate
3422 * for updating the contents of a page whose blocks may
3423 * have already been released, and we only want to zero
3424 * out the regions that correspond to those released blocks.
3426 * Returns zero on success or negative on failure.
3428 static int ext4_discard_partial_page_buffers_no_lock(handle_t
*handle
,
3429 struct inode
*inode
, struct page
*page
, loff_t from
,
3430 loff_t length
, int flags
)
3432 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3433 unsigned int offset
= from
& (PAGE_CACHE_SIZE
-1);
3434 unsigned int blocksize
, max
, pos
;
3436 struct buffer_head
*bh
;
3439 blocksize
= inode
->i_sb
->s_blocksize
;
3440 max
= PAGE_CACHE_SIZE
- offset
;
3442 if (index
!= page
->index
)
3446 * correct length if it does not fall between
3447 * 'from' and the end of the page
3449 if (length
> max
|| length
< 0)
3452 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3454 if (!page_has_buffers(page
))
3455 create_empty_buffers(page
, blocksize
, 0);
3457 /* Find the buffer that contains "offset" */
3458 bh
= page_buffers(page
);
3460 while (offset
>= pos
) {
3461 bh
= bh
->b_this_page
;
3467 while (pos
< offset
+ length
) {
3468 unsigned int end_of_block
, range_to_discard
;
3472 /* The length of space left to zero and unmap */
3473 range_to_discard
= offset
+ length
- pos
;
3475 /* The length of space until the end of the block */
3476 end_of_block
= blocksize
- (pos
& (blocksize
-1));
3479 * Do not unmap or zero past end of block
3480 * for this buffer head
3482 if (range_to_discard
> end_of_block
)
3483 range_to_discard
= end_of_block
;
3487 * Skip this buffer head if we are only zeroing unampped
3488 * regions of the page
3490 if (flags
& EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
&&
3494 /* If the range is block aligned, unmap */
3495 if (range_to_discard
== blocksize
) {
3496 clear_buffer_dirty(bh
);
3498 clear_buffer_mapped(bh
);
3499 clear_buffer_req(bh
);
3500 clear_buffer_new(bh
);
3501 clear_buffer_delay(bh
);
3502 clear_buffer_unwritten(bh
);
3503 clear_buffer_uptodate(bh
);
3504 zero_user(page
, pos
, range_to_discard
);
3505 BUFFER_TRACE(bh
, "Buffer discarded");
3510 * If this block is not completely contained in the range
3511 * to be discarded, then it is not going to be released. Because
3512 * we need to keep this block, we need to make sure this part
3513 * of the page is uptodate before we modify it by writeing
3514 * partial zeros on it.
3516 if (!buffer_mapped(bh
)) {
3518 * Buffer head must be mapped before we can read
3521 BUFFER_TRACE(bh
, "unmapped");
3522 ext4_get_block(inode
, iblock
, bh
, 0);
3523 /* unmapped? It's a hole - nothing to do */
3524 if (!buffer_mapped(bh
)) {
3525 BUFFER_TRACE(bh
, "still unmapped");
3530 /* Ok, it's mapped. Make sure it's up-to-date */
3531 if (PageUptodate(page
))
3532 set_buffer_uptodate(bh
);
3534 if (!buffer_uptodate(bh
)) {
3536 ll_rw_block(READ
, 1, &bh
);
3538 /* Uhhuh. Read error. Complain and punt.*/
3539 if (!buffer_uptodate(bh
))
3543 if (ext4_should_journal_data(inode
)) {
3544 BUFFER_TRACE(bh
, "get write access");
3545 err
= ext4_journal_get_write_access(handle
, bh
);
3550 zero_user(page
, pos
, range_to_discard
);
3553 if (ext4_should_journal_data(inode
)) {
3554 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3556 mark_buffer_dirty(bh
);
3558 BUFFER_TRACE(bh
, "Partial buffer zeroed");
3560 bh
= bh
->b_this_page
;
3562 pos
+= range_to_discard
;
3568 int ext4_can_truncate(struct inode
*inode
)
3570 if (S_ISREG(inode
->i_mode
))
3572 if (S_ISDIR(inode
->i_mode
))
3574 if (S_ISLNK(inode
->i_mode
))
3575 return !ext4_inode_is_fast_symlink(inode
);
3580 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3581 * associated with the given offset and length
3583 * @inode: File inode
3584 * @offset: The offset where the hole will begin
3585 * @len: The length of the hole
3587 * Returns: 0 on success or negative on failure
3590 int ext4_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
3592 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
3593 if (!S_ISREG(inode
->i_mode
))
3596 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3597 return ext4_ind_punch_hole(file
, offset
, length
);
3599 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) {
3600 /* TODO: Add support for bigalloc file systems */
3604 trace_ext4_punch_hole(inode
, offset
, length
);
3606 return ext4_ext_punch_hole(file
, offset
, length
);
3612 * We block out ext4_get_block() block instantiations across the entire
3613 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3614 * simultaneously on behalf of the same inode.
3616 * As we work through the truncate and commit bits of it to the journal there
3617 * is one core, guiding principle: the file's tree must always be consistent on
3618 * disk. We must be able to restart the truncate after a crash.
3620 * The file's tree may be transiently inconsistent in memory (although it
3621 * probably isn't), but whenever we close off and commit a journal transaction,
3622 * the contents of (the filesystem + the journal) must be consistent and
3623 * restartable. It's pretty simple, really: bottom up, right to left (although
3624 * left-to-right works OK too).
3626 * Note that at recovery time, journal replay occurs *before* the restart of
3627 * truncate against the orphan inode list.
3629 * The committed inode has the new, desired i_size (which is the same as
3630 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3631 * that this inode's truncate did not complete and it will again call
3632 * ext4_truncate() to have another go. So there will be instantiated blocks
3633 * to the right of the truncation point in a crashed ext4 filesystem. But
3634 * that's fine - as long as they are linked from the inode, the post-crash
3635 * ext4_truncate() run will find them and release them.
3637 void ext4_truncate(struct inode
*inode
)
3639 trace_ext4_truncate_enter(inode
);
3641 if (!ext4_can_truncate(inode
))
3644 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3646 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3647 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3649 if (ext4_has_inline_data(inode
)) {
3652 ext4_inline_data_truncate(inode
, &has_inline
);
3657 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3658 ext4_ext_truncate(inode
);
3660 ext4_ind_truncate(inode
);
3662 trace_ext4_truncate_exit(inode
);
3666 * ext4_get_inode_loc returns with an extra refcount against the inode's
3667 * underlying buffer_head on success. If 'in_mem' is true, we have all
3668 * data in memory that is needed to recreate the on-disk version of this
3671 static int __ext4_get_inode_loc(struct inode
*inode
,
3672 struct ext4_iloc
*iloc
, int in_mem
)
3674 struct ext4_group_desc
*gdp
;
3675 struct buffer_head
*bh
;
3676 struct super_block
*sb
= inode
->i_sb
;
3678 int inodes_per_block
, inode_offset
;
3681 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3684 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3685 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3690 * Figure out the offset within the block group inode table
3692 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3693 inode_offset
= ((inode
->i_ino
- 1) %
3694 EXT4_INODES_PER_GROUP(sb
));
3695 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3696 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3698 bh
= sb_getblk(sb
, block
);
3701 if (!buffer_uptodate(bh
)) {
3705 * If the buffer has the write error flag, we have failed
3706 * to write out another inode in the same block. In this
3707 * case, we don't have to read the block because we may
3708 * read the old inode data successfully.
3710 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3711 set_buffer_uptodate(bh
);
3713 if (buffer_uptodate(bh
)) {
3714 /* someone brought it uptodate while we waited */
3720 * If we have all information of the inode in memory and this
3721 * is the only valid inode in the block, we need not read the
3725 struct buffer_head
*bitmap_bh
;
3728 start
= inode_offset
& ~(inodes_per_block
- 1);
3730 /* Is the inode bitmap in cache? */
3731 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3732 if (unlikely(!bitmap_bh
))
3736 * If the inode bitmap isn't in cache then the
3737 * optimisation may end up performing two reads instead
3738 * of one, so skip it.
3740 if (!buffer_uptodate(bitmap_bh
)) {
3744 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3745 if (i
== inode_offset
)
3747 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3751 if (i
== start
+ inodes_per_block
) {
3752 /* all other inodes are free, so skip I/O */
3753 memset(bh
->b_data
, 0, bh
->b_size
);
3754 set_buffer_uptodate(bh
);
3762 * If we need to do any I/O, try to pre-readahead extra
3763 * blocks from the inode table.
3765 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3766 ext4_fsblk_t b
, end
, table
;
3769 table
= ext4_inode_table(sb
, gdp
);
3770 /* s_inode_readahead_blks is always a power of 2 */
3771 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
3774 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
3775 num
= EXT4_INODES_PER_GROUP(sb
);
3776 if (ext4_has_group_desc_csum(sb
))
3777 num
-= ext4_itable_unused_count(sb
, gdp
);
3778 table
+= num
/ inodes_per_block
;
3782 sb_breadahead(sb
, b
++);
3786 * There are other valid inodes in the buffer, this inode
3787 * has in-inode xattrs, or we don't have this inode in memory.
3788 * Read the block from disk.
3790 trace_ext4_load_inode(inode
);
3792 bh
->b_end_io
= end_buffer_read_sync
;
3793 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
3795 if (!buffer_uptodate(bh
)) {
3796 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3797 "unable to read itable block");
3807 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3809 /* We have all inode data except xattrs in memory here. */
3810 return __ext4_get_inode_loc(inode
, iloc
,
3811 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
3814 void ext4_set_inode_flags(struct inode
*inode
)
3816 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3818 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3819 if (flags
& EXT4_SYNC_FL
)
3820 inode
->i_flags
|= S_SYNC
;
3821 if (flags
& EXT4_APPEND_FL
)
3822 inode
->i_flags
|= S_APPEND
;
3823 if (flags
& EXT4_IMMUTABLE_FL
)
3824 inode
->i_flags
|= S_IMMUTABLE
;
3825 if (flags
& EXT4_NOATIME_FL
)
3826 inode
->i_flags
|= S_NOATIME
;
3827 if (flags
& EXT4_DIRSYNC_FL
)
3828 inode
->i_flags
|= S_DIRSYNC
;
3831 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3832 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3834 unsigned int vfs_fl
;
3835 unsigned long old_fl
, new_fl
;
3838 vfs_fl
= ei
->vfs_inode
.i_flags
;
3839 old_fl
= ei
->i_flags
;
3840 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3841 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
3843 if (vfs_fl
& S_SYNC
)
3844 new_fl
|= EXT4_SYNC_FL
;
3845 if (vfs_fl
& S_APPEND
)
3846 new_fl
|= EXT4_APPEND_FL
;
3847 if (vfs_fl
& S_IMMUTABLE
)
3848 new_fl
|= EXT4_IMMUTABLE_FL
;
3849 if (vfs_fl
& S_NOATIME
)
3850 new_fl
|= EXT4_NOATIME_FL
;
3851 if (vfs_fl
& S_DIRSYNC
)
3852 new_fl
|= EXT4_DIRSYNC_FL
;
3853 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
3856 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3857 struct ext4_inode_info
*ei
)
3860 struct inode
*inode
= &(ei
->vfs_inode
);
3861 struct super_block
*sb
= inode
->i_sb
;
3863 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3864 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3865 /* we are using combined 48 bit field */
3866 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
3867 le32_to_cpu(raw_inode
->i_blocks_lo
);
3868 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
3869 /* i_blocks represent file system block size */
3870 return i_blocks
<< (inode
->i_blkbits
- 9);
3875 return le32_to_cpu(raw_inode
->i_blocks_lo
);
3879 static inline void ext4_iget_extra_inode(struct inode
*inode
,
3880 struct ext4_inode
*raw_inode
,
3881 struct ext4_inode_info
*ei
)
3883 __le32
*magic
= (void *)raw_inode
+
3884 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
3885 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
3886 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
3887 ext4_find_inline_data_nolock(inode
);
3889 EXT4_I(inode
)->i_inline_off
= 0;
3892 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
3894 struct ext4_iloc iloc
;
3895 struct ext4_inode
*raw_inode
;
3896 struct ext4_inode_info
*ei
;
3897 struct inode
*inode
;
3898 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
3904 inode
= iget_locked(sb
, ino
);
3906 return ERR_PTR(-ENOMEM
);
3907 if (!(inode
->i_state
& I_NEW
))
3913 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
3916 raw_inode
= ext4_raw_inode(&iloc
);
3918 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3919 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
3920 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
3921 EXT4_INODE_SIZE(inode
->i_sb
)) {
3922 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
3923 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
3924 EXT4_INODE_SIZE(inode
->i_sb
));
3929 ei
->i_extra_isize
= 0;
3931 /* Precompute checksum seed for inode metadata */
3932 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3933 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
)) {
3934 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3936 __le32 inum
= cpu_to_le32(inode
->i_ino
);
3937 __le32 gen
= raw_inode
->i_generation
;
3938 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
3940 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
3944 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
3945 EXT4_ERROR_INODE(inode
, "checksum invalid");
3950 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
3951 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
3952 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
3953 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
3954 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
3955 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
3957 i_uid_write(inode
, i_uid
);
3958 i_gid_write(inode
, i_gid
);
3959 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
3961 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
3962 ei
->i_inline_off
= 0;
3963 ei
->i_dir_start_lookup
= 0;
3964 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
3965 /* We now have enough fields to check if the inode was active or not.
3966 * This is needed because nfsd might try to access dead inodes
3967 * the test is that same one that e2fsck uses
3968 * NeilBrown 1999oct15
3970 if (inode
->i_nlink
== 0) {
3971 if (inode
->i_mode
== 0 ||
3972 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
3973 /* this inode is deleted */
3977 /* The only unlinked inodes we let through here have
3978 * valid i_mode and are being read by the orphan
3979 * recovery code: that's fine, we're about to complete
3980 * the process of deleting those. */
3982 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
3983 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
3984 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
3985 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
3987 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
3988 inode
->i_size
= ext4_isize(raw_inode
);
3989 ei
->i_disksize
= inode
->i_size
;
3991 ei
->i_reserved_quota
= 0;
3993 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
3994 ei
->i_block_group
= iloc
.block_group
;
3995 ei
->i_last_alloc_group
= ~0;
3997 * NOTE! The in-memory inode i_data array is in little-endian order
3998 * even on big-endian machines: we do NOT byteswap the block numbers!
4000 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4001 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4002 INIT_LIST_HEAD(&ei
->i_orphan
);
4005 * Set transaction id's of transactions that have to be committed
4006 * to finish f[data]sync. We set them to currently running transaction
4007 * as we cannot be sure that the inode or some of its metadata isn't
4008 * part of the transaction - the inode could have been reclaimed and
4009 * now it is reread from disk.
4012 transaction_t
*transaction
;
4015 read_lock(&journal
->j_state_lock
);
4016 if (journal
->j_running_transaction
)
4017 transaction
= journal
->j_running_transaction
;
4019 transaction
= journal
->j_committing_transaction
;
4021 tid
= transaction
->t_tid
;
4023 tid
= journal
->j_commit_sequence
;
4024 read_unlock(&journal
->j_state_lock
);
4025 ei
->i_sync_tid
= tid
;
4026 ei
->i_datasync_tid
= tid
;
4029 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4030 if (ei
->i_extra_isize
== 0) {
4031 /* The extra space is currently unused. Use it. */
4032 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4033 EXT4_GOOD_OLD_INODE_SIZE
;
4035 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4039 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4040 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4041 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4042 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4044 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4045 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4046 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4048 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4052 if (ei
->i_file_acl
&&
4053 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4054 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4058 } else if (!ext4_has_inline_data(inode
)) {
4059 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4060 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4061 (S_ISLNK(inode
->i_mode
) &&
4062 !ext4_inode_is_fast_symlink(inode
))))
4063 /* Validate extent which is part of inode */
4064 ret
= ext4_ext_check_inode(inode
);
4065 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4066 (S_ISLNK(inode
->i_mode
) &&
4067 !ext4_inode_is_fast_symlink(inode
))) {
4068 /* Validate block references which are part of inode */
4069 ret
= ext4_ind_check_inode(inode
);
4075 if (S_ISREG(inode
->i_mode
)) {
4076 inode
->i_op
= &ext4_file_inode_operations
;
4077 inode
->i_fop
= &ext4_file_operations
;
4078 ext4_set_aops(inode
);
4079 } else if (S_ISDIR(inode
->i_mode
)) {
4080 inode
->i_op
= &ext4_dir_inode_operations
;
4081 inode
->i_fop
= &ext4_dir_operations
;
4082 } else if (S_ISLNK(inode
->i_mode
)) {
4083 if (ext4_inode_is_fast_symlink(inode
)) {
4084 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4085 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4086 sizeof(ei
->i_data
) - 1);
4088 inode
->i_op
= &ext4_symlink_inode_operations
;
4089 ext4_set_aops(inode
);
4091 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4092 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4093 inode
->i_op
= &ext4_special_inode_operations
;
4094 if (raw_inode
->i_block
[0])
4095 init_special_inode(inode
, inode
->i_mode
,
4096 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4098 init_special_inode(inode
, inode
->i_mode
,
4099 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4102 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4106 ext4_set_inode_flags(inode
);
4107 unlock_new_inode(inode
);
4113 return ERR_PTR(ret
);
4116 static int ext4_inode_blocks_set(handle_t
*handle
,
4117 struct ext4_inode
*raw_inode
,
4118 struct ext4_inode_info
*ei
)
4120 struct inode
*inode
= &(ei
->vfs_inode
);
4121 u64 i_blocks
= inode
->i_blocks
;
4122 struct super_block
*sb
= inode
->i_sb
;
4124 if (i_blocks
<= ~0U) {
4126 * i_blocks can be represented in a 32 bit variable
4127 * as multiple of 512 bytes
4129 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4130 raw_inode
->i_blocks_high
= 0;
4131 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4134 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4137 if (i_blocks
<= 0xffffffffffffULL
) {
4139 * i_blocks can be represented in a 48 bit variable
4140 * as multiple of 512 bytes
4142 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4143 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4144 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4146 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4147 /* i_block is stored in file system block size */
4148 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4149 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4150 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4156 * Post the struct inode info into an on-disk inode location in the
4157 * buffer-cache. This gobbles the caller's reference to the
4158 * buffer_head in the inode location struct.
4160 * The caller must have write access to iloc->bh.
4162 static int ext4_do_update_inode(handle_t
*handle
,
4163 struct inode
*inode
,
4164 struct ext4_iloc
*iloc
)
4166 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4167 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4168 struct buffer_head
*bh
= iloc
->bh
;
4169 int err
= 0, rc
, block
;
4170 int need_datasync
= 0;
4174 /* For fields not not tracking in the in-memory inode,
4175 * initialise them to zero for new inodes. */
4176 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4177 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4179 ext4_get_inode_flags(ei
);
4180 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4181 i_uid
= i_uid_read(inode
);
4182 i_gid
= i_gid_read(inode
);
4183 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4184 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4185 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4187 * Fix up interoperability with old kernels. Otherwise, old inodes get
4188 * re-used with the upper 16 bits of the uid/gid intact
4191 raw_inode
->i_uid_high
=
4192 cpu_to_le16(high_16_bits(i_uid
));
4193 raw_inode
->i_gid_high
=
4194 cpu_to_le16(high_16_bits(i_gid
));
4196 raw_inode
->i_uid_high
= 0;
4197 raw_inode
->i_gid_high
= 0;
4200 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4201 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4202 raw_inode
->i_uid_high
= 0;
4203 raw_inode
->i_gid_high
= 0;
4205 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4207 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4208 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4209 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4210 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4212 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4214 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4215 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4216 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4217 cpu_to_le32(EXT4_OS_HURD
))
4218 raw_inode
->i_file_acl_high
=
4219 cpu_to_le16(ei
->i_file_acl
>> 32);
4220 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4221 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4222 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4225 if (ei
->i_disksize
> 0x7fffffffULL
) {
4226 struct super_block
*sb
= inode
->i_sb
;
4227 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4228 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4229 EXT4_SB(sb
)->s_es
->s_rev_level
==
4230 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4231 /* If this is the first large file
4232 * created, add a flag to the superblock.
4234 err
= ext4_journal_get_write_access(handle
,
4235 EXT4_SB(sb
)->s_sbh
);
4238 ext4_update_dynamic_rev(sb
);
4239 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4240 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4241 ext4_handle_sync(handle
);
4242 err
= ext4_handle_dirty_super(handle
, sb
);
4245 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4246 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4247 if (old_valid_dev(inode
->i_rdev
)) {
4248 raw_inode
->i_block
[0] =
4249 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4250 raw_inode
->i_block
[1] = 0;
4252 raw_inode
->i_block
[0] = 0;
4253 raw_inode
->i_block
[1] =
4254 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4255 raw_inode
->i_block
[2] = 0;
4257 } else if (!ext4_has_inline_data(inode
)) {
4258 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4259 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4262 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4263 if (ei
->i_extra_isize
) {
4264 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4265 raw_inode
->i_version_hi
=
4266 cpu_to_le32(inode
->i_version
>> 32);
4267 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4270 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4272 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4273 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4276 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4278 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4281 ext4_std_error(inode
->i_sb
, err
);
4286 * ext4_write_inode()
4288 * We are called from a few places:
4290 * - Within generic_file_write() for O_SYNC files.
4291 * Here, there will be no transaction running. We wait for any running
4292 * transaction to commit.
4294 * - Within sys_sync(), kupdate and such.
4295 * We wait on commit, if tol to.
4297 * - Within prune_icache() (PF_MEMALLOC == true)
4298 * Here we simply return. We can't afford to block kswapd on the
4301 * In all cases it is actually safe for us to return without doing anything,
4302 * because the inode has been copied into a raw inode buffer in
4303 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4306 * Note that we are absolutely dependent upon all inode dirtiers doing the
4307 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4308 * which we are interested.
4310 * It would be a bug for them to not do this. The code:
4312 * mark_inode_dirty(inode)
4314 * inode->i_size = expr;
4316 * is in error because a kswapd-driven write_inode() could occur while
4317 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4318 * will no longer be on the superblock's dirty inode list.
4320 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4324 if (current
->flags
& PF_MEMALLOC
)
4327 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4328 if (ext4_journal_current_handle()) {
4329 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4334 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
4337 err
= ext4_force_commit(inode
->i_sb
);
4339 struct ext4_iloc iloc
;
4341 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4344 if (wbc
->sync_mode
== WB_SYNC_ALL
)
4345 sync_dirty_buffer(iloc
.bh
);
4346 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4347 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4348 "IO error syncing inode");
4357 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4358 * buffers that are attached to a page stradding i_size and are undergoing
4359 * commit. In that case we have to wait for commit to finish and try again.
4361 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4365 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4366 tid_t commit_tid
= 0;
4369 offset
= inode
->i_size
& (PAGE_CACHE_SIZE
- 1);
4371 * All buffers in the last page remain valid? Then there's nothing to
4372 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4375 if (offset
> PAGE_CACHE_SIZE
- (1 << inode
->i_blkbits
))
4378 page
= find_lock_page(inode
->i_mapping
,
4379 inode
->i_size
>> PAGE_CACHE_SHIFT
);
4382 ret
= __ext4_journalled_invalidatepage(page
, offset
);
4384 page_cache_release(page
);
4388 read_lock(&journal
->j_state_lock
);
4389 if (journal
->j_committing_transaction
)
4390 commit_tid
= journal
->j_committing_transaction
->t_tid
;
4391 read_unlock(&journal
->j_state_lock
);
4393 jbd2_log_wait_commit(journal
, commit_tid
);
4400 * Called from notify_change.
4402 * We want to trap VFS attempts to truncate the file as soon as
4403 * possible. In particular, we want to make sure that when the VFS
4404 * shrinks i_size, we put the inode on the orphan list and modify
4405 * i_disksize immediately, so that during the subsequent flushing of
4406 * dirty pages and freeing of disk blocks, we can guarantee that any
4407 * commit will leave the blocks being flushed in an unused state on
4408 * disk. (On recovery, the inode will get truncated and the blocks will
4409 * be freed, so we have a strong guarantee that no future commit will
4410 * leave these blocks visible to the user.)
4412 * Another thing we have to assure is that if we are in ordered mode
4413 * and inode is still attached to the committing transaction, we must
4414 * we start writeout of all the dirty pages which are being truncated.
4415 * This way we are sure that all the data written in the previous
4416 * transaction are already on disk (truncate waits for pages under
4419 * Called with inode->i_mutex down.
4421 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4423 struct inode
*inode
= dentry
->d_inode
;
4426 const unsigned int ia_valid
= attr
->ia_valid
;
4428 error
= inode_change_ok(inode
, attr
);
4432 if (is_quota_modification(inode
, attr
))
4433 dquot_initialize(inode
);
4434 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4435 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4438 /* (user+group)*(old+new) structure, inode write (sb,
4439 * inode block, ? - but truncate inode update has it) */
4440 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
4441 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
4442 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
4443 if (IS_ERR(handle
)) {
4444 error
= PTR_ERR(handle
);
4447 error
= dquot_transfer(inode
, attr
);
4449 ext4_journal_stop(handle
);
4452 /* Update corresponding info in inode so that everything is in
4453 * one transaction */
4454 if (attr
->ia_valid
& ATTR_UID
)
4455 inode
->i_uid
= attr
->ia_uid
;
4456 if (attr
->ia_valid
& ATTR_GID
)
4457 inode
->i_gid
= attr
->ia_gid
;
4458 error
= ext4_mark_inode_dirty(handle
, inode
);
4459 ext4_journal_stop(handle
);
4462 if (attr
->ia_valid
& ATTR_SIZE
) {
4464 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4465 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4467 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4472 if (S_ISREG(inode
->i_mode
) &&
4473 attr
->ia_valid
& ATTR_SIZE
&&
4474 (attr
->ia_size
< inode
->i_size
)) {
4477 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
4478 if (IS_ERR(handle
)) {
4479 error
= PTR_ERR(handle
);
4482 if (ext4_handle_valid(handle
)) {
4483 error
= ext4_orphan_add(handle
, inode
);
4486 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4487 rc
= ext4_mark_inode_dirty(handle
, inode
);
4490 ext4_journal_stop(handle
);
4492 if (ext4_should_order_data(inode
)) {
4493 error
= ext4_begin_ordered_truncate(inode
,
4496 /* Do as much error cleanup as possible */
4497 handle
= ext4_journal_start(inode
,
4499 if (IS_ERR(handle
)) {
4500 ext4_orphan_del(NULL
, inode
);
4503 ext4_orphan_del(handle
, inode
);
4505 ext4_journal_stop(handle
);
4511 if (attr
->ia_valid
& ATTR_SIZE
) {
4512 if (attr
->ia_size
!= inode
->i_size
) {
4513 loff_t oldsize
= inode
->i_size
;
4515 i_size_write(inode
, attr
->ia_size
);
4517 * Blocks are going to be removed from the inode. Wait
4518 * for dio in flight. Temporarily disable
4519 * dioread_nolock to prevent livelock.
4522 if (!ext4_should_journal_data(inode
)) {
4523 ext4_inode_block_unlocked_dio(inode
);
4524 inode_dio_wait(inode
);
4525 ext4_inode_resume_unlocked_dio(inode
);
4527 ext4_wait_for_tail_page_commit(inode
);
4530 * Truncate pagecache after we've waited for commit
4531 * in data=journal mode to make pages freeable.
4533 truncate_pagecache(inode
, oldsize
, inode
->i_size
);
4535 ext4_truncate(inode
);
4539 setattr_copy(inode
, attr
);
4540 mark_inode_dirty(inode
);
4544 * If the call to ext4_truncate failed to get a transaction handle at
4545 * all, we need to clean up the in-core orphan list manually.
4547 if (orphan
&& inode
->i_nlink
)
4548 ext4_orphan_del(NULL
, inode
);
4550 if (!rc
&& (ia_valid
& ATTR_MODE
))
4551 rc
= ext4_acl_chmod(inode
);
4554 ext4_std_error(inode
->i_sb
, error
);
4560 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4563 struct inode
*inode
;
4564 unsigned long delalloc_blocks
;
4566 inode
= dentry
->d_inode
;
4567 generic_fillattr(inode
, stat
);
4570 * We can't update i_blocks if the block allocation is delayed
4571 * otherwise in the case of system crash before the real block
4572 * allocation is done, we will have i_blocks inconsistent with
4573 * on-disk file blocks.
4574 * We always keep i_blocks updated together with real
4575 * allocation. But to not confuse with user, stat
4576 * will return the blocks that include the delayed allocation
4577 * blocks for this file.
4579 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
4580 EXT4_I(inode
)->i_reserved_data_blocks
);
4582 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4586 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4588 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4589 return ext4_ind_trans_blocks(inode
, nrblocks
, chunk
);
4590 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4594 * Account for index blocks, block groups bitmaps and block group
4595 * descriptor blocks if modify datablocks and index blocks
4596 * worse case, the indexs blocks spread over different block groups
4598 * If datablocks are discontiguous, they are possible to spread over
4599 * different block groups too. If they are contiguous, with flexbg,
4600 * they could still across block group boundary.
4602 * Also account for superblock, inode, quota and xattr blocks
4604 static int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4606 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4612 * How many index blocks need to touch to modify nrblocks?
4613 * The "Chunk" flag indicating whether the nrblocks is
4614 * physically contiguous on disk
4616 * For Direct IO and fallocate, they calls get_block to allocate
4617 * one single extent at a time, so they could set the "Chunk" flag
4619 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4624 * Now let's see how many group bitmaps and group descriptors need
4634 if (groups
> ngroups
)
4636 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4637 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4639 /* bitmaps and block group descriptor blocks */
4640 ret
+= groups
+ gdpblocks
;
4642 /* Blocks for super block, inode, quota and xattr blocks */
4643 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4649 * Calculate the total number of credits to reserve to fit
4650 * the modification of a single pages into a single transaction,
4651 * which may include multiple chunks of block allocations.
4653 * This could be called via ext4_write_begin()
4655 * We need to consider the worse case, when
4656 * one new block per extent.
4658 int ext4_writepage_trans_blocks(struct inode
*inode
)
4660 int bpp
= ext4_journal_blocks_per_page(inode
);
4663 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4665 /* Account for data blocks for journalled mode */
4666 if (ext4_should_journal_data(inode
))
4672 * Calculate the journal credits for a chunk of data modification.
4674 * This is called from DIO, fallocate or whoever calling
4675 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4677 * journal buffers for data blocks are not included here, as DIO
4678 * and fallocate do no need to journal data buffers.
4680 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4682 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4686 * The caller must have previously called ext4_reserve_inode_write().
4687 * Give this, we know that the caller already has write access to iloc->bh.
4689 int ext4_mark_iloc_dirty(handle_t
*handle
,
4690 struct inode
*inode
, struct ext4_iloc
*iloc
)
4694 if (IS_I_VERSION(inode
))
4695 inode_inc_iversion(inode
);
4697 /* the do_update_inode consumes one bh->b_count */
4700 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4701 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4707 * On success, We end up with an outstanding reference count against
4708 * iloc->bh. This _must_ be cleaned up later.
4712 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4713 struct ext4_iloc
*iloc
)
4717 err
= ext4_get_inode_loc(inode
, iloc
);
4719 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4720 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4726 ext4_std_error(inode
->i_sb
, err
);
4731 * Expand an inode by new_extra_isize bytes.
4732 * Returns 0 on success or negative error number on failure.
4734 static int ext4_expand_extra_isize(struct inode
*inode
,
4735 unsigned int new_extra_isize
,
4736 struct ext4_iloc iloc
,
4739 struct ext4_inode
*raw_inode
;
4740 struct ext4_xattr_ibody_header
*header
;
4742 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4745 raw_inode
= ext4_raw_inode(&iloc
);
4747 header
= IHDR(inode
, raw_inode
);
4749 /* No extended attributes present */
4750 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
4751 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4752 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4754 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4758 /* try to expand with EAs present */
4759 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4764 * What we do here is to mark the in-core inode as clean with respect to inode
4765 * dirtiness (it may still be data-dirty).
4766 * This means that the in-core inode may be reaped by prune_icache
4767 * without having to perform any I/O. This is a very good thing,
4768 * because *any* task may call prune_icache - even ones which
4769 * have a transaction open against a different journal.
4771 * Is this cheating? Not really. Sure, we haven't written the
4772 * inode out, but prune_icache isn't a user-visible syncing function.
4773 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4774 * we start and wait on commits.
4776 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4778 struct ext4_iloc iloc
;
4779 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4780 static unsigned int mnt_count
;
4784 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
4785 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4786 if (ext4_handle_valid(handle
) &&
4787 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4788 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
4790 * We need extra buffer credits since we may write into EA block
4791 * with this same handle. If journal_extend fails, then it will
4792 * only result in a minor loss of functionality for that inode.
4793 * If this is felt to be critical, then e2fsck should be run to
4794 * force a large enough s_min_extra_isize.
4796 if ((jbd2_journal_extend(handle
,
4797 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4798 ret
= ext4_expand_extra_isize(inode
,
4799 sbi
->s_want_extra_isize
,
4802 ext4_set_inode_state(inode
,
4803 EXT4_STATE_NO_EXPAND
);
4805 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4806 ext4_warning(inode
->i_sb
,
4807 "Unable to expand inode %lu. Delete"
4808 " some EAs or run e2fsck.",
4811 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4817 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4822 * ext4_dirty_inode() is called from __mark_inode_dirty()
4824 * We're really interested in the case where a file is being extended.
4825 * i_size has been changed by generic_commit_write() and we thus need
4826 * to include the updated inode in the current transaction.
4828 * Also, dquot_alloc_block() will always dirty the inode when blocks
4829 * are allocated to the file.
4831 * If the inode is marked synchronous, we don't honour that here - doing
4832 * so would cause a commit on atime updates, which we don't bother doing.
4833 * We handle synchronous inodes at the highest possible level.
4835 void ext4_dirty_inode(struct inode
*inode
, int flags
)
4839 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
4843 ext4_mark_inode_dirty(handle
, inode
);
4845 ext4_journal_stop(handle
);
4852 * Bind an inode's backing buffer_head into this transaction, to prevent
4853 * it from being flushed to disk early. Unlike
4854 * ext4_reserve_inode_write, this leaves behind no bh reference and
4855 * returns no iloc structure, so the caller needs to repeat the iloc
4856 * lookup to mark the inode dirty later.
4858 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4860 struct ext4_iloc iloc
;
4864 err
= ext4_get_inode_loc(inode
, &iloc
);
4866 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4867 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4869 err
= ext4_handle_dirty_metadata(handle
,
4875 ext4_std_error(inode
->i_sb
, err
);
4880 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4887 * We have to be very careful here: changing a data block's
4888 * journaling status dynamically is dangerous. If we write a
4889 * data block to the journal, change the status and then delete
4890 * that block, we risk forgetting to revoke the old log record
4891 * from the journal and so a subsequent replay can corrupt data.
4892 * So, first we make sure that the journal is empty and that
4893 * nobody is changing anything.
4896 journal
= EXT4_JOURNAL(inode
);
4899 if (is_journal_aborted(journal
))
4901 /* We have to allocate physical blocks for delalloc blocks
4902 * before flushing journal. otherwise delalloc blocks can not
4903 * be allocated any more. even more truncate on delalloc blocks
4904 * could trigger BUG by flushing delalloc blocks in journal.
4905 * There is no delalloc block in non-journal data mode.
4907 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
4908 err
= ext4_alloc_da_blocks(inode
);
4913 /* Wait for all existing dio workers */
4914 ext4_inode_block_unlocked_dio(inode
);
4915 inode_dio_wait(inode
);
4917 jbd2_journal_lock_updates(journal
);
4920 * OK, there are no updates running now, and all cached data is
4921 * synced to disk. We are now in a completely consistent state
4922 * which doesn't have anything in the journal, and we know that
4923 * no filesystem updates are running, so it is safe to modify
4924 * the inode's in-core data-journaling state flag now.
4928 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4930 jbd2_journal_flush(journal
);
4931 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4933 ext4_set_aops(inode
);
4935 jbd2_journal_unlock_updates(journal
);
4936 ext4_inode_resume_unlocked_dio(inode
);
4938 /* Finally we can mark the inode as dirty. */
4940 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
4942 return PTR_ERR(handle
);
4944 err
= ext4_mark_inode_dirty(handle
, inode
);
4945 ext4_handle_sync(handle
);
4946 ext4_journal_stop(handle
);
4947 ext4_std_error(inode
->i_sb
, err
);
4952 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4954 return !buffer_mapped(bh
);
4957 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
4959 struct page
*page
= vmf
->page
;
4963 struct file
*file
= vma
->vm_file
;
4964 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4965 struct address_space
*mapping
= inode
->i_mapping
;
4967 get_block_t
*get_block
;
4970 sb_start_pagefault(inode
->i_sb
);
4971 file_update_time(vma
->vm_file
);
4972 /* Delalloc case is easy... */
4973 if (test_opt(inode
->i_sb
, DELALLOC
) &&
4974 !ext4_should_journal_data(inode
) &&
4975 !ext4_nonda_switch(inode
->i_sb
)) {
4977 ret
= __block_page_mkwrite(vma
, vmf
,
4978 ext4_da_get_block_prep
);
4979 } while (ret
== -ENOSPC
&&
4980 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
4985 size
= i_size_read(inode
);
4986 /* Page got truncated from under us? */
4987 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
4989 ret
= VM_FAULT_NOPAGE
;
4993 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4994 len
= size
& ~PAGE_CACHE_MASK
;
4996 len
= PAGE_CACHE_SIZE
;
4998 * Return if we have all the buffers mapped. This avoids the need to do
4999 * journal_start/journal_stop which can block and take a long time
5001 if (page_has_buffers(page
)) {
5002 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5004 ext4_bh_unmapped
)) {
5005 /* Wait so that we don't change page under IO */
5006 wait_on_page_writeback(page
);
5007 ret
= VM_FAULT_LOCKED
;
5012 /* OK, we need to fill the hole... */
5013 if (ext4_should_dioread_nolock(inode
))
5014 get_block
= ext4_get_block_write
;
5016 get_block
= ext4_get_block
;
5018 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5019 ext4_writepage_trans_blocks(inode
));
5020 if (IS_ERR(handle
)) {
5021 ret
= VM_FAULT_SIGBUS
;
5024 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
5025 if (!ret
&& ext4_should_journal_data(inode
)) {
5026 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5027 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
5029 ret
= VM_FAULT_SIGBUS
;
5030 ext4_journal_stop(handle
);
5033 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5035 ext4_journal_stop(handle
);
5036 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5039 ret
= block_page_mkwrite_return(ret
);
5041 sb_end_pagefault(inode
->i_sb
);