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
);
486 * The ext4_map_blocks() function tries to look up the requested blocks,
487 * and returns if the blocks are already mapped.
489 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
490 * and store the allocated blocks in the result buffer head and mark it
493 * If file type is extents based, it will call ext4_ext_map_blocks(),
494 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
497 * On success, it returns the number of blocks being mapped or allocate.
498 * if create==0 and the blocks are pre-allocated and uninitialized block,
499 * the result buffer head is unmapped. If the create ==1, it will make sure
500 * the buffer head is mapped.
502 * It returns 0 if plain look up failed (blocks have not been allocated), in
503 * that case, buffer head is unmapped
505 * It returns the error in case of allocation failure.
507 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
508 struct ext4_map_blocks
*map
, int flags
)
510 struct extent_status es
;
514 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
515 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
516 (unsigned long) map
->m_lblk
);
518 /* Lookup extent status tree firstly */
519 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
520 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
521 map
->m_pblk
= ext4_es_pblock(&es
) +
522 map
->m_lblk
- es
.es_lblk
;
523 map
->m_flags
|= ext4_es_is_written(&es
) ?
524 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
525 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
526 if (retval
> map
->m_len
)
529 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
538 * Try to see if we can get the block without requesting a new
541 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
542 down_read((&EXT4_I(inode
)->i_data_sem
));
543 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
544 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
545 EXT4_GET_BLOCKS_KEEP_SIZE
);
547 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
548 EXT4_GET_BLOCKS_KEEP_SIZE
);
552 unsigned long long status
;
554 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
555 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
556 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
557 ext4_find_delalloc_range(inode
, map
->m_lblk
,
558 map
->m_lblk
+ map
->m_len
- 1))
559 status
|= EXTENT_STATUS_DELAYED
;
560 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
561 map
->m_len
, map
->m_pblk
, status
);
565 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
566 up_read((&EXT4_I(inode
)->i_data_sem
));
569 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
570 int ret
= check_block_validity(inode
, map
);
575 /* If it is only a block(s) look up */
576 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
580 * Returns if the blocks have already allocated
582 * Note that if blocks have been preallocated
583 * ext4_ext_get_block() returns the create = 0
584 * with buffer head unmapped.
586 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
590 * Here we clear m_flags because after allocating an new extent,
591 * it will be set again.
593 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
596 * New blocks allocate and/or writing to uninitialized extent
597 * will possibly result in updating i_data, so we take
598 * the write lock of i_data_sem, and call get_blocks()
599 * with create == 1 flag.
601 down_write((&EXT4_I(inode
)->i_data_sem
));
604 * if the caller is from delayed allocation writeout path
605 * we have already reserved fs blocks for allocation
606 * let the underlying get_block() function know to
607 * avoid double accounting
609 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
610 ext4_set_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
612 * We need to check for EXT4 here because migrate
613 * could have changed the inode type in between
615 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
616 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
618 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
620 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
622 * We allocated new blocks which will result in
623 * i_data's format changing. Force the migrate
624 * to fail by clearing migrate flags
626 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
630 * Update reserved blocks/metadata blocks after successful
631 * block allocation which had been deferred till now. We don't
632 * support fallocate for non extent files. So we can update
633 * reserve space here.
636 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
637 ext4_da_update_reserve_space(inode
, retval
, 1);
639 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
640 ext4_clear_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
644 unsigned long long status
;
646 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
647 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
648 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
649 ext4_find_delalloc_range(inode
, map
->m_lblk
,
650 map
->m_lblk
+ map
->m_len
- 1))
651 status
|= EXTENT_STATUS_DELAYED
;
652 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
653 map
->m_pblk
, status
);
658 up_write((&EXT4_I(inode
)->i_data_sem
));
659 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
660 int ret
= check_block_validity(inode
, map
);
667 /* Maximum number of blocks we map for direct IO at once. */
668 #define DIO_MAX_BLOCKS 4096
670 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
671 struct buffer_head
*bh
, int flags
)
673 handle_t
*handle
= ext4_journal_current_handle();
674 struct ext4_map_blocks map
;
675 int ret
= 0, started
= 0;
678 if (ext4_has_inline_data(inode
))
682 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
684 if (flags
&& !(flags
& EXT4_GET_BLOCKS_NO_LOCK
) && !handle
) {
685 /* Direct IO write... */
686 if (map
.m_len
> DIO_MAX_BLOCKS
)
687 map
.m_len
= DIO_MAX_BLOCKS
;
688 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
689 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
691 if (IS_ERR(handle
)) {
692 ret
= PTR_ERR(handle
);
698 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
700 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
701 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
702 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
706 ext4_journal_stop(handle
);
710 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
711 struct buffer_head
*bh
, int create
)
713 return _ext4_get_block(inode
, iblock
, bh
,
714 create
? EXT4_GET_BLOCKS_CREATE
: 0);
718 * `handle' can be NULL if create is zero
720 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
721 ext4_lblk_t block
, int create
, int *errp
)
723 struct ext4_map_blocks map
;
724 struct buffer_head
*bh
;
727 J_ASSERT(handle
!= NULL
|| create
== 0);
731 err
= ext4_map_blocks(handle
, inode
, &map
,
732 create
? EXT4_GET_BLOCKS_CREATE
: 0);
734 /* ensure we send some value back into *errp */
737 if (create
&& err
== 0)
738 err
= -ENOSPC
; /* should never happen */
744 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
749 if (map
.m_flags
& EXT4_MAP_NEW
) {
750 J_ASSERT(create
!= 0);
751 J_ASSERT(handle
!= NULL
);
754 * Now that we do not always journal data, we should
755 * keep in mind whether this should always journal the
756 * new buffer as metadata. For now, regular file
757 * writes use ext4_get_block instead, so it's not a
761 BUFFER_TRACE(bh
, "call get_create_access");
762 fatal
= ext4_journal_get_create_access(handle
, bh
);
763 if (!fatal
&& !buffer_uptodate(bh
)) {
764 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
765 set_buffer_uptodate(bh
);
768 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
769 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
773 BUFFER_TRACE(bh
, "not a new buffer");
783 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
784 ext4_lblk_t block
, int create
, int *err
)
786 struct buffer_head
*bh
;
788 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
791 if (buffer_uptodate(bh
))
793 ll_rw_block(READ
| REQ_META
| REQ_PRIO
, 1, &bh
);
795 if (buffer_uptodate(bh
))
802 int ext4_walk_page_buffers(handle_t
*handle
,
803 struct buffer_head
*head
,
807 int (*fn
)(handle_t
*handle
,
808 struct buffer_head
*bh
))
810 struct buffer_head
*bh
;
811 unsigned block_start
, block_end
;
812 unsigned blocksize
= head
->b_size
;
814 struct buffer_head
*next
;
816 for (bh
= head
, block_start
= 0;
817 ret
== 0 && (bh
!= head
|| !block_start
);
818 block_start
= block_end
, bh
= next
) {
819 next
= bh
->b_this_page
;
820 block_end
= block_start
+ blocksize
;
821 if (block_end
<= from
|| block_start
>= to
) {
822 if (partial
&& !buffer_uptodate(bh
))
826 err
= (*fn
)(handle
, bh
);
834 * To preserve ordering, it is essential that the hole instantiation and
835 * the data write be encapsulated in a single transaction. We cannot
836 * close off a transaction and start a new one between the ext4_get_block()
837 * and the commit_write(). So doing the jbd2_journal_start at the start of
838 * prepare_write() is the right place.
840 * Also, this function can nest inside ext4_writepage(). In that case, we
841 * *know* that ext4_writepage() has generated enough buffer credits to do the
842 * whole page. So we won't block on the journal in that case, which is good,
843 * because the caller may be PF_MEMALLOC.
845 * By accident, ext4 can be reentered when a transaction is open via
846 * quota file writes. If we were to commit the transaction while thus
847 * reentered, there can be a deadlock - we would be holding a quota
848 * lock, and the commit would never complete if another thread had a
849 * transaction open and was blocking on the quota lock - a ranking
852 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
853 * will _not_ run commit under these circumstances because handle->h_ref
854 * is elevated. We'll still have enough credits for the tiny quotafile
857 int do_journal_get_write_access(handle_t
*handle
,
858 struct buffer_head
*bh
)
860 int dirty
= buffer_dirty(bh
);
863 if (!buffer_mapped(bh
) || buffer_freed(bh
))
866 * __block_write_begin() could have dirtied some buffers. Clean
867 * the dirty bit as jbd2_journal_get_write_access() could complain
868 * otherwise about fs integrity issues. Setting of the dirty bit
869 * by __block_write_begin() isn't a real problem here as we clear
870 * the bit before releasing a page lock and thus writeback cannot
871 * ever write the buffer.
874 clear_buffer_dirty(bh
);
875 ret
= ext4_journal_get_write_access(handle
, bh
);
877 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
881 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
882 struct buffer_head
*bh_result
, int create
);
883 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
884 loff_t pos
, unsigned len
, unsigned flags
,
885 struct page
**pagep
, void **fsdata
)
887 struct inode
*inode
= mapping
->host
;
888 int ret
, needed_blocks
;
895 trace_ext4_write_begin(inode
, pos
, len
, flags
);
897 * Reserve one block more for addition to orphan list in case
898 * we allocate blocks but write fails for some reason
900 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
901 index
= pos
>> PAGE_CACHE_SHIFT
;
902 from
= pos
& (PAGE_CACHE_SIZE
- 1);
905 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
906 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
915 * grab_cache_page_write_begin() can take a long time if the
916 * system is thrashing due to memory pressure, or if the page
917 * is being written back. So grab it first before we start
918 * the transaction handle. This also allows us to allocate
919 * the page (if needed) without using GFP_NOFS.
922 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
928 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
929 if (IS_ERR(handle
)) {
930 page_cache_release(page
);
931 return PTR_ERR(handle
);
935 if (page
->mapping
!= mapping
) {
936 /* The page got truncated from under us */
938 page_cache_release(page
);
939 ext4_journal_stop(handle
);
942 wait_on_page_writeback(page
);
944 if (ext4_should_dioread_nolock(inode
))
945 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
947 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
949 if (!ret
&& ext4_should_journal_data(inode
)) {
950 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
952 do_journal_get_write_access
);
958 * __block_write_begin may have instantiated a few blocks
959 * outside i_size. Trim these off again. Don't need
960 * i_size_read because we hold i_mutex.
962 * Add inode to orphan list in case we crash before
965 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
966 ext4_orphan_add(handle
, inode
);
968 ext4_journal_stop(handle
);
969 if (pos
+ len
> inode
->i_size
) {
970 ext4_truncate_failed_write(inode
);
972 * If truncate failed early the inode might
973 * still be on the orphan list; we need to
974 * make sure the inode is removed from the
975 * orphan list in that case.
978 ext4_orphan_del(NULL
, inode
);
981 if (ret
== -ENOSPC
&&
982 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
984 page_cache_release(page
);
991 /* For write_end() in data=journal mode */
992 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
994 if (!buffer_mapped(bh
) || buffer_freed(bh
))
996 set_buffer_uptodate(bh
);
997 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1000 static int ext4_generic_write_end(struct file
*file
,
1001 struct address_space
*mapping
,
1002 loff_t pos
, unsigned len
, unsigned copied
,
1003 struct page
*page
, void *fsdata
)
1005 int i_size_changed
= 0;
1006 struct inode
*inode
= mapping
->host
;
1007 handle_t
*handle
= ext4_journal_current_handle();
1009 if (ext4_has_inline_data(inode
))
1010 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1013 copied
= block_write_end(file
, mapping
, pos
,
1014 len
, copied
, page
, fsdata
);
1017 * No need to use i_size_read() here, the i_size
1018 * cannot change under us because we hold i_mutex.
1020 * But it's important to update i_size while still holding page lock:
1021 * page writeout could otherwise come in and zero beyond i_size.
1023 if (pos
+ copied
> inode
->i_size
) {
1024 i_size_write(inode
, pos
+ copied
);
1028 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1029 /* We need to mark inode dirty even if
1030 * new_i_size is less that inode->i_size
1031 * bu greater than i_disksize.(hint delalloc)
1033 ext4_update_i_disksize(inode
, (pos
+ copied
));
1037 page_cache_release(page
);
1040 * Don't mark the inode dirty under page lock. First, it unnecessarily
1041 * makes the holding time of page lock longer. Second, it forces lock
1042 * ordering of page lock and transaction start for journaling
1046 ext4_mark_inode_dirty(handle
, inode
);
1052 * We need to pick up the new inode size which generic_commit_write gave us
1053 * `file' can be NULL - eg, when called from page_symlink().
1055 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1056 * buffers are managed internally.
1058 static int ext4_ordered_write_end(struct file
*file
,
1059 struct address_space
*mapping
,
1060 loff_t pos
, unsigned len
, unsigned copied
,
1061 struct page
*page
, void *fsdata
)
1063 handle_t
*handle
= ext4_journal_current_handle();
1064 struct inode
*inode
= mapping
->host
;
1067 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1068 ret
= ext4_jbd2_file_inode(handle
, inode
);
1071 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1074 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1075 /* if we have allocated more blocks and copied
1076 * less. We will have blocks allocated outside
1077 * inode->i_size. So truncate them
1079 ext4_orphan_add(handle
, inode
);
1084 page_cache_release(page
);
1087 ret2
= ext4_journal_stop(handle
);
1091 if (pos
+ len
> inode
->i_size
) {
1092 ext4_truncate_failed_write(inode
);
1094 * If truncate failed early the inode might still be
1095 * on the orphan list; we need to make sure the inode
1096 * is removed from the orphan list in that case.
1099 ext4_orphan_del(NULL
, inode
);
1103 return ret
? ret
: copied
;
1106 static int ext4_writeback_write_end(struct file
*file
,
1107 struct address_space
*mapping
,
1108 loff_t pos
, unsigned len
, unsigned copied
,
1109 struct page
*page
, void *fsdata
)
1111 handle_t
*handle
= ext4_journal_current_handle();
1112 struct inode
*inode
= mapping
->host
;
1115 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1116 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1119 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1120 /* if we have allocated more blocks and copied
1121 * less. We will have blocks allocated outside
1122 * inode->i_size. So truncate them
1124 ext4_orphan_add(handle
, inode
);
1129 ret2
= ext4_journal_stop(handle
);
1133 if (pos
+ len
> inode
->i_size
) {
1134 ext4_truncate_failed_write(inode
);
1136 * If truncate failed early the inode might still be
1137 * on the orphan list; we need to make sure the inode
1138 * is removed from the orphan list in that case.
1141 ext4_orphan_del(NULL
, inode
);
1144 return ret
? ret
: copied
;
1147 static int ext4_journalled_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
;
1159 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1160 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1163 BUG_ON(!ext4_handle_valid(handle
));
1165 if (ext4_has_inline_data(inode
))
1166 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1170 if (!PageUptodate(page
))
1172 page_zero_new_buffers(page
, from
+copied
, to
);
1175 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1176 to
, &partial
, write_end_fn
);
1178 SetPageUptodate(page
);
1180 new_i_size
= pos
+ copied
;
1181 if (new_i_size
> inode
->i_size
)
1182 i_size_write(inode
, pos
+copied
);
1183 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1184 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1185 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1186 ext4_update_i_disksize(inode
, new_i_size
);
1187 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1193 page_cache_release(page
);
1194 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1195 /* if we have allocated more blocks and copied
1196 * less. We will have blocks allocated outside
1197 * inode->i_size. So truncate them
1199 ext4_orphan_add(handle
, inode
);
1201 ret2
= ext4_journal_stop(handle
);
1204 if (pos
+ len
> inode
->i_size
) {
1205 ext4_truncate_failed_write(inode
);
1207 * If truncate failed early the inode might still be
1208 * on the orphan list; we need to make sure the inode
1209 * is removed from the orphan list in that case.
1212 ext4_orphan_del(NULL
, inode
);
1215 return ret
? ret
: copied
;
1219 * Reserve a single cluster located at lblock
1221 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1224 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1225 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1226 unsigned int md_needed
;
1228 ext4_lblk_t save_last_lblock
;
1232 * We will charge metadata quota at writeout time; this saves
1233 * us from metadata over-estimation, though we may go over by
1234 * a small amount in the end. Here we just reserve for data.
1236 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1241 * recalculate the amount of metadata blocks to reserve
1242 * in order to allocate nrblocks
1243 * worse case is one extent per block
1246 spin_lock(&ei
->i_block_reservation_lock
);
1248 * ext4_calc_metadata_amount() has side effects, which we have
1249 * to be prepared undo if we fail to claim space.
1251 save_len
= ei
->i_da_metadata_calc_len
;
1252 save_last_lblock
= ei
->i_da_metadata_calc_last_lblock
;
1253 md_needed
= EXT4_NUM_B2C(sbi
,
1254 ext4_calc_metadata_amount(inode
, lblock
));
1255 trace_ext4_da_reserve_space(inode
, md_needed
);
1258 * We do still charge estimated metadata to the sb though;
1259 * we cannot afford to run out of free blocks.
1261 if (ext4_claim_free_clusters(sbi
, md_needed
+ 1, 0)) {
1262 ei
->i_da_metadata_calc_len
= save_len
;
1263 ei
->i_da_metadata_calc_last_lblock
= save_last_lblock
;
1264 spin_unlock(&ei
->i_block_reservation_lock
);
1265 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1269 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1272 ei
->i_reserved_data_blocks
++;
1273 ei
->i_reserved_meta_blocks
+= md_needed
;
1274 spin_unlock(&ei
->i_block_reservation_lock
);
1276 return 0; /* success */
1279 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1281 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1282 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1285 return; /* Nothing to release, exit */
1287 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1289 trace_ext4_da_release_space(inode
, to_free
);
1290 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1292 * if there aren't enough reserved blocks, then the
1293 * counter is messed up somewhere. Since this
1294 * function is called from invalidate page, it's
1295 * harmless to return without any action.
1297 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1298 "ino %lu, to_free %d with only %d reserved "
1299 "data blocks", inode
->i_ino
, to_free
,
1300 ei
->i_reserved_data_blocks
);
1302 to_free
= ei
->i_reserved_data_blocks
;
1304 ei
->i_reserved_data_blocks
-= to_free
;
1306 if (ei
->i_reserved_data_blocks
== 0) {
1308 * We can release all of the reserved metadata blocks
1309 * only when we have written all of the delayed
1310 * allocation blocks.
1311 * Note that in case of bigalloc, i_reserved_meta_blocks,
1312 * i_reserved_data_blocks, etc. refer to number of clusters.
1314 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
1315 ei
->i_reserved_meta_blocks
);
1316 ei
->i_reserved_meta_blocks
= 0;
1317 ei
->i_da_metadata_calc_len
= 0;
1320 /* update fs dirty data blocks counter */
1321 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1323 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1325 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1328 static void ext4_da_page_release_reservation(struct page
*page
,
1329 unsigned long offset
)
1332 struct buffer_head
*head
, *bh
;
1333 unsigned int curr_off
= 0;
1334 struct inode
*inode
= page
->mapping
->host
;
1335 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1339 head
= page_buffers(page
);
1342 unsigned int next_off
= curr_off
+ bh
->b_size
;
1344 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1346 clear_buffer_delay(bh
);
1348 curr_off
= next_off
;
1349 } while ((bh
= bh
->b_this_page
) != head
);
1352 lblk
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1353 ext4_es_remove_extent(inode
, lblk
, to_release
);
1356 /* If we have released all the blocks belonging to a cluster, then we
1357 * need to release the reserved space for that cluster. */
1358 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1359 while (num_clusters
> 0) {
1360 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1361 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1362 if (sbi
->s_cluster_ratio
== 1 ||
1363 !ext4_find_delalloc_cluster(inode
, lblk
))
1364 ext4_da_release_space(inode
, 1);
1371 * Delayed allocation stuff
1375 * mpage_da_submit_io - walks through extent of pages and try to write
1376 * them with writepage() call back
1378 * @mpd->inode: inode
1379 * @mpd->first_page: first page of the extent
1380 * @mpd->next_page: page after the last page of the extent
1382 * By the time mpage_da_submit_io() is called we expect all blocks
1383 * to be allocated. this may be wrong if allocation failed.
1385 * As pages are already locked by write_cache_pages(), we can't use it
1387 static int mpage_da_submit_io(struct mpage_da_data
*mpd
,
1388 struct ext4_map_blocks
*map
)
1390 struct pagevec pvec
;
1391 unsigned long index
, end
;
1392 int ret
= 0, err
, nr_pages
, i
;
1393 struct inode
*inode
= mpd
->inode
;
1394 struct address_space
*mapping
= inode
->i_mapping
;
1395 loff_t size
= i_size_read(inode
);
1396 unsigned int len
, block_start
;
1397 struct buffer_head
*bh
, *page_bufs
= NULL
;
1398 sector_t pblock
= 0, cur_logical
= 0;
1399 struct ext4_io_submit io_submit
;
1401 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1402 memset(&io_submit
, 0, sizeof(io_submit
));
1404 * We need to start from the first_page to the next_page - 1
1405 * to make sure we also write the mapped dirty buffer_heads.
1406 * If we look at mpd->b_blocknr we would only be looking
1407 * at the currently mapped buffer_heads.
1409 index
= mpd
->first_page
;
1410 end
= mpd
->next_page
- 1;
1412 pagevec_init(&pvec
, 0);
1413 while (index
<= end
) {
1414 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1417 for (i
= 0; i
< nr_pages
; i
++) {
1419 struct page
*page
= pvec
.pages
[i
];
1421 index
= page
->index
;
1425 if (index
== size
>> PAGE_CACHE_SHIFT
)
1426 len
= size
& ~PAGE_CACHE_MASK
;
1428 len
= PAGE_CACHE_SIZE
;
1430 cur_logical
= index
<< (PAGE_CACHE_SHIFT
-
1432 pblock
= map
->m_pblk
+ (cur_logical
-
1437 BUG_ON(!PageLocked(page
));
1438 BUG_ON(PageWriteback(page
));
1440 bh
= page_bufs
= page_buffers(page
);
1443 if (map
&& (cur_logical
>= map
->m_lblk
) &&
1444 (cur_logical
<= (map
->m_lblk
+
1445 (map
->m_len
- 1)))) {
1446 if (buffer_delay(bh
)) {
1447 clear_buffer_delay(bh
);
1448 bh
->b_blocknr
= pblock
;
1450 if (buffer_unwritten(bh
) ||
1452 BUG_ON(bh
->b_blocknr
!= pblock
);
1453 if (map
->m_flags
& EXT4_MAP_UNINIT
)
1454 set_buffer_uninit(bh
);
1455 clear_buffer_unwritten(bh
);
1459 * skip page if block allocation undone and
1462 if (ext4_bh_delay_or_unwritten(NULL
, bh
))
1464 bh
= bh
->b_this_page
;
1465 block_start
+= bh
->b_size
;
1468 } while (bh
!= page_bufs
);
1475 clear_page_dirty_for_io(page
);
1476 err
= ext4_bio_write_page(&io_submit
, page
, len
,
1479 mpd
->pages_written
++;
1481 * In error case, we have to continue because
1482 * remaining pages are still locked
1487 pagevec_release(&pvec
);
1489 ext4_io_submit(&io_submit
);
1493 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
)
1497 struct pagevec pvec
;
1498 struct inode
*inode
= mpd
->inode
;
1499 struct address_space
*mapping
= inode
->i_mapping
;
1500 ext4_lblk_t start
, last
;
1502 index
= mpd
->first_page
;
1503 end
= mpd
->next_page
- 1;
1505 start
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1506 last
= end
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1507 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1509 pagevec_init(&pvec
, 0);
1510 while (index
<= end
) {
1511 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1514 for (i
= 0; i
< nr_pages
; i
++) {
1515 struct page
*page
= pvec
.pages
[i
];
1516 if (page
->index
> end
)
1518 BUG_ON(!PageLocked(page
));
1519 BUG_ON(PageWriteback(page
));
1520 block_invalidatepage(page
, 0);
1521 ClearPageUptodate(page
);
1524 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1525 pagevec_release(&pvec
);
1530 static void ext4_print_free_blocks(struct inode
*inode
)
1532 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1533 struct super_block
*sb
= inode
->i_sb
;
1535 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1536 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1537 ext4_count_free_clusters(inode
->i_sb
)));
1538 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1539 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1540 (long long) EXT4_C2B(EXT4_SB(inode
->i_sb
),
1541 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1542 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1543 (long long) EXT4_C2B(EXT4_SB(inode
->i_sb
),
1544 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1545 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1546 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1547 EXT4_I(inode
)->i_reserved_data_blocks
);
1548 ext4_msg(sb
, KERN_CRIT
, "i_reserved_meta_blocks=%u",
1549 EXT4_I(inode
)->i_reserved_meta_blocks
);
1554 * mpage_da_map_and_submit - go through given space, map them
1555 * if necessary, and then submit them for I/O
1557 * @mpd - bh describing space
1559 * The function skips space we know is already mapped to disk blocks.
1562 static void mpage_da_map_and_submit(struct mpage_da_data
*mpd
)
1564 int err
, blks
, get_blocks_flags
;
1565 struct ext4_map_blocks map
, *mapp
= NULL
;
1566 sector_t next
= mpd
->b_blocknr
;
1567 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1568 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
1569 handle_t
*handle
= NULL
;
1572 * If the blocks are mapped already, or we couldn't accumulate
1573 * any blocks, then proceed immediately to the submission stage.
1575 if ((mpd
->b_size
== 0) ||
1576 ((mpd
->b_state
& (1 << BH_Mapped
)) &&
1577 !(mpd
->b_state
& (1 << BH_Delay
)) &&
1578 !(mpd
->b_state
& (1 << BH_Unwritten
))))
1581 handle
= ext4_journal_current_handle();
1585 * Call ext4_map_blocks() to allocate any delayed allocation
1586 * blocks, or to convert an uninitialized extent to be
1587 * initialized (in the case where we have written into
1588 * one or more preallocated blocks).
1590 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1591 * indicate that we are on the delayed allocation path. This
1592 * affects functions in many different parts of the allocation
1593 * call path. This flag exists primarily because we don't
1594 * want to change *many* call functions, so ext4_map_blocks()
1595 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1596 * inode's allocation semaphore is taken.
1598 * If the blocks in questions were delalloc blocks, set
1599 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1600 * variables are updated after the blocks have been allocated.
1603 map
.m_len
= max_blocks
;
1604 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
1605 if (ext4_should_dioread_nolock(mpd
->inode
))
1606 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
1607 if (mpd
->b_state
& (1 << BH_Delay
))
1608 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
1610 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
1612 struct super_block
*sb
= mpd
->inode
->i_sb
;
1616 * If get block returns EAGAIN or ENOSPC and there
1617 * appears to be free blocks we will just let
1618 * mpage_da_submit_io() unlock all of the pages.
1623 if (err
== -ENOSPC
&& ext4_count_free_clusters(sb
)) {
1629 * get block failure will cause us to loop in
1630 * writepages, because a_ops->writepage won't be able
1631 * to make progress. The page will be redirtied by
1632 * writepage and writepages will again try to write
1635 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
1636 ext4_msg(sb
, KERN_CRIT
,
1637 "delayed block allocation failed for inode %lu "
1638 "at logical offset %llu with max blocks %zd "
1639 "with error %d", mpd
->inode
->i_ino
,
1640 (unsigned long long) next
,
1641 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1642 ext4_msg(sb
, KERN_CRIT
,
1643 "This should not happen!! Data will be lost");
1645 ext4_print_free_blocks(mpd
->inode
);
1647 /* invalidate all the pages */
1648 ext4_da_block_invalidatepages(mpd
);
1650 /* Mark this page range as having been completed */
1657 if (map
.m_flags
& EXT4_MAP_NEW
) {
1658 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
1661 for (i
= 0; i
< map
.m_len
; i
++)
1662 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
1666 * Update on-disk size along with block allocation.
1668 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
1669 if (disksize
> i_size_read(mpd
->inode
))
1670 disksize
= i_size_read(mpd
->inode
);
1671 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
1672 ext4_update_i_disksize(mpd
->inode
, disksize
);
1673 err
= ext4_mark_inode_dirty(handle
, mpd
->inode
);
1675 ext4_error(mpd
->inode
->i_sb
,
1676 "Failed to mark inode %lu dirty",
1681 mpage_da_submit_io(mpd
, mapp
);
1685 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1686 (1 << BH_Delay) | (1 << BH_Unwritten))
1689 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1691 * @mpd->lbh - extent of blocks
1692 * @logical - logical number of the block in the file
1693 * @b_state - b_state of the buffer head added
1695 * the function is used to collect contig. blocks in same state
1697 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, sector_t logical
,
1698 unsigned long b_state
)
1701 int blkbits
= mpd
->inode
->i_blkbits
;
1702 int nrblocks
= mpd
->b_size
>> blkbits
;
1705 * XXX Don't go larger than mballoc is willing to allocate
1706 * This is a stopgap solution. We eventually need to fold
1707 * mpage_da_submit_io() into this function and then call
1708 * ext4_map_blocks() multiple times in a loop
1710 if (nrblocks
>= (8*1024*1024 >> blkbits
))
1713 /* check if the reserved journal credits might overflow */
1714 if (!ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
)) {
1715 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1717 * With non-extent format we are limited by the journal
1718 * credit available. Total credit needed to insert
1719 * nrblocks contiguous blocks is dependent on the
1720 * nrblocks. So limit nrblocks.
1726 * First block in the extent
1728 if (mpd
->b_size
== 0) {
1729 mpd
->b_blocknr
= logical
;
1730 mpd
->b_size
= 1 << blkbits
;
1731 mpd
->b_state
= b_state
& BH_FLAGS
;
1735 next
= mpd
->b_blocknr
+ nrblocks
;
1737 * Can we merge the block to our big extent?
1739 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
1740 mpd
->b_size
+= 1 << blkbits
;
1746 * We couldn't merge the block to our extent, so we
1747 * need to flush current extent and start new one
1749 mpage_da_map_and_submit(mpd
);
1753 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1755 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1759 * This function is grabs code from the very beginning of
1760 * ext4_map_blocks, but assumes that the caller is from delayed write
1761 * time. This function looks up the requested blocks and sets the
1762 * buffer delay bit under the protection of i_data_sem.
1764 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1765 struct ext4_map_blocks
*map
,
1766 struct buffer_head
*bh
)
1768 struct extent_status es
;
1770 sector_t invalid_block
= ~((sector_t
) 0xffff);
1772 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1776 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1777 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1778 (unsigned long) map
->m_lblk
);
1780 /* Lookup extent status tree firstly */
1781 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1783 if (ext4_es_is_hole(&es
)) {
1785 down_read((&EXT4_I(inode
)->i_data_sem
));
1790 * Delayed extent could be allocated by fallocate.
1791 * So we need to check it.
1793 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1794 map_bh(bh
, inode
->i_sb
, invalid_block
);
1796 set_buffer_delay(bh
);
1800 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1801 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1802 if (retval
> map
->m_len
)
1803 retval
= map
->m_len
;
1804 map
->m_len
= retval
;
1805 if (ext4_es_is_written(&es
))
1806 map
->m_flags
|= EXT4_MAP_MAPPED
;
1807 else if (ext4_es_is_unwritten(&es
))
1808 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1816 * Try to see if we can get the block without requesting a new
1817 * file system block.
1819 down_read((&EXT4_I(inode
)->i_data_sem
));
1820 if (ext4_has_inline_data(inode
)) {
1822 * We will soon create blocks for this page, and let
1823 * us pretend as if the blocks aren't allocated yet.
1824 * In case of clusters, we have to handle the work
1825 * of mapping from cluster so that the reserved space
1826 * is calculated properly.
1828 if ((EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) &&
1829 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
1830 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
1832 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1833 retval
= ext4_ext_map_blocks(NULL
, inode
, map
,
1834 EXT4_GET_BLOCKS_NO_PUT_HOLE
);
1836 retval
= ext4_ind_map_blocks(NULL
, inode
, map
,
1837 EXT4_GET_BLOCKS_NO_PUT_HOLE
);
1843 * XXX: __block_prepare_write() unmaps passed block,
1846 /* If the block was allocated from previously allocated cluster,
1847 * then we dont need to reserve it again. */
1848 if (!(map
->m_flags
& EXT4_MAP_FROM_CLUSTER
)) {
1849 ret
= ext4_da_reserve_space(inode
, iblock
);
1851 /* not enough space to reserve */
1857 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1858 ~0, EXTENT_STATUS_DELAYED
);
1864 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1865 * and it should not appear on the bh->b_state.
1867 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
1869 map_bh(bh
, inode
->i_sb
, invalid_block
);
1871 set_buffer_delay(bh
);
1872 } else if (retval
> 0) {
1874 unsigned long long status
;
1876 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1877 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1878 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1879 map
->m_pblk
, status
);
1885 up_read((&EXT4_I(inode
)->i_data_sem
));
1891 * This is a special get_blocks_t callback which is used by
1892 * ext4_da_write_begin(). It will either return mapped block or
1893 * reserve space for a single block.
1895 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1896 * We also have b_blocknr = -1 and b_bdev initialized properly
1898 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1899 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1900 * initialized properly.
1902 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1903 struct buffer_head
*bh
, int create
)
1905 struct ext4_map_blocks map
;
1908 BUG_ON(create
== 0);
1909 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1911 map
.m_lblk
= iblock
;
1915 * first, we need to know whether the block is allocated already
1916 * preallocated blocks are unmapped but should treated
1917 * the same as allocated blocks.
1919 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1923 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1924 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1926 if (buffer_unwritten(bh
)) {
1927 /* A delayed write to unwritten bh should be marked
1928 * new and mapped. Mapped ensures that we don't do
1929 * get_block multiple times when we write to the same
1930 * offset and new ensures that we do proper zero out
1931 * for partial write.
1934 set_buffer_mapped(bh
);
1939 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1945 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1951 static int __ext4_journalled_writepage(struct page
*page
,
1954 struct address_space
*mapping
= page
->mapping
;
1955 struct inode
*inode
= mapping
->host
;
1956 struct buffer_head
*page_bufs
= NULL
;
1957 handle_t
*handle
= NULL
;
1958 int ret
= 0, err
= 0;
1959 int inline_data
= ext4_has_inline_data(inode
);
1960 struct buffer_head
*inode_bh
= NULL
;
1962 ClearPageChecked(page
);
1965 BUG_ON(page
->index
!= 0);
1966 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1967 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1968 if (inode_bh
== NULL
)
1971 page_bufs
= page_buffers(page
);
1976 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1979 /* As soon as we unlock the page, it can go away, but we have
1980 * references to buffers so we are safe */
1983 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1984 ext4_writepage_trans_blocks(inode
));
1985 if (IS_ERR(handle
)) {
1986 ret
= PTR_ERR(handle
);
1990 BUG_ON(!ext4_handle_valid(handle
));
1993 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1995 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1998 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1999 do_journal_get_write_access
);
2001 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2006 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
2007 err
= ext4_journal_stop(handle
);
2011 if (!ext4_has_inline_data(inode
))
2012 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
2014 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2021 * Note that we don't need to start a transaction unless we're journaling data
2022 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2023 * need to file the inode to the transaction's list in ordered mode because if
2024 * we are writing back data added by write(), the inode is already there and if
2025 * we are writing back data modified via mmap(), no one guarantees in which
2026 * transaction the data will hit the disk. In case we are journaling data, we
2027 * cannot start transaction directly because transaction start ranks above page
2028 * lock so we have to do some magic.
2030 * This function can get called via...
2031 * - ext4_da_writepages after taking page lock (have journal handle)
2032 * - journal_submit_inode_data_buffers (no journal handle)
2033 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2034 * - grab_page_cache when doing write_begin (have journal handle)
2036 * We don't do any block allocation in this function. If we have page with
2037 * multiple blocks we need to write those buffer_heads that are mapped. This
2038 * is important for mmaped based write. So if we do with blocksize 1K
2039 * truncate(f, 1024);
2040 * a = mmap(f, 0, 4096);
2042 * truncate(f, 4096);
2043 * we have in the page first buffer_head mapped via page_mkwrite call back
2044 * but other buffer_heads would be unmapped but dirty (dirty done via the
2045 * do_wp_page). So writepage should write the first block. If we modify
2046 * the mmap area beyond 1024 we will again get a page_fault and the
2047 * page_mkwrite callback will do the block allocation and mark the
2048 * buffer_heads mapped.
2050 * We redirty the page if we have any buffer_heads that is either delay or
2051 * unwritten in the page.
2053 * We can get recursively called as show below.
2055 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2058 * But since we don't do any block allocation we should not deadlock.
2059 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2061 static int ext4_writepage(struct page
*page
,
2062 struct writeback_control
*wbc
)
2067 struct buffer_head
*page_bufs
= NULL
;
2068 struct inode
*inode
= page
->mapping
->host
;
2069 struct ext4_io_submit io_submit
;
2071 trace_ext4_writepage(page
);
2072 size
= i_size_read(inode
);
2073 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2074 len
= size
& ~PAGE_CACHE_MASK
;
2076 len
= PAGE_CACHE_SIZE
;
2078 page_bufs
= page_buffers(page
);
2080 * We cannot do block allocation or other extent handling in this
2081 * function. If there are buffers needing that, we have to redirty
2082 * the page. But we may reach here when we do a journal commit via
2083 * journal_submit_inode_data_buffers() and in that case we must write
2084 * allocated buffers to achieve data=ordered mode guarantees.
2086 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2087 ext4_bh_delay_or_unwritten
)) {
2088 redirty_page_for_writepage(wbc
, page
);
2089 if (current
->flags
& PF_MEMALLOC
) {
2091 * For memory cleaning there's no point in writing only
2092 * some buffers. So just bail out. Warn if we came here
2093 * from direct reclaim.
2095 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2102 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2104 * It's mmapped pagecache. Add buffers and journal it. There
2105 * doesn't seem much point in redirtying the page here.
2107 return __ext4_journalled_writepage(page
, len
);
2109 memset(&io_submit
, 0, sizeof(io_submit
));
2110 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
);
2111 ext4_io_submit(&io_submit
);
2116 * This is called via ext4_da_writepages() to
2117 * calculate the total number of credits to reserve to fit
2118 * a single extent allocation into a single transaction,
2119 * ext4_da_writpeages() will loop calling this before
2120 * the block allocation.
2123 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2125 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2128 * With non-extent format the journal credit needed to
2129 * insert nrblocks contiguous block is dependent on
2130 * number of contiguous block. So we will limit
2131 * number of contiguous block to a sane value
2133 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
2134 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2135 max_blocks
= EXT4_MAX_TRANS_DATA
;
2137 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2141 * write_cache_pages_da - walk the list of dirty pages of the given
2142 * address space and accumulate pages that need writing, and call
2143 * mpage_da_map_and_submit to map a single contiguous memory region
2144 * and then write them.
2146 static int write_cache_pages_da(handle_t
*handle
,
2147 struct address_space
*mapping
,
2148 struct writeback_control
*wbc
,
2149 struct mpage_da_data
*mpd
,
2150 pgoff_t
*done_index
)
2152 struct buffer_head
*bh
, *head
;
2153 struct inode
*inode
= mapping
->host
;
2154 struct pagevec pvec
;
2155 unsigned int nr_pages
;
2158 long nr_to_write
= wbc
->nr_to_write
;
2159 int i
, tag
, ret
= 0;
2161 memset(mpd
, 0, sizeof(struct mpage_da_data
));
2164 pagevec_init(&pvec
, 0);
2165 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2166 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2168 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2169 tag
= PAGECACHE_TAG_TOWRITE
;
2171 tag
= PAGECACHE_TAG_DIRTY
;
2173 *done_index
= index
;
2174 while (index
<= end
) {
2175 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2176 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2180 for (i
= 0; i
< nr_pages
; i
++) {
2181 struct page
*page
= pvec
.pages
[i
];
2184 * At this point, the page may be truncated or
2185 * invalidated (changing page->mapping to NULL), or
2186 * even swizzled back from swapper_space to tmpfs file
2187 * mapping. However, page->index will not change
2188 * because we have a reference on the page.
2190 if (page
->index
> end
)
2193 *done_index
= page
->index
+ 1;
2196 * If we can't merge this page, and we have
2197 * accumulated an contiguous region, write it
2199 if ((mpd
->next_page
!= page
->index
) &&
2200 (mpd
->next_page
!= mpd
->first_page
)) {
2201 mpage_da_map_and_submit(mpd
);
2202 goto ret_extent_tail
;
2208 * If the page is no longer dirty, or its
2209 * mapping no longer corresponds to inode we
2210 * are writing (which means it has been
2211 * truncated or invalidated), or the page is
2212 * already under writeback and we are not
2213 * doing a data integrity writeback, skip the page
2215 if (!PageDirty(page
) ||
2216 (PageWriteback(page
) &&
2217 (wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2218 unlikely(page
->mapping
!= mapping
)) {
2223 wait_on_page_writeback(page
);
2224 BUG_ON(PageWriteback(page
));
2227 * If we have inline data and arrive here, it means that
2228 * we will soon create the block for the 1st page, so
2229 * we'd better clear the inline data here.
2231 if (ext4_has_inline_data(inode
)) {
2232 BUG_ON(ext4_test_inode_state(inode
,
2233 EXT4_STATE_MAY_INLINE_DATA
));
2234 ext4_destroy_inline_data(handle
, inode
);
2237 if (mpd
->next_page
!= page
->index
)
2238 mpd
->first_page
= page
->index
;
2239 mpd
->next_page
= page
->index
+ 1;
2240 logical
= (sector_t
) page
->index
<<
2241 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2243 /* Add all dirty buffers to mpd */
2244 head
= page_buffers(page
);
2247 BUG_ON(buffer_locked(bh
));
2249 * We need to try to allocate unmapped blocks
2250 * in the same page. Otherwise we won't make
2251 * progress with the page in ext4_writepage
2253 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2254 mpage_add_bh_to_extent(mpd
, logical
,
2257 goto ret_extent_tail
;
2258 } else if (buffer_dirty(bh
) &&
2259 buffer_mapped(bh
)) {
2261 * mapped dirty buffer. We need to
2262 * update the b_state because we look
2263 * at b_state in mpage_da_map_blocks.
2264 * We don't update b_size because if we
2265 * find an unmapped buffer_head later
2266 * we need to use the b_state flag of
2269 if (mpd
->b_size
== 0)
2271 bh
->b_state
& BH_FLAGS
;
2274 } while ((bh
= bh
->b_this_page
) != head
);
2276 if (nr_to_write
> 0) {
2278 if (nr_to_write
== 0 &&
2279 wbc
->sync_mode
== WB_SYNC_NONE
)
2281 * We stop writing back only if we are
2282 * not doing integrity sync. In case of
2283 * integrity sync we have to keep going
2284 * because someone may be concurrently
2285 * dirtying pages, and we might have
2286 * synced a lot of newly appeared dirty
2287 * pages, but have not synced all of the
2293 pagevec_release(&pvec
);
2298 ret
= MPAGE_DA_EXTENT_TAIL
;
2300 pagevec_release(&pvec
);
2306 static int ext4_da_writepages(struct address_space
*mapping
,
2307 struct writeback_control
*wbc
)
2310 int range_whole
= 0;
2311 handle_t
*handle
= NULL
;
2312 struct mpage_da_data mpd
;
2313 struct inode
*inode
= mapping
->host
;
2314 int pages_written
= 0;
2315 unsigned int max_pages
;
2316 int range_cyclic
, cycled
= 1, io_done
= 0;
2317 int needed_blocks
, ret
= 0;
2318 long desired_nr_to_write
, nr_to_writebump
= 0;
2319 loff_t range_start
= wbc
->range_start
;
2320 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2321 pgoff_t done_index
= 0;
2323 struct blk_plug plug
;
2325 trace_ext4_da_writepages(inode
, wbc
);
2328 * No pages to write? This is mainly a kludge to avoid starting
2329 * a transaction for special inodes like journal inode on last iput()
2330 * because that could violate lock ordering on umount
2332 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2336 * If the filesystem has aborted, it is read-only, so return
2337 * right away instead of dumping stack traces later on that
2338 * will obscure the real source of the problem. We test
2339 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2340 * the latter could be true if the filesystem is mounted
2341 * read-only, and in that case, ext4_da_writepages should
2342 * *never* be called, so if that ever happens, we would want
2345 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2348 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2351 range_cyclic
= wbc
->range_cyclic
;
2352 if (wbc
->range_cyclic
) {
2353 index
= mapping
->writeback_index
;
2356 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2357 wbc
->range_end
= LLONG_MAX
;
2358 wbc
->range_cyclic
= 0;
2361 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2362 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2366 * This works around two forms of stupidity. The first is in
2367 * the writeback code, which caps the maximum number of pages
2368 * written to be 1024 pages. This is wrong on multiple
2369 * levels; different architectues have a different page size,
2370 * which changes the maximum amount of data which gets
2371 * written. Secondly, 4 megabytes is way too small. XFS
2372 * forces this value to be 16 megabytes by multiplying
2373 * nr_to_write parameter by four, and then relies on its
2374 * allocator to allocate larger extents to make them
2375 * contiguous. Unfortunately this brings us to the second
2376 * stupidity, which is that ext4's mballoc code only allocates
2377 * at most 2048 blocks. So we force contiguous writes up to
2378 * the number of dirty blocks in the inode, or
2379 * sbi->max_writeback_mb_bump whichever is smaller.
2381 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2382 if (!range_cyclic
&& range_whole
) {
2383 if (wbc
->nr_to_write
== LONG_MAX
)
2384 desired_nr_to_write
= wbc
->nr_to_write
;
2386 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2388 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2390 if (desired_nr_to_write
> max_pages
)
2391 desired_nr_to_write
= max_pages
;
2393 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2394 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2395 wbc
->nr_to_write
= desired_nr_to_write
;
2399 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2400 tag_pages_for_writeback(mapping
, index
, end
);
2402 blk_start_plug(&plug
);
2403 while (!ret
&& wbc
->nr_to_write
> 0) {
2406 * we insert one extent at a time. So we need
2407 * credit needed for single extent allocation.
2408 * journalled mode is currently not supported
2411 BUG_ON(ext4_should_journal_data(inode
));
2412 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2414 /* start a new transaction*/
2415 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2417 if (IS_ERR(handle
)) {
2418 ret
= PTR_ERR(handle
);
2419 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2420 "%ld pages, ino %lu; err %d", __func__
,
2421 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2422 blk_finish_plug(&plug
);
2423 goto out_writepages
;
2427 * Now call write_cache_pages_da() to find the next
2428 * contiguous region of logical blocks that need
2429 * blocks to be allocated by ext4 and submit them.
2431 ret
= write_cache_pages_da(handle
, mapping
,
2432 wbc
, &mpd
, &done_index
);
2434 * If we have a contiguous extent of pages and we
2435 * haven't done the I/O yet, map the blocks and submit
2438 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2439 mpage_da_map_and_submit(&mpd
);
2440 ret
= MPAGE_DA_EXTENT_TAIL
;
2442 trace_ext4_da_write_pages(inode
, &mpd
);
2443 wbc
->nr_to_write
-= mpd
.pages_written
;
2445 ext4_journal_stop(handle
);
2447 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2448 /* commit the transaction which would
2449 * free blocks released in the transaction
2452 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2454 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2456 * Got one extent now try with rest of the pages.
2457 * If mpd.retval is set -EIO, journal is aborted.
2458 * So we don't need to write any more.
2460 pages_written
+= mpd
.pages_written
;
2463 } else if (wbc
->nr_to_write
)
2465 * There is no more writeout needed
2466 * or we requested for a noblocking writeout
2467 * and we found the device congested
2471 blk_finish_plug(&plug
);
2472 if (!io_done
&& !cycled
) {
2475 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2476 wbc
->range_end
= mapping
->writeback_index
- 1;
2481 wbc
->range_cyclic
= range_cyclic
;
2482 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2484 * set the writeback_index so that range_cyclic
2485 * mode will write it back later
2487 mapping
->writeback_index
= done_index
;
2490 wbc
->nr_to_write
-= nr_to_writebump
;
2491 wbc
->range_start
= range_start
;
2492 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2496 static int ext4_nonda_switch(struct super_block
*sb
)
2498 s64 free_blocks
, dirty_blocks
;
2499 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2502 * switch to non delalloc mode if we are running low
2503 * on free block. The free block accounting via percpu
2504 * counters can get slightly wrong with percpu_counter_batch getting
2505 * accumulated on each CPU without updating global counters
2506 * Delalloc need an accurate free block accounting. So switch
2507 * to non delalloc when we are near to error range.
2509 free_blocks
= EXT4_C2B(sbi
,
2510 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
));
2511 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2513 * Start pushing delalloc when 1/2 of free blocks are dirty.
2515 if (dirty_blocks
&& (free_blocks
< 2 * dirty_blocks
))
2516 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2518 if (2 * free_blocks
< 3 * dirty_blocks
||
2519 free_blocks
< (dirty_blocks
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2521 * free block count is less than 150% of dirty blocks
2522 * or free blocks is less than watermark
2529 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2530 loff_t pos
, unsigned len
, unsigned flags
,
2531 struct page
**pagep
, void **fsdata
)
2533 int ret
, retries
= 0;
2536 struct inode
*inode
= mapping
->host
;
2539 index
= pos
>> PAGE_CACHE_SHIFT
;
2541 if (ext4_nonda_switch(inode
->i_sb
)) {
2542 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2543 return ext4_write_begin(file
, mapping
, pos
,
2544 len
, flags
, pagep
, fsdata
);
2546 *fsdata
= (void *)0;
2547 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2549 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2550 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2560 * grab_cache_page_write_begin() can take a long time if the
2561 * system is thrashing due to memory pressure, or if the page
2562 * is being written back. So grab it first before we start
2563 * the transaction handle. This also allows us to allocate
2564 * the page (if needed) without using GFP_NOFS.
2567 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2573 * With delayed allocation, we don't log the i_disksize update
2574 * if there is delayed block allocation. But we still need
2575 * to journalling the i_disksize update if writes to the end
2576 * of file which has an already mapped buffer.
2579 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, 1);
2580 if (IS_ERR(handle
)) {
2581 page_cache_release(page
);
2582 return PTR_ERR(handle
);
2586 if (page
->mapping
!= mapping
) {
2587 /* The page got truncated from under us */
2589 page_cache_release(page
);
2590 ext4_journal_stop(handle
);
2593 /* In case writeback began while the page was unlocked */
2594 wait_on_page_writeback(page
);
2596 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2599 ext4_journal_stop(handle
);
2601 * block_write_begin may have instantiated a few blocks
2602 * outside i_size. Trim these off again. Don't need
2603 * i_size_read because we hold i_mutex.
2605 if (pos
+ len
> inode
->i_size
)
2606 ext4_truncate_failed_write(inode
);
2608 if (ret
== -ENOSPC
&&
2609 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2612 page_cache_release(page
);
2621 * Check if we should update i_disksize
2622 * when write to the end of file but not require block allocation
2624 static int ext4_da_should_update_i_disksize(struct page
*page
,
2625 unsigned long offset
)
2627 struct buffer_head
*bh
;
2628 struct inode
*inode
= page
->mapping
->host
;
2632 bh
= page_buffers(page
);
2633 idx
= offset
>> inode
->i_blkbits
;
2635 for (i
= 0; i
< idx
; i
++)
2636 bh
= bh
->b_this_page
;
2638 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2643 static int ext4_da_write_end(struct file
*file
,
2644 struct address_space
*mapping
,
2645 loff_t pos
, unsigned len
, unsigned copied
,
2646 struct page
*page
, void *fsdata
)
2648 struct inode
*inode
= mapping
->host
;
2650 handle_t
*handle
= ext4_journal_current_handle();
2652 unsigned long start
, end
;
2653 int write_mode
= (int)(unsigned long)fsdata
;
2655 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2656 switch (ext4_inode_journal_mode(inode
)) {
2657 case EXT4_INODE_ORDERED_DATA_MODE
:
2658 return ext4_ordered_write_end(file
, mapping
, pos
,
2659 len
, copied
, page
, fsdata
);
2660 case EXT4_INODE_WRITEBACK_DATA_MODE
:
2661 return ext4_writeback_write_end(file
, mapping
, pos
,
2662 len
, copied
, page
, fsdata
);
2668 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2669 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2670 end
= start
+ copied
- 1;
2673 * generic_write_end() will run mark_inode_dirty() if i_size
2674 * changes. So let's piggyback the i_disksize mark_inode_dirty
2677 new_i_size
= pos
+ copied
;
2678 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2679 if (ext4_has_inline_data(inode
) ||
2680 ext4_da_should_update_i_disksize(page
, end
)) {
2681 down_write(&EXT4_I(inode
)->i_data_sem
);
2682 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
2683 EXT4_I(inode
)->i_disksize
= new_i_size
;
2684 up_write(&EXT4_I(inode
)->i_data_sem
);
2685 /* We need to mark inode dirty even if
2686 * new_i_size is less that inode->i_size
2687 * bu greater than i_disksize.(hint delalloc)
2689 ext4_mark_inode_dirty(handle
, inode
);
2693 if (write_mode
!= CONVERT_INLINE_DATA
&&
2694 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
2695 ext4_has_inline_data(inode
))
2696 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
2699 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2705 ret2
= ext4_journal_stop(handle
);
2709 return ret
? ret
: copied
;
2712 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2715 * Drop reserved blocks
2717 BUG_ON(!PageLocked(page
));
2718 if (!page_has_buffers(page
))
2721 ext4_da_page_release_reservation(page
, offset
);
2724 ext4_invalidatepage(page
, offset
);
2730 * Force all delayed allocation blocks to be allocated for a given inode.
2732 int ext4_alloc_da_blocks(struct inode
*inode
)
2734 trace_ext4_alloc_da_blocks(inode
);
2736 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2737 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2741 * We do something simple for now. The filemap_flush() will
2742 * also start triggering a write of the data blocks, which is
2743 * not strictly speaking necessary (and for users of
2744 * laptop_mode, not even desirable). However, to do otherwise
2745 * would require replicating code paths in:
2747 * ext4_da_writepages() ->
2748 * write_cache_pages() ---> (via passed in callback function)
2749 * __mpage_da_writepage() -->
2750 * mpage_add_bh_to_extent()
2751 * mpage_da_map_blocks()
2753 * The problem is that write_cache_pages(), located in
2754 * mm/page-writeback.c, marks pages clean in preparation for
2755 * doing I/O, which is not desirable if we're not planning on
2758 * We could call write_cache_pages(), and then redirty all of
2759 * the pages by calling redirty_page_for_writepage() but that
2760 * would be ugly in the extreme. So instead we would need to
2761 * replicate parts of the code in the above functions,
2762 * simplifying them because we wouldn't actually intend to
2763 * write out the pages, but rather only collect contiguous
2764 * logical block extents, call the multi-block allocator, and
2765 * then update the buffer heads with the block allocations.
2767 * For now, though, we'll cheat by calling filemap_flush(),
2768 * which will map the blocks, and start the I/O, but not
2769 * actually wait for the I/O to complete.
2771 return filemap_flush(inode
->i_mapping
);
2775 * bmap() is special. It gets used by applications such as lilo and by
2776 * the swapper to find the on-disk block of a specific piece of data.
2778 * Naturally, this is dangerous if the block concerned is still in the
2779 * journal. If somebody makes a swapfile on an ext4 data-journaling
2780 * filesystem and enables swap, then they may get a nasty shock when the
2781 * data getting swapped to that swapfile suddenly gets overwritten by
2782 * the original zero's written out previously to the journal and
2783 * awaiting writeback in the kernel's buffer cache.
2785 * So, if we see any bmap calls here on a modified, data-journaled file,
2786 * take extra steps to flush any blocks which might be in the cache.
2788 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2790 struct inode
*inode
= mapping
->host
;
2795 * We can get here for an inline file via the FIBMAP ioctl
2797 if (ext4_has_inline_data(inode
))
2800 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2801 test_opt(inode
->i_sb
, DELALLOC
)) {
2803 * With delalloc we want to sync the file
2804 * so that we can make sure we allocate
2807 filemap_write_and_wait(mapping
);
2810 if (EXT4_JOURNAL(inode
) &&
2811 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2813 * This is a REALLY heavyweight approach, but the use of
2814 * bmap on dirty files is expected to be extremely rare:
2815 * only if we run lilo or swapon on a freshly made file
2816 * do we expect this to happen.
2818 * (bmap requires CAP_SYS_RAWIO so this does not
2819 * represent an unprivileged user DOS attack --- we'd be
2820 * in trouble if mortal users could trigger this path at
2823 * NB. EXT4_STATE_JDATA is not set on files other than
2824 * regular files. If somebody wants to bmap a directory
2825 * or symlink and gets confused because the buffer
2826 * hasn't yet been flushed to disk, they deserve
2827 * everything they get.
2830 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2831 journal
= EXT4_JOURNAL(inode
);
2832 jbd2_journal_lock_updates(journal
);
2833 err
= jbd2_journal_flush(journal
);
2834 jbd2_journal_unlock_updates(journal
);
2840 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2843 static int ext4_readpage(struct file
*file
, struct page
*page
)
2846 struct inode
*inode
= page
->mapping
->host
;
2848 trace_ext4_readpage(page
);
2850 if (ext4_has_inline_data(inode
))
2851 ret
= ext4_readpage_inline(inode
, page
);
2854 return mpage_readpage(page
, ext4_get_block
);
2860 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2861 struct list_head
*pages
, unsigned nr_pages
)
2863 struct inode
*inode
= mapping
->host
;
2865 /* If the file has inline data, no need to do readpages. */
2866 if (ext4_has_inline_data(inode
))
2869 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2872 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2874 trace_ext4_invalidatepage(page
, offset
);
2876 /* No journalling happens on data buffers when this function is used */
2877 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
2879 block_invalidatepage(page
, offset
);
2882 static int __ext4_journalled_invalidatepage(struct page
*page
,
2883 unsigned long offset
)
2885 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2887 trace_ext4_journalled_invalidatepage(page
, offset
);
2890 * If it's a full truncate we just forget about the pending dirtying
2893 ClearPageChecked(page
);
2895 return jbd2_journal_invalidatepage(journal
, page
, offset
);
2898 /* Wrapper for aops... */
2899 static void ext4_journalled_invalidatepage(struct page
*page
,
2900 unsigned long offset
)
2902 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
) < 0);
2905 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2907 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2909 trace_ext4_releasepage(page
);
2911 WARN_ON(PageChecked(page
));
2912 if (!page_has_buffers(page
))
2915 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2917 return try_to_free_buffers(page
);
2921 * ext4_get_block used when preparing for a DIO write or buffer write.
2922 * We allocate an uinitialized extent if blocks haven't been allocated.
2923 * The extent will be converted to initialized after the IO is complete.
2925 int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
2926 struct buffer_head
*bh_result
, int create
)
2928 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2929 inode
->i_ino
, create
);
2930 return _ext4_get_block(inode
, iblock
, bh_result
,
2931 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
2934 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
2935 struct buffer_head
*bh_result
, int create
)
2937 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2938 inode
->i_ino
, create
);
2939 return _ext4_get_block(inode
, iblock
, bh_result
,
2940 EXT4_GET_BLOCKS_NO_LOCK
);
2943 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
2944 ssize_t size
, void *private, int ret
,
2947 struct inode
*inode
= file_inode(iocb
->ki_filp
);
2948 ext4_io_end_t
*io_end
= iocb
->private;
2950 /* if not async direct IO or dio with 0 bytes write, just return */
2951 if (!io_end
|| !size
)
2954 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2955 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2956 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
2959 iocb
->private = NULL
;
2961 /* if not aio dio with unwritten extents, just free io and return */
2962 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
2963 ext4_free_io_end(io_end
);
2965 inode_dio_done(inode
);
2967 aio_complete(iocb
, ret
, 0);
2971 io_end
->offset
= offset
;
2972 io_end
->size
= size
;
2974 io_end
->iocb
= iocb
;
2975 io_end
->result
= ret
;
2978 ext4_add_complete_io(io_end
);
2982 * For ext4 extent files, ext4 will do direct-io write to holes,
2983 * preallocated extents, and those write extend the file, no need to
2984 * fall back to buffered IO.
2986 * For holes, we fallocate those blocks, mark them as uninitialized
2987 * If those blocks were preallocated, we mark sure they are split, but
2988 * still keep the range to write as uninitialized.
2990 * The unwritten extents will be converted to written when DIO is completed.
2991 * For async direct IO, since the IO may still pending when return, we
2992 * set up an end_io call back function, which will do the conversion
2993 * when async direct IO completed.
2995 * If the O_DIRECT write will extend the file then add this inode to the
2996 * orphan list. So recovery will truncate it back to the original size
2997 * if the machine crashes during the write.
3000 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3001 const struct iovec
*iov
, loff_t offset
,
3002 unsigned long nr_segs
)
3004 struct file
*file
= iocb
->ki_filp
;
3005 struct inode
*inode
= file
->f_mapping
->host
;
3007 size_t count
= iov_length(iov
, nr_segs
);
3009 get_block_t
*get_block_func
= NULL
;
3011 loff_t final_size
= offset
+ count
;
3013 /* Use the old path for reads and writes beyond i_size. */
3014 if (rw
!= WRITE
|| final_size
> inode
->i_size
)
3015 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3017 BUG_ON(iocb
->private == NULL
);
3019 /* If we do a overwrite dio, i_mutex locking can be released */
3020 overwrite
= *((int *)iocb
->private);
3023 atomic_inc(&inode
->i_dio_count
);
3024 down_read(&EXT4_I(inode
)->i_data_sem
);
3025 mutex_unlock(&inode
->i_mutex
);
3029 * We could direct write to holes and fallocate.
3031 * Allocated blocks to fill the hole are marked as
3032 * uninitialized to prevent parallel buffered read to expose
3033 * the stale data before DIO complete the data IO.
3035 * As to previously fallocated extents, ext4 get_block will
3036 * just simply mark the buffer mapped but still keep the
3037 * extents uninitialized.
3039 * For non AIO case, we will convert those unwritten extents
3040 * to written after return back from blockdev_direct_IO.
3042 * For async DIO, the conversion needs to be deferred when the
3043 * IO is completed. The ext4 end_io callback function will be
3044 * called to take care of the conversion work. Here for async
3045 * case, we allocate an io_end structure to hook to the iocb.
3047 iocb
->private = NULL
;
3048 ext4_inode_aio_set(inode
, NULL
);
3049 if (!is_sync_kiocb(iocb
)) {
3050 ext4_io_end_t
*io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
3055 io_end
->flag
|= EXT4_IO_END_DIRECT
;
3056 iocb
->private = io_end
;
3058 * we save the io structure for current async direct
3059 * IO, so that later ext4_map_blocks() could flag the
3060 * io structure whether there is a unwritten extents
3061 * needs to be converted when IO is completed.
3063 ext4_inode_aio_set(inode
, io_end
);
3067 get_block_func
= ext4_get_block_write_nolock
;
3069 get_block_func
= ext4_get_block_write
;
3070 dio_flags
= DIO_LOCKING
;
3072 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3073 inode
->i_sb
->s_bdev
, iov
,
3081 ext4_inode_aio_set(inode
, NULL
);
3083 * The io_end structure takes a reference to the inode, that
3084 * structure needs to be destroyed and the reference to the
3085 * inode need to be dropped, when IO is complete, even with 0
3086 * byte write, or failed.
3088 * In the successful AIO DIO case, the io_end structure will
3089 * be destroyed and the reference to the inode will be dropped
3090 * after the end_io call back function is called.
3092 * In the case there is 0 byte write, or error case, since VFS
3093 * direct IO won't invoke the end_io call back function, we
3094 * need to free the end_io structure here.
3096 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3097 ext4_free_io_end(iocb
->private);
3098 iocb
->private = NULL
;
3099 } else if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3100 EXT4_STATE_DIO_UNWRITTEN
)) {
3103 * for non AIO case, since the IO is already
3104 * completed, we could do the conversion right here
3106 err
= ext4_convert_unwritten_extents(inode
,
3110 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3114 /* take i_mutex locking again if we do a ovewrite dio */
3116 inode_dio_done(inode
);
3117 up_read(&EXT4_I(inode
)->i_data_sem
);
3118 mutex_lock(&inode
->i_mutex
);
3124 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3125 const struct iovec
*iov
, loff_t offset
,
3126 unsigned long nr_segs
)
3128 struct file
*file
= iocb
->ki_filp
;
3129 struct inode
*inode
= file
->f_mapping
->host
;
3133 * If we are doing data journalling we don't support O_DIRECT
3135 if (ext4_should_journal_data(inode
))
3138 /* Let buffer I/O handle the inline data case. */
3139 if (ext4_has_inline_data(inode
))
3142 trace_ext4_direct_IO_enter(inode
, offset
, iov_length(iov
, nr_segs
), rw
);
3143 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3144 ret
= ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3146 ret
= ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3147 trace_ext4_direct_IO_exit(inode
, offset
,
3148 iov_length(iov
, nr_segs
), rw
, ret
);
3153 * Pages can be marked dirty completely asynchronously from ext4's journalling
3154 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3155 * much here because ->set_page_dirty is called under VFS locks. The page is
3156 * not necessarily locked.
3158 * We cannot just dirty the page and leave attached buffers clean, because the
3159 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3160 * or jbddirty because all the journalling code will explode.
3162 * So what we do is to mark the page "pending dirty" and next time writepage
3163 * is called, propagate that into the buffers appropriately.
3165 static int ext4_journalled_set_page_dirty(struct page
*page
)
3167 SetPageChecked(page
);
3168 return __set_page_dirty_nobuffers(page
);
3171 static const struct address_space_operations ext4_ordered_aops
= {
3172 .readpage
= ext4_readpage
,
3173 .readpages
= ext4_readpages
,
3174 .writepage
= ext4_writepage
,
3175 .write_begin
= ext4_write_begin
,
3176 .write_end
= ext4_ordered_write_end
,
3178 .invalidatepage
= ext4_invalidatepage
,
3179 .releasepage
= ext4_releasepage
,
3180 .direct_IO
= ext4_direct_IO
,
3181 .migratepage
= buffer_migrate_page
,
3182 .is_partially_uptodate
= block_is_partially_uptodate
,
3183 .error_remove_page
= generic_error_remove_page
,
3186 static const struct address_space_operations ext4_writeback_aops
= {
3187 .readpage
= ext4_readpage
,
3188 .readpages
= ext4_readpages
,
3189 .writepage
= ext4_writepage
,
3190 .write_begin
= ext4_write_begin
,
3191 .write_end
= ext4_writeback_write_end
,
3193 .invalidatepage
= ext4_invalidatepage
,
3194 .releasepage
= ext4_releasepage
,
3195 .direct_IO
= ext4_direct_IO
,
3196 .migratepage
= buffer_migrate_page
,
3197 .is_partially_uptodate
= block_is_partially_uptodate
,
3198 .error_remove_page
= generic_error_remove_page
,
3201 static const struct address_space_operations ext4_journalled_aops
= {
3202 .readpage
= ext4_readpage
,
3203 .readpages
= ext4_readpages
,
3204 .writepage
= ext4_writepage
,
3205 .write_begin
= ext4_write_begin
,
3206 .write_end
= ext4_journalled_write_end
,
3207 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3209 .invalidatepage
= ext4_journalled_invalidatepage
,
3210 .releasepage
= ext4_releasepage
,
3211 .direct_IO
= ext4_direct_IO
,
3212 .is_partially_uptodate
= block_is_partially_uptodate
,
3213 .error_remove_page
= generic_error_remove_page
,
3216 static const struct address_space_operations ext4_da_aops
= {
3217 .readpage
= ext4_readpage
,
3218 .readpages
= ext4_readpages
,
3219 .writepage
= ext4_writepage
,
3220 .writepages
= ext4_da_writepages
,
3221 .write_begin
= ext4_da_write_begin
,
3222 .write_end
= ext4_da_write_end
,
3224 .invalidatepage
= ext4_da_invalidatepage
,
3225 .releasepage
= ext4_releasepage
,
3226 .direct_IO
= ext4_direct_IO
,
3227 .migratepage
= buffer_migrate_page
,
3228 .is_partially_uptodate
= block_is_partially_uptodate
,
3229 .error_remove_page
= generic_error_remove_page
,
3232 void ext4_set_aops(struct inode
*inode
)
3234 switch (ext4_inode_journal_mode(inode
)) {
3235 case EXT4_INODE_ORDERED_DATA_MODE
:
3236 if (test_opt(inode
->i_sb
, DELALLOC
))
3237 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3239 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3241 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3242 if (test_opt(inode
->i_sb
, DELALLOC
))
3243 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3245 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3247 case EXT4_INODE_JOURNAL_DATA_MODE
:
3248 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3257 * ext4_discard_partial_page_buffers()
3258 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3259 * This function finds and locks the page containing the offset
3260 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3261 * Calling functions that already have the page locked should call
3262 * ext4_discard_partial_page_buffers_no_lock directly.
3264 int ext4_discard_partial_page_buffers(handle_t
*handle
,
3265 struct address_space
*mapping
, loff_t from
,
3266 loff_t length
, int flags
)
3268 struct inode
*inode
= mapping
->host
;
3272 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3273 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3277 err
= ext4_discard_partial_page_buffers_no_lock(handle
, inode
, page
,
3278 from
, length
, flags
);
3281 page_cache_release(page
);
3286 * ext4_discard_partial_page_buffers_no_lock()
3287 * Zeros a page range of length 'length' starting from offset 'from'.
3288 * Buffer heads that correspond to the block aligned regions of the
3289 * zeroed range will be unmapped. Unblock aligned regions
3290 * will have the corresponding buffer head mapped if needed so that
3291 * that region of the page can be updated with the partial zero out.
3293 * This function assumes that the page has already been locked. The
3294 * The range to be discarded must be contained with in the given page.
3295 * If the specified range exceeds the end of the page it will be shortened
3296 * to the end of the page that corresponds to 'from'. This function is
3297 * appropriate for updating a page and it buffer heads to be unmapped and
3298 * zeroed for blocks that have been either released, or are going to be
3301 * handle: The journal handle
3302 * inode: The files inode
3303 * page: A locked page that contains the offset "from"
3304 * from: The starting byte offset (from the beginning of the file)
3305 * to begin discarding
3306 * len: The length of bytes to discard
3307 * flags: Optional flags that may be used:
3309 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3310 * Only zero the regions of the page whose buffer heads
3311 * have already been unmapped. This flag is appropriate
3312 * for updating the contents of a page whose blocks may
3313 * have already been released, and we only want to zero
3314 * out the regions that correspond to those released blocks.
3316 * Returns zero on success or negative on failure.
3318 static int ext4_discard_partial_page_buffers_no_lock(handle_t
*handle
,
3319 struct inode
*inode
, struct page
*page
, loff_t from
,
3320 loff_t length
, int flags
)
3322 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3323 unsigned int offset
= from
& (PAGE_CACHE_SIZE
-1);
3324 unsigned int blocksize
, max
, pos
;
3326 struct buffer_head
*bh
;
3329 blocksize
= inode
->i_sb
->s_blocksize
;
3330 max
= PAGE_CACHE_SIZE
- offset
;
3332 if (index
!= page
->index
)
3336 * correct length if it does not fall between
3337 * 'from' and the end of the page
3339 if (length
> max
|| length
< 0)
3342 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3344 if (!page_has_buffers(page
))
3345 create_empty_buffers(page
, blocksize
, 0);
3347 /* Find the buffer that contains "offset" */
3348 bh
= page_buffers(page
);
3350 while (offset
>= pos
) {
3351 bh
= bh
->b_this_page
;
3357 while (pos
< offset
+ length
) {
3358 unsigned int end_of_block
, range_to_discard
;
3362 /* The length of space left to zero and unmap */
3363 range_to_discard
= offset
+ length
- pos
;
3365 /* The length of space until the end of the block */
3366 end_of_block
= blocksize
- (pos
& (blocksize
-1));
3369 * Do not unmap or zero past end of block
3370 * for this buffer head
3372 if (range_to_discard
> end_of_block
)
3373 range_to_discard
= end_of_block
;
3377 * Skip this buffer head if we are only zeroing unampped
3378 * regions of the page
3380 if (flags
& EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
&&
3384 /* If the range is block aligned, unmap */
3385 if (range_to_discard
== blocksize
) {
3386 clear_buffer_dirty(bh
);
3388 clear_buffer_mapped(bh
);
3389 clear_buffer_req(bh
);
3390 clear_buffer_new(bh
);
3391 clear_buffer_delay(bh
);
3392 clear_buffer_unwritten(bh
);
3393 clear_buffer_uptodate(bh
);
3394 zero_user(page
, pos
, range_to_discard
);
3395 BUFFER_TRACE(bh
, "Buffer discarded");
3400 * If this block is not completely contained in the range
3401 * to be discarded, then it is not going to be released. Because
3402 * we need to keep this block, we need to make sure this part
3403 * of the page is uptodate before we modify it by writeing
3404 * partial zeros on it.
3406 if (!buffer_mapped(bh
)) {
3408 * Buffer head must be mapped before we can read
3411 BUFFER_TRACE(bh
, "unmapped");
3412 ext4_get_block(inode
, iblock
, bh
, 0);
3413 /* unmapped? It's a hole - nothing to do */
3414 if (!buffer_mapped(bh
)) {
3415 BUFFER_TRACE(bh
, "still unmapped");
3420 /* Ok, it's mapped. Make sure it's up-to-date */
3421 if (PageUptodate(page
))
3422 set_buffer_uptodate(bh
);
3424 if (!buffer_uptodate(bh
)) {
3426 ll_rw_block(READ
, 1, &bh
);
3428 /* Uhhuh. Read error. Complain and punt.*/
3429 if (!buffer_uptodate(bh
))
3433 if (ext4_should_journal_data(inode
)) {
3434 BUFFER_TRACE(bh
, "get write access");
3435 err
= ext4_journal_get_write_access(handle
, bh
);
3440 zero_user(page
, pos
, range_to_discard
);
3443 if (ext4_should_journal_data(inode
)) {
3444 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3446 mark_buffer_dirty(bh
);
3448 BUFFER_TRACE(bh
, "Partial buffer zeroed");
3450 bh
= bh
->b_this_page
;
3452 pos
+= range_to_discard
;
3458 int ext4_can_truncate(struct inode
*inode
)
3460 if (S_ISREG(inode
->i_mode
))
3462 if (S_ISDIR(inode
->i_mode
))
3464 if (S_ISLNK(inode
->i_mode
))
3465 return !ext4_inode_is_fast_symlink(inode
);
3470 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3471 * associated with the given offset and length
3473 * @inode: File inode
3474 * @offset: The offset where the hole will begin
3475 * @len: The length of the hole
3477 * Returns: 0 on success or negative on failure
3480 int ext4_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
3482 struct inode
*inode
= file_inode(file
);
3483 if (!S_ISREG(inode
->i_mode
))
3486 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3487 return ext4_ind_punch_hole(file
, offset
, length
);
3489 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) {
3490 /* TODO: Add support for bigalloc file systems */
3494 trace_ext4_punch_hole(inode
, offset
, length
);
3496 return ext4_ext_punch_hole(file
, offset
, length
);
3502 * We block out ext4_get_block() block instantiations across the entire
3503 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3504 * simultaneously on behalf of the same inode.
3506 * As we work through the truncate and commit bits of it to the journal there
3507 * is one core, guiding principle: the file's tree must always be consistent on
3508 * disk. We must be able to restart the truncate after a crash.
3510 * The file's tree may be transiently inconsistent in memory (although it
3511 * probably isn't), but whenever we close off and commit a journal transaction,
3512 * the contents of (the filesystem + the journal) must be consistent and
3513 * restartable. It's pretty simple, really: bottom up, right to left (although
3514 * left-to-right works OK too).
3516 * Note that at recovery time, journal replay occurs *before* the restart of
3517 * truncate against the orphan inode list.
3519 * The committed inode has the new, desired i_size (which is the same as
3520 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3521 * that this inode's truncate did not complete and it will again call
3522 * ext4_truncate() to have another go. So there will be instantiated blocks
3523 * to the right of the truncation point in a crashed ext4 filesystem. But
3524 * that's fine - as long as they are linked from the inode, the post-crash
3525 * ext4_truncate() run will find them and release them.
3527 void ext4_truncate(struct inode
*inode
)
3529 trace_ext4_truncate_enter(inode
);
3531 if (!ext4_can_truncate(inode
))
3534 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3536 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3537 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3539 if (ext4_has_inline_data(inode
)) {
3542 ext4_inline_data_truncate(inode
, &has_inline
);
3547 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3548 ext4_ext_truncate(inode
);
3550 ext4_ind_truncate(inode
);
3552 trace_ext4_truncate_exit(inode
);
3556 * ext4_get_inode_loc returns with an extra refcount against the inode's
3557 * underlying buffer_head on success. If 'in_mem' is true, we have all
3558 * data in memory that is needed to recreate the on-disk version of this
3561 static int __ext4_get_inode_loc(struct inode
*inode
,
3562 struct ext4_iloc
*iloc
, int in_mem
)
3564 struct ext4_group_desc
*gdp
;
3565 struct buffer_head
*bh
;
3566 struct super_block
*sb
= inode
->i_sb
;
3568 int inodes_per_block
, inode_offset
;
3571 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3574 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3575 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3580 * Figure out the offset within the block group inode table
3582 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3583 inode_offset
= ((inode
->i_ino
- 1) %
3584 EXT4_INODES_PER_GROUP(sb
));
3585 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3586 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3588 bh
= sb_getblk(sb
, block
);
3591 if (!buffer_uptodate(bh
)) {
3595 * If the buffer has the write error flag, we have failed
3596 * to write out another inode in the same block. In this
3597 * case, we don't have to read the block because we may
3598 * read the old inode data successfully.
3600 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3601 set_buffer_uptodate(bh
);
3603 if (buffer_uptodate(bh
)) {
3604 /* someone brought it uptodate while we waited */
3610 * If we have all information of the inode in memory and this
3611 * is the only valid inode in the block, we need not read the
3615 struct buffer_head
*bitmap_bh
;
3618 start
= inode_offset
& ~(inodes_per_block
- 1);
3620 /* Is the inode bitmap in cache? */
3621 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3622 if (unlikely(!bitmap_bh
))
3626 * If the inode bitmap isn't in cache then the
3627 * optimisation may end up performing two reads instead
3628 * of one, so skip it.
3630 if (!buffer_uptodate(bitmap_bh
)) {
3634 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3635 if (i
== inode_offset
)
3637 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3641 if (i
== start
+ inodes_per_block
) {
3642 /* all other inodes are free, so skip I/O */
3643 memset(bh
->b_data
, 0, bh
->b_size
);
3644 set_buffer_uptodate(bh
);
3652 * If we need to do any I/O, try to pre-readahead extra
3653 * blocks from the inode table.
3655 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3656 ext4_fsblk_t b
, end
, table
;
3659 table
= ext4_inode_table(sb
, gdp
);
3660 /* s_inode_readahead_blks is always a power of 2 */
3661 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
3664 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
3665 num
= EXT4_INODES_PER_GROUP(sb
);
3666 if (ext4_has_group_desc_csum(sb
))
3667 num
-= ext4_itable_unused_count(sb
, gdp
);
3668 table
+= num
/ inodes_per_block
;
3672 sb_breadahead(sb
, b
++);
3676 * There are other valid inodes in the buffer, this inode
3677 * has in-inode xattrs, or we don't have this inode in memory.
3678 * Read the block from disk.
3680 trace_ext4_load_inode(inode
);
3682 bh
->b_end_io
= end_buffer_read_sync
;
3683 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
3685 if (!buffer_uptodate(bh
)) {
3686 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3687 "unable to read itable block");
3697 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3699 /* We have all inode data except xattrs in memory here. */
3700 return __ext4_get_inode_loc(inode
, iloc
,
3701 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
3704 void ext4_set_inode_flags(struct inode
*inode
)
3706 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3708 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3709 if (flags
& EXT4_SYNC_FL
)
3710 inode
->i_flags
|= S_SYNC
;
3711 if (flags
& EXT4_APPEND_FL
)
3712 inode
->i_flags
|= S_APPEND
;
3713 if (flags
& EXT4_IMMUTABLE_FL
)
3714 inode
->i_flags
|= S_IMMUTABLE
;
3715 if (flags
& EXT4_NOATIME_FL
)
3716 inode
->i_flags
|= S_NOATIME
;
3717 if (flags
& EXT4_DIRSYNC_FL
)
3718 inode
->i_flags
|= S_DIRSYNC
;
3721 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3722 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3724 unsigned int vfs_fl
;
3725 unsigned long old_fl
, new_fl
;
3728 vfs_fl
= ei
->vfs_inode
.i_flags
;
3729 old_fl
= ei
->i_flags
;
3730 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3731 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
3733 if (vfs_fl
& S_SYNC
)
3734 new_fl
|= EXT4_SYNC_FL
;
3735 if (vfs_fl
& S_APPEND
)
3736 new_fl
|= EXT4_APPEND_FL
;
3737 if (vfs_fl
& S_IMMUTABLE
)
3738 new_fl
|= EXT4_IMMUTABLE_FL
;
3739 if (vfs_fl
& S_NOATIME
)
3740 new_fl
|= EXT4_NOATIME_FL
;
3741 if (vfs_fl
& S_DIRSYNC
)
3742 new_fl
|= EXT4_DIRSYNC_FL
;
3743 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
3746 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3747 struct ext4_inode_info
*ei
)
3750 struct inode
*inode
= &(ei
->vfs_inode
);
3751 struct super_block
*sb
= inode
->i_sb
;
3753 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3754 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3755 /* we are using combined 48 bit field */
3756 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
3757 le32_to_cpu(raw_inode
->i_blocks_lo
);
3758 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
3759 /* i_blocks represent file system block size */
3760 return i_blocks
<< (inode
->i_blkbits
- 9);
3765 return le32_to_cpu(raw_inode
->i_blocks_lo
);
3769 static inline void ext4_iget_extra_inode(struct inode
*inode
,
3770 struct ext4_inode
*raw_inode
,
3771 struct ext4_inode_info
*ei
)
3773 __le32
*magic
= (void *)raw_inode
+
3774 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
3775 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
3776 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
3777 ext4_find_inline_data_nolock(inode
);
3779 EXT4_I(inode
)->i_inline_off
= 0;
3782 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
3784 struct ext4_iloc iloc
;
3785 struct ext4_inode
*raw_inode
;
3786 struct ext4_inode_info
*ei
;
3787 struct inode
*inode
;
3788 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
3794 inode
= iget_locked(sb
, ino
);
3796 return ERR_PTR(-ENOMEM
);
3797 if (!(inode
->i_state
& I_NEW
))
3803 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
3806 raw_inode
= ext4_raw_inode(&iloc
);
3808 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3809 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
3810 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
3811 EXT4_INODE_SIZE(inode
->i_sb
)) {
3812 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
3813 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
3814 EXT4_INODE_SIZE(inode
->i_sb
));
3819 ei
->i_extra_isize
= 0;
3821 /* Precompute checksum seed for inode metadata */
3822 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3823 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
)) {
3824 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3826 __le32 inum
= cpu_to_le32(inode
->i_ino
);
3827 __le32 gen
= raw_inode
->i_generation
;
3828 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
3830 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
3834 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
3835 EXT4_ERROR_INODE(inode
, "checksum invalid");
3840 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
3841 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
3842 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
3843 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
3844 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
3845 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
3847 i_uid_write(inode
, i_uid
);
3848 i_gid_write(inode
, i_gid
);
3849 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
3851 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
3852 ei
->i_inline_off
= 0;
3853 ei
->i_dir_start_lookup
= 0;
3854 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
3855 /* We now have enough fields to check if the inode was active or not.
3856 * This is needed because nfsd might try to access dead inodes
3857 * the test is that same one that e2fsck uses
3858 * NeilBrown 1999oct15
3860 if (inode
->i_nlink
== 0) {
3861 if (inode
->i_mode
== 0 ||
3862 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
3863 /* this inode is deleted */
3867 /* The only unlinked inodes we let through here have
3868 * valid i_mode and are being read by the orphan
3869 * recovery code: that's fine, we're about to complete
3870 * the process of deleting those. */
3872 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
3873 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
3874 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
3875 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
3877 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
3878 inode
->i_size
= ext4_isize(raw_inode
);
3879 ei
->i_disksize
= inode
->i_size
;
3881 ei
->i_reserved_quota
= 0;
3883 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
3884 ei
->i_block_group
= iloc
.block_group
;
3885 ei
->i_last_alloc_group
= ~0;
3887 * NOTE! The in-memory inode i_data array is in little-endian order
3888 * even on big-endian machines: we do NOT byteswap the block numbers!
3890 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
3891 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
3892 INIT_LIST_HEAD(&ei
->i_orphan
);
3895 * Set transaction id's of transactions that have to be committed
3896 * to finish f[data]sync. We set them to currently running transaction
3897 * as we cannot be sure that the inode or some of its metadata isn't
3898 * part of the transaction - the inode could have been reclaimed and
3899 * now it is reread from disk.
3902 transaction_t
*transaction
;
3905 read_lock(&journal
->j_state_lock
);
3906 if (journal
->j_running_transaction
)
3907 transaction
= journal
->j_running_transaction
;
3909 transaction
= journal
->j_committing_transaction
;
3911 tid
= transaction
->t_tid
;
3913 tid
= journal
->j_commit_sequence
;
3914 read_unlock(&journal
->j_state_lock
);
3915 ei
->i_sync_tid
= tid
;
3916 ei
->i_datasync_tid
= tid
;
3919 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3920 if (ei
->i_extra_isize
== 0) {
3921 /* The extra space is currently unused. Use it. */
3922 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
3923 EXT4_GOOD_OLD_INODE_SIZE
;
3925 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
3929 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
3930 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
3931 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
3932 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
3934 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
3935 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3936 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
3938 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
3942 if (ei
->i_file_acl
&&
3943 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
3944 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
3948 } else if (!ext4_has_inline_data(inode
)) {
3949 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
3950 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
3951 (S_ISLNK(inode
->i_mode
) &&
3952 !ext4_inode_is_fast_symlink(inode
))))
3953 /* Validate extent which is part of inode */
3954 ret
= ext4_ext_check_inode(inode
);
3955 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
3956 (S_ISLNK(inode
->i_mode
) &&
3957 !ext4_inode_is_fast_symlink(inode
))) {
3958 /* Validate block references which are part of inode */
3959 ret
= ext4_ind_check_inode(inode
);
3965 if (S_ISREG(inode
->i_mode
)) {
3966 inode
->i_op
= &ext4_file_inode_operations
;
3967 inode
->i_fop
= &ext4_file_operations
;
3968 ext4_set_aops(inode
);
3969 } else if (S_ISDIR(inode
->i_mode
)) {
3970 inode
->i_op
= &ext4_dir_inode_operations
;
3971 inode
->i_fop
= &ext4_dir_operations
;
3972 } else if (S_ISLNK(inode
->i_mode
)) {
3973 if (ext4_inode_is_fast_symlink(inode
)) {
3974 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
3975 nd_terminate_link(ei
->i_data
, inode
->i_size
,
3976 sizeof(ei
->i_data
) - 1);
3978 inode
->i_op
= &ext4_symlink_inode_operations
;
3979 ext4_set_aops(inode
);
3981 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
3982 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
3983 inode
->i_op
= &ext4_special_inode_operations
;
3984 if (raw_inode
->i_block
[0])
3985 init_special_inode(inode
, inode
->i_mode
,
3986 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
3988 init_special_inode(inode
, inode
->i_mode
,
3989 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
3992 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
3996 ext4_set_inode_flags(inode
);
3997 unlock_new_inode(inode
);
4003 return ERR_PTR(ret
);
4006 static int ext4_inode_blocks_set(handle_t
*handle
,
4007 struct ext4_inode
*raw_inode
,
4008 struct ext4_inode_info
*ei
)
4010 struct inode
*inode
= &(ei
->vfs_inode
);
4011 u64 i_blocks
= inode
->i_blocks
;
4012 struct super_block
*sb
= inode
->i_sb
;
4014 if (i_blocks
<= ~0U) {
4016 * i_blocks can be represented in a 32 bit variable
4017 * as multiple of 512 bytes
4019 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4020 raw_inode
->i_blocks_high
= 0;
4021 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4024 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4027 if (i_blocks
<= 0xffffffffffffULL
) {
4029 * i_blocks can be represented in a 48 bit variable
4030 * as multiple of 512 bytes
4032 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4033 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4034 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4036 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4037 /* i_block is stored in file system block size */
4038 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4039 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4040 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4046 * Post the struct inode info into an on-disk inode location in the
4047 * buffer-cache. This gobbles the caller's reference to the
4048 * buffer_head in the inode location struct.
4050 * The caller must have write access to iloc->bh.
4052 static int ext4_do_update_inode(handle_t
*handle
,
4053 struct inode
*inode
,
4054 struct ext4_iloc
*iloc
)
4056 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4057 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4058 struct buffer_head
*bh
= iloc
->bh
;
4059 int err
= 0, rc
, block
;
4060 int need_datasync
= 0;
4064 /* For fields not not tracking in the in-memory inode,
4065 * initialise them to zero for new inodes. */
4066 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4067 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4069 ext4_get_inode_flags(ei
);
4070 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4071 i_uid
= i_uid_read(inode
);
4072 i_gid
= i_gid_read(inode
);
4073 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4074 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4075 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4077 * Fix up interoperability with old kernels. Otherwise, old inodes get
4078 * re-used with the upper 16 bits of the uid/gid intact
4081 raw_inode
->i_uid_high
=
4082 cpu_to_le16(high_16_bits(i_uid
));
4083 raw_inode
->i_gid_high
=
4084 cpu_to_le16(high_16_bits(i_gid
));
4086 raw_inode
->i_uid_high
= 0;
4087 raw_inode
->i_gid_high
= 0;
4090 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4091 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4092 raw_inode
->i_uid_high
= 0;
4093 raw_inode
->i_gid_high
= 0;
4095 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4097 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4098 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4099 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4100 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4102 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4104 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4105 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4106 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4107 cpu_to_le32(EXT4_OS_HURD
))
4108 raw_inode
->i_file_acl_high
=
4109 cpu_to_le16(ei
->i_file_acl
>> 32);
4110 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4111 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4112 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4115 if (ei
->i_disksize
> 0x7fffffffULL
) {
4116 struct super_block
*sb
= inode
->i_sb
;
4117 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4118 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4119 EXT4_SB(sb
)->s_es
->s_rev_level
==
4120 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4121 /* If this is the first large file
4122 * created, add a flag to the superblock.
4124 err
= ext4_journal_get_write_access(handle
,
4125 EXT4_SB(sb
)->s_sbh
);
4128 ext4_update_dynamic_rev(sb
);
4129 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4130 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4131 ext4_handle_sync(handle
);
4132 err
= ext4_handle_dirty_super(handle
, sb
);
4135 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4136 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4137 if (old_valid_dev(inode
->i_rdev
)) {
4138 raw_inode
->i_block
[0] =
4139 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4140 raw_inode
->i_block
[1] = 0;
4142 raw_inode
->i_block
[0] = 0;
4143 raw_inode
->i_block
[1] =
4144 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4145 raw_inode
->i_block
[2] = 0;
4147 } else if (!ext4_has_inline_data(inode
)) {
4148 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4149 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4152 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4153 if (ei
->i_extra_isize
) {
4154 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4155 raw_inode
->i_version_hi
=
4156 cpu_to_le32(inode
->i_version
>> 32);
4157 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4160 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4162 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4163 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4166 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4168 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4171 ext4_std_error(inode
->i_sb
, err
);
4176 * ext4_write_inode()
4178 * We are called from a few places:
4180 * - Within generic_file_write() for O_SYNC files.
4181 * Here, there will be no transaction running. We wait for any running
4182 * transaction to commit.
4184 * - Within sys_sync(), kupdate and such.
4185 * We wait on commit, if tol to.
4187 * - Within prune_icache() (PF_MEMALLOC == true)
4188 * Here we simply return. We can't afford to block kswapd on the
4191 * In all cases it is actually safe for us to return without doing anything,
4192 * because the inode has been copied into a raw inode buffer in
4193 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4196 * Note that we are absolutely dependent upon all inode dirtiers doing the
4197 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4198 * which we are interested.
4200 * It would be a bug for them to not do this. The code:
4202 * mark_inode_dirty(inode)
4204 * inode->i_size = expr;
4206 * is in error because a kswapd-driven write_inode() could occur while
4207 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4208 * will no longer be on the superblock's dirty inode list.
4210 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4214 if (current
->flags
& PF_MEMALLOC
)
4217 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4218 if (ext4_journal_current_handle()) {
4219 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4224 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
4227 err
= ext4_force_commit(inode
->i_sb
);
4229 struct ext4_iloc iloc
;
4231 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4234 if (wbc
->sync_mode
== WB_SYNC_ALL
)
4235 sync_dirty_buffer(iloc
.bh
);
4236 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4237 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4238 "IO error syncing inode");
4247 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4248 * buffers that are attached to a page stradding i_size and are undergoing
4249 * commit. In that case we have to wait for commit to finish and try again.
4251 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4255 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4256 tid_t commit_tid
= 0;
4259 offset
= inode
->i_size
& (PAGE_CACHE_SIZE
- 1);
4261 * All buffers in the last page remain valid? Then there's nothing to
4262 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4265 if (offset
> PAGE_CACHE_SIZE
- (1 << inode
->i_blkbits
))
4268 page
= find_lock_page(inode
->i_mapping
,
4269 inode
->i_size
>> PAGE_CACHE_SHIFT
);
4272 ret
= __ext4_journalled_invalidatepage(page
, offset
);
4274 page_cache_release(page
);
4278 read_lock(&journal
->j_state_lock
);
4279 if (journal
->j_committing_transaction
)
4280 commit_tid
= journal
->j_committing_transaction
->t_tid
;
4281 read_unlock(&journal
->j_state_lock
);
4283 jbd2_log_wait_commit(journal
, commit_tid
);
4290 * Called from notify_change.
4292 * We want to trap VFS attempts to truncate the file as soon as
4293 * possible. In particular, we want to make sure that when the VFS
4294 * shrinks i_size, we put the inode on the orphan list and modify
4295 * i_disksize immediately, so that during the subsequent flushing of
4296 * dirty pages and freeing of disk blocks, we can guarantee that any
4297 * commit will leave the blocks being flushed in an unused state on
4298 * disk. (On recovery, the inode will get truncated and the blocks will
4299 * be freed, so we have a strong guarantee that no future commit will
4300 * leave these blocks visible to the user.)
4302 * Another thing we have to assure is that if we are in ordered mode
4303 * and inode is still attached to the committing transaction, we must
4304 * we start writeout of all the dirty pages which are being truncated.
4305 * This way we are sure that all the data written in the previous
4306 * transaction are already on disk (truncate waits for pages under
4309 * Called with inode->i_mutex down.
4311 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4313 struct inode
*inode
= dentry
->d_inode
;
4316 const unsigned int ia_valid
= attr
->ia_valid
;
4318 error
= inode_change_ok(inode
, attr
);
4322 if (is_quota_modification(inode
, attr
))
4323 dquot_initialize(inode
);
4324 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4325 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4328 /* (user+group)*(old+new) structure, inode write (sb,
4329 * inode block, ? - but truncate inode update has it) */
4330 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
4331 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
4332 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
4333 if (IS_ERR(handle
)) {
4334 error
= PTR_ERR(handle
);
4337 error
= dquot_transfer(inode
, attr
);
4339 ext4_journal_stop(handle
);
4342 /* Update corresponding info in inode so that everything is in
4343 * one transaction */
4344 if (attr
->ia_valid
& ATTR_UID
)
4345 inode
->i_uid
= attr
->ia_uid
;
4346 if (attr
->ia_valid
& ATTR_GID
)
4347 inode
->i_gid
= attr
->ia_gid
;
4348 error
= ext4_mark_inode_dirty(handle
, inode
);
4349 ext4_journal_stop(handle
);
4352 if (attr
->ia_valid
& ATTR_SIZE
) {
4354 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4355 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4357 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4362 if (S_ISREG(inode
->i_mode
) &&
4363 attr
->ia_valid
& ATTR_SIZE
&&
4364 (attr
->ia_size
< inode
->i_size
)) {
4367 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
4368 if (IS_ERR(handle
)) {
4369 error
= PTR_ERR(handle
);
4372 if (ext4_handle_valid(handle
)) {
4373 error
= ext4_orphan_add(handle
, inode
);
4376 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4377 rc
= ext4_mark_inode_dirty(handle
, inode
);
4380 ext4_journal_stop(handle
);
4382 if (ext4_should_order_data(inode
)) {
4383 error
= ext4_begin_ordered_truncate(inode
,
4386 /* Do as much error cleanup as possible */
4387 handle
= ext4_journal_start(inode
,
4389 if (IS_ERR(handle
)) {
4390 ext4_orphan_del(NULL
, inode
);
4393 ext4_orphan_del(handle
, inode
);
4395 ext4_journal_stop(handle
);
4401 if (attr
->ia_valid
& ATTR_SIZE
) {
4402 if (attr
->ia_size
!= inode
->i_size
) {
4403 loff_t oldsize
= inode
->i_size
;
4405 i_size_write(inode
, attr
->ia_size
);
4407 * Blocks are going to be removed from the inode. Wait
4408 * for dio in flight. Temporarily disable
4409 * dioread_nolock to prevent livelock.
4412 if (!ext4_should_journal_data(inode
)) {
4413 ext4_inode_block_unlocked_dio(inode
);
4414 inode_dio_wait(inode
);
4415 ext4_inode_resume_unlocked_dio(inode
);
4417 ext4_wait_for_tail_page_commit(inode
);
4420 * Truncate pagecache after we've waited for commit
4421 * in data=journal mode to make pages freeable.
4423 truncate_pagecache(inode
, oldsize
, inode
->i_size
);
4425 ext4_truncate(inode
);
4429 setattr_copy(inode
, attr
);
4430 mark_inode_dirty(inode
);
4434 * If the call to ext4_truncate failed to get a transaction handle at
4435 * all, we need to clean up the in-core orphan list manually.
4437 if (orphan
&& inode
->i_nlink
)
4438 ext4_orphan_del(NULL
, inode
);
4440 if (!rc
&& (ia_valid
& ATTR_MODE
))
4441 rc
= ext4_acl_chmod(inode
);
4444 ext4_std_error(inode
->i_sb
, error
);
4450 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4453 struct inode
*inode
;
4454 unsigned long delalloc_blocks
;
4456 inode
= dentry
->d_inode
;
4457 generic_fillattr(inode
, stat
);
4460 * We can't update i_blocks if the block allocation is delayed
4461 * otherwise in the case of system crash before the real block
4462 * allocation is done, we will have i_blocks inconsistent with
4463 * on-disk file blocks.
4464 * We always keep i_blocks updated together with real
4465 * allocation. But to not confuse with user, stat
4466 * will return the blocks that include the delayed allocation
4467 * blocks for this file.
4469 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
4470 EXT4_I(inode
)->i_reserved_data_blocks
);
4472 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4476 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4478 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4479 return ext4_ind_trans_blocks(inode
, nrblocks
, chunk
);
4480 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4484 * Account for index blocks, block groups bitmaps and block group
4485 * descriptor blocks if modify datablocks and index blocks
4486 * worse case, the indexs blocks spread over different block groups
4488 * If datablocks are discontiguous, they are possible to spread over
4489 * different block groups too. If they are contiguous, with flexbg,
4490 * they could still across block group boundary.
4492 * Also account for superblock, inode, quota and xattr blocks
4494 static int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4496 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4502 * How many index blocks need to touch to modify nrblocks?
4503 * The "Chunk" flag indicating whether the nrblocks is
4504 * physically contiguous on disk
4506 * For Direct IO and fallocate, they calls get_block to allocate
4507 * one single extent at a time, so they could set the "Chunk" flag
4509 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4514 * Now let's see how many group bitmaps and group descriptors need
4524 if (groups
> ngroups
)
4526 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4527 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4529 /* bitmaps and block group descriptor blocks */
4530 ret
+= groups
+ gdpblocks
;
4532 /* Blocks for super block, inode, quota and xattr blocks */
4533 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4539 * Calculate the total number of credits to reserve to fit
4540 * the modification of a single pages into a single transaction,
4541 * which may include multiple chunks of block allocations.
4543 * This could be called via ext4_write_begin()
4545 * We need to consider the worse case, when
4546 * one new block per extent.
4548 int ext4_writepage_trans_blocks(struct inode
*inode
)
4550 int bpp
= ext4_journal_blocks_per_page(inode
);
4553 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4555 /* Account for data blocks for journalled mode */
4556 if (ext4_should_journal_data(inode
))
4562 * Calculate the journal credits for a chunk of data modification.
4564 * This is called from DIO, fallocate or whoever calling
4565 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4567 * journal buffers for data blocks are not included here, as DIO
4568 * and fallocate do no need to journal data buffers.
4570 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4572 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4576 * The caller must have previously called ext4_reserve_inode_write().
4577 * Give this, we know that the caller already has write access to iloc->bh.
4579 int ext4_mark_iloc_dirty(handle_t
*handle
,
4580 struct inode
*inode
, struct ext4_iloc
*iloc
)
4584 if (IS_I_VERSION(inode
))
4585 inode_inc_iversion(inode
);
4587 /* the do_update_inode consumes one bh->b_count */
4590 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4591 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4597 * On success, We end up with an outstanding reference count against
4598 * iloc->bh. This _must_ be cleaned up later.
4602 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4603 struct ext4_iloc
*iloc
)
4607 err
= ext4_get_inode_loc(inode
, iloc
);
4609 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4610 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4616 ext4_std_error(inode
->i_sb
, err
);
4621 * Expand an inode by new_extra_isize bytes.
4622 * Returns 0 on success or negative error number on failure.
4624 static int ext4_expand_extra_isize(struct inode
*inode
,
4625 unsigned int new_extra_isize
,
4626 struct ext4_iloc iloc
,
4629 struct ext4_inode
*raw_inode
;
4630 struct ext4_xattr_ibody_header
*header
;
4632 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4635 raw_inode
= ext4_raw_inode(&iloc
);
4637 header
= IHDR(inode
, raw_inode
);
4639 /* No extended attributes present */
4640 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
4641 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4642 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4644 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4648 /* try to expand with EAs present */
4649 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4654 * What we do here is to mark the in-core inode as clean with respect to inode
4655 * dirtiness (it may still be data-dirty).
4656 * This means that the in-core inode may be reaped by prune_icache
4657 * without having to perform any I/O. This is a very good thing,
4658 * because *any* task may call prune_icache - even ones which
4659 * have a transaction open against a different journal.
4661 * Is this cheating? Not really. Sure, we haven't written the
4662 * inode out, but prune_icache isn't a user-visible syncing function.
4663 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4664 * we start and wait on commits.
4666 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4668 struct ext4_iloc iloc
;
4669 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4670 static unsigned int mnt_count
;
4674 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
4675 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4676 if (ext4_handle_valid(handle
) &&
4677 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4678 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
4680 * We need extra buffer credits since we may write into EA block
4681 * with this same handle. If journal_extend fails, then it will
4682 * only result in a minor loss of functionality for that inode.
4683 * If this is felt to be critical, then e2fsck should be run to
4684 * force a large enough s_min_extra_isize.
4686 if ((jbd2_journal_extend(handle
,
4687 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4688 ret
= ext4_expand_extra_isize(inode
,
4689 sbi
->s_want_extra_isize
,
4692 ext4_set_inode_state(inode
,
4693 EXT4_STATE_NO_EXPAND
);
4695 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4696 ext4_warning(inode
->i_sb
,
4697 "Unable to expand inode %lu. Delete"
4698 " some EAs or run e2fsck.",
4701 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4707 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4712 * ext4_dirty_inode() is called from __mark_inode_dirty()
4714 * We're really interested in the case where a file is being extended.
4715 * i_size has been changed by generic_commit_write() and we thus need
4716 * to include the updated inode in the current transaction.
4718 * Also, dquot_alloc_block() will always dirty the inode when blocks
4719 * are allocated to the file.
4721 * If the inode is marked synchronous, we don't honour that here - doing
4722 * so would cause a commit on atime updates, which we don't bother doing.
4723 * We handle synchronous inodes at the highest possible level.
4725 void ext4_dirty_inode(struct inode
*inode
, int flags
)
4729 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
4733 ext4_mark_inode_dirty(handle
, inode
);
4735 ext4_journal_stop(handle
);
4742 * Bind an inode's backing buffer_head into this transaction, to prevent
4743 * it from being flushed to disk early. Unlike
4744 * ext4_reserve_inode_write, this leaves behind no bh reference and
4745 * returns no iloc structure, so the caller needs to repeat the iloc
4746 * lookup to mark the inode dirty later.
4748 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4750 struct ext4_iloc iloc
;
4754 err
= ext4_get_inode_loc(inode
, &iloc
);
4756 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4757 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4759 err
= ext4_handle_dirty_metadata(handle
,
4765 ext4_std_error(inode
->i_sb
, err
);
4770 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4777 * We have to be very careful here: changing a data block's
4778 * journaling status dynamically is dangerous. If we write a
4779 * data block to the journal, change the status and then delete
4780 * that block, we risk forgetting to revoke the old log record
4781 * from the journal and so a subsequent replay can corrupt data.
4782 * So, first we make sure that the journal is empty and that
4783 * nobody is changing anything.
4786 journal
= EXT4_JOURNAL(inode
);
4789 if (is_journal_aborted(journal
))
4791 /* We have to allocate physical blocks for delalloc blocks
4792 * before flushing journal. otherwise delalloc blocks can not
4793 * be allocated any more. even more truncate on delalloc blocks
4794 * could trigger BUG by flushing delalloc blocks in journal.
4795 * There is no delalloc block in non-journal data mode.
4797 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
4798 err
= ext4_alloc_da_blocks(inode
);
4803 /* Wait for all existing dio workers */
4804 ext4_inode_block_unlocked_dio(inode
);
4805 inode_dio_wait(inode
);
4807 jbd2_journal_lock_updates(journal
);
4810 * OK, there are no updates running now, and all cached data is
4811 * synced to disk. We are now in a completely consistent state
4812 * which doesn't have anything in the journal, and we know that
4813 * no filesystem updates are running, so it is safe to modify
4814 * the inode's in-core data-journaling state flag now.
4818 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4820 jbd2_journal_flush(journal
);
4821 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4823 ext4_set_aops(inode
);
4825 jbd2_journal_unlock_updates(journal
);
4826 ext4_inode_resume_unlocked_dio(inode
);
4828 /* Finally we can mark the inode as dirty. */
4830 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
4832 return PTR_ERR(handle
);
4834 err
= ext4_mark_inode_dirty(handle
, inode
);
4835 ext4_handle_sync(handle
);
4836 ext4_journal_stop(handle
);
4837 ext4_std_error(inode
->i_sb
, err
);
4842 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4844 return !buffer_mapped(bh
);
4847 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
4849 struct page
*page
= vmf
->page
;
4853 struct file
*file
= vma
->vm_file
;
4854 struct inode
*inode
= file_inode(file
);
4855 struct address_space
*mapping
= inode
->i_mapping
;
4857 get_block_t
*get_block
;
4860 sb_start_pagefault(inode
->i_sb
);
4861 file_update_time(vma
->vm_file
);
4862 /* Delalloc case is easy... */
4863 if (test_opt(inode
->i_sb
, DELALLOC
) &&
4864 !ext4_should_journal_data(inode
) &&
4865 !ext4_nonda_switch(inode
->i_sb
)) {
4867 ret
= __block_page_mkwrite(vma
, vmf
,
4868 ext4_da_get_block_prep
);
4869 } while (ret
== -ENOSPC
&&
4870 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
4875 size
= i_size_read(inode
);
4876 /* Page got truncated from under us? */
4877 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
4879 ret
= VM_FAULT_NOPAGE
;
4883 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4884 len
= size
& ~PAGE_CACHE_MASK
;
4886 len
= PAGE_CACHE_SIZE
;
4888 * Return if we have all the buffers mapped. This avoids the need to do
4889 * journal_start/journal_stop which can block and take a long time
4891 if (page_has_buffers(page
)) {
4892 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
4894 ext4_bh_unmapped
)) {
4895 /* Wait so that we don't change page under IO */
4896 wait_for_stable_page(page
);
4897 ret
= VM_FAULT_LOCKED
;
4902 /* OK, we need to fill the hole... */
4903 if (ext4_should_dioread_nolock(inode
))
4904 get_block
= ext4_get_block_write
;
4906 get_block
= ext4_get_block
;
4908 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
4909 ext4_writepage_trans_blocks(inode
));
4910 if (IS_ERR(handle
)) {
4911 ret
= VM_FAULT_SIGBUS
;
4914 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
4915 if (!ret
&& ext4_should_journal_data(inode
)) {
4916 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
4917 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
4919 ret
= VM_FAULT_SIGBUS
;
4920 ext4_journal_stop(handle
);
4923 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
4925 ext4_journal_stop(handle
);
4926 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
4929 ret
= block_page_mkwrite_return(ret
);
4931 sb_end_pagefault(inode
->i_sb
);