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>
40 #include <linux/aio.h>
42 #include "ext4_jbd2.h"
47 #include <trace/events/ext4.h>
49 #define MPAGE_DA_EXTENT_TAIL 0x01
51 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
52 struct ext4_inode_info
*ei
)
54 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
59 csum_lo
= le16_to_cpu(raw
->i_checksum_lo
);
60 raw
->i_checksum_lo
= 0;
61 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
62 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
63 csum_hi
= le16_to_cpu(raw
->i_checksum_hi
);
64 raw
->i_checksum_hi
= 0;
67 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
,
68 EXT4_INODE_SIZE(inode
->i_sb
));
70 raw
->i_checksum_lo
= cpu_to_le16(csum_lo
);
71 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
72 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
73 raw
->i_checksum_hi
= cpu_to_le16(csum_hi
);
78 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
79 struct ext4_inode_info
*ei
)
81 __u32 provided
, calculated
;
83 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
84 cpu_to_le32(EXT4_OS_LINUX
) ||
85 !EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
86 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
89 provided
= le16_to_cpu(raw
->i_checksum_lo
);
90 calculated
= ext4_inode_csum(inode
, raw
, ei
);
91 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
92 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
93 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
97 return provided
== calculated
;
100 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
101 struct ext4_inode_info
*ei
)
105 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
106 cpu_to_le32(EXT4_OS_LINUX
) ||
107 !EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
108 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
111 csum
= ext4_inode_csum(inode
, raw
, ei
);
112 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
113 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
114 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
115 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
118 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
121 trace_ext4_begin_ordered_truncate(inode
, new_size
);
123 * If jinode is zero, then we never opened the file for
124 * writing, so there's no need to call
125 * jbd2_journal_begin_ordered_truncate() since there's no
126 * outstanding writes we need to flush.
128 if (!EXT4_I(inode
)->jinode
)
130 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
131 EXT4_I(inode
)->jinode
,
135 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
136 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
137 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
138 static int ext4_discard_partial_page_buffers_no_lock(handle_t
*handle
,
139 struct inode
*inode
, struct page
*page
, loff_t from
,
140 loff_t length
, int flags
);
143 * Test whether an inode is a fast symlink.
145 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
147 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
148 (inode
->i_sb
->s_blocksize
>> 9) : 0;
150 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
154 * Restart the transaction associated with *handle. This does a commit,
155 * so before we call here everything must be consistently dirtied against
158 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
164 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
165 * moment, get_block can be called only for blocks inside i_size since
166 * page cache has been already dropped and writes are blocked by
167 * i_mutex. So we can safely drop the i_data_sem here.
169 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
170 jbd_debug(2, "restarting handle %p\n", handle
);
171 up_write(&EXT4_I(inode
)->i_data_sem
);
172 ret
= ext4_journal_restart(handle
, nblocks
);
173 down_write(&EXT4_I(inode
)->i_data_sem
);
174 ext4_discard_preallocations(inode
);
180 * Called at the last iput() if i_nlink is zero.
182 void ext4_evict_inode(struct inode
*inode
)
187 trace_ext4_evict_inode(inode
);
189 if (inode
->i_nlink
) {
191 * When journalling data dirty buffers are tracked only in the
192 * journal. So although mm thinks everything is clean and
193 * ready for reaping the inode might still have some pages to
194 * write in the running transaction or waiting to be
195 * checkpointed. Thus calling jbd2_journal_invalidatepage()
196 * (via truncate_inode_pages()) to discard these buffers can
197 * cause data loss. Also even if we did not discard these
198 * buffers, we would have no way to find them after the inode
199 * is reaped and thus user could see stale data if he tries to
200 * read them before the transaction is checkpointed. So be
201 * careful and force everything to disk here... We use
202 * ei->i_datasync_tid to store the newest transaction
203 * containing inode's data.
205 * Note that directories do not have this problem because they
206 * don't use page cache.
208 if (ext4_should_journal_data(inode
) &&
209 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
)) &&
210 inode
->i_ino
!= EXT4_JOURNAL_INO
) {
211 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
212 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
214 jbd2_complete_transaction(journal
, commit_tid
);
215 filemap_write_and_wait(&inode
->i_data
);
217 truncate_inode_pages(&inode
->i_data
, 0);
218 ext4_ioend_shutdown(inode
);
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);
228 ext4_ioend_shutdown(inode
);
230 if (is_bad_inode(inode
))
234 * Protect us against freezing - iput() caller didn't have to have any
235 * protection against it
237 sb_start_intwrite(inode
->i_sb
);
238 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
239 ext4_blocks_for_truncate(inode
)+3);
240 if (IS_ERR(handle
)) {
241 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
243 * If we're going to skip the normal cleanup, we still need to
244 * make sure that the in-core orphan linked list is properly
247 ext4_orphan_del(NULL
, inode
);
248 sb_end_intwrite(inode
->i_sb
);
253 ext4_handle_sync(handle
);
255 err
= ext4_mark_inode_dirty(handle
, inode
);
257 ext4_warning(inode
->i_sb
,
258 "couldn't mark inode dirty (err %d)", err
);
262 ext4_truncate(inode
);
265 * ext4_ext_truncate() doesn't reserve any slop when it
266 * restarts journal transactions; therefore there may not be
267 * enough credits left in the handle to remove the inode from
268 * the orphan list and set the dtime field.
270 if (!ext4_handle_has_enough_credits(handle
, 3)) {
271 err
= ext4_journal_extend(handle
, 3);
273 err
= ext4_journal_restart(handle
, 3);
275 ext4_warning(inode
->i_sb
,
276 "couldn't extend journal (err %d)", err
);
278 ext4_journal_stop(handle
);
279 ext4_orphan_del(NULL
, inode
);
280 sb_end_intwrite(inode
->i_sb
);
286 * Kill off the orphan record which ext4_truncate created.
287 * AKPM: I think this can be inside the above `if'.
288 * Note that ext4_orphan_del() has to be able to cope with the
289 * deletion of a non-existent orphan - this is because we don't
290 * know if ext4_truncate() actually created an orphan record.
291 * (Well, we could do this if we need to, but heck - it works)
293 ext4_orphan_del(handle
, inode
);
294 EXT4_I(inode
)->i_dtime
= get_seconds();
297 * One subtle ordering requirement: if anything has gone wrong
298 * (transaction abort, IO errors, whatever), then we can still
299 * do these next steps (the fs will already have been marked as
300 * having errors), but we can't free the inode if the mark_dirty
303 if (ext4_mark_inode_dirty(handle
, inode
))
304 /* If that failed, just do the required in-core inode clear. */
305 ext4_clear_inode(inode
);
307 ext4_free_inode(handle
, inode
);
308 ext4_journal_stop(handle
);
309 sb_end_intwrite(inode
->i_sb
);
312 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
316 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
318 return &EXT4_I(inode
)->i_reserved_quota
;
323 * Calculate the number of metadata blocks need to reserve
324 * to allocate a block located at @lblock
326 static int ext4_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
328 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
329 return ext4_ext_calc_metadata_amount(inode
, lblock
);
331 return ext4_ind_calc_metadata_amount(inode
, lblock
);
335 * Called with i_data_sem down, which is important since we can call
336 * ext4_discard_preallocations() from here.
338 void ext4_da_update_reserve_space(struct inode
*inode
,
339 int used
, int quota_claim
)
341 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
342 struct ext4_inode_info
*ei
= EXT4_I(inode
);
344 spin_lock(&ei
->i_block_reservation_lock
);
345 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
346 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
347 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
348 "with only %d reserved data blocks",
349 __func__
, inode
->i_ino
, used
,
350 ei
->i_reserved_data_blocks
);
352 used
= ei
->i_reserved_data_blocks
;
355 if (unlikely(ei
->i_allocated_meta_blocks
> ei
->i_reserved_meta_blocks
)) {
356 ext4_warning(inode
->i_sb
, "ino %lu, allocated %d "
357 "with only %d reserved metadata blocks "
358 "(releasing %d blocks with reserved %d data blocks)",
359 inode
->i_ino
, ei
->i_allocated_meta_blocks
,
360 ei
->i_reserved_meta_blocks
, used
,
361 ei
->i_reserved_data_blocks
);
363 ei
->i_allocated_meta_blocks
= ei
->i_reserved_meta_blocks
;
366 /* Update per-inode reservations */
367 ei
->i_reserved_data_blocks
-= used
;
368 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
369 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
370 used
+ ei
->i_allocated_meta_blocks
);
371 ei
->i_allocated_meta_blocks
= 0;
373 if (ei
->i_reserved_data_blocks
== 0) {
375 * We can release all of the reserved metadata blocks
376 * only when we have written all of the delayed
379 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
380 ei
->i_reserved_meta_blocks
);
381 ei
->i_reserved_meta_blocks
= 0;
382 ei
->i_da_metadata_calc_len
= 0;
384 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
386 /* Update quota subsystem for data blocks */
388 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
391 * We did fallocate with an offset that is already delayed
392 * allocated. So on delayed allocated writeback we should
393 * not re-claim the quota for fallocated blocks.
395 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
399 * If we have done all the pending block allocations and if
400 * there aren't any writers on the inode, we can discard the
401 * inode's preallocations.
403 if ((ei
->i_reserved_data_blocks
== 0) &&
404 (atomic_read(&inode
->i_writecount
) == 0))
405 ext4_discard_preallocations(inode
);
408 static int __check_block_validity(struct inode
*inode
, const char *func
,
410 struct ext4_map_blocks
*map
)
412 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
414 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
415 "lblock %lu mapped to illegal pblock "
416 "(length %d)", (unsigned long) map
->m_lblk
,
423 #define check_block_validity(inode, map) \
424 __check_block_validity((inode), __func__, __LINE__, (map))
427 * Return the number of contiguous dirty pages in a given inode
428 * starting at page frame idx.
430 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
431 unsigned int max_pages
)
433 struct address_space
*mapping
= inode
->i_mapping
;
437 int i
, nr_pages
, done
= 0;
441 pagevec_init(&pvec
, 0);
444 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
446 (pgoff_t
)PAGEVEC_SIZE
);
449 for (i
= 0; i
< nr_pages
; i
++) {
450 struct page
*page
= pvec
.pages
[i
];
451 struct buffer_head
*bh
, *head
;
454 if (unlikely(page
->mapping
!= mapping
) ||
456 PageWriteback(page
) ||
457 page
->index
!= idx
) {
462 if (page_has_buffers(page
)) {
463 bh
= head
= page_buffers(page
);
465 if (!buffer_delay(bh
) &&
466 !buffer_unwritten(bh
))
468 bh
= bh
->b_this_page
;
469 } while (!done
&& (bh
!= head
));
476 if (num
>= max_pages
) {
481 pagevec_release(&pvec
);
486 #ifdef ES_AGGRESSIVE_TEST
487 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
489 struct ext4_map_blocks
*es_map
,
490 struct ext4_map_blocks
*map
,
497 * There is a race window that the result is not the same.
498 * e.g. xfstests #223 when dioread_nolock enables. The reason
499 * is that we lookup a block mapping in extent status tree with
500 * out taking i_data_sem. So at the time the unwritten extent
501 * could be converted.
503 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
504 down_read((&EXT4_I(inode
)->i_data_sem
));
505 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
506 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
507 EXT4_GET_BLOCKS_KEEP_SIZE
);
509 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
510 EXT4_GET_BLOCKS_KEEP_SIZE
);
512 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
513 up_read((&EXT4_I(inode
)->i_data_sem
));
515 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
516 * because it shouldn't be marked in es_map->m_flags.
518 map
->m_flags
&= ~(EXT4_MAP_FROM_CLUSTER
| EXT4_MAP_BOUNDARY
);
521 * We don't check m_len because extent will be collpased in status
522 * tree. So the m_len might not equal.
524 if (es_map
->m_lblk
!= map
->m_lblk
||
525 es_map
->m_flags
!= map
->m_flags
||
526 es_map
->m_pblk
!= map
->m_pblk
) {
527 printk("ES cache assertation failed for inode: %lu "
528 "es_cached ex [%d/%d/%llu/%x] != "
529 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
530 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
531 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
532 map
->m_len
, map
->m_pblk
, map
->m_flags
,
536 #endif /* ES_AGGRESSIVE_TEST */
539 * The ext4_map_blocks() function tries to look up the requested blocks,
540 * and returns if the blocks are already mapped.
542 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
543 * and store the allocated blocks in the result buffer head and mark it
546 * If file type is extents based, it will call ext4_ext_map_blocks(),
547 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
550 * On success, it returns the number of blocks being mapped or allocate.
551 * if create==0 and the blocks are pre-allocated and uninitialized block,
552 * the result buffer head is unmapped. If the create ==1, it will make sure
553 * the buffer head is mapped.
555 * It returns 0 if plain look up failed (blocks have not been allocated), in
556 * that case, buffer head is unmapped
558 * It returns the error in case of allocation failure.
560 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
561 struct ext4_map_blocks
*map
, int flags
)
563 struct extent_status es
;
565 #ifdef ES_AGGRESSIVE_TEST
566 struct ext4_map_blocks orig_map
;
568 memcpy(&orig_map
, map
, sizeof(*map
));
572 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
573 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
574 (unsigned long) map
->m_lblk
);
576 /* Lookup extent status tree firstly */
577 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
578 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
579 map
->m_pblk
= ext4_es_pblock(&es
) +
580 map
->m_lblk
- es
.es_lblk
;
581 map
->m_flags
|= ext4_es_is_written(&es
) ?
582 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
583 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
584 if (retval
> map
->m_len
)
587 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
592 #ifdef ES_AGGRESSIVE_TEST
593 ext4_map_blocks_es_recheck(handle
, inode
, map
,
600 * Try to see if we can get the block without requesting a new
603 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
604 down_read((&EXT4_I(inode
)->i_data_sem
));
605 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
606 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
607 EXT4_GET_BLOCKS_KEEP_SIZE
);
609 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
610 EXT4_GET_BLOCKS_KEEP_SIZE
);
614 unsigned long long status
;
616 #ifdef ES_AGGRESSIVE_TEST
617 if (retval
!= map
->m_len
) {
618 printk("ES len assertation failed for inode: %lu "
619 "retval %d != map->m_len %d "
620 "in %s (lookup)\n", inode
->i_ino
, retval
,
621 map
->m_len
, __func__
);
625 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
626 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
627 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
628 ext4_find_delalloc_range(inode
, map
->m_lblk
,
629 map
->m_lblk
+ map
->m_len
- 1))
630 status
|= EXTENT_STATUS_DELAYED
;
631 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
632 map
->m_len
, map
->m_pblk
, status
);
636 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
637 up_read((&EXT4_I(inode
)->i_data_sem
));
640 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
641 int ret
= check_block_validity(inode
, map
);
646 /* If it is only a block(s) look up */
647 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
651 * Returns if the blocks have already allocated
653 * Note that if blocks have been preallocated
654 * ext4_ext_get_block() returns the create = 0
655 * with buffer head unmapped.
657 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
661 * Here we clear m_flags because after allocating an new extent,
662 * it will be set again.
664 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
667 * New blocks allocate and/or writing to uninitialized extent
668 * will possibly result in updating i_data, so we take
669 * the write lock of i_data_sem, and call get_blocks()
670 * with create == 1 flag.
672 down_write((&EXT4_I(inode
)->i_data_sem
));
675 * if the caller is from delayed allocation writeout path
676 * we have already reserved fs blocks for allocation
677 * let the underlying get_block() function know to
678 * avoid double accounting
680 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
681 ext4_set_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
683 * We need to check for EXT4 here because migrate
684 * could have changed the inode type in between
686 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
687 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
689 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
691 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
693 * We allocated new blocks which will result in
694 * i_data's format changing. Force the migrate
695 * to fail by clearing migrate flags
697 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
701 * Update reserved blocks/metadata blocks after successful
702 * block allocation which had been deferred till now. We don't
703 * support fallocate for non extent files. So we can update
704 * reserve space here.
707 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
708 ext4_da_update_reserve_space(inode
, retval
, 1);
710 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
711 ext4_clear_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
715 unsigned long long status
;
717 #ifdef ES_AGGRESSIVE_TEST
718 if (retval
!= map
->m_len
) {
719 printk("ES len assertation failed for inode: %lu "
720 "retval %d != map->m_len %d "
721 "in %s (allocation)\n", inode
->i_ino
, retval
,
722 map
->m_len
, __func__
);
727 * If the extent has been zeroed out, we don't need to update
728 * extent status tree.
730 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
731 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
732 if (ext4_es_is_written(&es
))
735 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
736 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
737 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
738 ext4_find_delalloc_range(inode
, map
->m_lblk
,
739 map
->m_lblk
+ map
->m_len
- 1))
740 status
|= EXTENT_STATUS_DELAYED
;
741 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
742 map
->m_pblk
, status
);
748 up_write((&EXT4_I(inode
)->i_data_sem
));
749 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
750 int ret
= check_block_validity(inode
, map
);
757 /* Maximum number of blocks we map for direct IO at once. */
758 #define DIO_MAX_BLOCKS 4096
760 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
761 struct buffer_head
*bh
, int flags
)
763 handle_t
*handle
= ext4_journal_current_handle();
764 struct ext4_map_blocks map
;
765 int ret
= 0, started
= 0;
768 if (ext4_has_inline_data(inode
))
772 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
774 if (flags
&& !(flags
& EXT4_GET_BLOCKS_NO_LOCK
) && !handle
) {
775 /* Direct IO write... */
776 if (map
.m_len
> DIO_MAX_BLOCKS
)
777 map
.m_len
= DIO_MAX_BLOCKS
;
778 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
779 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
781 if (IS_ERR(handle
)) {
782 ret
= PTR_ERR(handle
);
788 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
790 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
791 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
792 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
796 ext4_journal_stop(handle
);
800 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
801 struct buffer_head
*bh
, int create
)
803 return _ext4_get_block(inode
, iblock
, bh
,
804 create
? EXT4_GET_BLOCKS_CREATE
: 0);
808 * `handle' can be NULL if create is zero
810 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
811 ext4_lblk_t block
, int create
, int *errp
)
813 struct ext4_map_blocks map
;
814 struct buffer_head
*bh
;
817 J_ASSERT(handle
!= NULL
|| create
== 0);
821 err
= ext4_map_blocks(handle
, inode
, &map
,
822 create
? EXT4_GET_BLOCKS_CREATE
: 0);
824 /* ensure we send some value back into *errp */
827 if (create
&& err
== 0)
828 err
= -ENOSPC
; /* should never happen */
834 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
839 if (map
.m_flags
& EXT4_MAP_NEW
) {
840 J_ASSERT(create
!= 0);
841 J_ASSERT(handle
!= NULL
);
844 * Now that we do not always journal data, we should
845 * keep in mind whether this should always journal the
846 * new buffer as metadata. For now, regular file
847 * writes use ext4_get_block instead, so it's not a
851 BUFFER_TRACE(bh
, "call get_create_access");
852 fatal
= ext4_journal_get_create_access(handle
, bh
);
853 if (!fatal
&& !buffer_uptodate(bh
)) {
854 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
855 set_buffer_uptodate(bh
);
858 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
859 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
863 BUFFER_TRACE(bh
, "not a new buffer");
873 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
874 ext4_lblk_t block
, int create
, int *err
)
876 struct buffer_head
*bh
;
878 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
881 if (buffer_uptodate(bh
))
883 ll_rw_block(READ
| REQ_META
| REQ_PRIO
, 1, &bh
);
885 if (buffer_uptodate(bh
))
892 int ext4_walk_page_buffers(handle_t
*handle
,
893 struct buffer_head
*head
,
897 int (*fn
)(handle_t
*handle
,
898 struct buffer_head
*bh
))
900 struct buffer_head
*bh
;
901 unsigned block_start
, block_end
;
902 unsigned blocksize
= head
->b_size
;
904 struct buffer_head
*next
;
906 for (bh
= head
, block_start
= 0;
907 ret
== 0 && (bh
!= head
|| !block_start
);
908 block_start
= block_end
, bh
= next
) {
909 next
= bh
->b_this_page
;
910 block_end
= block_start
+ blocksize
;
911 if (block_end
<= from
|| block_start
>= to
) {
912 if (partial
&& !buffer_uptodate(bh
))
916 err
= (*fn
)(handle
, bh
);
924 * To preserve ordering, it is essential that the hole instantiation and
925 * the data write be encapsulated in a single transaction. We cannot
926 * close off a transaction and start a new one between the ext4_get_block()
927 * and the commit_write(). So doing the jbd2_journal_start at the start of
928 * prepare_write() is the right place.
930 * Also, this function can nest inside ext4_writepage(). In that case, we
931 * *know* that ext4_writepage() has generated enough buffer credits to do the
932 * whole page. So we won't block on the journal in that case, which is good,
933 * because the caller may be PF_MEMALLOC.
935 * By accident, ext4 can be reentered when a transaction is open via
936 * quota file writes. If we were to commit the transaction while thus
937 * reentered, there can be a deadlock - we would be holding a quota
938 * lock, and the commit would never complete if another thread had a
939 * transaction open and was blocking on the quota lock - a ranking
942 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
943 * will _not_ run commit under these circumstances because handle->h_ref
944 * is elevated. We'll still have enough credits for the tiny quotafile
947 int do_journal_get_write_access(handle_t
*handle
,
948 struct buffer_head
*bh
)
950 int dirty
= buffer_dirty(bh
);
953 if (!buffer_mapped(bh
) || buffer_freed(bh
))
956 * __block_write_begin() could have dirtied some buffers. Clean
957 * the dirty bit as jbd2_journal_get_write_access() could complain
958 * otherwise about fs integrity issues. Setting of the dirty bit
959 * by __block_write_begin() isn't a real problem here as we clear
960 * the bit before releasing a page lock and thus writeback cannot
961 * ever write the buffer.
964 clear_buffer_dirty(bh
);
965 ret
= ext4_journal_get_write_access(handle
, bh
);
967 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
971 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
972 struct buffer_head
*bh_result
, int create
);
973 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
974 loff_t pos
, unsigned len
, unsigned flags
,
975 struct page
**pagep
, void **fsdata
)
977 struct inode
*inode
= mapping
->host
;
978 int ret
, needed_blocks
;
985 trace_ext4_write_begin(inode
, pos
, len
, flags
);
987 * Reserve one block more for addition to orphan list in case
988 * we allocate blocks but write fails for some reason
990 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
991 index
= pos
>> PAGE_CACHE_SHIFT
;
992 from
= pos
& (PAGE_CACHE_SIZE
- 1);
995 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
996 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1005 * grab_cache_page_write_begin() can take a long time if the
1006 * system is thrashing due to memory pressure, or if the page
1007 * is being written back. So grab it first before we start
1008 * the transaction handle. This also allows us to allocate
1009 * the page (if needed) without using GFP_NOFS.
1012 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1018 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1019 if (IS_ERR(handle
)) {
1020 page_cache_release(page
);
1021 return PTR_ERR(handle
);
1025 if (page
->mapping
!= mapping
) {
1026 /* The page got truncated from under us */
1028 page_cache_release(page
);
1029 ext4_journal_stop(handle
);
1032 wait_on_page_writeback(page
);
1034 if (ext4_should_dioread_nolock(inode
))
1035 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
1037 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1039 if (!ret
&& ext4_should_journal_data(inode
)) {
1040 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1042 do_journal_get_write_access
);
1048 * __block_write_begin may have instantiated a few blocks
1049 * outside i_size. Trim these off again. Don't need
1050 * i_size_read because we hold i_mutex.
1052 * Add inode to orphan list in case we crash before
1055 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1056 ext4_orphan_add(handle
, inode
);
1058 ext4_journal_stop(handle
);
1059 if (pos
+ len
> inode
->i_size
) {
1060 ext4_truncate_failed_write(inode
);
1062 * If truncate failed early the inode might
1063 * still be on the orphan list; we need to
1064 * make sure the inode is removed from the
1065 * orphan list in that case.
1068 ext4_orphan_del(NULL
, inode
);
1071 if (ret
== -ENOSPC
&&
1072 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1074 page_cache_release(page
);
1081 /* For write_end() in data=journal mode */
1082 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1085 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1087 set_buffer_uptodate(bh
);
1088 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1089 clear_buffer_meta(bh
);
1090 clear_buffer_prio(bh
);
1095 * We need to pick up the new inode size which generic_commit_write gave us
1096 * `file' can be NULL - eg, when called from page_symlink().
1098 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1099 * buffers are managed internally.
1101 static int ext4_write_end(struct file
*file
,
1102 struct address_space
*mapping
,
1103 loff_t pos
, unsigned len
, unsigned copied
,
1104 struct page
*page
, void *fsdata
)
1106 handle_t
*handle
= ext4_journal_current_handle();
1107 struct inode
*inode
= mapping
->host
;
1109 int i_size_changed
= 0;
1111 trace_ext4_write_end(inode
, pos
, len
, copied
);
1112 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
)) {
1113 ret
= ext4_jbd2_file_inode(handle
, inode
);
1116 page_cache_release(page
);
1121 if (ext4_has_inline_data(inode
)) {
1122 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1128 copied
= block_write_end(file
, mapping
, pos
,
1129 len
, copied
, page
, fsdata
);
1132 * No need to use i_size_read() here, the i_size
1133 * cannot change under us because we hole i_mutex.
1135 * But it's important to update i_size while still holding page lock:
1136 * page writeout could otherwise come in and zero beyond i_size.
1138 if (pos
+ copied
> inode
->i_size
) {
1139 i_size_write(inode
, pos
+ copied
);
1143 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1144 /* We need to mark inode dirty even if
1145 * new_i_size is less that inode->i_size
1146 * but greater than i_disksize. (hint delalloc)
1148 ext4_update_i_disksize(inode
, (pos
+ copied
));
1152 page_cache_release(page
);
1155 * Don't mark the inode dirty under page lock. First, it unnecessarily
1156 * makes the holding time of page lock longer. Second, it forces lock
1157 * ordering of page lock and transaction start for journaling
1161 ext4_mark_inode_dirty(handle
, inode
);
1165 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1166 /* if we have allocated more blocks and copied
1167 * less. We will have blocks allocated outside
1168 * inode->i_size. So truncate them
1170 ext4_orphan_add(handle
, inode
);
1172 ret2
= ext4_journal_stop(handle
);
1176 if (pos
+ len
> inode
->i_size
) {
1177 ext4_truncate_failed_write(inode
);
1179 * If truncate failed early the inode might still be
1180 * on the orphan list; we need to make sure the inode
1181 * is removed from the orphan list in that case.
1184 ext4_orphan_del(NULL
, inode
);
1187 return ret
? ret
: copied
;
1190 static int ext4_journalled_write_end(struct file
*file
,
1191 struct address_space
*mapping
,
1192 loff_t pos
, unsigned len
, unsigned copied
,
1193 struct page
*page
, void *fsdata
)
1195 handle_t
*handle
= ext4_journal_current_handle();
1196 struct inode
*inode
= mapping
->host
;
1202 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1203 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1206 BUG_ON(!ext4_handle_valid(handle
));
1208 if (ext4_has_inline_data(inode
))
1209 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1213 if (!PageUptodate(page
))
1215 page_zero_new_buffers(page
, from
+copied
, to
);
1218 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1219 to
, &partial
, write_end_fn
);
1221 SetPageUptodate(page
);
1223 new_i_size
= pos
+ copied
;
1224 if (new_i_size
> inode
->i_size
)
1225 i_size_write(inode
, pos
+copied
);
1226 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1227 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1228 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1229 ext4_update_i_disksize(inode
, new_i_size
);
1230 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1236 page_cache_release(page
);
1237 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1238 /* if we have allocated more blocks and copied
1239 * less. We will have blocks allocated outside
1240 * inode->i_size. So truncate them
1242 ext4_orphan_add(handle
, inode
);
1244 ret2
= ext4_journal_stop(handle
);
1247 if (pos
+ len
> inode
->i_size
) {
1248 ext4_truncate_failed_write(inode
);
1250 * If truncate failed early the inode might still be
1251 * on the orphan list; we need to make sure the inode
1252 * is removed from the orphan list in that case.
1255 ext4_orphan_del(NULL
, inode
);
1258 return ret
? ret
: copied
;
1262 * Reserve a metadata for a single block located at lblock
1264 static int ext4_da_reserve_metadata(struct inode
*inode
, ext4_lblk_t lblock
)
1267 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1268 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1269 unsigned int md_needed
;
1270 ext4_lblk_t save_last_lblock
;
1274 * recalculate the amount of metadata blocks to reserve
1275 * in order to allocate nrblocks
1276 * worse case is one extent per block
1279 spin_lock(&ei
->i_block_reservation_lock
);
1281 * ext4_calc_metadata_amount() has side effects, which we have
1282 * to be prepared undo if we fail to claim space.
1284 save_len
= ei
->i_da_metadata_calc_len
;
1285 save_last_lblock
= ei
->i_da_metadata_calc_last_lblock
;
1286 md_needed
= EXT4_NUM_B2C(sbi
,
1287 ext4_calc_metadata_amount(inode
, lblock
));
1288 trace_ext4_da_reserve_space(inode
, md_needed
);
1291 * We do still charge estimated metadata to the sb though;
1292 * we cannot afford to run out of free blocks.
1294 if (ext4_claim_free_clusters(sbi
, md_needed
, 0)) {
1295 ei
->i_da_metadata_calc_len
= save_len
;
1296 ei
->i_da_metadata_calc_last_lblock
= save_last_lblock
;
1297 spin_unlock(&ei
->i_block_reservation_lock
);
1298 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1304 ei
->i_reserved_meta_blocks
+= md_needed
;
1305 spin_unlock(&ei
->i_block_reservation_lock
);
1307 return 0; /* success */
1311 * Reserve a single cluster located at lblock
1313 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1316 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1317 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1318 unsigned int md_needed
;
1320 ext4_lblk_t save_last_lblock
;
1324 * We will charge metadata quota at writeout time; this saves
1325 * us from metadata over-estimation, though we may go over by
1326 * a small amount in the end. Here we just reserve for data.
1328 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1333 * recalculate the amount of metadata blocks to reserve
1334 * in order to allocate nrblocks
1335 * worse case is one extent per block
1338 spin_lock(&ei
->i_block_reservation_lock
);
1340 * ext4_calc_metadata_amount() has side effects, which we have
1341 * to be prepared undo if we fail to claim space.
1343 save_len
= ei
->i_da_metadata_calc_len
;
1344 save_last_lblock
= ei
->i_da_metadata_calc_last_lblock
;
1345 md_needed
= EXT4_NUM_B2C(sbi
,
1346 ext4_calc_metadata_amount(inode
, lblock
));
1347 trace_ext4_da_reserve_space(inode
, md_needed
);
1350 * We do still charge estimated metadata to the sb though;
1351 * we cannot afford to run out of free blocks.
1353 if (ext4_claim_free_clusters(sbi
, md_needed
+ 1, 0)) {
1354 ei
->i_da_metadata_calc_len
= save_len
;
1355 ei
->i_da_metadata_calc_last_lblock
= save_last_lblock
;
1356 spin_unlock(&ei
->i_block_reservation_lock
);
1357 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1361 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1364 ei
->i_reserved_data_blocks
++;
1365 ei
->i_reserved_meta_blocks
+= md_needed
;
1366 spin_unlock(&ei
->i_block_reservation_lock
);
1368 return 0; /* success */
1371 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1373 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1374 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1377 return; /* Nothing to release, exit */
1379 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1381 trace_ext4_da_release_space(inode
, to_free
);
1382 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1384 * if there aren't enough reserved blocks, then the
1385 * counter is messed up somewhere. Since this
1386 * function is called from invalidate page, it's
1387 * harmless to return without any action.
1389 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1390 "ino %lu, to_free %d with only %d reserved "
1391 "data blocks", inode
->i_ino
, to_free
,
1392 ei
->i_reserved_data_blocks
);
1394 to_free
= ei
->i_reserved_data_blocks
;
1396 ei
->i_reserved_data_blocks
-= to_free
;
1398 if (ei
->i_reserved_data_blocks
== 0) {
1400 * We can release all of the reserved metadata blocks
1401 * only when we have written all of the delayed
1402 * allocation blocks.
1403 * Note that in case of bigalloc, i_reserved_meta_blocks,
1404 * i_reserved_data_blocks, etc. refer to number of clusters.
1406 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
1407 ei
->i_reserved_meta_blocks
);
1408 ei
->i_reserved_meta_blocks
= 0;
1409 ei
->i_da_metadata_calc_len
= 0;
1412 /* update fs dirty data blocks counter */
1413 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1415 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1417 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1420 static void ext4_da_page_release_reservation(struct page
*page
,
1421 unsigned long offset
)
1424 struct buffer_head
*head
, *bh
;
1425 unsigned int curr_off
= 0;
1426 struct inode
*inode
= page
->mapping
->host
;
1427 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1431 head
= page_buffers(page
);
1434 unsigned int next_off
= curr_off
+ bh
->b_size
;
1436 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1438 clear_buffer_delay(bh
);
1440 curr_off
= next_off
;
1441 } while ((bh
= bh
->b_this_page
) != head
);
1444 lblk
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1445 ext4_es_remove_extent(inode
, lblk
, to_release
);
1448 /* If we have released all the blocks belonging to a cluster, then we
1449 * need to release the reserved space for that cluster. */
1450 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1451 while (num_clusters
> 0) {
1452 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1453 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1454 if (sbi
->s_cluster_ratio
== 1 ||
1455 !ext4_find_delalloc_cluster(inode
, lblk
))
1456 ext4_da_release_space(inode
, 1);
1463 * Delayed allocation stuff
1467 * mpage_da_submit_io - walks through extent of pages and try to write
1468 * them with writepage() call back
1470 * @mpd->inode: inode
1471 * @mpd->first_page: first page of the extent
1472 * @mpd->next_page: page after the last page of the extent
1474 * By the time mpage_da_submit_io() is called we expect all blocks
1475 * to be allocated. this may be wrong if allocation failed.
1477 * As pages are already locked by write_cache_pages(), we can't use it
1479 static int mpage_da_submit_io(struct mpage_da_data
*mpd
,
1480 struct ext4_map_blocks
*map
)
1482 struct pagevec pvec
;
1483 unsigned long index
, end
;
1484 int ret
= 0, err
, nr_pages
, i
;
1485 struct inode
*inode
= mpd
->inode
;
1486 struct address_space
*mapping
= inode
->i_mapping
;
1487 loff_t size
= i_size_read(inode
);
1488 unsigned int len
, block_start
;
1489 struct buffer_head
*bh
, *page_bufs
= NULL
;
1490 sector_t pblock
= 0, cur_logical
= 0;
1491 struct ext4_io_submit io_submit
;
1493 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1494 memset(&io_submit
, 0, sizeof(io_submit
));
1496 * We need to start from the first_page to the next_page - 1
1497 * to make sure we also write the mapped dirty buffer_heads.
1498 * If we look at mpd->b_blocknr we would only be looking
1499 * at the currently mapped buffer_heads.
1501 index
= mpd
->first_page
;
1502 end
= mpd
->next_page
- 1;
1504 pagevec_init(&pvec
, 0);
1505 while (index
<= end
) {
1506 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1509 for (i
= 0; i
< nr_pages
; i
++) {
1511 struct page
*page
= pvec
.pages
[i
];
1513 index
= page
->index
;
1517 if (index
== size
>> PAGE_CACHE_SHIFT
)
1518 len
= size
& ~PAGE_CACHE_MASK
;
1520 len
= PAGE_CACHE_SIZE
;
1522 cur_logical
= index
<< (PAGE_CACHE_SHIFT
-
1524 pblock
= map
->m_pblk
+ (cur_logical
-
1529 BUG_ON(!PageLocked(page
));
1530 BUG_ON(PageWriteback(page
));
1532 bh
= page_bufs
= page_buffers(page
);
1535 if (map
&& (cur_logical
>= map
->m_lblk
) &&
1536 (cur_logical
<= (map
->m_lblk
+
1537 (map
->m_len
- 1)))) {
1538 if (buffer_delay(bh
)) {
1539 clear_buffer_delay(bh
);
1540 bh
->b_blocknr
= pblock
;
1542 if (buffer_unwritten(bh
) ||
1544 BUG_ON(bh
->b_blocknr
!= pblock
);
1545 if (map
->m_flags
& EXT4_MAP_UNINIT
)
1546 set_buffer_uninit(bh
);
1547 clear_buffer_unwritten(bh
);
1551 * skip page if block allocation undone and
1554 if (ext4_bh_delay_or_unwritten(NULL
, bh
))
1556 bh
= bh
->b_this_page
;
1557 block_start
+= bh
->b_size
;
1560 } while (bh
!= page_bufs
);
1567 clear_page_dirty_for_io(page
);
1568 err
= ext4_bio_write_page(&io_submit
, page
, len
,
1571 mpd
->pages_written
++;
1573 * In error case, we have to continue because
1574 * remaining pages are still locked
1579 pagevec_release(&pvec
);
1581 ext4_io_submit(&io_submit
);
1585 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
)
1589 struct pagevec pvec
;
1590 struct inode
*inode
= mpd
->inode
;
1591 struct address_space
*mapping
= inode
->i_mapping
;
1592 ext4_lblk_t start
, last
;
1594 index
= mpd
->first_page
;
1595 end
= mpd
->next_page
- 1;
1597 start
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1598 last
= end
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1599 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1601 pagevec_init(&pvec
, 0);
1602 while (index
<= end
) {
1603 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1606 for (i
= 0; i
< nr_pages
; i
++) {
1607 struct page
*page
= pvec
.pages
[i
];
1608 if (page
->index
> end
)
1610 BUG_ON(!PageLocked(page
));
1611 BUG_ON(PageWriteback(page
));
1612 block_invalidatepage(page
, 0);
1613 ClearPageUptodate(page
);
1616 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1617 pagevec_release(&pvec
);
1622 static void ext4_print_free_blocks(struct inode
*inode
)
1624 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1625 struct super_block
*sb
= inode
->i_sb
;
1626 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1628 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1629 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1630 ext4_count_free_clusters(sb
)));
1631 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1632 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1633 (long long) EXT4_C2B(EXT4_SB(sb
),
1634 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1635 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1636 (long long) EXT4_C2B(EXT4_SB(sb
),
1637 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1638 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1639 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1640 ei
->i_reserved_data_blocks
);
1641 ext4_msg(sb
, KERN_CRIT
, "i_reserved_meta_blocks=%u",
1642 ei
->i_reserved_meta_blocks
);
1643 ext4_msg(sb
, KERN_CRIT
, "i_allocated_meta_blocks=%u",
1644 ei
->i_allocated_meta_blocks
);
1649 * mpage_da_map_and_submit - go through given space, map them
1650 * if necessary, and then submit them for I/O
1652 * @mpd - bh describing space
1654 * The function skips space we know is already mapped to disk blocks.
1657 static void mpage_da_map_and_submit(struct mpage_da_data
*mpd
)
1659 int err
, blks
, get_blocks_flags
;
1660 struct ext4_map_blocks map
, *mapp
= NULL
;
1661 sector_t next
= mpd
->b_blocknr
;
1662 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1663 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
1664 handle_t
*handle
= NULL
;
1667 * If the blocks are mapped already, or we couldn't accumulate
1668 * any blocks, then proceed immediately to the submission stage.
1670 if ((mpd
->b_size
== 0) ||
1671 ((mpd
->b_state
& (1 << BH_Mapped
)) &&
1672 !(mpd
->b_state
& (1 << BH_Delay
)) &&
1673 !(mpd
->b_state
& (1 << BH_Unwritten
))))
1676 handle
= ext4_journal_current_handle();
1680 * Call ext4_map_blocks() to allocate any delayed allocation
1681 * blocks, or to convert an uninitialized extent to be
1682 * initialized (in the case where we have written into
1683 * one or more preallocated blocks).
1685 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1686 * indicate that we are on the delayed allocation path. This
1687 * affects functions in many different parts of the allocation
1688 * call path. This flag exists primarily because we don't
1689 * want to change *many* call functions, so ext4_map_blocks()
1690 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1691 * inode's allocation semaphore is taken.
1693 * If the blocks in questions were delalloc blocks, set
1694 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1695 * variables are updated after the blocks have been allocated.
1698 map
.m_len
= max_blocks
;
1700 * We're in delalloc path and it is possible that we're going to
1701 * need more metadata blocks than previously reserved. However
1702 * we must not fail because we're in writeback and there is
1703 * nothing we can do about it so it might result in data loss.
1704 * So use reserved blocks to allocate metadata if possible.
1706 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
1707 EXT4_GET_BLOCKS_METADATA_NOFAIL
;
1708 if (ext4_should_dioread_nolock(mpd
->inode
))
1709 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
1710 if (mpd
->b_state
& (1 << BH_Delay
))
1711 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
1714 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
1716 struct super_block
*sb
= mpd
->inode
->i_sb
;
1720 * If get block returns EAGAIN or ENOSPC and there
1721 * appears to be free blocks we will just let
1722 * mpage_da_submit_io() unlock all of the pages.
1727 if (err
== -ENOSPC
&& ext4_count_free_clusters(sb
)) {
1733 * get block failure will cause us to loop in
1734 * writepages, because a_ops->writepage won't be able
1735 * to make progress. The page will be redirtied by
1736 * writepage and writepages will again try to write
1739 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
1740 ext4_msg(sb
, KERN_CRIT
,
1741 "delayed block allocation failed for inode %lu "
1742 "at logical offset %llu with max blocks %zd "
1743 "with error %d", mpd
->inode
->i_ino
,
1744 (unsigned long long) next
,
1745 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1746 ext4_msg(sb
, KERN_CRIT
,
1747 "This should not happen!! Data will be lost");
1749 ext4_print_free_blocks(mpd
->inode
);
1751 /* invalidate all the pages */
1752 ext4_da_block_invalidatepages(mpd
);
1754 /* Mark this page range as having been completed */
1761 if (map
.m_flags
& EXT4_MAP_NEW
) {
1762 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
1765 for (i
= 0; i
< map
.m_len
; i
++)
1766 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
1770 * Update on-disk size along with block allocation.
1772 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
1773 if (disksize
> i_size_read(mpd
->inode
))
1774 disksize
= i_size_read(mpd
->inode
);
1775 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
1776 ext4_update_i_disksize(mpd
->inode
, disksize
);
1777 err
= ext4_mark_inode_dirty(handle
, mpd
->inode
);
1779 ext4_error(mpd
->inode
->i_sb
,
1780 "Failed to mark inode %lu dirty",
1785 mpage_da_submit_io(mpd
, mapp
);
1789 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1790 (1 << BH_Delay) | (1 << BH_Unwritten))
1793 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1795 * @mpd->lbh - extent of blocks
1796 * @logical - logical number of the block in the file
1797 * @b_state - b_state of the buffer head added
1799 * the function is used to collect contig. blocks in same state
1801 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, sector_t logical
,
1802 unsigned long b_state
)
1805 int blkbits
= mpd
->inode
->i_blkbits
;
1806 int nrblocks
= mpd
->b_size
>> blkbits
;
1809 * XXX Don't go larger than mballoc is willing to allocate
1810 * This is a stopgap solution. We eventually need to fold
1811 * mpage_da_submit_io() into this function and then call
1812 * ext4_map_blocks() multiple times in a loop
1814 if (nrblocks
>= (8*1024*1024 >> blkbits
))
1817 /* check if the reserved journal credits might overflow */
1818 if (!ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
)) {
1819 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1821 * With non-extent format we are limited by the journal
1822 * credit available. Total credit needed to insert
1823 * nrblocks contiguous blocks is dependent on the
1824 * nrblocks. So limit nrblocks.
1830 * First block in the extent
1832 if (mpd
->b_size
== 0) {
1833 mpd
->b_blocknr
= logical
;
1834 mpd
->b_size
= 1 << blkbits
;
1835 mpd
->b_state
= b_state
& BH_FLAGS
;
1839 next
= mpd
->b_blocknr
+ nrblocks
;
1841 * Can we merge the block to our big extent?
1843 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
1844 mpd
->b_size
+= 1 << blkbits
;
1850 * We couldn't merge the block to our extent, so we
1851 * need to flush current extent and start new one
1853 mpage_da_map_and_submit(mpd
);
1857 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1859 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1863 * This function is grabs code from the very beginning of
1864 * ext4_map_blocks, but assumes that the caller is from delayed write
1865 * time. This function looks up the requested blocks and sets the
1866 * buffer delay bit under the protection of i_data_sem.
1868 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1869 struct ext4_map_blocks
*map
,
1870 struct buffer_head
*bh
)
1872 struct extent_status es
;
1874 sector_t invalid_block
= ~((sector_t
) 0xffff);
1875 #ifdef ES_AGGRESSIVE_TEST
1876 struct ext4_map_blocks orig_map
;
1878 memcpy(&orig_map
, map
, sizeof(*map
));
1881 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1885 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1886 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1887 (unsigned long) map
->m_lblk
);
1889 /* Lookup extent status tree firstly */
1890 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1892 if (ext4_es_is_hole(&es
)) {
1894 down_read((&EXT4_I(inode
)->i_data_sem
));
1899 * Delayed extent could be allocated by fallocate.
1900 * So we need to check it.
1902 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1903 map_bh(bh
, inode
->i_sb
, invalid_block
);
1905 set_buffer_delay(bh
);
1909 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1910 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1911 if (retval
> map
->m_len
)
1912 retval
= map
->m_len
;
1913 map
->m_len
= retval
;
1914 if (ext4_es_is_written(&es
))
1915 map
->m_flags
|= EXT4_MAP_MAPPED
;
1916 else if (ext4_es_is_unwritten(&es
))
1917 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1921 #ifdef ES_AGGRESSIVE_TEST
1922 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1928 * Try to see if we can get the block without requesting a new
1929 * file system block.
1931 down_read((&EXT4_I(inode
)->i_data_sem
));
1932 if (ext4_has_inline_data(inode
)) {
1934 * We will soon create blocks for this page, and let
1935 * us pretend as if the blocks aren't allocated yet.
1936 * In case of clusters, we have to handle the work
1937 * of mapping from cluster so that the reserved space
1938 * is calculated properly.
1940 if ((EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) &&
1941 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
1942 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
1944 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1945 retval
= ext4_ext_map_blocks(NULL
, inode
, map
,
1946 EXT4_GET_BLOCKS_NO_PUT_HOLE
);
1948 retval
= ext4_ind_map_blocks(NULL
, inode
, map
,
1949 EXT4_GET_BLOCKS_NO_PUT_HOLE
);
1955 * XXX: __block_prepare_write() unmaps passed block,
1959 * If the block was allocated from previously allocated cluster,
1960 * then we don't need to reserve it again. However we still need
1961 * to reserve metadata for every block we're going to write.
1963 if (!(map
->m_flags
& EXT4_MAP_FROM_CLUSTER
)) {
1964 ret
= ext4_da_reserve_space(inode
, iblock
);
1966 /* not enough space to reserve */
1971 ret
= ext4_da_reserve_metadata(inode
, iblock
);
1973 /* not enough space to reserve */
1979 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1980 ~0, EXTENT_STATUS_DELAYED
);
1986 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1987 * and it should not appear on the bh->b_state.
1989 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
1991 map_bh(bh
, inode
->i_sb
, invalid_block
);
1993 set_buffer_delay(bh
);
1994 } else if (retval
> 0) {
1996 unsigned long long status
;
1998 #ifdef ES_AGGRESSIVE_TEST
1999 if (retval
!= map
->m_len
) {
2000 printk("ES len assertation failed for inode: %lu "
2001 "retval %d != map->m_len %d "
2002 "in %s (lookup)\n", inode
->i_ino
, retval
,
2003 map
->m_len
, __func__
);
2007 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
2008 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
2009 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
2010 map
->m_pblk
, status
);
2016 up_read((&EXT4_I(inode
)->i_data_sem
));
2022 * This is a special get_blocks_t callback which is used by
2023 * ext4_da_write_begin(). It will either return mapped block or
2024 * reserve space for a single block.
2026 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2027 * We also have b_blocknr = -1 and b_bdev initialized properly
2029 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2030 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2031 * initialized properly.
2033 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2034 struct buffer_head
*bh
, int create
)
2036 struct ext4_map_blocks map
;
2039 BUG_ON(create
== 0);
2040 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
2042 map
.m_lblk
= iblock
;
2046 * first, we need to know whether the block is allocated already
2047 * preallocated blocks are unmapped but should treated
2048 * the same as allocated blocks.
2050 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
2054 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
2055 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
2057 if (buffer_unwritten(bh
)) {
2058 /* A delayed write to unwritten bh should be marked
2059 * new and mapped. Mapped ensures that we don't do
2060 * get_block multiple times when we write to the same
2061 * offset and new ensures that we do proper zero out
2062 * for partial write.
2065 set_buffer_mapped(bh
);
2070 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2076 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2082 static int __ext4_journalled_writepage(struct page
*page
,
2085 struct address_space
*mapping
= page
->mapping
;
2086 struct inode
*inode
= mapping
->host
;
2087 struct buffer_head
*page_bufs
= NULL
;
2088 handle_t
*handle
= NULL
;
2089 int ret
= 0, err
= 0;
2090 int inline_data
= ext4_has_inline_data(inode
);
2091 struct buffer_head
*inode_bh
= NULL
;
2093 ClearPageChecked(page
);
2096 BUG_ON(page
->index
!= 0);
2097 BUG_ON(len
> ext4_get_max_inline_size(inode
));
2098 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
2099 if (inode_bh
== NULL
)
2102 page_bufs
= page_buffers(page
);
2107 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
2110 /* As soon as we unlock the page, it can go away, but we have
2111 * references to buffers so we are safe */
2114 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2115 ext4_writepage_trans_blocks(inode
));
2116 if (IS_ERR(handle
)) {
2117 ret
= PTR_ERR(handle
);
2121 BUG_ON(!ext4_handle_valid(handle
));
2124 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
2126 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
2129 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2130 do_journal_get_write_access
);
2132 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2137 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
2138 err
= ext4_journal_stop(handle
);
2142 if (!ext4_has_inline_data(inode
))
2143 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
2145 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2152 * Note that we don't need to start a transaction unless we're journaling data
2153 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2154 * need to file the inode to the transaction's list in ordered mode because if
2155 * we are writing back data added by write(), the inode is already there and if
2156 * we are writing back data modified via mmap(), no one guarantees in which
2157 * transaction the data will hit the disk. In case we are journaling data, we
2158 * cannot start transaction directly because transaction start ranks above page
2159 * lock so we have to do some magic.
2161 * This function can get called via...
2162 * - ext4_da_writepages after taking page lock (have journal handle)
2163 * - journal_submit_inode_data_buffers (no journal handle)
2164 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2165 * - grab_page_cache when doing write_begin (have journal handle)
2167 * We don't do any block allocation in this function. If we have page with
2168 * multiple blocks we need to write those buffer_heads that are mapped. This
2169 * is important for mmaped based write. So if we do with blocksize 1K
2170 * truncate(f, 1024);
2171 * a = mmap(f, 0, 4096);
2173 * truncate(f, 4096);
2174 * we have in the page first buffer_head mapped via page_mkwrite call back
2175 * but other buffer_heads would be unmapped but dirty (dirty done via the
2176 * do_wp_page). So writepage should write the first block. If we modify
2177 * the mmap area beyond 1024 we will again get a page_fault and the
2178 * page_mkwrite callback will do the block allocation and mark the
2179 * buffer_heads mapped.
2181 * We redirty the page if we have any buffer_heads that is either delay or
2182 * unwritten in the page.
2184 * We can get recursively called as show below.
2186 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2189 * But since we don't do any block allocation we should not deadlock.
2190 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2192 static int ext4_writepage(struct page
*page
,
2193 struct writeback_control
*wbc
)
2198 struct buffer_head
*page_bufs
= NULL
;
2199 struct inode
*inode
= page
->mapping
->host
;
2200 struct ext4_io_submit io_submit
;
2202 trace_ext4_writepage(page
);
2203 size
= i_size_read(inode
);
2204 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2205 len
= size
& ~PAGE_CACHE_MASK
;
2207 len
= PAGE_CACHE_SIZE
;
2209 page_bufs
= page_buffers(page
);
2211 * We cannot do block allocation or other extent handling in this
2212 * function. If there are buffers needing that, we have to redirty
2213 * the page. But we may reach here when we do a journal commit via
2214 * journal_submit_inode_data_buffers() and in that case we must write
2215 * allocated buffers to achieve data=ordered mode guarantees.
2217 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2218 ext4_bh_delay_or_unwritten
)) {
2219 redirty_page_for_writepage(wbc
, page
);
2220 if (current
->flags
& PF_MEMALLOC
) {
2222 * For memory cleaning there's no point in writing only
2223 * some buffers. So just bail out. Warn if we came here
2224 * from direct reclaim.
2226 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2233 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2235 * It's mmapped pagecache. Add buffers and journal it. There
2236 * doesn't seem much point in redirtying the page here.
2238 return __ext4_journalled_writepage(page
, len
);
2240 memset(&io_submit
, 0, sizeof(io_submit
));
2241 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
);
2242 ext4_io_submit(&io_submit
);
2247 * This is called via ext4_da_writepages() to
2248 * calculate the total number of credits to reserve to fit
2249 * a single extent allocation into a single transaction,
2250 * ext4_da_writpeages() will loop calling this before
2251 * the block allocation.
2254 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2256 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2259 * With non-extent format the journal credit needed to
2260 * insert nrblocks contiguous block is dependent on
2261 * number of contiguous block. So we will limit
2262 * number of contiguous block to a sane value
2264 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
2265 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2266 max_blocks
= EXT4_MAX_TRANS_DATA
;
2268 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2272 * write_cache_pages_da - walk the list of dirty pages of the given
2273 * address space and accumulate pages that need writing, and call
2274 * mpage_da_map_and_submit to map a single contiguous memory region
2275 * and then write them.
2277 static int write_cache_pages_da(handle_t
*handle
,
2278 struct address_space
*mapping
,
2279 struct writeback_control
*wbc
,
2280 struct mpage_da_data
*mpd
,
2281 pgoff_t
*done_index
)
2283 struct buffer_head
*bh
, *head
;
2284 struct inode
*inode
= mapping
->host
;
2285 struct pagevec pvec
;
2286 unsigned int nr_pages
;
2289 long nr_to_write
= wbc
->nr_to_write
;
2290 int i
, tag
, ret
= 0;
2292 memset(mpd
, 0, sizeof(struct mpage_da_data
));
2295 pagevec_init(&pvec
, 0);
2296 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2297 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2299 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2300 tag
= PAGECACHE_TAG_TOWRITE
;
2302 tag
= PAGECACHE_TAG_DIRTY
;
2304 *done_index
= index
;
2305 while (index
<= end
) {
2306 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2307 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2311 for (i
= 0; i
< nr_pages
; i
++) {
2312 struct page
*page
= pvec
.pages
[i
];
2315 * At this point, the page may be truncated or
2316 * invalidated (changing page->mapping to NULL), or
2317 * even swizzled back from swapper_space to tmpfs file
2318 * mapping. However, page->index will not change
2319 * because we have a reference on the page.
2321 if (page
->index
> end
)
2324 *done_index
= page
->index
+ 1;
2327 * If we can't merge this page, and we have
2328 * accumulated an contiguous region, write it
2330 if ((mpd
->next_page
!= page
->index
) &&
2331 (mpd
->next_page
!= mpd
->first_page
)) {
2332 mpage_da_map_and_submit(mpd
);
2333 goto ret_extent_tail
;
2339 * If the page is no longer dirty, or its
2340 * mapping no longer corresponds to inode we
2341 * are writing (which means it has been
2342 * truncated or invalidated), or the page is
2343 * already under writeback and we are not
2344 * doing a data integrity writeback, skip the page
2346 if (!PageDirty(page
) ||
2347 (PageWriteback(page
) &&
2348 (wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2349 unlikely(page
->mapping
!= mapping
)) {
2354 wait_on_page_writeback(page
);
2355 BUG_ON(PageWriteback(page
));
2358 * If we have inline data and arrive here, it means that
2359 * we will soon create the block for the 1st page, so
2360 * we'd better clear the inline data here.
2362 if (ext4_has_inline_data(inode
)) {
2363 BUG_ON(ext4_test_inode_state(inode
,
2364 EXT4_STATE_MAY_INLINE_DATA
));
2365 ext4_destroy_inline_data(handle
, inode
);
2368 if (mpd
->next_page
!= page
->index
)
2369 mpd
->first_page
= page
->index
;
2370 mpd
->next_page
= page
->index
+ 1;
2371 logical
= (sector_t
) page
->index
<<
2372 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2374 /* Add all dirty buffers to mpd */
2375 head
= page_buffers(page
);
2378 BUG_ON(buffer_locked(bh
));
2380 * We need to try to allocate unmapped blocks
2381 * in the same page. Otherwise we won't make
2382 * progress with the page in ext4_writepage
2384 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2385 mpage_add_bh_to_extent(mpd
, logical
,
2388 goto ret_extent_tail
;
2389 } else if (buffer_dirty(bh
) &&
2390 buffer_mapped(bh
)) {
2392 * mapped dirty buffer. We need to
2393 * update the b_state because we look
2394 * at b_state in mpage_da_map_blocks.
2395 * We don't update b_size because if we
2396 * find an unmapped buffer_head later
2397 * we need to use the b_state flag of
2400 if (mpd
->b_size
== 0)
2402 bh
->b_state
& BH_FLAGS
;
2405 } while ((bh
= bh
->b_this_page
) != head
);
2407 if (nr_to_write
> 0) {
2409 if (nr_to_write
== 0 &&
2410 wbc
->sync_mode
== WB_SYNC_NONE
)
2412 * We stop writing back only if we are
2413 * not doing integrity sync. In case of
2414 * integrity sync we have to keep going
2415 * because someone may be concurrently
2416 * dirtying pages, and we might have
2417 * synced a lot of newly appeared dirty
2418 * pages, but have not synced all of the
2424 pagevec_release(&pvec
);
2429 ret
= MPAGE_DA_EXTENT_TAIL
;
2431 pagevec_release(&pvec
);
2437 static int ext4_da_writepages(struct address_space
*mapping
,
2438 struct writeback_control
*wbc
)
2441 int range_whole
= 0;
2442 handle_t
*handle
= NULL
;
2443 struct mpage_da_data mpd
;
2444 struct inode
*inode
= mapping
->host
;
2445 int pages_written
= 0;
2446 unsigned int max_pages
;
2447 int range_cyclic
, cycled
= 1, io_done
= 0;
2448 int needed_blocks
, ret
= 0;
2449 long desired_nr_to_write
, nr_to_writebump
= 0;
2450 loff_t range_start
= wbc
->range_start
;
2451 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2452 pgoff_t done_index
= 0;
2454 struct blk_plug plug
;
2456 trace_ext4_da_writepages(inode
, wbc
);
2459 * No pages to write? This is mainly a kludge to avoid starting
2460 * a transaction for special inodes like journal inode on last iput()
2461 * because that could violate lock ordering on umount
2463 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2467 * If the filesystem has aborted, it is read-only, so return
2468 * right away instead of dumping stack traces later on that
2469 * will obscure the real source of the problem. We test
2470 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2471 * the latter could be true if the filesystem is mounted
2472 * read-only, and in that case, ext4_da_writepages should
2473 * *never* be called, so if that ever happens, we would want
2476 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2479 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2482 range_cyclic
= wbc
->range_cyclic
;
2483 if (wbc
->range_cyclic
) {
2484 index
= mapping
->writeback_index
;
2487 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2488 wbc
->range_end
= LLONG_MAX
;
2489 wbc
->range_cyclic
= 0;
2492 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2493 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2497 * This works around two forms of stupidity. The first is in
2498 * the writeback code, which caps the maximum number of pages
2499 * written to be 1024 pages. This is wrong on multiple
2500 * levels; different architectues have a different page size,
2501 * which changes the maximum amount of data which gets
2502 * written. Secondly, 4 megabytes is way too small. XFS
2503 * forces this value to be 16 megabytes by multiplying
2504 * nr_to_write parameter by four, and then relies on its
2505 * allocator to allocate larger extents to make them
2506 * contiguous. Unfortunately this brings us to the second
2507 * stupidity, which is that ext4's mballoc code only allocates
2508 * at most 2048 blocks. So we force contiguous writes up to
2509 * the number of dirty blocks in the inode, or
2510 * sbi->max_writeback_mb_bump whichever is smaller.
2512 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2513 if (!range_cyclic
&& range_whole
) {
2514 if (wbc
->nr_to_write
== LONG_MAX
)
2515 desired_nr_to_write
= wbc
->nr_to_write
;
2517 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2519 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2521 if (desired_nr_to_write
> max_pages
)
2522 desired_nr_to_write
= max_pages
;
2524 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2525 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2526 wbc
->nr_to_write
= desired_nr_to_write
;
2530 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2531 tag_pages_for_writeback(mapping
, index
, end
);
2533 blk_start_plug(&plug
);
2534 while (!ret
&& wbc
->nr_to_write
> 0) {
2537 * we insert one extent at a time. So we need
2538 * credit needed for single extent allocation.
2539 * journalled mode is currently not supported
2542 BUG_ON(ext4_should_journal_data(inode
));
2543 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2545 /* start a new transaction*/
2546 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2548 if (IS_ERR(handle
)) {
2549 ret
= PTR_ERR(handle
);
2550 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2551 "%ld pages, ino %lu; err %d", __func__
,
2552 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2553 blk_finish_plug(&plug
);
2554 goto out_writepages
;
2558 * Now call write_cache_pages_da() to find the next
2559 * contiguous region of logical blocks that need
2560 * blocks to be allocated by ext4 and submit them.
2562 ret
= write_cache_pages_da(handle
, mapping
,
2563 wbc
, &mpd
, &done_index
);
2565 * If we have a contiguous extent of pages and we
2566 * haven't done the I/O yet, map the blocks and submit
2569 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2570 mpage_da_map_and_submit(&mpd
);
2571 ret
= MPAGE_DA_EXTENT_TAIL
;
2573 trace_ext4_da_write_pages(inode
, &mpd
);
2574 wbc
->nr_to_write
-= mpd
.pages_written
;
2576 ext4_journal_stop(handle
);
2578 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2579 /* commit the transaction which would
2580 * free blocks released in the transaction
2583 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2585 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2587 * Got one extent now try with rest of the pages.
2588 * If mpd.retval is set -EIO, journal is aborted.
2589 * So we don't need to write any more.
2591 pages_written
+= mpd
.pages_written
;
2594 } else if (wbc
->nr_to_write
)
2596 * There is no more writeout needed
2597 * or we requested for a noblocking writeout
2598 * and we found the device congested
2602 blk_finish_plug(&plug
);
2603 if (!io_done
&& !cycled
) {
2606 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2607 wbc
->range_end
= mapping
->writeback_index
- 1;
2612 wbc
->range_cyclic
= range_cyclic
;
2613 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2615 * set the writeback_index so that range_cyclic
2616 * mode will write it back later
2618 mapping
->writeback_index
= done_index
;
2621 wbc
->nr_to_write
-= nr_to_writebump
;
2622 wbc
->range_start
= range_start
;
2623 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2627 static int ext4_nonda_switch(struct super_block
*sb
)
2629 s64 free_clusters
, dirty_clusters
;
2630 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2633 * switch to non delalloc mode if we are running low
2634 * on free block. The free block accounting via percpu
2635 * counters can get slightly wrong with percpu_counter_batch getting
2636 * accumulated on each CPU without updating global counters
2637 * Delalloc need an accurate free block accounting. So switch
2638 * to non delalloc when we are near to error range.
2641 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2643 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2645 * Start pushing delalloc when 1/2 of free blocks are dirty.
2647 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2648 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2650 if (2 * free_clusters
< 3 * dirty_clusters
||
2651 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2653 * free block count is less than 150% of dirty blocks
2654 * or free blocks is less than watermark
2661 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2662 loff_t pos
, unsigned len
, unsigned flags
,
2663 struct page
**pagep
, void **fsdata
)
2665 int ret
, retries
= 0;
2668 struct inode
*inode
= mapping
->host
;
2671 index
= pos
>> PAGE_CACHE_SHIFT
;
2673 if (ext4_nonda_switch(inode
->i_sb
)) {
2674 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2675 return ext4_write_begin(file
, mapping
, pos
,
2676 len
, flags
, pagep
, fsdata
);
2678 *fsdata
= (void *)0;
2679 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2681 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2682 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2692 * grab_cache_page_write_begin() can take a long time if the
2693 * system is thrashing due to memory pressure, or if the page
2694 * is being written back. So grab it first before we start
2695 * the transaction handle. This also allows us to allocate
2696 * the page (if needed) without using GFP_NOFS.
2699 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2705 * With delayed allocation, we don't log the i_disksize update
2706 * if there is delayed block allocation. But we still need
2707 * to journalling the i_disksize update if writes to the end
2708 * of file which has an already mapped buffer.
2711 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, 1);
2712 if (IS_ERR(handle
)) {
2713 page_cache_release(page
);
2714 return PTR_ERR(handle
);
2718 if (page
->mapping
!= mapping
) {
2719 /* The page got truncated from under us */
2721 page_cache_release(page
);
2722 ext4_journal_stop(handle
);
2725 /* In case writeback began while the page was unlocked */
2726 wait_on_page_writeback(page
);
2728 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2731 ext4_journal_stop(handle
);
2733 * block_write_begin may have instantiated a few blocks
2734 * outside i_size. Trim these off again. Don't need
2735 * i_size_read because we hold i_mutex.
2737 if (pos
+ len
> inode
->i_size
)
2738 ext4_truncate_failed_write(inode
);
2740 if (ret
== -ENOSPC
&&
2741 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2744 page_cache_release(page
);
2753 * Check if we should update i_disksize
2754 * when write to the end of file but not require block allocation
2756 static int ext4_da_should_update_i_disksize(struct page
*page
,
2757 unsigned long offset
)
2759 struct buffer_head
*bh
;
2760 struct inode
*inode
= page
->mapping
->host
;
2764 bh
= page_buffers(page
);
2765 idx
= offset
>> inode
->i_blkbits
;
2767 for (i
= 0; i
< idx
; i
++)
2768 bh
= bh
->b_this_page
;
2770 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2775 static int ext4_da_write_end(struct file
*file
,
2776 struct address_space
*mapping
,
2777 loff_t pos
, unsigned len
, unsigned copied
,
2778 struct page
*page
, void *fsdata
)
2780 struct inode
*inode
= mapping
->host
;
2782 handle_t
*handle
= ext4_journal_current_handle();
2784 unsigned long start
, end
;
2785 int write_mode
= (int)(unsigned long)fsdata
;
2787 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
2788 return ext4_write_end(file
, mapping
, pos
,
2789 len
, copied
, page
, fsdata
);
2791 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2792 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2793 end
= start
+ copied
- 1;
2796 * generic_write_end() will run mark_inode_dirty() if i_size
2797 * changes. So let's piggyback the i_disksize mark_inode_dirty
2800 new_i_size
= pos
+ copied
;
2801 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2802 if (ext4_has_inline_data(inode
) ||
2803 ext4_da_should_update_i_disksize(page
, end
)) {
2804 down_write(&EXT4_I(inode
)->i_data_sem
);
2805 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
2806 EXT4_I(inode
)->i_disksize
= new_i_size
;
2807 up_write(&EXT4_I(inode
)->i_data_sem
);
2808 /* We need to mark inode dirty even if
2809 * new_i_size is less that inode->i_size
2810 * bu greater than i_disksize.(hint delalloc)
2812 ext4_mark_inode_dirty(handle
, inode
);
2816 if (write_mode
!= CONVERT_INLINE_DATA
&&
2817 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
2818 ext4_has_inline_data(inode
))
2819 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
2822 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2828 ret2
= ext4_journal_stop(handle
);
2832 return ret
? ret
: copied
;
2835 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2838 * Drop reserved blocks
2840 BUG_ON(!PageLocked(page
));
2841 if (!page_has_buffers(page
))
2844 ext4_da_page_release_reservation(page
, offset
);
2847 ext4_invalidatepage(page
, offset
);
2853 * Force all delayed allocation blocks to be allocated for a given inode.
2855 int ext4_alloc_da_blocks(struct inode
*inode
)
2857 trace_ext4_alloc_da_blocks(inode
);
2859 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2860 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2864 * We do something simple for now. The filemap_flush() will
2865 * also start triggering a write of the data blocks, which is
2866 * not strictly speaking necessary (and for users of
2867 * laptop_mode, not even desirable). However, to do otherwise
2868 * would require replicating code paths in:
2870 * ext4_da_writepages() ->
2871 * write_cache_pages() ---> (via passed in callback function)
2872 * __mpage_da_writepage() -->
2873 * mpage_add_bh_to_extent()
2874 * mpage_da_map_blocks()
2876 * The problem is that write_cache_pages(), located in
2877 * mm/page-writeback.c, marks pages clean in preparation for
2878 * doing I/O, which is not desirable if we're not planning on
2881 * We could call write_cache_pages(), and then redirty all of
2882 * the pages by calling redirty_page_for_writepage() but that
2883 * would be ugly in the extreme. So instead we would need to
2884 * replicate parts of the code in the above functions,
2885 * simplifying them because we wouldn't actually intend to
2886 * write out the pages, but rather only collect contiguous
2887 * logical block extents, call the multi-block allocator, and
2888 * then update the buffer heads with the block allocations.
2890 * For now, though, we'll cheat by calling filemap_flush(),
2891 * which will map the blocks, and start the I/O, but not
2892 * actually wait for the I/O to complete.
2894 return filemap_flush(inode
->i_mapping
);
2898 * bmap() is special. It gets used by applications such as lilo and by
2899 * the swapper to find the on-disk block of a specific piece of data.
2901 * Naturally, this is dangerous if the block concerned is still in the
2902 * journal. If somebody makes a swapfile on an ext4 data-journaling
2903 * filesystem and enables swap, then they may get a nasty shock when the
2904 * data getting swapped to that swapfile suddenly gets overwritten by
2905 * the original zero's written out previously to the journal and
2906 * awaiting writeback in the kernel's buffer cache.
2908 * So, if we see any bmap calls here on a modified, data-journaled file,
2909 * take extra steps to flush any blocks which might be in the cache.
2911 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2913 struct inode
*inode
= mapping
->host
;
2918 * We can get here for an inline file via the FIBMAP ioctl
2920 if (ext4_has_inline_data(inode
))
2923 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2924 test_opt(inode
->i_sb
, DELALLOC
)) {
2926 * With delalloc we want to sync the file
2927 * so that we can make sure we allocate
2930 filemap_write_and_wait(mapping
);
2933 if (EXT4_JOURNAL(inode
) &&
2934 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2936 * This is a REALLY heavyweight approach, but the use of
2937 * bmap on dirty files is expected to be extremely rare:
2938 * only if we run lilo or swapon on a freshly made file
2939 * do we expect this to happen.
2941 * (bmap requires CAP_SYS_RAWIO so this does not
2942 * represent an unprivileged user DOS attack --- we'd be
2943 * in trouble if mortal users could trigger this path at
2946 * NB. EXT4_STATE_JDATA is not set on files other than
2947 * regular files. If somebody wants to bmap a directory
2948 * or symlink and gets confused because the buffer
2949 * hasn't yet been flushed to disk, they deserve
2950 * everything they get.
2953 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2954 journal
= EXT4_JOURNAL(inode
);
2955 jbd2_journal_lock_updates(journal
);
2956 err
= jbd2_journal_flush(journal
);
2957 jbd2_journal_unlock_updates(journal
);
2963 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2966 static int ext4_readpage(struct file
*file
, struct page
*page
)
2969 struct inode
*inode
= page
->mapping
->host
;
2971 trace_ext4_readpage(page
);
2973 if (ext4_has_inline_data(inode
))
2974 ret
= ext4_readpage_inline(inode
, page
);
2977 return mpage_readpage(page
, ext4_get_block
);
2983 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2984 struct list_head
*pages
, unsigned nr_pages
)
2986 struct inode
*inode
= mapping
->host
;
2988 /* If the file has inline data, no need to do readpages. */
2989 if (ext4_has_inline_data(inode
))
2992 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2995 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2997 trace_ext4_invalidatepage(page
, offset
);
2999 /* No journalling happens on data buffers when this function is used */
3000 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3002 block_invalidatepage(page
, offset
);
3005 static int __ext4_journalled_invalidatepage(struct page
*page
,
3006 unsigned long offset
)
3008 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3010 trace_ext4_journalled_invalidatepage(page
, offset
);
3013 * If it's a full truncate we just forget about the pending dirtying
3016 ClearPageChecked(page
);
3018 return jbd2_journal_invalidatepage(journal
, page
, offset
);
3021 /* Wrapper for aops... */
3022 static void ext4_journalled_invalidatepage(struct page
*page
,
3023 unsigned long offset
)
3025 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
) < 0);
3028 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3030 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3032 trace_ext4_releasepage(page
);
3034 /* Page has dirty journalled data -> cannot release */
3035 if (PageChecked(page
))
3038 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3040 return try_to_free_buffers(page
);
3044 * ext4_get_block used when preparing for a DIO write or buffer write.
3045 * We allocate an uinitialized extent if blocks haven't been allocated.
3046 * The extent will be converted to initialized after the IO is complete.
3048 int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
3049 struct buffer_head
*bh_result
, int create
)
3051 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3052 inode
->i_ino
, create
);
3053 return _ext4_get_block(inode
, iblock
, bh_result
,
3054 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
3057 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
3058 struct buffer_head
*bh_result
, int create
)
3060 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3061 inode
->i_ino
, create
);
3062 return _ext4_get_block(inode
, iblock
, bh_result
,
3063 EXT4_GET_BLOCKS_NO_LOCK
);
3066 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3067 ssize_t size
, void *private, int ret
,
3070 struct inode
*inode
= file_inode(iocb
->ki_filp
);
3071 ext4_io_end_t
*io_end
= iocb
->private;
3073 /* if not async direct IO or dio with 0 bytes write, just return */
3074 if (!io_end
|| !size
)
3077 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3078 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3079 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3082 iocb
->private = NULL
;
3084 /* if not aio dio with unwritten extents, just free io and return */
3085 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
3086 ext4_free_io_end(io_end
);
3088 inode_dio_done(inode
);
3090 aio_complete(iocb
, ret
, 0);
3094 io_end
->offset
= offset
;
3095 io_end
->size
= size
;
3097 io_end
->iocb
= iocb
;
3098 io_end
->result
= ret
;
3101 ext4_add_complete_io(io_end
);
3105 * For ext4 extent files, ext4 will do direct-io write to holes,
3106 * preallocated extents, and those write extend the file, no need to
3107 * fall back to buffered IO.
3109 * For holes, we fallocate those blocks, mark them as uninitialized
3110 * If those blocks were preallocated, we mark sure they are split, but
3111 * still keep the range to write as uninitialized.
3113 * The unwritten extents will be converted to written when DIO is completed.
3114 * For async direct IO, since the IO may still pending when return, we
3115 * set up an end_io call back function, which will do the conversion
3116 * when async direct IO completed.
3118 * If the O_DIRECT write will extend the file then add this inode to the
3119 * orphan list. So recovery will truncate it back to the original size
3120 * if the machine crashes during the write.
3123 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3124 const struct iovec
*iov
, loff_t offset
,
3125 unsigned long nr_segs
)
3127 struct file
*file
= iocb
->ki_filp
;
3128 struct inode
*inode
= file
->f_mapping
->host
;
3130 size_t count
= iov_length(iov
, nr_segs
);
3132 get_block_t
*get_block_func
= NULL
;
3134 loff_t final_size
= offset
+ count
;
3136 /* Use the old path for reads and writes beyond i_size. */
3137 if (rw
!= WRITE
|| final_size
> inode
->i_size
)
3138 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3140 BUG_ON(iocb
->private == NULL
);
3142 /* If we do a overwrite dio, i_mutex locking can be released */
3143 overwrite
= *((int *)iocb
->private);
3146 atomic_inc(&inode
->i_dio_count
);
3147 down_read(&EXT4_I(inode
)->i_data_sem
);
3148 mutex_unlock(&inode
->i_mutex
);
3152 * We could direct write to holes and fallocate.
3154 * Allocated blocks to fill the hole are marked as
3155 * uninitialized to prevent parallel buffered read to expose
3156 * the stale data before DIO complete the data IO.
3158 * As to previously fallocated extents, ext4 get_block will
3159 * just simply mark the buffer mapped but still keep the
3160 * extents uninitialized.
3162 * For non AIO case, we will convert those unwritten extents
3163 * to written after return back from blockdev_direct_IO.
3165 * For async DIO, the conversion needs to be deferred when the
3166 * IO is completed. The ext4 end_io callback function will be
3167 * called to take care of the conversion work. Here for async
3168 * case, we allocate an io_end structure to hook to the iocb.
3170 iocb
->private = NULL
;
3171 ext4_inode_aio_set(inode
, NULL
);
3172 if (!is_sync_kiocb(iocb
)) {
3173 ext4_io_end_t
*io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
3178 io_end
->flag
|= EXT4_IO_END_DIRECT
;
3179 iocb
->private = io_end
;
3181 * we save the io structure for current async direct
3182 * IO, so that later ext4_map_blocks() could flag the
3183 * io structure whether there is a unwritten extents
3184 * needs to be converted when IO is completed.
3186 ext4_inode_aio_set(inode
, io_end
);
3190 get_block_func
= ext4_get_block_write_nolock
;
3192 get_block_func
= ext4_get_block_write
;
3193 dio_flags
= DIO_LOCKING
;
3195 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3196 inode
->i_sb
->s_bdev
, iov
,
3204 ext4_inode_aio_set(inode
, NULL
);
3206 * The io_end structure takes a reference to the inode, that
3207 * structure needs to be destroyed and the reference to the
3208 * inode need to be dropped, when IO is complete, even with 0
3209 * byte write, or failed.
3211 * In the successful AIO DIO case, the io_end structure will
3212 * be destroyed and the reference to the inode will be dropped
3213 * after the end_io call back function is called.
3215 * In the case there is 0 byte write, or error case, since VFS
3216 * direct IO won't invoke the end_io call back function, we
3217 * need to free the end_io structure here.
3219 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3220 ext4_free_io_end(iocb
->private);
3221 iocb
->private = NULL
;
3222 } else if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3223 EXT4_STATE_DIO_UNWRITTEN
)) {
3226 * for non AIO case, since the IO is already
3227 * completed, we could do the conversion right here
3229 err
= ext4_convert_unwritten_extents(inode
,
3233 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3237 /* take i_mutex locking again if we do a ovewrite dio */
3239 inode_dio_done(inode
);
3240 up_read(&EXT4_I(inode
)->i_data_sem
);
3241 mutex_lock(&inode
->i_mutex
);
3247 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3248 const struct iovec
*iov
, loff_t offset
,
3249 unsigned long nr_segs
)
3251 struct file
*file
= iocb
->ki_filp
;
3252 struct inode
*inode
= file
->f_mapping
->host
;
3256 * If we are doing data journalling we don't support O_DIRECT
3258 if (ext4_should_journal_data(inode
))
3261 /* Let buffer I/O handle the inline data case. */
3262 if (ext4_has_inline_data(inode
))
3265 trace_ext4_direct_IO_enter(inode
, offset
, iov_length(iov
, nr_segs
), rw
);
3266 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3267 ret
= ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3269 ret
= ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3270 trace_ext4_direct_IO_exit(inode
, offset
,
3271 iov_length(iov
, nr_segs
), rw
, ret
);
3276 * Pages can be marked dirty completely asynchronously from ext4's journalling
3277 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3278 * much here because ->set_page_dirty is called under VFS locks. The page is
3279 * not necessarily locked.
3281 * We cannot just dirty the page and leave attached buffers clean, because the
3282 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3283 * or jbddirty because all the journalling code will explode.
3285 * So what we do is to mark the page "pending dirty" and next time writepage
3286 * is called, propagate that into the buffers appropriately.
3288 static int ext4_journalled_set_page_dirty(struct page
*page
)
3290 SetPageChecked(page
);
3291 return __set_page_dirty_nobuffers(page
);
3294 static const struct address_space_operations ext4_aops
= {
3295 .readpage
= ext4_readpage
,
3296 .readpages
= ext4_readpages
,
3297 .writepage
= ext4_writepage
,
3298 .write_begin
= ext4_write_begin
,
3299 .write_end
= ext4_write_end
,
3301 .invalidatepage
= ext4_invalidatepage
,
3302 .releasepage
= ext4_releasepage
,
3303 .direct_IO
= ext4_direct_IO
,
3304 .migratepage
= buffer_migrate_page
,
3305 .is_partially_uptodate
= block_is_partially_uptodate
,
3306 .error_remove_page
= generic_error_remove_page
,
3309 static const struct address_space_operations ext4_journalled_aops
= {
3310 .readpage
= ext4_readpage
,
3311 .readpages
= ext4_readpages
,
3312 .writepage
= ext4_writepage
,
3313 .write_begin
= ext4_write_begin
,
3314 .write_end
= ext4_journalled_write_end
,
3315 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3317 .invalidatepage
= ext4_journalled_invalidatepage
,
3318 .releasepage
= ext4_releasepage
,
3319 .direct_IO
= ext4_direct_IO
,
3320 .is_partially_uptodate
= block_is_partially_uptodate
,
3321 .error_remove_page
= generic_error_remove_page
,
3324 static const struct address_space_operations ext4_da_aops
= {
3325 .readpage
= ext4_readpage
,
3326 .readpages
= ext4_readpages
,
3327 .writepage
= ext4_writepage
,
3328 .writepages
= ext4_da_writepages
,
3329 .write_begin
= ext4_da_write_begin
,
3330 .write_end
= ext4_da_write_end
,
3332 .invalidatepage
= ext4_da_invalidatepage
,
3333 .releasepage
= ext4_releasepage
,
3334 .direct_IO
= ext4_direct_IO
,
3335 .migratepage
= buffer_migrate_page
,
3336 .is_partially_uptodate
= block_is_partially_uptodate
,
3337 .error_remove_page
= generic_error_remove_page
,
3340 void ext4_set_aops(struct inode
*inode
)
3342 switch (ext4_inode_journal_mode(inode
)) {
3343 case EXT4_INODE_ORDERED_DATA_MODE
:
3344 ext4_set_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3346 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3347 ext4_clear_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3349 case EXT4_INODE_JOURNAL_DATA_MODE
:
3350 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3355 if (test_opt(inode
->i_sb
, DELALLOC
))
3356 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3358 inode
->i_mapping
->a_ops
= &ext4_aops
;
3363 * ext4_discard_partial_page_buffers()
3364 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3365 * This function finds and locks the page containing the offset
3366 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3367 * Calling functions that already have the page locked should call
3368 * ext4_discard_partial_page_buffers_no_lock directly.
3370 int ext4_discard_partial_page_buffers(handle_t
*handle
,
3371 struct address_space
*mapping
, loff_t from
,
3372 loff_t length
, int flags
)
3374 struct inode
*inode
= mapping
->host
;
3378 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3379 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3383 err
= ext4_discard_partial_page_buffers_no_lock(handle
, inode
, page
,
3384 from
, length
, flags
);
3387 page_cache_release(page
);
3392 * ext4_discard_partial_page_buffers_no_lock()
3393 * Zeros a page range of length 'length' starting from offset 'from'.
3394 * Buffer heads that correspond to the block aligned regions of the
3395 * zeroed range will be unmapped. Unblock aligned regions
3396 * will have the corresponding buffer head mapped if needed so that
3397 * that region of the page can be updated with the partial zero out.
3399 * This function assumes that the page has already been locked. The
3400 * The range to be discarded must be contained with in the given page.
3401 * If the specified range exceeds the end of the page it will be shortened
3402 * to the end of the page that corresponds to 'from'. This function is
3403 * appropriate for updating a page and it buffer heads to be unmapped and
3404 * zeroed for blocks that have been either released, or are going to be
3407 * handle: The journal handle
3408 * inode: The files inode
3409 * page: A locked page that contains the offset "from"
3410 * from: The starting byte offset (from the beginning of the file)
3411 * to begin discarding
3412 * len: The length of bytes to discard
3413 * flags: Optional flags that may be used:
3415 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3416 * Only zero the regions of the page whose buffer heads
3417 * have already been unmapped. This flag is appropriate
3418 * for updating the contents of a page whose blocks may
3419 * have already been released, and we only want to zero
3420 * out the regions that correspond to those released blocks.
3422 * Returns zero on success or negative on failure.
3424 static int ext4_discard_partial_page_buffers_no_lock(handle_t
*handle
,
3425 struct inode
*inode
, struct page
*page
, loff_t from
,
3426 loff_t length
, int flags
)
3428 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3429 unsigned int offset
= from
& (PAGE_CACHE_SIZE
-1);
3430 unsigned int blocksize
, max
, pos
;
3432 struct buffer_head
*bh
;
3435 blocksize
= inode
->i_sb
->s_blocksize
;
3436 max
= PAGE_CACHE_SIZE
- offset
;
3438 if (index
!= page
->index
)
3442 * correct length if it does not fall between
3443 * 'from' and the end of the page
3445 if (length
> max
|| length
< 0)
3448 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3450 if (!page_has_buffers(page
))
3451 create_empty_buffers(page
, blocksize
, 0);
3453 /* Find the buffer that contains "offset" */
3454 bh
= page_buffers(page
);
3456 while (offset
>= pos
) {
3457 bh
= bh
->b_this_page
;
3463 while (pos
< offset
+ length
) {
3464 unsigned int end_of_block
, range_to_discard
;
3468 /* The length of space left to zero and unmap */
3469 range_to_discard
= offset
+ length
- pos
;
3471 /* The length of space until the end of the block */
3472 end_of_block
= blocksize
- (pos
& (blocksize
-1));
3475 * Do not unmap or zero past end of block
3476 * for this buffer head
3478 if (range_to_discard
> end_of_block
)
3479 range_to_discard
= end_of_block
;
3483 * Skip this buffer head if we are only zeroing unampped
3484 * regions of the page
3486 if (flags
& EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
&&
3490 /* If the range is block aligned, unmap */
3491 if (range_to_discard
== blocksize
) {
3492 clear_buffer_dirty(bh
);
3494 clear_buffer_mapped(bh
);
3495 clear_buffer_req(bh
);
3496 clear_buffer_new(bh
);
3497 clear_buffer_delay(bh
);
3498 clear_buffer_unwritten(bh
);
3499 clear_buffer_uptodate(bh
);
3500 zero_user(page
, pos
, range_to_discard
);
3501 BUFFER_TRACE(bh
, "Buffer discarded");
3506 * If this block is not completely contained in the range
3507 * to be discarded, then it is not going to be released. Because
3508 * we need to keep this block, we need to make sure this part
3509 * of the page is uptodate before we modify it by writeing
3510 * partial zeros on it.
3512 if (!buffer_mapped(bh
)) {
3514 * Buffer head must be mapped before we can read
3517 BUFFER_TRACE(bh
, "unmapped");
3518 ext4_get_block(inode
, iblock
, bh
, 0);
3519 /* unmapped? It's a hole - nothing to do */
3520 if (!buffer_mapped(bh
)) {
3521 BUFFER_TRACE(bh
, "still unmapped");
3526 /* Ok, it's mapped. Make sure it's up-to-date */
3527 if (PageUptodate(page
))
3528 set_buffer_uptodate(bh
);
3530 if (!buffer_uptodate(bh
)) {
3532 ll_rw_block(READ
, 1, &bh
);
3534 /* Uhhuh. Read error. Complain and punt.*/
3535 if (!buffer_uptodate(bh
))
3539 if (ext4_should_journal_data(inode
)) {
3540 BUFFER_TRACE(bh
, "get write access");
3541 err
= ext4_journal_get_write_access(handle
, bh
);
3546 zero_user(page
, pos
, range_to_discard
);
3549 if (ext4_should_journal_data(inode
)) {
3550 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3552 mark_buffer_dirty(bh
);
3554 BUFFER_TRACE(bh
, "Partial buffer zeroed");
3556 bh
= bh
->b_this_page
;
3558 pos
+= range_to_discard
;
3564 int ext4_can_truncate(struct inode
*inode
)
3566 if (S_ISREG(inode
->i_mode
))
3568 if (S_ISDIR(inode
->i_mode
))
3570 if (S_ISLNK(inode
->i_mode
))
3571 return !ext4_inode_is_fast_symlink(inode
);
3576 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3577 * associated with the given offset and length
3579 * @inode: File inode
3580 * @offset: The offset where the hole will begin
3581 * @len: The length of the hole
3583 * Returns: 0 on success or negative on failure
3586 int ext4_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
3588 struct inode
*inode
= file_inode(file
);
3589 struct super_block
*sb
= inode
->i_sb
;
3590 ext4_lblk_t first_block
, stop_block
;
3591 struct address_space
*mapping
= inode
->i_mapping
;
3592 loff_t first_page
, last_page
, page_len
;
3593 loff_t first_page_offset
, last_page_offset
;
3595 unsigned int credits
;
3598 if (!S_ISREG(inode
->i_mode
))
3601 if (EXT4_SB(sb
)->s_cluster_ratio
> 1) {
3602 /* TODO: Add support for bigalloc file systems */
3606 trace_ext4_punch_hole(inode
, offset
, length
);
3609 * Write out all dirty pages to avoid race conditions
3610 * Then release them.
3612 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3613 ret
= filemap_write_and_wait_range(mapping
, offset
,
3614 offset
+ length
- 1);
3619 mutex_lock(&inode
->i_mutex
);
3620 /* It's not possible punch hole on append only file */
3621 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
)) {
3625 if (IS_SWAPFILE(inode
)) {
3630 /* No need to punch hole beyond i_size */
3631 if (offset
>= inode
->i_size
)
3635 * If the hole extends beyond i_size, set the hole
3636 * to end after the page that contains i_size
3638 if (offset
+ length
> inode
->i_size
) {
3639 length
= inode
->i_size
+
3640 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
3644 first_page
= (offset
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
3645 last_page
= (offset
+ length
) >> PAGE_CACHE_SHIFT
;
3647 first_page_offset
= first_page
<< PAGE_CACHE_SHIFT
;
3648 last_page_offset
= last_page
<< PAGE_CACHE_SHIFT
;
3650 /* Now release the pages */
3651 if (last_page_offset
> first_page_offset
) {
3652 truncate_pagecache_range(inode
, first_page_offset
,
3653 last_page_offset
- 1);
3656 /* Wait all existing dio workers, newcomers will block on i_mutex */
3657 ext4_inode_block_unlocked_dio(inode
);
3658 ret
= ext4_flush_unwritten_io(inode
);
3661 inode_dio_wait(inode
);
3663 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3664 credits
= ext4_writepage_trans_blocks(inode
);
3666 credits
= ext4_blocks_for_truncate(inode
);
3667 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3668 if (IS_ERR(handle
)) {
3669 ret
= PTR_ERR(handle
);
3670 ext4_std_error(sb
, ret
);
3675 * Now we need to zero out the non-page-aligned data in the
3676 * pages at the start and tail of the hole, and unmap the
3677 * buffer heads for the block aligned regions of the page that
3678 * were completely zeroed.
3680 if (first_page
> last_page
) {
3682 * If the file space being truncated is contained
3683 * within a page just zero out and unmap the middle of
3686 ret
= ext4_discard_partial_page_buffers(handle
,
3687 mapping
, offset
, length
, 0);
3693 * zero out and unmap the partial page that contains
3694 * the start of the hole
3696 page_len
= first_page_offset
- offset
;
3698 ret
= ext4_discard_partial_page_buffers(handle
, mapping
,
3699 offset
, page_len
, 0);
3705 * zero out and unmap the partial page that contains
3706 * the end of the hole
3708 page_len
= offset
+ length
- last_page_offset
;
3710 ret
= ext4_discard_partial_page_buffers(handle
, mapping
,
3711 last_page_offset
, page_len
, 0);
3718 * If i_size is contained in the last page, we need to
3719 * unmap and zero the partial page after i_size
3721 if (inode
->i_size
>> PAGE_CACHE_SHIFT
== last_page
&&
3722 inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
3723 page_len
= PAGE_CACHE_SIZE
-
3724 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
3727 ret
= ext4_discard_partial_page_buffers(handle
,
3728 mapping
, inode
->i_size
, page_len
, 0);
3735 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
3736 EXT4_BLOCK_SIZE_BITS(sb
);
3737 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
3739 /* If there are no blocks to remove, return now */
3740 if (first_block
>= stop_block
)
3743 down_write(&EXT4_I(inode
)->i_data_sem
);
3744 ext4_discard_preallocations(inode
);
3746 ret
= ext4_es_remove_extent(inode
, first_block
,
3747 stop_block
- first_block
);
3749 up_write(&EXT4_I(inode
)->i_data_sem
);
3753 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3754 ret
= ext4_ext_remove_space(inode
, first_block
,
3757 ret
= ext4_free_hole_blocks(handle
, inode
, first_block
,
3760 ext4_discard_preallocations(inode
);
3761 up_write(&EXT4_I(inode
)->i_data_sem
);
3763 ext4_handle_sync(handle
);
3764 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3765 ext4_mark_inode_dirty(handle
, inode
);
3767 ext4_journal_stop(handle
);
3769 ext4_inode_resume_unlocked_dio(inode
);
3771 mutex_unlock(&inode
->i_mutex
);
3778 * We block out ext4_get_block() block instantiations across the entire
3779 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3780 * simultaneously on behalf of the same inode.
3782 * As we work through the truncate and commit bits of it to the journal there
3783 * is one core, guiding principle: the file's tree must always be consistent on
3784 * disk. We must be able to restart the truncate after a crash.
3786 * The file's tree may be transiently inconsistent in memory (although it
3787 * probably isn't), but whenever we close off and commit a journal transaction,
3788 * the contents of (the filesystem + the journal) must be consistent and
3789 * restartable. It's pretty simple, really: bottom up, right to left (although
3790 * left-to-right works OK too).
3792 * Note that at recovery time, journal replay occurs *before* the restart of
3793 * truncate against the orphan inode list.
3795 * The committed inode has the new, desired i_size (which is the same as
3796 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3797 * that this inode's truncate did not complete and it will again call
3798 * ext4_truncate() to have another go. So there will be instantiated blocks
3799 * to the right of the truncation point in a crashed ext4 filesystem. But
3800 * that's fine - as long as they are linked from the inode, the post-crash
3801 * ext4_truncate() run will find them and release them.
3803 void ext4_truncate(struct inode
*inode
)
3805 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3806 unsigned int credits
;
3808 struct address_space
*mapping
= inode
->i_mapping
;
3812 * There is a possibility that we're either freeing the inode
3813 * or it completely new indode. In those cases we might not
3814 * have i_mutex locked because it's not necessary.
3816 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
3817 WARN_ON(!mutex_is_locked(&inode
->i_mutex
));
3818 trace_ext4_truncate_enter(inode
);
3820 if (!ext4_can_truncate(inode
))
3823 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3825 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3826 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3828 if (ext4_has_inline_data(inode
)) {
3831 ext4_inline_data_truncate(inode
, &has_inline
);
3837 * finish any pending end_io work so we won't run the risk of
3838 * converting any truncated blocks to initialized later
3840 ext4_flush_unwritten_io(inode
);
3842 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3843 credits
= ext4_writepage_trans_blocks(inode
);
3845 credits
= ext4_blocks_for_truncate(inode
);
3847 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3848 if (IS_ERR(handle
)) {
3849 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
3853 if (inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
3854 page_len
= PAGE_CACHE_SIZE
-
3855 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
3857 if (ext4_discard_partial_page_buffers(handle
,
3858 mapping
, inode
->i_size
, page_len
, 0))
3863 * We add the inode to the orphan list, so that if this
3864 * truncate spans multiple transactions, and we crash, we will
3865 * resume the truncate when the filesystem recovers. It also
3866 * marks the inode dirty, to catch the new size.
3868 * Implication: the file must always be in a sane, consistent
3869 * truncatable state while each transaction commits.
3871 if (ext4_orphan_add(handle
, inode
))
3874 down_write(&EXT4_I(inode
)->i_data_sem
);
3876 ext4_discard_preallocations(inode
);
3878 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3879 ext4_ext_truncate(handle
, inode
);
3881 ext4_ind_truncate(handle
, inode
);
3883 up_write(&ei
->i_data_sem
);
3886 ext4_handle_sync(handle
);
3890 * If this was a simple ftruncate() and the file will remain alive,
3891 * then we need to clear up the orphan record which we created above.
3892 * However, if this was a real unlink then we were called by
3893 * ext4_delete_inode(), and we allow that function to clean up the
3894 * orphan info for us.
3897 ext4_orphan_del(handle
, inode
);
3899 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3900 ext4_mark_inode_dirty(handle
, inode
);
3901 ext4_journal_stop(handle
);
3903 trace_ext4_truncate_exit(inode
);
3907 * ext4_get_inode_loc returns with an extra refcount against the inode's
3908 * underlying buffer_head on success. If 'in_mem' is true, we have all
3909 * data in memory that is needed to recreate the on-disk version of this
3912 static int __ext4_get_inode_loc(struct inode
*inode
,
3913 struct ext4_iloc
*iloc
, int in_mem
)
3915 struct ext4_group_desc
*gdp
;
3916 struct buffer_head
*bh
;
3917 struct super_block
*sb
= inode
->i_sb
;
3919 int inodes_per_block
, inode_offset
;
3922 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3925 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3926 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3931 * Figure out the offset within the block group inode table
3933 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3934 inode_offset
= ((inode
->i_ino
- 1) %
3935 EXT4_INODES_PER_GROUP(sb
));
3936 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3937 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3939 bh
= sb_getblk(sb
, block
);
3942 if (!buffer_uptodate(bh
)) {
3946 * If the buffer has the write error flag, we have failed
3947 * to write out another inode in the same block. In this
3948 * case, we don't have to read the block because we may
3949 * read the old inode data successfully.
3951 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3952 set_buffer_uptodate(bh
);
3954 if (buffer_uptodate(bh
)) {
3955 /* someone brought it uptodate while we waited */
3961 * If we have all information of the inode in memory and this
3962 * is the only valid inode in the block, we need not read the
3966 struct buffer_head
*bitmap_bh
;
3969 start
= inode_offset
& ~(inodes_per_block
- 1);
3971 /* Is the inode bitmap in cache? */
3972 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3973 if (unlikely(!bitmap_bh
))
3977 * If the inode bitmap isn't in cache then the
3978 * optimisation may end up performing two reads instead
3979 * of one, so skip it.
3981 if (!buffer_uptodate(bitmap_bh
)) {
3985 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3986 if (i
== inode_offset
)
3988 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3992 if (i
== start
+ inodes_per_block
) {
3993 /* all other inodes are free, so skip I/O */
3994 memset(bh
->b_data
, 0, bh
->b_size
);
3995 set_buffer_uptodate(bh
);
4003 * If we need to do any I/O, try to pre-readahead extra
4004 * blocks from the inode table.
4006 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4007 ext4_fsblk_t b
, end
, table
;
4009 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4011 table
= ext4_inode_table(sb
, gdp
);
4012 /* s_inode_readahead_blks is always a power of 2 */
4013 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4017 num
= EXT4_INODES_PER_GROUP(sb
);
4018 if (ext4_has_group_desc_csum(sb
))
4019 num
-= ext4_itable_unused_count(sb
, gdp
);
4020 table
+= num
/ inodes_per_block
;
4024 sb_breadahead(sb
, b
++);
4028 * There are other valid inodes in the buffer, this inode
4029 * has in-inode xattrs, or we don't have this inode in memory.
4030 * Read the block from disk.
4032 trace_ext4_load_inode(inode
);
4034 bh
->b_end_io
= end_buffer_read_sync
;
4035 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
4037 if (!buffer_uptodate(bh
)) {
4038 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4039 "unable to read itable block");
4049 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4051 /* We have all inode data except xattrs in memory here. */
4052 return __ext4_get_inode_loc(inode
, iloc
,
4053 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4056 void ext4_set_inode_flags(struct inode
*inode
)
4058 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4060 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
4061 if (flags
& EXT4_SYNC_FL
)
4062 inode
->i_flags
|= S_SYNC
;
4063 if (flags
& EXT4_APPEND_FL
)
4064 inode
->i_flags
|= S_APPEND
;
4065 if (flags
& EXT4_IMMUTABLE_FL
)
4066 inode
->i_flags
|= S_IMMUTABLE
;
4067 if (flags
& EXT4_NOATIME_FL
)
4068 inode
->i_flags
|= S_NOATIME
;
4069 if (flags
& EXT4_DIRSYNC_FL
)
4070 inode
->i_flags
|= S_DIRSYNC
;
4073 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4074 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4076 unsigned int vfs_fl
;
4077 unsigned long old_fl
, new_fl
;
4080 vfs_fl
= ei
->vfs_inode
.i_flags
;
4081 old_fl
= ei
->i_flags
;
4082 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4083 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4085 if (vfs_fl
& S_SYNC
)
4086 new_fl
|= EXT4_SYNC_FL
;
4087 if (vfs_fl
& S_APPEND
)
4088 new_fl
|= EXT4_APPEND_FL
;
4089 if (vfs_fl
& S_IMMUTABLE
)
4090 new_fl
|= EXT4_IMMUTABLE_FL
;
4091 if (vfs_fl
& S_NOATIME
)
4092 new_fl
|= EXT4_NOATIME_FL
;
4093 if (vfs_fl
& S_DIRSYNC
)
4094 new_fl
|= EXT4_DIRSYNC_FL
;
4095 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4098 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4099 struct ext4_inode_info
*ei
)
4102 struct inode
*inode
= &(ei
->vfs_inode
);
4103 struct super_block
*sb
= inode
->i_sb
;
4105 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4106 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4107 /* we are using combined 48 bit field */
4108 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4109 le32_to_cpu(raw_inode
->i_blocks_lo
);
4110 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4111 /* i_blocks represent file system block size */
4112 return i_blocks
<< (inode
->i_blkbits
- 9);
4117 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4121 static inline void ext4_iget_extra_inode(struct inode
*inode
,
4122 struct ext4_inode
*raw_inode
,
4123 struct ext4_inode_info
*ei
)
4125 __le32
*magic
= (void *)raw_inode
+
4126 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4127 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4128 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4129 ext4_find_inline_data_nolock(inode
);
4131 EXT4_I(inode
)->i_inline_off
= 0;
4134 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4136 struct ext4_iloc iloc
;
4137 struct ext4_inode
*raw_inode
;
4138 struct ext4_inode_info
*ei
;
4139 struct inode
*inode
;
4140 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4146 inode
= iget_locked(sb
, ino
);
4148 return ERR_PTR(-ENOMEM
);
4149 if (!(inode
->i_state
& I_NEW
))
4155 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4158 raw_inode
= ext4_raw_inode(&iloc
);
4160 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4161 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4162 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4163 EXT4_INODE_SIZE(inode
->i_sb
)) {
4164 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
4165 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
4166 EXT4_INODE_SIZE(inode
->i_sb
));
4171 ei
->i_extra_isize
= 0;
4173 /* Precompute checksum seed for inode metadata */
4174 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4175 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
)) {
4176 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4178 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4179 __le32 gen
= raw_inode
->i_generation
;
4180 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4182 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4186 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4187 EXT4_ERROR_INODE(inode
, "checksum invalid");
4192 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4193 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4194 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4195 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4196 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4197 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4199 i_uid_write(inode
, i_uid
);
4200 i_gid_write(inode
, i_gid
);
4201 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4203 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4204 ei
->i_inline_off
= 0;
4205 ei
->i_dir_start_lookup
= 0;
4206 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4207 /* We now have enough fields to check if the inode was active or not.
4208 * This is needed because nfsd might try to access dead inodes
4209 * the test is that same one that e2fsck uses
4210 * NeilBrown 1999oct15
4212 if (inode
->i_nlink
== 0) {
4213 if ((inode
->i_mode
== 0 ||
4214 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4215 ino
!= EXT4_BOOT_LOADER_INO
) {
4216 /* this inode is deleted */
4220 /* The only unlinked inodes we let through here have
4221 * valid i_mode and are being read by the orphan
4222 * recovery code: that's fine, we're about to complete
4223 * the process of deleting those.
4224 * OR it is the EXT4_BOOT_LOADER_INO which is
4225 * not initialized on a new filesystem. */
4227 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4228 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4229 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4230 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4232 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4233 inode
->i_size
= ext4_isize(raw_inode
);
4234 ei
->i_disksize
= inode
->i_size
;
4236 ei
->i_reserved_quota
= 0;
4238 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4239 ei
->i_block_group
= iloc
.block_group
;
4240 ei
->i_last_alloc_group
= ~0;
4242 * NOTE! The in-memory inode i_data array is in little-endian order
4243 * even on big-endian machines: we do NOT byteswap the block numbers!
4245 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4246 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4247 INIT_LIST_HEAD(&ei
->i_orphan
);
4250 * Set transaction id's of transactions that have to be committed
4251 * to finish f[data]sync. We set them to currently running transaction
4252 * as we cannot be sure that the inode or some of its metadata isn't
4253 * part of the transaction - the inode could have been reclaimed and
4254 * now it is reread from disk.
4257 transaction_t
*transaction
;
4260 read_lock(&journal
->j_state_lock
);
4261 if (journal
->j_running_transaction
)
4262 transaction
= journal
->j_running_transaction
;
4264 transaction
= journal
->j_committing_transaction
;
4266 tid
= transaction
->t_tid
;
4268 tid
= journal
->j_commit_sequence
;
4269 read_unlock(&journal
->j_state_lock
);
4270 ei
->i_sync_tid
= tid
;
4271 ei
->i_datasync_tid
= tid
;
4274 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4275 if (ei
->i_extra_isize
== 0) {
4276 /* The extra space is currently unused. Use it. */
4277 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4278 EXT4_GOOD_OLD_INODE_SIZE
;
4280 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4284 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4285 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4286 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4287 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4289 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4290 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4291 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4293 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4297 if (ei
->i_file_acl
&&
4298 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4299 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4303 } else if (!ext4_has_inline_data(inode
)) {
4304 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4305 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4306 (S_ISLNK(inode
->i_mode
) &&
4307 !ext4_inode_is_fast_symlink(inode
))))
4308 /* Validate extent which is part of inode */
4309 ret
= ext4_ext_check_inode(inode
);
4310 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4311 (S_ISLNK(inode
->i_mode
) &&
4312 !ext4_inode_is_fast_symlink(inode
))) {
4313 /* Validate block references which are part of inode */
4314 ret
= ext4_ind_check_inode(inode
);
4320 if (S_ISREG(inode
->i_mode
)) {
4321 inode
->i_op
= &ext4_file_inode_operations
;
4322 inode
->i_fop
= &ext4_file_operations
;
4323 ext4_set_aops(inode
);
4324 } else if (S_ISDIR(inode
->i_mode
)) {
4325 inode
->i_op
= &ext4_dir_inode_operations
;
4326 inode
->i_fop
= &ext4_dir_operations
;
4327 } else if (S_ISLNK(inode
->i_mode
)) {
4328 if (ext4_inode_is_fast_symlink(inode
)) {
4329 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4330 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4331 sizeof(ei
->i_data
) - 1);
4333 inode
->i_op
= &ext4_symlink_inode_operations
;
4334 ext4_set_aops(inode
);
4336 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4337 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4338 inode
->i_op
= &ext4_special_inode_operations
;
4339 if (raw_inode
->i_block
[0])
4340 init_special_inode(inode
, inode
->i_mode
,
4341 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4343 init_special_inode(inode
, inode
->i_mode
,
4344 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4345 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4346 make_bad_inode(inode
);
4349 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4353 ext4_set_inode_flags(inode
);
4354 unlock_new_inode(inode
);
4360 return ERR_PTR(ret
);
4363 static int ext4_inode_blocks_set(handle_t
*handle
,
4364 struct ext4_inode
*raw_inode
,
4365 struct ext4_inode_info
*ei
)
4367 struct inode
*inode
= &(ei
->vfs_inode
);
4368 u64 i_blocks
= inode
->i_blocks
;
4369 struct super_block
*sb
= inode
->i_sb
;
4371 if (i_blocks
<= ~0U) {
4373 * i_blocks can be represented in a 32 bit variable
4374 * as multiple of 512 bytes
4376 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4377 raw_inode
->i_blocks_high
= 0;
4378 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4381 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4384 if (i_blocks
<= 0xffffffffffffULL
) {
4386 * i_blocks can be represented in a 48 bit variable
4387 * as multiple of 512 bytes
4389 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4390 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4391 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4393 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4394 /* i_block is stored in file system block size */
4395 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4396 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4397 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4403 * Post the struct inode info into an on-disk inode location in the
4404 * buffer-cache. This gobbles the caller's reference to the
4405 * buffer_head in the inode location struct.
4407 * The caller must have write access to iloc->bh.
4409 static int ext4_do_update_inode(handle_t
*handle
,
4410 struct inode
*inode
,
4411 struct ext4_iloc
*iloc
)
4413 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4414 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4415 struct buffer_head
*bh
= iloc
->bh
;
4416 int err
= 0, rc
, block
;
4417 int need_datasync
= 0;
4421 /* For fields not not tracking in the in-memory inode,
4422 * initialise them to zero for new inodes. */
4423 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4424 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4426 ext4_get_inode_flags(ei
);
4427 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4428 i_uid
= i_uid_read(inode
);
4429 i_gid
= i_gid_read(inode
);
4430 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4431 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4432 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4434 * Fix up interoperability with old kernels. Otherwise, old inodes get
4435 * re-used with the upper 16 bits of the uid/gid intact
4438 raw_inode
->i_uid_high
=
4439 cpu_to_le16(high_16_bits(i_uid
));
4440 raw_inode
->i_gid_high
=
4441 cpu_to_le16(high_16_bits(i_gid
));
4443 raw_inode
->i_uid_high
= 0;
4444 raw_inode
->i_gid_high
= 0;
4447 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4448 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4449 raw_inode
->i_uid_high
= 0;
4450 raw_inode
->i_gid_high
= 0;
4452 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4454 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4455 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4456 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4457 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4459 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4461 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4462 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4463 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4464 cpu_to_le32(EXT4_OS_HURD
))
4465 raw_inode
->i_file_acl_high
=
4466 cpu_to_le16(ei
->i_file_acl
>> 32);
4467 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4468 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4469 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4472 if (ei
->i_disksize
> 0x7fffffffULL
) {
4473 struct super_block
*sb
= inode
->i_sb
;
4474 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4475 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4476 EXT4_SB(sb
)->s_es
->s_rev_level
==
4477 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4478 /* If this is the first large file
4479 * created, add a flag to the superblock.
4481 err
= ext4_journal_get_write_access(handle
,
4482 EXT4_SB(sb
)->s_sbh
);
4485 ext4_update_dynamic_rev(sb
);
4486 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4487 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4488 ext4_handle_sync(handle
);
4489 err
= ext4_handle_dirty_super(handle
, sb
);
4492 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4493 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4494 if (old_valid_dev(inode
->i_rdev
)) {
4495 raw_inode
->i_block
[0] =
4496 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4497 raw_inode
->i_block
[1] = 0;
4499 raw_inode
->i_block
[0] = 0;
4500 raw_inode
->i_block
[1] =
4501 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4502 raw_inode
->i_block
[2] = 0;
4504 } else if (!ext4_has_inline_data(inode
)) {
4505 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4506 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4509 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4510 if (ei
->i_extra_isize
) {
4511 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4512 raw_inode
->i_version_hi
=
4513 cpu_to_le32(inode
->i_version
>> 32);
4514 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4517 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4519 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4520 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4523 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4525 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4528 ext4_std_error(inode
->i_sb
, err
);
4533 * ext4_write_inode()
4535 * We are called from a few places:
4537 * - Within generic_file_write() for O_SYNC files.
4538 * Here, there will be no transaction running. We wait for any running
4539 * transaction to commit.
4541 * - Within sys_sync(), kupdate and such.
4542 * We wait on commit, if tol to.
4544 * - Within prune_icache() (PF_MEMALLOC == true)
4545 * Here we simply return. We can't afford to block kswapd on the
4548 * In all cases it is actually safe for us to return without doing anything,
4549 * because the inode has been copied into a raw inode buffer in
4550 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4553 * Note that we are absolutely dependent upon all inode dirtiers doing the
4554 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4555 * which we are interested.
4557 * It would be a bug for them to not do this. The code:
4559 * mark_inode_dirty(inode)
4561 * inode->i_size = expr;
4563 * is in error because a kswapd-driven write_inode() could occur while
4564 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4565 * will no longer be on the superblock's dirty inode list.
4567 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4571 if (current
->flags
& PF_MEMALLOC
)
4574 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4575 if (ext4_journal_current_handle()) {
4576 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4581 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
4584 err
= ext4_force_commit(inode
->i_sb
);
4586 struct ext4_iloc iloc
;
4588 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4591 if (wbc
->sync_mode
== WB_SYNC_ALL
)
4592 sync_dirty_buffer(iloc
.bh
);
4593 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4594 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4595 "IO error syncing inode");
4604 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4605 * buffers that are attached to a page stradding i_size and are undergoing
4606 * commit. In that case we have to wait for commit to finish and try again.
4608 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4612 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4613 tid_t commit_tid
= 0;
4616 offset
= inode
->i_size
& (PAGE_CACHE_SIZE
- 1);
4618 * All buffers in the last page remain valid? Then there's nothing to
4619 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4622 if (offset
> PAGE_CACHE_SIZE
- (1 << inode
->i_blkbits
))
4625 page
= find_lock_page(inode
->i_mapping
,
4626 inode
->i_size
>> PAGE_CACHE_SHIFT
);
4629 ret
= __ext4_journalled_invalidatepage(page
, offset
);
4631 page_cache_release(page
);
4635 read_lock(&journal
->j_state_lock
);
4636 if (journal
->j_committing_transaction
)
4637 commit_tid
= journal
->j_committing_transaction
->t_tid
;
4638 read_unlock(&journal
->j_state_lock
);
4640 jbd2_log_wait_commit(journal
, commit_tid
);
4647 * Called from notify_change.
4649 * We want to trap VFS attempts to truncate the file as soon as
4650 * possible. In particular, we want to make sure that when the VFS
4651 * shrinks i_size, we put the inode on the orphan list and modify
4652 * i_disksize immediately, so that during the subsequent flushing of
4653 * dirty pages and freeing of disk blocks, we can guarantee that any
4654 * commit will leave the blocks being flushed in an unused state on
4655 * disk. (On recovery, the inode will get truncated and the blocks will
4656 * be freed, so we have a strong guarantee that no future commit will
4657 * leave these blocks visible to the user.)
4659 * Another thing we have to assure is that if we are in ordered mode
4660 * and inode is still attached to the committing transaction, we must
4661 * we start writeout of all the dirty pages which are being truncated.
4662 * This way we are sure that all the data written in the previous
4663 * transaction are already on disk (truncate waits for pages under
4666 * Called with inode->i_mutex down.
4668 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4670 struct inode
*inode
= dentry
->d_inode
;
4673 const unsigned int ia_valid
= attr
->ia_valid
;
4675 error
= inode_change_ok(inode
, attr
);
4679 if (is_quota_modification(inode
, attr
))
4680 dquot_initialize(inode
);
4681 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4682 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4685 /* (user+group)*(old+new) structure, inode write (sb,
4686 * inode block, ? - but truncate inode update has it) */
4687 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
4688 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
4689 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
4690 if (IS_ERR(handle
)) {
4691 error
= PTR_ERR(handle
);
4694 error
= dquot_transfer(inode
, attr
);
4696 ext4_journal_stop(handle
);
4699 /* Update corresponding info in inode so that everything is in
4700 * one transaction */
4701 if (attr
->ia_valid
& ATTR_UID
)
4702 inode
->i_uid
= attr
->ia_uid
;
4703 if (attr
->ia_valid
& ATTR_GID
)
4704 inode
->i_gid
= attr
->ia_gid
;
4705 error
= ext4_mark_inode_dirty(handle
, inode
);
4706 ext4_journal_stop(handle
);
4709 if (attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
!= inode
->i_size
) {
4711 loff_t oldsize
= inode
->i_size
;
4713 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4714 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4716 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4719 if (S_ISREG(inode
->i_mode
) &&
4720 (attr
->ia_size
< inode
->i_size
)) {
4721 if (ext4_should_order_data(inode
)) {
4722 error
= ext4_begin_ordered_truncate(inode
,
4727 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
4728 if (IS_ERR(handle
)) {
4729 error
= PTR_ERR(handle
);
4732 if (ext4_handle_valid(handle
)) {
4733 error
= ext4_orphan_add(handle
, inode
);
4736 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4737 rc
= ext4_mark_inode_dirty(handle
, inode
);
4740 ext4_journal_stop(handle
);
4742 ext4_orphan_del(NULL
, inode
);
4747 i_size_write(inode
, attr
->ia_size
);
4749 * Blocks are going to be removed from the inode. Wait
4750 * for dio in flight. Temporarily disable
4751 * dioread_nolock to prevent livelock.
4754 if (!ext4_should_journal_data(inode
)) {
4755 ext4_inode_block_unlocked_dio(inode
);
4756 inode_dio_wait(inode
);
4757 ext4_inode_resume_unlocked_dio(inode
);
4759 ext4_wait_for_tail_page_commit(inode
);
4762 * Truncate pagecache after we've waited for commit
4763 * in data=journal mode to make pages freeable.
4765 truncate_pagecache(inode
, oldsize
, inode
->i_size
);
4768 * We want to call ext4_truncate() even if attr->ia_size ==
4769 * inode->i_size for cases like truncation of fallocated space
4771 if (attr
->ia_valid
& ATTR_SIZE
)
4772 ext4_truncate(inode
);
4775 setattr_copy(inode
, attr
);
4776 mark_inode_dirty(inode
);
4780 * If the call to ext4_truncate failed to get a transaction handle at
4781 * all, we need to clean up the in-core orphan list manually.
4783 if (orphan
&& inode
->i_nlink
)
4784 ext4_orphan_del(NULL
, inode
);
4786 if (!rc
&& (ia_valid
& ATTR_MODE
))
4787 rc
= ext4_acl_chmod(inode
);
4790 ext4_std_error(inode
->i_sb
, error
);
4796 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4799 struct inode
*inode
;
4800 unsigned long long delalloc_blocks
;
4802 inode
= dentry
->d_inode
;
4803 generic_fillattr(inode
, stat
);
4806 * We can't update i_blocks if the block allocation is delayed
4807 * otherwise in the case of system crash before the real block
4808 * allocation is done, we will have i_blocks inconsistent with
4809 * on-disk file blocks.
4810 * We always keep i_blocks updated together with real
4811 * allocation. But to not confuse with user, stat
4812 * will return the blocks that include the delayed allocation
4813 * blocks for this file.
4815 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
4816 EXT4_I(inode
)->i_reserved_data_blocks
);
4818 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
-9);
4822 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4824 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4825 return ext4_ind_trans_blocks(inode
, nrblocks
, chunk
);
4826 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4830 * Account for index blocks, block groups bitmaps and block group
4831 * descriptor blocks if modify datablocks and index blocks
4832 * worse case, the indexs blocks spread over different block groups
4834 * If datablocks are discontiguous, they are possible to spread over
4835 * different block groups too. If they are contiguous, with flexbg,
4836 * they could still across block group boundary.
4838 * Also account for superblock, inode, quota and xattr blocks
4840 static int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4842 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4848 * How many index blocks need to touch to modify nrblocks?
4849 * The "Chunk" flag indicating whether the nrblocks is
4850 * physically contiguous on disk
4852 * For Direct IO and fallocate, they calls get_block to allocate
4853 * one single extent at a time, so they could set the "Chunk" flag
4855 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4860 * Now let's see how many group bitmaps and group descriptors need
4870 if (groups
> ngroups
)
4872 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4873 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4875 /* bitmaps and block group descriptor blocks */
4876 ret
+= groups
+ gdpblocks
;
4878 /* Blocks for super block, inode, quota and xattr blocks */
4879 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4885 * Calculate the total number of credits to reserve to fit
4886 * the modification of a single pages into a single transaction,
4887 * which may include multiple chunks of block allocations.
4889 * This could be called via ext4_write_begin()
4891 * We need to consider the worse case, when
4892 * one new block per extent.
4894 int ext4_writepage_trans_blocks(struct inode
*inode
)
4896 int bpp
= ext4_journal_blocks_per_page(inode
);
4899 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4901 /* Account for data blocks for journalled mode */
4902 if (ext4_should_journal_data(inode
))
4908 * Calculate the journal credits for a chunk of data modification.
4910 * This is called from DIO, fallocate or whoever calling
4911 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4913 * journal buffers for data blocks are not included here, as DIO
4914 * and fallocate do no need to journal data buffers.
4916 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4918 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4922 * The caller must have previously called ext4_reserve_inode_write().
4923 * Give this, we know that the caller already has write access to iloc->bh.
4925 int ext4_mark_iloc_dirty(handle_t
*handle
,
4926 struct inode
*inode
, struct ext4_iloc
*iloc
)
4930 if (IS_I_VERSION(inode
))
4931 inode_inc_iversion(inode
);
4933 /* the do_update_inode consumes one bh->b_count */
4936 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4937 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4943 * On success, We end up with an outstanding reference count against
4944 * iloc->bh. This _must_ be cleaned up later.
4948 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4949 struct ext4_iloc
*iloc
)
4953 err
= ext4_get_inode_loc(inode
, iloc
);
4955 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4956 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4962 ext4_std_error(inode
->i_sb
, err
);
4967 * Expand an inode by new_extra_isize bytes.
4968 * Returns 0 on success or negative error number on failure.
4970 static int ext4_expand_extra_isize(struct inode
*inode
,
4971 unsigned int new_extra_isize
,
4972 struct ext4_iloc iloc
,
4975 struct ext4_inode
*raw_inode
;
4976 struct ext4_xattr_ibody_header
*header
;
4978 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4981 raw_inode
= ext4_raw_inode(&iloc
);
4983 header
= IHDR(inode
, raw_inode
);
4985 /* No extended attributes present */
4986 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
4987 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4988 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4990 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4994 /* try to expand with EAs present */
4995 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5000 * What we do here is to mark the in-core inode as clean with respect to inode
5001 * dirtiness (it may still be data-dirty).
5002 * This means that the in-core inode may be reaped by prune_icache
5003 * without having to perform any I/O. This is a very good thing,
5004 * because *any* task may call prune_icache - even ones which
5005 * have a transaction open against a different journal.
5007 * Is this cheating? Not really. Sure, we haven't written the
5008 * inode out, but prune_icache isn't a user-visible syncing function.
5009 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5010 * we start and wait on commits.
5012 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5014 struct ext4_iloc iloc
;
5015 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5016 static unsigned int mnt_count
;
5020 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5021 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5022 if (ext4_handle_valid(handle
) &&
5023 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5024 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5026 * We need extra buffer credits since we may write into EA block
5027 * with this same handle. If journal_extend fails, then it will
5028 * only result in a minor loss of functionality for that inode.
5029 * If this is felt to be critical, then e2fsck should be run to
5030 * force a large enough s_min_extra_isize.
5032 if ((jbd2_journal_extend(handle
,
5033 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5034 ret
= ext4_expand_extra_isize(inode
,
5035 sbi
->s_want_extra_isize
,
5038 ext4_set_inode_state(inode
,
5039 EXT4_STATE_NO_EXPAND
);
5041 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5042 ext4_warning(inode
->i_sb
,
5043 "Unable to expand inode %lu. Delete"
5044 " some EAs or run e2fsck.",
5047 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5053 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5058 * ext4_dirty_inode() is called from __mark_inode_dirty()
5060 * We're really interested in the case where a file is being extended.
5061 * i_size has been changed by generic_commit_write() and we thus need
5062 * to include the updated inode in the current transaction.
5064 * Also, dquot_alloc_block() will always dirty the inode when blocks
5065 * are allocated to the file.
5067 * If the inode is marked synchronous, we don't honour that here - doing
5068 * so would cause a commit on atime updates, which we don't bother doing.
5069 * We handle synchronous inodes at the highest possible level.
5071 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5075 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5079 ext4_mark_inode_dirty(handle
, inode
);
5081 ext4_journal_stop(handle
);
5088 * Bind an inode's backing buffer_head into this transaction, to prevent
5089 * it from being flushed to disk early. Unlike
5090 * ext4_reserve_inode_write, this leaves behind no bh reference and
5091 * returns no iloc structure, so the caller needs to repeat the iloc
5092 * lookup to mark the inode dirty later.
5094 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5096 struct ext4_iloc iloc
;
5100 err
= ext4_get_inode_loc(inode
, &iloc
);
5102 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5103 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5105 err
= ext4_handle_dirty_metadata(handle
,
5111 ext4_std_error(inode
->i_sb
, err
);
5116 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5123 * We have to be very careful here: changing a data block's
5124 * journaling status dynamically is dangerous. If we write a
5125 * data block to the journal, change the status and then delete
5126 * that block, we risk forgetting to revoke the old log record
5127 * from the journal and so a subsequent replay can corrupt data.
5128 * So, first we make sure that the journal is empty and that
5129 * nobody is changing anything.
5132 journal
= EXT4_JOURNAL(inode
);
5135 if (is_journal_aborted(journal
))
5137 /* We have to allocate physical blocks for delalloc blocks
5138 * before flushing journal. otherwise delalloc blocks can not
5139 * be allocated any more. even more truncate on delalloc blocks
5140 * could trigger BUG by flushing delalloc blocks in journal.
5141 * There is no delalloc block in non-journal data mode.
5143 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
5144 err
= ext4_alloc_da_blocks(inode
);
5149 /* Wait for all existing dio workers */
5150 ext4_inode_block_unlocked_dio(inode
);
5151 inode_dio_wait(inode
);
5153 jbd2_journal_lock_updates(journal
);
5156 * OK, there are no updates running now, and all cached data is
5157 * synced to disk. We are now in a completely consistent state
5158 * which doesn't have anything in the journal, and we know that
5159 * no filesystem updates are running, so it is safe to modify
5160 * the inode's in-core data-journaling state flag now.
5164 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5166 jbd2_journal_flush(journal
);
5167 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5169 ext4_set_aops(inode
);
5171 jbd2_journal_unlock_updates(journal
);
5172 ext4_inode_resume_unlocked_dio(inode
);
5174 /* Finally we can mark the inode as dirty. */
5176 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5178 return PTR_ERR(handle
);
5180 err
= ext4_mark_inode_dirty(handle
, inode
);
5181 ext4_handle_sync(handle
);
5182 ext4_journal_stop(handle
);
5183 ext4_std_error(inode
->i_sb
, err
);
5188 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5190 return !buffer_mapped(bh
);
5193 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5195 struct page
*page
= vmf
->page
;
5199 struct file
*file
= vma
->vm_file
;
5200 struct inode
*inode
= file_inode(file
);
5201 struct address_space
*mapping
= inode
->i_mapping
;
5203 get_block_t
*get_block
;
5206 sb_start_pagefault(inode
->i_sb
);
5207 file_update_time(vma
->vm_file
);
5208 /* Delalloc case is easy... */
5209 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5210 !ext4_should_journal_data(inode
) &&
5211 !ext4_nonda_switch(inode
->i_sb
)) {
5213 ret
= __block_page_mkwrite(vma
, vmf
,
5214 ext4_da_get_block_prep
);
5215 } while (ret
== -ENOSPC
&&
5216 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5221 size
= i_size_read(inode
);
5222 /* Page got truncated from under us? */
5223 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5225 ret
= VM_FAULT_NOPAGE
;
5229 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5230 len
= size
& ~PAGE_CACHE_MASK
;
5232 len
= PAGE_CACHE_SIZE
;
5234 * Return if we have all the buffers mapped. This avoids the need to do
5235 * journal_start/journal_stop which can block and take a long time
5237 if (page_has_buffers(page
)) {
5238 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5240 ext4_bh_unmapped
)) {
5241 /* Wait so that we don't change page under IO */
5242 wait_for_stable_page(page
);
5243 ret
= VM_FAULT_LOCKED
;
5248 /* OK, we need to fill the hole... */
5249 if (ext4_should_dioread_nolock(inode
))
5250 get_block
= ext4_get_block_write
;
5252 get_block
= ext4_get_block
;
5254 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5255 ext4_writepage_trans_blocks(inode
));
5256 if (IS_ERR(handle
)) {
5257 ret
= VM_FAULT_SIGBUS
;
5260 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
5261 if (!ret
&& ext4_should_journal_data(inode
)) {
5262 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5263 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
5265 ret
= VM_FAULT_SIGBUS
;
5266 ext4_journal_stop(handle
);
5269 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5271 ext4_journal_stop(handle
);
5272 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5275 ret
= block_page_mkwrite_return(ret
);
5277 sb_end_pagefault(inode
->i_sb
);