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 int offset
,
136 unsigned int length
);
137 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
138 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
139 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
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);
219 WARN_ON(atomic_read(&EXT4_I(inode
)->i_ioend_count
));
223 if (!is_bad_inode(inode
))
224 dquot_initialize(inode
);
226 if (ext4_should_order_data(inode
))
227 ext4_begin_ordered_truncate(inode
, 0);
228 truncate_inode_pages(&inode
->i_data
, 0);
230 WARN_ON(atomic_read(&EXT4_I(inode
)->i_ioend_count
));
231 if (is_bad_inode(inode
))
235 * Protect us against freezing - iput() caller didn't have to have any
236 * protection against it
238 sb_start_intwrite(inode
->i_sb
);
239 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
240 ext4_blocks_for_truncate(inode
)+3);
241 if (IS_ERR(handle
)) {
242 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
244 * If we're going to skip the normal cleanup, we still need to
245 * make sure that the in-core orphan linked list is properly
248 ext4_orphan_del(NULL
, inode
);
249 sb_end_intwrite(inode
->i_sb
);
254 ext4_handle_sync(handle
);
256 err
= ext4_mark_inode_dirty(handle
, inode
);
258 ext4_warning(inode
->i_sb
,
259 "couldn't mark inode dirty (err %d)", err
);
263 ext4_truncate(inode
);
266 * ext4_ext_truncate() doesn't reserve any slop when it
267 * restarts journal transactions; therefore there may not be
268 * enough credits left in the handle to remove the inode from
269 * the orphan list and set the dtime field.
271 if (!ext4_handle_has_enough_credits(handle
, 3)) {
272 err
= ext4_journal_extend(handle
, 3);
274 err
= ext4_journal_restart(handle
, 3);
276 ext4_warning(inode
->i_sb
,
277 "couldn't extend journal (err %d)", err
);
279 ext4_journal_stop(handle
);
280 ext4_orphan_del(NULL
, inode
);
281 sb_end_intwrite(inode
->i_sb
);
287 * Kill off the orphan record which ext4_truncate created.
288 * AKPM: I think this can be inside the above `if'.
289 * Note that ext4_orphan_del() has to be able to cope with the
290 * deletion of a non-existent orphan - this is because we don't
291 * know if ext4_truncate() actually created an orphan record.
292 * (Well, we could do this if we need to, but heck - it works)
294 ext4_orphan_del(handle
, inode
);
295 EXT4_I(inode
)->i_dtime
= get_seconds();
298 * One subtle ordering requirement: if anything has gone wrong
299 * (transaction abort, IO errors, whatever), then we can still
300 * do these next steps (the fs will already have been marked as
301 * having errors), but we can't free the inode if the mark_dirty
304 if (ext4_mark_inode_dirty(handle
, inode
))
305 /* If that failed, just do the required in-core inode clear. */
306 ext4_clear_inode(inode
);
308 ext4_free_inode(handle
, inode
);
309 ext4_journal_stop(handle
);
310 sb_end_intwrite(inode
->i_sb
);
313 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
317 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
319 return &EXT4_I(inode
)->i_reserved_quota
;
324 * Calculate the number of metadata blocks need to reserve
325 * to allocate a block located at @lblock
327 static int ext4_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
329 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
330 return ext4_ext_calc_metadata_amount(inode
, lblock
);
332 return ext4_ind_calc_metadata_amount(inode
, lblock
);
336 * Called with i_data_sem down, which is important since we can call
337 * ext4_discard_preallocations() from here.
339 void ext4_da_update_reserve_space(struct inode
*inode
,
340 int used
, int quota_claim
)
342 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
343 struct ext4_inode_info
*ei
= EXT4_I(inode
);
345 spin_lock(&ei
->i_block_reservation_lock
);
346 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
347 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
348 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
349 "with only %d reserved data blocks",
350 __func__
, inode
->i_ino
, used
,
351 ei
->i_reserved_data_blocks
);
353 used
= ei
->i_reserved_data_blocks
;
356 if (unlikely(ei
->i_allocated_meta_blocks
> ei
->i_reserved_meta_blocks
)) {
357 ext4_warning(inode
->i_sb
, "ino %lu, allocated %d "
358 "with only %d reserved metadata blocks "
359 "(releasing %d blocks with reserved %d data blocks)",
360 inode
->i_ino
, ei
->i_allocated_meta_blocks
,
361 ei
->i_reserved_meta_blocks
, used
,
362 ei
->i_reserved_data_blocks
);
364 ei
->i_allocated_meta_blocks
= ei
->i_reserved_meta_blocks
;
367 /* Update per-inode reservations */
368 ei
->i_reserved_data_blocks
-= used
;
369 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
370 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
371 used
+ ei
->i_allocated_meta_blocks
);
372 ei
->i_allocated_meta_blocks
= 0;
374 if (ei
->i_reserved_data_blocks
== 0) {
376 * We can release all of the reserved metadata blocks
377 * only when we have written all of the delayed
380 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
381 ei
->i_reserved_meta_blocks
);
382 ei
->i_reserved_meta_blocks
= 0;
383 ei
->i_da_metadata_calc_len
= 0;
385 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
387 /* Update quota subsystem for data blocks */
389 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
392 * We did fallocate with an offset that is already delayed
393 * allocated. So on delayed allocated writeback we should
394 * not re-claim the quota for fallocated blocks.
396 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
400 * If we have done all the pending block allocations and if
401 * there aren't any writers on the inode, we can discard the
402 * inode's preallocations.
404 if ((ei
->i_reserved_data_blocks
== 0) &&
405 (atomic_read(&inode
->i_writecount
) == 0))
406 ext4_discard_preallocations(inode
);
409 static int __check_block_validity(struct inode
*inode
, const char *func
,
411 struct ext4_map_blocks
*map
)
413 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
415 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
416 "lblock %lu mapped to illegal pblock "
417 "(length %d)", (unsigned long) map
->m_lblk
,
424 #define check_block_validity(inode, map) \
425 __check_block_validity((inode), __func__, __LINE__, (map))
427 #ifdef ES_AGGRESSIVE_TEST
428 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
430 struct ext4_map_blocks
*es_map
,
431 struct ext4_map_blocks
*map
,
438 * There is a race window that the result is not the same.
439 * e.g. xfstests #223 when dioread_nolock enables. The reason
440 * is that we lookup a block mapping in extent status tree with
441 * out taking i_data_sem. So at the time the unwritten extent
442 * could be converted.
444 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
445 down_read((&EXT4_I(inode
)->i_data_sem
));
446 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
447 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
448 EXT4_GET_BLOCKS_KEEP_SIZE
);
450 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
451 EXT4_GET_BLOCKS_KEEP_SIZE
);
453 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
454 up_read((&EXT4_I(inode
)->i_data_sem
));
456 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
457 * because it shouldn't be marked in es_map->m_flags.
459 map
->m_flags
&= ~(EXT4_MAP_FROM_CLUSTER
| EXT4_MAP_BOUNDARY
);
462 * We don't check m_len because extent will be collpased in status
463 * tree. So the m_len might not equal.
465 if (es_map
->m_lblk
!= map
->m_lblk
||
466 es_map
->m_flags
!= map
->m_flags
||
467 es_map
->m_pblk
!= map
->m_pblk
) {
468 printk("ES cache assertion failed for inode: %lu "
469 "es_cached ex [%d/%d/%llu/%x] != "
470 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
471 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
472 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
473 map
->m_len
, map
->m_pblk
, map
->m_flags
,
477 #endif /* ES_AGGRESSIVE_TEST */
480 * The ext4_map_blocks() function tries to look up the requested blocks,
481 * and returns if the blocks are already mapped.
483 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
484 * and store the allocated blocks in the result buffer head and mark it
487 * If file type is extents based, it will call ext4_ext_map_blocks(),
488 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
491 * On success, it returns the number of blocks being mapped or allocate.
492 * if create==0 and the blocks are pre-allocated and uninitialized block,
493 * the result buffer head is unmapped. If the create ==1, it will make sure
494 * the buffer head is mapped.
496 * It returns 0 if plain look up failed (blocks have not been allocated), in
497 * that case, buffer head is unmapped
499 * It returns the error in case of allocation failure.
501 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
502 struct ext4_map_blocks
*map
, int flags
)
504 struct extent_status es
;
506 #ifdef ES_AGGRESSIVE_TEST
507 struct ext4_map_blocks orig_map
;
509 memcpy(&orig_map
, map
, sizeof(*map
));
513 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
514 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
515 (unsigned long) map
->m_lblk
);
517 /* Lookup extent status tree firstly */
518 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
519 ext4_es_lru_add(inode
);
520 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
521 map
->m_pblk
= ext4_es_pblock(&es
) +
522 map
->m_lblk
- es
.es_lblk
;
523 map
->m_flags
|= ext4_es_is_written(&es
) ?
524 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
525 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
526 if (retval
> map
->m_len
)
529 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
534 #ifdef ES_AGGRESSIVE_TEST
535 ext4_map_blocks_es_recheck(handle
, inode
, map
,
542 * Try to see if we can get the block without requesting a new
545 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
546 down_read((&EXT4_I(inode
)->i_data_sem
));
547 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
548 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
549 EXT4_GET_BLOCKS_KEEP_SIZE
);
551 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
552 EXT4_GET_BLOCKS_KEEP_SIZE
);
556 unsigned long long status
;
558 #ifdef ES_AGGRESSIVE_TEST
559 if (retval
!= map
->m_len
) {
560 printk("ES len assertion failed for inode: %lu "
561 "retval %d != map->m_len %d "
562 "in %s (lookup)\n", inode
->i_ino
, retval
,
563 map
->m_len
, __func__
);
567 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
568 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
569 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
570 ext4_find_delalloc_range(inode
, map
->m_lblk
,
571 map
->m_lblk
+ map
->m_len
- 1))
572 status
|= EXTENT_STATUS_DELAYED
;
573 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
574 map
->m_len
, map
->m_pblk
, status
);
578 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
579 up_read((&EXT4_I(inode
)->i_data_sem
));
582 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
583 int ret
= check_block_validity(inode
, map
);
588 /* If it is only a block(s) look up */
589 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
593 * Returns if the blocks have already allocated
595 * Note that if blocks have been preallocated
596 * ext4_ext_get_block() returns the create = 0
597 * with buffer head unmapped.
599 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
603 * Here we clear m_flags because after allocating an new extent,
604 * it will be set again.
606 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
609 * New blocks allocate and/or writing to uninitialized extent
610 * will possibly result in updating i_data, so we take
611 * the write lock of i_data_sem, and call get_blocks()
612 * with create == 1 flag.
614 down_write((&EXT4_I(inode
)->i_data_sem
));
617 * if the caller is from delayed allocation writeout path
618 * we have already reserved fs blocks for allocation
619 * let the underlying get_block() function know to
620 * avoid double accounting
622 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
623 ext4_set_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
625 * We need to check for EXT4 here because migrate
626 * could have changed the inode type in between
628 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
629 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
631 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
633 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
635 * We allocated new blocks which will result in
636 * i_data's format changing. Force the migrate
637 * to fail by clearing migrate flags
639 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
643 * Update reserved blocks/metadata blocks after successful
644 * block allocation which had been deferred till now. We don't
645 * support fallocate for non extent files. So we can update
646 * reserve space here.
649 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
650 ext4_da_update_reserve_space(inode
, retval
, 1);
652 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
653 ext4_clear_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
657 unsigned long long status
;
659 #ifdef ES_AGGRESSIVE_TEST
660 if (retval
!= map
->m_len
) {
661 printk("ES len assertion failed for inode: %lu "
662 "retval %d != map->m_len %d "
663 "in %s (allocation)\n", inode
->i_ino
, retval
,
664 map
->m_len
, __func__
);
669 * If the extent has been zeroed out, we don't need to update
670 * extent status tree.
672 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
673 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
674 if (ext4_es_is_written(&es
))
677 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
678 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
679 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
680 ext4_find_delalloc_range(inode
, map
->m_lblk
,
681 map
->m_lblk
+ map
->m_len
- 1))
682 status
|= EXTENT_STATUS_DELAYED
;
683 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
684 map
->m_pblk
, status
);
690 up_write((&EXT4_I(inode
)->i_data_sem
));
691 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
692 int ret
= check_block_validity(inode
, map
);
699 /* Maximum number of blocks we map for direct IO at once. */
700 #define DIO_MAX_BLOCKS 4096
702 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
703 struct buffer_head
*bh
, int flags
)
705 handle_t
*handle
= ext4_journal_current_handle();
706 struct ext4_map_blocks map
;
707 int ret
= 0, started
= 0;
710 if (ext4_has_inline_data(inode
))
714 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
716 if (flags
&& !(flags
& EXT4_GET_BLOCKS_NO_LOCK
) && !handle
) {
717 /* Direct IO write... */
718 if (map
.m_len
> DIO_MAX_BLOCKS
)
719 map
.m_len
= DIO_MAX_BLOCKS
;
720 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
721 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
723 if (IS_ERR(handle
)) {
724 ret
= PTR_ERR(handle
);
730 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
732 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
733 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
734 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
738 ext4_journal_stop(handle
);
742 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
743 struct buffer_head
*bh
, int create
)
745 return _ext4_get_block(inode
, iblock
, bh
,
746 create
? EXT4_GET_BLOCKS_CREATE
: 0);
750 * `handle' can be NULL if create is zero
752 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
753 ext4_lblk_t block
, int create
, int *errp
)
755 struct ext4_map_blocks map
;
756 struct buffer_head
*bh
;
759 J_ASSERT(handle
!= NULL
|| create
== 0);
763 err
= ext4_map_blocks(handle
, inode
, &map
,
764 create
? EXT4_GET_BLOCKS_CREATE
: 0);
766 /* ensure we send some value back into *errp */
769 if (create
&& err
== 0)
770 err
= -ENOSPC
; /* should never happen */
776 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
781 if (map
.m_flags
& EXT4_MAP_NEW
) {
782 J_ASSERT(create
!= 0);
783 J_ASSERT(handle
!= NULL
);
786 * Now that we do not always journal data, we should
787 * keep in mind whether this should always journal the
788 * new buffer as metadata. For now, regular file
789 * writes use ext4_get_block instead, so it's not a
793 BUFFER_TRACE(bh
, "call get_create_access");
794 fatal
= ext4_journal_get_create_access(handle
, bh
);
795 if (!fatal
&& !buffer_uptodate(bh
)) {
796 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
797 set_buffer_uptodate(bh
);
800 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
801 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
805 BUFFER_TRACE(bh
, "not a new buffer");
815 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
816 ext4_lblk_t block
, int create
, int *err
)
818 struct buffer_head
*bh
;
820 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
823 if (buffer_uptodate(bh
))
825 ll_rw_block(READ
| REQ_META
| REQ_PRIO
, 1, &bh
);
827 if (buffer_uptodate(bh
))
834 int ext4_walk_page_buffers(handle_t
*handle
,
835 struct buffer_head
*head
,
839 int (*fn
)(handle_t
*handle
,
840 struct buffer_head
*bh
))
842 struct buffer_head
*bh
;
843 unsigned block_start
, block_end
;
844 unsigned blocksize
= head
->b_size
;
846 struct buffer_head
*next
;
848 for (bh
= head
, block_start
= 0;
849 ret
== 0 && (bh
!= head
|| !block_start
);
850 block_start
= block_end
, bh
= next
) {
851 next
= bh
->b_this_page
;
852 block_end
= block_start
+ blocksize
;
853 if (block_end
<= from
|| block_start
>= to
) {
854 if (partial
&& !buffer_uptodate(bh
))
858 err
= (*fn
)(handle
, bh
);
866 * To preserve ordering, it is essential that the hole instantiation and
867 * the data write be encapsulated in a single transaction. We cannot
868 * close off a transaction and start a new one between the ext4_get_block()
869 * and the commit_write(). So doing the jbd2_journal_start at the start of
870 * prepare_write() is the right place.
872 * Also, this function can nest inside ext4_writepage(). In that case, we
873 * *know* that ext4_writepage() has generated enough buffer credits to do the
874 * whole page. So we won't block on the journal in that case, which is good,
875 * because the caller may be PF_MEMALLOC.
877 * By accident, ext4 can be reentered when a transaction is open via
878 * quota file writes. If we were to commit the transaction while thus
879 * reentered, there can be a deadlock - we would be holding a quota
880 * lock, and the commit would never complete if another thread had a
881 * transaction open and was blocking on the quota lock - a ranking
884 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
885 * will _not_ run commit under these circumstances because handle->h_ref
886 * is elevated. We'll still have enough credits for the tiny quotafile
889 int do_journal_get_write_access(handle_t
*handle
,
890 struct buffer_head
*bh
)
892 int dirty
= buffer_dirty(bh
);
895 if (!buffer_mapped(bh
) || buffer_freed(bh
))
898 * __block_write_begin() could have dirtied some buffers. Clean
899 * the dirty bit as jbd2_journal_get_write_access() could complain
900 * otherwise about fs integrity issues. Setting of the dirty bit
901 * by __block_write_begin() isn't a real problem here as we clear
902 * the bit before releasing a page lock and thus writeback cannot
903 * ever write the buffer.
906 clear_buffer_dirty(bh
);
907 ret
= ext4_journal_get_write_access(handle
, bh
);
909 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
913 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
914 struct buffer_head
*bh_result
, int create
);
915 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
916 loff_t pos
, unsigned len
, unsigned flags
,
917 struct page
**pagep
, void **fsdata
)
919 struct inode
*inode
= mapping
->host
;
920 int ret
, needed_blocks
;
927 trace_ext4_write_begin(inode
, pos
, len
, flags
);
929 * Reserve one block more for addition to orphan list in case
930 * we allocate blocks but write fails for some reason
932 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
933 index
= pos
>> PAGE_CACHE_SHIFT
;
934 from
= pos
& (PAGE_CACHE_SIZE
- 1);
937 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
938 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
947 * grab_cache_page_write_begin() can take a long time if the
948 * system is thrashing due to memory pressure, or if the page
949 * is being written back. So grab it first before we start
950 * the transaction handle. This also allows us to allocate
951 * the page (if needed) without using GFP_NOFS.
954 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
960 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
961 if (IS_ERR(handle
)) {
962 page_cache_release(page
);
963 return PTR_ERR(handle
);
967 if (page
->mapping
!= mapping
) {
968 /* The page got truncated from under us */
970 page_cache_release(page
);
971 ext4_journal_stop(handle
);
974 wait_on_page_writeback(page
);
976 if (ext4_should_dioread_nolock(inode
))
977 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
979 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
981 if (!ret
&& ext4_should_journal_data(inode
)) {
982 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
984 do_journal_get_write_access
);
990 * __block_write_begin may have instantiated a few blocks
991 * outside i_size. Trim these off again. Don't need
992 * i_size_read because we hold i_mutex.
994 * Add inode to orphan list in case we crash before
997 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
998 ext4_orphan_add(handle
, inode
);
1000 ext4_journal_stop(handle
);
1001 if (pos
+ len
> inode
->i_size
) {
1002 ext4_truncate_failed_write(inode
);
1004 * If truncate failed early the inode might
1005 * still be on the orphan list; we need to
1006 * make sure the inode is removed from the
1007 * orphan list in that case.
1010 ext4_orphan_del(NULL
, inode
);
1013 if (ret
== -ENOSPC
&&
1014 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1016 page_cache_release(page
);
1023 /* For write_end() in data=journal mode */
1024 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1027 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1029 set_buffer_uptodate(bh
);
1030 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1031 clear_buffer_meta(bh
);
1032 clear_buffer_prio(bh
);
1037 * We need to pick up the new inode size which generic_commit_write gave us
1038 * `file' can be NULL - eg, when called from page_symlink().
1040 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1041 * buffers are managed internally.
1043 static int ext4_write_end(struct file
*file
,
1044 struct address_space
*mapping
,
1045 loff_t pos
, unsigned len
, unsigned copied
,
1046 struct page
*page
, void *fsdata
)
1048 handle_t
*handle
= ext4_journal_current_handle();
1049 struct inode
*inode
= mapping
->host
;
1051 int i_size_changed
= 0;
1053 trace_ext4_write_end(inode
, pos
, len
, copied
);
1054 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
)) {
1055 ret
= ext4_jbd2_file_inode(handle
, inode
);
1058 page_cache_release(page
);
1063 if (ext4_has_inline_data(inode
)) {
1064 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1070 copied
= block_write_end(file
, mapping
, pos
,
1071 len
, copied
, page
, fsdata
);
1074 * No need to use i_size_read() here, the i_size
1075 * cannot change under us because we hole i_mutex.
1077 * But it's important to update i_size while still holding page lock:
1078 * page writeout could otherwise come in and zero beyond i_size.
1080 if (pos
+ copied
> inode
->i_size
) {
1081 i_size_write(inode
, pos
+ copied
);
1085 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1086 /* We need to mark inode dirty even if
1087 * new_i_size is less that inode->i_size
1088 * but greater than i_disksize. (hint delalloc)
1090 ext4_update_i_disksize(inode
, (pos
+ copied
));
1094 page_cache_release(page
);
1097 * Don't mark the inode dirty under page lock. First, it unnecessarily
1098 * makes the holding time of page lock longer. Second, it forces lock
1099 * ordering of page lock and transaction start for journaling
1103 ext4_mark_inode_dirty(handle
, inode
);
1105 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1106 /* if we have allocated more blocks and copied
1107 * less. We will have blocks allocated outside
1108 * inode->i_size. So truncate them
1110 ext4_orphan_add(handle
, inode
);
1112 ret2
= ext4_journal_stop(handle
);
1116 if (pos
+ len
> inode
->i_size
) {
1117 ext4_truncate_failed_write(inode
);
1119 * If truncate failed early the inode might still be
1120 * on the orphan list; we need to make sure the inode
1121 * is removed from the orphan list in that case.
1124 ext4_orphan_del(NULL
, inode
);
1127 return ret
? ret
: copied
;
1130 static int ext4_journalled_write_end(struct file
*file
,
1131 struct address_space
*mapping
,
1132 loff_t pos
, unsigned len
, unsigned copied
,
1133 struct page
*page
, void *fsdata
)
1135 handle_t
*handle
= ext4_journal_current_handle();
1136 struct inode
*inode
= mapping
->host
;
1142 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1143 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1146 BUG_ON(!ext4_handle_valid(handle
));
1148 if (ext4_has_inline_data(inode
))
1149 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1153 if (!PageUptodate(page
))
1155 page_zero_new_buffers(page
, from
+copied
, to
);
1158 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1159 to
, &partial
, write_end_fn
);
1161 SetPageUptodate(page
);
1163 new_i_size
= pos
+ copied
;
1164 if (new_i_size
> inode
->i_size
)
1165 i_size_write(inode
, pos
+copied
);
1166 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1167 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1168 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1169 ext4_update_i_disksize(inode
, new_i_size
);
1170 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1176 page_cache_release(page
);
1177 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1178 /* if we have allocated more blocks and copied
1179 * less. We will have blocks allocated outside
1180 * inode->i_size. So truncate them
1182 ext4_orphan_add(handle
, inode
);
1184 ret2
= ext4_journal_stop(handle
);
1187 if (pos
+ len
> inode
->i_size
) {
1188 ext4_truncate_failed_write(inode
);
1190 * If truncate failed early the inode might still be
1191 * on the orphan list; we need to make sure the inode
1192 * is removed from the orphan list in that case.
1195 ext4_orphan_del(NULL
, inode
);
1198 return ret
? ret
: copied
;
1202 * Reserve a metadata for a single block located at lblock
1204 static int ext4_da_reserve_metadata(struct inode
*inode
, ext4_lblk_t lblock
)
1207 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1208 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1209 unsigned int md_needed
;
1210 ext4_lblk_t save_last_lblock
;
1214 * recalculate the amount of metadata blocks to reserve
1215 * in order to allocate nrblocks
1216 * worse case is one extent per block
1219 spin_lock(&ei
->i_block_reservation_lock
);
1221 * ext4_calc_metadata_amount() has side effects, which we have
1222 * to be prepared undo if we fail to claim space.
1224 save_len
= ei
->i_da_metadata_calc_len
;
1225 save_last_lblock
= ei
->i_da_metadata_calc_last_lblock
;
1226 md_needed
= EXT4_NUM_B2C(sbi
,
1227 ext4_calc_metadata_amount(inode
, lblock
));
1228 trace_ext4_da_reserve_space(inode
, md_needed
);
1231 * We do still charge estimated metadata to the sb though;
1232 * we cannot afford to run out of free blocks.
1234 if (ext4_claim_free_clusters(sbi
, md_needed
, 0)) {
1235 ei
->i_da_metadata_calc_len
= save_len
;
1236 ei
->i_da_metadata_calc_last_lblock
= save_last_lblock
;
1237 spin_unlock(&ei
->i_block_reservation_lock
);
1238 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1244 ei
->i_reserved_meta_blocks
+= md_needed
;
1245 spin_unlock(&ei
->i_block_reservation_lock
);
1247 return 0; /* success */
1251 * Reserve a single cluster located at lblock
1253 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1256 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1257 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1258 unsigned int md_needed
;
1260 ext4_lblk_t save_last_lblock
;
1264 * We will charge metadata quota at writeout time; this saves
1265 * us from metadata over-estimation, though we may go over by
1266 * a small amount in the end. Here we just reserve for data.
1268 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1273 * recalculate the amount of metadata blocks to reserve
1274 * in order to allocate nrblocks
1275 * worse case is one extent per block
1278 spin_lock(&ei
->i_block_reservation_lock
);
1280 * ext4_calc_metadata_amount() has side effects, which we have
1281 * to be prepared undo if we fail to claim space.
1283 save_len
= ei
->i_da_metadata_calc_len
;
1284 save_last_lblock
= ei
->i_da_metadata_calc_last_lblock
;
1285 md_needed
= EXT4_NUM_B2C(sbi
,
1286 ext4_calc_metadata_amount(inode
, lblock
));
1287 trace_ext4_da_reserve_space(inode
, md_needed
);
1290 * We do still charge estimated metadata to the sb though;
1291 * we cannot afford to run out of free blocks.
1293 if (ext4_claim_free_clusters(sbi
, md_needed
+ 1, 0)) {
1294 ei
->i_da_metadata_calc_len
= save_len
;
1295 ei
->i_da_metadata_calc_last_lblock
= save_last_lblock
;
1296 spin_unlock(&ei
->i_block_reservation_lock
);
1297 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1301 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1304 ei
->i_reserved_data_blocks
++;
1305 ei
->i_reserved_meta_blocks
+= md_needed
;
1306 spin_unlock(&ei
->i_block_reservation_lock
);
1308 return 0; /* success */
1311 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1313 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1314 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1317 return; /* Nothing to release, exit */
1319 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1321 trace_ext4_da_release_space(inode
, to_free
);
1322 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1324 * if there aren't enough reserved blocks, then the
1325 * counter is messed up somewhere. Since this
1326 * function is called from invalidate page, it's
1327 * harmless to return without any action.
1329 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1330 "ino %lu, to_free %d with only %d reserved "
1331 "data blocks", inode
->i_ino
, to_free
,
1332 ei
->i_reserved_data_blocks
);
1334 to_free
= ei
->i_reserved_data_blocks
;
1336 ei
->i_reserved_data_blocks
-= to_free
;
1338 if (ei
->i_reserved_data_blocks
== 0) {
1340 * We can release all of the reserved metadata blocks
1341 * only when we have written all of the delayed
1342 * allocation blocks.
1343 * Note that in case of bigalloc, i_reserved_meta_blocks,
1344 * i_reserved_data_blocks, etc. refer to number of clusters.
1346 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
1347 ei
->i_reserved_meta_blocks
);
1348 ei
->i_reserved_meta_blocks
= 0;
1349 ei
->i_da_metadata_calc_len
= 0;
1352 /* update fs dirty data blocks counter */
1353 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1355 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1357 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1360 static void ext4_da_page_release_reservation(struct page
*page
,
1361 unsigned int offset
,
1362 unsigned int length
)
1365 struct buffer_head
*head
, *bh
;
1366 unsigned int curr_off
= 0;
1367 struct inode
*inode
= page
->mapping
->host
;
1368 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1369 unsigned int stop
= offset
+ length
;
1373 BUG_ON(stop
> PAGE_CACHE_SIZE
|| stop
< length
);
1375 head
= page_buffers(page
);
1378 unsigned int next_off
= curr_off
+ bh
->b_size
;
1380 if (next_off
> stop
)
1383 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1385 clear_buffer_delay(bh
);
1387 curr_off
= next_off
;
1388 } while ((bh
= bh
->b_this_page
) != head
);
1391 lblk
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1392 ext4_es_remove_extent(inode
, lblk
, to_release
);
1395 /* If we have released all the blocks belonging to a cluster, then we
1396 * need to release the reserved space for that cluster. */
1397 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1398 while (num_clusters
> 0) {
1399 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1400 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1401 if (sbi
->s_cluster_ratio
== 1 ||
1402 !ext4_find_delalloc_cluster(inode
, lblk
))
1403 ext4_da_release_space(inode
, 1);
1410 * Delayed allocation stuff
1413 struct mpage_da_data
{
1414 struct inode
*inode
;
1415 struct writeback_control
*wbc
;
1417 pgoff_t first_page
; /* The first page to write */
1418 pgoff_t next_page
; /* Current page to examine */
1419 pgoff_t last_page
; /* Last page to examine */
1421 * Extent to map - this can be after first_page because that can be
1422 * fully mapped. We somewhat abuse m_flags to store whether the extent
1423 * is delalloc or unwritten.
1425 struct ext4_map_blocks map
;
1426 struct ext4_io_submit io_submit
; /* IO submission data */
1429 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1434 struct pagevec pvec
;
1435 struct inode
*inode
= mpd
->inode
;
1436 struct address_space
*mapping
= inode
->i_mapping
;
1438 /* This is necessary when next_page == 0. */
1439 if (mpd
->first_page
>= mpd
->next_page
)
1442 index
= mpd
->first_page
;
1443 end
= mpd
->next_page
- 1;
1445 ext4_lblk_t start
, last
;
1446 start
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1447 last
= end
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1448 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1451 pagevec_init(&pvec
, 0);
1452 while (index
<= end
) {
1453 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1456 for (i
= 0; i
< nr_pages
; i
++) {
1457 struct page
*page
= pvec
.pages
[i
];
1458 if (page
->index
> end
)
1460 BUG_ON(!PageLocked(page
));
1461 BUG_ON(PageWriteback(page
));
1463 block_invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
1464 ClearPageUptodate(page
);
1468 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1469 pagevec_release(&pvec
);
1473 static void ext4_print_free_blocks(struct inode
*inode
)
1475 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1476 struct super_block
*sb
= inode
->i_sb
;
1477 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1479 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1480 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1481 ext4_count_free_clusters(sb
)));
1482 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1483 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1484 (long long) EXT4_C2B(EXT4_SB(sb
),
1485 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1486 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1487 (long long) EXT4_C2B(EXT4_SB(sb
),
1488 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1489 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1490 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1491 ei
->i_reserved_data_blocks
);
1492 ext4_msg(sb
, KERN_CRIT
, "i_reserved_meta_blocks=%u",
1493 ei
->i_reserved_meta_blocks
);
1494 ext4_msg(sb
, KERN_CRIT
, "i_allocated_meta_blocks=%u",
1495 ei
->i_allocated_meta_blocks
);
1499 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1501 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1505 * This function is grabs code from the very beginning of
1506 * ext4_map_blocks, but assumes that the caller is from delayed write
1507 * time. This function looks up the requested blocks and sets the
1508 * buffer delay bit under the protection of i_data_sem.
1510 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1511 struct ext4_map_blocks
*map
,
1512 struct buffer_head
*bh
)
1514 struct extent_status es
;
1516 sector_t invalid_block
= ~((sector_t
) 0xffff);
1517 #ifdef ES_AGGRESSIVE_TEST
1518 struct ext4_map_blocks orig_map
;
1520 memcpy(&orig_map
, map
, sizeof(*map
));
1523 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1527 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1528 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1529 (unsigned long) map
->m_lblk
);
1531 /* Lookup extent status tree firstly */
1532 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1533 ext4_es_lru_add(inode
);
1534 if (ext4_es_is_hole(&es
)) {
1536 down_read((&EXT4_I(inode
)->i_data_sem
));
1541 * Delayed extent could be allocated by fallocate.
1542 * So we need to check it.
1544 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1545 map_bh(bh
, inode
->i_sb
, invalid_block
);
1547 set_buffer_delay(bh
);
1551 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1552 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1553 if (retval
> map
->m_len
)
1554 retval
= map
->m_len
;
1555 map
->m_len
= retval
;
1556 if (ext4_es_is_written(&es
))
1557 map
->m_flags
|= EXT4_MAP_MAPPED
;
1558 else if (ext4_es_is_unwritten(&es
))
1559 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1563 #ifdef ES_AGGRESSIVE_TEST
1564 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1570 * Try to see if we can get the block without requesting a new
1571 * file system block.
1573 down_read((&EXT4_I(inode
)->i_data_sem
));
1574 if (ext4_has_inline_data(inode
)) {
1576 * We will soon create blocks for this page, and let
1577 * us pretend as if the blocks aren't allocated yet.
1578 * In case of clusters, we have to handle the work
1579 * of mapping from cluster so that the reserved space
1580 * is calculated properly.
1582 if ((EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) &&
1583 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
1584 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
1586 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1587 retval
= ext4_ext_map_blocks(NULL
, inode
, map
,
1588 EXT4_GET_BLOCKS_NO_PUT_HOLE
);
1590 retval
= ext4_ind_map_blocks(NULL
, inode
, map
,
1591 EXT4_GET_BLOCKS_NO_PUT_HOLE
);
1597 * XXX: __block_prepare_write() unmaps passed block,
1601 * If the block was allocated from previously allocated cluster,
1602 * then we don't need to reserve it again. However we still need
1603 * to reserve metadata for every block we're going to write.
1605 if (!(map
->m_flags
& EXT4_MAP_FROM_CLUSTER
)) {
1606 ret
= ext4_da_reserve_space(inode
, iblock
);
1608 /* not enough space to reserve */
1613 ret
= ext4_da_reserve_metadata(inode
, iblock
);
1615 /* not enough space to reserve */
1621 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1622 ~0, EXTENT_STATUS_DELAYED
);
1628 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1629 * and it should not appear on the bh->b_state.
1631 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
1633 map_bh(bh
, inode
->i_sb
, invalid_block
);
1635 set_buffer_delay(bh
);
1636 } else if (retval
> 0) {
1638 unsigned long long status
;
1640 #ifdef ES_AGGRESSIVE_TEST
1641 if (retval
!= map
->m_len
) {
1642 printk("ES len assertion failed for inode: %lu "
1643 "retval %d != map->m_len %d "
1644 "in %s (lookup)\n", inode
->i_ino
, retval
,
1645 map
->m_len
, __func__
);
1649 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1650 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1651 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1652 map
->m_pblk
, status
);
1658 up_read((&EXT4_I(inode
)->i_data_sem
));
1664 * This is a special get_blocks_t callback which is used by
1665 * ext4_da_write_begin(). It will either return mapped block or
1666 * reserve space for a single block.
1668 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1669 * We also have b_blocknr = -1 and b_bdev initialized properly
1671 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1672 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1673 * initialized properly.
1675 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1676 struct buffer_head
*bh
, int create
)
1678 struct ext4_map_blocks map
;
1681 BUG_ON(create
== 0);
1682 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1684 map
.m_lblk
= iblock
;
1688 * first, we need to know whether the block is allocated already
1689 * preallocated blocks are unmapped but should treated
1690 * the same as allocated blocks.
1692 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1696 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1697 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1699 if (buffer_unwritten(bh
)) {
1700 /* A delayed write to unwritten bh should be marked
1701 * new and mapped. Mapped ensures that we don't do
1702 * get_block multiple times when we write to the same
1703 * offset and new ensures that we do proper zero out
1704 * for partial write.
1707 set_buffer_mapped(bh
);
1712 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1718 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1724 static int __ext4_journalled_writepage(struct page
*page
,
1727 struct address_space
*mapping
= page
->mapping
;
1728 struct inode
*inode
= mapping
->host
;
1729 struct buffer_head
*page_bufs
= NULL
;
1730 handle_t
*handle
= NULL
;
1731 int ret
= 0, err
= 0;
1732 int inline_data
= ext4_has_inline_data(inode
);
1733 struct buffer_head
*inode_bh
= NULL
;
1735 ClearPageChecked(page
);
1738 BUG_ON(page
->index
!= 0);
1739 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1740 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1741 if (inode_bh
== NULL
)
1744 page_bufs
= page_buffers(page
);
1749 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1752 /* As soon as we unlock the page, it can go away, but we have
1753 * references to buffers so we are safe */
1756 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1757 ext4_writepage_trans_blocks(inode
));
1758 if (IS_ERR(handle
)) {
1759 ret
= PTR_ERR(handle
);
1763 BUG_ON(!ext4_handle_valid(handle
));
1766 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1768 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1771 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1772 do_journal_get_write_access
);
1774 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1779 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1780 err
= ext4_journal_stop(handle
);
1784 if (!ext4_has_inline_data(inode
))
1785 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1787 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1794 * Note that we don't need to start a transaction unless we're journaling data
1795 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1796 * need to file the inode to the transaction's list in ordered mode because if
1797 * we are writing back data added by write(), the inode is already there and if
1798 * we are writing back data modified via mmap(), no one guarantees in which
1799 * transaction the data will hit the disk. In case we are journaling data, we
1800 * cannot start transaction directly because transaction start ranks above page
1801 * lock so we have to do some magic.
1803 * This function can get called via...
1804 * - ext4_writepages after taking page lock (have journal handle)
1805 * - journal_submit_inode_data_buffers (no journal handle)
1806 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1807 * - grab_page_cache when doing write_begin (have journal handle)
1809 * We don't do any block allocation in this function. If we have page with
1810 * multiple blocks we need to write those buffer_heads that are mapped. This
1811 * is important for mmaped based write. So if we do with blocksize 1K
1812 * truncate(f, 1024);
1813 * a = mmap(f, 0, 4096);
1815 * truncate(f, 4096);
1816 * we have in the page first buffer_head mapped via page_mkwrite call back
1817 * but other buffer_heads would be unmapped but dirty (dirty done via the
1818 * do_wp_page). So writepage should write the first block. If we modify
1819 * the mmap area beyond 1024 we will again get a page_fault and the
1820 * page_mkwrite callback will do the block allocation and mark the
1821 * buffer_heads mapped.
1823 * We redirty the page if we have any buffer_heads that is either delay or
1824 * unwritten in the page.
1826 * We can get recursively called as show below.
1828 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1831 * But since we don't do any block allocation we should not deadlock.
1832 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1834 static int ext4_writepage(struct page
*page
,
1835 struct writeback_control
*wbc
)
1840 struct buffer_head
*page_bufs
= NULL
;
1841 struct inode
*inode
= page
->mapping
->host
;
1842 struct ext4_io_submit io_submit
;
1844 trace_ext4_writepage(page
);
1845 size
= i_size_read(inode
);
1846 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1847 len
= size
& ~PAGE_CACHE_MASK
;
1849 len
= PAGE_CACHE_SIZE
;
1851 page_bufs
= page_buffers(page
);
1853 * We cannot do block allocation or other extent handling in this
1854 * function. If there are buffers needing that, we have to redirty
1855 * the page. But we may reach here when we do a journal commit via
1856 * journal_submit_inode_data_buffers() and in that case we must write
1857 * allocated buffers to achieve data=ordered mode guarantees.
1859 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
1860 ext4_bh_delay_or_unwritten
)) {
1861 redirty_page_for_writepage(wbc
, page
);
1862 if (current
->flags
& PF_MEMALLOC
) {
1864 * For memory cleaning there's no point in writing only
1865 * some buffers. So just bail out. Warn if we came here
1866 * from direct reclaim.
1868 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
1875 if (PageChecked(page
) && ext4_should_journal_data(inode
))
1877 * It's mmapped pagecache. Add buffers and journal it. There
1878 * doesn't seem much point in redirtying the page here.
1880 return __ext4_journalled_writepage(page
, len
);
1882 ext4_io_submit_init(&io_submit
, wbc
);
1883 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
1884 if (!io_submit
.io_end
) {
1885 redirty_page_for_writepage(wbc
, page
);
1889 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
);
1890 ext4_io_submit(&io_submit
);
1891 /* Drop io_end reference we got from init */
1892 ext4_put_io_end_defer(io_submit
.io_end
);
1896 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1899 * mballoc gives us at most this number of blocks...
1900 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1901 * The rest of mballoc seems to handle chunks upto full group size.
1903 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1906 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1908 * @mpd - extent of blocks
1909 * @lblk - logical number of the block in the file
1910 * @b_state - b_state of the buffer head added
1912 * the function is used to collect contig. blocks in same state
1914 static int mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
1915 unsigned long b_state
)
1917 struct ext4_map_blocks
*map
= &mpd
->map
;
1919 /* Don't go larger than mballoc is willing to allocate */
1920 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
1923 /* First block in the extent? */
1924 if (map
->m_len
== 0) {
1927 map
->m_flags
= b_state
& BH_FLAGS
;
1931 /* Can we merge the block to our big extent? */
1932 if (lblk
== map
->m_lblk
+ map
->m_len
&&
1933 (b_state
& BH_FLAGS
) == map
->m_flags
) {
1940 static bool add_page_bufs_to_extent(struct mpage_da_data
*mpd
,
1941 struct buffer_head
*head
,
1942 struct buffer_head
*bh
,
1945 struct inode
*inode
= mpd
->inode
;
1946 ext4_lblk_t blocks
= (i_size_read(inode
) + (1 << inode
->i_blkbits
) - 1)
1947 >> inode
->i_blkbits
;
1950 BUG_ON(buffer_locked(bh
));
1952 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
1953 (!buffer_delay(bh
) && !buffer_unwritten(bh
)) ||
1955 /* Found extent to map? */
1962 if (!mpage_add_bh_to_extent(mpd
, lblk
, bh
->b_state
))
1964 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
1968 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
1971 loff_t size
= i_size_read(mpd
->inode
);
1974 BUG_ON(page
->index
!= mpd
->first_page
);
1975 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1976 len
= size
& ~PAGE_CACHE_MASK
;
1978 len
= PAGE_CACHE_SIZE
;
1979 clear_page_dirty_for_io(page
);
1980 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
);
1982 mpd
->wbc
->nr_to_write
--;
1989 * mpage_map_buffers - update buffers corresponding to changed extent and
1990 * submit fully mapped pages for IO
1992 * @mpd - description of extent to map, on return next extent to map
1994 * Scan buffers corresponding to changed extent (we expect corresponding pages
1995 * to be already locked) and update buffer state according to new extent state.
1996 * We map delalloc buffers to their physical location, clear unwritten bits,
1997 * and mark buffers as uninit when we perform writes to uninitialized extents
1998 * and do extent conversion after IO is finished. If the last page is not fully
1999 * mapped, we update @map to the next extent in the last page that needs
2000 * mapping. Otherwise we submit the page for IO.
2002 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2004 struct pagevec pvec
;
2006 struct inode
*inode
= mpd
->inode
;
2007 struct buffer_head
*head
, *bh
;
2008 int bpp_bits
= PAGE_CACHE_SHIFT
- inode
->i_blkbits
;
2009 ext4_lblk_t blocks
= (i_size_read(inode
) + (1 << inode
->i_blkbits
) - 1)
2010 >> inode
->i_blkbits
;
2016 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2017 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2018 lblk
= start
<< bpp_bits
;
2019 pblock
= mpd
->map
.m_pblk
;
2021 pagevec_init(&pvec
, 0);
2022 while (start
<= end
) {
2023 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, start
,
2027 for (i
= 0; i
< nr_pages
; i
++) {
2028 struct page
*page
= pvec
.pages
[i
];
2030 if (page
->index
> end
)
2032 /* Upto 'end' pages must be contiguous */
2033 BUG_ON(page
->index
!= start
);
2034 bh
= head
= page_buffers(page
);
2036 if (lblk
< mpd
->map
.m_lblk
)
2038 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2040 * Buffer after end of mapped extent.
2041 * Find next buffer in the page to map.
2044 mpd
->map
.m_flags
= 0;
2045 add_page_bufs_to_extent(mpd
, head
, bh
,
2047 pagevec_release(&pvec
);
2050 if (buffer_delay(bh
)) {
2051 clear_buffer_delay(bh
);
2052 bh
->b_blocknr
= pblock
++;
2054 clear_buffer_unwritten(bh
);
2055 } while (++lblk
< blocks
&&
2056 (bh
= bh
->b_this_page
) != head
);
2059 * FIXME: This is going to break if dioread_nolock
2060 * supports blocksize < pagesize as we will try to
2061 * convert potentially unmapped parts of inode.
2063 mpd
->io_submit
.io_end
->size
+= PAGE_CACHE_SIZE
;
2064 /* Page fully mapped - let IO run! */
2065 err
= mpage_submit_page(mpd
, page
);
2067 pagevec_release(&pvec
);
2072 pagevec_release(&pvec
);
2074 /* Extent fully mapped and matches with page boundary. We are done. */
2076 mpd
->map
.m_flags
= 0;
2080 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2082 struct inode
*inode
= mpd
->inode
;
2083 struct ext4_map_blocks
*map
= &mpd
->map
;
2084 int get_blocks_flags
;
2087 trace_ext4_da_write_pages_extent(inode
, map
);
2089 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2090 * to convert an uninitialized extent to be initialized (in the case
2091 * where we have written into one or more preallocated blocks). It is
2092 * possible that we're going to need more metadata blocks than
2093 * previously reserved. However we must not fail because we're in
2094 * writeback and there is nothing we can do about it so it might result
2095 * in data loss. So use reserved blocks to allocate metadata if
2098 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if the blocks
2099 * in question are delalloc blocks. This affects functions in many
2100 * different parts of the allocation call path. This flag exists
2101 * primarily because we don't want to change *many* call functions, so
2102 * ext4_map_blocks() will set the EXT4_STATE_DELALLOC_RESERVED flag
2103 * once the inode's allocation semaphore is taken.
2105 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2106 EXT4_GET_BLOCKS_METADATA_NOFAIL
;
2107 if (ext4_should_dioread_nolock(inode
))
2108 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2109 if (map
->m_flags
& (1 << BH_Delay
))
2110 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2112 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2115 if (map
->m_flags
& EXT4_MAP_UNINIT
) {
2116 if (!mpd
->io_submit
.io_end
->handle
&&
2117 ext4_handle_valid(handle
)) {
2118 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2119 handle
->h_rsv_handle
= NULL
;
2121 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2124 BUG_ON(map
->m_len
== 0);
2125 if (map
->m_flags
& EXT4_MAP_NEW
) {
2126 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2129 for (i
= 0; i
< map
->m_len
; i
++)
2130 unmap_underlying_metadata(bdev
, map
->m_pblk
+ i
);
2136 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2137 * mpd->len and submit pages underlying it for IO
2139 * @handle - handle for journal operations
2140 * @mpd - extent to map
2142 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2143 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2144 * them to initialized or split the described range from larger unwritten
2145 * extent. Note that we need not map all the described range since allocation
2146 * can return less blocks or the range is covered by more unwritten extents. We
2147 * cannot map more because we are limited by reserved transaction credits. On
2148 * the other hand we always make sure that the last touched page is fully
2149 * mapped so that it can be written out (and thus forward progress is
2150 * guaranteed). After mapping we submit all mapped pages for IO.
2152 static int mpage_map_and_submit_extent(handle_t
*handle
,
2153 struct mpage_da_data
*mpd
,
2154 bool *give_up_on_write
)
2156 struct inode
*inode
= mpd
->inode
;
2157 struct ext4_map_blocks
*map
= &mpd
->map
;
2161 mpd
->io_submit
.io_end
->offset
=
2162 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2164 err
= mpage_map_one_extent(handle
, mpd
);
2166 struct super_block
*sb
= inode
->i_sb
;
2168 if (EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2169 goto invalidate_dirty_pages
;
2171 * Let the uper layers retry transient errors.
2172 * In the case of ENOSPC, if ext4_count_free_blocks()
2173 * is non-zero, a commit should free up blocks.
2175 if ((err
== -ENOMEM
) ||
2176 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
)))
2178 ext4_msg(sb
, KERN_CRIT
,
2179 "Delayed block allocation failed for "
2180 "inode %lu at logical offset %llu with"
2181 " max blocks %u with error %d",
2183 (unsigned long long)map
->m_lblk
,
2184 (unsigned)map
->m_len
, -err
);
2185 ext4_msg(sb
, KERN_CRIT
,
2186 "This should not happen!! Data will "
2189 ext4_print_free_blocks(inode
);
2190 invalidate_dirty_pages
:
2191 *give_up_on_write
= true;
2195 * Update buffer state, submit mapped pages, and get us new
2198 err
= mpage_map_and_submit_buffers(mpd
);
2201 } while (map
->m_len
);
2203 /* Update on-disk size after IO is submitted */
2204 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_CACHE_SHIFT
;
2205 if (disksize
> i_size_read(inode
))
2206 disksize
= i_size_read(inode
);
2207 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2210 ext4_update_i_disksize(inode
, disksize
);
2211 err2
= ext4_mark_inode_dirty(handle
, inode
);
2213 ext4_error(inode
->i_sb
,
2214 "Failed to mark inode %lu dirty",
2223 * Calculate the total number of credits to reserve for one writepages
2224 * iteration. This is called from ext4_writepages(). We map an extent of
2225 * upto MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2226 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2227 * bpp - 1 blocks in bpp different extents.
2229 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2231 int bpp
= ext4_journal_blocks_per_page(inode
);
2233 return ext4_meta_trans_blocks(inode
,
2234 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2238 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2239 * and underlying extent to map
2241 * @mpd - where to look for pages
2243 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2244 * IO immediately. When we find a page which isn't mapped we start accumulating
2245 * extent of buffers underlying these pages that needs mapping (formed by
2246 * either delayed or unwritten buffers). We also lock the pages containing
2247 * these buffers. The extent found is returned in @mpd structure (starting at
2248 * mpd->lblk with length mpd->len blocks).
2250 * Note that this function can attach bios to one io_end structure which are
2251 * neither logically nor physically contiguous. Although it may seem as an
2252 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2253 * case as we need to track IO to all buffers underlying a page in one io_end.
2255 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2257 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2258 struct pagevec pvec
;
2259 unsigned int nr_pages
;
2260 pgoff_t index
= mpd
->first_page
;
2261 pgoff_t end
= mpd
->last_page
;
2264 int blkbits
= mpd
->inode
->i_blkbits
;
2266 struct buffer_head
*head
;
2268 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2269 tag
= PAGECACHE_TAG_TOWRITE
;
2271 tag
= PAGECACHE_TAG_DIRTY
;
2273 pagevec_init(&pvec
, 0);
2275 mpd
->next_page
= index
;
2276 while (index
<= end
) {
2277 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2278 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2282 for (i
= 0; i
< nr_pages
; i
++) {
2283 struct page
*page
= pvec
.pages
[i
];
2286 * At this point, the page may be truncated or
2287 * invalidated (changing page->mapping to NULL), or
2288 * even swizzled back from swapper_space to tmpfs file
2289 * mapping. However, page->index will not change
2290 * because we have a reference on the page.
2292 if (page
->index
> end
)
2295 /* If we can't merge this page, we are done. */
2296 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2301 * If the page is no longer dirty, or its mapping no
2302 * longer corresponds to inode we are writing (which
2303 * means it has been truncated or invalidated), or the
2304 * page is already under writeback and we are not doing
2305 * a data integrity writeback, skip the page
2307 if (!PageDirty(page
) ||
2308 (PageWriteback(page
) &&
2309 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2310 unlikely(page
->mapping
!= mapping
)) {
2315 wait_on_page_writeback(page
);
2316 BUG_ON(PageWriteback(page
));
2318 if (mpd
->map
.m_len
== 0)
2319 mpd
->first_page
= page
->index
;
2320 mpd
->next_page
= page
->index
+ 1;
2321 /* Add all dirty buffers to mpd */
2322 lblk
= ((ext4_lblk_t
)page
->index
) <<
2323 (PAGE_CACHE_SHIFT
- blkbits
);
2324 head
= page_buffers(page
);
2325 if (!add_page_bufs_to_extent(mpd
, head
, head
, lblk
))
2327 /* So far everything mapped? Submit the page for IO. */
2328 if (mpd
->map
.m_len
== 0) {
2329 err
= mpage_submit_page(mpd
, page
);
2335 * Accumulated enough dirty pages? This doesn't apply
2336 * to WB_SYNC_ALL mode. For integrity sync we have to
2337 * keep going because someone may be concurrently
2338 * dirtying pages, and we might have synced a lot of
2339 * newly appeared dirty pages, but have not synced all
2340 * of the old dirty pages.
2342 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&&
2343 mpd
->next_page
- mpd
->first_page
>=
2344 mpd
->wbc
->nr_to_write
)
2347 pagevec_release(&pvec
);
2352 pagevec_release(&pvec
);
2356 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2359 struct address_space
*mapping
= data
;
2360 int ret
= ext4_writepage(page
, wbc
);
2361 mapping_set_error(mapping
, ret
);
2365 static int ext4_writepages(struct address_space
*mapping
,
2366 struct writeback_control
*wbc
)
2368 pgoff_t writeback_index
= 0;
2369 long nr_to_write
= wbc
->nr_to_write
;
2370 int range_whole
= 0;
2372 handle_t
*handle
= NULL
;
2373 struct mpage_da_data mpd
;
2374 struct inode
*inode
= mapping
->host
;
2375 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2376 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2378 struct blk_plug plug
;
2379 bool give_up_on_write
= false;
2381 trace_ext4_writepages(inode
, wbc
);
2384 * No pages to write? This is mainly a kludge to avoid starting
2385 * a transaction for special inodes like journal inode on last iput()
2386 * because that could violate lock ordering on umount
2388 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2391 if (ext4_should_journal_data(inode
)) {
2392 struct blk_plug plug
;
2395 blk_start_plug(&plug
);
2396 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2397 blk_finish_plug(&plug
);
2402 * If the filesystem has aborted, it is read-only, so return
2403 * right away instead of dumping stack traces later on that
2404 * will obscure the real source of the problem. We test
2405 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2406 * the latter could be true if the filesystem is mounted
2407 * read-only, and in that case, ext4_writepages should
2408 * *never* be called, so if that ever happens, we would want
2411 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2414 if (ext4_should_dioread_nolock(inode
)) {
2416 * We may need to convert upto one extent per block in
2417 * the page and we may dirty the inode.
2419 rsv_blocks
= 1 + (PAGE_CACHE_SIZE
>> inode
->i_blkbits
);
2423 * If we have inline data and arrive here, it means that
2424 * we will soon create the block for the 1st page, so
2425 * we'd better clear the inline data here.
2427 if (ext4_has_inline_data(inode
)) {
2428 /* Just inode will be modified... */
2429 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2430 if (IS_ERR(handle
)) {
2431 ret
= PTR_ERR(handle
);
2432 goto out_writepages
;
2434 BUG_ON(ext4_test_inode_state(inode
,
2435 EXT4_STATE_MAY_INLINE_DATA
));
2436 ext4_destroy_inline_data(handle
, inode
);
2437 ext4_journal_stop(handle
);
2440 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2443 if (wbc
->range_cyclic
) {
2444 writeback_index
= mapping
->writeback_index
;
2445 if (writeback_index
)
2447 mpd
.first_page
= writeback_index
;
2450 mpd
.first_page
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2451 mpd
.last_page
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2456 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2458 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2459 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2461 blk_start_plug(&plug
);
2462 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2463 /* For each extent of pages we use new io_end */
2464 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2465 if (!mpd
.io_submit
.io_end
) {
2471 * We have two constraints: We find one extent to map and we
2472 * must always write out whole page (makes a difference when
2473 * blocksize < pagesize) so that we don't block on IO when we
2474 * try to write out the rest of the page. Journalled mode is
2475 * not supported by delalloc.
2477 BUG_ON(ext4_should_journal_data(inode
));
2478 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2480 /* start a new transaction */
2481 handle
= ext4_journal_start_with_reserve(inode
,
2482 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2483 if (IS_ERR(handle
)) {
2484 ret
= PTR_ERR(handle
);
2485 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2486 "%ld pages, ino %lu; err %d", __func__
,
2487 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2488 /* Release allocated io_end */
2489 ext4_put_io_end(mpd
.io_submit
.io_end
);
2493 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2494 ret
= mpage_prepare_extent_to_map(&mpd
);
2497 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2501 * We scanned the whole range (or exhausted
2502 * nr_to_write), submitted what was mapped and
2503 * didn't find anything needing mapping. We are
2509 ext4_journal_stop(handle
);
2510 /* Submit prepared bio */
2511 ext4_io_submit(&mpd
.io_submit
);
2512 /* Unlock pages we didn't use */
2513 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2514 /* Drop our io_end reference we got from init */
2515 ext4_put_io_end(mpd
.io_submit
.io_end
);
2517 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2519 * Commit the transaction which would
2520 * free blocks released in the transaction
2523 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2527 /* Fatal error - ENOMEM, EIO... */
2531 blk_finish_plug(&plug
);
2532 if (!ret
&& !cycled
) {
2534 mpd
.last_page
= writeback_index
- 1;
2540 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2542 * Set the writeback_index so that range_cyclic
2543 * mode will write it back later
2545 mapping
->writeback_index
= mpd
.first_page
;
2548 trace_ext4_writepages_result(inode
, wbc
, ret
,
2549 nr_to_write
- wbc
->nr_to_write
);
2553 static int ext4_nonda_switch(struct super_block
*sb
)
2555 s64 free_clusters
, dirty_clusters
;
2556 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2559 * switch to non delalloc mode if we are running low
2560 * on free block. The free block accounting via percpu
2561 * counters can get slightly wrong with percpu_counter_batch getting
2562 * accumulated on each CPU without updating global counters
2563 * Delalloc need an accurate free block accounting. So switch
2564 * to non delalloc when we are near to error range.
2567 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2569 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2571 * Start pushing delalloc when 1/2 of free blocks are dirty.
2573 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2574 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2576 if (2 * free_clusters
< 3 * dirty_clusters
||
2577 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2579 * free block count is less than 150% of dirty blocks
2580 * or free blocks is less than watermark
2587 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2588 loff_t pos
, unsigned len
, unsigned flags
,
2589 struct page
**pagep
, void **fsdata
)
2591 int ret
, retries
= 0;
2594 struct inode
*inode
= mapping
->host
;
2597 index
= pos
>> PAGE_CACHE_SHIFT
;
2599 if (ext4_nonda_switch(inode
->i_sb
)) {
2600 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2601 return ext4_write_begin(file
, mapping
, pos
,
2602 len
, flags
, pagep
, fsdata
);
2604 *fsdata
= (void *)0;
2605 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2607 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2608 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2618 * grab_cache_page_write_begin() can take a long time if the
2619 * system is thrashing due to memory pressure, or if the page
2620 * is being written back. So grab it first before we start
2621 * the transaction handle. This also allows us to allocate
2622 * the page (if needed) without using GFP_NOFS.
2625 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2631 * With delayed allocation, we don't log the i_disksize update
2632 * if there is delayed block allocation. But we still need
2633 * to journalling the i_disksize update if writes to the end
2634 * of file which has an already mapped buffer.
2637 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, 1);
2638 if (IS_ERR(handle
)) {
2639 page_cache_release(page
);
2640 return PTR_ERR(handle
);
2644 if (page
->mapping
!= mapping
) {
2645 /* The page got truncated from under us */
2647 page_cache_release(page
);
2648 ext4_journal_stop(handle
);
2651 /* In case writeback began while the page was unlocked */
2652 wait_on_page_writeback(page
);
2654 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2657 ext4_journal_stop(handle
);
2659 * block_write_begin may have instantiated a few blocks
2660 * outside i_size. Trim these off again. Don't need
2661 * i_size_read because we hold i_mutex.
2663 if (pos
+ len
> inode
->i_size
)
2664 ext4_truncate_failed_write(inode
);
2666 if (ret
== -ENOSPC
&&
2667 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2670 page_cache_release(page
);
2679 * Check if we should update i_disksize
2680 * when write to the end of file but not require block allocation
2682 static int ext4_da_should_update_i_disksize(struct page
*page
,
2683 unsigned long offset
)
2685 struct buffer_head
*bh
;
2686 struct inode
*inode
= page
->mapping
->host
;
2690 bh
= page_buffers(page
);
2691 idx
= offset
>> inode
->i_blkbits
;
2693 for (i
= 0; i
< idx
; i
++)
2694 bh
= bh
->b_this_page
;
2696 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2701 static int ext4_da_write_end(struct file
*file
,
2702 struct address_space
*mapping
,
2703 loff_t pos
, unsigned len
, unsigned copied
,
2704 struct page
*page
, void *fsdata
)
2706 struct inode
*inode
= mapping
->host
;
2708 handle_t
*handle
= ext4_journal_current_handle();
2710 unsigned long start
, end
;
2711 int write_mode
= (int)(unsigned long)fsdata
;
2713 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
2714 return ext4_write_end(file
, mapping
, pos
,
2715 len
, copied
, page
, fsdata
);
2717 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2718 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2719 end
= start
+ copied
- 1;
2722 * generic_write_end() will run mark_inode_dirty() if i_size
2723 * changes. So let's piggyback the i_disksize mark_inode_dirty
2726 new_i_size
= pos
+ copied
;
2727 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2728 if (ext4_has_inline_data(inode
) ||
2729 ext4_da_should_update_i_disksize(page
, end
)) {
2730 down_write(&EXT4_I(inode
)->i_data_sem
);
2731 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
2732 EXT4_I(inode
)->i_disksize
= new_i_size
;
2733 up_write(&EXT4_I(inode
)->i_data_sem
);
2734 /* We need to mark inode dirty even if
2735 * new_i_size is less that inode->i_size
2736 * bu greater than i_disksize.(hint delalloc)
2738 ext4_mark_inode_dirty(handle
, inode
);
2742 if (write_mode
!= CONVERT_INLINE_DATA
&&
2743 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
2744 ext4_has_inline_data(inode
))
2745 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
2748 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2754 ret2
= ext4_journal_stop(handle
);
2758 return ret
? ret
: copied
;
2761 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
2762 unsigned int length
)
2765 * Drop reserved blocks
2767 BUG_ON(!PageLocked(page
));
2768 if (!page_has_buffers(page
))
2771 ext4_da_page_release_reservation(page
, offset
, length
);
2774 ext4_invalidatepage(page
, offset
, length
);
2780 * Force all delayed allocation blocks to be allocated for a given inode.
2782 int ext4_alloc_da_blocks(struct inode
*inode
)
2784 trace_ext4_alloc_da_blocks(inode
);
2786 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2787 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2791 * We do something simple for now. The filemap_flush() will
2792 * also start triggering a write of the data blocks, which is
2793 * not strictly speaking necessary (and for users of
2794 * laptop_mode, not even desirable). However, to do otherwise
2795 * would require replicating code paths in:
2797 * ext4_writepages() ->
2798 * write_cache_pages() ---> (via passed in callback function)
2799 * __mpage_da_writepage() -->
2800 * mpage_add_bh_to_extent()
2801 * mpage_da_map_blocks()
2803 * The problem is that write_cache_pages(), located in
2804 * mm/page-writeback.c, marks pages clean in preparation for
2805 * doing I/O, which is not desirable if we're not planning on
2808 * We could call write_cache_pages(), and then redirty all of
2809 * the pages by calling redirty_page_for_writepage() but that
2810 * would be ugly in the extreme. So instead we would need to
2811 * replicate parts of the code in the above functions,
2812 * simplifying them because we wouldn't actually intend to
2813 * write out the pages, but rather only collect contiguous
2814 * logical block extents, call the multi-block allocator, and
2815 * then update the buffer heads with the block allocations.
2817 * For now, though, we'll cheat by calling filemap_flush(),
2818 * which will map the blocks, and start the I/O, but not
2819 * actually wait for the I/O to complete.
2821 return filemap_flush(inode
->i_mapping
);
2825 * bmap() is special. It gets used by applications such as lilo and by
2826 * the swapper to find the on-disk block of a specific piece of data.
2828 * Naturally, this is dangerous if the block concerned is still in the
2829 * journal. If somebody makes a swapfile on an ext4 data-journaling
2830 * filesystem and enables swap, then they may get a nasty shock when the
2831 * data getting swapped to that swapfile suddenly gets overwritten by
2832 * the original zero's written out previously to the journal and
2833 * awaiting writeback in the kernel's buffer cache.
2835 * So, if we see any bmap calls here on a modified, data-journaled file,
2836 * take extra steps to flush any blocks which might be in the cache.
2838 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2840 struct inode
*inode
= mapping
->host
;
2845 * We can get here for an inline file via the FIBMAP ioctl
2847 if (ext4_has_inline_data(inode
))
2850 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2851 test_opt(inode
->i_sb
, DELALLOC
)) {
2853 * With delalloc we want to sync the file
2854 * so that we can make sure we allocate
2857 filemap_write_and_wait(mapping
);
2860 if (EXT4_JOURNAL(inode
) &&
2861 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2863 * This is a REALLY heavyweight approach, but the use of
2864 * bmap on dirty files is expected to be extremely rare:
2865 * only if we run lilo or swapon on a freshly made file
2866 * do we expect this to happen.
2868 * (bmap requires CAP_SYS_RAWIO so this does not
2869 * represent an unprivileged user DOS attack --- we'd be
2870 * in trouble if mortal users could trigger this path at
2873 * NB. EXT4_STATE_JDATA is not set on files other than
2874 * regular files. If somebody wants to bmap a directory
2875 * or symlink and gets confused because the buffer
2876 * hasn't yet been flushed to disk, they deserve
2877 * everything they get.
2880 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2881 journal
= EXT4_JOURNAL(inode
);
2882 jbd2_journal_lock_updates(journal
);
2883 err
= jbd2_journal_flush(journal
);
2884 jbd2_journal_unlock_updates(journal
);
2890 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2893 static int ext4_readpage(struct file
*file
, struct page
*page
)
2896 struct inode
*inode
= page
->mapping
->host
;
2898 trace_ext4_readpage(page
);
2900 if (ext4_has_inline_data(inode
))
2901 ret
= ext4_readpage_inline(inode
, page
);
2904 return mpage_readpage(page
, ext4_get_block
);
2910 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2911 struct list_head
*pages
, unsigned nr_pages
)
2913 struct inode
*inode
= mapping
->host
;
2915 /* If the file has inline data, no need to do readpages. */
2916 if (ext4_has_inline_data(inode
))
2919 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2922 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
2923 unsigned int length
)
2925 trace_ext4_invalidatepage(page
, offset
, length
);
2927 /* No journalling happens on data buffers when this function is used */
2928 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
2930 block_invalidatepage(page
, offset
, length
);
2933 static int __ext4_journalled_invalidatepage(struct page
*page
,
2934 unsigned int offset
,
2935 unsigned int length
)
2937 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2939 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
2942 * If it's a full truncate we just forget about the pending dirtying
2944 if (offset
== 0 && length
== PAGE_CACHE_SIZE
)
2945 ClearPageChecked(page
);
2947 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
2950 /* Wrapper for aops... */
2951 static void ext4_journalled_invalidatepage(struct page
*page
,
2952 unsigned int offset
,
2953 unsigned int length
)
2955 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
2958 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2960 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2962 trace_ext4_releasepage(page
);
2964 /* Page has dirty journalled data -> cannot release */
2965 if (PageChecked(page
))
2968 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2970 return try_to_free_buffers(page
);
2974 * ext4_get_block used when preparing for a DIO write or buffer write.
2975 * We allocate an uinitialized extent if blocks haven't been allocated.
2976 * The extent will be converted to initialized after the IO is complete.
2978 int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
2979 struct buffer_head
*bh_result
, int create
)
2981 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2982 inode
->i_ino
, create
);
2983 return _ext4_get_block(inode
, iblock
, bh_result
,
2984 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
2987 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
2988 struct buffer_head
*bh_result
, int create
)
2990 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2991 inode
->i_ino
, create
);
2992 return _ext4_get_block(inode
, iblock
, bh_result
,
2993 EXT4_GET_BLOCKS_NO_LOCK
);
2996 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
2997 ssize_t size
, void *private, int ret
,
3000 struct inode
*inode
= file_inode(iocb
->ki_filp
);
3001 ext4_io_end_t
*io_end
= iocb
->private;
3003 /* if not async direct IO just return */
3005 inode_dio_done(inode
);
3007 aio_complete(iocb
, ret
, 0);
3011 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3012 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3013 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3016 iocb
->private = NULL
;
3017 io_end
->offset
= offset
;
3018 io_end
->size
= size
;
3020 io_end
->iocb
= iocb
;
3021 io_end
->result
= ret
;
3023 ext4_put_io_end_defer(io_end
);
3027 * For ext4 extent files, ext4 will do direct-io write to holes,
3028 * preallocated extents, and those write extend the file, no need to
3029 * fall back to buffered IO.
3031 * For holes, we fallocate those blocks, mark them as uninitialized
3032 * If those blocks were preallocated, we mark sure they are split, but
3033 * still keep the range to write as uninitialized.
3035 * The unwritten extents will be converted to written when DIO is completed.
3036 * For async direct IO, since the IO may still pending when return, we
3037 * set up an end_io call back function, which will do the conversion
3038 * when async direct IO completed.
3040 * If the O_DIRECT write will extend the file then add this inode to the
3041 * orphan list. So recovery will truncate it back to the original size
3042 * if the machine crashes during the write.
3045 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3046 const struct iovec
*iov
, loff_t offset
,
3047 unsigned long nr_segs
)
3049 struct file
*file
= iocb
->ki_filp
;
3050 struct inode
*inode
= file
->f_mapping
->host
;
3052 size_t count
= iov_length(iov
, nr_segs
);
3054 get_block_t
*get_block_func
= NULL
;
3056 loff_t final_size
= offset
+ count
;
3057 ext4_io_end_t
*io_end
= NULL
;
3059 /* Use the old path for reads and writes beyond i_size. */
3060 if (rw
!= WRITE
|| final_size
> inode
->i_size
)
3061 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3063 BUG_ON(iocb
->private == NULL
);
3066 * Make all waiters for direct IO properly wait also for extent
3067 * conversion. This also disallows race between truncate() and
3068 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3071 atomic_inc(&inode
->i_dio_count
);
3073 /* If we do a overwrite dio, i_mutex locking can be released */
3074 overwrite
= *((int *)iocb
->private);
3077 down_read(&EXT4_I(inode
)->i_data_sem
);
3078 mutex_unlock(&inode
->i_mutex
);
3082 * We could direct write to holes and fallocate.
3084 * Allocated blocks to fill the hole are marked as
3085 * uninitialized to prevent parallel buffered read to expose
3086 * the stale data before DIO complete the data IO.
3088 * As to previously fallocated extents, ext4 get_block will
3089 * just simply mark the buffer mapped but still keep the
3090 * extents uninitialized.
3092 * For non AIO case, we will convert those unwritten extents
3093 * to written after return back from blockdev_direct_IO.
3095 * For async DIO, the conversion needs to be deferred when the
3096 * IO is completed. The ext4 end_io callback function will be
3097 * called to take care of the conversion work. Here for async
3098 * case, we allocate an io_end structure to hook to the iocb.
3100 iocb
->private = NULL
;
3101 ext4_inode_aio_set(inode
, NULL
);
3102 if (!is_sync_kiocb(iocb
)) {
3103 io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
3108 io_end
->flag
|= EXT4_IO_END_DIRECT
;
3110 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3112 iocb
->private = ext4_get_io_end(io_end
);
3114 * we save the io structure for current async direct
3115 * IO, so that later ext4_map_blocks() could flag the
3116 * io structure whether there is a unwritten extents
3117 * needs to be converted when IO is completed.
3119 ext4_inode_aio_set(inode
, io_end
);
3123 get_block_func
= ext4_get_block_write_nolock
;
3125 get_block_func
= ext4_get_block_write
;
3126 dio_flags
= DIO_LOCKING
;
3128 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3129 inode
->i_sb
->s_bdev
, iov
,
3137 * Put our reference to io_end. This can free the io_end structure e.g.
3138 * in sync IO case or in case of error. It can even perform extent
3139 * conversion if all bios we submitted finished before we got here.
3140 * Note that in that case iocb->private can be already set to NULL
3144 ext4_inode_aio_set(inode
, NULL
);
3145 ext4_put_io_end(io_end
);
3147 * When no IO was submitted ext4_end_io_dio() was not
3148 * called so we have to put iocb's reference.
3150 if (ret
<= 0 && ret
!= -EIOCBQUEUED
&& iocb
->private) {
3151 WARN_ON(iocb
->private != io_end
);
3152 WARN_ON(io_end
->flag
& EXT4_IO_END_UNWRITTEN
);
3153 WARN_ON(io_end
->iocb
);
3155 * Generic code already did inode_dio_done() so we
3156 * have to clear EXT4_IO_END_DIRECT to not do it for
3160 ext4_put_io_end(io_end
);
3161 iocb
->private = NULL
;
3164 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3165 EXT4_STATE_DIO_UNWRITTEN
)) {
3168 * for non AIO case, since the IO is already
3169 * completed, we could do the conversion right here
3171 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3175 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3180 inode_dio_done(inode
);
3181 /* take i_mutex locking again if we do a ovewrite dio */
3183 up_read(&EXT4_I(inode
)->i_data_sem
);
3184 mutex_lock(&inode
->i_mutex
);
3190 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3191 const struct iovec
*iov
, loff_t offset
,
3192 unsigned long nr_segs
)
3194 struct file
*file
= iocb
->ki_filp
;
3195 struct inode
*inode
= file
->f_mapping
->host
;
3199 * If we are doing data journalling we don't support O_DIRECT
3201 if (ext4_should_journal_data(inode
))
3204 /* Let buffer I/O handle the inline data case. */
3205 if (ext4_has_inline_data(inode
))
3208 trace_ext4_direct_IO_enter(inode
, offset
, iov_length(iov
, nr_segs
), rw
);
3209 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3210 ret
= ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3212 ret
= ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3213 trace_ext4_direct_IO_exit(inode
, offset
,
3214 iov_length(iov
, nr_segs
), rw
, ret
);
3219 * Pages can be marked dirty completely asynchronously from ext4's journalling
3220 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3221 * much here because ->set_page_dirty is called under VFS locks. The page is
3222 * not necessarily locked.
3224 * We cannot just dirty the page and leave attached buffers clean, because the
3225 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3226 * or jbddirty because all the journalling code will explode.
3228 * So what we do is to mark the page "pending dirty" and next time writepage
3229 * is called, propagate that into the buffers appropriately.
3231 static int ext4_journalled_set_page_dirty(struct page
*page
)
3233 SetPageChecked(page
);
3234 return __set_page_dirty_nobuffers(page
);
3237 static const struct address_space_operations ext4_aops
= {
3238 .readpage
= ext4_readpage
,
3239 .readpages
= ext4_readpages
,
3240 .writepage
= ext4_writepage
,
3241 .writepages
= ext4_writepages
,
3242 .write_begin
= ext4_write_begin
,
3243 .write_end
= ext4_write_end
,
3245 .invalidatepage
= ext4_invalidatepage
,
3246 .releasepage
= ext4_releasepage
,
3247 .direct_IO
= ext4_direct_IO
,
3248 .migratepage
= buffer_migrate_page
,
3249 .is_partially_uptodate
= block_is_partially_uptodate
,
3250 .error_remove_page
= generic_error_remove_page
,
3253 static const struct address_space_operations ext4_journalled_aops
= {
3254 .readpage
= ext4_readpage
,
3255 .readpages
= ext4_readpages
,
3256 .writepage
= ext4_writepage
,
3257 .writepages
= ext4_writepages
,
3258 .write_begin
= ext4_write_begin
,
3259 .write_end
= ext4_journalled_write_end
,
3260 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3262 .invalidatepage
= ext4_journalled_invalidatepage
,
3263 .releasepage
= ext4_releasepage
,
3264 .direct_IO
= ext4_direct_IO
,
3265 .is_partially_uptodate
= block_is_partially_uptodate
,
3266 .error_remove_page
= generic_error_remove_page
,
3269 static const struct address_space_operations ext4_da_aops
= {
3270 .readpage
= ext4_readpage
,
3271 .readpages
= ext4_readpages
,
3272 .writepage
= ext4_writepage
,
3273 .writepages
= ext4_writepages
,
3274 .write_begin
= ext4_da_write_begin
,
3275 .write_end
= ext4_da_write_end
,
3277 .invalidatepage
= ext4_da_invalidatepage
,
3278 .releasepage
= ext4_releasepage
,
3279 .direct_IO
= ext4_direct_IO
,
3280 .migratepage
= buffer_migrate_page
,
3281 .is_partially_uptodate
= block_is_partially_uptodate
,
3282 .error_remove_page
= generic_error_remove_page
,
3285 void ext4_set_aops(struct inode
*inode
)
3287 switch (ext4_inode_journal_mode(inode
)) {
3288 case EXT4_INODE_ORDERED_DATA_MODE
:
3289 ext4_set_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3291 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3292 ext4_clear_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3294 case EXT4_INODE_JOURNAL_DATA_MODE
:
3295 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3300 if (test_opt(inode
->i_sb
, DELALLOC
))
3301 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3303 inode
->i_mapping
->a_ops
= &ext4_aops
;
3307 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3308 * up to the end of the block which corresponds to `from'.
3309 * This required during truncate. We need to physically zero the tail end
3310 * of that block so it doesn't yield old data if the file is later grown.
3312 int ext4_block_truncate_page(handle_t
*handle
,
3313 struct address_space
*mapping
, loff_t from
)
3315 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3318 struct inode
*inode
= mapping
->host
;
3320 blocksize
= inode
->i_sb
->s_blocksize
;
3321 length
= blocksize
- (offset
& (blocksize
- 1));
3323 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3327 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3328 * starting from file offset 'from'. The range to be zero'd must
3329 * be contained with in one block. If the specified range exceeds
3330 * the end of the block it will be shortened to end of the block
3331 * that cooresponds to 'from'
3333 int ext4_block_zero_page_range(handle_t
*handle
,
3334 struct address_space
*mapping
, loff_t from
, loff_t length
)
3336 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3337 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3338 unsigned blocksize
, max
, pos
;
3340 struct inode
*inode
= mapping
->host
;
3341 struct buffer_head
*bh
;
3345 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3346 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3350 blocksize
= inode
->i_sb
->s_blocksize
;
3351 max
= blocksize
- (offset
& (blocksize
- 1));
3354 * correct length if it does not fall between
3355 * 'from' and the end of the block
3357 if (length
> max
|| length
< 0)
3360 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3362 if (!page_has_buffers(page
))
3363 create_empty_buffers(page
, blocksize
, 0);
3365 /* Find the buffer that contains "offset" */
3366 bh
= page_buffers(page
);
3368 while (offset
>= pos
) {
3369 bh
= bh
->b_this_page
;
3373 if (buffer_freed(bh
)) {
3374 BUFFER_TRACE(bh
, "freed: skip");
3377 if (!buffer_mapped(bh
)) {
3378 BUFFER_TRACE(bh
, "unmapped");
3379 ext4_get_block(inode
, iblock
, bh
, 0);
3380 /* unmapped? It's a hole - nothing to do */
3381 if (!buffer_mapped(bh
)) {
3382 BUFFER_TRACE(bh
, "still unmapped");
3387 /* Ok, it's mapped. Make sure it's up-to-date */
3388 if (PageUptodate(page
))
3389 set_buffer_uptodate(bh
);
3391 if (!buffer_uptodate(bh
)) {
3393 ll_rw_block(READ
, 1, &bh
);
3395 /* Uhhuh. Read error. Complain and punt. */
3396 if (!buffer_uptodate(bh
))
3399 if (ext4_should_journal_data(inode
)) {
3400 BUFFER_TRACE(bh
, "get write access");
3401 err
= ext4_journal_get_write_access(handle
, bh
);
3405 zero_user(page
, offset
, length
);
3406 BUFFER_TRACE(bh
, "zeroed end of block");
3408 if (ext4_should_journal_data(inode
)) {
3409 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3412 mark_buffer_dirty(bh
);
3413 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
))
3414 err
= ext4_jbd2_file_inode(handle
, inode
);
3419 page_cache_release(page
);
3423 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
3424 loff_t lstart
, loff_t length
)
3426 struct super_block
*sb
= inode
->i_sb
;
3427 struct address_space
*mapping
= inode
->i_mapping
;
3428 unsigned partial_start
, partial_end
;
3429 ext4_fsblk_t start
, end
;
3430 loff_t byte_end
= (lstart
+ length
- 1);
3433 partial_start
= lstart
& (sb
->s_blocksize
- 1);
3434 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
3436 start
= lstart
>> sb
->s_blocksize_bits
;
3437 end
= byte_end
>> sb
->s_blocksize_bits
;
3439 /* Handle partial zero within the single block */
3441 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
3442 err
= ext4_block_zero_page_range(handle
, mapping
,
3446 /* Handle partial zero out on the start of the range */
3447 if (partial_start
) {
3448 err
= ext4_block_zero_page_range(handle
, mapping
,
3449 lstart
, sb
->s_blocksize
);
3453 /* Handle partial zero out on the end of the range */
3454 if (partial_end
!= sb
->s_blocksize
- 1)
3455 err
= ext4_block_zero_page_range(handle
, mapping
,
3456 byte_end
- partial_end
,
3461 int ext4_can_truncate(struct inode
*inode
)
3463 if (S_ISREG(inode
->i_mode
))
3465 if (S_ISDIR(inode
->i_mode
))
3467 if (S_ISLNK(inode
->i_mode
))
3468 return !ext4_inode_is_fast_symlink(inode
);
3473 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3474 * associated with the given offset and length
3476 * @inode: File inode
3477 * @offset: The offset where the hole will begin
3478 * @len: The length of the hole
3480 * Returns: 0 on success or negative on failure
3483 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
3485 struct super_block
*sb
= inode
->i_sb
;
3486 ext4_lblk_t first_block
, stop_block
;
3487 struct address_space
*mapping
= inode
->i_mapping
;
3488 loff_t first_block_offset
, last_block_offset
;
3490 unsigned int credits
;
3493 if (!S_ISREG(inode
->i_mode
))
3496 if (EXT4_SB(sb
)->s_cluster_ratio
> 1) {
3497 /* TODO: Add support for bigalloc file systems */
3501 trace_ext4_punch_hole(inode
, offset
, length
);
3504 * Write out all dirty pages to avoid race conditions
3505 * Then release them.
3507 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3508 ret
= filemap_write_and_wait_range(mapping
, offset
,
3509 offset
+ length
- 1);
3514 mutex_lock(&inode
->i_mutex
);
3515 /* It's not possible punch hole on append only file */
3516 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
)) {
3520 if (IS_SWAPFILE(inode
)) {
3525 /* No need to punch hole beyond i_size */
3526 if (offset
>= inode
->i_size
)
3530 * If the hole extends beyond i_size, set the hole
3531 * to end after the page that contains i_size
3533 if (offset
+ length
> inode
->i_size
) {
3534 length
= inode
->i_size
+
3535 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
3539 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
3540 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
3542 /* Now release the pages and zero block aligned part of pages*/
3543 if (last_block_offset
> first_block_offset
)
3544 truncate_pagecache_range(inode
, first_block_offset
,
3547 /* Wait all existing dio workers, newcomers will block on i_mutex */
3548 ext4_inode_block_unlocked_dio(inode
);
3549 inode_dio_wait(inode
);
3551 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3552 credits
= ext4_writepage_trans_blocks(inode
);
3554 credits
= ext4_blocks_for_truncate(inode
);
3555 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3556 if (IS_ERR(handle
)) {
3557 ret
= PTR_ERR(handle
);
3558 ext4_std_error(sb
, ret
);
3562 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
3567 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
3568 EXT4_BLOCK_SIZE_BITS(sb
);
3569 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
3571 /* If there are no blocks to remove, return now */
3572 if (first_block
>= stop_block
)
3575 down_write(&EXT4_I(inode
)->i_data_sem
);
3576 ext4_discard_preallocations(inode
);
3578 ret
= ext4_es_remove_extent(inode
, first_block
,
3579 stop_block
- first_block
);
3581 up_write(&EXT4_I(inode
)->i_data_sem
);
3585 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3586 ret
= ext4_ext_remove_space(inode
, first_block
,
3589 ret
= ext4_free_hole_blocks(handle
, inode
, first_block
,
3592 ext4_discard_preallocations(inode
);
3593 up_write(&EXT4_I(inode
)->i_data_sem
);
3595 ext4_handle_sync(handle
);
3596 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3597 ext4_mark_inode_dirty(handle
, inode
);
3599 ext4_journal_stop(handle
);
3601 ext4_inode_resume_unlocked_dio(inode
);
3603 mutex_unlock(&inode
->i_mutex
);
3610 * We block out ext4_get_block() block instantiations across the entire
3611 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3612 * simultaneously on behalf of the same inode.
3614 * As we work through the truncate and commit bits of it to the journal there
3615 * is one core, guiding principle: the file's tree must always be consistent on
3616 * disk. We must be able to restart the truncate after a crash.
3618 * The file's tree may be transiently inconsistent in memory (although it
3619 * probably isn't), but whenever we close off and commit a journal transaction,
3620 * the contents of (the filesystem + the journal) must be consistent and
3621 * restartable. It's pretty simple, really: bottom up, right to left (although
3622 * left-to-right works OK too).
3624 * Note that at recovery time, journal replay occurs *before* the restart of
3625 * truncate against the orphan inode list.
3627 * The committed inode has the new, desired i_size (which is the same as
3628 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3629 * that this inode's truncate did not complete and it will again call
3630 * ext4_truncate() to have another go. So there will be instantiated blocks
3631 * to the right of the truncation point in a crashed ext4 filesystem. But
3632 * that's fine - as long as they are linked from the inode, the post-crash
3633 * ext4_truncate() run will find them and release them.
3635 void ext4_truncate(struct inode
*inode
)
3637 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3638 unsigned int credits
;
3640 struct address_space
*mapping
= inode
->i_mapping
;
3643 * There is a possibility that we're either freeing the inode
3644 * or it completely new indode. In those cases we might not
3645 * have i_mutex locked because it's not necessary.
3647 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
3648 WARN_ON(!mutex_is_locked(&inode
->i_mutex
));
3649 trace_ext4_truncate_enter(inode
);
3651 if (!ext4_can_truncate(inode
))
3654 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3656 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3657 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3659 if (ext4_has_inline_data(inode
)) {
3662 ext4_inline_data_truncate(inode
, &has_inline
);
3667 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3668 credits
= ext4_writepage_trans_blocks(inode
);
3670 credits
= ext4_blocks_for_truncate(inode
);
3672 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3673 if (IS_ERR(handle
)) {
3674 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
3678 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
3679 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
3682 * We add the inode to the orphan list, so that if this
3683 * truncate spans multiple transactions, and we crash, we will
3684 * resume the truncate when the filesystem recovers. It also
3685 * marks the inode dirty, to catch the new size.
3687 * Implication: the file must always be in a sane, consistent
3688 * truncatable state while each transaction commits.
3690 if (ext4_orphan_add(handle
, inode
))
3693 down_write(&EXT4_I(inode
)->i_data_sem
);
3695 ext4_discard_preallocations(inode
);
3697 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3698 ext4_ext_truncate(handle
, inode
);
3700 ext4_ind_truncate(handle
, inode
);
3702 up_write(&ei
->i_data_sem
);
3705 ext4_handle_sync(handle
);
3709 * If this was a simple ftruncate() and the file will remain alive,
3710 * then we need to clear up the orphan record which we created above.
3711 * However, if this was a real unlink then we were called by
3712 * ext4_delete_inode(), and we allow that function to clean up the
3713 * orphan info for us.
3716 ext4_orphan_del(handle
, inode
);
3718 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3719 ext4_mark_inode_dirty(handle
, inode
);
3720 ext4_journal_stop(handle
);
3722 trace_ext4_truncate_exit(inode
);
3726 * ext4_get_inode_loc returns with an extra refcount against the inode's
3727 * underlying buffer_head on success. If 'in_mem' is true, we have all
3728 * data in memory that is needed to recreate the on-disk version of this
3731 static int __ext4_get_inode_loc(struct inode
*inode
,
3732 struct ext4_iloc
*iloc
, int in_mem
)
3734 struct ext4_group_desc
*gdp
;
3735 struct buffer_head
*bh
;
3736 struct super_block
*sb
= inode
->i_sb
;
3738 int inodes_per_block
, inode_offset
;
3741 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3744 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3745 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3750 * Figure out the offset within the block group inode table
3752 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3753 inode_offset
= ((inode
->i_ino
- 1) %
3754 EXT4_INODES_PER_GROUP(sb
));
3755 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3756 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3758 bh
= sb_getblk(sb
, block
);
3761 if (!buffer_uptodate(bh
)) {
3765 * If the buffer has the write error flag, we have failed
3766 * to write out another inode in the same block. In this
3767 * case, we don't have to read the block because we may
3768 * read the old inode data successfully.
3770 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3771 set_buffer_uptodate(bh
);
3773 if (buffer_uptodate(bh
)) {
3774 /* someone brought it uptodate while we waited */
3780 * If we have all information of the inode in memory and this
3781 * is the only valid inode in the block, we need not read the
3785 struct buffer_head
*bitmap_bh
;
3788 start
= inode_offset
& ~(inodes_per_block
- 1);
3790 /* Is the inode bitmap in cache? */
3791 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3792 if (unlikely(!bitmap_bh
))
3796 * If the inode bitmap isn't in cache then the
3797 * optimisation may end up performing two reads instead
3798 * of one, so skip it.
3800 if (!buffer_uptodate(bitmap_bh
)) {
3804 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3805 if (i
== inode_offset
)
3807 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3811 if (i
== start
+ inodes_per_block
) {
3812 /* all other inodes are free, so skip I/O */
3813 memset(bh
->b_data
, 0, bh
->b_size
);
3814 set_buffer_uptodate(bh
);
3822 * If we need to do any I/O, try to pre-readahead extra
3823 * blocks from the inode table.
3825 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3826 ext4_fsblk_t b
, end
, table
;
3828 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
3830 table
= ext4_inode_table(sb
, gdp
);
3831 /* s_inode_readahead_blks is always a power of 2 */
3832 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
3836 num
= EXT4_INODES_PER_GROUP(sb
);
3837 if (ext4_has_group_desc_csum(sb
))
3838 num
-= ext4_itable_unused_count(sb
, gdp
);
3839 table
+= num
/ inodes_per_block
;
3843 sb_breadahead(sb
, b
++);
3847 * There are other valid inodes in the buffer, this inode
3848 * has in-inode xattrs, or we don't have this inode in memory.
3849 * Read the block from disk.
3851 trace_ext4_load_inode(inode
);
3853 bh
->b_end_io
= end_buffer_read_sync
;
3854 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
3856 if (!buffer_uptodate(bh
)) {
3857 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3858 "unable to read itable block");
3868 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3870 /* We have all inode data except xattrs in memory here. */
3871 return __ext4_get_inode_loc(inode
, iloc
,
3872 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
3875 void ext4_set_inode_flags(struct inode
*inode
)
3877 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3879 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3880 if (flags
& EXT4_SYNC_FL
)
3881 inode
->i_flags
|= S_SYNC
;
3882 if (flags
& EXT4_APPEND_FL
)
3883 inode
->i_flags
|= S_APPEND
;
3884 if (flags
& EXT4_IMMUTABLE_FL
)
3885 inode
->i_flags
|= S_IMMUTABLE
;
3886 if (flags
& EXT4_NOATIME_FL
)
3887 inode
->i_flags
|= S_NOATIME
;
3888 if (flags
& EXT4_DIRSYNC_FL
)
3889 inode
->i_flags
|= S_DIRSYNC
;
3892 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3893 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3895 unsigned int vfs_fl
;
3896 unsigned long old_fl
, new_fl
;
3899 vfs_fl
= ei
->vfs_inode
.i_flags
;
3900 old_fl
= ei
->i_flags
;
3901 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3902 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
3904 if (vfs_fl
& S_SYNC
)
3905 new_fl
|= EXT4_SYNC_FL
;
3906 if (vfs_fl
& S_APPEND
)
3907 new_fl
|= EXT4_APPEND_FL
;
3908 if (vfs_fl
& S_IMMUTABLE
)
3909 new_fl
|= EXT4_IMMUTABLE_FL
;
3910 if (vfs_fl
& S_NOATIME
)
3911 new_fl
|= EXT4_NOATIME_FL
;
3912 if (vfs_fl
& S_DIRSYNC
)
3913 new_fl
|= EXT4_DIRSYNC_FL
;
3914 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
3917 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3918 struct ext4_inode_info
*ei
)
3921 struct inode
*inode
= &(ei
->vfs_inode
);
3922 struct super_block
*sb
= inode
->i_sb
;
3924 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3925 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3926 /* we are using combined 48 bit field */
3927 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
3928 le32_to_cpu(raw_inode
->i_blocks_lo
);
3929 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
3930 /* i_blocks represent file system block size */
3931 return i_blocks
<< (inode
->i_blkbits
- 9);
3936 return le32_to_cpu(raw_inode
->i_blocks_lo
);
3940 static inline void ext4_iget_extra_inode(struct inode
*inode
,
3941 struct ext4_inode
*raw_inode
,
3942 struct ext4_inode_info
*ei
)
3944 __le32
*magic
= (void *)raw_inode
+
3945 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
3946 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
3947 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
3948 ext4_find_inline_data_nolock(inode
);
3950 EXT4_I(inode
)->i_inline_off
= 0;
3953 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
3955 struct ext4_iloc iloc
;
3956 struct ext4_inode
*raw_inode
;
3957 struct ext4_inode_info
*ei
;
3958 struct inode
*inode
;
3959 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
3965 inode
= iget_locked(sb
, ino
);
3967 return ERR_PTR(-ENOMEM
);
3968 if (!(inode
->i_state
& I_NEW
))
3974 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
3977 raw_inode
= ext4_raw_inode(&iloc
);
3979 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3980 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
3981 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
3982 EXT4_INODE_SIZE(inode
->i_sb
)) {
3983 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
3984 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
3985 EXT4_INODE_SIZE(inode
->i_sb
));
3990 ei
->i_extra_isize
= 0;
3992 /* Precompute checksum seed for inode metadata */
3993 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3994 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
)) {
3995 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3997 __le32 inum
= cpu_to_le32(inode
->i_ino
);
3998 __le32 gen
= raw_inode
->i_generation
;
3999 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4001 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4005 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4006 EXT4_ERROR_INODE(inode
, "checksum invalid");
4011 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4012 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4013 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4014 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4015 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4016 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4018 i_uid_write(inode
, i_uid
);
4019 i_gid_write(inode
, i_gid
);
4020 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4022 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4023 ei
->i_inline_off
= 0;
4024 ei
->i_dir_start_lookup
= 0;
4025 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4026 /* We now have enough fields to check if the inode was active or not.
4027 * This is needed because nfsd might try to access dead inodes
4028 * the test is that same one that e2fsck uses
4029 * NeilBrown 1999oct15
4031 if (inode
->i_nlink
== 0) {
4032 if ((inode
->i_mode
== 0 ||
4033 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4034 ino
!= EXT4_BOOT_LOADER_INO
) {
4035 /* this inode is deleted */
4039 /* The only unlinked inodes we let through here have
4040 * valid i_mode and are being read by the orphan
4041 * recovery code: that's fine, we're about to complete
4042 * the process of deleting those.
4043 * OR it is the EXT4_BOOT_LOADER_INO which is
4044 * not initialized on a new filesystem. */
4046 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4047 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4048 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4049 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4051 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4052 inode
->i_size
= ext4_isize(raw_inode
);
4053 ei
->i_disksize
= inode
->i_size
;
4055 ei
->i_reserved_quota
= 0;
4057 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4058 ei
->i_block_group
= iloc
.block_group
;
4059 ei
->i_last_alloc_group
= ~0;
4061 * NOTE! The in-memory inode i_data array is in little-endian order
4062 * even on big-endian machines: we do NOT byteswap the block numbers!
4064 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4065 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4066 INIT_LIST_HEAD(&ei
->i_orphan
);
4069 * Set transaction id's of transactions that have to be committed
4070 * to finish f[data]sync. We set them to currently running transaction
4071 * as we cannot be sure that the inode or some of its metadata isn't
4072 * part of the transaction - the inode could have been reclaimed and
4073 * now it is reread from disk.
4076 transaction_t
*transaction
;
4079 read_lock(&journal
->j_state_lock
);
4080 if (journal
->j_running_transaction
)
4081 transaction
= journal
->j_running_transaction
;
4083 transaction
= journal
->j_committing_transaction
;
4085 tid
= transaction
->t_tid
;
4087 tid
= journal
->j_commit_sequence
;
4088 read_unlock(&journal
->j_state_lock
);
4089 ei
->i_sync_tid
= tid
;
4090 ei
->i_datasync_tid
= tid
;
4093 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4094 if (ei
->i_extra_isize
== 0) {
4095 /* The extra space is currently unused. Use it. */
4096 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4097 EXT4_GOOD_OLD_INODE_SIZE
;
4099 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4103 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4104 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4105 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4106 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4108 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4109 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4110 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4112 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4116 if (ei
->i_file_acl
&&
4117 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4118 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4122 } else if (!ext4_has_inline_data(inode
)) {
4123 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4124 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4125 (S_ISLNK(inode
->i_mode
) &&
4126 !ext4_inode_is_fast_symlink(inode
))))
4127 /* Validate extent which is part of inode */
4128 ret
= ext4_ext_check_inode(inode
);
4129 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4130 (S_ISLNK(inode
->i_mode
) &&
4131 !ext4_inode_is_fast_symlink(inode
))) {
4132 /* Validate block references which are part of inode */
4133 ret
= ext4_ind_check_inode(inode
);
4139 if (S_ISREG(inode
->i_mode
)) {
4140 inode
->i_op
= &ext4_file_inode_operations
;
4141 inode
->i_fop
= &ext4_file_operations
;
4142 ext4_set_aops(inode
);
4143 } else if (S_ISDIR(inode
->i_mode
)) {
4144 inode
->i_op
= &ext4_dir_inode_operations
;
4145 inode
->i_fop
= &ext4_dir_operations
;
4146 } else if (S_ISLNK(inode
->i_mode
)) {
4147 if (ext4_inode_is_fast_symlink(inode
)) {
4148 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4149 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4150 sizeof(ei
->i_data
) - 1);
4152 inode
->i_op
= &ext4_symlink_inode_operations
;
4153 ext4_set_aops(inode
);
4155 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4156 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4157 inode
->i_op
= &ext4_special_inode_operations
;
4158 if (raw_inode
->i_block
[0])
4159 init_special_inode(inode
, inode
->i_mode
,
4160 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4162 init_special_inode(inode
, inode
->i_mode
,
4163 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4164 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4165 make_bad_inode(inode
);
4168 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4172 ext4_set_inode_flags(inode
);
4173 unlock_new_inode(inode
);
4179 return ERR_PTR(ret
);
4182 static int ext4_inode_blocks_set(handle_t
*handle
,
4183 struct ext4_inode
*raw_inode
,
4184 struct ext4_inode_info
*ei
)
4186 struct inode
*inode
= &(ei
->vfs_inode
);
4187 u64 i_blocks
= inode
->i_blocks
;
4188 struct super_block
*sb
= inode
->i_sb
;
4190 if (i_blocks
<= ~0U) {
4192 * i_blocks can be represented in a 32 bit variable
4193 * as multiple of 512 bytes
4195 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4196 raw_inode
->i_blocks_high
= 0;
4197 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4200 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4203 if (i_blocks
<= 0xffffffffffffULL
) {
4205 * i_blocks can be represented in a 48 bit variable
4206 * as multiple of 512 bytes
4208 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4209 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4210 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4212 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4213 /* i_block is stored in file system block size */
4214 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4215 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4216 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4222 * Post the struct inode info into an on-disk inode location in the
4223 * buffer-cache. This gobbles the caller's reference to the
4224 * buffer_head in the inode location struct.
4226 * The caller must have write access to iloc->bh.
4228 static int ext4_do_update_inode(handle_t
*handle
,
4229 struct inode
*inode
,
4230 struct ext4_iloc
*iloc
)
4232 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4233 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4234 struct buffer_head
*bh
= iloc
->bh
;
4235 int err
= 0, rc
, block
;
4236 int need_datasync
= 0;
4240 /* For fields not not tracking in the in-memory inode,
4241 * initialise them to zero for new inodes. */
4242 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4243 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4245 ext4_get_inode_flags(ei
);
4246 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4247 i_uid
= i_uid_read(inode
);
4248 i_gid
= i_gid_read(inode
);
4249 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4250 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4251 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4253 * Fix up interoperability with old kernels. Otherwise, old inodes get
4254 * re-used with the upper 16 bits of the uid/gid intact
4257 raw_inode
->i_uid_high
=
4258 cpu_to_le16(high_16_bits(i_uid
));
4259 raw_inode
->i_gid_high
=
4260 cpu_to_le16(high_16_bits(i_gid
));
4262 raw_inode
->i_uid_high
= 0;
4263 raw_inode
->i_gid_high
= 0;
4266 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4267 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4268 raw_inode
->i_uid_high
= 0;
4269 raw_inode
->i_gid_high
= 0;
4271 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4273 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4274 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4275 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4276 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4278 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4280 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4281 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4282 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4283 cpu_to_le32(EXT4_OS_HURD
))
4284 raw_inode
->i_file_acl_high
=
4285 cpu_to_le16(ei
->i_file_acl
>> 32);
4286 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4287 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4288 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4291 if (ei
->i_disksize
> 0x7fffffffULL
) {
4292 struct super_block
*sb
= inode
->i_sb
;
4293 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4294 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4295 EXT4_SB(sb
)->s_es
->s_rev_level
==
4296 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4297 /* If this is the first large file
4298 * created, add a flag to the superblock.
4300 err
= ext4_journal_get_write_access(handle
,
4301 EXT4_SB(sb
)->s_sbh
);
4304 ext4_update_dynamic_rev(sb
);
4305 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4306 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4307 ext4_handle_sync(handle
);
4308 err
= ext4_handle_dirty_super(handle
, sb
);
4311 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4312 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4313 if (old_valid_dev(inode
->i_rdev
)) {
4314 raw_inode
->i_block
[0] =
4315 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4316 raw_inode
->i_block
[1] = 0;
4318 raw_inode
->i_block
[0] = 0;
4319 raw_inode
->i_block
[1] =
4320 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4321 raw_inode
->i_block
[2] = 0;
4323 } else if (!ext4_has_inline_data(inode
)) {
4324 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4325 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4328 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4329 if (ei
->i_extra_isize
) {
4330 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4331 raw_inode
->i_version_hi
=
4332 cpu_to_le32(inode
->i_version
>> 32);
4333 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4336 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4338 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4339 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4342 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4344 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4347 ext4_std_error(inode
->i_sb
, err
);
4352 * ext4_write_inode()
4354 * We are called from a few places:
4356 * - Within generic_file_write() for O_SYNC files.
4357 * Here, there will be no transaction running. We wait for any running
4358 * transaction to commit.
4360 * - Within sys_sync(), kupdate and such.
4361 * We wait on commit, if tol to.
4363 * - Within prune_icache() (PF_MEMALLOC == true)
4364 * Here we simply return. We can't afford to block kswapd on the
4367 * In all cases it is actually safe for us to return without doing anything,
4368 * because the inode has been copied into a raw inode buffer in
4369 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4372 * Note that we are absolutely dependent upon all inode dirtiers doing the
4373 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4374 * which we are interested.
4376 * It would be a bug for them to not do this. The code:
4378 * mark_inode_dirty(inode)
4380 * inode->i_size = expr;
4382 * is in error because a kswapd-driven write_inode() could occur while
4383 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4384 * will no longer be on the superblock's dirty inode list.
4386 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4390 if (current
->flags
& PF_MEMALLOC
)
4393 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4394 if (ext4_journal_current_handle()) {
4395 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4400 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
4403 err
= ext4_force_commit(inode
->i_sb
);
4405 struct ext4_iloc iloc
;
4407 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4410 if (wbc
->sync_mode
== WB_SYNC_ALL
)
4411 sync_dirty_buffer(iloc
.bh
);
4412 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4413 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4414 "IO error syncing inode");
4423 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4424 * buffers that are attached to a page stradding i_size and are undergoing
4425 * commit. In that case we have to wait for commit to finish and try again.
4427 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4431 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4432 tid_t commit_tid
= 0;
4435 offset
= inode
->i_size
& (PAGE_CACHE_SIZE
- 1);
4437 * All buffers in the last page remain valid? Then there's nothing to
4438 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4441 if (offset
> PAGE_CACHE_SIZE
- (1 << inode
->i_blkbits
))
4444 page
= find_lock_page(inode
->i_mapping
,
4445 inode
->i_size
>> PAGE_CACHE_SHIFT
);
4448 ret
= __ext4_journalled_invalidatepage(page
, offset
,
4449 PAGE_CACHE_SIZE
- offset
);
4451 page_cache_release(page
);
4455 read_lock(&journal
->j_state_lock
);
4456 if (journal
->j_committing_transaction
)
4457 commit_tid
= journal
->j_committing_transaction
->t_tid
;
4458 read_unlock(&journal
->j_state_lock
);
4460 jbd2_log_wait_commit(journal
, commit_tid
);
4467 * Called from notify_change.
4469 * We want to trap VFS attempts to truncate the file as soon as
4470 * possible. In particular, we want to make sure that when the VFS
4471 * shrinks i_size, we put the inode on the orphan list and modify
4472 * i_disksize immediately, so that during the subsequent flushing of
4473 * dirty pages and freeing of disk blocks, we can guarantee that any
4474 * commit will leave the blocks being flushed in an unused state on
4475 * disk. (On recovery, the inode will get truncated and the blocks will
4476 * be freed, so we have a strong guarantee that no future commit will
4477 * leave these blocks visible to the user.)
4479 * Another thing we have to assure is that if we are in ordered mode
4480 * and inode is still attached to the committing transaction, we must
4481 * we start writeout of all the dirty pages which are being truncated.
4482 * This way we are sure that all the data written in the previous
4483 * transaction are already on disk (truncate waits for pages under
4486 * Called with inode->i_mutex down.
4488 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4490 struct inode
*inode
= dentry
->d_inode
;
4493 const unsigned int ia_valid
= attr
->ia_valid
;
4495 error
= inode_change_ok(inode
, attr
);
4499 if (is_quota_modification(inode
, attr
))
4500 dquot_initialize(inode
);
4501 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4502 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4505 /* (user+group)*(old+new) structure, inode write (sb,
4506 * inode block, ? - but truncate inode update has it) */
4507 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
4508 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
4509 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
4510 if (IS_ERR(handle
)) {
4511 error
= PTR_ERR(handle
);
4514 error
= dquot_transfer(inode
, attr
);
4516 ext4_journal_stop(handle
);
4519 /* Update corresponding info in inode so that everything is in
4520 * one transaction */
4521 if (attr
->ia_valid
& ATTR_UID
)
4522 inode
->i_uid
= attr
->ia_uid
;
4523 if (attr
->ia_valid
& ATTR_GID
)
4524 inode
->i_gid
= attr
->ia_gid
;
4525 error
= ext4_mark_inode_dirty(handle
, inode
);
4526 ext4_journal_stop(handle
);
4529 if (attr
->ia_valid
& ATTR_SIZE
) {
4531 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4532 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4534 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4539 if (S_ISREG(inode
->i_mode
) &&
4540 attr
->ia_valid
& ATTR_SIZE
&&
4541 (attr
->ia_size
< inode
->i_size
)) {
4544 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
4545 if (IS_ERR(handle
)) {
4546 error
= PTR_ERR(handle
);
4549 if (ext4_handle_valid(handle
)) {
4550 error
= ext4_orphan_add(handle
, inode
);
4553 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4554 rc
= ext4_mark_inode_dirty(handle
, inode
);
4557 ext4_journal_stop(handle
);
4559 if (ext4_should_order_data(inode
)) {
4560 error
= ext4_begin_ordered_truncate(inode
,
4563 /* Do as much error cleanup as possible */
4564 handle
= ext4_journal_start(inode
,
4566 if (IS_ERR(handle
)) {
4567 ext4_orphan_del(NULL
, inode
);
4570 ext4_orphan_del(handle
, inode
);
4572 ext4_journal_stop(handle
);
4578 if (attr
->ia_valid
& ATTR_SIZE
) {
4579 if (attr
->ia_size
!= inode
->i_size
) {
4580 loff_t oldsize
= inode
->i_size
;
4582 i_size_write(inode
, attr
->ia_size
);
4584 * Blocks are going to be removed from the inode. Wait
4585 * for dio in flight. Temporarily disable
4586 * dioread_nolock to prevent livelock.
4589 if (!ext4_should_journal_data(inode
)) {
4590 ext4_inode_block_unlocked_dio(inode
);
4591 inode_dio_wait(inode
);
4592 ext4_inode_resume_unlocked_dio(inode
);
4594 ext4_wait_for_tail_page_commit(inode
);
4597 * Truncate pagecache after we've waited for commit
4598 * in data=journal mode to make pages freeable.
4600 truncate_pagecache(inode
, oldsize
, inode
->i_size
);
4602 ext4_truncate(inode
);
4606 setattr_copy(inode
, attr
);
4607 mark_inode_dirty(inode
);
4611 * If the call to ext4_truncate failed to get a transaction handle at
4612 * all, we need to clean up the in-core orphan list manually.
4614 if (orphan
&& inode
->i_nlink
)
4615 ext4_orphan_del(NULL
, inode
);
4617 if (!rc
&& (ia_valid
& ATTR_MODE
))
4618 rc
= ext4_acl_chmod(inode
);
4621 ext4_std_error(inode
->i_sb
, error
);
4627 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4630 struct inode
*inode
;
4631 unsigned long long delalloc_blocks
;
4633 inode
= dentry
->d_inode
;
4634 generic_fillattr(inode
, stat
);
4637 * We can't update i_blocks if the block allocation is delayed
4638 * otherwise in the case of system crash before the real block
4639 * allocation is done, we will have i_blocks inconsistent with
4640 * on-disk file blocks.
4641 * We always keep i_blocks updated together with real
4642 * allocation. But to not confuse with user, stat
4643 * will return the blocks that include the delayed allocation
4644 * blocks for this file.
4646 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
4647 EXT4_I(inode
)->i_reserved_data_blocks
);
4649 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
-9);
4653 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
4656 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4657 return ext4_ind_trans_blocks(inode
, lblocks
);
4658 return ext4_ext_index_trans_blocks(inode
, pextents
);
4662 * Account for index blocks, block groups bitmaps and block group
4663 * descriptor blocks if modify datablocks and index blocks
4664 * worse case, the indexs blocks spread over different block groups
4666 * If datablocks are discontiguous, they are possible to spread over
4667 * different block groups too. If they are contiguous, with flexbg,
4668 * they could still across block group boundary.
4670 * Also account for superblock, inode, quota and xattr blocks
4672 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
4675 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4681 * How many index blocks need to touch to map @lblocks logical blocks
4682 * to @pextents physical extents?
4684 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
4689 * Now let's see how many group bitmaps and group descriptors need
4692 groups
= idxblocks
+ pextents
;
4694 if (groups
> ngroups
)
4696 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4697 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4699 /* bitmaps and block group descriptor blocks */
4700 ret
+= groups
+ gdpblocks
;
4702 /* Blocks for super block, inode, quota and xattr blocks */
4703 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4709 * Calculate the total number of credits to reserve to fit
4710 * the modification of a single pages into a single transaction,
4711 * which may include multiple chunks of block allocations.
4713 * This could be called via ext4_write_begin()
4715 * We need to consider the worse case, when
4716 * one new block per extent.
4718 int ext4_writepage_trans_blocks(struct inode
*inode
)
4720 int bpp
= ext4_journal_blocks_per_page(inode
);
4723 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
4725 /* Account for data blocks for journalled mode */
4726 if (ext4_should_journal_data(inode
))
4732 * Calculate the journal credits for a chunk of data modification.
4734 * This is called from DIO, fallocate or whoever calling
4735 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4737 * journal buffers for data blocks are not included here, as DIO
4738 * and fallocate do no need to journal data buffers.
4740 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4742 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4746 * The caller must have previously called ext4_reserve_inode_write().
4747 * Give this, we know that the caller already has write access to iloc->bh.
4749 int ext4_mark_iloc_dirty(handle_t
*handle
,
4750 struct inode
*inode
, struct ext4_iloc
*iloc
)
4754 if (IS_I_VERSION(inode
))
4755 inode_inc_iversion(inode
);
4757 /* the do_update_inode consumes one bh->b_count */
4760 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4761 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4767 * On success, We end up with an outstanding reference count against
4768 * iloc->bh. This _must_ be cleaned up later.
4772 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4773 struct ext4_iloc
*iloc
)
4777 err
= ext4_get_inode_loc(inode
, iloc
);
4779 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4780 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4786 ext4_std_error(inode
->i_sb
, err
);
4791 * Expand an inode by new_extra_isize bytes.
4792 * Returns 0 on success or negative error number on failure.
4794 static int ext4_expand_extra_isize(struct inode
*inode
,
4795 unsigned int new_extra_isize
,
4796 struct ext4_iloc iloc
,
4799 struct ext4_inode
*raw_inode
;
4800 struct ext4_xattr_ibody_header
*header
;
4802 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4805 raw_inode
= ext4_raw_inode(&iloc
);
4807 header
= IHDR(inode
, raw_inode
);
4809 /* No extended attributes present */
4810 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
4811 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4812 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4814 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4818 /* try to expand with EAs present */
4819 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4824 * What we do here is to mark the in-core inode as clean with respect to inode
4825 * dirtiness (it may still be data-dirty).
4826 * This means that the in-core inode may be reaped by prune_icache
4827 * without having to perform any I/O. This is a very good thing,
4828 * because *any* task may call prune_icache - even ones which
4829 * have a transaction open against a different journal.
4831 * Is this cheating? Not really. Sure, we haven't written the
4832 * inode out, but prune_icache isn't a user-visible syncing function.
4833 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4834 * we start and wait on commits.
4836 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4838 struct ext4_iloc iloc
;
4839 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4840 static unsigned int mnt_count
;
4844 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
4845 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4846 if (ext4_handle_valid(handle
) &&
4847 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4848 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
4850 * We need extra buffer credits since we may write into EA block
4851 * with this same handle. If journal_extend fails, then it will
4852 * only result in a minor loss of functionality for that inode.
4853 * If this is felt to be critical, then e2fsck should be run to
4854 * force a large enough s_min_extra_isize.
4856 if ((jbd2_journal_extend(handle
,
4857 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4858 ret
= ext4_expand_extra_isize(inode
,
4859 sbi
->s_want_extra_isize
,
4862 ext4_set_inode_state(inode
,
4863 EXT4_STATE_NO_EXPAND
);
4865 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4866 ext4_warning(inode
->i_sb
,
4867 "Unable to expand inode %lu. Delete"
4868 " some EAs or run e2fsck.",
4871 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4877 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4882 * ext4_dirty_inode() is called from __mark_inode_dirty()
4884 * We're really interested in the case where a file is being extended.
4885 * i_size has been changed by generic_commit_write() and we thus need
4886 * to include the updated inode in the current transaction.
4888 * Also, dquot_alloc_block() will always dirty the inode when blocks
4889 * are allocated to the file.
4891 * If the inode is marked synchronous, we don't honour that here - doing
4892 * so would cause a commit on atime updates, which we don't bother doing.
4893 * We handle synchronous inodes at the highest possible level.
4895 void ext4_dirty_inode(struct inode
*inode
, int flags
)
4899 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
4903 ext4_mark_inode_dirty(handle
, inode
);
4905 ext4_journal_stop(handle
);
4912 * Bind an inode's backing buffer_head into this transaction, to prevent
4913 * it from being flushed to disk early. Unlike
4914 * ext4_reserve_inode_write, this leaves behind no bh reference and
4915 * returns no iloc structure, so the caller needs to repeat the iloc
4916 * lookup to mark the inode dirty later.
4918 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4920 struct ext4_iloc iloc
;
4924 err
= ext4_get_inode_loc(inode
, &iloc
);
4926 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4927 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4929 err
= ext4_handle_dirty_metadata(handle
,
4935 ext4_std_error(inode
->i_sb
, err
);
4940 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4947 * We have to be very careful here: changing a data block's
4948 * journaling status dynamically is dangerous. If we write a
4949 * data block to the journal, change the status and then delete
4950 * that block, we risk forgetting to revoke the old log record
4951 * from the journal and so a subsequent replay can corrupt data.
4952 * So, first we make sure that the journal is empty and that
4953 * nobody is changing anything.
4956 journal
= EXT4_JOURNAL(inode
);
4959 if (is_journal_aborted(journal
))
4961 /* We have to allocate physical blocks for delalloc blocks
4962 * before flushing journal. otherwise delalloc blocks can not
4963 * be allocated any more. even more truncate on delalloc blocks
4964 * could trigger BUG by flushing delalloc blocks in journal.
4965 * There is no delalloc block in non-journal data mode.
4967 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
4968 err
= ext4_alloc_da_blocks(inode
);
4973 /* Wait for all existing dio workers */
4974 ext4_inode_block_unlocked_dio(inode
);
4975 inode_dio_wait(inode
);
4977 jbd2_journal_lock_updates(journal
);
4980 * OK, there are no updates running now, and all cached data is
4981 * synced to disk. We are now in a completely consistent state
4982 * which doesn't have anything in the journal, and we know that
4983 * no filesystem updates are running, so it is safe to modify
4984 * the inode's in-core data-journaling state flag now.
4988 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4990 jbd2_journal_flush(journal
);
4991 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4993 ext4_set_aops(inode
);
4995 jbd2_journal_unlock_updates(journal
);
4996 ext4_inode_resume_unlocked_dio(inode
);
4998 /* Finally we can mark the inode as dirty. */
5000 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5002 return PTR_ERR(handle
);
5004 err
= ext4_mark_inode_dirty(handle
, inode
);
5005 ext4_handle_sync(handle
);
5006 ext4_journal_stop(handle
);
5007 ext4_std_error(inode
->i_sb
, err
);
5012 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5014 return !buffer_mapped(bh
);
5017 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5019 struct page
*page
= vmf
->page
;
5023 struct file
*file
= vma
->vm_file
;
5024 struct inode
*inode
= file_inode(file
);
5025 struct address_space
*mapping
= inode
->i_mapping
;
5027 get_block_t
*get_block
;
5030 sb_start_pagefault(inode
->i_sb
);
5031 file_update_time(vma
->vm_file
);
5032 /* Delalloc case is easy... */
5033 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5034 !ext4_should_journal_data(inode
) &&
5035 !ext4_nonda_switch(inode
->i_sb
)) {
5037 ret
= __block_page_mkwrite(vma
, vmf
,
5038 ext4_da_get_block_prep
);
5039 } while (ret
== -ENOSPC
&&
5040 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5045 size
= i_size_read(inode
);
5046 /* Page got truncated from under us? */
5047 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5049 ret
= VM_FAULT_NOPAGE
;
5053 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5054 len
= size
& ~PAGE_CACHE_MASK
;
5056 len
= PAGE_CACHE_SIZE
;
5058 * Return if we have all the buffers mapped. This avoids the need to do
5059 * journal_start/journal_stop which can block and take a long time
5061 if (page_has_buffers(page
)) {
5062 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5064 ext4_bh_unmapped
)) {
5065 /* Wait so that we don't change page under IO */
5066 wait_for_stable_page(page
);
5067 ret
= VM_FAULT_LOCKED
;
5072 /* OK, we need to fill the hole... */
5073 if (ext4_should_dioread_nolock(inode
))
5074 get_block
= ext4_get_block_write
;
5076 get_block
= ext4_get_block
;
5078 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5079 ext4_writepage_trans_blocks(inode
));
5080 if (IS_ERR(handle
)) {
5081 ret
= VM_FAULT_SIGBUS
;
5084 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
5085 if (!ret
&& ext4_should_journal_data(inode
)) {
5086 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5087 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
5089 ret
= VM_FAULT_SIGBUS
;
5090 ext4_journal_stop(handle
);
5093 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5095 ext4_journal_stop(handle
);
5096 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5099 ret
= block_page_mkwrite_return(ret
);
5101 sb_end_pagefault(inode
->i_sb
);