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
);
558 if (unlikely(retval
!= map
->m_len
)) {
559 ext4_warning(inode
->i_sb
,
560 "ES len assertion failed for inode "
561 "%lu: retval %d != map->m_len %d",
562 inode
->i_ino
, retval
, map
->m_len
);
566 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
567 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
568 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
569 ext4_find_delalloc_range(inode
, map
->m_lblk
,
570 map
->m_lblk
+ map
->m_len
- 1))
571 status
|= EXTENT_STATUS_DELAYED
;
572 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
573 map
->m_len
, map
->m_pblk
, status
);
577 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
578 up_read((&EXT4_I(inode
)->i_data_sem
));
581 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
582 int ret
= check_block_validity(inode
, map
);
587 /* If it is only a block(s) look up */
588 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
592 * Returns if the blocks have already allocated
594 * Note that if blocks have been preallocated
595 * ext4_ext_get_block() returns the create = 0
596 * with buffer head unmapped.
598 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
602 * Here we clear m_flags because after allocating an new extent,
603 * it will be set again.
605 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
608 * New blocks allocate and/or writing to uninitialized extent
609 * will possibly result in updating i_data, so we take
610 * the write lock of i_data_sem, and call get_blocks()
611 * with create == 1 flag.
613 down_write((&EXT4_I(inode
)->i_data_sem
));
616 * if the caller is from delayed allocation writeout path
617 * we have already reserved fs blocks for allocation
618 * let the underlying get_block() function know to
619 * avoid double accounting
621 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
622 ext4_set_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
624 * We need to check for EXT4 here because migrate
625 * could have changed the inode type in between
627 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
628 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
630 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
632 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
634 * We allocated new blocks which will result in
635 * i_data's format changing. Force the migrate
636 * to fail by clearing migrate flags
638 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
642 * Update reserved blocks/metadata blocks after successful
643 * block allocation which had been deferred till now. We don't
644 * support fallocate for non extent files. So we can update
645 * reserve space here.
648 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
649 ext4_da_update_reserve_space(inode
, retval
, 1);
651 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
652 ext4_clear_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
658 if (unlikely(retval
!= map
->m_len
)) {
659 ext4_warning(inode
->i_sb
,
660 "ES len assertion failed for inode "
661 "%lu: retval %d != map->m_len %d",
662 inode
->i_ino
, retval
, map
->m_len
);
667 * If the extent has been zeroed out, we don't need to update
668 * extent status tree.
670 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
671 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
672 if (ext4_es_is_written(&es
))
675 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
676 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
677 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
678 ext4_find_delalloc_range(inode
, map
->m_lblk
,
679 map
->m_lblk
+ map
->m_len
- 1))
680 status
|= EXTENT_STATUS_DELAYED
;
681 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
682 map
->m_pblk
, status
);
688 up_write((&EXT4_I(inode
)->i_data_sem
));
689 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
690 int ret
= check_block_validity(inode
, map
);
697 /* Maximum number of blocks we map for direct IO at once. */
698 #define DIO_MAX_BLOCKS 4096
700 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
701 struct buffer_head
*bh
, int flags
)
703 handle_t
*handle
= ext4_journal_current_handle();
704 struct ext4_map_blocks map
;
705 int ret
= 0, started
= 0;
708 if (ext4_has_inline_data(inode
))
712 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
714 if (flags
&& !(flags
& EXT4_GET_BLOCKS_NO_LOCK
) && !handle
) {
715 /* Direct IO write... */
716 if (map
.m_len
> DIO_MAX_BLOCKS
)
717 map
.m_len
= DIO_MAX_BLOCKS
;
718 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
719 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
721 if (IS_ERR(handle
)) {
722 ret
= PTR_ERR(handle
);
728 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
730 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
731 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
732 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
736 ext4_journal_stop(handle
);
740 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
741 struct buffer_head
*bh
, int create
)
743 return _ext4_get_block(inode
, iblock
, bh
,
744 create
? EXT4_GET_BLOCKS_CREATE
: 0);
748 * `handle' can be NULL if create is zero
750 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
751 ext4_lblk_t block
, int create
, int *errp
)
753 struct ext4_map_blocks map
;
754 struct buffer_head
*bh
;
757 J_ASSERT(handle
!= NULL
|| create
== 0);
761 err
= ext4_map_blocks(handle
, inode
, &map
,
762 create
? EXT4_GET_BLOCKS_CREATE
: 0);
764 /* ensure we send some value back into *errp */
767 if (create
&& err
== 0)
768 err
= -ENOSPC
; /* should never happen */
774 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
779 if (map
.m_flags
& EXT4_MAP_NEW
) {
780 J_ASSERT(create
!= 0);
781 J_ASSERT(handle
!= NULL
);
784 * Now that we do not always journal data, we should
785 * keep in mind whether this should always journal the
786 * new buffer as metadata. For now, regular file
787 * writes use ext4_get_block instead, so it's not a
791 BUFFER_TRACE(bh
, "call get_create_access");
792 fatal
= ext4_journal_get_create_access(handle
, bh
);
793 if (!fatal
&& !buffer_uptodate(bh
)) {
794 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
795 set_buffer_uptodate(bh
);
798 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
799 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
803 BUFFER_TRACE(bh
, "not a new buffer");
813 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
814 ext4_lblk_t block
, int create
, int *err
)
816 struct buffer_head
*bh
;
818 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
821 if (buffer_uptodate(bh
))
823 ll_rw_block(READ
| REQ_META
| REQ_PRIO
, 1, &bh
);
825 if (buffer_uptodate(bh
))
832 int ext4_walk_page_buffers(handle_t
*handle
,
833 struct buffer_head
*head
,
837 int (*fn
)(handle_t
*handle
,
838 struct buffer_head
*bh
))
840 struct buffer_head
*bh
;
841 unsigned block_start
, block_end
;
842 unsigned blocksize
= head
->b_size
;
844 struct buffer_head
*next
;
846 for (bh
= head
, block_start
= 0;
847 ret
== 0 && (bh
!= head
|| !block_start
);
848 block_start
= block_end
, bh
= next
) {
849 next
= bh
->b_this_page
;
850 block_end
= block_start
+ blocksize
;
851 if (block_end
<= from
|| block_start
>= to
) {
852 if (partial
&& !buffer_uptodate(bh
))
856 err
= (*fn
)(handle
, bh
);
864 * To preserve ordering, it is essential that the hole instantiation and
865 * the data write be encapsulated in a single transaction. We cannot
866 * close off a transaction and start a new one between the ext4_get_block()
867 * and the commit_write(). So doing the jbd2_journal_start at the start of
868 * prepare_write() is the right place.
870 * Also, this function can nest inside ext4_writepage(). In that case, we
871 * *know* that ext4_writepage() has generated enough buffer credits to do the
872 * whole page. So we won't block on the journal in that case, which is good,
873 * because the caller may be PF_MEMALLOC.
875 * By accident, ext4 can be reentered when a transaction is open via
876 * quota file writes. If we were to commit the transaction while thus
877 * reentered, there can be a deadlock - we would be holding a quota
878 * lock, and the commit would never complete if another thread had a
879 * transaction open and was blocking on the quota lock - a ranking
882 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
883 * will _not_ run commit under these circumstances because handle->h_ref
884 * is elevated. We'll still have enough credits for the tiny quotafile
887 int do_journal_get_write_access(handle_t
*handle
,
888 struct buffer_head
*bh
)
890 int dirty
= buffer_dirty(bh
);
893 if (!buffer_mapped(bh
) || buffer_freed(bh
))
896 * __block_write_begin() could have dirtied some buffers. Clean
897 * the dirty bit as jbd2_journal_get_write_access() could complain
898 * otherwise about fs integrity issues. Setting of the dirty bit
899 * by __block_write_begin() isn't a real problem here as we clear
900 * the bit before releasing a page lock and thus writeback cannot
901 * ever write the buffer.
904 clear_buffer_dirty(bh
);
905 ret
= ext4_journal_get_write_access(handle
, bh
);
907 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
911 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
912 struct buffer_head
*bh_result
, int create
);
913 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
914 loff_t pos
, unsigned len
, unsigned flags
,
915 struct page
**pagep
, void **fsdata
)
917 struct inode
*inode
= mapping
->host
;
918 int ret
, needed_blocks
;
925 trace_ext4_write_begin(inode
, pos
, len
, flags
);
927 * Reserve one block more for addition to orphan list in case
928 * we allocate blocks but write fails for some reason
930 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
931 index
= pos
>> PAGE_CACHE_SHIFT
;
932 from
= pos
& (PAGE_CACHE_SIZE
- 1);
935 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
936 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
945 * grab_cache_page_write_begin() can take a long time if the
946 * system is thrashing due to memory pressure, or if the page
947 * is being written back. So grab it first before we start
948 * the transaction handle. This also allows us to allocate
949 * the page (if needed) without using GFP_NOFS.
952 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
958 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
959 if (IS_ERR(handle
)) {
960 page_cache_release(page
);
961 return PTR_ERR(handle
);
965 if (page
->mapping
!= mapping
) {
966 /* The page got truncated from under us */
968 page_cache_release(page
);
969 ext4_journal_stop(handle
);
972 /* In case writeback began while the page was unlocked */
973 wait_for_stable_page(page
);
975 if (ext4_should_dioread_nolock(inode
))
976 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
978 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
980 if (!ret
&& ext4_should_journal_data(inode
)) {
981 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
983 do_journal_get_write_access
);
989 * __block_write_begin may have instantiated a few blocks
990 * outside i_size. Trim these off again. Don't need
991 * i_size_read because we hold i_mutex.
993 * Add inode to orphan list in case we crash before
996 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
997 ext4_orphan_add(handle
, inode
);
999 ext4_journal_stop(handle
);
1000 if (pos
+ len
> inode
->i_size
) {
1001 ext4_truncate_failed_write(inode
);
1003 * If truncate failed early the inode might
1004 * still be on the orphan list; we need to
1005 * make sure the inode is removed from the
1006 * orphan list in that case.
1009 ext4_orphan_del(NULL
, inode
);
1012 if (ret
== -ENOSPC
&&
1013 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1015 page_cache_release(page
);
1022 /* For write_end() in data=journal mode */
1023 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1026 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1028 set_buffer_uptodate(bh
);
1029 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1030 clear_buffer_meta(bh
);
1031 clear_buffer_prio(bh
);
1036 * We need to pick up the new inode size which generic_commit_write gave us
1037 * `file' can be NULL - eg, when called from page_symlink().
1039 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1040 * buffers are managed internally.
1042 static int ext4_write_end(struct file
*file
,
1043 struct address_space
*mapping
,
1044 loff_t pos
, unsigned len
, unsigned copied
,
1045 struct page
*page
, void *fsdata
)
1047 handle_t
*handle
= ext4_journal_current_handle();
1048 struct inode
*inode
= mapping
->host
;
1050 int i_size_changed
= 0;
1052 trace_ext4_write_end(inode
, pos
, len
, copied
);
1053 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
)) {
1054 ret
= ext4_jbd2_file_inode(handle
, inode
);
1057 page_cache_release(page
);
1062 if (ext4_has_inline_data(inode
)) {
1063 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1069 copied
= block_write_end(file
, mapping
, pos
,
1070 len
, copied
, page
, fsdata
);
1073 * No need to use i_size_read() here, the i_size
1074 * cannot change under us because we hole i_mutex.
1076 * But it's important to update i_size while still holding page lock:
1077 * page writeout could otherwise come in and zero beyond i_size.
1079 if (pos
+ copied
> inode
->i_size
) {
1080 i_size_write(inode
, pos
+ copied
);
1084 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1085 /* We need to mark inode dirty even if
1086 * new_i_size is less that inode->i_size
1087 * but greater than i_disksize. (hint delalloc)
1089 ext4_update_i_disksize(inode
, (pos
+ copied
));
1093 page_cache_release(page
);
1096 * Don't mark the inode dirty under page lock. First, it unnecessarily
1097 * makes the holding time of page lock longer. Second, it forces lock
1098 * ordering of page lock and transaction start for journaling
1102 ext4_mark_inode_dirty(handle
, inode
);
1104 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1105 /* if we have allocated more blocks and copied
1106 * less. We will have blocks allocated outside
1107 * inode->i_size. So truncate them
1109 ext4_orphan_add(handle
, inode
);
1111 ret2
= ext4_journal_stop(handle
);
1115 if (pos
+ len
> inode
->i_size
) {
1116 ext4_truncate_failed_write(inode
);
1118 * If truncate failed early the inode might still be
1119 * on the orphan list; we need to make sure the inode
1120 * is removed from the orphan list in that case.
1123 ext4_orphan_del(NULL
, inode
);
1126 return ret
? ret
: copied
;
1129 static int ext4_journalled_write_end(struct file
*file
,
1130 struct address_space
*mapping
,
1131 loff_t pos
, unsigned len
, unsigned copied
,
1132 struct page
*page
, void *fsdata
)
1134 handle_t
*handle
= ext4_journal_current_handle();
1135 struct inode
*inode
= mapping
->host
;
1141 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1142 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1145 BUG_ON(!ext4_handle_valid(handle
));
1147 if (ext4_has_inline_data(inode
))
1148 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1152 if (!PageUptodate(page
))
1154 page_zero_new_buffers(page
, from
+copied
, to
);
1157 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1158 to
, &partial
, write_end_fn
);
1160 SetPageUptodate(page
);
1162 new_i_size
= pos
+ copied
;
1163 if (new_i_size
> inode
->i_size
)
1164 i_size_write(inode
, pos
+copied
);
1165 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1166 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1167 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1168 ext4_update_i_disksize(inode
, new_i_size
);
1169 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1175 page_cache_release(page
);
1176 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1177 /* if we have allocated more blocks and copied
1178 * less. We will have blocks allocated outside
1179 * inode->i_size. So truncate them
1181 ext4_orphan_add(handle
, inode
);
1183 ret2
= ext4_journal_stop(handle
);
1186 if (pos
+ len
> inode
->i_size
) {
1187 ext4_truncate_failed_write(inode
);
1189 * If truncate failed early the inode might still be
1190 * on the orphan list; we need to make sure the inode
1191 * is removed from the orphan list in that case.
1194 ext4_orphan_del(NULL
, inode
);
1197 return ret
? ret
: copied
;
1201 * Reserve a metadata for a single block located at lblock
1203 static int ext4_da_reserve_metadata(struct inode
*inode
, ext4_lblk_t lblock
)
1206 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1207 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1208 unsigned int md_needed
;
1209 ext4_lblk_t save_last_lblock
;
1213 * recalculate the amount of metadata blocks to reserve
1214 * in order to allocate nrblocks
1215 * worse case is one extent per block
1218 spin_lock(&ei
->i_block_reservation_lock
);
1220 * ext4_calc_metadata_amount() has side effects, which we have
1221 * to be prepared undo if we fail to claim space.
1223 save_len
= ei
->i_da_metadata_calc_len
;
1224 save_last_lblock
= ei
->i_da_metadata_calc_last_lblock
;
1225 md_needed
= EXT4_NUM_B2C(sbi
,
1226 ext4_calc_metadata_amount(inode
, lblock
));
1227 trace_ext4_da_reserve_space(inode
, md_needed
);
1230 * We do still charge estimated metadata to the sb though;
1231 * we cannot afford to run out of free blocks.
1233 if (ext4_claim_free_clusters(sbi
, md_needed
, 0)) {
1234 ei
->i_da_metadata_calc_len
= save_len
;
1235 ei
->i_da_metadata_calc_last_lblock
= save_last_lblock
;
1236 spin_unlock(&ei
->i_block_reservation_lock
);
1237 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1243 ei
->i_reserved_meta_blocks
+= md_needed
;
1244 spin_unlock(&ei
->i_block_reservation_lock
);
1246 return 0; /* success */
1250 * Reserve a single cluster located at lblock
1252 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1255 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1256 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1257 unsigned int md_needed
;
1259 ext4_lblk_t save_last_lblock
;
1263 * We will charge metadata quota at writeout time; this saves
1264 * us from metadata over-estimation, though we may go over by
1265 * a small amount in the end. Here we just reserve for data.
1267 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1272 * recalculate the amount of metadata blocks to reserve
1273 * in order to allocate nrblocks
1274 * worse case is one extent per block
1277 spin_lock(&ei
->i_block_reservation_lock
);
1279 * ext4_calc_metadata_amount() has side effects, which we have
1280 * to be prepared undo if we fail to claim space.
1282 save_len
= ei
->i_da_metadata_calc_len
;
1283 save_last_lblock
= ei
->i_da_metadata_calc_last_lblock
;
1284 md_needed
= EXT4_NUM_B2C(sbi
,
1285 ext4_calc_metadata_amount(inode
, lblock
));
1286 trace_ext4_da_reserve_space(inode
, md_needed
);
1289 * We do still charge estimated metadata to the sb though;
1290 * we cannot afford to run out of free blocks.
1292 if (ext4_claim_free_clusters(sbi
, md_needed
+ 1, 0)) {
1293 ei
->i_da_metadata_calc_len
= save_len
;
1294 ei
->i_da_metadata_calc_last_lblock
= save_last_lblock
;
1295 spin_unlock(&ei
->i_block_reservation_lock
);
1296 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1300 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1303 ei
->i_reserved_data_blocks
++;
1304 ei
->i_reserved_meta_blocks
+= md_needed
;
1305 spin_unlock(&ei
->i_block_reservation_lock
);
1307 return 0; /* success */
1310 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1312 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1313 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1316 return; /* Nothing to release, exit */
1318 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1320 trace_ext4_da_release_space(inode
, to_free
);
1321 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1323 * if there aren't enough reserved blocks, then the
1324 * counter is messed up somewhere. Since this
1325 * function is called from invalidate page, it's
1326 * harmless to return without any action.
1328 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1329 "ino %lu, to_free %d with only %d reserved "
1330 "data blocks", inode
->i_ino
, to_free
,
1331 ei
->i_reserved_data_blocks
);
1333 to_free
= ei
->i_reserved_data_blocks
;
1335 ei
->i_reserved_data_blocks
-= to_free
;
1337 if (ei
->i_reserved_data_blocks
== 0) {
1339 * We can release all of the reserved metadata blocks
1340 * only when we have written all of the delayed
1341 * allocation blocks.
1342 * Note that in case of bigalloc, i_reserved_meta_blocks,
1343 * i_reserved_data_blocks, etc. refer to number of clusters.
1345 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
1346 ei
->i_reserved_meta_blocks
);
1347 ei
->i_reserved_meta_blocks
= 0;
1348 ei
->i_da_metadata_calc_len
= 0;
1351 /* update fs dirty data blocks counter */
1352 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1354 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1356 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1359 static void ext4_da_page_release_reservation(struct page
*page
,
1360 unsigned int offset
,
1361 unsigned int length
)
1364 struct buffer_head
*head
, *bh
;
1365 unsigned int curr_off
= 0;
1366 struct inode
*inode
= page
->mapping
->host
;
1367 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1368 unsigned int stop
= offset
+ length
;
1372 BUG_ON(stop
> PAGE_CACHE_SIZE
|| stop
< length
);
1374 head
= page_buffers(page
);
1377 unsigned int next_off
= curr_off
+ bh
->b_size
;
1379 if (next_off
> stop
)
1382 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1384 clear_buffer_delay(bh
);
1386 curr_off
= next_off
;
1387 } while ((bh
= bh
->b_this_page
) != head
);
1390 lblk
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1391 ext4_es_remove_extent(inode
, lblk
, to_release
);
1394 /* If we have released all the blocks belonging to a cluster, then we
1395 * need to release the reserved space for that cluster. */
1396 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1397 while (num_clusters
> 0) {
1398 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1399 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1400 if (sbi
->s_cluster_ratio
== 1 ||
1401 !ext4_find_delalloc_cluster(inode
, lblk
))
1402 ext4_da_release_space(inode
, 1);
1409 * Delayed allocation stuff
1412 struct mpage_da_data
{
1413 struct inode
*inode
;
1414 struct writeback_control
*wbc
;
1416 pgoff_t first_page
; /* The first page to write */
1417 pgoff_t next_page
; /* Current page to examine */
1418 pgoff_t last_page
; /* Last page to examine */
1420 * Extent to map - this can be after first_page because that can be
1421 * fully mapped. We somewhat abuse m_flags to store whether the extent
1422 * is delalloc or unwritten.
1424 struct ext4_map_blocks map
;
1425 struct ext4_io_submit io_submit
; /* IO submission data */
1428 static void mpage_release_unused_pages(struct mpage_da_data
*mpd
,
1433 struct pagevec pvec
;
1434 struct inode
*inode
= mpd
->inode
;
1435 struct address_space
*mapping
= inode
->i_mapping
;
1437 /* This is necessary when next_page == 0. */
1438 if (mpd
->first_page
>= mpd
->next_page
)
1441 index
= mpd
->first_page
;
1442 end
= mpd
->next_page
- 1;
1444 ext4_lblk_t start
, last
;
1445 start
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1446 last
= end
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1447 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1450 pagevec_init(&pvec
, 0);
1451 while (index
<= end
) {
1452 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1455 for (i
= 0; i
< nr_pages
; i
++) {
1456 struct page
*page
= pvec
.pages
[i
];
1457 if (page
->index
> end
)
1459 BUG_ON(!PageLocked(page
));
1460 BUG_ON(PageWriteback(page
));
1462 block_invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
1463 ClearPageUptodate(page
);
1467 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1468 pagevec_release(&pvec
);
1472 static void ext4_print_free_blocks(struct inode
*inode
)
1474 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1475 struct super_block
*sb
= inode
->i_sb
;
1476 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1478 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1479 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1480 ext4_count_free_clusters(sb
)));
1481 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1482 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1483 (long long) EXT4_C2B(EXT4_SB(sb
),
1484 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1485 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1486 (long long) EXT4_C2B(EXT4_SB(sb
),
1487 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1488 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1489 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1490 ei
->i_reserved_data_blocks
);
1491 ext4_msg(sb
, KERN_CRIT
, "i_reserved_meta_blocks=%u",
1492 ei
->i_reserved_meta_blocks
);
1493 ext4_msg(sb
, KERN_CRIT
, "i_allocated_meta_blocks=%u",
1494 ei
->i_allocated_meta_blocks
);
1498 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1500 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1504 * This function is grabs code from the very beginning of
1505 * ext4_map_blocks, but assumes that the caller is from delayed write
1506 * time. This function looks up the requested blocks and sets the
1507 * buffer delay bit under the protection of i_data_sem.
1509 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1510 struct ext4_map_blocks
*map
,
1511 struct buffer_head
*bh
)
1513 struct extent_status es
;
1515 sector_t invalid_block
= ~((sector_t
) 0xffff);
1516 #ifdef ES_AGGRESSIVE_TEST
1517 struct ext4_map_blocks orig_map
;
1519 memcpy(&orig_map
, map
, sizeof(*map
));
1522 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1526 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1527 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1528 (unsigned long) map
->m_lblk
);
1530 /* Lookup extent status tree firstly */
1531 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1532 ext4_es_lru_add(inode
);
1533 if (ext4_es_is_hole(&es
)) {
1535 down_read((&EXT4_I(inode
)->i_data_sem
));
1540 * Delayed extent could be allocated by fallocate.
1541 * So we need to check it.
1543 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1544 map_bh(bh
, inode
->i_sb
, invalid_block
);
1546 set_buffer_delay(bh
);
1550 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1551 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1552 if (retval
> map
->m_len
)
1553 retval
= map
->m_len
;
1554 map
->m_len
= retval
;
1555 if (ext4_es_is_written(&es
))
1556 map
->m_flags
|= EXT4_MAP_MAPPED
;
1557 else if (ext4_es_is_unwritten(&es
))
1558 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1562 #ifdef ES_AGGRESSIVE_TEST
1563 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1569 * Try to see if we can get the block without requesting a new
1570 * file system block.
1572 down_read((&EXT4_I(inode
)->i_data_sem
));
1573 if (ext4_has_inline_data(inode
)) {
1575 * We will soon create blocks for this page, and let
1576 * us pretend as if the blocks aren't allocated yet.
1577 * In case of clusters, we have to handle the work
1578 * of mapping from cluster so that the reserved space
1579 * is calculated properly.
1581 if ((EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) &&
1582 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
1583 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
1585 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1586 retval
= ext4_ext_map_blocks(NULL
, inode
, map
,
1587 EXT4_GET_BLOCKS_NO_PUT_HOLE
);
1589 retval
= ext4_ind_map_blocks(NULL
, inode
, map
,
1590 EXT4_GET_BLOCKS_NO_PUT_HOLE
);
1596 * XXX: __block_prepare_write() unmaps passed block,
1600 * If the block was allocated from previously allocated cluster,
1601 * then we don't need to reserve it again. However we still need
1602 * to reserve metadata for every block we're going to write.
1604 if (!(map
->m_flags
& EXT4_MAP_FROM_CLUSTER
)) {
1605 ret
= ext4_da_reserve_space(inode
, iblock
);
1607 /* not enough space to reserve */
1612 ret
= ext4_da_reserve_metadata(inode
, iblock
);
1614 /* not enough space to reserve */
1620 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1621 ~0, EXTENT_STATUS_DELAYED
);
1627 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1628 * and it should not appear on the bh->b_state.
1630 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
1632 map_bh(bh
, inode
->i_sb
, invalid_block
);
1634 set_buffer_delay(bh
);
1635 } else if (retval
> 0) {
1637 unsigned int status
;
1639 if (unlikely(retval
!= map
->m_len
)) {
1640 ext4_warning(inode
->i_sb
,
1641 "ES len assertion failed for inode "
1642 "%lu: retval %d != map->m_len %d",
1643 inode
->i_ino
, retval
, map
->m_len
);
1647 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
1648 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
1649 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1650 map
->m_pblk
, status
);
1656 up_read((&EXT4_I(inode
)->i_data_sem
));
1662 * This is a special get_blocks_t callback which is used by
1663 * ext4_da_write_begin(). It will either return mapped block or
1664 * reserve space for a single block.
1666 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1667 * We also have b_blocknr = -1 and b_bdev initialized properly
1669 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1670 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1671 * initialized properly.
1673 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1674 struct buffer_head
*bh
, int create
)
1676 struct ext4_map_blocks map
;
1679 BUG_ON(create
== 0);
1680 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1682 map
.m_lblk
= iblock
;
1686 * first, we need to know whether the block is allocated already
1687 * preallocated blocks are unmapped but should treated
1688 * the same as allocated blocks.
1690 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1694 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1695 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1697 if (buffer_unwritten(bh
)) {
1698 /* A delayed write to unwritten bh should be marked
1699 * new and mapped. Mapped ensures that we don't do
1700 * get_block multiple times when we write to the same
1701 * offset and new ensures that we do proper zero out
1702 * for partial write.
1705 set_buffer_mapped(bh
);
1710 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1716 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1722 static int __ext4_journalled_writepage(struct page
*page
,
1725 struct address_space
*mapping
= page
->mapping
;
1726 struct inode
*inode
= mapping
->host
;
1727 struct buffer_head
*page_bufs
= NULL
;
1728 handle_t
*handle
= NULL
;
1729 int ret
= 0, err
= 0;
1730 int inline_data
= ext4_has_inline_data(inode
);
1731 struct buffer_head
*inode_bh
= NULL
;
1733 ClearPageChecked(page
);
1736 BUG_ON(page
->index
!= 0);
1737 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1738 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1739 if (inode_bh
== NULL
)
1742 page_bufs
= page_buffers(page
);
1747 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1750 /* As soon as we unlock the page, it can go away, but we have
1751 * references to buffers so we are safe */
1754 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
1755 ext4_writepage_trans_blocks(inode
));
1756 if (IS_ERR(handle
)) {
1757 ret
= PTR_ERR(handle
);
1761 BUG_ON(!ext4_handle_valid(handle
));
1764 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1766 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1769 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1770 do_journal_get_write_access
);
1772 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1777 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1778 err
= ext4_journal_stop(handle
);
1782 if (!ext4_has_inline_data(inode
))
1783 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1785 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1792 * Note that we don't need to start a transaction unless we're journaling data
1793 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1794 * need to file the inode to the transaction's list in ordered mode because if
1795 * we are writing back data added by write(), the inode is already there and if
1796 * we are writing back data modified via mmap(), no one guarantees in which
1797 * transaction the data will hit the disk. In case we are journaling data, we
1798 * cannot start transaction directly because transaction start ranks above page
1799 * lock so we have to do some magic.
1801 * This function can get called via...
1802 * - ext4_writepages after taking page lock (have journal handle)
1803 * - journal_submit_inode_data_buffers (no journal handle)
1804 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1805 * - grab_page_cache when doing write_begin (have journal handle)
1807 * We don't do any block allocation in this function. If we have page with
1808 * multiple blocks we need to write those buffer_heads that are mapped. This
1809 * is important for mmaped based write. So if we do with blocksize 1K
1810 * truncate(f, 1024);
1811 * a = mmap(f, 0, 4096);
1813 * truncate(f, 4096);
1814 * we have in the page first buffer_head mapped via page_mkwrite call back
1815 * but other buffer_heads would be unmapped but dirty (dirty done via the
1816 * do_wp_page). So writepage should write the first block. If we modify
1817 * the mmap area beyond 1024 we will again get a page_fault and the
1818 * page_mkwrite callback will do the block allocation and mark the
1819 * buffer_heads mapped.
1821 * We redirty the page if we have any buffer_heads that is either delay or
1822 * unwritten in the page.
1824 * We can get recursively called as show below.
1826 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1829 * But since we don't do any block allocation we should not deadlock.
1830 * Page also have the dirty flag cleared so we don't get recurive page_lock.
1832 static int ext4_writepage(struct page
*page
,
1833 struct writeback_control
*wbc
)
1838 struct buffer_head
*page_bufs
= NULL
;
1839 struct inode
*inode
= page
->mapping
->host
;
1840 struct ext4_io_submit io_submit
;
1842 trace_ext4_writepage(page
);
1843 size
= i_size_read(inode
);
1844 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1845 len
= size
& ~PAGE_CACHE_MASK
;
1847 len
= PAGE_CACHE_SIZE
;
1849 page_bufs
= page_buffers(page
);
1851 * We cannot do block allocation or other extent handling in this
1852 * function. If there are buffers needing that, we have to redirty
1853 * the page. But we may reach here when we do a journal commit via
1854 * journal_submit_inode_data_buffers() and in that case we must write
1855 * allocated buffers to achieve data=ordered mode guarantees.
1857 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
1858 ext4_bh_delay_or_unwritten
)) {
1859 redirty_page_for_writepage(wbc
, page
);
1860 if (current
->flags
& PF_MEMALLOC
) {
1862 * For memory cleaning there's no point in writing only
1863 * some buffers. So just bail out. Warn if we came here
1864 * from direct reclaim.
1866 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
1873 if (PageChecked(page
) && ext4_should_journal_data(inode
))
1875 * It's mmapped pagecache. Add buffers and journal it. There
1876 * doesn't seem much point in redirtying the page here.
1878 return __ext4_journalled_writepage(page
, len
);
1880 ext4_io_submit_init(&io_submit
, wbc
);
1881 io_submit
.io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
1882 if (!io_submit
.io_end
) {
1883 redirty_page_for_writepage(wbc
, page
);
1887 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
);
1888 ext4_io_submit(&io_submit
);
1889 /* Drop io_end reference we got from init */
1890 ext4_put_io_end_defer(io_submit
.io_end
);
1894 static int mpage_submit_page(struct mpage_da_data
*mpd
, struct page
*page
)
1897 loff_t size
= i_size_read(mpd
->inode
);
1900 BUG_ON(page
->index
!= mpd
->first_page
);
1901 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
1902 len
= size
& ~PAGE_CACHE_MASK
;
1904 len
= PAGE_CACHE_SIZE
;
1905 clear_page_dirty_for_io(page
);
1906 err
= ext4_bio_write_page(&mpd
->io_submit
, page
, len
, mpd
->wbc
);
1908 mpd
->wbc
->nr_to_write
--;
1914 #define BH_FLAGS ((1 << BH_Unwritten) | (1 << BH_Delay))
1917 * mballoc gives us at most this number of blocks...
1918 * XXX: That seems to be only a limitation of ext4_mb_normalize_request().
1919 * The rest of mballoc seems to handle chunks up to full group size.
1921 #define MAX_WRITEPAGES_EXTENT_LEN 2048
1924 * mpage_add_bh_to_extent - try to add bh to extent of blocks to map
1926 * @mpd - extent of blocks
1927 * @lblk - logical number of the block in the file
1928 * @bh - buffer head we want to add to the extent
1930 * The function is used to collect contig. blocks in the same state. If the
1931 * buffer doesn't require mapping for writeback and we haven't started the
1932 * extent of buffers to map yet, the function returns 'true' immediately - the
1933 * caller can write the buffer right away. Otherwise the function returns true
1934 * if the block has been added to the extent, false if the block couldn't be
1937 static bool mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, ext4_lblk_t lblk
,
1938 struct buffer_head
*bh
)
1940 struct ext4_map_blocks
*map
= &mpd
->map
;
1942 /* Buffer that doesn't need mapping for writeback? */
1943 if (!buffer_dirty(bh
) || !buffer_mapped(bh
) ||
1944 (!buffer_delay(bh
) && !buffer_unwritten(bh
))) {
1945 /* So far no extent to map => we write the buffer right away */
1946 if (map
->m_len
== 0)
1951 /* First block in the extent? */
1952 if (map
->m_len
== 0) {
1955 map
->m_flags
= bh
->b_state
& BH_FLAGS
;
1959 /* Don't go larger than mballoc is willing to allocate */
1960 if (map
->m_len
>= MAX_WRITEPAGES_EXTENT_LEN
)
1963 /* Can we merge the block to our big extent? */
1964 if (lblk
== map
->m_lblk
+ map
->m_len
&&
1965 (bh
->b_state
& BH_FLAGS
) == map
->m_flags
) {
1973 * mpage_process_page_bufs - submit page buffers for IO or add them to extent
1975 * @mpd - extent of blocks for mapping
1976 * @head - the first buffer in the page
1977 * @bh - buffer we should start processing from
1978 * @lblk - logical number of the block in the file corresponding to @bh
1980 * Walk through page buffers from @bh upto @head (exclusive) and either submit
1981 * the page for IO if all buffers in this page were mapped and there's no
1982 * accumulated extent of buffers to map or add buffers in the page to the
1983 * extent of buffers to map. The function returns 1 if the caller can continue
1984 * by processing the next page, 0 if it should stop adding buffers to the
1985 * extent to map because we cannot extend it anymore. It can also return value
1986 * < 0 in case of error during IO submission.
1988 static int mpage_process_page_bufs(struct mpage_da_data
*mpd
,
1989 struct buffer_head
*head
,
1990 struct buffer_head
*bh
,
1993 struct inode
*inode
= mpd
->inode
;
1995 ext4_lblk_t blocks
= (i_size_read(inode
) + (1 << inode
->i_blkbits
) - 1)
1996 >> inode
->i_blkbits
;
1999 BUG_ON(buffer_locked(bh
));
2001 if (lblk
>= blocks
|| !mpage_add_bh_to_extent(mpd
, lblk
, bh
)) {
2002 /* Found extent to map? */
2005 /* Everything mapped so far and we hit EOF */
2008 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2009 /* So far everything mapped? Submit the page for IO. */
2010 if (mpd
->map
.m_len
== 0) {
2011 err
= mpage_submit_page(mpd
, head
->b_page
);
2015 return lblk
< blocks
;
2019 * mpage_map_buffers - update buffers corresponding to changed extent and
2020 * submit fully mapped pages for IO
2022 * @mpd - description of extent to map, on return next extent to map
2024 * Scan buffers corresponding to changed extent (we expect corresponding pages
2025 * to be already locked) and update buffer state according to new extent state.
2026 * We map delalloc buffers to their physical location, clear unwritten bits,
2027 * and mark buffers as uninit when we perform writes to uninitialized extents
2028 * and do extent conversion after IO is finished. If the last page is not fully
2029 * mapped, we update @map to the next extent in the last page that needs
2030 * mapping. Otherwise we submit the page for IO.
2032 static int mpage_map_and_submit_buffers(struct mpage_da_data
*mpd
)
2034 struct pagevec pvec
;
2036 struct inode
*inode
= mpd
->inode
;
2037 struct buffer_head
*head
, *bh
;
2038 int bpp_bits
= PAGE_CACHE_SHIFT
- inode
->i_blkbits
;
2044 start
= mpd
->map
.m_lblk
>> bpp_bits
;
2045 end
= (mpd
->map
.m_lblk
+ mpd
->map
.m_len
- 1) >> bpp_bits
;
2046 lblk
= start
<< bpp_bits
;
2047 pblock
= mpd
->map
.m_pblk
;
2049 pagevec_init(&pvec
, 0);
2050 while (start
<= end
) {
2051 nr_pages
= pagevec_lookup(&pvec
, inode
->i_mapping
, start
,
2055 for (i
= 0; i
< nr_pages
; i
++) {
2056 struct page
*page
= pvec
.pages
[i
];
2058 if (page
->index
> end
)
2060 /* Up to 'end' pages must be contiguous */
2061 BUG_ON(page
->index
!= start
);
2062 bh
= head
= page_buffers(page
);
2064 if (lblk
< mpd
->map
.m_lblk
)
2066 if (lblk
>= mpd
->map
.m_lblk
+ mpd
->map
.m_len
) {
2068 * Buffer after end of mapped extent.
2069 * Find next buffer in the page to map.
2072 mpd
->map
.m_flags
= 0;
2074 * FIXME: If dioread_nolock supports
2075 * blocksize < pagesize, we need to make
2076 * sure we add size mapped so far to
2077 * io_end->size as the following call
2078 * can submit the page for IO.
2080 err
= mpage_process_page_bufs(mpd
, head
,
2082 pagevec_release(&pvec
);
2087 if (buffer_delay(bh
)) {
2088 clear_buffer_delay(bh
);
2089 bh
->b_blocknr
= pblock
++;
2091 clear_buffer_unwritten(bh
);
2092 } while (lblk
++, (bh
= bh
->b_this_page
) != head
);
2095 * FIXME: This is going to break if dioread_nolock
2096 * supports blocksize < pagesize as we will try to
2097 * convert potentially unmapped parts of inode.
2099 mpd
->io_submit
.io_end
->size
+= PAGE_CACHE_SIZE
;
2100 /* Page fully mapped - let IO run! */
2101 err
= mpage_submit_page(mpd
, page
);
2103 pagevec_release(&pvec
);
2108 pagevec_release(&pvec
);
2110 /* Extent fully mapped and matches with page boundary. We are done. */
2112 mpd
->map
.m_flags
= 0;
2116 static int mpage_map_one_extent(handle_t
*handle
, struct mpage_da_data
*mpd
)
2118 struct inode
*inode
= mpd
->inode
;
2119 struct ext4_map_blocks
*map
= &mpd
->map
;
2120 int get_blocks_flags
;
2123 trace_ext4_da_write_pages_extent(inode
, map
);
2125 * Call ext4_map_blocks() to allocate any delayed allocation blocks, or
2126 * to convert an uninitialized extent to be initialized (in the case
2127 * where we have written into one or more preallocated blocks). It is
2128 * possible that we're going to need more metadata blocks than
2129 * previously reserved. However we must not fail because we're in
2130 * writeback and there is nothing we can do about it so it might result
2131 * in data loss. So use reserved blocks to allocate metadata if
2134 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE if the blocks
2135 * in question are delalloc blocks. This affects functions in many
2136 * different parts of the allocation call path. This flag exists
2137 * primarily because we don't want to change *many* call functions, so
2138 * ext4_map_blocks() will set the EXT4_STATE_DELALLOC_RESERVED flag
2139 * once the inode's allocation semaphore is taken.
2141 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
2142 EXT4_GET_BLOCKS_METADATA_NOFAIL
;
2143 if (ext4_should_dioread_nolock(inode
))
2144 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
2145 if (map
->m_flags
& (1 << BH_Delay
))
2146 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
2148 err
= ext4_map_blocks(handle
, inode
, map
, get_blocks_flags
);
2151 if (map
->m_flags
& EXT4_MAP_UNINIT
) {
2152 if (!mpd
->io_submit
.io_end
->handle
&&
2153 ext4_handle_valid(handle
)) {
2154 mpd
->io_submit
.io_end
->handle
= handle
->h_rsv_handle
;
2155 handle
->h_rsv_handle
= NULL
;
2157 ext4_set_io_unwritten_flag(inode
, mpd
->io_submit
.io_end
);
2160 BUG_ON(map
->m_len
== 0);
2161 if (map
->m_flags
& EXT4_MAP_NEW
) {
2162 struct block_device
*bdev
= inode
->i_sb
->s_bdev
;
2165 for (i
= 0; i
< map
->m_len
; i
++)
2166 unmap_underlying_metadata(bdev
, map
->m_pblk
+ i
);
2172 * mpage_map_and_submit_extent - map extent starting at mpd->lblk of length
2173 * mpd->len and submit pages underlying it for IO
2175 * @handle - handle for journal operations
2176 * @mpd - extent to map
2178 * The function maps extent starting at mpd->lblk of length mpd->len. If it is
2179 * delayed, blocks are allocated, if it is unwritten, we may need to convert
2180 * them to initialized or split the described range from larger unwritten
2181 * extent. Note that we need not map all the described range since allocation
2182 * can return less blocks or the range is covered by more unwritten extents. We
2183 * cannot map more because we are limited by reserved transaction credits. On
2184 * the other hand we always make sure that the last touched page is fully
2185 * mapped so that it can be written out (and thus forward progress is
2186 * guaranteed). After mapping we submit all mapped pages for IO.
2188 static int mpage_map_and_submit_extent(handle_t
*handle
,
2189 struct mpage_da_data
*mpd
,
2190 bool *give_up_on_write
)
2192 struct inode
*inode
= mpd
->inode
;
2193 struct ext4_map_blocks
*map
= &mpd
->map
;
2197 mpd
->io_submit
.io_end
->offset
=
2198 ((loff_t
)map
->m_lblk
) << inode
->i_blkbits
;
2200 err
= mpage_map_one_extent(handle
, mpd
);
2202 struct super_block
*sb
= inode
->i_sb
;
2204 if (EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)
2205 goto invalidate_dirty_pages
;
2207 * Let the uper layers retry transient errors.
2208 * In the case of ENOSPC, if ext4_count_free_blocks()
2209 * is non-zero, a commit should free up blocks.
2211 if ((err
== -ENOMEM
) ||
2212 (err
== -ENOSPC
&& ext4_count_free_clusters(sb
)))
2214 ext4_msg(sb
, KERN_CRIT
,
2215 "Delayed block allocation failed for "
2216 "inode %lu at logical offset %llu with"
2217 " max blocks %u with error %d",
2219 (unsigned long long)map
->m_lblk
,
2220 (unsigned)map
->m_len
, -err
);
2221 ext4_msg(sb
, KERN_CRIT
,
2222 "This should not happen!! Data will "
2225 ext4_print_free_blocks(inode
);
2226 invalidate_dirty_pages
:
2227 *give_up_on_write
= true;
2231 * Update buffer state, submit mapped pages, and get us new
2234 err
= mpage_map_and_submit_buffers(mpd
);
2237 } while (map
->m_len
);
2239 /* Update on-disk size after IO is submitted */
2240 disksize
= ((loff_t
)mpd
->first_page
) << PAGE_CACHE_SHIFT
;
2241 if (disksize
> EXT4_I(inode
)->i_disksize
) {
2244 ext4_wb_update_i_disksize(inode
, disksize
);
2245 err2
= ext4_mark_inode_dirty(handle
, inode
);
2247 ext4_error(inode
->i_sb
,
2248 "Failed to mark inode %lu dirty",
2257 * Calculate the total number of credits to reserve for one writepages
2258 * iteration. This is called from ext4_writepages(). We map an extent of
2259 * up to MAX_WRITEPAGES_EXTENT_LEN blocks and then we go on and finish mapping
2260 * the last partial page. So in total we can map MAX_WRITEPAGES_EXTENT_LEN +
2261 * bpp - 1 blocks in bpp different extents.
2263 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2265 int bpp
= ext4_journal_blocks_per_page(inode
);
2267 return ext4_meta_trans_blocks(inode
,
2268 MAX_WRITEPAGES_EXTENT_LEN
+ bpp
- 1, bpp
);
2272 * mpage_prepare_extent_to_map - find & lock contiguous range of dirty pages
2273 * and underlying extent to map
2275 * @mpd - where to look for pages
2277 * Walk dirty pages in the mapping. If they are fully mapped, submit them for
2278 * IO immediately. When we find a page which isn't mapped we start accumulating
2279 * extent of buffers underlying these pages that needs mapping (formed by
2280 * either delayed or unwritten buffers). We also lock the pages containing
2281 * these buffers. The extent found is returned in @mpd structure (starting at
2282 * mpd->lblk with length mpd->len blocks).
2284 * Note that this function can attach bios to one io_end structure which are
2285 * neither logically nor physically contiguous. Although it may seem as an
2286 * unnecessary complication, it is actually inevitable in blocksize < pagesize
2287 * case as we need to track IO to all buffers underlying a page in one io_end.
2289 static int mpage_prepare_extent_to_map(struct mpage_da_data
*mpd
)
2291 struct address_space
*mapping
= mpd
->inode
->i_mapping
;
2292 struct pagevec pvec
;
2293 unsigned int nr_pages
;
2294 pgoff_t index
= mpd
->first_page
;
2295 pgoff_t end
= mpd
->last_page
;
2298 int blkbits
= mpd
->inode
->i_blkbits
;
2300 struct buffer_head
*head
;
2302 if (mpd
->wbc
->sync_mode
== WB_SYNC_ALL
|| mpd
->wbc
->tagged_writepages
)
2303 tag
= PAGECACHE_TAG_TOWRITE
;
2305 tag
= PAGECACHE_TAG_DIRTY
;
2307 pagevec_init(&pvec
, 0);
2309 mpd
->next_page
= index
;
2310 while (index
<= end
) {
2311 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2312 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2316 for (i
= 0; i
< nr_pages
; i
++) {
2317 struct page
*page
= pvec
.pages
[i
];
2320 * At this point, the page may be truncated or
2321 * invalidated (changing page->mapping to NULL), or
2322 * even swizzled back from swapper_space to tmpfs file
2323 * mapping. However, page->index will not change
2324 * because we have a reference on the page.
2326 if (page
->index
> end
)
2329 /* If we can't merge this page, we are done. */
2330 if (mpd
->map
.m_len
> 0 && mpd
->next_page
!= page
->index
)
2335 * If the page is no longer dirty, or its mapping no
2336 * longer corresponds to inode we are writing (which
2337 * means it has been truncated or invalidated), or the
2338 * page is already under writeback and we are not doing
2339 * a data integrity writeback, skip the page
2341 if (!PageDirty(page
) ||
2342 (PageWriteback(page
) &&
2343 (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2344 unlikely(page
->mapping
!= mapping
)) {
2349 wait_on_page_writeback(page
);
2350 BUG_ON(PageWriteback(page
));
2352 if (mpd
->map
.m_len
== 0)
2353 mpd
->first_page
= page
->index
;
2354 mpd
->next_page
= page
->index
+ 1;
2355 /* Add all dirty buffers to mpd */
2356 lblk
= ((ext4_lblk_t
)page
->index
) <<
2357 (PAGE_CACHE_SHIFT
- blkbits
);
2358 head
= page_buffers(page
);
2359 err
= mpage_process_page_bufs(mpd
, head
, head
, lblk
);
2365 * Accumulated enough dirty pages? This doesn't apply
2366 * to WB_SYNC_ALL mode. For integrity sync we have to
2367 * keep going because someone may be concurrently
2368 * dirtying pages, and we might have synced a lot of
2369 * newly appeared dirty pages, but have not synced all
2370 * of the old dirty pages.
2372 if (mpd
->wbc
->sync_mode
== WB_SYNC_NONE
&&
2373 mpd
->next_page
- mpd
->first_page
>=
2374 mpd
->wbc
->nr_to_write
)
2377 pagevec_release(&pvec
);
2382 pagevec_release(&pvec
);
2386 static int __writepage(struct page
*page
, struct writeback_control
*wbc
,
2389 struct address_space
*mapping
= data
;
2390 int ret
= ext4_writepage(page
, wbc
);
2391 mapping_set_error(mapping
, ret
);
2395 static int ext4_writepages(struct address_space
*mapping
,
2396 struct writeback_control
*wbc
)
2398 pgoff_t writeback_index
= 0;
2399 long nr_to_write
= wbc
->nr_to_write
;
2400 int range_whole
= 0;
2402 handle_t
*handle
= NULL
;
2403 struct mpage_da_data mpd
;
2404 struct inode
*inode
= mapping
->host
;
2405 int needed_blocks
, rsv_blocks
= 0, ret
= 0;
2406 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2408 struct blk_plug plug
;
2409 bool give_up_on_write
= false;
2411 trace_ext4_writepages(inode
, wbc
);
2414 * No pages to write? This is mainly a kludge to avoid starting
2415 * a transaction for special inodes like journal inode on last iput()
2416 * because that could violate lock ordering on umount
2418 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2421 if (ext4_should_journal_data(inode
)) {
2422 struct blk_plug plug
;
2425 blk_start_plug(&plug
);
2426 ret
= write_cache_pages(mapping
, wbc
, __writepage
, mapping
);
2427 blk_finish_plug(&plug
);
2432 * If the filesystem has aborted, it is read-only, so return
2433 * right away instead of dumping stack traces later on that
2434 * will obscure the real source of the problem. We test
2435 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2436 * the latter could be true if the filesystem is mounted
2437 * read-only, and in that case, ext4_writepages should
2438 * *never* be called, so if that ever happens, we would want
2441 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2444 if (ext4_should_dioread_nolock(inode
)) {
2446 * We may need to convert up to one extent per block in
2447 * the page and we may dirty the inode.
2449 rsv_blocks
= 1 + (PAGE_CACHE_SIZE
>> inode
->i_blkbits
);
2453 * If we have inline data and arrive here, it means that
2454 * we will soon create the block for the 1st page, so
2455 * we'd better clear the inline data here.
2457 if (ext4_has_inline_data(inode
)) {
2458 /* Just inode will be modified... */
2459 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
2460 if (IS_ERR(handle
)) {
2461 ret
= PTR_ERR(handle
);
2462 goto out_writepages
;
2464 BUG_ON(ext4_test_inode_state(inode
,
2465 EXT4_STATE_MAY_INLINE_DATA
));
2466 ext4_destroy_inline_data(handle
, inode
);
2467 ext4_journal_stop(handle
);
2470 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2473 if (wbc
->range_cyclic
) {
2474 writeback_index
= mapping
->writeback_index
;
2475 if (writeback_index
)
2477 mpd
.first_page
= writeback_index
;
2480 mpd
.first_page
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2481 mpd
.last_page
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2486 ext4_io_submit_init(&mpd
.io_submit
, wbc
);
2488 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2489 tag_pages_for_writeback(mapping
, mpd
.first_page
, mpd
.last_page
);
2491 blk_start_plug(&plug
);
2492 while (!done
&& mpd
.first_page
<= mpd
.last_page
) {
2493 /* For each extent of pages we use new io_end */
2494 mpd
.io_submit
.io_end
= ext4_init_io_end(inode
, GFP_KERNEL
);
2495 if (!mpd
.io_submit
.io_end
) {
2501 * We have two constraints: We find one extent to map and we
2502 * must always write out whole page (makes a difference when
2503 * blocksize < pagesize) so that we don't block on IO when we
2504 * try to write out the rest of the page. Journalled mode is
2505 * not supported by delalloc.
2507 BUG_ON(ext4_should_journal_data(inode
));
2508 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2510 /* start a new transaction */
2511 handle
= ext4_journal_start_with_reserve(inode
,
2512 EXT4_HT_WRITE_PAGE
, needed_blocks
, rsv_blocks
);
2513 if (IS_ERR(handle
)) {
2514 ret
= PTR_ERR(handle
);
2515 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2516 "%ld pages, ino %lu; err %d", __func__
,
2517 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2518 /* Release allocated io_end */
2519 ext4_put_io_end(mpd
.io_submit
.io_end
);
2523 trace_ext4_da_write_pages(inode
, mpd
.first_page
, mpd
.wbc
);
2524 ret
= mpage_prepare_extent_to_map(&mpd
);
2527 ret
= mpage_map_and_submit_extent(handle
, &mpd
,
2531 * We scanned the whole range (or exhausted
2532 * nr_to_write), submitted what was mapped and
2533 * didn't find anything needing mapping. We are
2539 ext4_journal_stop(handle
);
2540 /* Submit prepared bio */
2541 ext4_io_submit(&mpd
.io_submit
);
2542 /* Unlock pages we didn't use */
2543 mpage_release_unused_pages(&mpd
, give_up_on_write
);
2544 /* Drop our io_end reference we got from init */
2545 ext4_put_io_end(mpd
.io_submit
.io_end
);
2547 if (ret
== -ENOSPC
&& sbi
->s_journal
) {
2549 * Commit the transaction which would
2550 * free blocks released in the transaction
2553 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2557 /* Fatal error - ENOMEM, EIO... */
2561 blk_finish_plug(&plug
);
2562 if (!ret
&& !cycled
) {
2564 mpd
.last_page
= writeback_index
- 1;
2570 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2572 * Set the writeback_index so that range_cyclic
2573 * mode will write it back later
2575 mapping
->writeback_index
= mpd
.first_page
;
2578 trace_ext4_writepages_result(inode
, wbc
, ret
,
2579 nr_to_write
- wbc
->nr_to_write
);
2583 static int ext4_nonda_switch(struct super_block
*sb
)
2585 s64 free_clusters
, dirty_clusters
;
2586 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2589 * switch to non delalloc mode if we are running low
2590 * on free block. The free block accounting via percpu
2591 * counters can get slightly wrong with percpu_counter_batch getting
2592 * accumulated on each CPU without updating global counters
2593 * Delalloc need an accurate free block accounting. So switch
2594 * to non delalloc when we are near to error range.
2597 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2599 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2601 * Start pushing delalloc when 1/2 of free blocks are dirty.
2603 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2604 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2606 if (2 * free_clusters
< 3 * dirty_clusters
||
2607 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2609 * free block count is less than 150% of dirty blocks
2610 * or free blocks is less than watermark
2617 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2618 loff_t pos
, unsigned len
, unsigned flags
,
2619 struct page
**pagep
, void **fsdata
)
2621 int ret
, retries
= 0;
2624 struct inode
*inode
= mapping
->host
;
2627 index
= pos
>> PAGE_CACHE_SHIFT
;
2629 if (ext4_nonda_switch(inode
->i_sb
)) {
2630 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2631 return ext4_write_begin(file
, mapping
, pos
,
2632 len
, flags
, pagep
, fsdata
);
2634 *fsdata
= (void *)0;
2635 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2637 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2638 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2648 * grab_cache_page_write_begin() can take a long time if the
2649 * system is thrashing due to memory pressure, or if the page
2650 * is being written back. So grab it first before we start
2651 * the transaction handle. This also allows us to allocate
2652 * the page (if needed) without using GFP_NOFS.
2655 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2661 * With delayed allocation, we don't log the i_disksize update
2662 * if there is delayed block allocation. But we still need
2663 * to journalling the i_disksize update if writes to the end
2664 * of file which has an already mapped buffer.
2667 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, 1);
2668 if (IS_ERR(handle
)) {
2669 page_cache_release(page
);
2670 return PTR_ERR(handle
);
2674 if (page
->mapping
!= mapping
) {
2675 /* The page got truncated from under us */
2677 page_cache_release(page
);
2678 ext4_journal_stop(handle
);
2681 /* In case writeback began while the page was unlocked */
2682 wait_for_stable_page(page
);
2684 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2687 ext4_journal_stop(handle
);
2689 * block_write_begin may have instantiated a few blocks
2690 * outside i_size. Trim these off again. Don't need
2691 * i_size_read because we hold i_mutex.
2693 if (pos
+ len
> inode
->i_size
)
2694 ext4_truncate_failed_write(inode
);
2696 if (ret
== -ENOSPC
&&
2697 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2700 page_cache_release(page
);
2709 * Check if we should update i_disksize
2710 * when write to the end of file but not require block allocation
2712 static int ext4_da_should_update_i_disksize(struct page
*page
,
2713 unsigned long offset
)
2715 struct buffer_head
*bh
;
2716 struct inode
*inode
= page
->mapping
->host
;
2720 bh
= page_buffers(page
);
2721 idx
= offset
>> inode
->i_blkbits
;
2723 for (i
= 0; i
< idx
; i
++)
2724 bh
= bh
->b_this_page
;
2726 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2731 static int ext4_da_write_end(struct file
*file
,
2732 struct address_space
*mapping
,
2733 loff_t pos
, unsigned len
, unsigned copied
,
2734 struct page
*page
, void *fsdata
)
2736 struct inode
*inode
= mapping
->host
;
2738 handle_t
*handle
= ext4_journal_current_handle();
2740 unsigned long start
, end
;
2741 int write_mode
= (int)(unsigned long)fsdata
;
2743 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
2744 return ext4_write_end(file
, mapping
, pos
,
2745 len
, copied
, page
, fsdata
);
2747 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2748 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2749 end
= start
+ copied
- 1;
2752 * generic_write_end() will run mark_inode_dirty() if i_size
2753 * changes. So let's piggyback the i_disksize mark_inode_dirty
2756 new_i_size
= pos
+ copied
;
2757 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2758 if (ext4_has_inline_data(inode
) ||
2759 ext4_da_should_update_i_disksize(page
, end
)) {
2760 down_write(&EXT4_I(inode
)->i_data_sem
);
2761 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
2762 EXT4_I(inode
)->i_disksize
= new_i_size
;
2763 up_write(&EXT4_I(inode
)->i_data_sem
);
2764 /* We need to mark inode dirty even if
2765 * new_i_size is less that inode->i_size
2766 * bu greater than i_disksize.(hint delalloc)
2768 ext4_mark_inode_dirty(handle
, inode
);
2772 if (write_mode
!= CONVERT_INLINE_DATA
&&
2773 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
2774 ext4_has_inline_data(inode
))
2775 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
2778 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2784 ret2
= ext4_journal_stop(handle
);
2788 return ret
? ret
: copied
;
2791 static void ext4_da_invalidatepage(struct page
*page
, unsigned int offset
,
2792 unsigned int length
)
2795 * Drop reserved blocks
2797 BUG_ON(!PageLocked(page
));
2798 if (!page_has_buffers(page
))
2801 ext4_da_page_release_reservation(page
, offset
, length
);
2804 ext4_invalidatepage(page
, offset
, length
);
2810 * Force all delayed allocation blocks to be allocated for a given inode.
2812 int ext4_alloc_da_blocks(struct inode
*inode
)
2814 trace_ext4_alloc_da_blocks(inode
);
2816 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2817 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2821 * We do something simple for now. The filemap_flush() will
2822 * also start triggering a write of the data blocks, which is
2823 * not strictly speaking necessary (and for users of
2824 * laptop_mode, not even desirable). However, to do otherwise
2825 * would require replicating code paths in:
2827 * ext4_writepages() ->
2828 * write_cache_pages() ---> (via passed in callback function)
2829 * __mpage_da_writepage() -->
2830 * mpage_add_bh_to_extent()
2831 * mpage_da_map_blocks()
2833 * The problem is that write_cache_pages(), located in
2834 * mm/page-writeback.c, marks pages clean in preparation for
2835 * doing I/O, which is not desirable if we're not planning on
2838 * We could call write_cache_pages(), and then redirty all of
2839 * the pages by calling redirty_page_for_writepage() but that
2840 * would be ugly in the extreme. So instead we would need to
2841 * replicate parts of the code in the above functions,
2842 * simplifying them because we wouldn't actually intend to
2843 * write out the pages, but rather only collect contiguous
2844 * logical block extents, call the multi-block allocator, and
2845 * then update the buffer heads with the block allocations.
2847 * For now, though, we'll cheat by calling filemap_flush(),
2848 * which will map the blocks, and start the I/O, but not
2849 * actually wait for the I/O to complete.
2851 return filemap_flush(inode
->i_mapping
);
2855 * bmap() is special. It gets used by applications such as lilo and by
2856 * the swapper to find the on-disk block of a specific piece of data.
2858 * Naturally, this is dangerous if the block concerned is still in the
2859 * journal. If somebody makes a swapfile on an ext4 data-journaling
2860 * filesystem and enables swap, then they may get a nasty shock when the
2861 * data getting swapped to that swapfile suddenly gets overwritten by
2862 * the original zero's written out previously to the journal and
2863 * awaiting writeback in the kernel's buffer cache.
2865 * So, if we see any bmap calls here on a modified, data-journaled file,
2866 * take extra steps to flush any blocks which might be in the cache.
2868 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2870 struct inode
*inode
= mapping
->host
;
2875 * We can get here for an inline file via the FIBMAP ioctl
2877 if (ext4_has_inline_data(inode
))
2880 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2881 test_opt(inode
->i_sb
, DELALLOC
)) {
2883 * With delalloc we want to sync the file
2884 * so that we can make sure we allocate
2887 filemap_write_and_wait(mapping
);
2890 if (EXT4_JOURNAL(inode
) &&
2891 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2893 * This is a REALLY heavyweight approach, but the use of
2894 * bmap on dirty files is expected to be extremely rare:
2895 * only if we run lilo or swapon on a freshly made file
2896 * do we expect this to happen.
2898 * (bmap requires CAP_SYS_RAWIO so this does not
2899 * represent an unprivileged user DOS attack --- we'd be
2900 * in trouble if mortal users could trigger this path at
2903 * NB. EXT4_STATE_JDATA is not set on files other than
2904 * regular files. If somebody wants to bmap a directory
2905 * or symlink and gets confused because the buffer
2906 * hasn't yet been flushed to disk, they deserve
2907 * everything they get.
2910 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2911 journal
= EXT4_JOURNAL(inode
);
2912 jbd2_journal_lock_updates(journal
);
2913 err
= jbd2_journal_flush(journal
);
2914 jbd2_journal_unlock_updates(journal
);
2920 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2923 static int ext4_readpage(struct file
*file
, struct page
*page
)
2926 struct inode
*inode
= page
->mapping
->host
;
2928 trace_ext4_readpage(page
);
2930 if (ext4_has_inline_data(inode
))
2931 ret
= ext4_readpage_inline(inode
, page
);
2934 return mpage_readpage(page
, ext4_get_block
);
2940 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2941 struct list_head
*pages
, unsigned nr_pages
)
2943 struct inode
*inode
= mapping
->host
;
2945 /* If the file has inline data, no need to do readpages. */
2946 if (ext4_has_inline_data(inode
))
2949 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2952 static void ext4_invalidatepage(struct page
*page
, unsigned int offset
,
2953 unsigned int length
)
2955 trace_ext4_invalidatepage(page
, offset
, length
);
2957 /* No journalling happens on data buffers when this function is used */
2958 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
2960 block_invalidatepage(page
, offset
, length
);
2963 static int __ext4_journalled_invalidatepage(struct page
*page
,
2964 unsigned int offset
,
2965 unsigned int length
)
2967 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2969 trace_ext4_journalled_invalidatepage(page
, offset
, length
);
2972 * If it's a full truncate we just forget about the pending dirtying
2974 if (offset
== 0 && length
== PAGE_CACHE_SIZE
)
2975 ClearPageChecked(page
);
2977 return jbd2_journal_invalidatepage(journal
, page
, offset
, length
);
2980 /* Wrapper for aops... */
2981 static void ext4_journalled_invalidatepage(struct page
*page
,
2982 unsigned int offset
,
2983 unsigned int length
)
2985 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
, length
) < 0);
2988 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2990 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2992 trace_ext4_releasepage(page
);
2994 /* Page has dirty journalled data -> cannot release */
2995 if (PageChecked(page
))
2998 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3000 return try_to_free_buffers(page
);
3004 * ext4_get_block used when preparing for a DIO write or buffer write.
3005 * We allocate an uinitialized extent if blocks haven't been allocated.
3006 * The extent will be converted to initialized after the IO is complete.
3008 int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
3009 struct buffer_head
*bh_result
, int create
)
3011 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3012 inode
->i_ino
, create
);
3013 return _ext4_get_block(inode
, iblock
, bh_result
,
3014 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
3017 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
3018 struct buffer_head
*bh_result
, int create
)
3020 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3021 inode
->i_ino
, create
);
3022 return _ext4_get_block(inode
, iblock
, bh_result
,
3023 EXT4_GET_BLOCKS_NO_LOCK
);
3026 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3027 ssize_t size
, void *private, int ret
,
3030 struct inode
*inode
= file_inode(iocb
->ki_filp
);
3031 ext4_io_end_t
*io_end
= iocb
->private;
3033 /* if not async direct IO just return */
3035 inode_dio_done(inode
);
3037 aio_complete(iocb
, ret
, 0);
3041 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3042 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3043 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3046 iocb
->private = NULL
;
3047 io_end
->offset
= offset
;
3048 io_end
->size
= size
;
3050 io_end
->iocb
= iocb
;
3051 io_end
->result
= ret
;
3053 ext4_put_io_end_defer(io_end
);
3057 * For ext4 extent files, ext4 will do direct-io write to holes,
3058 * preallocated extents, and those write extend the file, no need to
3059 * fall back to buffered IO.
3061 * For holes, we fallocate those blocks, mark them as uninitialized
3062 * If those blocks were preallocated, we mark sure they are split, but
3063 * still keep the range to write as uninitialized.
3065 * The unwritten extents will be converted to written when DIO is completed.
3066 * For async direct IO, since the IO may still pending when return, we
3067 * set up an end_io call back function, which will do the conversion
3068 * when async direct IO completed.
3070 * If the O_DIRECT write will extend the file then add this inode to the
3071 * orphan list. So recovery will truncate it back to the original size
3072 * if the machine crashes during the write.
3075 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3076 const struct iovec
*iov
, loff_t offset
,
3077 unsigned long nr_segs
)
3079 struct file
*file
= iocb
->ki_filp
;
3080 struct inode
*inode
= file
->f_mapping
->host
;
3082 size_t count
= iov_length(iov
, nr_segs
);
3084 get_block_t
*get_block_func
= NULL
;
3086 loff_t final_size
= offset
+ count
;
3087 ext4_io_end_t
*io_end
= NULL
;
3089 /* Use the old path for reads and writes beyond i_size. */
3090 if (rw
!= WRITE
|| final_size
> inode
->i_size
)
3091 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3093 BUG_ON(iocb
->private == NULL
);
3096 * Make all waiters for direct IO properly wait also for extent
3097 * conversion. This also disallows race between truncate() and
3098 * overwrite DIO as i_dio_count needs to be incremented under i_mutex.
3101 atomic_inc(&inode
->i_dio_count
);
3103 /* If we do a overwrite dio, i_mutex locking can be released */
3104 overwrite
= *((int *)iocb
->private);
3107 down_read(&EXT4_I(inode
)->i_data_sem
);
3108 mutex_unlock(&inode
->i_mutex
);
3112 * We could direct write to holes and fallocate.
3114 * Allocated blocks to fill the hole are marked as
3115 * uninitialized to prevent parallel buffered read to expose
3116 * the stale data before DIO complete the data IO.
3118 * As to previously fallocated extents, ext4 get_block will
3119 * just simply mark the buffer mapped but still keep the
3120 * extents uninitialized.
3122 * For non AIO case, we will convert those unwritten extents
3123 * to written after return back from blockdev_direct_IO.
3125 * For async DIO, the conversion needs to be deferred when the
3126 * IO is completed. The ext4 end_io callback function will be
3127 * called to take care of the conversion work. Here for async
3128 * case, we allocate an io_end structure to hook to the iocb.
3130 iocb
->private = NULL
;
3131 ext4_inode_aio_set(inode
, NULL
);
3132 if (!is_sync_kiocb(iocb
)) {
3133 io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
3138 io_end
->flag
|= EXT4_IO_END_DIRECT
;
3140 * Grab reference for DIO. Will be dropped in ext4_end_io_dio()
3142 iocb
->private = ext4_get_io_end(io_end
);
3144 * we save the io structure for current async direct
3145 * IO, so that later ext4_map_blocks() could flag the
3146 * io structure whether there is a unwritten extents
3147 * needs to be converted when IO is completed.
3149 ext4_inode_aio_set(inode
, io_end
);
3153 get_block_func
= ext4_get_block_write_nolock
;
3155 get_block_func
= ext4_get_block_write
;
3156 dio_flags
= DIO_LOCKING
;
3158 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3159 inode
->i_sb
->s_bdev
, iov
,
3167 * Put our reference to io_end. This can free the io_end structure e.g.
3168 * in sync IO case or in case of error. It can even perform extent
3169 * conversion if all bios we submitted finished before we got here.
3170 * Note that in that case iocb->private can be already set to NULL
3174 ext4_inode_aio_set(inode
, NULL
);
3175 ext4_put_io_end(io_end
);
3177 * When no IO was submitted ext4_end_io_dio() was not
3178 * called so we have to put iocb's reference.
3180 if (ret
<= 0 && ret
!= -EIOCBQUEUED
&& iocb
->private) {
3181 WARN_ON(iocb
->private != io_end
);
3182 WARN_ON(io_end
->flag
& EXT4_IO_END_UNWRITTEN
);
3183 WARN_ON(io_end
->iocb
);
3185 * Generic code already did inode_dio_done() so we
3186 * have to clear EXT4_IO_END_DIRECT to not do it for
3190 ext4_put_io_end(io_end
);
3191 iocb
->private = NULL
;
3194 if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3195 EXT4_STATE_DIO_UNWRITTEN
)) {
3198 * for non AIO case, since the IO is already
3199 * completed, we could do the conversion right here
3201 err
= ext4_convert_unwritten_extents(NULL
, inode
,
3205 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3210 inode_dio_done(inode
);
3211 /* take i_mutex locking again if we do a ovewrite dio */
3213 up_read(&EXT4_I(inode
)->i_data_sem
);
3214 mutex_lock(&inode
->i_mutex
);
3220 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3221 const struct iovec
*iov
, loff_t offset
,
3222 unsigned long nr_segs
)
3224 struct file
*file
= iocb
->ki_filp
;
3225 struct inode
*inode
= file
->f_mapping
->host
;
3229 * If we are doing data journalling we don't support O_DIRECT
3231 if (ext4_should_journal_data(inode
))
3234 /* Let buffer I/O handle the inline data case. */
3235 if (ext4_has_inline_data(inode
))
3238 trace_ext4_direct_IO_enter(inode
, offset
, iov_length(iov
, nr_segs
), rw
);
3239 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3240 ret
= ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3242 ret
= ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3243 trace_ext4_direct_IO_exit(inode
, offset
,
3244 iov_length(iov
, nr_segs
), rw
, ret
);
3249 * Pages can be marked dirty completely asynchronously from ext4's journalling
3250 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3251 * much here because ->set_page_dirty is called under VFS locks. The page is
3252 * not necessarily locked.
3254 * We cannot just dirty the page and leave attached buffers clean, because the
3255 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3256 * or jbddirty because all the journalling code will explode.
3258 * So what we do is to mark the page "pending dirty" and next time writepage
3259 * is called, propagate that into the buffers appropriately.
3261 static int ext4_journalled_set_page_dirty(struct page
*page
)
3263 SetPageChecked(page
);
3264 return __set_page_dirty_nobuffers(page
);
3267 static const struct address_space_operations ext4_aops
= {
3268 .readpage
= ext4_readpage
,
3269 .readpages
= ext4_readpages
,
3270 .writepage
= ext4_writepage
,
3271 .writepages
= ext4_writepages
,
3272 .write_begin
= ext4_write_begin
,
3273 .write_end
= ext4_write_end
,
3275 .invalidatepage
= ext4_invalidatepage
,
3276 .releasepage
= ext4_releasepage
,
3277 .direct_IO
= ext4_direct_IO
,
3278 .migratepage
= buffer_migrate_page
,
3279 .is_partially_uptodate
= block_is_partially_uptodate
,
3280 .error_remove_page
= generic_error_remove_page
,
3283 static const struct address_space_operations ext4_journalled_aops
= {
3284 .readpage
= ext4_readpage
,
3285 .readpages
= ext4_readpages
,
3286 .writepage
= ext4_writepage
,
3287 .writepages
= ext4_writepages
,
3288 .write_begin
= ext4_write_begin
,
3289 .write_end
= ext4_journalled_write_end
,
3290 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3292 .invalidatepage
= ext4_journalled_invalidatepage
,
3293 .releasepage
= ext4_releasepage
,
3294 .direct_IO
= ext4_direct_IO
,
3295 .is_partially_uptodate
= block_is_partially_uptodate
,
3296 .error_remove_page
= generic_error_remove_page
,
3299 static const struct address_space_operations ext4_da_aops
= {
3300 .readpage
= ext4_readpage
,
3301 .readpages
= ext4_readpages
,
3302 .writepage
= ext4_writepage
,
3303 .writepages
= ext4_writepages
,
3304 .write_begin
= ext4_da_write_begin
,
3305 .write_end
= ext4_da_write_end
,
3307 .invalidatepage
= ext4_da_invalidatepage
,
3308 .releasepage
= ext4_releasepage
,
3309 .direct_IO
= ext4_direct_IO
,
3310 .migratepage
= buffer_migrate_page
,
3311 .is_partially_uptodate
= block_is_partially_uptodate
,
3312 .error_remove_page
= generic_error_remove_page
,
3315 void ext4_set_aops(struct inode
*inode
)
3317 switch (ext4_inode_journal_mode(inode
)) {
3318 case EXT4_INODE_ORDERED_DATA_MODE
:
3319 ext4_set_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3321 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3322 ext4_clear_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3324 case EXT4_INODE_JOURNAL_DATA_MODE
:
3325 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3330 if (test_opt(inode
->i_sb
, DELALLOC
))
3331 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3333 inode
->i_mapping
->a_ops
= &ext4_aops
;
3337 * ext4_block_truncate_page() zeroes out a mapping from file offset `from'
3338 * up to the end of the block which corresponds to `from'.
3339 * This required during truncate. We need to physically zero the tail end
3340 * of that block so it doesn't yield old data if the file is later grown.
3342 int ext4_block_truncate_page(handle_t
*handle
,
3343 struct address_space
*mapping
, loff_t from
)
3345 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3348 struct inode
*inode
= mapping
->host
;
3350 blocksize
= inode
->i_sb
->s_blocksize
;
3351 length
= blocksize
- (offset
& (blocksize
- 1));
3353 return ext4_block_zero_page_range(handle
, mapping
, from
, length
);
3357 * ext4_block_zero_page_range() zeros out a mapping of length 'length'
3358 * starting from file offset 'from'. The range to be zero'd must
3359 * be contained with in one block. If the specified range exceeds
3360 * the end of the block it will be shortened to end of the block
3361 * that cooresponds to 'from'
3363 int ext4_block_zero_page_range(handle_t
*handle
,
3364 struct address_space
*mapping
, loff_t from
, loff_t length
)
3366 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3367 unsigned offset
= from
& (PAGE_CACHE_SIZE
-1);
3368 unsigned blocksize
, max
, pos
;
3370 struct inode
*inode
= mapping
->host
;
3371 struct buffer_head
*bh
;
3375 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3376 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3380 blocksize
= inode
->i_sb
->s_blocksize
;
3381 max
= blocksize
- (offset
& (blocksize
- 1));
3384 * correct length if it does not fall between
3385 * 'from' and the end of the block
3387 if (length
> max
|| length
< 0)
3390 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3392 if (!page_has_buffers(page
))
3393 create_empty_buffers(page
, blocksize
, 0);
3395 /* Find the buffer that contains "offset" */
3396 bh
= page_buffers(page
);
3398 while (offset
>= pos
) {
3399 bh
= bh
->b_this_page
;
3403 if (buffer_freed(bh
)) {
3404 BUFFER_TRACE(bh
, "freed: skip");
3407 if (!buffer_mapped(bh
)) {
3408 BUFFER_TRACE(bh
, "unmapped");
3409 ext4_get_block(inode
, iblock
, bh
, 0);
3410 /* unmapped? It's a hole - nothing to do */
3411 if (!buffer_mapped(bh
)) {
3412 BUFFER_TRACE(bh
, "still unmapped");
3417 /* Ok, it's mapped. Make sure it's up-to-date */
3418 if (PageUptodate(page
))
3419 set_buffer_uptodate(bh
);
3421 if (!buffer_uptodate(bh
)) {
3423 ll_rw_block(READ
, 1, &bh
);
3425 /* Uhhuh. Read error. Complain and punt. */
3426 if (!buffer_uptodate(bh
))
3429 if (ext4_should_journal_data(inode
)) {
3430 BUFFER_TRACE(bh
, "get write access");
3431 err
= ext4_journal_get_write_access(handle
, bh
);
3435 zero_user(page
, offset
, length
);
3436 BUFFER_TRACE(bh
, "zeroed end of block");
3438 if (ext4_should_journal_data(inode
)) {
3439 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3442 mark_buffer_dirty(bh
);
3443 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
))
3444 err
= ext4_jbd2_file_inode(handle
, inode
);
3449 page_cache_release(page
);
3453 int ext4_zero_partial_blocks(handle_t
*handle
, struct inode
*inode
,
3454 loff_t lstart
, loff_t length
)
3456 struct super_block
*sb
= inode
->i_sb
;
3457 struct address_space
*mapping
= inode
->i_mapping
;
3458 unsigned partial_start
, partial_end
;
3459 ext4_fsblk_t start
, end
;
3460 loff_t byte_end
= (lstart
+ length
- 1);
3463 partial_start
= lstart
& (sb
->s_blocksize
- 1);
3464 partial_end
= byte_end
& (sb
->s_blocksize
- 1);
3466 start
= lstart
>> sb
->s_blocksize_bits
;
3467 end
= byte_end
>> sb
->s_blocksize_bits
;
3469 /* Handle partial zero within the single block */
3471 (partial_start
|| (partial_end
!= sb
->s_blocksize
- 1))) {
3472 err
= ext4_block_zero_page_range(handle
, mapping
,
3476 /* Handle partial zero out on the start of the range */
3477 if (partial_start
) {
3478 err
= ext4_block_zero_page_range(handle
, mapping
,
3479 lstart
, sb
->s_blocksize
);
3483 /* Handle partial zero out on the end of the range */
3484 if (partial_end
!= sb
->s_blocksize
- 1)
3485 err
= ext4_block_zero_page_range(handle
, mapping
,
3486 byte_end
- partial_end
,
3491 int ext4_can_truncate(struct inode
*inode
)
3493 if (S_ISREG(inode
->i_mode
))
3495 if (S_ISDIR(inode
->i_mode
))
3497 if (S_ISLNK(inode
->i_mode
))
3498 return !ext4_inode_is_fast_symlink(inode
);
3503 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3504 * associated with the given offset and length
3506 * @inode: File inode
3507 * @offset: The offset where the hole will begin
3508 * @len: The length of the hole
3510 * Returns: 0 on success or negative on failure
3513 int ext4_punch_hole(struct inode
*inode
, loff_t offset
, loff_t length
)
3515 struct super_block
*sb
= inode
->i_sb
;
3516 ext4_lblk_t first_block
, stop_block
;
3517 struct address_space
*mapping
= inode
->i_mapping
;
3518 loff_t first_block_offset
, last_block_offset
;
3520 unsigned int credits
;
3523 if (!S_ISREG(inode
->i_mode
))
3526 if (EXT4_SB(sb
)->s_cluster_ratio
> 1) {
3527 /* TODO: Add support for bigalloc file systems */
3531 trace_ext4_punch_hole(inode
, offset
, length
);
3534 * Write out all dirty pages to avoid race conditions
3535 * Then release them.
3537 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3538 ret
= filemap_write_and_wait_range(mapping
, offset
,
3539 offset
+ length
- 1);
3544 mutex_lock(&inode
->i_mutex
);
3545 /* It's not possible punch hole on append only file */
3546 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
)) {
3550 if (IS_SWAPFILE(inode
)) {
3555 /* No need to punch hole beyond i_size */
3556 if (offset
>= inode
->i_size
)
3560 * If the hole extends beyond i_size, set the hole
3561 * to end after the page that contains i_size
3563 if (offset
+ length
> inode
->i_size
) {
3564 length
= inode
->i_size
+
3565 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
3569 if (offset
& (sb
->s_blocksize
- 1) ||
3570 (offset
+ length
) & (sb
->s_blocksize
- 1)) {
3572 * Attach jinode to inode for jbd2 if we do any zeroing of
3575 ret
= ext4_inode_attach_jinode(inode
);
3581 first_block_offset
= round_up(offset
, sb
->s_blocksize
);
3582 last_block_offset
= round_down((offset
+ length
), sb
->s_blocksize
) - 1;
3584 /* Now release the pages and zero block aligned part of pages*/
3585 if (last_block_offset
> first_block_offset
)
3586 truncate_pagecache_range(inode
, first_block_offset
,
3589 /* Wait all existing dio workers, newcomers will block on i_mutex */
3590 ext4_inode_block_unlocked_dio(inode
);
3591 inode_dio_wait(inode
);
3593 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3594 credits
= ext4_writepage_trans_blocks(inode
);
3596 credits
= ext4_blocks_for_truncate(inode
);
3597 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3598 if (IS_ERR(handle
)) {
3599 ret
= PTR_ERR(handle
);
3600 ext4_std_error(sb
, ret
);
3604 ret
= ext4_zero_partial_blocks(handle
, inode
, offset
,
3609 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
3610 EXT4_BLOCK_SIZE_BITS(sb
);
3611 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
3613 /* If there are no blocks to remove, return now */
3614 if (first_block
>= stop_block
)
3617 down_write(&EXT4_I(inode
)->i_data_sem
);
3618 ext4_discard_preallocations(inode
);
3620 ret
= ext4_es_remove_extent(inode
, first_block
,
3621 stop_block
- first_block
);
3623 up_write(&EXT4_I(inode
)->i_data_sem
);
3627 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3628 ret
= ext4_ext_remove_space(inode
, first_block
,
3631 ret
= ext4_free_hole_blocks(handle
, inode
, first_block
,
3634 ext4_discard_preallocations(inode
);
3635 up_write(&EXT4_I(inode
)->i_data_sem
);
3637 ext4_handle_sync(handle
);
3638 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3639 ext4_mark_inode_dirty(handle
, inode
);
3641 ext4_journal_stop(handle
);
3643 ext4_inode_resume_unlocked_dio(inode
);
3645 mutex_unlock(&inode
->i_mutex
);
3649 int ext4_inode_attach_jinode(struct inode
*inode
)
3651 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3652 struct jbd2_inode
*jinode
;
3654 if (ei
->jinode
|| !EXT4_SB(inode
->i_sb
)->s_journal
)
3657 jinode
= jbd2_alloc_inode(GFP_KERNEL
);
3658 spin_lock(&inode
->i_lock
);
3661 spin_unlock(&inode
->i_lock
);
3664 ei
->jinode
= jinode
;
3665 jbd2_journal_init_jbd_inode(ei
->jinode
, inode
);
3668 spin_unlock(&inode
->i_lock
);
3669 if (unlikely(jinode
!= NULL
))
3670 jbd2_free_inode(jinode
);
3677 * We block out ext4_get_block() block instantiations across the entire
3678 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3679 * simultaneously on behalf of the same inode.
3681 * As we work through the truncate and commit bits of it to the journal there
3682 * is one core, guiding principle: the file's tree must always be consistent on
3683 * disk. We must be able to restart the truncate after a crash.
3685 * The file's tree may be transiently inconsistent in memory (although it
3686 * probably isn't), but whenever we close off and commit a journal transaction,
3687 * the contents of (the filesystem + the journal) must be consistent and
3688 * restartable. It's pretty simple, really: bottom up, right to left (although
3689 * left-to-right works OK too).
3691 * Note that at recovery time, journal replay occurs *before* the restart of
3692 * truncate against the orphan inode list.
3694 * The committed inode has the new, desired i_size (which is the same as
3695 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3696 * that this inode's truncate did not complete and it will again call
3697 * ext4_truncate() to have another go. So there will be instantiated blocks
3698 * to the right of the truncation point in a crashed ext4 filesystem. But
3699 * that's fine - as long as they are linked from the inode, the post-crash
3700 * ext4_truncate() run will find them and release them.
3702 void ext4_truncate(struct inode
*inode
)
3704 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3705 unsigned int credits
;
3707 struct address_space
*mapping
= inode
->i_mapping
;
3710 * There is a possibility that we're either freeing the inode
3711 * or it completely new indode. In those cases we might not
3712 * have i_mutex locked because it's not necessary.
3714 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
3715 WARN_ON(!mutex_is_locked(&inode
->i_mutex
));
3716 trace_ext4_truncate_enter(inode
);
3718 if (!ext4_can_truncate(inode
))
3721 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3723 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3724 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3726 if (ext4_has_inline_data(inode
)) {
3729 ext4_inline_data_truncate(inode
, &has_inline
);
3734 /* If we zero-out tail of the page, we have to create jinode for jbd2 */
3735 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1)) {
3736 if (ext4_inode_attach_jinode(inode
) < 0)
3740 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3741 credits
= ext4_writepage_trans_blocks(inode
);
3743 credits
= ext4_blocks_for_truncate(inode
);
3745 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3746 if (IS_ERR(handle
)) {
3747 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
3751 if (inode
->i_size
& (inode
->i_sb
->s_blocksize
- 1))
3752 ext4_block_truncate_page(handle
, mapping
, inode
->i_size
);
3755 * We add the inode to the orphan list, so that if this
3756 * truncate spans multiple transactions, and we crash, we will
3757 * resume the truncate when the filesystem recovers. It also
3758 * marks the inode dirty, to catch the new size.
3760 * Implication: the file must always be in a sane, consistent
3761 * truncatable state while each transaction commits.
3763 if (ext4_orphan_add(handle
, inode
))
3766 down_write(&EXT4_I(inode
)->i_data_sem
);
3768 ext4_discard_preallocations(inode
);
3770 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3771 ext4_ext_truncate(handle
, inode
);
3773 ext4_ind_truncate(handle
, inode
);
3775 up_write(&ei
->i_data_sem
);
3778 ext4_handle_sync(handle
);
3782 * If this was a simple ftruncate() and the file will remain alive,
3783 * then we need to clear up the orphan record which we created above.
3784 * However, if this was a real unlink then we were called by
3785 * ext4_delete_inode(), and we allow that function to clean up the
3786 * orphan info for us.
3789 ext4_orphan_del(handle
, inode
);
3791 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3792 ext4_mark_inode_dirty(handle
, inode
);
3793 ext4_journal_stop(handle
);
3795 trace_ext4_truncate_exit(inode
);
3799 * ext4_get_inode_loc returns with an extra refcount against the inode's
3800 * underlying buffer_head on success. If 'in_mem' is true, we have all
3801 * data in memory that is needed to recreate the on-disk version of this
3804 static int __ext4_get_inode_loc(struct inode
*inode
,
3805 struct ext4_iloc
*iloc
, int in_mem
)
3807 struct ext4_group_desc
*gdp
;
3808 struct buffer_head
*bh
;
3809 struct super_block
*sb
= inode
->i_sb
;
3811 int inodes_per_block
, inode_offset
;
3814 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3817 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3818 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3823 * Figure out the offset within the block group inode table
3825 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3826 inode_offset
= ((inode
->i_ino
- 1) %
3827 EXT4_INODES_PER_GROUP(sb
));
3828 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3829 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3831 bh
= sb_getblk(sb
, block
);
3834 if (!buffer_uptodate(bh
)) {
3838 * If the buffer has the write error flag, we have failed
3839 * to write out another inode in the same block. In this
3840 * case, we don't have to read the block because we may
3841 * read the old inode data successfully.
3843 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3844 set_buffer_uptodate(bh
);
3846 if (buffer_uptodate(bh
)) {
3847 /* someone brought it uptodate while we waited */
3853 * If we have all information of the inode in memory and this
3854 * is the only valid inode in the block, we need not read the
3858 struct buffer_head
*bitmap_bh
;
3861 start
= inode_offset
& ~(inodes_per_block
- 1);
3863 /* Is the inode bitmap in cache? */
3864 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3865 if (unlikely(!bitmap_bh
))
3869 * If the inode bitmap isn't in cache then the
3870 * optimisation may end up performing two reads instead
3871 * of one, so skip it.
3873 if (!buffer_uptodate(bitmap_bh
)) {
3877 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3878 if (i
== inode_offset
)
3880 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3884 if (i
== start
+ inodes_per_block
) {
3885 /* all other inodes are free, so skip I/O */
3886 memset(bh
->b_data
, 0, bh
->b_size
);
3887 set_buffer_uptodate(bh
);
3895 * If we need to do any I/O, try to pre-readahead extra
3896 * blocks from the inode table.
3898 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3899 ext4_fsblk_t b
, end
, table
;
3901 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
3903 table
= ext4_inode_table(sb
, gdp
);
3904 /* s_inode_readahead_blks is always a power of 2 */
3905 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
3909 num
= EXT4_INODES_PER_GROUP(sb
);
3910 if (ext4_has_group_desc_csum(sb
))
3911 num
-= ext4_itable_unused_count(sb
, gdp
);
3912 table
+= num
/ inodes_per_block
;
3916 sb_breadahead(sb
, b
++);
3920 * There are other valid inodes in the buffer, this inode
3921 * has in-inode xattrs, or we don't have this inode in memory.
3922 * Read the block from disk.
3924 trace_ext4_load_inode(inode
);
3926 bh
->b_end_io
= end_buffer_read_sync
;
3927 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
3929 if (!buffer_uptodate(bh
)) {
3930 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3931 "unable to read itable block");
3941 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3943 /* We have all inode data except xattrs in memory here. */
3944 return __ext4_get_inode_loc(inode
, iloc
,
3945 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
3948 void ext4_set_inode_flags(struct inode
*inode
)
3950 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3952 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3953 if (flags
& EXT4_SYNC_FL
)
3954 inode
->i_flags
|= S_SYNC
;
3955 if (flags
& EXT4_APPEND_FL
)
3956 inode
->i_flags
|= S_APPEND
;
3957 if (flags
& EXT4_IMMUTABLE_FL
)
3958 inode
->i_flags
|= S_IMMUTABLE
;
3959 if (flags
& EXT4_NOATIME_FL
)
3960 inode
->i_flags
|= S_NOATIME
;
3961 if (flags
& EXT4_DIRSYNC_FL
)
3962 inode
->i_flags
|= S_DIRSYNC
;
3965 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3966 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3968 unsigned int vfs_fl
;
3969 unsigned long old_fl
, new_fl
;
3972 vfs_fl
= ei
->vfs_inode
.i_flags
;
3973 old_fl
= ei
->i_flags
;
3974 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3975 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
3977 if (vfs_fl
& S_SYNC
)
3978 new_fl
|= EXT4_SYNC_FL
;
3979 if (vfs_fl
& S_APPEND
)
3980 new_fl
|= EXT4_APPEND_FL
;
3981 if (vfs_fl
& S_IMMUTABLE
)
3982 new_fl
|= EXT4_IMMUTABLE_FL
;
3983 if (vfs_fl
& S_NOATIME
)
3984 new_fl
|= EXT4_NOATIME_FL
;
3985 if (vfs_fl
& S_DIRSYNC
)
3986 new_fl
|= EXT4_DIRSYNC_FL
;
3987 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
3990 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3991 struct ext4_inode_info
*ei
)
3994 struct inode
*inode
= &(ei
->vfs_inode
);
3995 struct super_block
*sb
= inode
->i_sb
;
3997 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3998 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3999 /* we are using combined 48 bit field */
4000 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4001 le32_to_cpu(raw_inode
->i_blocks_lo
);
4002 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4003 /* i_blocks represent file system block size */
4004 return i_blocks
<< (inode
->i_blkbits
- 9);
4009 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4013 static inline void ext4_iget_extra_inode(struct inode
*inode
,
4014 struct ext4_inode
*raw_inode
,
4015 struct ext4_inode_info
*ei
)
4017 __le32
*magic
= (void *)raw_inode
+
4018 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4019 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4020 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4021 ext4_find_inline_data_nolock(inode
);
4023 EXT4_I(inode
)->i_inline_off
= 0;
4026 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4028 struct ext4_iloc iloc
;
4029 struct ext4_inode
*raw_inode
;
4030 struct ext4_inode_info
*ei
;
4031 struct inode
*inode
;
4032 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4038 inode
= iget_locked(sb
, ino
);
4040 return ERR_PTR(-ENOMEM
);
4041 if (!(inode
->i_state
& I_NEW
))
4047 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4050 raw_inode
= ext4_raw_inode(&iloc
);
4052 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4053 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4054 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4055 EXT4_INODE_SIZE(inode
->i_sb
)) {
4056 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
4057 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
4058 EXT4_INODE_SIZE(inode
->i_sb
));
4063 ei
->i_extra_isize
= 0;
4065 /* Precompute checksum seed for inode metadata */
4066 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4067 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
)) {
4068 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4070 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4071 __le32 gen
= raw_inode
->i_generation
;
4072 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4074 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4078 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4079 EXT4_ERROR_INODE(inode
, "checksum invalid");
4084 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4085 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4086 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4087 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4088 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4089 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4091 i_uid_write(inode
, i_uid
);
4092 i_gid_write(inode
, i_gid
);
4093 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4095 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4096 ei
->i_inline_off
= 0;
4097 ei
->i_dir_start_lookup
= 0;
4098 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4099 /* We now have enough fields to check if the inode was active or not.
4100 * This is needed because nfsd might try to access dead inodes
4101 * the test is that same one that e2fsck uses
4102 * NeilBrown 1999oct15
4104 if (inode
->i_nlink
== 0) {
4105 if ((inode
->i_mode
== 0 ||
4106 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4107 ino
!= EXT4_BOOT_LOADER_INO
) {
4108 /* this inode is deleted */
4112 /* The only unlinked inodes we let through here have
4113 * valid i_mode and are being read by the orphan
4114 * recovery code: that's fine, we're about to complete
4115 * the process of deleting those.
4116 * OR it is the EXT4_BOOT_LOADER_INO which is
4117 * not initialized on a new filesystem. */
4119 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4120 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4121 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4122 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4124 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4125 inode
->i_size
= ext4_isize(raw_inode
);
4126 ei
->i_disksize
= inode
->i_size
;
4128 ei
->i_reserved_quota
= 0;
4130 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4131 ei
->i_block_group
= iloc
.block_group
;
4132 ei
->i_last_alloc_group
= ~0;
4134 * NOTE! The in-memory inode i_data array is in little-endian order
4135 * even on big-endian machines: we do NOT byteswap the block numbers!
4137 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4138 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4139 INIT_LIST_HEAD(&ei
->i_orphan
);
4142 * Set transaction id's of transactions that have to be committed
4143 * to finish f[data]sync. We set them to currently running transaction
4144 * as we cannot be sure that the inode or some of its metadata isn't
4145 * part of the transaction - the inode could have been reclaimed and
4146 * now it is reread from disk.
4149 transaction_t
*transaction
;
4152 read_lock(&journal
->j_state_lock
);
4153 if (journal
->j_running_transaction
)
4154 transaction
= journal
->j_running_transaction
;
4156 transaction
= journal
->j_committing_transaction
;
4158 tid
= transaction
->t_tid
;
4160 tid
= journal
->j_commit_sequence
;
4161 read_unlock(&journal
->j_state_lock
);
4162 ei
->i_sync_tid
= tid
;
4163 ei
->i_datasync_tid
= tid
;
4166 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4167 if (ei
->i_extra_isize
== 0) {
4168 /* The extra space is currently unused. Use it. */
4169 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4170 EXT4_GOOD_OLD_INODE_SIZE
;
4172 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4176 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4177 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4178 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4179 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4181 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4182 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4183 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4185 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4189 if (ei
->i_file_acl
&&
4190 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4191 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4195 } else if (!ext4_has_inline_data(inode
)) {
4196 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4197 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4198 (S_ISLNK(inode
->i_mode
) &&
4199 !ext4_inode_is_fast_symlink(inode
))))
4200 /* Validate extent which is part of inode */
4201 ret
= ext4_ext_check_inode(inode
);
4202 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4203 (S_ISLNK(inode
->i_mode
) &&
4204 !ext4_inode_is_fast_symlink(inode
))) {
4205 /* Validate block references which are part of inode */
4206 ret
= ext4_ind_check_inode(inode
);
4212 if (S_ISREG(inode
->i_mode
)) {
4213 inode
->i_op
= &ext4_file_inode_operations
;
4214 inode
->i_fop
= &ext4_file_operations
;
4215 ext4_set_aops(inode
);
4216 } else if (S_ISDIR(inode
->i_mode
)) {
4217 inode
->i_op
= &ext4_dir_inode_operations
;
4218 inode
->i_fop
= &ext4_dir_operations
;
4219 } else if (S_ISLNK(inode
->i_mode
)) {
4220 if (ext4_inode_is_fast_symlink(inode
)) {
4221 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4222 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4223 sizeof(ei
->i_data
) - 1);
4225 inode
->i_op
= &ext4_symlink_inode_operations
;
4226 ext4_set_aops(inode
);
4228 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4229 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4230 inode
->i_op
= &ext4_special_inode_operations
;
4231 if (raw_inode
->i_block
[0])
4232 init_special_inode(inode
, inode
->i_mode
,
4233 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4235 init_special_inode(inode
, inode
->i_mode
,
4236 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4237 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4238 make_bad_inode(inode
);
4241 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4245 ext4_set_inode_flags(inode
);
4246 unlock_new_inode(inode
);
4252 return ERR_PTR(ret
);
4255 static int ext4_inode_blocks_set(handle_t
*handle
,
4256 struct ext4_inode
*raw_inode
,
4257 struct ext4_inode_info
*ei
)
4259 struct inode
*inode
= &(ei
->vfs_inode
);
4260 u64 i_blocks
= inode
->i_blocks
;
4261 struct super_block
*sb
= inode
->i_sb
;
4263 if (i_blocks
<= ~0U) {
4265 * i_blocks can be represented in a 32 bit variable
4266 * as multiple of 512 bytes
4268 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4269 raw_inode
->i_blocks_high
= 0;
4270 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4273 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4276 if (i_blocks
<= 0xffffffffffffULL
) {
4278 * i_blocks can be represented in a 48 bit variable
4279 * as multiple of 512 bytes
4281 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4282 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4283 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4285 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4286 /* i_block is stored in file system block size */
4287 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4288 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4289 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4295 * Post the struct inode info into an on-disk inode location in the
4296 * buffer-cache. This gobbles the caller's reference to the
4297 * buffer_head in the inode location struct.
4299 * The caller must have write access to iloc->bh.
4301 static int ext4_do_update_inode(handle_t
*handle
,
4302 struct inode
*inode
,
4303 struct ext4_iloc
*iloc
)
4305 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4306 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4307 struct buffer_head
*bh
= iloc
->bh
;
4308 int err
= 0, rc
, block
;
4309 int need_datasync
= 0;
4313 /* For fields not not tracking in the in-memory inode,
4314 * initialise them to zero for new inodes. */
4315 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4316 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4318 ext4_get_inode_flags(ei
);
4319 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4320 i_uid
= i_uid_read(inode
);
4321 i_gid
= i_gid_read(inode
);
4322 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4323 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4324 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4326 * Fix up interoperability with old kernels. Otherwise, old inodes get
4327 * re-used with the upper 16 bits of the uid/gid intact
4330 raw_inode
->i_uid_high
=
4331 cpu_to_le16(high_16_bits(i_uid
));
4332 raw_inode
->i_gid_high
=
4333 cpu_to_le16(high_16_bits(i_gid
));
4335 raw_inode
->i_uid_high
= 0;
4336 raw_inode
->i_gid_high
= 0;
4339 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4340 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4341 raw_inode
->i_uid_high
= 0;
4342 raw_inode
->i_gid_high
= 0;
4344 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4346 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4347 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4348 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4349 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4351 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4353 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4354 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4355 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4356 cpu_to_le32(EXT4_OS_HURD
))
4357 raw_inode
->i_file_acl_high
=
4358 cpu_to_le16(ei
->i_file_acl
>> 32);
4359 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4360 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4361 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4364 if (ei
->i_disksize
> 0x7fffffffULL
) {
4365 struct super_block
*sb
= inode
->i_sb
;
4366 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4367 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4368 EXT4_SB(sb
)->s_es
->s_rev_level
==
4369 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4370 /* If this is the first large file
4371 * created, add a flag to the superblock.
4373 err
= ext4_journal_get_write_access(handle
,
4374 EXT4_SB(sb
)->s_sbh
);
4377 ext4_update_dynamic_rev(sb
);
4378 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4379 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4380 ext4_handle_sync(handle
);
4381 err
= ext4_handle_dirty_super(handle
, sb
);
4384 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4385 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4386 if (old_valid_dev(inode
->i_rdev
)) {
4387 raw_inode
->i_block
[0] =
4388 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4389 raw_inode
->i_block
[1] = 0;
4391 raw_inode
->i_block
[0] = 0;
4392 raw_inode
->i_block
[1] =
4393 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4394 raw_inode
->i_block
[2] = 0;
4396 } else if (!ext4_has_inline_data(inode
)) {
4397 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4398 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4401 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4402 if (ei
->i_extra_isize
) {
4403 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4404 raw_inode
->i_version_hi
=
4405 cpu_to_le32(inode
->i_version
>> 32);
4406 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4409 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4411 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4412 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4415 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4417 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4420 ext4_std_error(inode
->i_sb
, err
);
4425 * ext4_write_inode()
4427 * We are called from a few places:
4429 * - Within generic_file_write() for O_SYNC files.
4430 * Here, there will be no transaction running. We wait for any running
4431 * transaction to commit.
4433 * - Within sys_sync(), kupdate and such.
4434 * We wait on commit, if tol to.
4436 * - Within prune_icache() (PF_MEMALLOC == true)
4437 * Here we simply return. We can't afford to block kswapd on the
4440 * In all cases it is actually safe for us to return without doing anything,
4441 * because the inode has been copied into a raw inode buffer in
4442 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4445 * Note that we are absolutely dependent upon all inode dirtiers doing the
4446 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4447 * which we are interested.
4449 * It would be a bug for them to not do this. The code:
4451 * mark_inode_dirty(inode)
4453 * inode->i_size = expr;
4455 * is in error because a kswapd-driven write_inode() could occur while
4456 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4457 * will no longer be on the superblock's dirty inode list.
4459 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4463 if (current
->flags
& PF_MEMALLOC
)
4466 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4467 if (ext4_journal_current_handle()) {
4468 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4473 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
4476 err
= ext4_force_commit(inode
->i_sb
);
4478 struct ext4_iloc iloc
;
4480 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4483 if (wbc
->sync_mode
== WB_SYNC_ALL
)
4484 sync_dirty_buffer(iloc
.bh
);
4485 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4486 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4487 "IO error syncing inode");
4496 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4497 * buffers that are attached to a page stradding i_size and are undergoing
4498 * commit. In that case we have to wait for commit to finish and try again.
4500 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4504 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4505 tid_t commit_tid
= 0;
4508 offset
= inode
->i_size
& (PAGE_CACHE_SIZE
- 1);
4510 * All buffers in the last page remain valid? Then there's nothing to
4511 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4514 if (offset
> PAGE_CACHE_SIZE
- (1 << inode
->i_blkbits
))
4517 page
= find_lock_page(inode
->i_mapping
,
4518 inode
->i_size
>> PAGE_CACHE_SHIFT
);
4521 ret
= __ext4_journalled_invalidatepage(page
, offset
,
4522 PAGE_CACHE_SIZE
- offset
);
4524 page_cache_release(page
);
4528 read_lock(&journal
->j_state_lock
);
4529 if (journal
->j_committing_transaction
)
4530 commit_tid
= journal
->j_committing_transaction
->t_tid
;
4531 read_unlock(&journal
->j_state_lock
);
4533 jbd2_log_wait_commit(journal
, commit_tid
);
4540 * Called from notify_change.
4542 * We want to trap VFS attempts to truncate the file as soon as
4543 * possible. In particular, we want to make sure that when the VFS
4544 * shrinks i_size, we put the inode on the orphan list and modify
4545 * i_disksize immediately, so that during the subsequent flushing of
4546 * dirty pages and freeing of disk blocks, we can guarantee that any
4547 * commit will leave the blocks being flushed in an unused state on
4548 * disk. (On recovery, the inode will get truncated and the blocks will
4549 * be freed, so we have a strong guarantee that no future commit will
4550 * leave these blocks visible to the user.)
4552 * Another thing we have to assure is that if we are in ordered mode
4553 * and inode is still attached to the committing transaction, we must
4554 * we start writeout of all the dirty pages which are being truncated.
4555 * This way we are sure that all the data written in the previous
4556 * transaction are already on disk (truncate waits for pages under
4559 * Called with inode->i_mutex down.
4561 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4563 struct inode
*inode
= dentry
->d_inode
;
4566 const unsigned int ia_valid
= attr
->ia_valid
;
4568 error
= inode_change_ok(inode
, attr
);
4572 if (is_quota_modification(inode
, attr
))
4573 dquot_initialize(inode
);
4574 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4575 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4578 /* (user+group)*(old+new) structure, inode write (sb,
4579 * inode block, ? - but truncate inode update has it) */
4580 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
4581 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
4582 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
4583 if (IS_ERR(handle
)) {
4584 error
= PTR_ERR(handle
);
4587 error
= dquot_transfer(inode
, attr
);
4589 ext4_journal_stop(handle
);
4592 /* Update corresponding info in inode so that everything is in
4593 * one transaction */
4594 if (attr
->ia_valid
& ATTR_UID
)
4595 inode
->i_uid
= attr
->ia_uid
;
4596 if (attr
->ia_valid
& ATTR_GID
)
4597 inode
->i_gid
= attr
->ia_gid
;
4598 error
= ext4_mark_inode_dirty(handle
, inode
);
4599 ext4_journal_stop(handle
);
4602 if (attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
!= inode
->i_size
) {
4604 loff_t oldsize
= inode
->i_size
;
4606 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4607 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4609 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4612 if (S_ISREG(inode
->i_mode
) &&
4613 (attr
->ia_size
< inode
->i_size
)) {
4614 if (ext4_should_order_data(inode
)) {
4615 error
= ext4_begin_ordered_truncate(inode
,
4620 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
4621 if (IS_ERR(handle
)) {
4622 error
= PTR_ERR(handle
);
4625 if (ext4_handle_valid(handle
)) {
4626 error
= ext4_orphan_add(handle
, inode
);
4629 down_write(&EXT4_I(inode
)->i_data_sem
);
4630 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4631 rc
= ext4_mark_inode_dirty(handle
, inode
);
4635 * We have to update i_size under i_data_sem together
4636 * with i_disksize to avoid races with writeback code
4637 * running ext4_wb_update_i_disksize().
4640 i_size_write(inode
, attr
->ia_size
);
4641 up_write(&EXT4_I(inode
)->i_data_sem
);
4642 ext4_journal_stop(handle
);
4644 ext4_orphan_del(NULL
, inode
);
4648 i_size_write(inode
, attr
->ia_size
);
4651 * Blocks are going to be removed from the inode. Wait
4652 * for dio in flight. Temporarily disable
4653 * dioread_nolock to prevent livelock.
4656 if (!ext4_should_journal_data(inode
)) {
4657 ext4_inode_block_unlocked_dio(inode
);
4658 inode_dio_wait(inode
);
4659 ext4_inode_resume_unlocked_dio(inode
);
4661 ext4_wait_for_tail_page_commit(inode
);
4664 * Truncate pagecache after we've waited for commit
4665 * in data=journal mode to make pages freeable.
4667 truncate_pagecache(inode
, oldsize
, inode
->i_size
);
4670 * We want to call ext4_truncate() even if attr->ia_size ==
4671 * inode->i_size for cases like truncation of fallocated space
4673 if (attr
->ia_valid
& ATTR_SIZE
)
4674 ext4_truncate(inode
);
4677 setattr_copy(inode
, attr
);
4678 mark_inode_dirty(inode
);
4682 * If the call to ext4_truncate failed to get a transaction handle at
4683 * all, we need to clean up the in-core orphan list manually.
4685 if (orphan
&& inode
->i_nlink
)
4686 ext4_orphan_del(NULL
, inode
);
4688 if (!rc
&& (ia_valid
& ATTR_MODE
))
4689 rc
= ext4_acl_chmod(inode
);
4692 ext4_std_error(inode
->i_sb
, error
);
4698 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4701 struct inode
*inode
;
4702 unsigned long long delalloc_blocks
;
4704 inode
= dentry
->d_inode
;
4705 generic_fillattr(inode
, stat
);
4708 * We can't update i_blocks if the block allocation is delayed
4709 * otherwise in the case of system crash before the real block
4710 * allocation is done, we will have i_blocks inconsistent with
4711 * on-disk file blocks.
4712 * We always keep i_blocks updated together with real
4713 * allocation. But to not confuse with user, stat
4714 * will return the blocks that include the delayed allocation
4715 * blocks for this file.
4717 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
4718 EXT4_I(inode
)->i_reserved_data_blocks
);
4720 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
-9);
4724 static int ext4_index_trans_blocks(struct inode
*inode
, int lblocks
,
4727 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4728 return ext4_ind_trans_blocks(inode
, lblocks
);
4729 return ext4_ext_index_trans_blocks(inode
, pextents
);
4733 * Account for index blocks, block groups bitmaps and block group
4734 * descriptor blocks if modify datablocks and index blocks
4735 * worse case, the indexs blocks spread over different block groups
4737 * If datablocks are discontiguous, they are possible to spread over
4738 * different block groups too. If they are contiguous, with flexbg,
4739 * they could still across block group boundary.
4741 * Also account for superblock, inode, quota and xattr blocks
4743 static int ext4_meta_trans_blocks(struct inode
*inode
, int lblocks
,
4746 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4752 * How many index blocks need to touch to map @lblocks logical blocks
4753 * to @pextents physical extents?
4755 idxblocks
= ext4_index_trans_blocks(inode
, lblocks
, pextents
);
4760 * Now let's see how many group bitmaps and group descriptors need
4763 groups
= idxblocks
+ pextents
;
4765 if (groups
> ngroups
)
4767 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4768 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4770 /* bitmaps and block group descriptor blocks */
4771 ret
+= groups
+ gdpblocks
;
4773 /* Blocks for super block, inode, quota and xattr blocks */
4774 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4780 * Calculate the total number of credits to reserve to fit
4781 * the modification of a single pages into a single transaction,
4782 * which may include multiple chunks of block allocations.
4784 * This could be called via ext4_write_begin()
4786 * We need to consider the worse case, when
4787 * one new block per extent.
4789 int ext4_writepage_trans_blocks(struct inode
*inode
)
4791 int bpp
= ext4_journal_blocks_per_page(inode
);
4794 ret
= ext4_meta_trans_blocks(inode
, bpp
, bpp
);
4796 /* Account for data blocks for journalled mode */
4797 if (ext4_should_journal_data(inode
))
4803 * Calculate the journal credits for a chunk of data modification.
4805 * This is called from DIO, fallocate or whoever calling
4806 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4808 * journal buffers for data blocks are not included here, as DIO
4809 * and fallocate do no need to journal data buffers.
4811 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4813 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4817 * The caller must have previously called ext4_reserve_inode_write().
4818 * Give this, we know that the caller already has write access to iloc->bh.
4820 int ext4_mark_iloc_dirty(handle_t
*handle
,
4821 struct inode
*inode
, struct ext4_iloc
*iloc
)
4825 if (IS_I_VERSION(inode
))
4826 inode_inc_iversion(inode
);
4828 /* the do_update_inode consumes one bh->b_count */
4831 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4832 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4838 * On success, We end up with an outstanding reference count against
4839 * iloc->bh. This _must_ be cleaned up later.
4843 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4844 struct ext4_iloc
*iloc
)
4848 err
= ext4_get_inode_loc(inode
, iloc
);
4850 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4851 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4857 ext4_std_error(inode
->i_sb
, err
);
4862 * Expand an inode by new_extra_isize bytes.
4863 * Returns 0 on success or negative error number on failure.
4865 static int ext4_expand_extra_isize(struct inode
*inode
,
4866 unsigned int new_extra_isize
,
4867 struct ext4_iloc iloc
,
4870 struct ext4_inode
*raw_inode
;
4871 struct ext4_xattr_ibody_header
*header
;
4873 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4876 raw_inode
= ext4_raw_inode(&iloc
);
4878 header
= IHDR(inode
, raw_inode
);
4880 /* No extended attributes present */
4881 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
4882 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4883 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4885 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4889 /* try to expand with EAs present */
4890 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4895 * What we do here is to mark the in-core inode as clean with respect to inode
4896 * dirtiness (it may still be data-dirty).
4897 * This means that the in-core inode may be reaped by prune_icache
4898 * without having to perform any I/O. This is a very good thing,
4899 * because *any* task may call prune_icache - even ones which
4900 * have a transaction open against a different journal.
4902 * Is this cheating? Not really. Sure, we haven't written the
4903 * inode out, but prune_icache isn't a user-visible syncing function.
4904 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4905 * we start and wait on commits.
4907 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4909 struct ext4_iloc iloc
;
4910 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4911 static unsigned int mnt_count
;
4915 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
4916 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4917 if (ext4_handle_valid(handle
) &&
4918 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4919 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
4921 * We need extra buffer credits since we may write into EA block
4922 * with this same handle. If journal_extend fails, then it will
4923 * only result in a minor loss of functionality for that inode.
4924 * If this is felt to be critical, then e2fsck should be run to
4925 * force a large enough s_min_extra_isize.
4927 if ((jbd2_journal_extend(handle
,
4928 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4929 ret
= ext4_expand_extra_isize(inode
,
4930 sbi
->s_want_extra_isize
,
4933 ext4_set_inode_state(inode
,
4934 EXT4_STATE_NO_EXPAND
);
4936 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4937 ext4_warning(inode
->i_sb
,
4938 "Unable to expand inode %lu. Delete"
4939 " some EAs or run e2fsck.",
4942 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4948 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4953 * ext4_dirty_inode() is called from __mark_inode_dirty()
4955 * We're really interested in the case where a file is being extended.
4956 * i_size has been changed by generic_commit_write() and we thus need
4957 * to include the updated inode in the current transaction.
4959 * Also, dquot_alloc_block() will always dirty the inode when blocks
4960 * are allocated to the file.
4962 * If the inode is marked synchronous, we don't honour that here - doing
4963 * so would cause a commit on atime updates, which we don't bother doing.
4964 * We handle synchronous inodes at the highest possible level.
4966 void ext4_dirty_inode(struct inode
*inode
, int flags
)
4970 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
4974 ext4_mark_inode_dirty(handle
, inode
);
4976 ext4_journal_stop(handle
);
4983 * Bind an inode's backing buffer_head into this transaction, to prevent
4984 * it from being flushed to disk early. Unlike
4985 * ext4_reserve_inode_write, this leaves behind no bh reference and
4986 * returns no iloc structure, so the caller needs to repeat the iloc
4987 * lookup to mark the inode dirty later.
4989 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4991 struct ext4_iloc iloc
;
4995 err
= ext4_get_inode_loc(inode
, &iloc
);
4997 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4998 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5000 err
= ext4_handle_dirty_metadata(handle
,
5006 ext4_std_error(inode
->i_sb
, err
);
5011 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5018 * We have to be very careful here: changing a data block's
5019 * journaling status dynamically is dangerous. If we write a
5020 * data block to the journal, change the status and then delete
5021 * that block, we risk forgetting to revoke the old log record
5022 * from the journal and so a subsequent replay can corrupt data.
5023 * So, first we make sure that the journal is empty and that
5024 * nobody is changing anything.
5027 journal
= EXT4_JOURNAL(inode
);
5030 if (is_journal_aborted(journal
))
5032 /* We have to allocate physical blocks for delalloc blocks
5033 * before flushing journal. otherwise delalloc blocks can not
5034 * be allocated any more. even more truncate on delalloc blocks
5035 * could trigger BUG by flushing delalloc blocks in journal.
5036 * There is no delalloc block in non-journal data mode.
5038 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
5039 err
= ext4_alloc_da_blocks(inode
);
5044 /* Wait for all existing dio workers */
5045 ext4_inode_block_unlocked_dio(inode
);
5046 inode_dio_wait(inode
);
5048 jbd2_journal_lock_updates(journal
);
5051 * OK, there are no updates running now, and all cached data is
5052 * synced to disk. We are now in a completely consistent state
5053 * which doesn't have anything in the journal, and we know that
5054 * no filesystem updates are running, so it is safe to modify
5055 * the inode's in-core data-journaling state flag now.
5059 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5061 jbd2_journal_flush(journal
);
5062 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5064 ext4_set_aops(inode
);
5066 jbd2_journal_unlock_updates(journal
);
5067 ext4_inode_resume_unlocked_dio(inode
);
5069 /* Finally we can mark the inode as dirty. */
5071 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5073 return PTR_ERR(handle
);
5075 err
= ext4_mark_inode_dirty(handle
, inode
);
5076 ext4_handle_sync(handle
);
5077 ext4_journal_stop(handle
);
5078 ext4_std_error(inode
->i_sb
, err
);
5083 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5085 return !buffer_mapped(bh
);
5088 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5090 struct page
*page
= vmf
->page
;
5094 struct file
*file
= vma
->vm_file
;
5095 struct inode
*inode
= file_inode(file
);
5096 struct address_space
*mapping
= inode
->i_mapping
;
5098 get_block_t
*get_block
;
5101 sb_start_pagefault(inode
->i_sb
);
5102 file_update_time(vma
->vm_file
);
5103 /* Delalloc case is easy... */
5104 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5105 !ext4_should_journal_data(inode
) &&
5106 !ext4_nonda_switch(inode
->i_sb
)) {
5108 ret
= __block_page_mkwrite(vma
, vmf
,
5109 ext4_da_get_block_prep
);
5110 } while (ret
== -ENOSPC
&&
5111 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5116 size
= i_size_read(inode
);
5117 /* Page got truncated from under us? */
5118 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5120 ret
= VM_FAULT_NOPAGE
;
5124 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5125 len
= size
& ~PAGE_CACHE_MASK
;
5127 len
= PAGE_CACHE_SIZE
;
5129 * Return if we have all the buffers mapped. This avoids the need to do
5130 * journal_start/journal_stop which can block and take a long time
5132 if (page_has_buffers(page
)) {
5133 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5135 ext4_bh_unmapped
)) {
5136 /* Wait so that we don't change page under IO */
5137 wait_for_stable_page(page
);
5138 ret
= VM_FAULT_LOCKED
;
5143 /* OK, we need to fill the hole... */
5144 if (ext4_should_dioread_nolock(inode
))
5145 get_block
= ext4_get_block_write
;
5147 get_block
= ext4_get_block
;
5149 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5150 ext4_writepage_trans_blocks(inode
));
5151 if (IS_ERR(handle
)) {
5152 ret
= VM_FAULT_SIGBUS
;
5155 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
5156 if (!ret
&& ext4_should_journal_data(inode
)) {
5157 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5158 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
5160 ret
= VM_FAULT_SIGBUS
;
5161 ext4_journal_stop(handle
);
5164 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5166 ext4_journal_stop(handle
);
5167 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5170 ret
= block_page_mkwrite_return(ret
);
5172 sb_end_pagefault(inode
->i_sb
);