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>
41 #include <linux/bitops.h>
43 #include "ext4_jbd2.h"
48 #include <trace/events/ext4.h>
49 #include <linux/blkdev.h>
51 #define MPAGE_DA_EXTENT_TAIL 0x01
53 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
54 struct ext4_inode_info
*ei
)
56 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
61 csum_lo
= le16_to_cpu(raw
->i_checksum_lo
);
62 raw
->i_checksum_lo
= 0;
63 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
64 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
65 csum_hi
= le16_to_cpu(raw
->i_checksum_hi
);
66 raw
->i_checksum_hi
= 0;
69 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
,
70 EXT4_INODE_SIZE(inode
->i_sb
));
72 raw
->i_checksum_lo
= cpu_to_le16(csum_lo
);
73 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
74 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
75 raw
->i_checksum_hi
= cpu_to_le16(csum_hi
);
80 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
81 struct ext4_inode_info
*ei
)
83 __u32 provided
, calculated
;
85 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
86 cpu_to_le32(EXT4_OS_LINUX
) ||
87 !EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
88 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
91 provided
= le16_to_cpu(raw
->i_checksum_lo
);
92 calculated
= ext4_inode_csum(inode
, raw
, ei
);
93 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
94 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
95 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
99 return provided
== calculated
;
102 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
103 struct ext4_inode_info
*ei
)
107 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
108 cpu_to_le32(EXT4_OS_LINUX
) ||
109 !EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
110 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
113 csum
= ext4_inode_csum(inode
, raw
, ei
);
114 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
115 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
116 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
117 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
120 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
123 trace_ext4_begin_ordered_truncate(inode
, new_size
);
125 * If jinode is zero, then we never opened the file for
126 * writing, so there's no need to call
127 * jbd2_journal_begin_ordered_truncate() since there's no
128 * outstanding writes we need to flush.
130 if (!EXT4_I(inode
)->jinode
)
132 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
133 EXT4_I(inode
)->jinode
,
137 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
138 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
139 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
140 static int ext4_discard_partial_page_buffers_no_lock(handle_t
*handle
,
141 struct inode
*inode
, struct page
*page
, loff_t from
,
142 loff_t length
, int flags
);
145 * Test whether an inode is a fast symlink.
147 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
149 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
150 (inode
->i_sb
->s_blocksize
>> 9) : 0;
152 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
156 * Restart the transaction associated with *handle. This does a commit,
157 * so before we call here everything must be consistently dirtied against
160 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
166 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
167 * moment, get_block can be called only for blocks inside i_size since
168 * page cache has been already dropped and writes are blocked by
169 * i_mutex. So we can safely drop the i_data_sem here.
171 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
172 jbd_debug(2, "restarting handle %p\n", handle
);
173 up_write(&EXT4_I(inode
)->i_data_sem
);
174 ret
= ext4_journal_restart(handle
, nblocks
);
175 down_write(&EXT4_I(inode
)->i_data_sem
);
176 ext4_discard_preallocations(inode
);
182 * Called at the last iput() if i_nlink is zero.
184 void ext4_evict_inode(struct inode
*inode
)
189 trace_ext4_evict_inode(inode
);
191 if (inode
->i_nlink
) {
193 * When journalling data dirty buffers are tracked only in the
194 * journal. So although mm thinks everything is clean and
195 * ready for reaping the inode might still have some pages to
196 * write in the running transaction or waiting to be
197 * checkpointed. Thus calling jbd2_journal_invalidatepage()
198 * (via truncate_inode_pages()) to discard these buffers can
199 * cause data loss. Also even if we did not discard these
200 * buffers, we would have no way to find them after the inode
201 * is reaped and thus user could see stale data if he tries to
202 * read them before the transaction is checkpointed. So be
203 * careful and force everything to disk here... We use
204 * ei->i_datasync_tid to store the newest transaction
205 * containing inode's data.
207 * Note that directories do not have this problem because they
208 * don't use page cache.
210 if (inode
->i_ino
!= EXT4_JOURNAL_INO
&&
211 ext4_should_journal_data(inode
) &&
212 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
))) {
213 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
214 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
216 jbd2_complete_transaction(journal
, commit_tid
);
217 filemap_write_and_wait(&inode
->i_data
);
219 truncate_inode_pages(&inode
->i_data
, 0);
220 ext4_ioend_shutdown(inode
);
224 if (!is_bad_inode(inode
))
225 dquot_initialize(inode
);
227 if (ext4_should_order_data(inode
))
228 ext4_begin_ordered_truncate(inode
, 0);
229 truncate_inode_pages(&inode
->i_data
, 0);
230 ext4_ioend_shutdown(inode
);
232 if (is_bad_inode(inode
))
236 * Protect us against freezing - iput() caller didn't have to have any
237 * protection against it
239 sb_start_intwrite(inode
->i_sb
);
240 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
,
241 ext4_blocks_for_truncate(inode
)+3);
242 if (IS_ERR(handle
)) {
243 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
245 * If we're going to skip the normal cleanup, we still need to
246 * make sure that the in-core orphan linked list is properly
249 ext4_orphan_del(NULL
, inode
);
250 sb_end_intwrite(inode
->i_sb
);
255 ext4_handle_sync(handle
);
257 err
= ext4_mark_inode_dirty(handle
, inode
);
259 ext4_warning(inode
->i_sb
,
260 "couldn't mark inode dirty (err %d)", err
);
264 ext4_truncate(inode
);
267 * ext4_ext_truncate() doesn't reserve any slop when it
268 * restarts journal transactions; therefore there may not be
269 * enough credits left in the handle to remove the inode from
270 * the orphan list and set the dtime field.
272 if (!ext4_handle_has_enough_credits(handle
, 3)) {
273 err
= ext4_journal_extend(handle
, 3);
275 err
= ext4_journal_restart(handle
, 3);
277 ext4_warning(inode
->i_sb
,
278 "couldn't extend journal (err %d)", err
);
280 ext4_journal_stop(handle
);
281 ext4_orphan_del(NULL
, inode
);
282 sb_end_intwrite(inode
->i_sb
);
288 * Kill off the orphan record which ext4_truncate created.
289 * AKPM: I think this can be inside the above `if'.
290 * Note that ext4_orphan_del() has to be able to cope with the
291 * deletion of a non-existent orphan - this is because we don't
292 * know if ext4_truncate() actually created an orphan record.
293 * (Well, we could do this if we need to, but heck - it works)
295 ext4_orphan_del(handle
, inode
);
296 EXT4_I(inode
)->i_dtime
= get_seconds();
299 * One subtle ordering requirement: if anything has gone wrong
300 * (transaction abort, IO errors, whatever), then we can still
301 * do these next steps (the fs will already have been marked as
302 * having errors), but we can't free the inode if the mark_dirty
305 if (ext4_mark_inode_dirty(handle
, inode
))
306 /* If that failed, just do the required in-core inode clear. */
307 ext4_clear_inode(inode
);
309 ext4_free_inode(handle
, inode
);
310 ext4_journal_stop(handle
);
311 sb_end_intwrite(inode
->i_sb
);
314 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
318 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
320 return &EXT4_I(inode
)->i_reserved_quota
;
325 * Calculate the number of metadata blocks need to reserve
326 * to allocate a block located at @lblock
328 static int ext4_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
330 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
331 return ext4_ext_calc_metadata_amount(inode
, lblock
);
333 return ext4_ind_calc_metadata_amount(inode
, lblock
);
337 * Called with i_data_sem down, which is important since we can call
338 * ext4_discard_preallocations() from here.
340 void ext4_da_update_reserve_space(struct inode
*inode
,
341 int used
, int quota_claim
)
343 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
344 struct ext4_inode_info
*ei
= EXT4_I(inode
);
346 spin_lock(&ei
->i_block_reservation_lock
);
347 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
348 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
349 ext4_warning(inode
->i_sb
, "%s: ino %lu, used %d "
350 "with only %d reserved data blocks",
351 __func__
, inode
->i_ino
, used
,
352 ei
->i_reserved_data_blocks
);
354 used
= ei
->i_reserved_data_blocks
;
357 if (unlikely(ei
->i_allocated_meta_blocks
> ei
->i_reserved_meta_blocks
)) {
358 ext4_warning(inode
->i_sb
, "ino %lu, allocated %d "
359 "with only %d reserved metadata blocks "
360 "(releasing %d blocks with reserved %d data blocks)",
361 inode
->i_ino
, ei
->i_allocated_meta_blocks
,
362 ei
->i_reserved_meta_blocks
, used
,
363 ei
->i_reserved_data_blocks
);
365 ei
->i_allocated_meta_blocks
= ei
->i_reserved_meta_blocks
;
368 /* Update per-inode reservations */
369 ei
->i_reserved_data_blocks
-= used
;
370 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
371 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
372 used
+ ei
->i_allocated_meta_blocks
);
373 ei
->i_allocated_meta_blocks
= 0;
375 if (ei
->i_reserved_data_blocks
== 0) {
377 * We can release all of the reserved metadata blocks
378 * only when we have written all of the delayed
381 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
382 ei
->i_reserved_meta_blocks
);
383 ei
->i_reserved_meta_blocks
= 0;
384 ei
->i_da_metadata_calc_len
= 0;
386 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
388 /* Update quota subsystem for data blocks */
390 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
393 * We did fallocate with an offset that is already delayed
394 * allocated. So on delayed allocated writeback we should
395 * not re-claim the quota for fallocated blocks.
397 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
401 * If we have done all the pending block allocations and if
402 * there aren't any writers on the inode, we can discard the
403 * inode's preallocations.
405 if ((ei
->i_reserved_data_blocks
== 0) &&
406 (atomic_read(&inode
->i_writecount
) == 0))
407 ext4_discard_preallocations(inode
);
410 static int __check_block_validity(struct inode
*inode
, const char *func
,
412 struct ext4_map_blocks
*map
)
414 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
416 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
417 "lblock %lu mapped to illegal pblock "
418 "(length %d)", (unsigned long) map
->m_lblk
,
425 #define check_block_validity(inode, map) \
426 __check_block_validity((inode), __func__, __LINE__, (map))
429 * Return the number of contiguous dirty pages in a given inode
430 * starting at page frame idx.
432 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
433 unsigned int max_pages
)
435 struct address_space
*mapping
= inode
->i_mapping
;
439 int i
, nr_pages
, done
= 0;
443 pagevec_init(&pvec
, 0);
446 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
448 (pgoff_t
)PAGEVEC_SIZE
);
451 for (i
= 0; i
< nr_pages
; i
++) {
452 struct page
*page
= pvec
.pages
[i
];
453 struct buffer_head
*bh
, *head
;
456 if (unlikely(page
->mapping
!= mapping
) ||
458 PageWriteback(page
) ||
459 page
->index
!= idx
) {
464 if (page_has_buffers(page
)) {
465 bh
= head
= page_buffers(page
);
467 if (!buffer_delay(bh
) &&
468 !buffer_unwritten(bh
))
470 bh
= bh
->b_this_page
;
471 } while (!done
&& (bh
!= head
));
478 if (num
>= max_pages
) {
483 pagevec_release(&pvec
);
488 #ifdef ES_AGGRESSIVE_TEST
489 static void ext4_map_blocks_es_recheck(handle_t
*handle
,
491 struct ext4_map_blocks
*es_map
,
492 struct ext4_map_blocks
*map
,
499 * There is a race window that the result is not the same.
500 * e.g. xfstests #223 when dioread_nolock enables. The reason
501 * is that we lookup a block mapping in extent status tree with
502 * out taking i_data_sem. So at the time the unwritten extent
503 * could be converted.
505 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
506 down_read((&EXT4_I(inode
)->i_data_sem
));
507 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
508 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
509 EXT4_GET_BLOCKS_KEEP_SIZE
);
511 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
512 EXT4_GET_BLOCKS_KEEP_SIZE
);
514 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
515 up_read((&EXT4_I(inode
)->i_data_sem
));
517 * Clear EXT4_MAP_FROM_CLUSTER and EXT4_MAP_BOUNDARY flag
518 * because it shouldn't be marked in es_map->m_flags.
520 map
->m_flags
&= ~(EXT4_MAP_FROM_CLUSTER
| EXT4_MAP_BOUNDARY
);
523 * We don't check m_len because extent will be collpased in status
524 * tree. So the m_len might not equal.
526 if (es_map
->m_lblk
!= map
->m_lblk
||
527 es_map
->m_flags
!= map
->m_flags
||
528 es_map
->m_pblk
!= map
->m_pblk
) {
529 printk("ES cache assertation failed for inode: %lu "
530 "es_cached ex [%d/%d/%llu/%x] != "
531 "found ex [%d/%d/%llu/%x] retval %d flags %x\n",
532 inode
->i_ino
, es_map
->m_lblk
, es_map
->m_len
,
533 es_map
->m_pblk
, es_map
->m_flags
, map
->m_lblk
,
534 map
->m_len
, map
->m_pblk
, map
->m_flags
,
538 #endif /* ES_AGGRESSIVE_TEST */
541 * The ext4_map_blocks() function tries to look up the requested blocks,
542 * and returns if the blocks are already mapped.
544 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
545 * and store the allocated blocks in the result buffer head and mark it
548 * If file type is extents based, it will call ext4_ext_map_blocks(),
549 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
552 * On success, it returns the number of blocks being mapped or allocate.
553 * if create==0 and the blocks are pre-allocated and uninitialized block,
554 * the result buffer head is unmapped. If the create ==1, it will make sure
555 * the buffer head is mapped.
557 * It returns 0 if plain look up failed (blocks have not been allocated), in
558 * that case, buffer head is unmapped
560 * It returns the error in case of allocation failure.
562 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
563 struct ext4_map_blocks
*map
, int flags
)
565 struct extent_status es
;
567 #ifdef ES_AGGRESSIVE_TEST
568 struct ext4_map_blocks orig_map
;
570 memcpy(&orig_map
, map
, sizeof(*map
));
574 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
575 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
576 (unsigned long) map
->m_lblk
);
578 /* Lookup extent status tree firstly */
579 if (ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
580 if (ext4_es_is_written(&es
) || ext4_es_is_unwritten(&es
)) {
581 map
->m_pblk
= ext4_es_pblock(&es
) +
582 map
->m_lblk
- es
.es_lblk
;
583 map
->m_flags
|= ext4_es_is_written(&es
) ?
584 EXT4_MAP_MAPPED
: EXT4_MAP_UNWRITTEN
;
585 retval
= es
.es_len
- (map
->m_lblk
- es
.es_lblk
);
586 if (retval
> map
->m_len
)
589 } else if (ext4_es_is_delayed(&es
) || ext4_es_is_hole(&es
)) {
594 #ifdef ES_AGGRESSIVE_TEST
595 ext4_map_blocks_es_recheck(handle
, inode
, map
,
602 * Try to see if we can get the block without requesting a new
605 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
606 down_read((&EXT4_I(inode
)->i_data_sem
));
607 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
608 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
609 EXT4_GET_BLOCKS_KEEP_SIZE
);
611 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
612 EXT4_GET_BLOCKS_KEEP_SIZE
);
616 unsigned long long status
;
618 #ifdef ES_AGGRESSIVE_TEST
619 if (retval
!= map
->m_len
) {
620 printk("ES len assertation failed for inode: %lu "
621 "retval %d != map->m_len %d "
622 "in %s (lookup)\n", inode
->i_ino
, retval
,
623 map
->m_len
, __func__
);
627 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
628 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
629 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
630 !(status
& EXTENT_STATUS_WRITTEN
) &&
631 ext4_find_delalloc_range(inode
, map
->m_lblk
,
632 map
->m_lblk
+ map
->m_len
- 1))
633 status
|= EXTENT_STATUS_DELAYED
;
634 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
,
635 map
->m_len
, map
->m_pblk
, status
);
639 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
640 up_read((&EXT4_I(inode
)->i_data_sem
));
643 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
644 int ret
= check_block_validity(inode
, map
);
649 /* If it is only a block(s) look up */
650 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
654 * Returns if the blocks have already allocated
656 * Note that if blocks have been preallocated
657 * ext4_ext_get_block() returns the create = 0
658 * with buffer head unmapped.
660 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
664 * Here we clear m_flags because after allocating an new extent,
665 * it will be set again.
667 map
->m_flags
&= ~EXT4_MAP_FLAGS
;
670 * New blocks allocate and/or writing to uninitialized extent
671 * will possibly result in updating i_data, so we take
672 * the write lock of i_data_sem, and call get_blocks()
673 * with create == 1 flag.
675 down_write((&EXT4_I(inode
)->i_data_sem
));
678 * if the caller is from delayed allocation writeout path
679 * we have already reserved fs blocks for allocation
680 * let the underlying get_block() function know to
681 * avoid double accounting
683 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
684 ext4_set_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
686 * We need to check for EXT4 here because migrate
687 * could have changed the inode type in between
689 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
690 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
692 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
694 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
696 * We allocated new blocks which will result in
697 * i_data's format changing. Force the migrate
698 * to fail by clearing migrate flags
700 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
704 * Update reserved blocks/metadata blocks after successful
705 * block allocation which had been deferred till now. We don't
706 * support fallocate for non extent files. So we can update
707 * reserve space here.
710 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
711 ext4_da_update_reserve_space(inode
, retval
, 1);
713 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
714 ext4_clear_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
718 unsigned long long status
;
720 #ifdef ES_AGGRESSIVE_TEST
721 if (retval
!= map
->m_len
) {
722 printk("ES len assertation failed for inode: %lu "
723 "retval %d != map->m_len %d "
724 "in %s (allocation)\n", inode
->i_ino
, retval
,
725 map
->m_len
, __func__
);
730 * If the extent has been zeroed out, we don't need to update
731 * extent status tree.
733 if ((flags
& EXT4_GET_BLOCKS_PRE_IO
) &&
734 ext4_es_lookup_extent(inode
, map
->m_lblk
, &es
)) {
735 if (ext4_es_is_written(&es
))
738 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
739 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
740 if (!(flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) &&
741 !(status
& EXTENT_STATUS_WRITTEN
) &&
742 ext4_find_delalloc_range(inode
, map
->m_lblk
,
743 map
->m_lblk
+ map
->m_len
- 1))
744 status
|= EXTENT_STATUS_DELAYED
;
745 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
746 map
->m_pblk
, status
);
752 up_write((&EXT4_I(inode
)->i_data_sem
));
753 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
754 int ret
= check_block_validity(inode
, map
);
761 /* Maximum number of blocks we map for direct IO at once. */
762 #define DIO_MAX_BLOCKS 4096
764 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
765 struct buffer_head
*bh
, int flags
)
767 handle_t
*handle
= ext4_journal_current_handle();
768 struct ext4_map_blocks map
;
769 int ret
= 0, started
= 0;
772 if (ext4_has_inline_data(inode
))
776 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
778 if (flags
&& !(flags
& EXT4_GET_BLOCKS_NO_LOCK
) && !handle
) {
779 /* Direct IO write... */
780 if (map
.m_len
> DIO_MAX_BLOCKS
)
781 map
.m_len
= DIO_MAX_BLOCKS
;
782 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
783 handle
= ext4_journal_start(inode
, EXT4_HT_MAP_BLOCKS
,
785 if (IS_ERR(handle
)) {
786 ret
= PTR_ERR(handle
);
792 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
794 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
795 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
796 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
800 ext4_journal_stop(handle
);
804 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
805 struct buffer_head
*bh
, int create
)
807 return _ext4_get_block(inode
, iblock
, bh
,
808 create
? EXT4_GET_BLOCKS_CREATE
: 0);
812 * `handle' can be NULL if create is zero
814 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
815 ext4_lblk_t block
, int create
, int *errp
)
817 struct ext4_map_blocks map
;
818 struct buffer_head
*bh
;
821 J_ASSERT(handle
!= NULL
|| create
== 0);
825 err
= ext4_map_blocks(handle
, inode
, &map
,
826 create
? EXT4_GET_BLOCKS_CREATE
: 0);
828 /* ensure we send some value back into *errp */
831 if (create
&& err
== 0)
832 err
= -ENOSPC
; /* should never happen */
838 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
843 if (map
.m_flags
& EXT4_MAP_NEW
) {
844 J_ASSERT(create
!= 0);
845 J_ASSERT(handle
!= NULL
);
848 * Now that we do not always journal data, we should
849 * keep in mind whether this should always journal the
850 * new buffer as metadata. For now, regular file
851 * writes use ext4_get_block instead, so it's not a
855 BUFFER_TRACE(bh
, "call get_create_access");
856 fatal
= ext4_journal_get_create_access(handle
, bh
);
857 if (!fatal
&& !buffer_uptodate(bh
)) {
858 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
859 set_buffer_uptodate(bh
);
862 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
863 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
867 BUFFER_TRACE(bh
, "not a new buffer");
877 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
878 ext4_lblk_t block
, int create
, int *err
)
880 struct buffer_head
*bh
;
882 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
885 if (buffer_uptodate(bh
))
887 ll_rw_block(READ
| REQ_META
| REQ_PRIO
, 1, &bh
);
889 if (buffer_uptodate(bh
))
896 int ext4_walk_page_buffers(handle_t
*handle
,
897 struct buffer_head
*head
,
901 int (*fn
)(handle_t
*handle
,
902 struct buffer_head
*bh
))
904 struct buffer_head
*bh
;
905 unsigned block_start
, block_end
;
906 unsigned blocksize
= head
->b_size
;
908 struct buffer_head
*next
;
910 for (bh
= head
, block_start
= 0;
911 ret
== 0 && (bh
!= head
|| !block_start
);
912 block_start
= block_end
, bh
= next
) {
913 next
= bh
->b_this_page
;
914 block_end
= block_start
+ blocksize
;
915 if (block_end
<= from
|| block_start
>= to
) {
916 if (partial
&& !buffer_uptodate(bh
))
920 err
= (*fn
)(handle
, bh
);
928 * To preserve ordering, it is essential that the hole instantiation and
929 * the data write be encapsulated in a single transaction. We cannot
930 * close off a transaction and start a new one between the ext4_get_block()
931 * and the commit_write(). So doing the jbd2_journal_start at the start of
932 * prepare_write() is the right place.
934 * Also, this function can nest inside ext4_writepage(). In that case, we
935 * *know* that ext4_writepage() has generated enough buffer credits to do the
936 * whole page. So we won't block on the journal in that case, which is good,
937 * because the caller may be PF_MEMALLOC.
939 * By accident, ext4 can be reentered when a transaction is open via
940 * quota file writes. If we were to commit the transaction while thus
941 * reentered, there can be a deadlock - we would be holding a quota
942 * lock, and the commit would never complete if another thread had a
943 * transaction open and was blocking on the quota lock - a ranking
946 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
947 * will _not_ run commit under these circumstances because handle->h_ref
948 * is elevated. We'll still have enough credits for the tiny quotafile
951 int do_journal_get_write_access(handle_t
*handle
,
952 struct buffer_head
*bh
)
954 int dirty
= buffer_dirty(bh
);
957 if (!buffer_mapped(bh
) || buffer_freed(bh
))
960 * __block_write_begin() could have dirtied some buffers. Clean
961 * the dirty bit as jbd2_journal_get_write_access() could complain
962 * otherwise about fs integrity issues. Setting of the dirty bit
963 * by __block_write_begin() isn't a real problem here as we clear
964 * the bit before releasing a page lock and thus writeback cannot
965 * ever write the buffer.
968 clear_buffer_dirty(bh
);
969 ret
= ext4_journal_get_write_access(handle
, bh
);
971 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
975 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
976 struct buffer_head
*bh_result
, int create
);
977 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
978 loff_t pos
, unsigned len
, unsigned flags
,
979 struct page
**pagep
, void **fsdata
)
981 struct inode
*inode
= mapping
->host
;
982 int ret
, needed_blocks
;
988 #if defined(FEATURE_STORAGE_PID_LOGGER)
989 extern unsigned char *page_logger
;
990 struct page_pid_logger
*tmp_logger
;
991 unsigned long page_index
;
992 extern spinlock_t g_locker
;
993 unsigned long g_flags
;
996 trace_ext4_write_begin(inode
, pos
, len
, flags
);
998 * Reserve one block more for addition to orphan list in case
999 * we allocate blocks but write fails for some reason
1001 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
1002 index
= pos
>> PAGE_CACHE_SHIFT
;
1003 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1006 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
1007 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
1016 * grab_cache_page_write_begin() can take a long time if the
1017 * system is thrashing due to memory pressure, or if the page
1018 * is being written back. So grab it first before we start
1019 * the transaction handle. This also allows us to allocate
1020 * the page (if needed) without using GFP_NOFS.
1023 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
1029 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
, needed_blocks
);
1030 if (IS_ERR(handle
)) {
1031 page_cache_release(page
);
1032 return PTR_ERR(handle
);
1036 if (page
->mapping
!= mapping
) {
1037 /* The page got truncated from under us */
1039 page_cache_release(page
);
1040 ext4_journal_stop(handle
);
1043 /* In case writeback began while the page was unlocked */
1044 wait_for_stable_page(page
);
1046 if (ext4_should_dioread_nolock(inode
))
1047 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
1049 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
1051 if (!ret
&& ext4_should_journal_data(inode
)) {
1052 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
1054 do_journal_get_write_access
);
1060 * __block_write_begin may have instantiated a few blocks
1061 * outside i_size. Trim these off again. Don't need
1062 * i_size_read because we hold i_mutex.
1064 * Add inode to orphan list in case we crash before
1067 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1068 ext4_orphan_add(handle
, inode
);
1070 ext4_journal_stop(handle
);
1071 if (pos
+ len
> inode
->i_size
) {
1072 ext4_truncate_failed_write(inode
);
1074 * If truncate failed early the inode might
1075 * still be on the orphan list; we need to
1076 * make sure the inode is removed from the
1077 * orphan list in that case.
1080 ext4_orphan_del(NULL
, inode
);
1083 if (ret
== -ENOSPC
&&
1084 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
1086 page_cache_release(page
);
1090 #if defined(FEATURE_STORAGE_PID_LOGGER)
1091 if( page_logger
&& (*pagep
)) {
1092 //#if defined(CONFIG_FLATMEM)
1093 //page_index = (unsigned long)((*pagep) - mem_map) ;
1095 page_index
= (unsigned long)(__page_to_pfn(*pagep
))- PHYS_PFN_OFFSET
;
1097 tmp_logger
=((struct page_pid_logger
*)page_logger
) + page_index
;
1098 spin_lock_irqsave(&g_locker
, g_flags
);
1099 if( page_index
< num_physpages
) {
1100 if( tmp_logger
->pid1
== 0XFFFF)
1101 tmp_logger
->pid1
= current
->pid
;
1102 else if( tmp_logger
->pid1
!= current
->pid
)
1103 tmp_logger
->pid2
= current
->pid
;
1105 spin_unlock_irqrestore(&g_locker
, g_flags
);
1111 /* For write_end() in data=journal mode */
1112 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
1115 if (!buffer_mapped(bh
) || buffer_freed(bh
))
1117 set_buffer_uptodate(bh
);
1118 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
1119 clear_buffer_meta(bh
);
1120 clear_buffer_prio(bh
);
1125 * We need to pick up the new inode size which generic_commit_write gave us
1126 * `file' can be NULL - eg, when called from page_symlink().
1128 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1129 * buffers are managed internally.
1131 static int ext4_write_end(struct file
*file
,
1132 struct address_space
*mapping
,
1133 loff_t pos
, unsigned len
, unsigned copied
,
1134 struct page
*page
, void *fsdata
)
1136 handle_t
*handle
= ext4_journal_current_handle();
1137 struct inode
*inode
= mapping
->host
;
1139 int i_size_changed
= 0;
1141 trace_ext4_write_end(inode
, pos
, len
, copied
);
1142 if (ext4_test_inode_state(inode
, EXT4_STATE_ORDERED_MODE
)) {
1143 ret
= ext4_jbd2_file_inode(handle
, inode
);
1146 page_cache_release(page
);
1151 if (ext4_has_inline_data(inode
)) {
1152 ret
= ext4_write_inline_data_end(inode
, pos
, len
,
1158 copied
= block_write_end(file
, mapping
, pos
,
1159 len
, copied
, page
, fsdata
);
1162 * No need to use i_size_read() here, the i_size
1163 * cannot change under us because we hole i_mutex.
1165 * But it's important to update i_size while still holding page lock:
1166 * page writeout could otherwise come in and zero beyond i_size.
1168 if (pos
+ copied
> inode
->i_size
) {
1169 i_size_write(inode
, pos
+ copied
);
1173 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
1174 /* We need to mark inode dirty even if
1175 * new_i_size is less that inode->i_size
1176 * but greater than i_disksize. (hint delalloc)
1178 ext4_update_i_disksize(inode
, (pos
+ copied
));
1182 page_cache_release(page
);
1185 * Don't mark the inode dirty under page lock. First, it unnecessarily
1186 * makes the holding time of page lock longer. Second, it forces lock
1187 * ordering of page lock and transaction start for journaling
1191 ext4_mark_inode_dirty(handle
, inode
);
1195 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1196 /* if we have allocated more blocks and copied
1197 * less. We will have blocks allocated outside
1198 * inode->i_size. So truncate them
1200 ext4_orphan_add(handle
, inode
);
1202 ret2
= ext4_journal_stop(handle
);
1206 if (pos
+ len
> inode
->i_size
) {
1207 ext4_truncate_failed_write(inode
);
1209 * If truncate failed early the inode might still be
1210 * on the orphan list; we need to make sure the inode
1211 * is removed from the orphan list in that case.
1214 ext4_orphan_del(NULL
, inode
);
1217 return ret
? ret
: copied
;
1220 static int ext4_journalled_write_end(struct file
*file
,
1221 struct address_space
*mapping
,
1222 loff_t pos
, unsigned len
, unsigned copied
,
1223 struct page
*page
, void *fsdata
)
1225 handle_t
*handle
= ext4_journal_current_handle();
1226 struct inode
*inode
= mapping
->host
;
1232 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1233 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1236 BUG_ON(!ext4_handle_valid(handle
));
1238 if (ext4_has_inline_data(inode
))
1239 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1243 if (!PageUptodate(page
))
1245 page_zero_new_buffers(page
, from
+copied
, to
);
1248 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1249 to
, &partial
, write_end_fn
);
1251 SetPageUptodate(page
);
1253 new_i_size
= pos
+ copied
;
1254 if (new_i_size
> inode
->i_size
)
1255 i_size_write(inode
, pos
+copied
);
1256 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1257 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1258 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1259 ext4_update_i_disksize(inode
, new_i_size
);
1260 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1266 page_cache_release(page
);
1267 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1268 /* if we have allocated more blocks and copied
1269 * less. We will have blocks allocated outside
1270 * inode->i_size. So truncate them
1272 ext4_orphan_add(handle
, inode
);
1274 ret2
= ext4_journal_stop(handle
);
1277 if (pos
+ len
> inode
->i_size
) {
1278 ext4_truncate_failed_write(inode
);
1280 * If truncate failed early the inode might still be
1281 * on the orphan list; we need to make sure the inode
1282 * is removed from the orphan list in that case.
1285 ext4_orphan_del(NULL
, inode
);
1288 return ret
? ret
: copied
;
1292 * Reserve a metadata for a single block located at lblock
1294 static int ext4_da_reserve_metadata(struct inode
*inode
, ext4_lblk_t lblock
)
1296 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1297 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1298 unsigned int md_needed
;
1299 ext4_lblk_t save_last_lblock
;
1303 * recalculate the amount of metadata blocks to reserve
1304 * in order to allocate nrblocks
1305 * worse case is one extent per block
1307 spin_lock(&ei
->i_block_reservation_lock
);
1309 * ext4_calc_metadata_amount() has side effects, which we have
1310 * to be prepared undo if we fail to claim space.
1312 save_len
= ei
->i_da_metadata_calc_len
;
1313 save_last_lblock
= ei
->i_da_metadata_calc_last_lblock
;
1314 md_needed
= EXT4_NUM_B2C(sbi
,
1315 ext4_calc_metadata_amount(inode
, lblock
));
1316 trace_ext4_da_reserve_space(inode
, md_needed
);
1319 * We do still charge estimated metadata to the sb though;
1320 * we cannot afford to run out of free blocks.
1322 if (ext4_claim_free_clusters(sbi
, md_needed
, 0)) {
1323 ei
->i_da_metadata_calc_len
= save_len
;
1324 ei
->i_da_metadata_calc_last_lblock
= save_last_lblock
;
1325 spin_unlock(&ei
->i_block_reservation_lock
);
1328 ei
->i_reserved_meta_blocks
+= md_needed
;
1329 spin_unlock(&ei
->i_block_reservation_lock
);
1331 return 0; /* success */
1335 * Reserve a single cluster located at lblock
1337 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1339 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1340 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1341 unsigned int md_needed
;
1343 ext4_lblk_t save_last_lblock
;
1347 * We will charge metadata quota at writeout time; this saves
1348 * us from metadata over-estimation, though we may go over by
1349 * a small amount in the end. Here we just reserve for data.
1351 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1356 * recalculate the amount of metadata blocks to reserve
1357 * in order to allocate nrblocks
1358 * worse case is one extent per block
1360 spin_lock(&ei
->i_block_reservation_lock
);
1362 * ext4_calc_metadata_amount() has side effects, which we have
1363 * to be prepared undo if we fail to claim space.
1365 save_len
= ei
->i_da_metadata_calc_len
;
1366 save_last_lblock
= ei
->i_da_metadata_calc_last_lblock
;
1367 md_needed
= EXT4_NUM_B2C(sbi
,
1368 ext4_calc_metadata_amount(inode
, lblock
));
1369 trace_ext4_da_reserve_space(inode
, md_needed
);
1372 * We do still charge estimated metadata to the sb though;
1373 * we cannot afford to run out of free blocks.
1375 if (ext4_claim_free_clusters(sbi
, md_needed
+ 1, 0)) {
1376 ei
->i_da_metadata_calc_len
= save_len
;
1377 ei
->i_da_metadata_calc_last_lblock
= save_last_lblock
;
1378 spin_unlock(&ei
->i_block_reservation_lock
);
1379 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1382 ei
->i_reserved_data_blocks
++;
1383 ei
->i_reserved_meta_blocks
+= md_needed
;
1384 spin_unlock(&ei
->i_block_reservation_lock
);
1386 return 0; /* success */
1389 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1391 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1392 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1395 return; /* Nothing to release, exit */
1397 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1399 trace_ext4_da_release_space(inode
, to_free
);
1400 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1402 * if there aren't enough reserved blocks, then the
1403 * counter is messed up somewhere. Since this
1404 * function is called from invalidate page, it's
1405 * harmless to return without any action.
1407 ext4_warning(inode
->i_sb
, "ext4_da_release_space: "
1408 "ino %lu, to_free %d with only %d reserved "
1409 "data blocks", inode
->i_ino
, to_free
,
1410 ei
->i_reserved_data_blocks
);
1412 to_free
= ei
->i_reserved_data_blocks
;
1414 ei
->i_reserved_data_blocks
-= to_free
;
1416 if (ei
->i_reserved_data_blocks
== 0) {
1418 * We can release all of the reserved metadata blocks
1419 * only when we have written all of the delayed
1420 * allocation blocks.
1421 * Note that in case of bigalloc, i_reserved_meta_blocks,
1422 * i_reserved_data_blocks, etc. refer to number of clusters.
1424 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
1425 ei
->i_reserved_meta_blocks
);
1426 ei
->i_reserved_meta_blocks
= 0;
1427 ei
->i_da_metadata_calc_len
= 0;
1430 /* update fs dirty data blocks counter */
1431 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1433 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1435 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1438 static void ext4_da_page_release_reservation(struct page
*page
,
1439 unsigned long offset
)
1441 int to_release
= 0, contiguous_blks
= 0;
1442 struct buffer_head
*head
, *bh
;
1443 unsigned int curr_off
= 0;
1444 struct inode
*inode
= page
->mapping
->host
;
1445 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1449 head
= page_buffers(page
);
1452 unsigned int next_off
= curr_off
+ bh
->b_size
;
1454 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1457 clear_buffer_delay(bh
);
1458 } else if (contiguous_blks
) {
1459 lblk
= page
->index
<<
1460 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1461 lblk
+= (curr_off
>> inode
->i_blkbits
) -
1463 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1464 contiguous_blks
= 0;
1466 curr_off
= next_off
;
1467 } while ((bh
= bh
->b_this_page
) != head
);
1469 if (contiguous_blks
) {
1470 lblk
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1471 lblk
+= (curr_off
>> inode
->i_blkbits
) - contiguous_blks
;
1472 ext4_es_remove_extent(inode
, lblk
, contiguous_blks
);
1475 /* If we have released all the blocks belonging to a cluster, then we
1476 * need to release the reserved space for that cluster. */
1477 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1478 while (num_clusters
> 0) {
1479 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1480 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1481 if (sbi
->s_cluster_ratio
== 1 ||
1482 !ext4_find_delalloc_cluster(inode
, lblk
))
1483 ext4_da_release_space(inode
, 1);
1490 * Delayed allocation stuff
1494 * mpage_da_submit_io - walks through extent of pages and try to write
1495 * them with writepage() call back
1497 * @mpd->inode: inode
1498 * @mpd->first_page: first page of the extent
1499 * @mpd->next_page: page after the last page of the extent
1501 * By the time mpage_da_submit_io() is called we expect all blocks
1502 * to be allocated. this may be wrong if allocation failed.
1504 * As pages are already locked by write_cache_pages(), we can't use it
1506 static int mpage_da_submit_io(struct mpage_da_data
*mpd
,
1507 struct ext4_map_blocks
*map
)
1509 struct pagevec pvec
;
1510 unsigned long index
, end
;
1511 int ret
= 0, err
, nr_pages
, i
;
1512 struct inode
*inode
= mpd
->inode
;
1513 struct address_space
*mapping
= inode
->i_mapping
;
1514 loff_t size
= i_size_read(inode
);
1515 unsigned int len
, block_start
;
1516 struct buffer_head
*bh
, *page_bufs
= NULL
;
1517 sector_t pblock
= 0, cur_logical
= 0;
1518 struct ext4_io_submit io_submit
;
1520 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1521 memset(&io_submit
, 0, sizeof(io_submit
));
1523 * We need to start from the first_page to the next_page - 1
1524 * to make sure we also write the mapped dirty buffer_heads.
1525 * If we look at mpd->b_blocknr we would only be looking
1526 * at the currently mapped buffer_heads.
1528 index
= mpd
->first_page
;
1529 end
= mpd
->next_page
- 1;
1531 pagevec_init(&pvec
, 0);
1532 while (index
<= end
) {
1533 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1536 for (i
= 0; i
< nr_pages
; i
++) {
1538 struct page
*page
= pvec
.pages
[i
];
1540 index
= page
->index
;
1544 if (index
== size
>> PAGE_CACHE_SHIFT
)
1545 len
= size
& ~PAGE_CACHE_MASK
;
1547 len
= PAGE_CACHE_SIZE
;
1549 cur_logical
= index
<< (PAGE_CACHE_SHIFT
-
1551 pblock
= map
->m_pblk
+ (cur_logical
-
1556 BUG_ON(!PageLocked(page
));
1557 BUG_ON(PageWriteback(page
));
1559 bh
= page_bufs
= page_buffers(page
);
1562 if (map
&& (cur_logical
>= map
->m_lblk
) &&
1563 (cur_logical
<= (map
->m_lblk
+
1564 (map
->m_len
- 1)))) {
1565 if (buffer_delay(bh
)) {
1566 clear_buffer_delay(bh
);
1567 bh
->b_blocknr
= pblock
;
1569 if (buffer_unwritten(bh
) ||
1571 BUG_ON(bh
->b_blocknr
!= pblock
);
1572 if (map
->m_flags
& EXT4_MAP_UNINIT
)
1573 set_buffer_uninit(bh
);
1574 clear_buffer_unwritten(bh
);
1578 * skip page if block allocation undone and
1581 if (ext4_bh_delay_or_unwritten(NULL
, bh
))
1583 bh
= bh
->b_this_page
;
1584 block_start
+= bh
->b_size
;
1587 } while (bh
!= page_bufs
);
1594 clear_page_dirty_for_io(page
);
1595 err
= ext4_bio_write_page(&io_submit
, page
, len
,
1598 mpd
->pages_written
++;
1600 * In error case, we have to continue because
1601 * remaining pages are still locked
1606 pagevec_release(&pvec
);
1608 ext4_io_submit(&io_submit
);
1612 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
)
1616 struct pagevec pvec
;
1617 struct inode
*inode
= mpd
->inode
;
1618 struct address_space
*mapping
= inode
->i_mapping
;
1619 ext4_lblk_t start
, last
;
1621 index
= mpd
->first_page
;
1622 end
= mpd
->next_page
- 1;
1624 start
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1625 last
= end
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1626 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1628 pagevec_init(&pvec
, 0);
1629 while (index
<= end
) {
1630 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1633 for (i
= 0; i
< nr_pages
; i
++) {
1634 struct page
*page
= pvec
.pages
[i
];
1635 if (page
->index
> end
)
1637 BUG_ON(!PageLocked(page
));
1638 BUG_ON(PageWriteback(page
));
1639 block_invalidatepage(page
, 0);
1640 ClearPageUptodate(page
);
1643 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1644 pagevec_release(&pvec
);
1649 static void ext4_print_free_blocks(struct inode
*inode
)
1651 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1652 struct super_block
*sb
= inode
->i_sb
;
1653 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1655 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1656 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1657 ext4_count_free_clusters(sb
)));
1658 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1659 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1660 (long long) EXT4_C2B(EXT4_SB(sb
),
1661 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1662 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1663 (long long) EXT4_C2B(EXT4_SB(sb
),
1664 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1665 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1666 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1667 ei
->i_reserved_data_blocks
);
1668 ext4_msg(sb
, KERN_CRIT
, "i_reserved_meta_blocks=%u",
1669 ei
->i_reserved_meta_blocks
);
1670 ext4_msg(sb
, KERN_CRIT
, "i_allocated_meta_blocks=%u",
1671 ei
->i_allocated_meta_blocks
);
1676 * mpage_da_map_and_submit - go through given space, map them
1677 * if necessary, and then submit them for I/O
1679 * @mpd - bh describing space
1681 * The function skips space we know is already mapped to disk blocks.
1684 static void mpage_da_map_and_submit(struct mpage_da_data
*mpd
)
1686 int err
, blks
, get_blocks_flags
;
1687 struct ext4_map_blocks map
, *mapp
= NULL
;
1688 sector_t next
= mpd
->b_blocknr
;
1689 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1690 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
1691 handle_t
*handle
= NULL
;
1694 * If the blocks are mapped already, or we couldn't accumulate
1695 * any blocks, then proceed immediately to the submission stage.
1697 if ((mpd
->b_size
== 0) ||
1698 ((mpd
->b_state
& (1 << BH_Mapped
)) &&
1699 !(mpd
->b_state
& (1 << BH_Delay
)) &&
1700 !(mpd
->b_state
& (1 << BH_Unwritten
))))
1703 handle
= ext4_journal_current_handle();
1707 * Call ext4_map_blocks() to allocate any delayed allocation
1708 * blocks, or to convert an uninitialized extent to be
1709 * initialized (in the case where we have written into
1710 * one or more preallocated blocks).
1712 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1713 * indicate that we are on the delayed allocation path. This
1714 * affects functions in many different parts of the allocation
1715 * call path. This flag exists primarily because we don't
1716 * want to change *many* call functions, so ext4_map_blocks()
1717 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1718 * inode's allocation semaphore is taken.
1720 * If the blocks in questions were delalloc blocks, set
1721 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1722 * variables are updated after the blocks have been allocated.
1725 map
.m_len
= max_blocks
;
1727 * We're in delalloc path and it is possible that we're going to
1728 * need more metadata blocks than previously reserved. However
1729 * we must not fail because we're in writeback and there is
1730 * nothing we can do about it so it might result in data loss.
1731 * So use reserved blocks to allocate metadata if possible.
1733 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
1734 EXT4_GET_BLOCKS_METADATA_NOFAIL
;
1735 if (ext4_should_dioread_nolock(mpd
->inode
))
1736 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
1737 if (mpd
->b_state
& (1 << BH_Delay
))
1738 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
1741 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
1743 struct super_block
*sb
= mpd
->inode
->i_sb
;
1747 * If get block returns EAGAIN or ENOSPC and there
1748 * appears to be free blocks we will just let
1749 * mpage_da_submit_io() unlock all of the pages.
1754 if (err
== -ENOSPC
&& ext4_count_free_clusters(sb
)) {
1760 * get block failure will cause us to loop in
1761 * writepages, because a_ops->writepage won't be able
1762 * to make progress. The page will be redirtied by
1763 * writepage and writepages will again try to write
1766 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
1767 ext4_msg(sb
, KERN_CRIT
,
1768 "delayed block allocation failed for inode %lu "
1769 "at logical offset %llu with max blocks %zd "
1770 "with error %d", mpd
->inode
->i_ino
,
1771 (unsigned long long) next
,
1772 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1773 ext4_msg(sb
, KERN_CRIT
,
1774 "This should not happen!! Data will be lost");
1776 ext4_print_free_blocks(mpd
->inode
);
1778 /* invalidate all the pages */
1779 ext4_da_block_invalidatepages(mpd
);
1781 /* Mark this page range as having been completed */
1788 if (map
.m_flags
& EXT4_MAP_NEW
) {
1789 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
1792 for (i
= 0; i
< map
.m_len
; i
++)
1793 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
1797 * Update on-disk size along with block allocation.
1799 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
1800 if (disksize
> i_size_read(mpd
->inode
))
1801 disksize
= i_size_read(mpd
->inode
);
1802 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
1803 ext4_update_i_disksize(mpd
->inode
, disksize
);
1804 err
= ext4_mark_inode_dirty(handle
, mpd
->inode
);
1806 ext4_error(mpd
->inode
->i_sb
,
1807 "Failed to mark inode %lu dirty",
1812 mpage_da_submit_io(mpd
, mapp
);
1816 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1817 (1 << BH_Delay) | (1 << BH_Unwritten))
1820 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1822 * @mpd->lbh - extent of blocks
1823 * @logical - logical number of the block in the file
1824 * @b_state - b_state of the buffer head added
1826 * the function is used to collect contig. blocks in same state
1828 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, sector_t logical
,
1829 unsigned long b_state
)
1832 int blkbits
= mpd
->inode
->i_blkbits
;
1833 int nrblocks
= mpd
->b_size
>> blkbits
;
1836 * XXX Don't go larger than mballoc is willing to allocate
1837 * This is a stopgap solution. We eventually need to fold
1838 * mpage_da_submit_io() into this function and then call
1839 * ext4_map_blocks() multiple times in a loop
1841 if (nrblocks
>= (8*1024*1024 >> blkbits
))
1844 /* check if the reserved journal credits might overflow */
1845 if (!ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
)) {
1846 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1848 * With non-extent format we are limited by the journal
1849 * credit available. Total credit needed to insert
1850 * nrblocks contiguous blocks is dependent on the
1851 * nrblocks. So limit nrblocks.
1857 * First block in the extent
1859 if (mpd
->b_size
== 0) {
1860 mpd
->b_blocknr
= logical
;
1861 mpd
->b_size
= 1 << blkbits
;
1862 mpd
->b_state
= b_state
& BH_FLAGS
;
1866 next
= mpd
->b_blocknr
+ nrblocks
;
1868 * Can we merge the block to our big extent?
1870 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
1871 mpd
->b_size
+= 1 << blkbits
;
1877 * We couldn't merge the block to our extent, so we
1878 * need to flush current extent and start new one
1880 mpage_da_map_and_submit(mpd
);
1884 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1886 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1890 * This function is grabs code from the very beginning of
1891 * ext4_map_blocks, but assumes that the caller is from delayed write
1892 * time. This function looks up the requested blocks and sets the
1893 * buffer delay bit under the protection of i_data_sem.
1895 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1896 struct ext4_map_blocks
*map
,
1897 struct buffer_head
*bh
)
1899 struct extent_status es
;
1901 sector_t invalid_block
= ~((sector_t
) 0xffff);
1902 #ifdef ES_AGGRESSIVE_TEST
1903 struct ext4_map_blocks orig_map
;
1905 memcpy(&orig_map
, map
, sizeof(*map
));
1908 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1912 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1913 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1914 (unsigned long) map
->m_lblk
);
1916 /* Lookup extent status tree firstly */
1917 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1919 if (ext4_es_is_hole(&es
)) {
1921 down_read((&EXT4_I(inode
)->i_data_sem
));
1926 * Delayed extent could be allocated by fallocate.
1927 * So we need to check it.
1929 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1930 map_bh(bh
, inode
->i_sb
, invalid_block
);
1932 set_buffer_delay(bh
);
1936 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1937 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1938 if (retval
> map
->m_len
)
1939 retval
= map
->m_len
;
1940 map
->m_len
= retval
;
1941 if (ext4_es_is_written(&es
))
1942 map
->m_flags
|= EXT4_MAP_MAPPED
;
1943 else if (ext4_es_is_unwritten(&es
))
1944 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1948 #ifdef ES_AGGRESSIVE_TEST
1949 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1955 * Try to see if we can get the block without requesting a new
1956 * file system block.
1958 down_read((&EXT4_I(inode
)->i_data_sem
));
1959 if (ext4_has_inline_data(inode
)) {
1961 * We will soon create blocks for this page, and let
1962 * us pretend as if the blocks aren't allocated yet.
1963 * In case of clusters, we have to handle the work
1964 * of mapping from cluster so that the reserved space
1965 * is calculated properly.
1967 if ((EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) &&
1968 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
1969 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
1971 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1972 retval
= ext4_ext_map_blocks(NULL
, inode
, map
,
1973 EXT4_GET_BLOCKS_NO_PUT_HOLE
);
1975 retval
= ext4_ind_map_blocks(NULL
, inode
, map
,
1976 EXT4_GET_BLOCKS_NO_PUT_HOLE
);
1982 * XXX: __block_prepare_write() unmaps passed block,
1986 * If the block was allocated from previously allocated cluster,
1987 * then we don't need to reserve it again. However we still need
1988 * to reserve metadata for every block we're going to write.
1990 if (!(map
->m_flags
& EXT4_MAP_FROM_CLUSTER
)) {
1991 ret
= ext4_da_reserve_space(inode
, iblock
);
1993 /* not enough space to reserve */
1998 ret
= ext4_da_reserve_metadata(inode
, iblock
);
2000 /* not enough space to reserve */
2006 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
2007 ~0, EXTENT_STATUS_DELAYED
);
2013 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
2014 * and it should not appear on the bh->b_state.
2016 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
2018 map_bh(bh
, inode
->i_sb
, invalid_block
);
2020 set_buffer_delay(bh
);
2021 } else if (retval
> 0) {
2023 unsigned long long status
;
2025 #ifdef ES_AGGRESSIVE_TEST
2026 if (retval
!= map
->m_len
) {
2027 printk("ES len assertation failed for inode: %lu "
2028 "retval %d != map->m_len %d "
2029 "in %s (lookup)\n", inode
->i_ino
, retval
,
2030 map
->m_len
, __func__
);
2034 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
2035 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
2036 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
2037 map
->m_pblk
, status
);
2043 up_read((&EXT4_I(inode
)->i_data_sem
));
2049 * This is a special get_blocks_t callback which is used by
2050 * ext4_da_write_begin(). It will either return mapped block or
2051 * reserve space for a single block.
2053 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2054 * We also have b_blocknr = -1 and b_bdev initialized properly
2056 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2057 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2058 * initialized properly.
2060 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2061 struct buffer_head
*bh
, int create
)
2063 struct ext4_map_blocks map
;
2066 BUG_ON(create
== 0);
2067 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
2069 map
.m_lblk
= iblock
;
2073 * first, we need to know whether the block is allocated already
2074 * preallocated blocks are unmapped but should treated
2075 * the same as allocated blocks.
2077 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
2081 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
2082 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
2084 if (buffer_unwritten(bh
)) {
2085 /* A delayed write to unwritten bh should be marked
2086 * new and mapped. Mapped ensures that we don't do
2087 * get_block multiple times when we write to the same
2088 * offset and new ensures that we do proper zero out
2089 * for partial write.
2092 set_buffer_mapped(bh
);
2097 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2103 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2109 static int __ext4_journalled_writepage(struct page
*page
,
2112 struct address_space
*mapping
= page
->mapping
;
2113 struct inode
*inode
= mapping
->host
;
2114 struct buffer_head
*page_bufs
= NULL
;
2115 handle_t
*handle
= NULL
;
2116 int ret
= 0, err
= 0;
2117 int inline_data
= ext4_has_inline_data(inode
);
2118 struct buffer_head
*inode_bh
= NULL
;
2120 ClearPageChecked(page
);
2123 BUG_ON(page
->index
!= 0);
2124 BUG_ON(len
> ext4_get_max_inline_size(inode
));
2125 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
2126 if (inode_bh
== NULL
)
2129 page_bufs
= page_buffers(page
);
2134 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
2138 * We need to release the page lock before we start the
2139 * journal, so grab a reference so the page won't disappear
2140 * out from under us.
2145 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2146 ext4_writepage_trans_blocks(inode
));
2147 if (IS_ERR(handle
)) {
2148 ret
= PTR_ERR(handle
);
2150 goto out_no_pagelock
;
2152 BUG_ON(!ext4_handle_valid(handle
));
2156 if (page
->mapping
!= mapping
) {
2157 /* The page got truncated from under us */
2158 ext4_journal_stop(handle
);
2164 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
2166 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
2169 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2170 do_journal_get_write_access
);
2172 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2177 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
2178 err
= ext4_journal_stop(handle
);
2182 if (!ext4_has_inline_data(inode
))
2183 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
2185 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2194 * Note that we don't need to start a transaction unless we're journaling data
2195 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2196 * need to file the inode to the transaction's list in ordered mode because if
2197 * we are writing back data added by write(), the inode is already there and if
2198 * we are writing back data modified via mmap(), no one guarantees in which
2199 * transaction the data will hit the disk. In case we are journaling data, we
2200 * cannot start transaction directly because transaction start ranks above page
2201 * lock so we have to do some magic.
2203 * This function can get called via...
2204 * - ext4_da_writepages after taking page lock (have journal handle)
2205 * - journal_submit_inode_data_buffers (no journal handle)
2206 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2207 * - grab_page_cache when doing write_begin (have journal handle)
2209 * We don't do any block allocation in this function. If we have page with
2210 * multiple blocks we need to write those buffer_heads that are mapped. This
2211 * is important for mmaped based write. So if we do with blocksize 1K
2212 * truncate(f, 1024);
2213 * a = mmap(f, 0, 4096);
2215 * truncate(f, 4096);
2216 * we have in the page first buffer_head mapped via page_mkwrite call back
2217 * but other buffer_heads would be unmapped but dirty (dirty done via the
2218 * do_wp_page). So writepage should write the first block. If we modify
2219 * the mmap area beyond 1024 we will again get a page_fault and the
2220 * page_mkwrite callback will do the block allocation and mark the
2221 * buffer_heads mapped.
2223 * We redirty the page if we have any buffer_heads that is either delay or
2224 * unwritten in the page.
2226 * We can get recursively called as show below.
2228 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2231 * But since we don't do any block allocation we should not deadlock.
2232 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2234 static int ext4_writepage(struct page
*page
,
2235 struct writeback_control
*wbc
)
2240 struct buffer_head
*page_bufs
= NULL
;
2241 struct inode
*inode
= page
->mapping
->host
;
2242 struct ext4_io_submit io_submit
;
2244 trace_ext4_writepage(page
);
2245 size
= i_size_read(inode
);
2246 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2247 len
= size
& ~PAGE_CACHE_MASK
;
2249 len
= PAGE_CACHE_SIZE
;
2251 page_bufs
= page_buffers(page
);
2253 * We cannot do block allocation or other extent handling in this
2254 * function. If there are buffers needing that, we have to redirty
2255 * the page. But we may reach here when we do a journal commit via
2256 * journal_submit_inode_data_buffers() and in that case we must write
2257 * allocated buffers to achieve data=ordered mode guarantees.
2259 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2260 ext4_bh_delay_or_unwritten
)) {
2261 redirty_page_for_writepage(wbc
, page
);
2262 if (current
->flags
& PF_MEMALLOC
) {
2264 * For memory cleaning there's no point in writing only
2265 * some buffers. So just bail out. Warn if we came here
2266 * from direct reclaim.
2268 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2275 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2277 * It's mmapped pagecache. Add buffers and journal it. There
2278 * doesn't seem much point in redirtying the page here.
2280 return __ext4_journalled_writepage(page
, len
);
2282 memset(&io_submit
, 0, sizeof(io_submit
));
2283 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
);
2284 ext4_io_submit(&io_submit
);
2289 * This is called via ext4_da_writepages() to
2290 * calculate the total number of credits to reserve to fit
2291 * a single extent allocation into a single transaction,
2292 * ext4_da_writpeages() will loop calling this before
2293 * the block allocation.
2296 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2298 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2301 * With non-extent format the journal credit needed to
2302 * insert nrblocks contiguous block is dependent on
2303 * number of contiguous block. So we will limit
2304 * number of contiguous block to a sane value
2306 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
2307 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2308 max_blocks
= EXT4_MAX_TRANS_DATA
;
2310 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2314 * write_cache_pages_da - walk the list of dirty pages of the given
2315 * address space and accumulate pages that need writing, and call
2316 * mpage_da_map_and_submit to map a single contiguous memory region
2317 * and then write them.
2319 static int write_cache_pages_da(handle_t
*handle
,
2320 struct address_space
*mapping
,
2321 struct writeback_control
*wbc
,
2322 struct mpage_da_data
*mpd
,
2323 pgoff_t
*done_index
)
2325 struct buffer_head
*bh
, *head
;
2326 struct inode
*inode
= mapping
->host
;
2327 struct pagevec pvec
;
2328 unsigned int nr_pages
;
2331 long nr_to_write
= wbc
->nr_to_write
;
2332 int i
, tag
, ret
= 0;
2334 memset(mpd
, 0, sizeof(struct mpage_da_data
));
2337 pagevec_init(&pvec
, 0);
2338 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2339 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2341 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2342 tag
= PAGECACHE_TAG_TOWRITE
;
2344 tag
= PAGECACHE_TAG_DIRTY
;
2346 *done_index
= index
;
2347 while (index
<= end
) {
2348 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2349 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2353 for (i
= 0; i
< nr_pages
; i
++) {
2354 struct page
*page
= pvec
.pages
[i
];
2357 * At this point, the page may be truncated or
2358 * invalidated (changing page->mapping to NULL), or
2359 * even swizzled back from swapper_space to tmpfs file
2360 * mapping. However, page->index will not change
2361 * because we have a reference on the page.
2363 if (page
->index
> end
)
2366 *done_index
= page
->index
+ 1;
2369 * If we can't merge this page, and we have
2370 * accumulated an contiguous region, write it
2372 if ((mpd
->next_page
!= page
->index
) &&
2373 (mpd
->next_page
!= mpd
->first_page
)) {
2374 mpage_da_map_and_submit(mpd
);
2375 goto ret_extent_tail
;
2381 * If the page is no longer dirty, or its
2382 * mapping no longer corresponds to inode we
2383 * are writing (which means it has been
2384 * truncated or invalidated), or the page is
2385 * already under writeback and we are not
2386 * doing a data integrity writeback, skip the page
2388 if (!PageDirty(page
) ||
2389 (PageWriteback(page
) &&
2390 (wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2391 unlikely(page
->mapping
!= mapping
)) {
2396 wait_on_page_writeback(page
);
2397 BUG_ON(PageWriteback(page
));
2400 * If we have inline data and arrive here, it means that
2401 * we will soon create the block for the 1st page, so
2402 * we'd better clear the inline data here.
2404 if (ext4_has_inline_data(inode
)) {
2405 BUG_ON(ext4_test_inode_state(inode
,
2406 EXT4_STATE_MAY_INLINE_DATA
));
2407 ext4_destroy_inline_data(handle
, inode
);
2410 if (mpd
->next_page
!= page
->index
)
2411 mpd
->first_page
= page
->index
;
2412 mpd
->next_page
= page
->index
+ 1;
2413 logical
= (sector_t
) page
->index
<<
2414 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2416 /* Add all dirty buffers to mpd */
2417 head
= page_buffers(page
);
2420 BUG_ON(buffer_locked(bh
));
2422 * We need to try to allocate unmapped blocks
2423 * in the same page. Otherwise we won't make
2424 * progress with the page in ext4_writepage
2426 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2427 mpage_add_bh_to_extent(mpd
, logical
,
2430 goto ret_extent_tail
;
2431 } else if (buffer_dirty(bh
) &&
2432 buffer_mapped(bh
)) {
2434 * mapped dirty buffer. We need to
2435 * update the b_state because we look
2436 * at b_state in mpage_da_map_blocks.
2437 * We don't update b_size because if we
2438 * find an unmapped buffer_head later
2439 * we need to use the b_state flag of
2442 if (mpd
->b_size
== 0)
2444 bh
->b_state
& BH_FLAGS
;
2447 } while ((bh
= bh
->b_this_page
) != head
);
2449 if (nr_to_write
> 0) {
2451 if (nr_to_write
== 0 &&
2452 wbc
->sync_mode
== WB_SYNC_NONE
)
2454 * We stop writing back only if we are
2455 * not doing integrity sync. In case of
2456 * integrity sync we have to keep going
2457 * because someone may be concurrently
2458 * dirtying pages, and we might have
2459 * synced a lot of newly appeared dirty
2460 * pages, but have not synced all of the
2466 pagevec_release(&pvec
);
2471 ret
= MPAGE_DA_EXTENT_TAIL
;
2473 pagevec_release(&pvec
);
2479 static int ext4_da_writepages(struct address_space
*mapping
,
2480 struct writeback_control
*wbc
)
2483 int range_whole
= 0;
2484 handle_t
*handle
= NULL
;
2485 struct mpage_da_data mpd
;
2486 struct inode
*inode
= mapping
->host
;
2487 int pages_written
= 0;
2488 unsigned int max_pages
;
2489 int range_cyclic
, cycled
= 1, io_done
= 0;
2490 int needed_blocks
, ret
= 0;
2491 long desired_nr_to_write
, nr_to_writebump
= 0;
2492 loff_t range_start
= wbc
->range_start
;
2493 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2494 pgoff_t done_index
= 0;
2496 struct blk_plug plug
;
2498 trace_ext4_da_writepages(inode
, wbc
);
2501 * No pages to write? This is mainly a kludge to avoid starting
2502 * a transaction for special inodes like journal inode on last iput()
2503 * because that could violate lock ordering on umount
2505 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2509 * If the filesystem has aborted, it is read-only, so return
2510 * right away instead of dumping stack traces later on that
2511 * will obscure the real source of the problem. We test
2512 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2513 * the latter could be true if the filesystem is mounted
2514 * read-only, and in that case, ext4_da_writepages should
2515 * *never* be called, so if that ever happens, we would want
2518 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2521 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2524 range_cyclic
= wbc
->range_cyclic
;
2525 if (wbc
->range_cyclic
) {
2526 index
= mapping
->writeback_index
;
2529 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2530 wbc
->range_end
= LLONG_MAX
;
2531 wbc
->range_cyclic
= 0;
2534 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2535 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2539 * This works around two forms of stupidity. The first is in
2540 * the writeback code, which caps the maximum number of pages
2541 * written to be 1024 pages. This is wrong on multiple
2542 * levels; different architectues have a different page size,
2543 * which changes the maximum amount of data which gets
2544 * written. Secondly, 4 megabytes is way too small. XFS
2545 * forces this value to be 16 megabytes by multiplying
2546 * nr_to_write parameter by four, and then relies on its
2547 * allocator to allocate larger extents to make them
2548 * contiguous. Unfortunately this brings us to the second
2549 * stupidity, which is that ext4's mballoc code only allocates
2550 * at most 2048 blocks. So we force contiguous writes up to
2551 * the number of dirty blocks in the inode, or
2552 * sbi->max_writeback_mb_bump whichever is smaller.
2554 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2555 if (!range_cyclic
&& range_whole
) {
2556 if (wbc
->nr_to_write
== LONG_MAX
)
2557 desired_nr_to_write
= wbc
->nr_to_write
;
2559 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2561 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2563 if (desired_nr_to_write
> max_pages
)
2564 desired_nr_to_write
= max_pages
;
2566 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2567 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2568 wbc
->nr_to_write
= desired_nr_to_write
;
2572 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2573 tag_pages_for_writeback(mapping
, index
, end
);
2575 blk_start_plug(&plug
);
2576 while (!ret
&& wbc
->nr_to_write
> 0) {
2579 * we insert one extent at a time. So we need
2580 * credit needed for single extent allocation.
2581 * journalled mode is currently not supported
2584 BUG_ON(ext4_should_journal_data(inode
));
2585 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2587 /* start a new transaction*/
2588 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2590 if (IS_ERR(handle
)) {
2591 ret
= PTR_ERR(handle
);
2592 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2593 "%ld pages, ino %lu; err %d", __func__
,
2594 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2595 blk_finish_plug(&plug
);
2596 goto out_writepages
;
2600 * Now call write_cache_pages_da() to find the next
2601 * contiguous region of logical blocks that need
2602 * blocks to be allocated by ext4 and submit them.
2604 ret
= write_cache_pages_da(handle
, mapping
,
2605 wbc
, &mpd
, &done_index
);
2607 * If we have a contiguous extent of pages and we
2608 * haven't done the I/O yet, map the blocks and submit
2611 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2612 mpage_da_map_and_submit(&mpd
);
2613 ret
= MPAGE_DA_EXTENT_TAIL
;
2615 trace_ext4_da_write_pages(inode
, &mpd
);
2616 wbc
->nr_to_write
-= mpd
.pages_written
;
2618 ext4_journal_stop(handle
);
2620 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2621 /* commit the transaction which would
2622 * free blocks released in the transaction
2625 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2627 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2629 * Got one extent now try with rest of the pages.
2630 * If mpd.retval is set -EIO, journal is aborted.
2631 * So we don't need to write any more.
2633 pages_written
+= mpd
.pages_written
;
2636 } else if (wbc
->nr_to_write
)
2638 * There is no more writeout needed
2639 * or we requested for a noblocking writeout
2640 * and we found the device congested
2644 blk_finish_plug(&plug
);
2645 if (!io_done
&& !cycled
) {
2648 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2649 wbc
->range_end
= mapping
->writeback_index
- 1;
2654 wbc
->range_cyclic
= range_cyclic
;
2655 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2657 * set the writeback_index so that range_cyclic
2658 * mode will write it back later
2660 mapping
->writeback_index
= done_index
;
2663 wbc
->nr_to_write
-= nr_to_writebump
;
2664 wbc
->range_start
= range_start
;
2665 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2669 static int ext4_nonda_switch(struct super_block
*sb
)
2671 s64 free_clusters
, dirty_clusters
;
2672 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2675 * switch to non delalloc mode if we are running low
2676 * on free block. The free block accounting via percpu
2677 * counters can get slightly wrong with percpu_counter_batch getting
2678 * accumulated on each CPU without updating global counters
2679 * Delalloc need an accurate free block accounting. So switch
2680 * to non delalloc when we are near to error range.
2683 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2685 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2687 * Start pushing delalloc when 1/2 of free blocks are dirty.
2689 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2690 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2692 if (2 * free_clusters
< 3 * dirty_clusters
||
2693 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2695 * free block count is less than 150% of dirty blocks
2696 * or free blocks is less than watermark
2703 /* We always reserve for an inode update; the superblock could be there too */
2704 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2706 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
2707 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
)))
2710 if (pos
+ len
<= 0x7fffffffULL
)
2713 /* We might need to update the superblock to set LARGE_FILE */
2717 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2718 loff_t pos
, unsigned len
, unsigned flags
,
2719 struct page
**pagep
, void **fsdata
)
2721 int ret
, retries
= 0;
2724 struct inode
*inode
= mapping
->host
;
2727 index
= pos
>> PAGE_CACHE_SHIFT
;
2729 if (ext4_nonda_switch(inode
->i_sb
)) {
2730 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2731 return ext4_write_begin(file
, mapping
, pos
,
2732 len
, flags
, pagep
, fsdata
);
2734 *fsdata
= (void *)0;
2735 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2737 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2738 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2748 * grab_cache_page_write_begin() can take a long time if the
2749 * system is thrashing due to memory pressure, or if the page
2750 * is being written back. So grab it first before we start
2751 * the transaction handle. This also allows us to allocate
2752 * the page (if needed) without using GFP_NOFS.
2755 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2761 * With delayed allocation, we don't log the i_disksize update
2762 * if there is delayed block allocation. But we still need
2763 * to journalling the i_disksize update if writes to the end
2764 * of file which has an already mapped buffer.
2767 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2768 ext4_da_write_credits(inode
, pos
, len
));
2769 if (IS_ERR(handle
)) {
2770 page_cache_release(page
);
2771 return PTR_ERR(handle
);
2775 if (page
->mapping
!= mapping
) {
2776 /* The page got truncated from under us */
2778 page_cache_release(page
);
2779 ext4_journal_stop(handle
);
2782 /* In case writeback began while the page was unlocked */
2783 wait_for_stable_page(page
);
2785 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2788 ext4_journal_stop(handle
);
2790 * block_write_begin may have instantiated a few blocks
2791 * outside i_size. Trim these off again. Don't need
2792 * i_size_read because we hold i_mutex.
2794 if (pos
+ len
> inode
->i_size
)
2795 ext4_truncate_failed_write(inode
);
2797 if (ret
== -ENOSPC
&&
2798 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2801 page_cache_release(page
);
2810 * Check if we should update i_disksize
2811 * when write to the end of file but not require block allocation
2813 static int ext4_da_should_update_i_disksize(struct page
*page
,
2814 unsigned long offset
)
2816 struct buffer_head
*bh
;
2817 struct inode
*inode
= page
->mapping
->host
;
2821 bh
= page_buffers(page
);
2822 idx
= offset
>> inode
->i_blkbits
;
2824 for (i
= 0; i
< idx
; i
++)
2825 bh
= bh
->b_this_page
;
2827 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2832 static int ext4_da_write_end(struct file
*file
,
2833 struct address_space
*mapping
,
2834 loff_t pos
, unsigned len
, unsigned copied
,
2835 struct page
*page
, void *fsdata
)
2837 struct inode
*inode
= mapping
->host
;
2839 handle_t
*handle
= ext4_journal_current_handle();
2841 unsigned long start
, end
;
2842 int write_mode
= (int)(unsigned long)fsdata
;
2844 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
2845 return ext4_write_end(file
, mapping
, pos
,
2846 len
, copied
, page
, fsdata
);
2848 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2849 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2850 end
= start
+ copied
- 1;
2853 * generic_write_end() will run mark_inode_dirty() if i_size
2854 * changes. So let's piggyback the i_disksize mark_inode_dirty
2857 new_i_size
= pos
+ copied
;
2858 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2859 if (ext4_has_inline_data(inode
) ||
2860 ext4_da_should_update_i_disksize(page
, end
)) {
2861 down_write(&EXT4_I(inode
)->i_data_sem
);
2862 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
2863 EXT4_I(inode
)->i_disksize
= new_i_size
;
2864 up_write(&EXT4_I(inode
)->i_data_sem
);
2865 /* We need to mark inode dirty even if
2866 * new_i_size is less that inode->i_size
2867 * bu greater than i_disksize.(hint delalloc)
2869 ext4_mark_inode_dirty(handle
, inode
);
2873 if (write_mode
!= CONVERT_INLINE_DATA
&&
2874 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
2875 ext4_has_inline_data(inode
))
2876 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
2879 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2885 ret2
= ext4_journal_stop(handle
);
2889 return ret
? ret
: copied
;
2892 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2895 * Drop reserved blocks
2897 BUG_ON(!PageLocked(page
));
2898 if (!page_has_buffers(page
))
2901 ext4_da_page_release_reservation(page
, offset
);
2904 ext4_invalidatepage(page
, offset
);
2910 * Force all delayed allocation blocks to be allocated for a given inode.
2912 int ext4_alloc_da_blocks(struct inode
*inode
)
2914 trace_ext4_alloc_da_blocks(inode
);
2916 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2917 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2921 * We do something simple for now. The filemap_flush() will
2922 * also start triggering a write of the data blocks, which is
2923 * not strictly speaking necessary (and for users of
2924 * laptop_mode, not even desirable). However, to do otherwise
2925 * would require replicating code paths in:
2927 * ext4_da_writepages() ->
2928 * write_cache_pages() ---> (via passed in callback function)
2929 * __mpage_da_writepage() -->
2930 * mpage_add_bh_to_extent()
2931 * mpage_da_map_blocks()
2933 * The problem is that write_cache_pages(), located in
2934 * mm/page-writeback.c, marks pages clean in preparation for
2935 * doing I/O, which is not desirable if we're not planning on
2938 * We could call write_cache_pages(), and then redirty all of
2939 * the pages by calling redirty_page_for_writepage() but that
2940 * would be ugly in the extreme. So instead we would need to
2941 * replicate parts of the code in the above functions,
2942 * simplifying them because we wouldn't actually intend to
2943 * write out the pages, but rather only collect contiguous
2944 * logical block extents, call the multi-block allocator, and
2945 * then update the buffer heads with the block allocations.
2947 * For now, though, we'll cheat by calling filemap_flush(),
2948 * which will map the blocks, and start the I/O, but not
2949 * actually wait for the I/O to complete.
2951 return filemap_flush(inode
->i_mapping
);
2955 * bmap() is special. It gets used by applications such as lilo and by
2956 * the swapper to find the on-disk block of a specific piece of data.
2958 * Naturally, this is dangerous if the block concerned is still in the
2959 * journal. If somebody makes a swapfile on an ext4 data-journaling
2960 * filesystem and enables swap, then they may get a nasty shock when the
2961 * data getting swapped to that swapfile suddenly gets overwritten by
2962 * the original zero's written out previously to the journal and
2963 * awaiting writeback in the kernel's buffer cache.
2965 * So, if we see any bmap calls here on a modified, data-journaled file,
2966 * take extra steps to flush any blocks which might be in the cache.
2968 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2970 struct inode
*inode
= mapping
->host
;
2975 * We can get here for an inline file via the FIBMAP ioctl
2977 if (ext4_has_inline_data(inode
))
2980 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2981 test_opt(inode
->i_sb
, DELALLOC
)) {
2983 * With delalloc we want to sync the file
2984 * so that we can make sure we allocate
2987 filemap_write_and_wait(mapping
);
2990 if (EXT4_JOURNAL(inode
) &&
2991 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2993 * This is a REALLY heavyweight approach, but the use of
2994 * bmap on dirty files is expected to be extremely rare:
2995 * only if we run lilo or swapon on a freshly made file
2996 * do we expect this to happen.
2998 * (bmap requires CAP_SYS_RAWIO so this does not
2999 * represent an unprivileged user DOS attack --- we'd be
3000 * in trouble if mortal users could trigger this path at
3003 * NB. EXT4_STATE_JDATA is not set on files other than
3004 * regular files. If somebody wants to bmap a directory
3005 * or symlink and gets confused because the buffer
3006 * hasn't yet been flushed to disk, they deserve
3007 * everything they get.
3010 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
3011 journal
= EXT4_JOURNAL(inode
);
3012 jbd2_journal_lock_updates(journal
);
3013 err
= jbd2_journal_flush(journal
);
3014 jbd2_journal_unlock_updates(journal
);
3020 return generic_block_bmap(mapping
, block
, ext4_get_block
);
3023 static int ext4_readpage(struct file
*file
, struct page
*page
)
3026 struct inode
*inode
= page
->mapping
->host
;
3028 trace_ext4_readpage(page
);
3030 if (ext4_has_inline_data(inode
))
3031 ret
= ext4_readpage_inline(inode
, page
);
3034 return mpage_readpage(page
, ext4_get_block
);
3040 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3041 struct list_head
*pages
, unsigned nr_pages
)
3043 struct inode
*inode
= mapping
->host
;
3045 /* If the file has inline data, no need to do readpages. */
3046 if (ext4_has_inline_data(inode
))
3049 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3052 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3054 trace_ext4_invalidatepage(page
, offset
);
3056 /* No journalling happens on data buffers when this function is used */
3057 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3059 block_invalidatepage(page
, offset
);
3062 static int __ext4_journalled_invalidatepage(struct page
*page
,
3063 unsigned long offset
)
3065 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3067 trace_ext4_journalled_invalidatepage(page
, offset
);
3070 * If it's a full truncate we just forget about the pending dirtying
3073 ClearPageChecked(page
);
3075 return jbd2_journal_invalidatepage(journal
, page
, offset
);
3078 /* Wrapper for aops... */
3079 static void ext4_journalled_invalidatepage(struct page
*page
,
3080 unsigned long offset
)
3082 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
) < 0);
3085 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3087 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3089 trace_ext4_releasepage(page
);
3091 /* Page has dirty journalled data -> cannot release */
3092 if (PageChecked(page
))
3095 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3097 return try_to_free_buffers(page
);
3101 * ext4_get_block used when preparing for a DIO write or buffer write.
3102 * We allocate an uinitialized extent if blocks haven't been allocated.
3103 * The extent will be converted to initialized after the IO is complete.
3105 int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
3106 struct buffer_head
*bh_result
, int create
)
3108 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3109 inode
->i_ino
, create
);
3110 return _ext4_get_block(inode
, iblock
, bh_result
,
3111 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
3114 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
3115 struct buffer_head
*bh_result
, int create
)
3117 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3118 inode
->i_ino
, create
);
3119 return _ext4_get_block(inode
, iblock
, bh_result
,
3120 EXT4_GET_BLOCKS_NO_LOCK
);
3123 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3124 ssize_t size
, void *private, int ret
,
3127 struct inode
*inode
= file_inode(iocb
->ki_filp
);
3128 ext4_io_end_t
*io_end
= iocb
->private;
3130 /* if not async direct IO or dio with 0 bytes write, just return */
3131 if (!io_end
|| !size
)
3134 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3135 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3136 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3139 iocb
->private = NULL
;
3141 /* if not aio dio with unwritten extents, just free io and return */
3142 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
3143 ext4_free_io_end(io_end
);
3145 inode_dio_done(inode
);
3147 aio_complete(iocb
, ret
, 0);
3151 io_end
->offset
= offset
;
3152 io_end
->size
= size
;
3154 io_end
->iocb
= iocb
;
3155 io_end
->result
= ret
;
3158 ext4_add_complete_io(io_end
);
3162 * For ext4 extent files, ext4 will do direct-io write to holes,
3163 * preallocated extents, and those write extend the file, no need to
3164 * fall back to buffered IO.
3166 * For holes, we fallocate those blocks, mark them as uninitialized
3167 * If those blocks were preallocated, we mark sure they are split, but
3168 * still keep the range to write as uninitialized.
3170 * The unwritten extents will be converted to written when DIO is completed.
3171 * For async direct IO, since the IO may still pending when return, we
3172 * set up an end_io call back function, which will do the conversion
3173 * when async direct IO completed.
3175 * If the O_DIRECT write will extend the file then add this inode to the
3176 * orphan list. So recovery will truncate it back to the original size
3177 * if the machine crashes during the write.
3180 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3181 const struct iovec
*iov
, loff_t offset
,
3182 unsigned long nr_segs
)
3184 struct file
*file
= iocb
->ki_filp
;
3185 struct inode
*inode
= file
->f_mapping
->host
;
3187 size_t count
= iov_length(iov
, nr_segs
);
3189 get_block_t
*get_block_func
= NULL
;
3191 loff_t final_size
= offset
+ count
;
3193 /* Use the old path for reads and writes beyond i_size. */
3194 if (rw
!= WRITE
|| final_size
> inode
->i_size
)
3195 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3197 BUG_ON(iocb
->private == NULL
);
3199 /* If we do a overwrite dio, i_mutex locking can be released */
3200 overwrite
= *((int *)iocb
->private);
3203 atomic_inc(&inode
->i_dio_count
);
3204 down_read(&EXT4_I(inode
)->i_data_sem
);
3205 mutex_unlock(&inode
->i_mutex
);
3209 * We could direct write to holes and fallocate.
3211 * Allocated blocks to fill the hole are marked as
3212 * uninitialized to prevent parallel buffered read to expose
3213 * the stale data before DIO complete the data IO.
3215 * As to previously fallocated extents, ext4 get_block will
3216 * just simply mark the buffer mapped but still keep the
3217 * extents uninitialized.
3219 * For non AIO case, we will convert those unwritten extents
3220 * to written after return back from blockdev_direct_IO.
3222 * For async DIO, the conversion needs to be deferred when the
3223 * IO is completed. The ext4 end_io callback function will be
3224 * called to take care of the conversion work. Here for async
3225 * case, we allocate an io_end structure to hook to the iocb.
3227 iocb
->private = NULL
;
3228 ext4_inode_aio_set(inode
, NULL
);
3229 if (!is_sync_kiocb(iocb
)) {
3230 ext4_io_end_t
*io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
3235 io_end
->flag
|= EXT4_IO_END_DIRECT
;
3236 iocb
->private = io_end
;
3238 * we save the io structure for current async direct
3239 * IO, so that later ext4_map_blocks() could flag the
3240 * io structure whether there is a unwritten extents
3241 * needs to be converted when IO is completed.
3243 ext4_inode_aio_set(inode
, io_end
);
3247 get_block_func
= ext4_get_block_write_nolock
;
3249 get_block_func
= ext4_get_block_write
;
3250 dio_flags
= DIO_LOCKING
;
3252 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3253 inode
->i_sb
->s_bdev
, iov
,
3261 ext4_inode_aio_set(inode
, NULL
);
3263 * The io_end structure takes a reference to the inode, that
3264 * structure needs to be destroyed and the reference to the
3265 * inode need to be dropped, when IO is complete, even with 0
3266 * byte write, or failed.
3268 * In the successful AIO DIO case, the io_end structure will
3269 * be destroyed and the reference to the inode will be dropped
3270 * after the end_io call back function is called.
3272 * In the case there is 0 byte write, or error case, since VFS
3273 * direct IO won't invoke the end_io call back function, we
3274 * need to free the end_io structure here.
3276 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3277 ext4_free_io_end(iocb
->private);
3278 iocb
->private = NULL
;
3279 } else if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3280 EXT4_STATE_DIO_UNWRITTEN
)) {
3283 * for non AIO case, since the IO is already
3284 * completed, we could do the conversion right here
3286 err
= ext4_convert_unwritten_extents(inode
,
3290 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3294 /* take i_mutex locking again if we do a ovewrite dio */
3296 inode_dio_done(inode
);
3297 up_read(&EXT4_I(inode
)->i_data_sem
);
3298 mutex_lock(&inode
->i_mutex
);
3304 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3305 const struct iovec
*iov
, loff_t offset
,
3306 unsigned long nr_segs
)
3308 struct file
*file
= iocb
->ki_filp
;
3309 struct inode
*inode
= file
->f_mapping
->host
;
3313 * If we are doing data journalling we don't support O_DIRECT
3315 if (ext4_should_journal_data(inode
))
3318 /* Let buffer I/O handle the inline data case. */
3319 if (ext4_has_inline_data(inode
))
3322 trace_ext4_direct_IO_enter(inode
, offset
, iov_length(iov
, nr_segs
), rw
);
3323 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3324 ret
= ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3326 ret
= ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3327 trace_ext4_direct_IO_exit(inode
, offset
,
3328 iov_length(iov
, nr_segs
), rw
, ret
);
3333 * Pages can be marked dirty completely asynchronously from ext4's journalling
3334 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3335 * much here because ->set_page_dirty is called under VFS locks. The page is
3336 * not necessarily locked.
3338 * We cannot just dirty the page and leave attached buffers clean, because the
3339 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3340 * or jbddirty because all the journalling code will explode.
3342 * So what we do is to mark the page "pending dirty" and next time writepage
3343 * is called, propagate that into the buffers appropriately.
3345 static int ext4_journalled_set_page_dirty(struct page
*page
)
3347 SetPageChecked(page
);
3348 return __set_page_dirty_nobuffers(page
);
3351 static const struct address_space_operations ext4_aops
= {
3352 .readpage
= ext4_readpage
,
3353 .readpages
= ext4_readpages
,
3354 .writepage
= ext4_writepage
,
3355 .write_begin
= ext4_write_begin
,
3356 .write_end
= ext4_write_end
,
3358 .invalidatepage
= ext4_invalidatepage
,
3359 .releasepage
= ext4_releasepage
,
3360 .direct_IO
= ext4_direct_IO
,
3361 .migratepage
= buffer_migrate_page
,
3362 .is_partially_uptodate
= block_is_partially_uptodate
,
3363 .error_remove_page
= generic_error_remove_page
,
3366 static const struct address_space_operations ext4_journalled_aops
= {
3367 .readpage
= ext4_readpage
,
3368 .readpages
= ext4_readpages
,
3369 .writepage
= ext4_writepage
,
3370 .write_begin
= ext4_write_begin
,
3371 .write_end
= ext4_journalled_write_end
,
3372 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3374 .invalidatepage
= ext4_journalled_invalidatepage
,
3375 .releasepage
= ext4_releasepage
,
3376 .direct_IO
= ext4_direct_IO
,
3377 .is_partially_uptodate
= block_is_partially_uptodate
,
3378 .error_remove_page
= generic_error_remove_page
,
3381 static const struct address_space_operations ext4_da_aops
= {
3382 .readpage
= ext4_readpage
,
3383 .readpages
= ext4_readpages
,
3384 .writepage
= ext4_writepage
,
3385 .writepages
= ext4_da_writepages
,
3386 .write_begin
= ext4_da_write_begin
,
3387 .write_end
= ext4_da_write_end
,
3389 .invalidatepage
= ext4_da_invalidatepage
,
3390 .releasepage
= ext4_releasepage
,
3391 .direct_IO
= ext4_direct_IO
,
3392 .migratepage
= buffer_migrate_page
,
3393 .is_partially_uptodate
= block_is_partially_uptodate
,
3394 .error_remove_page
= generic_error_remove_page
,
3397 void ext4_set_aops(struct inode
*inode
)
3399 switch (ext4_inode_journal_mode(inode
)) {
3400 case EXT4_INODE_ORDERED_DATA_MODE
:
3401 ext4_set_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3403 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3404 ext4_clear_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3406 case EXT4_INODE_JOURNAL_DATA_MODE
:
3407 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3412 if (test_opt(inode
->i_sb
, DELALLOC
))
3413 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3415 inode
->i_mapping
->a_ops
= &ext4_aops
;
3420 * ext4_discard_partial_page_buffers()
3421 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3422 * This function finds and locks the page containing the offset
3423 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3424 * Calling functions that already have the page locked should call
3425 * ext4_discard_partial_page_buffers_no_lock directly.
3427 int ext4_discard_partial_page_buffers(handle_t
*handle
,
3428 struct address_space
*mapping
, loff_t from
,
3429 loff_t length
, int flags
)
3431 struct inode
*inode
= mapping
->host
;
3435 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3436 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3440 err
= ext4_discard_partial_page_buffers_no_lock(handle
, inode
, page
,
3441 from
, length
, flags
);
3444 page_cache_release(page
);
3449 * ext4_discard_partial_page_buffers_no_lock()
3450 * Zeros a page range of length 'length' starting from offset 'from'.
3451 * Buffer heads that correspond to the block aligned regions of the
3452 * zeroed range will be unmapped. Unblock aligned regions
3453 * will have the corresponding buffer head mapped if needed so that
3454 * that region of the page can be updated with the partial zero out.
3456 * This function assumes that the page has already been locked. The
3457 * The range to be discarded must be contained with in the given page.
3458 * If the specified range exceeds the end of the page it will be shortened
3459 * to the end of the page that corresponds to 'from'. This function is
3460 * appropriate for updating a page and it buffer heads to be unmapped and
3461 * zeroed for blocks that have been either released, or are going to be
3464 * handle: The journal handle
3465 * inode: The files inode
3466 * page: A locked page that contains the offset "from"
3467 * from: The starting byte offset (from the beginning of the file)
3468 * to begin discarding
3469 * len: The length of bytes to discard
3470 * flags: Optional flags that may be used:
3472 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3473 * Only zero the regions of the page whose buffer heads
3474 * have already been unmapped. This flag is appropriate
3475 * for updating the contents of a page whose blocks may
3476 * have already been released, and we only want to zero
3477 * out the regions that correspond to those released blocks.
3479 * Returns zero on success or negative on failure.
3481 static int ext4_discard_partial_page_buffers_no_lock(handle_t
*handle
,
3482 struct inode
*inode
, struct page
*page
, loff_t from
,
3483 loff_t length
, int flags
)
3485 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3486 unsigned int offset
= from
& (PAGE_CACHE_SIZE
-1);
3487 unsigned int blocksize
, max
, pos
;
3489 struct buffer_head
*bh
;
3492 blocksize
= inode
->i_sb
->s_blocksize
;
3493 max
= PAGE_CACHE_SIZE
- offset
;
3495 if (index
!= page
->index
)
3499 * correct length if it does not fall between
3500 * 'from' and the end of the page
3502 if (length
> max
|| length
< 0)
3505 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3507 if (!page_has_buffers(page
))
3508 create_empty_buffers(page
, blocksize
, 0);
3510 /* Find the buffer that contains "offset" */
3511 bh
= page_buffers(page
);
3513 while (offset
>= pos
) {
3514 bh
= bh
->b_this_page
;
3520 while (pos
< offset
+ length
) {
3521 unsigned int end_of_block
, range_to_discard
;
3525 /* The length of space left to zero and unmap */
3526 range_to_discard
= offset
+ length
- pos
;
3528 /* The length of space until the end of the block */
3529 end_of_block
= blocksize
- (pos
& (blocksize
-1));
3532 * Do not unmap or zero past end of block
3533 * for this buffer head
3535 if (range_to_discard
> end_of_block
)
3536 range_to_discard
= end_of_block
;
3540 * Skip this buffer head if we are only zeroing unampped
3541 * regions of the page
3543 if (flags
& EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
&&
3547 /* If the range is block aligned, unmap */
3548 if (range_to_discard
== blocksize
) {
3549 clear_buffer_dirty(bh
);
3551 clear_buffer_mapped(bh
);
3552 clear_buffer_req(bh
);
3553 clear_buffer_new(bh
);
3554 clear_buffer_delay(bh
);
3555 clear_buffer_unwritten(bh
);
3556 clear_buffer_uptodate(bh
);
3557 zero_user(page
, pos
, range_to_discard
);
3558 BUFFER_TRACE(bh
, "Buffer discarded");
3563 * If this block is not completely contained in the range
3564 * to be discarded, then it is not going to be released. Because
3565 * we need to keep this block, we need to make sure this part
3566 * of the page is uptodate before we modify it by writeing
3567 * partial zeros on it.
3569 if (!buffer_mapped(bh
)) {
3571 * Buffer head must be mapped before we can read
3574 BUFFER_TRACE(bh
, "unmapped");
3575 ext4_get_block(inode
, iblock
, bh
, 0);
3576 /* unmapped? It's a hole - nothing to do */
3577 if (!buffer_mapped(bh
)) {
3578 BUFFER_TRACE(bh
, "still unmapped");
3583 /* Ok, it's mapped. Make sure it's up-to-date */
3584 if (PageUptodate(page
))
3585 set_buffer_uptodate(bh
);
3587 if (!buffer_uptodate(bh
)) {
3589 ll_rw_block(READ
, 1, &bh
);
3591 /* Uhhuh. Read error. Complain and punt.*/
3592 if (!buffer_uptodate(bh
))
3596 if (ext4_should_journal_data(inode
)) {
3597 BUFFER_TRACE(bh
, "get write access");
3598 err
= ext4_journal_get_write_access(handle
, bh
);
3603 zero_user(page
, pos
, range_to_discard
);
3606 if (ext4_should_journal_data(inode
)) {
3607 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3609 mark_buffer_dirty(bh
);
3611 BUFFER_TRACE(bh
, "Partial buffer zeroed");
3613 bh
= bh
->b_this_page
;
3615 pos
+= range_to_discard
;
3621 int ext4_can_truncate(struct inode
*inode
)
3623 if (S_ISREG(inode
->i_mode
))
3625 if (S_ISDIR(inode
->i_mode
))
3627 if (S_ISLNK(inode
->i_mode
))
3628 return !ext4_inode_is_fast_symlink(inode
);
3633 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3634 * associated with the given offset and length
3636 * @inode: File inode
3637 * @offset: The offset where the hole will begin
3638 * @len: The length of the hole
3640 * Returns: 0 on success or negative on failure
3643 int ext4_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
3646 struct inode
*inode
= file_inode(file
);
3647 struct super_block
*sb
= inode
->i_sb
;
3648 ext4_lblk_t first_block
, stop_block
;
3649 struct address_space
*mapping
= inode
->i_mapping
;
3650 loff_t first_page
, last_page
, page_len
;
3651 loff_t first_page_offset
, last_page_offset
;
3653 unsigned int credits
;
3656 if (!S_ISREG(inode
->i_mode
))
3659 if (EXT4_SB(sb
)->s_cluster_ratio
> 1) {
3660 /* TODO: Add support for bigalloc file systems */
3664 trace_ext4_punch_hole(inode
, offset
, length
);
3667 * Write out all dirty pages to avoid race conditions
3668 * Then release them.
3670 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3671 ret
= filemap_write_and_wait_range(mapping
, offset
,
3672 offset
+ length
- 1);
3677 mutex_lock(&inode
->i_mutex
);
3678 /* It's not possible punch hole on append only file */
3679 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
)) {
3683 if (IS_SWAPFILE(inode
)) {
3688 /* No need to punch hole beyond i_size */
3689 if (offset
>= inode
->i_size
)
3693 * If the hole extends beyond i_size, set the hole
3694 * to end after the page that contains i_size
3696 if (offset
+ length
> inode
->i_size
) {
3697 length
= inode
->i_size
+
3698 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
3702 first_page
= (offset
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
3703 last_page
= (offset
+ length
) >> PAGE_CACHE_SHIFT
;
3705 first_page_offset
= first_page
<< PAGE_CACHE_SHIFT
;
3706 last_page_offset
= last_page
<< PAGE_CACHE_SHIFT
;
3708 /* Now release the pages */
3709 if (last_page_offset
> first_page_offset
) {
3710 truncate_pagecache_range(inode
, first_page_offset
,
3711 last_page_offset
- 1);
3714 /* Wait all existing dio workers, newcomers will block on i_mutex */
3715 ext4_inode_block_unlocked_dio(inode
);
3716 ret
= ext4_flush_unwritten_io(inode
);
3719 inode_dio_wait(inode
);
3721 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3722 credits
= ext4_writepage_trans_blocks(inode
);
3724 credits
= ext4_blocks_for_truncate(inode
);
3725 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3726 if (IS_ERR(handle
)) {
3727 ret
= PTR_ERR(handle
);
3728 ext4_std_error(sb
, ret
);
3733 * Now we need to zero out the non-page-aligned data in the
3734 * pages at the start and tail of the hole, and unmap the
3735 * buffer heads for the block aligned regions of the page that
3736 * were completely zeroed.
3738 if (first_page
> last_page
) {
3740 * If the file space being truncated is contained
3741 * within a page just zero out and unmap the middle of
3744 ret
= ext4_discard_partial_page_buffers(handle
,
3745 mapping
, offset
, length
, 0);
3751 * zero out and unmap the partial page that contains
3752 * the start of the hole
3754 page_len
= first_page_offset
- offset
;
3756 ret
= ext4_discard_partial_page_buffers(handle
, mapping
,
3757 offset
, page_len
, 0);
3763 * zero out and unmap the partial page that contains
3764 * the end of the hole
3766 page_len
= offset
+ length
- last_page_offset
;
3768 ret
= ext4_discard_partial_page_buffers(handle
, mapping
,
3769 last_page_offset
, page_len
, 0);
3776 * If i_size is contained in the last page, we need to
3777 * unmap and zero the partial page after i_size
3779 if (inode
->i_size
>> PAGE_CACHE_SHIFT
== last_page
&&
3780 inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
3781 page_len
= PAGE_CACHE_SIZE
-
3782 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
3785 ret
= ext4_discard_partial_page_buffers(handle
,
3786 mapping
, inode
->i_size
, page_len
, 0);
3793 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
3794 EXT4_BLOCK_SIZE_BITS(sb
);
3795 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
3797 /* If there are no blocks to remove, return now */
3798 if (first_block
>= stop_block
)
3801 down_write(&EXT4_I(inode
)->i_data_sem
);
3802 ext4_discard_preallocations(inode
);
3804 ret
= ext4_es_remove_extent(inode
, first_block
,
3805 stop_block
- first_block
);
3807 up_write(&EXT4_I(inode
)->i_data_sem
);
3811 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3812 ret
= ext4_ext_remove_space(inode
, first_block
,
3815 ret
= ext4_free_hole_blocks(handle
, inode
, first_block
,
3818 ext4_discard_preallocations(inode
);
3819 up_write(&EXT4_I(inode
)->i_data_sem
);
3821 ext4_handle_sync(handle
);
3822 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3823 ext4_mark_inode_dirty(handle
, inode
);
3825 ext4_journal_stop(handle
);
3827 ext4_inode_resume_unlocked_dio(inode
);
3829 mutex_unlock(&inode
->i_mutex
);
3833 * Disabled as per b/28760453
3842 * We block out ext4_get_block() block instantiations across the entire
3843 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3844 * simultaneously on behalf of the same inode.
3846 * As we work through the truncate and commit bits of it to the journal there
3847 * is one core, guiding principle: the file's tree must always be consistent on
3848 * disk. We must be able to restart the truncate after a crash.
3850 * The file's tree may be transiently inconsistent in memory (although it
3851 * probably isn't), but whenever we close off and commit a journal transaction,
3852 * the contents of (the filesystem + the journal) must be consistent and
3853 * restartable. It's pretty simple, really: bottom up, right to left (although
3854 * left-to-right works OK too).
3856 * Note that at recovery time, journal replay occurs *before* the restart of
3857 * truncate against the orphan inode list.
3859 * The committed inode has the new, desired i_size (which is the same as
3860 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3861 * that this inode's truncate did not complete and it will again call
3862 * ext4_truncate() to have another go. So there will be instantiated blocks
3863 * to the right of the truncation point in a crashed ext4 filesystem. But
3864 * that's fine - as long as they are linked from the inode, the post-crash
3865 * ext4_truncate() run will find them and release them.
3867 void ext4_truncate(struct inode
*inode
)
3869 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3870 unsigned int credits
;
3872 struct address_space
*mapping
= inode
->i_mapping
;
3876 * There is a possibility that we're either freeing the inode
3877 * or it completely new indode. In those cases we might not
3878 * have i_mutex locked because it's not necessary.
3880 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
3881 WARN_ON(!mutex_is_locked(&inode
->i_mutex
));
3882 trace_ext4_truncate_enter(inode
);
3884 if (!ext4_can_truncate(inode
))
3887 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3889 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3890 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3892 if (ext4_has_inline_data(inode
)) {
3895 ext4_inline_data_truncate(inode
, &has_inline
);
3901 * finish any pending end_io work so we won't run the risk of
3902 * converting any truncated blocks to initialized later
3904 ext4_flush_unwritten_io(inode
);
3906 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3907 credits
= ext4_writepage_trans_blocks(inode
);
3909 credits
= ext4_blocks_for_truncate(inode
);
3911 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3912 if (IS_ERR(handle
)) {
3913 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
3917 if (inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
3918 page_len
= PAGE_CACHE_SIZE
-
3919 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
3921 if (ext4_discard_partial_page_buffers(handle
,
3922 mapping
, inode
->i_size
, page_len
, 0))
3927 * We add the inode to the orphan list, so that if this
3928 * truncate spans multiple transactions, and we crash, we will
3929 * resume the truncate when the filesystem recovers. It also
3930 * marks the inode dirty, to catch the new size.
3932 * Implication: the file must always be in a sane, consistent
3933 * truncatable state while each transaction commits.
3935 if (ext4_orphan_add(handle
, inode
))
3938 down_write(&EXT4_I(inode
)->i_data_sem
);
3940 ext4_discard_preallocations(inode
);
3942 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3943 ext4_ext_truncate(handle
, inode
);
3945 ext4_ind_truncate(handle
, inode
);
3947 up_write(&ei
->i_data_sem
);
3950 ext4_handle_sync(handle
);
3954 * If this was a simple ftruncate() and the file will remain alive,
3955 * then we need to clear up the orphan record which we created above.
3956 * However, if this was a real unlink then we were called by
3957 * ext4_delete_inode(), and we allow that function to clean up the
3958 * orphan info for us.
3961 ext4_orphan_del(handle
, inode
);
3963 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3964 ext4_mark_inode_dirty(handle
, inode
);
3965 ext4_journal_stop(handle
);
3967 trace_ext4_truncate_exit(inode
);
3971 * ext4_get_inode_loc returns with an extra refcount against the inode's
3972 * underlying buffer_head on success. If 'in_mem' is true, we have all
3973 * data in memory that is needed to recreate the on-disk version of this
3976 static int __ext4_get_inode_loc(struct inode
*inode
,
3977 struct ext4_iloc
*iloc
, int in_mem
)
3979 struct ext4_group_desc
*gdp
;
3980 struct buffer_head
*bh
;
3981 struct super_block
*sb
= inode
->i_sb
;
3983 int inodes_per_block
, inode_offset
;
3986 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3989 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3990 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3995 * Figure out the offset within the block group inode table
3997 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3998 inode_offset
= ((inode
->i_ino
- 1) %
3999 EXT4_INODES_PER_GROUP(sb
));
4000 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
4001 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
4003 bh
= sb_getblk(sb
, block
);
4006 if (!buffer_uptodate(bh
)) {
4010 * If the buffer has the write error flag, we have failed
4011 * to write out another inode in the same block. In this
4012 * case, we don't have to read the block because we may
4013 * read the old inode data successfully.
4015 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
4016 set_buffer_uptodate(bh
);
4018 if (buffer_uptodate(bh
)) {
4019 /* someone brought it uptodate while we waited */
4025 * If we have all information of the inode in memory and this
4026 * is the only valid inode in the block, we need not read the
4030 struct buffer_head
*bitmap_bh
;
4033 start
= inode_offset
& ~(inodes_per_block
- 1);
4035 /* Is the inode bitmap in cache? */
4036 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4037 if (unlikely(!bitmap_bh
))
4041 * If the inode bitmap isn't in cache then the
4042 * optimisation may end up performing two reads instead
4043 * of one, so skip it.
4045 if (!buffer_uptodate(bitmap_bh
)) {
4049 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4050 if (i
== inode_offset
)
4052 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4056 if (i
== start
+ inodes_per_block
) {
4057 /* all other inodes are free, so skip I/O */
4058 memset(bh
->b_data
, 0, bh
->b_size
);
4059 set_buffer_uptodate(bh
);
4067 * If we need to do any I/O, try to pre-readahead extra
4068 * blocks from the inode table.
4070 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4071 ext4_fsblk_t b
, end
, table
;
4073 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4075 table
= ext4_inode_table(sb
, gdp
);
4076 /* s_inode_readahead_blks is always a power of 2 */
4077 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4081 num
= EXT4_INODES_PER_GROUP(sb
);
4082 if (ext4_has_group_desc_csum(sb
))
4083 num
-= ext4_itable_unused_count(sb
, gdp
);
4084 table
+= num
/ inodes_per_block
;
4088 sb_breadahead(sb
, b
++);
4092 * There are other valid inodes in the buffer, this inode
4093 * has in-inode xattrs, or we don't have this inode in memory.
4094 * Read the block from disk.
4096 trace_ext4_load_inode(inode
);
4098 bh
->b_end_io
= end_buffer_read_sync
;
4099 #ifdef FEATURE_STORAGE_META_LOG
4100 if( bh
&& bh
->b_bdev
&& bh
->b_bdev
->bd_disk
)
4101 set_metadata_rw_status(bh
->b_bdev
->bd_disk
->first_minor
, WAIT_READ_CNT
);
4103 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
4105 if (!buffer_uptodate(bh
)) {
4106 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4107 "unable to read itable block");
4117 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4119 /* We have all inode data except xattrs in memory here. */
4120 return __ext4_get_inode_loc(inode
, iloc
,
4121 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4124 void ext4_set_inode_flags(struct inode
*inode
)
4126 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4127 unsigned int new_fl
= 0;
4129 if (flags
& EXT4_SYNC_FL
)
4131 if (flags
& EXT4_APPEND_FL
)
4133 if (flags
& EXT4_IMMUTABLE_FL
)
4134 new_fl
|= S_IMMUTABLE
;
4135 if (flags
& EXT4_NOATIME_FL
)
4136 new_fl
|= S_NOATIME
;
4137 if (flags
& EXT4_DIRSYNC_FL
)
4138 new_fl
|= S_DIRSYNC
;
4139 set_mask_bits(&inode
->i_flags
,
4140 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
, new_fl
);
4143 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4144 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4146 unsigned int vfs_fl
;
4147 unsigned long old_fl
, new_fl
;
4150 vfs_fl
= ei
->vfs_inode
.i_flags
;
4151 old_fl
= ei
->i_flags
;
4152 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4153 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4155 if (vfs_fl
& S_SYNC
)
4156 new_fl
|= EXT4_SYNC_FL
;
4157 if (vfs_fl
& S_APPEND
)
4158 new_fl
|= EXT4_APPEND_FL
;
4159 if (vfs_fl
& S_IMMUTABLE
)
4160 new_fl
|= EXT4_IMMUTABLE_FL
;
4161 if (vfs_fl
& S_NOATIME
)
4162 new_fl
|= EXT4_NOATIME_FL
;
4163 if (vfs_fl
& S_DIRSYNC
)
4164 new_fl
|= EXT4_DIRSYNC_FL
;
4165 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4168 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4169 struct ext4_inode_info
*ei
)
4172 struct inode
*inode
= &(ei
->vfs_inode
);
4173 struct super_block
*sb
= inode
->i_sb
;
4175 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4176 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4177 /* we are using combined 48 bit field */
4178 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4179 le32_to_cpu(raw_inode
->i_blocks_lo
);
4180 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4181 /* i_blocks represent file system block size */
4182 return i_blocks
<< (inode
->i_blkbits
- 9);
4187 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4191 static inline void ext4_iget_extra_inode(struct inode
*inode
,
4192 struct ext4_inode
*raw_inode
,
4193 struct ext4_inode_info
*ei
)
4195 __le32
*magic
= (void *)raw_inode
+
4196 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4197 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4198 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4199 ext4_find_inline_data_nolock(inode
);
4201 EXT4_I(inode
)->i_inline_off
= 0;
4204 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4206 struct ext4_iloc iloc
;
4207 struct ext4_inode
*raw_inode
;
4208 struct ext4_inode_info
*ei
;
4209 struct inode
*inode
;
4210 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4216 inode
= iget_locked(sb
, ino
);
4218 return ERR_PTR(-ENOMEM
);
4219 if (!(inode
->i_state
& I_NEW
))
4225 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4228 raw_inode
= ext4_raw_inode(&iloc
);
4230 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4231 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4232 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4233 EXT4_INODE_SIZE(inode
->i_sb
)) {
4234 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
4235 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
4236 EXT4_INODE_SIZE(inode
->i_sb
));
4241 ei
->i_extra_isize
= 0;
4243 /* Precompute checksum seed for inode metadata */
4244 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4245 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
)) {
4246 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4248 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4249 __le32 gen
= raw_inode
->i_generation
;
4250 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4252 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4256 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4257 EXT4_ERROR_INODE(inode
, "checksum invalid");
4262 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4263 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4264 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4265 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4266 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4267 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4269 i_uid_write(inode
, i_uid
);
4270 i_gid_write(inode
, i_gid
);
4271 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4273 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4274 ei
->i_inline_off
= 0;
4275 ei
->i_dir_start_lookup
= 0;
4276 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4277 /* We now have enough fields to check if the inode was active or not.
4278 * This is needed because nfsd might try to access dead inodes
4279 * the test is that same one that e2fsck uses
4280 * NeilBrown 1999oct15
4282 if (inode
->i_nlink
== 0) {
4283 if ((inode
->i_mode
== 0 ||
4284 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4285 ino
!= EXT4_BOOT_LOADER_INO
) {
4286 /* this inode is deleted */
4290 /* The only unlinked inodes we let through here have
4291 * valid i_mode and are being read by the orphan
4292 * recovery code: that's fine, we're about to complete
4293 * the process of deleting those.
4294 * OR it is the EXT4_BOOT_LOADER_INO which is
4295 * not initialized on a new filesystem. */
4297 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4298 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4299 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4300 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4302 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4303 inode
->i_size
= ext4_isize(raw_inode
);
4304 ei
->i_disksize
= inode
->i_size
;
4306 ei
->i_reserved_quota
= 0;
4308 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4309 ei
->i_block_group
= iloc
.block_group
;
4310 ei
->i_last_alloc_group
= ~0;
4312 * NOTE! The in-memory inode i_data array is in little-endian order
4313 * even on big-endian machines: we do NOT byteswap the block numbers!
4315 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4316 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4317 INIT_LIST_HEAD(&ei
->i_orphan
);
4320 * Set transaction id's of transactions that have to be committed
4321 * to finish f[data]sync. We set them to currently running transaction
4322 * as we cannot be sure that the inode or some of its metadata isn't
4323 * part of the transaction - the inode could have been reclaimed and
4324 * now it is reread from disk.
4327 transaction_t
*transaction
;
4330 read_lock(&journal
->j_state_lock
);
4331 if (journal
->j_running_transaction
)
4332 transaction
= journal
->j_running_transaction
;
4334 transaction
= journal
->j_committing_transaction
;
4336 tid
= transaction
->t_tid
;
4338 tid
= journal
->j_commit_sequence
;
4339 read_unlock(&journal
->j_state_lock
);
4340 ei
->i_sync_tid
= tid
;
4341 ei
->i_datasync_tid
= tid
;
4344 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4345 if (ei
->i_extra_isize
== 0) {
4346 /* The extra space is currently unused. Use it. */
4347 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4348 EXT4_GOOD_OLD_INODE_SIZE
;
4350 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4354 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4355 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4356 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4357 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4359 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4360 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4361 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4363 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4367 if (ei
->i_file_acl
&&
4368 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4369 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4373 } else if (!ext4_has_inline_data(inode
)) {
4374 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4375 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4376 (S_ISLNK(inode
->i_mode
) &&
4377 !ext4_inode_is_fast_symlink(inode
))))
4378 /* Validate extent which is part of inode */
4379 ret
= ext4_ext_check_inode(inode
);
4380 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4381 (S_ISLNK(inode
->i_mode
) &&
4382 !ext4_inode_is_fast_symlink(inode
))) {
4383 /* Validate block references which are part of inode */
4384 ret
= ext4_ind_check_inode(inode
);
4390 if (S_ISREG(inode
->i_mode
)) {
4391 inode
->i_op
= &ext4_file_inode_operations
;
4392 inode
->i_fop
= &ext4_file_operations
;
4393 ext4_set_aops(inode
);
4394 } else if (S_ISDIR(inode
->i_mode
)) {
4395 inode
->i_op
= &ext4_dir_inode_operations
;
4396 inode
->i_fop
= &ext4_dir_operations
;
4397 } else if (S_ISLNK(inode
->i_mode
)) {
4398 if (ext4_inode_is_fast_symlink(inode
)) {
4399 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4400 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4401 sizeof(ei
->i_data
) - 1);
4403 inode
->i_op
= &ext4_symlink_inode_operations
;
4404 ext4_set_aops(inode
);
4406 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4407 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4408 inode
->i_op
= &ext4_special_inode_operations
;
4409 if (raw_inode
->i_block
[0])
4410 init_special_inode(inode
, inode
->i_mode
,
4411 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4413 init_special_inode(inode
, inode
->i_mode
,
4414 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4415 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4416 make_bad_inode(inode
);
4419 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4423 ext4_set_inode_flags(inode
);
4424 unlock_new_inode(inode
);
4430 return ERR_PTR(ret
);
4433 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4435 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4436 return ERR_PTR(-EIO
);
4437 return ext4_iget(sb
, ino
);
4440 static int ext4_inode_blocks_set(handle_t
*handle
,
4441 struct ext4_inode
*raw_inode
,
4442 struct ext4_inode_info
*ei
)
4444 struct inode
*inode
= &(ei
->vfs_inode
);
4445 u64 i_blocks
= inode
->i_blocks
;
4446 struct super_block
*sb
= inode
->i_sb
;
4448 if (i_blocks
<= ~0U) {
4450 * i_blocks can be represented in a 32 bit variable
4451 * as multiple of 512 bytes
4453 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4454 raw_inode
->i_blocks_high
= 0;
4455 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4458 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4461 if (i_blocks
<= 0xffffffffffffULL
) {
4463 * i_blocks can be represented in a 48 bit variable
4464 * as multiple of 512 bytes
4466 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4467 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4468 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4470 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4471 /* i_block is stored in file system block size */
4472 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4473 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4474 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4480 * Post the struct inode info into an on-disk inode location in the
4481 * buffer-cache. This gobbles the caller's reference to the
4482 * buffer_head in the inode location struct.
4484 * The caller must have write access to iloc->bh.
4486 static int ext4_do_update_inode(handle_t
*handle
,
4487 struct inode
*inode
,
4488 struct ext4_iloc
*iloc
)
4490 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4491 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4492 struct buffer_head
*bh
= iloc
->bh
;
4493 int err
= 0, rc
, block
;
4494 int need_datasync
= 0;
4498 /* For fields not not tracking in the in-memory inode,
4499 * initialise them to zero for new inodes. */
4500 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4501 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4503 ext4_get_inode_flags(ei
);
4504 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4505 i_uid
= i_uid_read(inode
);
4506 i_gid
= i_gid_read(inode
);
4507 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4508 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4509 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4511 * Fix up interoperability with old kernels. Otherwise, old inodes get
4512 * re-used with the upper 16 bits of the uid/gid intact
4515 raw_inode
->i_uid_high
=
4516 cpu_to_le16(high_16_bits(i_uid
));
4517 raw_inode
->i_gid_high
=
4518 cpu_to_le16(high_16_bits(i_gid
));
4520 raw_inode
->i_uid_high
= 0;
4521 raw_inode
->i_gid_high
= 0;
4524 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4525 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4526 raw_inode
->i_uid_high
= 0;
4527 raw_inode
->i_gid_high
= 0;
4529 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4531 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4532 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4533 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4534 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4536 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4538 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4539 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4540 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4541 cpu_to_le32(EXT4_OS_HURD
))
4542 raw_inode
->i_file_acl_high
=
4543 cpu_to_le16(ei
->i_file_acl
>> 32);
4544 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4545 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4546 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4549 if (ei
->i_disksize
> 0x7fffffffULL
) {
4550 struct super_block
*sb
= inode
->i_sb
;
4551 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4552 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4553 EXT4_SB(sb
)->s_es
->s_rev_level
==
4554 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4555 /* If this is the first large file
4556 * created, add a flag to the superblock.
4558 err
= ext4_journal_get_write_access(handle
,
4559 EXT4_SB(sb
)->s_sbh
);
4562 ext4_update_dynamic_rev(sb
);
4563 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4564 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4565 ext4_handle_sync(handle
);
4566 err
= ext4_handle_dirty_super(handle
, sb
);
4569 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4570 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4571 if (old_valid_dev(inode
->i_rdev
)) {
4572 raw_inode
->i_block
[0] =
4573 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4574 raw_inode
->i_block
[1] = 0;
4576 raw_inode
->i_block
[0] = 0;
4577 raw_inode
->i_block
[1] =
4578 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4579 raw_inode
->i_block
[2] = 0;
4581 } else if (!ext4_has_inline_data(inode
)) {
4582 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4583 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4586 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4587 if (ei
->i_extra_isize
) {
4588 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4589 raw_inode
->i_version_hi
=
4590 cpu_to_le32(inode
->i_version
>> 32);
4591 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4594 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4596 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4597 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4600 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4602 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4605 ext4_std_error(inode
->i_sb
, err
);
4610 * ext4_write_inode()
4612 * We are called from a few places:
4614 * - Within generic_file_write() for O_SYNC files.
4615 * Here, there will be no transaction running. We wait for any running
4616 * transaction to commit.
4618 * - Within sys_sync(), kupdate and such.
4619 * We wait on commit, if tol to.
4621 * - Within prune_icache() (PF_MEMALLOC == true)
4622 * Here we simply return. We can't afford to block kswapd on the
4625 * In all cases it is actually safe for us to return without doing anything,
4626 * because the inode has been copied into a raw inode buffer in
4627 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4630 * Note that we are absolutely dependent upon all inode dirtiers doing the
4631 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4632 * which we are interested.
4634 * It would be a bug for them to not do this. The code:
4636 * mark_inode_dirty(inode)
4638 * inode->i_size = expr;
4640 * is in error because a kswapd-driven write_inode() could occur while
4641 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4642 * will no longer be on the superblock's dirty inode list.
4644 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4648 if (current
->flags
& PF_MEMALLOC
)
4651 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4652 if (ext4_journal_current_handle()) {
4653 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4658 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
4661 err
= ext4_force_commit(inode
->i_sb
);
4663 struct ext4_iloc iloc
;
4665 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4668 if (wbc
->sync_mode
== WB_SYNC_ALL
)
4669 sync_dirty_buffer(iloc
.bh
);
4670 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4671 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4672 "IO error syncing inode");
4681 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4682 * buffers that are attached to a page stradding i_size and are undergoing
4683 * commit. In that case we have to wait for commit to finish and try again.
4685 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4689 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4690 tid_t commit_tid
= 0;
4693 offset
= inode
->i_size
& (PAGE_CACHE_SIZE
- 1);
4695 * All buffers in the last page remain valid? Then there's nothing to
4696 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4699 if (offset
> PAGE_CACHE_SIZE
- (1 << inode
->i_blkbits
))
4702 page
= find_lock_page(inode
->i_mapping
,
4703 inode
->i_size
>> PAGE_CACHE_SHIFT
);
4706 ret
= __ext4_journalled_invalidatepage(page
, offset
);
4708 page_cache_release(page
);
4712 read_lock(&journal
->j_state_lock
);
4713 if (journal
->j_committing_transaction
)
4714 commit_tid
= journal
->j_committing_transaction
->t_tid
;
4715 read_unlock(&journal
->j_state_lock
);
4717 jbd2_log_wait_commit(journal
, commit_tid
);
4724 * Called from notify_change.
4726 * We want to trap VFS attempts to truncate the file as soon as
4727 * possible. In particular, we want to make sure that when the VFS
4728 * shrinks i_size, we put the inode on the orphan list and modify
4729 * i_disksize immediately, so that during the subsequent flushing of
4730 * dirty pages and freeing of disk blocks, we can guarantee that any
4731 * commit will leave the blocks being flushed in an unused state on
4732 * disk. (On recovery, the inode will get truncated and the blocks will
4733 * be freed, so we have a strong guarantee that no future commit will
4734 * leave these blocks visible to the user.)
4736 * Another thing we have to assure is that if we are in ordered mode
4737 * and inode is still attached to the committing transaction, we must
4738 * we start writeout of all the dirty pages which are being truncated.
4739 * This way we are sure that all the data written in the previous
4740 * transaction are already on disk (truncate waits for pages under
4743 * Called with inode->i_mutex down.
4745 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4747 struct inode
*inode
= dentry
->d_inode
;
4750 const unsigned int ia_valid
= attr
->ia_valid
;
4752 error
= inode_change_ok(inode
, attr
);
4756 if (is_quota_modification(inode
, attr
))
4757 dquot_initialize(inode
);
4758 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4759 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4762 /* (user+group)*(old+new) structure, inode write (sb,
4763 * inode block, ? - but truncate inode update has it) */
4764 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
4765 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
4766 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
4767 if (IS_ERR(handle
)) {
4768 error
= PTR_ERR(handle
);
4771 error
= dquot_transfer(inode
, attr
);
4773 ext4_journal_stop(handle
);
4776 /* Update corresponding info in inode so that everything is in
4777 * one transaction */
4778 if (attr
->ia_valid
& ATTR_UID
)
4779 inode
->i_uid
= attr
->ia_uid
;
4780 if (attr
->ia_valid
& ATTR_GID
)
4781 inode
->i_gid
= attr
->ia_gid
;
4782 error
= ext4_mark_inode_dirty(handle
, inode
);
4783 ext4_journal_stop(handle
);
4786 if (attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
!= inode
->i_size
) {
4788 loff_t oldsize
= inode
->i_size
;
4790 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4791 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4793 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4797 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
4798 inode_inc_iversion(inode
);
4800 if (S_ISREG(inode
->i_mode
) &&
4801 (attr
->ia_size
< inode
->i_size
)) {
4802 if (ext4_should_order_data(inode
)) {
4803 error
= ext4_begin_ordered_truncate(inode
,
4808 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
4809 if (IS_ERR(handle
)) {
4810 error
= PTR_ERR(handle
);
4813 if (ext4_handle_valid(handle
)) {
4814 error
= ext4_orphan_add(handle
, inode
);
4817 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4818 rc
= ext4_mark_inode_dirty(handle
, inode
);
4821 ext4_journal_stop(handle
);
4823 ext4_orphan_del(NULL
, inode
);
4828 i_size_write(inode
, attr
->ia_size
);
4830 * Blocks are going to be removed from the inode. Wait
4831 * for dio in flight. Temporarily disable
4832 * dioread_nolock to prevent livelock.
4835 if (!ext4_should_journal_data(inode
)) {
4836 ext4_inode_block_unlocked_dio(inode
);
4837 inode_dio_wait(inode
);
4838 ext4_inode_resume_unlocked_dio(inode
);
4840 ext4_wait_for_tail_page_commit(inode
);
4843 * Truncate pagecache after we've waited for commit
4844 * in data=journal mode to make pages freeable.
4846 truncate_pagecache(inode
, oldsize
, inode
->i_size
);
4849 * We want to call ext4_truncate() even if attr->ia_size ==
4850 * inode->i_size for cases like truncation of fallocated space
4852 if (attr
->ia_valid
& ATTR_SIZE
)
4853 ext4_truncate(inode
);
4856 setattr_copy(inode
, attr
);
4857 mark_inode_dirty(inode
);
4861 * If the call to ext4_truncate failed to get a transaction handle at
4862 * all, we need to clean up the in-core orphan list manually.
4864 if (orphan
&& inode
->i_nlink
)
4865 ext4_orphan_del(NULL
, inode
);
4867 if (!rc
&& (ia_valid
& ATTR_MODE
))
4868 rc
= ext4_acl_chmod(inode
);
4871 ext4_std_error(inode
->i_sb
, error
);
4877 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4880 struct inode
*inode
;
4881 unsigned long long delalloc_blocks
;
4883 inode
= dentry
->d_inode
;
4884 generic_fillattr(inode
, stat
);
4887 * We can't update i_blocks if the block allocation is delayed
4888 * otherwise in the case of system crash before the real block
4889 * allocation is done, we will have i_blocks inconsistent with
4890 * on-disk file blocks.
4891 * We always keep i_blocks updated together with real
4892 * allocation. But to not confuse with user, stat
4893 * will return the blocks that include the delayed allocation
4894 * blocks for this file.
4896 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
4897 EXT4_I(inode
)->i_reserved_data_blocks
);
4899 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
-9);
4903 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4905 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4906 return ext4_ind_trans_blocks(inode
, nrblocks
, chunk
);
4907 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4911 * Account for index blocks, block groups bitmaps and block group
4912 * descriptor blocks if modify datablocks and index blocks
4913 * worse case, the indexs blocks spread over different block groups
4915 * If datablocks are discontiguous, they are possible to spread over
4916 * different block groups too. If they are contiguous, with flexbg,
4917 * they could still across block group boundary.
4919 * Also account for superblock, inode, quota and xattr blocks
4921 static int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4923 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4929 * How many index blocks need to touch to modify nrblocks?
4930 * The "Chunk" flag indicating whether the nrblocks is
4931 * physically contiguous on disk
4933 * For Direct IO and fallocate, they calls get_block to allocate
4934 * one single extent at a time, so they could set the "Chunk" flag
4936 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4941 * Now let's see how many group bitmaps and group descriptors need
4951 if (groups
> ngroups
)
4953 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4954 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4956 /* bitmaps and block group descriptor blocks */
4957 ret
+= groups
+ gdpblocks
;
4959 /* Blocks for super block, inode, quota and xattr blocks */
4960 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4966 * Calculate the total number of credits to reserve to fit
4967 * the modification of a single pages into a single transaction,
4968 * which may include multiple chunks of block allocations.
4970 * This could be called via ext4_write_begin()
4972 * We need to consider the worse case, when
4973 * one new block per extent.
4975 int ext4_writepage_trans_blocks(struct inode
*inode
)
4977 int bpp
= ext4_journal_blocks_per_page(inode
);
4980 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4982 /* Account for data blocks for journalled mode */
4983 if (ext4_should_journal_data(inode
))
4989 * Calculate the journal credits for a chunk of data modification.
4991 * This is called from DIO, fallocate or whoever calling
4992 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4994 * journal buffers for data blocks are not included here, as DIO
4995 * and fallocate do no need to journal data buffers.
4997 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4999 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
5003 * The caller must have previously called ext4_reserve_inode_write().
5004 * Give this, we know that the caller already has write access to iloc->bh.
5006 int ext4_mark_iloc_dirty(handle_t
*handle
,
5007 struct inode
*inode
, struct ext4_iloc
*iloc
)
5011 if (IS_I_VERSION(inode
))
5012 inode_inc_iversion(inode
);
5014 /* the do_update_inode consumes one bh->b_count */
5017 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
5018 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5024 * On success, We end up with an outstanding reference count against
5025 * iloc->bh. This _must_ be cleaned up later.
5029 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5030 struct ext4_iloc
*iloc
)
5034 err
= ext4_get_inode_loc(inode
, iloc
);
5036 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5037 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5043 ext4_std_error(inode
->i_sb
, err
);
5048 * Expand an inode by new_extra_isize bytes.
5049 * Returns 0 on success or negative error number on failure.
5051 static int ext4_expand_extra_isize(struct inode
*inode
,
5052 unsigned int new_extra_isize
,
5053 struct ext4_iloc iloc
,
5056 struct ext4_inode
*raw_inode
;
5057 struct ext4_xattr_ibody_header
*header
;
5059 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5062 raw_inode
= ext4_raw_inode(&iloc
);
5064 header
= IHDR(inode
, raw_inode
);
5066 /* No extended attributes present */
5067 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5068 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5069 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5071 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5075 /* try to expand with EAs present */
5076 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5081 * What we do here is to mark the in-core inode as clean with respect to inode
5082 * dirtiness (it may still be data-dirty).
5083 * This means that the in-core inode may be reaped by prune_icache
5084 * without having to perform any I/O. This is a very good thing,
5085 * because *any* task may call prune_icache - even ones which
5086 * have a transaction open against a different journal.
5088 * Is this cheating? Not really. Sure, we haven't written the
5089 * inode out, but prune_icache isn't a user-visible syncing function.
5090 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5091 * we start and wait on commits.
5093 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5095 struct ext4_iloc iloc
;
5096 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5097 static unsigned int mnt_count
;
5101 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5102 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5105 if (ext4_handle_valid(handle
) &&
5106 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5107 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5109 * We need extra buffer credits since we may write into EA block
5110 * with this same handle. If journal_extend fails, then it will
5111 * only result in a minor loss of functionality for that inode.
5112 * If this is felt to be critical, then e2fsck should be run to
5113 * force a large enough s_min_extra_isize.
5115 if ((jbd2_journal_extend(handle
,
5116 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5117 ret
= ext4_expand_extra_isize(inode
,
5118 sbi
->s_want_extra_isize
,
5121 ext4_set_inode_state(inode
,
5122 EXT4_STATE_NO_EXPAND
);
5124 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5125 ext4_warning(inode
->i_sb
,
5126 "Unable to expand inode %lu. Delete"
5127 " some EAs or run e2fsck.",
5130 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5135 return ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5139 * ext4_dirty_inode() is called from __mark_inode_dirty()
5141 * We're really interested in the case where a file is being extended.
5142 * i_size has been changed by generic_commit_write() and we thus need
5143 * to include the updated inode in the current transaction.
5145 * Also, dquot_alloc_block() will always dirty the inode when blocks
5146 * are allocated to the file.
5148 * If the inode is marked synchronous, we don't honour that here - doing
5149 * so would cause a commit on atime updates, which we don't bother doing.
5150 * We handle synchronous inodes at the highest possible level.
5152 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5156 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5160 ext4_mark_inode_dirty(handle
, inode
);
5162 ext4_journal_stop(handle
);
5169 * Bind an inode's backing buffer_head into this transaction, to prevent
5170 * it from being flushed to disk early. Unlike
5171 * ext4_reserve_inode_write, this leaves behind no bh reference and
5172 * returns no iloc structure, so the caller needs to repeat the iloc
5173 * lookup to mark the inode dirty later.
5175 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5177 struct ext4_iloc iloc
;
5181 err
= ext4_get_inode_loc(inode
, &iloc
);
5183 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5184 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5186 err
= ext4_handle_dirty_metadata(handle
,
5192 ext4_std_error(inode
->i_sb
, err
);
5197 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5204 * We have to be very careful here: changing a data block's
5205 * journaling status dynamically is dangerous. If we write a
5206 * data block to the journal, change the status and then delete
5207 * that block, we risk forgetting to revoke the old log record
5208 * from the journal and so a subsequent replay can corrupt data.
5209 * So, first we make sure that the journal is empty and that
5210 * nobody is changing anything.
5213 journal
= EXT4_JOURNAL(inode
);
5216 if (is_journal_aborted(journal
))
5218 /* We have to allocate physical blocks for delalloc blocks
5219 * before flushing journal. otherwise delalloc blocks can not
5220 * be allocated any more. even more truncate on delalloc blocks
5221 * could trigger BUG by flushing delalloc blocks in journal.
5222 * There is no delalloc block in non-journal data mode.
5224 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
5225 err
= ext4_alloc_da_blocks(inode
);
5230 /* Wait for all existing dio workers */
5231 ext4_inode_block_unlocked_dio(inode
);
5232 inode_dio_wait(inode
);
5234 jbd2_journal_lock_updates(journal
);
5237 * OK, there are no updates running now, and all cached data is
5238 * synced to disk. We are now in a completely consistent state
5239 * which doesn't have anything in the journal, and we know that
5240 * no filesystem updates are running, so it is safe to modify
5241 * the inode's in-core data-journaling state flag now.
5245 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5247 jbd2_journal_flush(journal
);
5248 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5250 ext4_set_aops(inode
);
5252 jbd2_journal_unlock_updates(journal
);
5253 ext4_inode_resume_unlocked_dio(inode
);
5255 /* Finally we can mark the inode as dirty. */
5257 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5259 return PTR_ERR(handle
);
5261 err
= ext4_mark_inode_dirty(handle
, inode
);
5262 ext4_handle_sync(handle
);
5263 ext4_journal_stop(handle
);
5264 ext4_std_error(inode
->i_sb
, err
);
5269 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5271 return !buffer_mapped(bh
);
5274 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5276 struct page
*page
= vmf
->page
;
5280 struct file
*file
= vma
->vm_file
;
5281 struct inode
*inode
= file_inode(file
);
5282 struct address_space
*mapping
= inode
->i_mapping
;
5284 get_block_t
*get_block
;
5287 sb_start_pagefault(inode
->i_sb
);
5288 file_update_time(vma
->vm_file
);
5289 /* Delalloc case is easy... */
5290 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5291 !ext4_should_journal_data(inode
) &&
5292 !ext4_nonda_switch(inode
->i_sb
)) {
5294 ret
= __block_page_mkwrite(vma
, vmf
,
5295 ext4_da_get_block_prep
);
5296 } while (ret
== -ENOSPC
&&
5297 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5302 size
= i_size_read(inode
);
5303 /* Page got truncated from under us? */
5304 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5306 ret
= VM_FAULT_NOPAGE
;
5310 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5311 len
= size
& ~PAGE_CACHE_MASK
;
5313 len
= PAGE_CACHE_SIZE
;
5315 * Return if we have all the buffers mapped. This avoids the need to do
5316 * journal_start/journal_stop which can block and take a long time
5318 if (page_has_buffers(page
)) {
5319 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5321 ext4_bh_unmapped
)) {
5322 /* Wait so that we don't change page under IO */
5323 wait_for_stable_page(page
);
5324 ret
= VM_FAULT_LOCKED
;
5329 /* OK, we need to fill the hole... */
5330 if (ext4_should_dioread_nolock(inode
))
5331 get_block
= ext4_get_block_write
;
5333 get_block
= ext4_get_block
;
5335 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5336 ext4_writepage_trans_blocks(inode
));
5337 if (IS_ERR(handle
)) {
5338 ret
= VM_FAULT_SIGBUS
;
5341 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
5342 if (!ret
&& ext4_should_journal_data(inode
)) {
5343 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5344 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
5346 ret
= VM_FAULT_SIGBUS
;
5347 ext4_journal_stop(handle
);
5350 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5352 ext4_journal_stop(handle
);
5353 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5356 ret
= block_page_mkwrite_return(ret
);
5358 sb_end_pagefault(inode
->i_sb
);