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 (ext4_should_journal_data(inode
) &&
211 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
)) &&
212 inode
->i_ino
!= EXT4_JOURNAL_INO
) {
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
)
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
))) {
1456 clear_buffer_delay(bh
);
1458 curr_off
= next_off
;
1459 } while ((bh
= bh
->b_this_page
) != head
);
1462 lblk
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1463 ext4_es_remove_extent(inode
, lblk
, to_release
);
1466 /* If we have released all the blocks belonging to a cluster, then we
1467 * need to release the reserved space for that cluster. */
1468 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1469 while (num_clusters
> 0) {
1470 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1471 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1472 if (sbi
->s_cluster_ratio
== 1 ||
1473 !ext4_find_delalloc_cluster(inode
, lblk
))
1474 ext4_da_release_space(inode
, 1);
1481 * Delayed allocation stuff
1485 * mpage_da_submit_io - walks through extent of pages and try to write
1486 * them with writepage() call back
1488 * @mpd->inode: inode
1489 * @mpd->first_page: first page of the extent
1490 * @mpd->next_page: page after the last page of the extent
1492 * By the time mpage_da_submit_io() is called we expect all blocks
1493 * to be allocated. this may be wrong if allocation failed.
1495 * As pages are already locked by write_cache_pages(), we can't use it
1497 static int mpage_da_submit_io(struct mpage_da_data
*mpd
,
1498 struct ext4_map_blocks
*map
)
1500 struct pagevec pvec
;
1501 unsigned long index
, end
;
1502 int ret
= 0, err
, nr_pages
, i
;
1503 struct inode
*inode
= mpd
->inode
;
1504 struct address_space
*mapping
= inode
->i_mapping
;
1505 loff_t size
= i_size_read(inode
);
1506 unsigned int len
, block_start
;
1507 struct buffer_head
*bh
, *page_bufs
= NULL
;
1508 sector_t pblock
= 0, cur_logical
= 0;
1509 struct ext4_io_submit io_submit
;
1511 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1512 memset(&io_submit
, 0, sizeof(io_submit
));
1514 * We need to start from the first_page to the next_page - 1
1515 * to make sure we also write the mapped dirty buffer_heads.
1516 * If we look at mpd->b_blocknr we would only be looking
1517 * at the currently mapped buffer_heads.
1519 index
= mpd
->first_page
;
1520 end
= mpd
->next_page
- 1;
1522 pagevec_init(&pvec
, 0);
1523 while (index
<= end
) {
1524 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1527 for (i
= 0; i
< nr_pages
; i
++) {
1529 struct page
*page
= pvec
.pages
[i
];
1531 index
= page
->index
;
1535 if (index
== size
>> PAGE_CACHE_SHIFT
)
1536 len
= size
& ~PAGE_CACHE_MASK
;
1538 len
= PAGE_CACHE_SIZE
;
1540 cur_logical
= index
<< (PAGE_CACHE_SHIFT
-
1542 pblock
= map
->m_pblk
+ (cur_logical
-
1547 BUG_ON(!PageLocked(page
));
1548 BUG_ON(PageWriteback(page
));
1550 bh
= page_bufs
= page_buffers(page
);
1553 if (map
&& (cur_logical
>= map
->m_lblk
) &&
1554 (cur_logical
<= (map
->m_lblk
+
1555 (map
->m_len
- 1)))) {
1556 if (buffer_delay(bh
)) {
1557 clear_buffer_delay(bh
);
1558 bh
->b_blocknr
= pblock
;
1560 if (buffer_unwritten(bh
) ||
1562 BUG_ON(bh
->b_blocknr
!= pblock
);
1563 if (map
->m_flags
& EXT4_MAP_UNINIT
)
1564 set_buffer_uninit(bh
);
1565 clear_buffer_unwritten(bh
);
1569 * skip page if block allocation undone and
1572 if (ext4_bh_delay_or_unwritten(NULL
, bh
))
1574 bh
= bh
->b_this_page
;
1575 block_start
+= bh
->b_size
;
1578 } while (bh
!= page_bufs
);
1585 clear_page_dirty_for_io(page
);
1586 err
= ext4_bio_write_page(&io_submit
, page
, len
,
1589 mpd
->pages_written
++;
1591 * In error case, we have to continue because
1592 * remaining pages are still locked
1597 pagevec_release(&pvec
);
1599 ext4_io_submit(&io_submit
);
1603 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
)
1607 struct pagevec pvec
;
1608 struct inode
*inode
= mpd
->inode
;
1609 struct address_space
*mapping
= inode
->i_mapping
;
1610 ext4_lblk_t start
, last
;
1612 index
= mpd
->first_page
;
1613 end
= mpd
->next_page
- 1;
1615 start
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1616 last
= end
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1617 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1619 pagevec_init(&pvec
, 0);
1620 while (index
<= end
) {
1621 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1624 for (i
= 0; i
< nr_pages
; i
++) {
1625 struct page
*page
= pvec
.pages
[i
];
1626 if (page
->index
> end
)
1628 BUG_ON(!PageLocked(page
));
1629 BUG_ON(PageWriteback(page
));
1630 block_invalidatepage(page
, 0);
1631 ClearPageUptodate(page
);
1634 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1635 pagevec_release(&pvec
);
1640 static void ext4_print_free_blocks(struct inode
*inode
)
1642 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1643 struct super_block
*sb
= inode
->i_sb
;
1644 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1646 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1647 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1648 ext4_count_free_clusters(sb
)));
1649 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1650 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1651 (long long) EXT4_C2B(EXT4_SB(sb
),
1652 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1653 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1654 (long long) EXT4_C2B(EXT4_SB(sb
),
1655 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1656 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1657 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1658 ei
->i_reserved_data_blocks
);
1659 ext4_msg(sb
, KERN_CRIT
, "i_reserved_meta_blocks=%u",
1660 ei
->i_reserved_meta_blocks
);
1661 ext4_msg(sb
, KERN_CRIT
, "i_allocated_meta_blocks=%u",
1662 ei
->i_allocated_meta_blocks
);
1667 * mpage_da_map_and_submit - go through given space, map them
1668 * if necessary, and then submit them for I/O
1670 * @mpd - bh describing space
1672 * The function skips space we know is already mapped to disk blocks.
1675 static void mpage_da_map_and_submit(struct mpage_da_data
*mpd
)
1677 int err
, blks
, get_blocks_flags
;
1678 struct ext4_map_blocks map
, *mapp
= NULL
;
1679 sector_t next
= mpd
->b_blocknr
;
1680 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1681 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
1682 handle_t
*handle
= NULL
;
1685 * If the blocks are mapped already, or we couldn't accumulate
1686 * any blocks, then proceed immediately to the submission stage.
1688 if ((mpd
->b_size
== 0) ||
1689 ((mpd
->b_state
& (1 << BH_Mapped
)) &&
1690 !(mpd
->b_state
& (1 << BH_Delay
)) &&
1691 !(mpd
->b_state
& (1 << BH_Unwritten
))))
1694 handle
= ext4_journal_current_handle();
1698 * Call ext4_map_blocks() to allocate any delayed allocation
1699 * blocks, or to convert an uninitialized extent to be
1700 * initialized (in the case where we have written into
1701 * one or more preallocated blocks).
1703 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1704 * indicate that we are on the delayed allocation path. This
1705 * affects functions in many different parts of the allocation
1706 * call path. This flag exists primarily because we don't
1707 * want to change *many* call functions, so ext4_map_blocks()
1708 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1709 * inode's allocation semaphore is taken.
1711 * If the blocks in questions were delalloc blocks, set
1712 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1713 * variables are updated after the blocks have been allocated.
1716 map
.m_len
= max_blocks
;
1718 * We're in delalloc path and it is possible that we're going to
1719 * need more metadata blocks than previously reserved. However
1720 * we must not fail because we're in writeback and there is
1721 * nothing we can do about it so it might result in data loss.
1722 * So use reserved blocks to allocate metadata if possible.
1724 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
|
1725 EXT4_GET_BLOCKS_METADATA_NOFAIL
;
1726 if (ext4_should_dioread_nolock(mpd
->inode
))
1727 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
1728 if (mpd
->b_state
& (1 << BH_Delay
))
1729 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
1732 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
1734 struct super_block
*sb
= mpd
->inode
->i_sb
;
1738 * If get block returns EAGAIN or ENOSPC and there
1739 * appears to be free blocks we will just let
1740 * mpage_da_submit_io() unlock all of the pages.
1745 if (err
== -ENOSPC
&& ext4_count_free_clusters(sb
)) {
1751 * get block failure will cause us to loop in
1752 * writepages, because a_ops->writepage won't be able
1753 * to make progress. The page will be redirtied by
1754 * writepage and writepages will again try to write
1757 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
1758 ext4_msg(sb
, KERN_CRIT
,
1759 "delayed block allocation failed for inode %lu "
1760 "at logical offset %llu with max blocks %zd "
1761 "with error %d", mpd
->inode
->i_ino
,
1762 (unsigned long long) next
,
1763 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1764 ext4_msg(sb
, KERN_CRIT
,
1765 "This should not happen!! Data will be lost");
1767 ext4_print_free_blocks(mpd
->inode
);
1769 /* invalidate all the pages */
1770 ext4_da_block_invalidatepages(mpd
);
1772 /* Mark this page range as having been completed */
1779 if (map
.m_flags
& EXT4_MAP_NEW
) {
1780 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
1783 for (i
= 0; i
< map
.m_len
; i
++)
1784 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
1788 * Update on-disk size along with block allocation.
1790 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
1791 if (disksize
> i_size_read(mpd
->inode
))
1792 disksize
= i_size_read(mpd
->inode
);
1793 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
1794 ext4_update_i_disksize(mpd
->inode
, disksize
);
1795 err
= ext4_mark_inode_dirty(handle
, mpd
->inode
);
1797 ext4_error(mpd
->inode
->i_sb
,
1798 "Failed to mark inode %lu dirty",
1803 mpage_da_submit_io(mpd
, mapp
);
1807 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1808 (1 << BH_Delay) | (1 << BH_Unwritten))
1811 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1813 * @mpd->lbh - extent of blocks
1814 * @logical - logical number of the block in the file
1815 * @b_state - b_state of the buffer head added
1817 * the function is used to collect contig. blocks in same state
1819 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
, sector_t logical
,
1820 unsigned long b_state
)
1823 int blkbits
= mpd
->inode
->i_blkbits
;
1824 int nrblocks
= mpd
->b_size
>> blkbits
;
1827 * XXX Don't go larger than mballoc is willing to allocate
1828 * This is a stopgap solution. We eventually need to fold
1829 * mpage_da_submit_io() into this function and then call
1830 * ext4_map_blocks() multiple times in a loop
1832 if (nrblocks
>= (8*1024*1024 >> blkbits
))
1835 /* check if the reserved journal credits might overflow */
1836 if (!ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
)) {
1837 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1839 * With non-extent format we are limited by the journal
1840 * credit available. Total credit needed to insert
1841 * nrblocks contiguous blocks is dependent on the
1842 * nrblocks. So limit nrblocks.
1848 * First block in the extent
1850 if (mpd
->b_size
== 0) {
1851 mpd
->b_blocknr
= logical
;
1852 mpd
->b_size
= 1 << blkbits
;
1853 mpd
->b_state
= b_state
& BH_FLAGS
;
1857 next
= mpd
->b_blocknr
+ nrblocks
;
1859 * Can we merge the block to our big extent?
1861 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
1862 mpd
->b_size
+= 1 << blkbits
;
1868 * We couldn't merge the block to our extent, so we
1869 * need to flush current extent and start new one
1871 mpage_da_map_and_submit(mpd
);
1875 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1877 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1881 * This function is grabs code from the very beginning of
1882 * ext4_map_blocks, but assumes that the caller is from delayed write
1883 * time. This function looks up the requested blocks and sets the
1884 * buffer delay bit under the protection of i_data_sem.
1886 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1887 struct ext4_map_blocks
*map
,
1888 struct buffer_head
*bh
)
1890 struct extent_status es
;
1892 sector_t invalid_block
= ~((sector_t
) 0xffff);
1893 #ifdef ES_AGGRESSIVE_TEST
1894 struct ext4_map_blocks orig_map
;
1896 memcpy(&orig_map
, map
, sizeof(*map
));
1899 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1903 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1904 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1905 (unsigned long) map
->m_lblk
);
1907 /* Lookup extent status tree firstly */
1908 if (ext4_es_lookup_extent(inode
, iblock
, &es
)) {
1910 if (ext4_es_is_hole(&es
)) {
1912 down_read((&EXT4_I(inode
)->i_data_sem
));
1917 * Delayed extent could be allocated by fallocate.
1918 * So we need to check it.
1920 if (ext4_es_is_delayed(&es
) && !ext4_es_is_unwritten(&es
)) {
1921 map_bh(bh
, inode
->i_sb
, invalid_block
);
1923 set_buffer_delay(bh
);
1927 map
->m_pblk
= ext4_es_pblock(&es
) + iblock
- es
.es_lblk
;
1928 retval
= es
.es_len
- (iblock
- es
.es_lblk
);
1929 if (retval
> map
->m_len
)
1930 retval
= map
->m_len
;
1931 map
->m_len
= retval
;
1932 if (ext4_es_is_written(&es
))
1933 map
->m_flags
|= EXT4_MAP_MAPPED
;
1934 else if (ext4_es_is_unwritten(&es
))
1935 map
->m_flags
|= EXT4_MAP_UNWRITTEN
;
1939 #ifdef ES_AGGRESSIVE_TEST
1940 ext4_map_blocks_es_recheck(NULL
, inode
, map
, &orig_map
, 0);
1946 * Try to see if we can get the block without requesting a new
1947 * file system block.
1949 down_read((&EXT4_I(inode
)->i_data_sem
));
1950 if (ext4_has_inline_data(inode
)) {
1952 * We will soon create blocks for this page, and let
1953 * us pretend as if the blocks aren't allocated yet.
1954 * In case of clusters, we have to handle the work
1955 * of mapping from cluster so that the reserved space
1956 * is calculated properly.
1958 if ((EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) &&
1959 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
1960 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
1962 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1963 retval
= ext4_ext_map_blocks(NULL
, inode
, map
,
1964 EXT4_GET_BLOCKS_NO_PUT_HOLE
);
1966 retval
= ext4_ind_map_blocks(NULL
, inode
, map
,
1967 EXT4_GET_BLOCKS_NO_PUT_HOLE
);
1973 * XXX: __block_prepare_write() unmaps passed block,
1977 * If the block was allocated from previously allocated cluster,
1978 * then we don't need to reserve it again. However we still need
1979 * to reserve metadata for every block we're going to write.
1981 if (!(map
->m_flags
& EXT4_MAP_FROM_CLUSTER
)) {
1982 ret
= ext4_da_reserve_space(inode
, iblock
);
1984 /* not enough space to reserve */
1989 ret
= ext4_da_reserve_metadata(inode
, iblock
);
1991 /* not enough space to reserve */
1997 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
1998 ~0, EXTENT_STATUS_DELAYED
);
2004 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
2005 * and it should not appear on the bh->b_state.
2007 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
2009 map_bh(bh
, inode
->i_sb
, invalid_block
);
2011 set_buffer_delay(bh
);
2012 } else if (retval
> 0) {
2014 unsigned long long status
;
2016 #ifdef ES_AGGRESSIVE_TEST
2017 if (retval
!= map
->m_len
) {
2018 printk("ES len assertation failed for inode: %lu "
2019 "retval %d != map->m_len %d "
2020 "in %s (lookup)\n", inode
->i_ino
, retval
,
2021 map
->m_len
, __func__
);
2025 status
= map
->m_flags
& EXT4_MAP_UNWRITTEN
?
2026 EXTENT_STATUS_UNWRITTEN
: EXTENT_STATUS_WRITTEN
;
2027 ret
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
,
2028 map
->m_pblk
, status
);
2034 up_read((&EXT4_I(inode
)->i_data_sem
));
2040 * This is a special get_blocks_t callback which is used by
2041 * ext4_da_write_begin(). It will either return mapped block or
2042 * reserve space for a single block.
2044 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
2045 * We also have b_blocknr = -1 and b_bdev initialized properly
2047 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
2048 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
2049 * initialized properly.
2051 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
2052 struct buffer_head
*bh
, int create
)
2054 struct ext4_map_blocks map
;
2057 BUG_ON(create
== 0);
2058 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
2060 map
.m_lblk
= iblock
;
2064 * first, we need to know whether the block is allocated already
2065 * preallocated blocks are unmapped but should treated
2066 * the same as allocated blocks.
2068 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
2072 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
2073 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
2075 if (buffer_unwritten(bh
)) {
2076 /* A delayed write to unwritten bh should be marked
2077 * new and mapped. Mapped ensures that we don't do
2078 * get_block multiple times when we write to the same
2079 * offset and new ensures that we do proper zero out
2080 * for partial write.
2083 set_buffer_mapped(bh
);
2088 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
2094 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
2100 static int __ext4_journalled_writepage(struct page
*page
,
2103 struct address_space
*mapping
= page
->mapping
;
2104 struct inode
*inode
= mapping
->host
;
2105 struct buffer_head
*page_bufs
= NULL
;
2106 handle_t
*handle
= NULL
;
2107 int ret
= 0, err
= 0;
2108 int inline_data
= ext4_has_inline_data(inode
);
2109 struct buffer_head
*inode_bh
= NULL
;
2111 ClearPageChecked(page
);
2114 BUG_ON(page
->index
!= 0);
2115 BUG_ON(len
> ext4_get_max_inline_size(inode
));
2116 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
2117 if (inode_bh
== NULL
)
2120 page_bufs
= page_buffers(page
);
2125 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
2128 /* As soon as we unlock the page, it can go away, but we have
2129 * references to buffers so we are safe */
2132 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2133 ext4_writepage_trans_blocks(inode
));
2134 if (IS_ERR(handle
)) {
2135 ret
= PTR_ERR(handle
);
2139 BUG_ON(!ext4_handle_valid(handle
));
2142 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
2144 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
2147 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2148 do_journal_get_write_access
);
2150 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
2155 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
2156 err
= ext4_journal_stop(handle
);
2160 if (!ext4_has_inline_data(inode
))
2161 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
2163 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
2170 * Note that we don't need to start a transaction unless we're journaling data
2171 * because we should have holes filled from ext4_page_mkwrite(). We even don't
2172 * need to file the inode to the transaction's list in ordered mode because if
2173 * we are writing back data added by write(), the inode is already there and if
2174 * we are writing back data modified via mmap(), no one guarantees in which
2175 * transaction the data will hit the disk. In case we are journaling data, we
2176 * cannot start transaction directly because transaction start ranks above page
2177 * lock so we have to do some magic.
2179 * This function can get called via...
2180 * - ext4_da_writepages after taking page lock (have journal handle)
2181 * - journal_submit_inode_data_buffers (no journal handle)
2182 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
2183 * - grab_page_cache when doing write_begin (have journal handle)
2185 * We don't do any block allocation in this function. If we have page with
2186 * multiple blocks we need to write those buffer_heads that are mapped. This
2187 * is important for mmaped based write. So if we do with blocksize 1K
2188 * truncate(f, 1024);
2189 * a = mmap(f, 0, 4096);
2191 * truncate(f, 4096);
2192 * we have in the page first buffer_head mapped via page_mkwrite call back
2193 * but other buffer_heads would be unmapped but dirty (dirty done via the
2194 * do_wp_page). So writepage should write the first block. If we modify
2195 * the mmap area beyond 1024 we will again get a page_fault and the
2196 * page_mkwrite callback will do the block allocation and mark the
2197 * buffer_heads mapped.
2199 * We redirty the page if we have any buffer_heads that is either delay or
2200 * unwritten in the page.
2202 * We can get recursively called as show below.
2204 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2207 * But since we don't do any block allocation we should not deadlock.
2208 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2210 static int ext4_writepage(struct page
*page
,
2211 struct writeback_control
*wbc
)
2216 struct buffer_head
*page_bufs
= NULL
;
2217 struct inode
*inode
= page
->mapping
->host
;
2218 struct ext4_io_submit io_submit
;
2220 trace_ext4_writepage(page
);
2221 size
= i_size_read(inode
);
2222 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2223 len
= size
& ~PAGE_CACHE_MASK
;
2225 len
= PAGE_CACHE_SIZE
;
2227 page_bufs
= page_buffers(page
);
2229 * We cannot do block allocation or other extent handling in this
2230 * function. If there are buffers needing that, we have to redirty
2231 * the page. But we may reach here when we do a journal commit via
2232 * journal_submit_inode_data_buffers() and in that case we must write
2233 * allocated buffers to achieve data=ordered mode guarantees.
2235 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2236 ext4_bh_delay_or_unwritten
)) {
2237 redirty_page_for_writepage(wbc
, page
);
2238 if (current
->flags
& PF_MEMALLOC
) {
2240 * For memory cleaning there's no point in writing only
2241 * some buffers. So just bail out. Warn if we came here
2242 * from direct reclaim.
2244 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
))
2251 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2253 * It's mmapped pagecache. Add buffers and journal it. There
2254 * doesn't seem much point in redirtying the page here.
2256 return __ext4_journalled_writepage(page
, len
);
2258 memset(&io_submit
, 0, sizeof(io_submit
));
2259 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
);
2260 ext4_io_submit(&io_submit
);
2265 * This is called via ext4_da_writepages() to
2266 * calculate the total number of credits to reserve to fit
2267 * a single extent allocation into a single transaction,
2268 * ext4_da_writpeages() will loop calling this before
2269 * the block allocation.
2272 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2274 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2277 * With non-extent format the journal credit needed to
2278 * insert nrblocks contiguous block is dependent on
2279 * number of contiguous block. So we will limit
2280 * number of contiguous block to a sane value
2282 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
2283 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2284 max_blocks
= EXT4_MAX_TRANS_DATA
;
2286 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2290 * write_cache_pages_da - walk the list of dirty pages of the given
2291 * address space and accumulate pages that need writing, and call
2292 * mpage_da_map_and_submit to map a single contiguous memory region
2293 * and then write them.
2295 static int write_cache_pages_da(handle_t
*handle
,
2296 struct address_space
*mapping
,
2297 struct writeback_control
*wbc
,
2298 struct mpage_da_data
*mpd
,
2299 pgoff_t
*done_index
)
2301 struct buffer_head
*bh
, *head
;
2302 struct inode
*inode
= mapping
->host
;
2303 struct pagevec pvec
;
2304 unsigned int nr_pages
;
2307 long nr_to_write
= wbc
->nr_to_write
;
2308 int i
, tag
, ret
= 0;
2310 memset(mpd
, 0, sizeof(struct mpage_da_data
));
2313 pagevec_init(&pvec
, 0);
2314 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2315 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2317 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2318 tag
= PAGECACHE_TAG_TOWRITE
;
2320 tag
= PAGECACHE_TAG_DIRTY
;
2322 *done_index
= index
;
2323 while (index
<= end
) {
2324 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2325 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2329 for (i
= 0; i
< nr_pages
; i
++) {
2330 struct page
*page
= pvec
.pages
[i
];
2333 * At this point, the page may be truncated or
2334 * invalidated (changing page->mapping to NULL), or
2335 * even swizzled back from swapper_space to tmpfs file
2336 * mapping. However, page->index will not change
2337 * because we have a reference on the page.
2339 if (page
->index
> end
)
2342 *done_index
= page
->index
+ 1;
2345 * If we can't merge this page, and we have
2346 * accumulated an contiguous region, write it
2348 if ((mpd
->next_page
!= page
->index
) &&
2349 (mpd
->next_page
!= mpd
->first_page
)) {
2350 mpage_da_map_and_submit(mpd
);
2351 goto ret_extent_tail
;
2357 * If the page is no longer dirty, or its
2358 * mapping no longer corresponds to inode we
2359 * are writing (which means it has been
2360 * truncated or invalidated), or the page is
2361 * already under writeback and we are not
2362 * doing a data integrity writeback, skip the page
2364 if (!PageDirty(page
) ||
2365 (PageWriteback(page
) &&
2366 (wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2367 unlikely(page
->mapping
!= mapping
)) {
2372 wait_on_page_writeback(page
);
2373 BUG_ON(PageWriteback(page
));
2376 * If we have inline data and arrive here, it means that
2377 * we will soon create the block for the 1st page, so
2378 * we'd better clear the inline data here.
2380 if (ext4_has_inline_data(inode
)) {
2381 BUG_ON(ext4_test_inode_state(inode
,
2382 EXT4_STATE_MAY_INLINE_DATA
));
2383 ext4_destroy_inline_data(handle
, inode
);
2386 if (mpd
->next_page
!= page
->index
)
2387 mpd
->first_page
= page
->index
;
2388 mpd
->next_page
= page
->index
+ 1;
2389 logical
= (sector_t
) page
->index
<<
2390 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2392 /* Add all dirty buffers to mpd */
2393 head
= page_buffers(page
);
2396 BUG_ON(buffer_locked(bh
));
2398 * We need to try to allocate unmapped blocks
2399 * in the same page. Otherwise we won't make
2400 * progress with the page in ext4_writepage
2402 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2403 mpage_add_bh_to_extent(mpd
, logical
,
2406 goto ret_extent_tail
;
2407 } else if (buffer_dirty(bh
) &&
2408 buffer_mapped(bh
)) {
2410 * mapped dirty buffer. We need to
2411 * update the b_state because we look
2412 * at b_state in mpage_da_map_blocks.
2413 * We don't update b_size because if we
2414 * find an unmapped buffer_head later
2415 * we need to use the b_state flag of
2418 if (mpd
->b_size
== 0)
2420 bh
->b_state
& BH_FLAGS
;
2423 } while ((bh
= bh
->b_this_page
) != head
);
2425 if (nr_to_write
> 0) {
2427 if (nr_to_write
== 0 &&
2428 wbc
->sync_mode
== WB_SYNC_NONE
)
2430 * We stop writing back only if we are
2431 * not doing integrity sync. In case of
2432 * integrity sync we have to keep going
2433 * because someone may be concurrently
2434 * dirtying pages, and we might have
2435 * synced a lot of newly appeared dirty
2436 * pages, but have not synced all of the
2442 pagevec_release(&pvec
);
2447 ret
= MPAGE_DA_EXTENT_TAIL
;
2449 pagevec_release(&pvec
);
2455 static int ext4_da_writepages(struct address_space
*mapping
,
2456 struct writeback_control
*wbc
)
2459 int range_whole
= 0;
2460 handle_t
*handle
= NULL
;
2461 struct mpage_da_data mpd
;
2462 struct inode
*inode
= mapping
->host
;
2463 int pages_written
= 0;
2464 unsigned int max_pages
;
2465 int range_cyclic
, cycled
= 1, io_done
= 0;
2466 int needed_blocks
, ret
= 0;
2467 long desired_nr_to_write
, nr_to_writebump
= 0;
2468 loff_t range_start
= wbc
->range_start
;
2469 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2470 pgoff_t done_index
= 0;
2472 struct blk_plug plug
;
2474 trace_ext4_da_writepages(inode
, wbc
);
2477 * No pages to write? This is mainly a kludge to avoid starting
2478 * a transaction for special inodes like journal inode on last iput()
2479 * because that could violate lock ordering on umount
2481 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2485 * If the filesystem has aborted, it is read-only, so return
2486 * right away instead of dumping stack traces later on that
2487 * will obscure the real source of the problem. We test
2488 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2489 * the latter could be true if the filesystem is mounted
2490 * read-only, and in that case, ext4_da_writepages should
2491 * *never* be called, so if that ever happens, we would want
2494 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2497 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2500 range_cyclic
= wbc
->range_cyclic
;
2501 if (wbc
->range_cyclic
) {
2502 index
= mapping
->writeback_index
;
2505 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2506 wbc
->range_end
= LLONG_MAX
;
2507 wbc
->range_cyclic
= 0;
2510 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2511 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2515 * This works around two forms of stupidity. The first is in
2516 * the writeback code, which caps the maximum number of pages
2517 * written to be 1024 pages. This is wrong on multiple
2518 * levels; different architectues have a different page size,
2519 * which changes the maximum amount of data which gets
2520 * written. Secondly, 4 megabytes is way too small. XFS
2521 * forces this value to be 16 megabytes by multiplying
2522 * nr_to_write parameter by four, and then relies on its
2523 * allocator to allocate larger extents to make them
2524 * contiguous. Unfortunately this brings us to the second
2525 * stupidity, which is that ext4's mballoc code only allocates
2526 * at most 2048 blocks. So we force contiguous writes up to
2527 * the number of dirty blocks in the inode, or
2528 * sbi->max_writeback_mb_bump whichever is smaller.
2530 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2531 if (!range_cyclic
&& range_whole
) {
2532 if (wbc
->nr_to_write
== LONG_MAX
)
2533 desired_nr_to_write
= wbc
->nr_to_write
;
2535 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2537 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2539 if (desired_nr_to_write
> max_pages
)
2540 desired_nr_to_write
= max_pages
;
2542 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2543 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2544 wbc
->nr_to_write
= desired_nr_to_write
;
2548 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2549 tag_pages_for_writeback(mapping
, index
, end
);
2551 blk_start_plug(&plug
);
2552 while (!ret
&& wbc
->nr_to_write
> 0) {
2555 * we insert one extent at a time. So we need
2556 * credit needed for single extent allocation.
2557 * journalled mode is currently not supported
2560 BUG_ON(ext4_should_journal_data(inode
));
2561 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2563 /* start a new transaction*/
2564 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2566 if (IS_ERR(handle
)) {
2567 ret
= PTR_ERR(handle
);
2568 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2569 "%ld pages, ino %lu; err %d", __func__
,
2570 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2571 blk_finish_plug(&plug
);
2572 goto out_writepages
;
2576 * Now call write_cache_pages_da() to find the next
2577 * contiguous region of logical blocks that need
2578 * blocks to be allocated by ext4 and submit them.
2580 ret
= write_cache_pages_da(handle
, mapping
,
2581 wbc
, &mpd
, &done_index
);
2583 * If we have a contiguous extent of pages and we
2584 * haven't done the I/O yet, map the blocks and submit
2587 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2588 mpage_da_map_and_submit(&mpd
);
2589 ret
= MPAGE_DA_EXTENT_TAIL
;
2591 trace_ext4_da_write_pages(inode
, &mpd
);
2592 wbc
->nr_to_write
-= mpd
.pages_written
;
2594 ext4_journal_stop(handle
);
2596 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2597 /* commit the transaction which would
2598 * free blocks released in the transaction
2601 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2603 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2605 * Got one extent now try with rest of the pages.
2606 * If mpd.retval is set -EIO, journal is aborted.
2607 * So we don't need to write any more.
2609 pages_written
+= mpd
.pages_written
;
2612 } else if (wbc
->nr_to_write
)
2614 * There is no more writeout needed
2615 * or we requested for a noblocking writeout
2616 * and we found the device congested
2620 blk_finish_plug(&plug
);
2621 if (!io_done
&& !cycled
) {
2624 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2625 wbc
->range_end
= mapping
->writeback_index
- 1;
2630 wbc
->range_cyclic
= range_cyclic
;
2631 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2633 * set the writeback_index so that range_cyclic
2634 * mode will write it back later
2636 mapping
->writeback_index
= done_index
;
2639 wbc
->nr_to_write
-= nr_to_writebump
;
2640 wbc
->range_start
= range_start
;
2641 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2645 static int ext4_nonda_switch(struct super_block
*sb
)
2647 s64 free_clusters
, dirty_clusters
;
2648 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2651 * switch to non delalloc mode if we are running low
2652 * on free block. The free block accounting via percpu
2653 * counters can get slightly wrong with percpu_counter_batch getting
2654 * accumulated on each CPU without updating global counters
2655 * Delalloc need an accurate free block accounting. So switch
2656 * to non delalloc when we are near to error range.
2659 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
);
2661 percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2663 * Start pushing delalloc when 1/2 of free blocks are dirty.
2665 if (dirty_clusters
&& (free_clusters
< 2 * dirty_clusters
))
2666 try_to_writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2668 if (2 * free_clusters
< 3 * dirty_clusters
||
2669 free_clusters
< (dirty_clusters
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2671 * free block count is less than 150% of dirty blocks
2672 * or free blocks is less than watermark
2679 /* We always reserve for an inode update; the superblock could be there too */
2680 static int ext4_da_write_credits(struct inode
*inode
, loff_t pos
, unsigned len
)
2682 if (likely(EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
2683 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
)))
2686 if (pos
+ len
<= 0x7fffffffULL
)
2689 /* We might need to update the superblock to set LARGE_FILE */
2693 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2694 loff_t pos
, unsigned len
, unsigned flags
,
2695 struct page
**pagep
, void **fsdata
)
2697 int ret
, retries
= 0;
2700 struct inode
*inode
= mapping
->host
;
2703 index
= pos
>> PAGE_CACHE_SHIFT
;
2705 if (ext4_nonda_switch(inode
->i_sb
)) {
2706 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2707 return ext4_write_begin(file
, mapping
, pos
,
2708 len
, flags
, pagep
, fsdata
);
2710 *fsdata
= (void *)0;
2711 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2713 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2714 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2724 * grab_cache_page_write_begin() can take a long time if the
2725 * system is thrashing due to memory pressure, or if the page
2726 * is being written back. So grab it first before we start
2727 * the transaction handle. This also allows us to allocate
2728 * the page (if needed) without using GFP_NOFS.
2731 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2737 * With delayed allocation, we don't log the i_disksize update
2738 * if there is delayed block allocation. But we still need
2739 * to journalling the i_disksize update if writes to the end
2740 * of file which has an already mapped buffer.
2743 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
2744 ext4_da_write_credits(inode
, pos
, len
));
2745 if (IS_ERR(handle
)) {
2746 page_cache_release(page
);
2747 return PTR_ERR(handle
);
2751 if (page
->mapping
!= mapping
) {
2752 /* The page got truncated from under us */
2754 page_cache_release(page
);
2755 ext4_journal_stop(handle
);
2758 /* In case writeback began while the page was unlocked */
2759 wait_for_stable_page(page
);
2761 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2764 ext4_journal_stop(handle
);
2766 * block_write_begin may have instantiated a few blocks
2767 * outside i_size. Trim these off again. Don't need
2768 * i_size_read because we hold i_mutex.
2770 if (pos
+ len
> inode
->i_size
)
2771 ext4_truncate_failed_write(inode
);
2773 if (ret
== -ENOSPC
&&
2774 ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2777 page_cache_release(page
);
2786 * Check if we should update i_disksize
2787 * when write to the end of file but not require block allocation
2789 static int ext4_da_should_update_i_disksize(struct page
*page
,
2790 unsigned long offset
)
2792 struct buffer_head
*bh
;
2793 struct inode
*inode
= page
->mapping
->host
;
2797 bh
= page_buffers(page
);
2798 idx
= offset
>> inode
->i_blkbits
;
2800 for (i
= 0; i
< idx
; i
++)
2801 bh
= bh
->b_this_page
;
2803 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2808 static int ext4_da_write_end(struct file
*file
,
2809 struct address_space
*mapping
,
2810 loff_t pos
, unsigned len
, unsigned copied
,
2811 struct page
*page
, void *fsdata
)
2813 struct inode
*inode
= mapping
->host
;
2815 handle_t
*handle
= ext4_journal_current_handle();
2817 unsigned long start
, end
;
2818 int write_mode
= (int)(unsigned long)fsdata
;
2820 if (write_mode
== FALL_BACK_TO_NONDELALLOC
)
2821 return ext4_write_end(file
, mapping
, pos
,
2822 len
, copied
, page
, fsdata
);
2824 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2825 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2826 end
= start
+ copied
- 1;
2829 * generic_write_end() will run mark_inode_dirty() if i_size
2830 * changes. So let's piggyback the i_disksize mark_inode_dirty
2833 new_i_size
= pos
+ copied
;
2834 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2835 if (ext4_has_inline_data(inode
) ||
2836 ext4_da_should_update_i_disksize(page
, end
)) {
2837 down_write(&EXT4_I(inode
)->i_data_sem
);
2838 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
2839 EXT4_I(inode
)->i_disksize
= new_i_size
;
2840 up_write(&EXT4_I(inode
)->i_data_sem
);
2841 /* We need to mark inode dirty even if
2842 * new_i_size is less that inode->i_size
2843 * bu greater than i_disksize.(hint delalloc)
2845 ext4_mark_inode_dirty(handle
, inode
);
2849 if (write_mode
!= CONVERT_INLINE_DATA
&&
2850 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
2851 ext4_has_inline_data(inode
))
2852 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
2855 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2861 ret2
= ext4_journal_stop(handle
);
2865 return ret
? ret
: copied
;
2868 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2871 * Drop reserved blocks
2873 BUG_ON(!PageLocked(page
));
2874 if (!page_has_buffers(page
))
2877 ext4_da_page_release_reservation(page
, offset
);
2880 ext4_invalidatepage(page
, offset
);
2886 * Force all delayed allocation blocks to be allocated for a given inode.
2888 int ext4_alloc_da_blocks(struct inode
*inode
)
2890 trace_ext4_alloc_da_blocks(inode
);
2892 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2893 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2897 * We do something simple for now. The filemap_flush() will
2898 * also start triggering a write of the data blocks, which is
2899 * not strictly speaking necessary (and for users of
2900 * laptop_mode, not even desirable). However, to do otherwise
2901 * would require replicating code paths in:
2903 * ext4_da_writepages() ->
2904 * write_cache_pages() ---> (via passed in callback function)
2905 * __mpage_da_writepage() -->
2906 * mpage_add_bh_to_extent()
2907 * mpage_da_map_blocks()
2909 * The problem is that write_cache_pages(), located in
2910 * mm/page-writeback.c, marks pages clean in preparation for
2911 * doing I/O, which is not desirable if we're not planning on
2914 * We could call write_cache_pages(), and then redirty all of
2915 * the pages by calling redirty_page_for_writepage() but that
2916 * would be ugly in the extreme. So instead we would need to
2917 * replicate parts of the code in the above functions,
2918 * simplifying them because we wouldn't actually intend to
2919 * write out the pages, but rather only collect contiguous
2920 * logical block extents, call the multi-block allocator, and
2921 * then update the buffer heads with the block allocations.
2923 * For now, though, we'll cheat by calling filemap_flush(),
2924 * which will map the blocks, and start the I/O, but not
2925 * actually wait for the I/O to complete.
2927 return filemap_flush(inode
->i_mapping
);
2931 * bmap() is special. It gets used by applications such as lilo and by
2932 * the swapper to find the on-disk block of a specific piece of data.
2934 * Naturally, this is dangerous if the block concerned is still in the
2935 * journal. If somebody makes a swapfile on an ext4 data-journaling
2936 * filesystem and enables swap, then they may get a nasty shock when the
2937 * data getting swapped to that swapfile suddenly gets overwritten by
2938 * the original zero's written out previously to the journal and
2939 * awaiting writeback in the kernel's buffer cache.
2941 * So, if we see any bmap calls here on a modified, data-journaled file,
2942 * take extra steps to flush any blocks which might be in the cache.
2944 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2946 struct inode
*inode
= mapping
->host
;
2951 * We can get here for an inline file via the FIBMAP ioctl
2953 if (ext4_has_inline_data(inode
))
2956 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2957 test_opt(inode
->i_sb
, DELALLOC
)) {
2959 * With delalloc we want to sync the file
2960 * so that we can make sure we allocate
2963 filemap_write_and_wait(mapping
);
2966 if (EXT4_JOURNAL(inode
) &&
2967 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2969 * This is a REALLY heavyweight approach, but the use of
2970 * bmap on dirty files is expected to be extremely rare:
2971 * only if we run lilo or swapon on a freshly made file
2972 * do we expect this to happen.
2974 * (bmap requires CAP_SYS_RAWIO so this does not
2975 * represent an unprivileged user DOS attack --- we'd be
2976 * in trouble if mortal users could trigger this path at
2979 * NB. EXT4_STATE_JDATA is not set on files other than
2980 * regular files. If somebody wants to bmap a directory
2981 * or symlink and gets confused because the buffer
2982 * hasn't yet been flushed to disk, they deserve
2983 * everything they get.
2986 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2987 journal
= EXT4_JOURNAL(inode
);
2988 jbd2_journal_lock_updates(journal
);
2989 err
= jbd2_journal_flush(journal
);
2990 jbd2_journal_unlock_updates(journal
);
2996 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2999 static int ext4_readpage(struct file
*file
, struct page
*page
)
3002 struct inode
*inode
= page
->mapping
->host
;
3004 trace_ext4_readpage(page
);
3006 if (ext4_has_inline_data(inode
))
3007 ret
= ext4_readpage_inline(inode
, page
);
3010 return mpage_readpage(page
, ext4_get_block
);
3016 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
3017 struct list_head
*pages
, unsigned nr_pages
)
3019 struct inode
*inode
= mapping
->host
;
3021 /* If the file has inline data, no need to do readpages. */
3022 if (ext4_has_inline_data(inode
))
3025 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
3028 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
3030 trace_ext4_invalidatepage(page
, offset
);
3032 /* No journalling happens on data buffers when this function is used */
3033 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
3035 block_invalidatepage(page
, offset
);
3038 static int __ext4_journalled_invalidatepage(struct page
*page
,
3039 unsigned long offset
)
3041 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3043 trace_ext4_journalled_invalidatepage(page
, offset
);
3046 * If it's a full truncate we just forget about the pending dirtying
3049 ClearPageChecked(page
);
3051 return jbd2_journal_invalidatepage(journal
, page
, offset
);
3054 /* Wrapper for aops... */
3055 static void ext4_journalled_invalidatepage(struct page
*page
,
3056 unsigned long offset
)
3058 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
) < 0);
3061 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
3063 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
3065 trace_ext4_releasepage(page
);
3067 /* Page has dirty journalled data -> cannot release */
3068 if (PageChecked(page
))
3071 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
3073 return try_to_free_buffers(page
);
3077 * ext4_get_block used when preparing for a DIO write or buffer write.
3078 * We allocate an uinitialized extent if blocks haven't been allocated.
3079 * The extent will be converted to initialized after the IO is complete.
3081 int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
3082 struct buffer_head
*bh_result
, int create
)
3084 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
3085 inode
->i_ino
, create
);
3086 return _ext4_get_block(inode
, iblock
, bh_result
,
3087 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
3090 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
3091 struct buffer_head
*bh_result
, int create
)
3093 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
3094 inode
->i_ino
, create
);
3095 return _ext4_get_block(inode
, iblock
, bh_result
,
3096 EXT4_GET_BLOCKS_NO_LOCK
);
3099 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
3100 ssize_t size
, void *private, int ret
,
3103 struct inode
*inode
= file_inode(iocb
->ki_filp
);
3104 ext4_io_end_t
*io_end
= iocb
->private;
3106 /* if not async direct IO or dio with 0 bytes write, just return */
3107 if (!io_end
|| !size
)
3110 ext_debug("ext4_end_io_dio(): io_end 0x%p "
3111 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
3112 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
3115 iocb
->private = NULL
;
3117 /* if not aio dio with unwritten extents, just free io and return */
3118 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
3119 ext4_free_io_end(io_end
);
3121 inode_dio_done(inode
);
3123 aio_complete(iocb
, ret
, 0);
3127 io_end
->offset
= offset
;
3128 io_end
->size
= size
;
3130 io_end
->iocb
= iocb
;
3131 io_end
->result
= ret
;
3134 ext4_add_complete_io(io_end
);
3138 * For ext4 extent files, ext4 will do direct-io write to holes,
3139 * preallocated extents, and those write extend the file, no need to
3140 * fall back to buffered IO.
3142 * For holes, we fallocate those blocks, mark them as uninitialized
3143 * If those blocks were preallocated, we mark sure they are split, but
3144 * still keep the range to write as uninitialized.
3146 * The unwritten extents will be converted to written when DIO is completed.
3147 * For async direct IO, since the IO may still pending when return, we
3148 * set up an end_io call back function, which will do the conversion
3149 * when async direct IO completed.
3151 * If the O_DIRECT write will extend the file then add this inode to the
3152 * orphan list. So recovery will truncate it back to the original size
3153 * if the machine crashes during the write.
3156 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3157 const struct iovec
*iov
, loff_t offset
,
3158 unsigned long nr_segs
)
3160 struct file
*file
= iocb
->ki_filp
;
3161 struct inode
*inode
= file
->f_mapping
->host
;
3163 size_t count
= iov_length(iov
, nr_segs
);
3165 get_block_t
*get_block_func
= NULL
;
3167 loff_t final_size
= offset
+ count
;
3169 /* Use the old path for reads and writes beyond i_size. */
3170 if (rw
!= WRITE
|| final_size
> inode
->i_size
)
3171 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3173 BUG_ON(iocb
->private == NULL
);
3175 /* If we do a overwrite dio, i_mutex locking can be released */
3176 overwrite
= *((int *)iocb
->private);
3179 atomic_inc(&inode
->i_dio_count
);
3180 down_read(&EXT4_I(inode
)->i_data_sem
);
3181 mutex_unlock(&inode
->i_mutex
);
3185 * We could direct write to holes and fallocate.
3187 * Allocated blocks to fill the hole are marked as
3188 * uninitialized to prevent parallel buffered read to expose
3189 * the stale data before DIO complete the data IO.
3191 * As to previously fallocated extents, ext4 get_block will
3192 * just simply mark the buffer mapped but still keep the
3193 * extents uninitialized.
3195 * For non AIO case, we will convert those unwritten extents
3196 * to written after return back from blockdev_direct_IO.
3198 * For async DIO, the conversion needs to be deferred when the
3199 * IO is completed. The ext4 end_io callback function will be
3200 * called to take care of the conversion work. Here for async
3201 * case, we allocate an io_end structure to hook to the iocb.
3203 iocb
->private = NULL
;
3204 ext4_inode_aio_set(inode
, NULL
);
3205 if (!is_sync_kiocb(iocb
)) {
3206 ext4_io_end_t
*io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
3211 io_end
->flag
|= EXT4_IO_END_DIRECT
;
3212 iocb
->private = io_end
;
3214 * we save the io structure for current async direct
3215 * IO, so that later ext4_map_blocks() could flag the
3216 * io structure whether there is a unwritten extents
3217 * needs to be converted when IO is completed.
3219 ext4_inode_aio_set(inode
, io_end
);
3223 get_block_func
= ext4_get_block_write_nolock
;
3225 get_block_func
= ext4_get_block_write
;
3226 dio_flags
= DIO_LOCKING
;
3228 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3229 inode
->i_sb
->s_bdev
, iov
,
3237 ext4_inode_aio_set(inode
, NULL
);
3239 * The io_end structure takes a reference to the inode, that
3240 * structure needs to be destroyed and the reference to the
3241 * inode need to be dropped, when IO is complete, even with 0
3242 * byte write, or failed.
3244 * In the successful AIO DIO case, the io_end structure will
3245 * be destroyed and the reference to the inode will be dropped
3246 * after the end_io call back function is called.
3248 * In the case there is 0 byte write, or error case, since VFS
3249 * direct IO won't invoke the end_io call back function, we
3250 * need to free the end_io structure here.
3252 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3253 ext4_free_io_end(iocb
->private);
3254 iocb
->private = NULL
;
3255 } else if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3256 EXT4_STATE_DIO_UNWRITTEN
)) {
3259 * for non AIO case, since the IO is already
3260 * completed, we could do the conversion right here
3262 err
= ext4_convert_unwritten_extents(inode
,
3266 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3270 /* take i_mutex locking again if we do a ovewrite dio */
3272 inode_dio_done(inode
);
3273 up_read(&EXT4_I(inode
)->i_data_sem
);
3274 mutex_lock(&inode
->i_mutex
);
3280 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3281 const struct iovec
*iov
, loff_t offset
,
3282 unsigned long nr_segs
)
3284 struct file
*file
= iocb
->ki_filp
;
3285 struct inode
*inode
= file
->f_mapping
->host
;
3289 * If we are doing data journalling we don't support O_DIRECT
3291 if (ext4_should_journal_data(inode
))
3294 /* Let buffer I/O handle the inline data case. */
3295 if (ext4_has_inline_data(inode
))
3298 trace_ext4_direct_IO_enter(inode
, offset
, iov_length(iov
, nr_segs
), rw
);
3299 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3300 ret
= ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3302 ret
= ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3303 trace_ext4_direct_IO_exit(inode
, offset
,
3304 iov_length(iov
, nr_segs
), rw
, ret
);
3309 * Pages can be marked dirty completely asynchronously from ext4's journalling
3310 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3311 * much here because ->set_page_dirty is called under VFS locks. The page is
3312 * not necessarily locked.
3314 * We cannot just dirty the page and leave attached buffers clean, because the
3315 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3316 * or jbddirty because all the journalling code will explode.
3318 * So what we do is to mark the page "pending dirty" and next time writepage
3319 * is called, propagate that into the buffers appropriately.
3321 static int ext4_journalled_set_page_dirty(struct page
*page
)
3323 SetPageChecked(page
);
3324 return __set_page_dirty_nobuffers(page
);
3327 static const struct address_space_operations ext4_aops
= {
3328 .readpage
= ext4_readpage
,
3329 .readpages
= ext4_readpages
,
3330 .writepage
= ext4_writepage
,
3331 .write_begin
= ext4_write_begin
,
3332 .write_end
= ext4_write_end
,
3334 .invalidatepage
= ext4_invalidatepage
,
3335 .releasepage
= ext4_releasepage
,
3336 .direct_IO
= ext4_direct_IO
,
3337 .migratepage
= buffer_migrate_page
,
3338 .is_partially_uptodate
= block_is_partially_uptodate
,
3339 .error_remove_page
= generic_error_remove_page
,
3342 static const struct address_space_operations ext4_journalled_aops
= {
3343 .readpage
= ext4_readpage
,
3344 .readpages
= ext4_readpages
,
3345 .writepage
= ext4_writepage
,
3346 .write_begin
= ext4_write_begin
,
3347 .write_end
= ext4_journalled_write_end
,
3348 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3350 .invalidatepage
= ext4_journalled_invalidatepage
,
3351 .releasepage
= ext4_releasepage
,
3352 .direct_IO
= ext4_direct_IO
,
3353 .is_partially_uptodate
= block_is_partially_uptodate
,
3354 .error_remove_page
= generic_error_remove_page
,
3357 static const struct address_space_operations ext4_da_aops
= {
3358 .readpage
= ext4_readpage
,
3359 .readpages
= ext4_readpages
,
3360 .writepage
= ext4_writepage
,
3361 .writepages
= ext4_da_writepages
,
3362 .write_begin
= ext4_da_write_begin
,
3363 .write_end
= ext4_da_write_end
,
3365 .invalidatepage
= ext4_da_invalidatepage
,
3366 .releasepage
= ext4_releasepage
,
3367 .direct_IO
= ext4_direct_IO
,
3368 .migratepage
= buffer_migrate_page
,
3369 .is_partially_uptodate
= block_is_partially_uptodate
,
3370 .error_remove_page
= generic_error_remove_page
,
3373 void ext4_set_aops(struct inode
*inode
)
3375 switch (ext4_inode_journal_mode(inode
)) {
3376 case EXT4_INODE_ORDERED_DATA_MODE
:
3377 ext4_set_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3379 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3380 ext4_clear_inode_state(inode
, EXT4_STATE_ORDERED_MODE
);
3382 case EXT4_INODE_JOURNAL_DATA_MODE
:
3383 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3388 if (test_opt(inode
->i_sb
, DELALLOC
))
3389 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3391 inode
->i_mapping
->a_ops
= &ext4_aops
;
3396 * ext4_discard_partial_page_buffers()
3397 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3398 * This function finds and locks the page containing the offset
3399 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3400 * Calling functions that already have the page locked should call
3401 * ext4_discard_partial_page_buffers_no_lock directly.
3403 int ext4_discard_partial_page_buffers(handle_t
*handle
,
3404 struct address_space
*mapping
, loff_t from
,
3405 loff_t length
, int flags
)
3407 struct inode
*inode
= mapping
->host
;
3411 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3412 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3416 err
= ext4_discard_partial_page_buffers_no_lock(handle
, inode
, page
,
3417 from
, length
, flags
);
3420 page_cache_release(page
);
3425 * ext4_discard_partial_page_buffers_no_lock()
3426 * Zeros a page range of length 'length' starting from offset 'from'.
3427 * Buffer heads that correspond to the block aligned regions of the
3428 * zeroed range will be unmapped. Unblock aligned regions
3429 * will have the corresponding buffer head mapped if needed so that
3430 * that region of the page can be updated with the partial zero out.
3432 * This function assumes that the page has already been locked. The
3433 * The range to be discarded must be contained with in the given page.
3434 * If the specified range exceeds the end of the page it will be shortened
3435 * to the end of the page that corresponds to 'from'. This function is
3436 * appropriate for updating a page and it buffer heads to be unmapped and
3437 * zeroed for blocks that have been either released, or are going to be
3440 * handle: The journal handle
3441 * inode: The files inode
3442 * page: A locked page that contains the offset "from"
3443 * from: The starting byte offset (from the beginning of the file)
3444 * to begin discarding
3445 * len: The length of bytes to discard
3446 * flags: Optional flags that may be used:
3448 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3449 * Only zero the regions of the page whose buffer heads
3450 * have already been unmapped. This flag is appropriate
3451 * for updating the contents of a page whose blocks may
3452 * have already been released, and we only want to zero
3453 * out the regions that correspond to those released blocks.
3455 * Returns zero on success or negative on failure.
3457 static int ext4_discard_partial_page_buffers_no_lock(handle_t
*handle
,
3458 struct inode
*inode
, struct page
*page
, loff_t from
,
3459 loff_t length
, int flags
)
3461 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3462 unsigned int offset
= from
& (PAGE_CACHE_SIZE
-1);
3463 unsigned int blocksize
, max
, pos
;
3465 struct buffer_head
*bh
;
3468 blocksize
= inode
->i_sb
->s_blocksize
;
3469 max
= PAGE_CACHE_SIZE
- offset
;
3471 if (index
!= page
->index
)
3475 * correct length if it does not fall between
3476 * 'from' and the end of the page
3478 if (length
> max
|| length
< 0)
3481 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3483 if (!page_has_buffers(page
))
3484 create_empty_buffers(page
, blocksize
, 0);
3486 /* Find the buffer that contains "offset" */
3487 bh
= page_buffers(page
);
3489 while (offset
>= pos
) {
3490 bh
= bh
->b_this_page
;
3496 while (pos
< offset
+ length
) {
3497 unsigned int end_of_block
, range_to_discard
;
3501 /* The length of space left to zero and unmap */
3502 range_to_discard
= offset
+ length
- pos
;
3504 /* The length of space until the end of the block */
3505 end_of_block
= blocksize
- (pos
& (blocksize
-1));
3508 * Do not unmap or zero past end of block
3509 * for this buffer head
3511 if (range_to_discard
> end_of_block
)
3512 range_to_discard
= end_of_block
;
3516 * Skip this buffer head if we are only zeroing unampped
3517 * regions of the page
3519 if (flags
& EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
&&
3523 /* If the range is block aligned, unmap */
3524 if (range_to_discard
== blocksize
) {
3525 clear_buffer_dirty(bh
);
3527 clear_buffer_mapped(bh
);
3528 clear_buffer_req(bh
);
3529 clear_buffer_new(bh
);
3530 clear_buffer_delay(bh
);
3531 clear_buffer_unwritten(bh
);
3532 clear_buffer_uptodate(bh
);
3533 zero_user(page
, pos
, range_to_discard
);
3534 BUFFER_TRACE(bh
, "Buffer discarded");
3539 * If this block is not completely contained in the range
3540 * to be discarded, then it is not going to be released. Because
3541 * we need to keep this block, we need to make sure this part
3542 * of the page is uptodate before we modify it by writeing
3543 * partial zeros on it.
3545 if (!buffer_mapped(bh
)) {
3547 * Buffer head must be mapped before we can read
3550 BUFFER_TRACE(bh
, "unmapped");
3551 ext4_get_block(inode
, iblock
, bh
, 0);
3552 /* unmapped? It's a hole - nothing to do */
3553 if (!buffer_mapped(bh
)) {
3554 BUFFER_TRACE(bh
, "still unmapped");
3559 /* Ok, it's mapped. Make sure it's up-to-date */
3560 if (PageUptodate(page
))
3561 set_buffer_uptodate(bh
);
3563 if (!buffer_uptodate(bh
)) {
3565 ll_rw_block(READ
, 1, &bh
);
3567 /* Uhhuh. Read error. Complain and punt.*/
3568 if (!buffer_uptodate(bh
))
3572 if (ext4_should_journal_data(inode
)) {
3573 BUFFER_TRACE(bh
, "get write access");
3574 err
= ext4_journal_get_write_access(handle
, bh
);
3579 zero_user(page
, pos
, range_to_discard
);
3582 if (ext4_should_journal_data(inode
)) {
3583 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3585 mark_buffer_dirty(bh
);
3587 BUFFER_TRACE(bh
, "Partial buffer zeroed");
3589 bh
= bh
->b_this_page
;
3591 pos
+= range_to_discard
;
3597 int ext4_can_truncate(struct inode
*inode
)
3599 if (S_ISREG(inode
->i_mode
))
3601 if (S_ISDIR(inode
->i_mode
))
3603 if (S_ISLNK(inode
->i_mode
))
3604 return !ext4_inode_is_fast_symlink(inode
);
3609 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3610 * associated with the given offset and length
3612 * @inode: File inode
3613 * @offset: The offset where the hole will begin
3614 * @len: The length of the hole
3616 * Returns: 0 on success or negative on failure
3619 int ext4_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
3622 struct inode
*inode
= file_inode(file
);
3623 struct super_block
*sb
= inode
->i_sb
;
3624 ext4_lblk_t first_block
, stop_block
;
3625 struct address_space
*mapping
= inode
->i_mapping
;
3626 loff_t first_page
, last_page
, page_len
;
3627 loff_t first_page_offset
, last_page_offset
;
3629 unsigned int credits
;
3632 if (!S_ISREG(inode
->i_mode
))
3635 if (EXT4_SB(sb
)->s_cluster_ratio
> 1) {
3636 /* TODO: Add support for bigalloc file systems */
3640 trace_ext4_punch_hole(inode
, offset
, length
);
3643 * Write out all dirty pages to avoid race conditions
3644 * Then release them.
3646 if (mapping
->nrpages
&& mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
3647 ret
= filemap_write_and_wait_range(mapping
, offset
,
3648 offset
+ length
- 1);
3653 mutex_lock(&inode
->i_mutex
);
3654 /* It's not possible punch hole on append only file */
3655 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
)) {
3659 if (IS_SWAPFILE(inode
)) {
3664 /* No need to punch hole beyond i_size */
3665 if (offset
>= inode
->i_size
)
3669 * If the hole extends beyond i_size, set the hole
3670 * to end after the page that contains i_size
3672 if (offset
+ length
> inode
->i_size
) {
3673 length
= inode
->i_size
+
3674 PAGE_CACHE_SIZE
- (inode
->i_size
& (PAGE_CACHE_SIZE
- 1)) -
3678 first_page
= (offset
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
3679 last_page
= (offset
+ length
) >> PAGE_CACHE_SHIFT
;
3681 first_page_offset
= first_page
<< PAGE_CACHE_SHIFT
;
3682 last_page_offset
= last_page
<< PAGE_CACHE_SHIFT
;
3684 /* Now release the pages */
3685 if (last_page_offset
> first_page_offset
) {
3686 truncate_pagecache_range(inode
, first_page_offset
,
3687 last_page_offset
- 1);
3690 /* Wait all existing dio workers, newcomers will block on i_mutex */
3691 ext4_inode_block_unlocked_dio(inode
);
3692 ret
= ext4_flush_unwritten_io(inode
);
3695 inode_dio_wait(inode
);
3697 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3698 credits
= ext4_writepage_trans_blocks(inode
);
3700 credits
= ext4_blocks_for_truncate(inode
);
3701 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3702 if (IS_ERR(handle
)) {
3703 ret
= PTR_ERR(handle
);
3704 ext4_std_error(sb
, ret
);
3709 * Now we need to zero out the non-page-aligned data in the
3710 * pages at the start and tail of the hole, and unmap the
3711 * buffer heads for the block aligned regions of the page that
3712 * were completely zeroed.
3714 if (first_page
> last_page
) {
3716 * If the file space being truncated is contained
3717 * within a page just zero out and unmap the middle of
3720 ret
= ext4_discard_partial_page_buffers(handle
,
3721 mapping
, offset
, length
, 0);
3727 * zero out and unmap the partial page that contains
3728 * the start of the hole
3730 page_len
= first_page_offset
- offset
;
3732 ret
= ext4_discard_partial_page_buffers(handle
, mapping
,
3733 offset
, page_len
, 0);
3739 * zero out and unmap the partial page that contains
3740 * the end of the hole
3742 page_len
= offset
+ length
- last_page_offset
;
3744 ret
= ext4_discard_partial_page_buffers(handle
, mapping
,
3745 last_page_offset
, page_len
, 0);
3752 * If i_size is contained in the last page, we need to
3753 * unmap and zero the partial page after i_size
3755 if (inode
->i_size
>> PAGE_CACHE_SHIFT
== last_page
&&
3756 inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
3757 page_len
= PAGE_CACHE_SIZE
-
3758 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
3761 ret
= ext4_discard_partial_page_buffers(handle
,
3762 mapping
, inode
->i_size
, page_len
, 0);
3769 first_block
= (offset
+ sb
->s_blocksize
- 1) >>
3770 EXT4_BLOCK_SIZE_BITS(sb
);
3771 stop_block
= (offset
+ length
) >> EXT4_BLOCK_SIZE_BITS(sb
);
3773 /* If there are no blocks to remove, return now */
3774 if (first_block
>= stop_block
)
3777 down_write(&EXT4_I(inode
)->i_data_sem
);
3778 ext4_discard_preallocations(inode
);
3780 ret
= ext4_es_remove_extent(inode
, first_block
,
3781 stop_block
- first_block
);
3783 up_write(&EXT4_I(inode
)->i_data_sem
);
3787 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3788 ret
= ext4_ext_remove_space(inode
, first_block
,
3791 ret
= ext4_free_hole_blocks(handle
, inode
, first_block
,
3794 ext4_discard_preallocations(inode
);
3795 up_write(&EXT4_I(inode
)->i_data_sem
);
3797 ext4_handle_sync(handle
);
3798 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3799 ext4_mark_inode_dirty(handle
, inode
);
3801 ext4_journal_stop(handle
);
3803 ext4_inode_resume_unlocked_dio(inode
);
3805 mutex_unlock(&inode
->i_mutex
);
3809 * Disabled as per b/28760453
3818 * We block out ext4_get_block() block instantiations across the entire
3819 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3820 * simultaneously on behalf of the same inode.
3822 * As we work through the truncate and commit bits of it to the journal there
3823 * is one core, guiding principle: the file's tree must always be consistent on
3824 * disk. We must be able to restart the truncate after a crash.
3826 * The file's tree may be transiently inconsistent in memory (although it
3827 * probably isn't), but whenever we close off and commit a journal transaction,
3828 * the contents of (the filesystem + the journal) must be consistent and
3829 * restartable. It's pretty simple, really: bottom up, right to left (although
3830 * left-to-right works OK too).
3832 * Note that at recovery time, journal replay occurs *before* the restart of
3833 * truncate against the orphan inode list.
3835 * The committed inode has the new, desired i_size (which is the same as
3836 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3837 * that this inode's truncate did not complete and it will again call
3838 * ext4_truncate() to have another go. So there will be instantiated blocks
3839 * to the right of the truncation point in a crashed ext4 filesystem. But
3840 * that's fine - as long as they are linked from the inode, the post-crash
3841 * ext4_truncate() run will find them and release them.
3843 void ext4_truncate(struct inode
*inode
)
3845 struct ext4_inode_info
*ei
= EXT4_I(inode
);
3846 unsigned int credits
;
3848 struct address_space
*mapping
= inode
->i_mapping
;
3852 * There is a possibility that we're either freeing the inode
3853 * or it completely new indode. In those cases we might not
3854 * have i_mutex locked because it's not necessary.
3856 if (!(inode
->i_state
& (I_NEW
|I_FREEING
)))
3857 WARN_ON(!mutex_is_locked(&inode
->i_mutex
));
3858 trace_ext4_truncate_enter(inode
);
3860 if (!ext4_can_truncate(inode
))
3863 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3865 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3866 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3868 if (ext4_has_inline_data(inode
)) {
3871 ext4_inline_data_truncate(inode
, &has_inline
);
3877 * finish any pending end_io work so we won't run the risk of
3878 * converting any truncated blocks to initialized later
3880 ext4_flush_unwritten_io(inode
);
3882 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3883 credits
= ext4_writepage_trans_blocks(inode
);
3885 credits
= ext4_blocks_for_truncate(inode
);
3887 handle
= ext4_journal_start(inode
, EXT4_HT_TRUNCATE
, credits
);
3888 if (IS_ERR(handle
)) {
3889 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
3893 if (inode
->i_size
% PAGE_CACHE_SIZE
!= 0) {
3894 page_len
= PAGE_CACHE_SIZE
-
3895 (inode
->i_size
& (PAGE_CACHE_SIZE
- 1));
3897 if (ext4_discard_partial_page_buffers(handle
,
3898 mapping
, inode
->i_size
, page_len
, 0))
3903 * We add the inode to the orphan list, so that if this
3904 * truncate spans multiple transactions, and we crash, we will
3905 * resume the truncate when the filesystem recovers. It also
3906 * marks the inode dirty, to catch the new size.
3908 * Implication: the file must always be in a sane, consistent
3909 * truncatable state while each transaction commits.
3911 if (ext4_orphan_add(handle
, inode
))
3914 down_write(&EXT4_I(inode
)->i_data_sem
);
3916 ext4_discard_preallocations(inode
);
3918 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3919 ext4_ext_truncate(handle
, inode
);
3921 ext4_ind_truncate(handle
, inode
);
3923 up_write(&ei
->i_data_sem
);
3926 ext4_handle_sync(handle
);
3930 * If this was a simple ftruncate() and the file will remain alive,
3931 * then we need to clear up the orphan record which we created above.
3932 * However, if this was a real unlink then we were called by
3933 * ext4_delete_inode(), and we allow that function to clean up the
3934 * orphan info for us.
3937 ext4_orphan_del(handle
, inode
);
3939 inode
->i_mtime
= inode
->i_ctime
= ext4_current_time(inode
);
3940 ext4_mark_inode_dirty(handle
, inode
);
3941 ext4_journal_stop(handle
);
3943 trace_ext4_truncate_exit(inode
);
3947 * ext4_get_inode_loc returns with an extra refcount against the inode's
3948 * underlying buffer_head on success. If 'in_mem' is true, we have all
3949 * data in memory that is needed to recreate the on-disk version of this
3952 static int __ext4_get_inode_loc(struct inode
*inode
,
3953 struct ext4_iloc
*iloc
, int in_mem
)
3955 struct ext4_group_desc
*gdp
;
3956 struct buffer_head
*bh
;
3957 struct super_block
*sb
= inode
->i_sb
;
3959 int inodes_per_block
, inode_offset
;
3962 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3965 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3966 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3971 * Figure out the offset within the block group inode table
3973 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3974 inode_offset
= ((inode
->i_ino
- 1) %
3975 EXT4_INODES_PER_GROUP(sb
));
3976 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3977 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3979 bh
= sb_getblk(sb
, block
);
3982 if (!buffer_uptodate(bh
)) {
3986 * If the buffer has the write error flag, we have failed
3987 * to write out another inode in the same block. In this
3988 * case, we don't have to read the block because we may
3989 * read the old inode data successfully.
3991 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3992 set_buffer_uptodate(bh
);
3994 if (buffer_uptodate(bh
)) {
3995 /* someone brought it uptodate while we waited */
4001 * If we have all information of the inode in memory and this
4002 * is the only valid inode in the block, we need not read the
4006 struct buffer_head
*bitmap_bh
;
4009 start
= inode_offset
& ~(inodes_per_block
- 1);
4011 /* Is the inode bitmap in cache? */
4012 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
4013 if (unlikely(!bitmap_bh
))
4017 * If the inode bitmap isn't in cache then the
4018 * optimisation may end up performing two reads instead
4019 * of one, so skip it.
4021 if (!buffer_uptodate(bitmap_bh
)) {
4025 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
4026 if (i
== inode_offset
)
4028 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
4032 if (i
== start
+ inodes_per_block
) {
4033 /* all other inodes are free, so skip I/O */
4034 memset(bh
->b_data
, 0, bh
->b_size
);
4035 set_buffer_uptodate(bh
);
4043 * If we need to do any I/O, try to pre-readahead extra
4044 * blocks from the inode table.
4046 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
4047 ext4_fsblk_t b
, end
, table
;
4049 __u32 ra_blks
= EXT4_SB(sb
)->s_inode_readahead_blks
;
4051 table
= ext4_inode_table(sb
, gdp
);
4052 /* s_inode_readahead_blks is always a power of 2 */
4053 b
= block
& ~((ext4_fsblk_t
) ra_blks
- 1);
4057 num
= EXT4_INODES_PER_GROUP(sb
);
4058 if (ext4_has_group_desc_csum(sb
))
4059 num
-= ext4_itable_unused_count(sb
, gdp
);
4060 table
+= num
/ inodes_per_block
;
4064 sb_breadahead(sb
, b
++);
4068 * There are other valid inodes in the buffer, this inode
4069 * has in-inode xattrs, or we don't have this inode in memory.
4070 * Read the block from disk.
4072 trace_ext4_load_inode(inode
);
4074 bh
->b_end_io
= end_buffer_read_sync
;
4075 #ifdef FEATURE_STORAGE_META_LOG
4076 if( bh
&& bh
->b_bdev
&& bh
->b_bdev
->bd_disk
)
4077 set_metadata_rw_status(bh
->b_bdev
->bd_disk
->first_minor
, WAIT_READ_CNT
);
4079 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
4081 if (!buffer_uptodate(bh
)) {
4082 EXT4_ERROR_INODE_BLOCK(inode
, block
,
4083 "unable to read itable block");
4093 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
4095 /* We have all inode data except xattrs in memory here. */
4096 return __ext4_get_inode_loc(inode
, iloc
,
4097 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
4100 void ext4_set_inode_flags(struct inode
*inode
)
4102 unsigned int flags
= EXT4_I(inode
)->i_flags
;
4103 unsigned int new_fl
= 0;
4105 if (flags
& EXT4_SYNC_FL
)
4107 if (flags
& EXT4_APPEND_FL
)
4109 if (flags
& EXT4_IMMUTABLE_FL
)
4110 new_fl
|= S_IMMUTABLE
;
4111 if (flags
& EXT4_NOATIME_FL
)
4112 new_fl
|= S_NOATIME
;
4113 if (flags
& EXT4_DIRSYNC_FL
)
4114 new_fl
|= S_DIRSYNC
;
4115 set_mask_bits(&inode
->i_flags
,
4116 S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
, new_fl
);
4119 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
4120 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
4122 unsigned int vfs_fl
;
4123 unsigned long old_fl
, new_fl
;
4126 vfs_fl
= ei
->vfs_inode
.i_flags
;
4127 old_fl
= ei
->i_flags
;
4128 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
4129 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
4131 if (vfs_fl
& S_SYNC
)
4132 new_fl
|= EXT4_SYNC_FL
;
4133 if (vfs_fl
& S_APPEND
)
4134 new_fl
|= EXT4_APPEND_FL
;
4135 if (vfs_fl
& S_IMMUTABLE
)
4136 new_fl
|= EXT4_IMMUTABLE_FL
;
4137 if (vfs_fl
& S_NOATIME
)
4138 new_fl
|= EXT4_NOATIME_FL
;
4139 if (vfs_fl
& S_DIRSYNC
)
4140 new_fl
|= EXT4_DIRSYNC_FL
;
4141 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
4144 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
4145 struct ext4_inode_info
*ei
)
4148 struct inode
*inode
= &(ei
->vfs_inode
);
4149 struct super_block
*sb
= inode
->i_sb
;
4151 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4152 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
4153 /* we are using combined 48 bit field */
4154 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
4155 le32_to_cpu(raw_inode
->i_blocks_lo
);
4156 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
4157 /* i_blocks represent file system block size */
4158 return i_blocks
<< (inode
->i_blkbits
- 9);
4163 return le32_to_cpu(raw_inode
->i_blocks_lo
);
4167 static inline void ext4_iget_extra_inode(struct inode
*inode
,
4168 struct ext4_inode
*raw_inode
,
4169 struct ext4_inode_info
*ei
)
4171 __le32
*magic
= (void *)raw_inode
+
4172 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
4173 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4174 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
4175 ext4_find_inline_data_nolock(inode
);
4177 EXT4_I(inode
)->i_inline_off
= 0;
4180 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
4182 struct ext4_iloc iloc
;
4183 struct ext4_inode
*raw_inode
;
4184 struct ext4_inode_info
*ei
;
4185 struct inode
*inode
;
4186 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
4192 inode
= iget_locked(sb
, ino
);
4194 return ERR_PTR(-ENOMEM
);
4195 if (!(inode
->i_state
& I_NEW
))
4201 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4204 raw_inode
= ext4_raw_inode(&iloc
);
4206 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4207 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
4208 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
4209 EXT4_INODE_SIZE(inode
->i_sb
)) {
4210 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
4211 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
4212 EXT4_INODE_SIZE(inode
->i_sb
));
4217 ei
->i_extra_isize
= 0;
4219 /* Precompute checksum seed for inode metadata */
4220 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4221 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
)) {
4222 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4224 __le32 inum
= cpu_to_le32(inode
->i_ino
);
4225 __le32 gen
= raw_inode
->i_generation
;
4226 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
4228 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
4232 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
4233 EXT4_ERROR_INODE(inode
, "checksum invalid");
4238 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
4239 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
4240 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
4241 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4242 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
4243 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
4245 i_uid_write(inode
, i_uid
);
4246 i_gid_write(inode
, i_gid
);
4247 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
4249 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
4250 ei
->i_inline_off
= 0;
4251 ei
->i_dir_start_lookup
= 0;
4252 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
4253 /* We now have enough fields to check if the inode was active or not.
4254 * This is needed because nfsd might try to access dead inodes
4255 * the test is that same one that e2fsck uses
4256 * NeilBrown 1999oct15
4258 if (inode
->i_nlink
== 0) {
4259 if ((inode
->i_mode
== 0 ||
4260 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) &&
4261 ino
!= EXT4_BOOT_LOADER_INO
) {
4262 /* this inode is deleted */
4266 /* The only unlinked inodes we let through here have
4267 * valid i_mode and are being read by the orphan
4268 * recovery code: that's fine, we're about to complete
4269 * the process of deleting those.
4270 * OR it is the EXT4_BOOT_LOADER_INO which is
4271 * not initialized on a new filesystem. */
4273 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
4274 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
4275 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
4276 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
4278 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
4279 inode
->i_size
= ext4_isize(raw_inode
);
4280 ei
->i_disksize
= inode
->i_size
;
4282 ei
->i_reserved_quota
= 0;
4284 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
4285 ei
->i_block_group
= iloc
.block_group
;
4286 ei
->i_last_alloc_group
= ~0;
4288 * NOTE! The in-memory inode i_data array is in little-endian order
4289 * even on big-endian machines: we do NOT byteswap the block numbers!
4291 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4292 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
4293 INIT_LIST_HEAD(&ei
->i_orphan
);
4296 * Set transaction id's of transactions that have to be committed
4297 * to finish f[data]sync. We set them to currently running transaction
4298 * as we cannot be sure that the inode or some of its metadata isn't
4299 * part of the transaction - the inode could have been reclaimed and
4300 * now it is reread from disk.
4303 transaction_t
*transaction
;
4306 read_lock(&journal
->j_state_lock
);
4307 if (journal
->j_running_transaction
)
4308 transaction
= journal
->j_running_transaction
;
4310 transaction
= journal
->j_committing_transaction
;
4312 tid
= transaction
->t_tid
;
4314 tid
= journal
->j_commit_sequence
;
4315 read_unlock(&journal
->j_state_lock
);
4316 ei
->i_sync_tid
= tid
;
4317 ei
->i_datasync_tid
= tid
;
4320 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4321 if (ei
->i_extra_isize
== 0) {
4322 /* The extra space is currently unused. Use it. */
4323 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
4324 EXT4_GOOD_OLD_INODE_SIZE
;
4326 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
4330 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
4331 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
4332 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
4333 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
4335 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
4336 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
4337 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4339 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
4343 if (ei
->i_file_acl
&&
4344 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
4345 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
4349 } else if (!ext4_has_inline_data(inode
)) {
4350 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
4351 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4352 (S_ISLNK(inode
->i_mode
) &&
4353 !ext4_inode_is_fast_symlink(inode
))))
4354 /* Validate extent which is part of inode */
4355 ret
= ext4_ext_check_inode(inode
);
4356 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
4357 (S_ISLNK(inode
->i_mode
) &&
4358 !ext4_inode_is_fast_symlink(inode
))) {
4359 /* Validate block references which are part of inode */
4360 ret
= ext4_ind_check_inode(inode
);
4366 if (S_ISREG(inode
->i_mode
)) {
4367 inode
->i_op
= &ext4_file_inode_operations
;
4368 inode
->i_fop
= &ext4_file_operations
;
4369 ext4_set_aops(inode
);
4370 } else if (S_ISDIR(inode
->i_mode
)) {
4371 inode
->i_op
= &ext4_dir_inode_operations
;
4372 inode
->i_fop
= &ext4_dir_operations
;
4373 } else if (S_ISLNK(inode
->i_mode
)) {
4374 if (ext4_inode_is_fast_symlink(inode
)) {
4375 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
4376 nd_terminate_link(ei
->i_data
, inode
->i_size
,
4377 sizeof(ei
->i_data
) - 1);
4379 inode
->i_op
= &ext4_symlink_inode_operations
;
4380 ext4_set_aops(inode
);
4382 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
4383 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
4384 inode
->i_op
= &ext4_special_inode_operations
;
4385 if (raw_inode
->i_block
[0])
4386 init_special_inode(inode
, inode
->i_mode
,
4387 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
4389 init_special_inode(inode
, inode
->i_mode
,
4390 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
4391 } else if (ino
== EXT4_BOOT_LOADER_INO
) {
4392 make_bad_inode(inode
);
4395 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
4399 ext4_set_inode_flags(inode
);
4400 unlock_new_inode(inode
);
4406 return ERR_PTR(ret
);
4409 struct inode
*ext4_iget_normal(struct super_block
*sb
, unsigned long ino
)
4411 if (ino
< EXT4_FIRST_INO(sb
) && ino
!= EXT4_ROOT_INO
)
4412 return ERR_PTR(-EIO
);
4413 return ext4_iget(sb
, ino
);
4416 static int ext4_inode_blocks_set(handle_t
*handle
,
4417 struct ext4_inode
*raw_inode
,
4418 struct ext4_inode_info
*ei
)
4420 struct inode
*inode
= &(ei
->vfs_inode
);
4421 u64 i_blocks
= inode
->i_blocks
;
4422 struct super_block
*sb
= inode
->i_sb
;
4424 if (i_blocks
<= ~0U) {
4426 * i_blocks can be represented in a 32 bit variable
4427 * as multiple of 512 bytes
4429 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4430 raw_inode
->i_blocks_high
= 0;
4431 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4434 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4437 if (i_blocks
<= 0xffffffffffffULL
) {
4439 * i_blocks can be represented in a 48 bit variable
4440 * as multiple of 512 bytes
4442 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4443 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4444 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4446 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4447 /* i_block is stored in file system block size */
4448 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4449 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4450 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4456 * Post the struct inode info into an on-disk inode location in the
4457 * buffer-cache. This gobbles the caller's reference to the
4458 * buffer_head in the inode location struct.
4460 * The caller must have write access to iloc->bh.
4462 static int ext4_do_update_inode(handle_t
*handle
,
4463 struct inode
*inode
,
4464 struct ext4_iloc
*iloc
)
4466 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4467 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4468 struct buffer_head
*bh
= iloc
->bh
;
4469 int err
= 0, rc
, block
;
4470 int need_datasync
= 0;
4474 /* For fields not not tracking in the in-memory inode,
4475 * initialise them to zero for new inodes. */
4476 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4477 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4479 ext4_get_inode_flags(ei
);
4480 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4481 i_uid
= i_uid_read(inode
);
4482 i_gid
= i_gid_read(inode
);
4483 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4484 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4485 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4487 * Fix up interoperability with old kernels. Otherwise, old inodes get
4488 * re-used with the upper 16 bits of the uid/gid intact
4491 raw_inode
->i_uid_high
=
4492 cpu_to_le16(high_16_bits(i_uid
));
4493 raw_inode
->i_gid_high
=
4494 cpu_to_le16(high_16_bits(i_gid
));
4496 raw_inode
->i_uid_high
= 0;
4497 raw_inode
->i_gid_high
= 0;
4500 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4501 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4502 raw_inode
->i_uid_high
= 0;
4503 raw_inode
->i_gid_high
= 0;
4505 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4507 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4508 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4509 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4510 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4512 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4514 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4515 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4516 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4517 cpu_to_le32(EXT4_OS_HURD
))
4518 raw_inode
->i_file_acl_high
=
4519 cpu_to_le16(ei
->i_file_acl
>> 32);
4520 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4521 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4522 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4525 if (ei
->i_disksize
> 0x7fffffffULL
) {
4526 struct super_block
*sb
= inode
->i_sb
;
4527 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4528 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4529 EXT4_SB(sb
)->s_es
->s_rev_level
==
4530 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4531 /* If this is the first large file
4532 * created, add a flag to the superblock.
4534 err
= ext4_journal_get_write_access(handle
,
4535 EXT4_SB(sb
)->s_sbh
);
4538 ext4_update_dynamic_rev(sb
);
4539 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4540 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4541 ext4_handle_sync(handle
);
4542 err
= ext4_handle_dirty_super(handle
, sb
);
4545 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4546 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4547 if (old_valid_dev(inode
->i_rdev
)) {
4548 raw_inode
->i_block
[0] =
4549 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4550 raw_inode
->i_block
[1] = 0;
4552 raw_inode
->i_block
[0] = 0;
4553 raw_inode
->i_block
[1] =
4554 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4555 raw_inode
->i_block
[2] = 0;
4557 } else if (!ext4_has_inline_data(inode
)) {
4558 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4559 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4562 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4563 if (ei
->i_extra_isize
) {
4564 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4565 raw_inode
->i_version_hi
=
4566 cpu_to_le32(inode
->i_version
>> 32);
4567 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4570 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4572 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4573 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4576 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4578 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4581 ext4_std_error(inode
->i_sb
, err
);
4586 * ext4_write_inode()
4588 * We are called from a few places:
4590 * - Within generic_file_write() for O_SYNC files.
4591 * Here, there will be no transaction running. We wait for any running
4592 * transaction to commit.
4594 * - Within sys_sync(), kupdate and such.
4595 * We wait on commit, if tol to.
4597 * - Within prune_icache() (PF_MEMALLOC == true)
4598 * Here we simply return. We can't afford to block kswapd on the
4601 * In all cases it is actually safe for us to return without doing anything,
4602 * because the inode has been copied into a raw inode buffer in
4603 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4606 * Note that we are absolutely dependent upon all inode dirtiers doing the
4607 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4608 * which we are interested.
4610 * It would be a bug for them to not do this. The code:
4612 * mark_inode_dirty(inode)
4614 * inode->i_size = expr;
4616 * is in error because a kswapd-driven write_inode() could occur while
4617 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4618 * will no longer be on the superblock's dirty inode list.
4620 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4624 if (current
->flags
& PF_MEMALLOC
)
4627 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4628 if (ext4_journal_current_handle()) {
4629 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4634 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
4637 err
= ext4_force_commit(inode
->i_sb
);
4639 struct ext4_iloc iloc
;
4641 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4644 if (wbc
->sync_mode
== WB_SYNC_ALL
)
4645 sync_dirty_buffer(iloc
.bh
);
4646 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4647 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4648 "IO error syncing inode");
4657 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4658 * buffers that are attached to a page stradding i_size and are undergoing
4659 * commit. In that case we have to wait for commit to finish and try again.
4661 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4665 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4666 tid_t commit_tid
= 0;
4669 offset
= inode
->i_size
& (PAGE_CACHE_SIZE
- 1);
4671 * All buffers in the last page remain valid? Then there's nothing to
4672 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4675 if (offset
> PAGE_CACHE_SIZE
- (1 << inode
->i_blkbits
))
4678 page
= find_lock_page(inode
->i_mapping
,
4679 inode
->i_size
>> PAGE_CACHE_SHIFT
);
4682 ret
= __ext4_journalled_invalidatepage(page
, offset
);
4684 page_cache_release(page
);
4688 read_lock(&journal
->j_state_lock
);
4689 if (journal
->j_committing_transaction
)
4690 commit_tid
= journal
->j_committing_transaction
->t_tid
;
4691 read_unlock(&journal
->j_state_lock
);
4693 jbd2_log_wait_commit(journal
, commit_tid
);
4700 * Called from notify_change.
4702 * We want to trap VFS attempts to truncate the file as soon as
4703 * possible. In particular, we want to make sure that when the VFS
4704 * shrinks i_size, we put the inode on the orphan list and modify
4705 * i_disksize immediately, so that during the subsequent flushing of
4706 * dirty pages and freeing of disk blocks, we can guarantee that any
4707 * commit will leave the blocks being flushed in an unused state on
4708 * disk. (On recovery, the inode will get truncated and the blocks will
4709 * be freed, so we have a strong guarantee that no future commit will
4710 * leave these blocks visible to the user.)
4712 * Another thing we have to assure is that if we are in ordered mode
4713 * and inode is still attached to the committing transaction, we must
4714 * we start writeout of all the dirty pages which are being truncated.
4715 * This way we are sure that all the data written in the previous
4716 * transaction are already on disk (truncate waits for pages under
4719 * Called with inode->i_mutex down.
4721 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4723 struct inode
*inode
= dentry
->d_inode
;
4726 const unsigned int ia_valid
= attr
->ia_valid
;
4728 error
= inode_change_ok(inode
, attr
);
4732 if (is_quota_modification(inode
, attr
))
4733 dquot_initialize(inode
);
4734 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4735 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4738 /* (user+group)*(old+new) structure, inode write (sb,
4739 * inode block, ? - but truncate inode update has it) */
4740 handle
= ext4_journal_start(inode
, EXT4_HT_QUOTA
,
4741 (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
) +
4742 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
)) + 3);
4743 if (IS_ERR(handle
)) {
4744 error
= PTR_ERR(handle
);
4747 error
= dquot_transfer(inode
, attr
);
4749 ext4_journal_stop(handle
);
4752 /* Update corresponding info in inode so that everything is in
4753 * one transaction */
4754 if (attr
->ia_valid
& ATTR_UID
)
4755 inode
->i_uid
= attr
->ia_uid
;
4756 if (attr
->ia_valid
& ATTR_GID
)
4757 inode
->i_gid
= attr
->ia_gid
;
4758 error
= ext4_mark_inode_dirty(handle
, inode
);
4759 ext4_journal_stop(handle
);
4762 if (attr
->ia_valid
& ATTR_SIZE
&& attr
->ia_size
!= inode
->i_size
) {
4764 loff_t oldsize
= inode
->i_size
;
4766 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4767 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4769 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4773 if (IS_I_VERSION(inode
) && attr
->ia_size
!= inode
->i_size
)
4774 inode_inc_iversion(inode
);
4776 if (S_ISREG(inode
->i_mode
) &&
4777 (attr
->ia_size
< inode
->i_size
)) {
4778 if (ext4_should_order_data(inode
)) {
4779 error
= ext4_begin_ordered_truncate(inode
,
4784 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 3);
4785 if (IS_ERR(handle
)) {
4786 error
= PTR_ERR(handle
);
4789 if (ext4_handle_valid(handle
)) {
4790 error
= ext4_orphan_add(handle
, inode
);
4793 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4794 rc
= ext4_mark_inode_dirty(handle
, inode
);
4797 ext4_journal_stop(handle
);
4799 ext4_orphan_del(NULL
, inode
);
4804 i_size_write(inode
, attr
->ia_size
);
4806 * Blocks are going to be removed from the inode. Wait
4807 * for dio in flight. Temporarily disable
4808 * dioread_nolock to prevent livelock.
4811 if (!ext4_should_journal_data(inode
)) {
4812 ext4_inode_block_unlocked_dio(inode
);
4813 inode_dio_wait(inode
);
4814 ext4_inode_resume_unlocked_dio(inode
);
4816 ext4_wait_for_tail_page_commit(inode
);
4819 * Truncate pagecache after we've waited for commit
4820 * in data=journal mode to make pages freeable.
4822 truncate_pagecache(inode
, oldsize
, inode
->i_size
);
4825 * We want to call ext4_truncate() even if attr->ia_size ==
4826 * inode->i_size for cases like truncation of fallocated space
4828 if (attr
->ia_valid
& ATTR_SIZE
)
4829 ext4_truncate(inode
);
4832 setattr_copy(inode
, attr
);
4833 mark_inode_dirty(inode
);
4837 * If the call to ext4_truncate failed to get a transaction handle at
4838 * all, we need to clean up the in-core orphan list manually.
4840 if (orphan
&& inode
->i_nlink
)
4841 ext4_orphan_del(NULL
, inode
);
4843 if (!rc
&& (ia_valid
& ATTR_MODE
))
4844 rc
= ext4_acl_chmod(inode
);
4847 ext4_std_error(inode
->i_sb
, error
);
4853 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4856 struct inode
*inode
;
4857 unsigned long long delalloc_blocks
;
4859 inode
= dentry
->d_inode
;
4860 generic_fillattr(inode
, stat
);
4863 * We can't update i_blocks if the block allocation is delayed
4864 * otherwise in the case of system crash before the real block
4865 * allocation is done, we will have i_blocks inconsistent with
4866 * on-disk file blocks.
4867 * We always keep i_blocks updated together with real
4868 * allocation. But to not confuse with user, stat
4869 * will return the blocks that include the delayed allocation
4870 * blocks for this file.
4872 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
4873 EXT4_I(inode
)->i_reserved_data_blocks
);
4875 stat
->blocks
+= delalloc_blocks
<< (inode
->i_sb
->s_blocksize_bits
-9);
4879 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4881 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4882 return ext4_ind_trans_blocks(inode
, nrblocks
, chunk
);
4883 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4887 * Account for index blocks, block groups bitmaps and block group
4888 * descriptor blocks if modify datablocks and index blocks
4889 * worse case, the indexs blocks spread over different block groups
4891 * If datablocks are discontiguous, they are possible to spread over
4892 * different block groups too. If they are contiguous, with flexbg,
4893 * they could still across block group boundary.
4895 * Also account for superblock, inode, quota and xattr blocks
4897 static int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4899 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4905 * How many index blocks need to touch to modify nrblocks?
4906 * The "Chunk" flag indicating whether the nrblocks is
4907 * physically contiguous on disk
4909 * For Direct IO and fallocate, they calls get_block to allocate
4910 * one single extent at a time, so they could set the "Chunk" flag
4912 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4917 * Now let's see how many group bitmaps and group descriptors need
4927 if (groups
> ngroups
)
4929 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4930 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4932 /* bitmaps and block group descriptor blocks */
4933 ret
+= groups
+ gdpblocks
;
4935 /* Blocks for super block, inode, quota and xattr blocks */
4936 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4942 * Calculate the total number of credits to reserve to fit
4943 * the modification of a single pages into a single transaction,
4944 * which may include multiple chunks of block allocations.
4946 * This could be called via ext4_write_begin()
4948 * We need to consider the worse case, when
4949 * one new block per extent.
4951 int ext4_writepage_trans_blocks(struct inode
*inode
)
4953 int bpp
= ext4_journal_blocks_per_page(inode
);
4956 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4958 /* Account for data blocks for journalled mode */
4959 if (ext4_should_journal_data(inode
))
4965 * Calculate the journal credits for a chunk of data modification.
4967 * This is called from DIO, fallocate or whoever calling
4968 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4970 * journal buffers for data blocks are not included here, as DIO
4971 * and fallocate do no need to journal data buffers.
4973 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4975 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4979 * The caller must have previously called ext4_reserve_inode_write().
4980 * Give this, we know that the caller already has write access to iloc->bh.
4982 int ext4_mark_iloc_dirty(handle_t
*handle
,
4983 struct inode
*inode
, struct ext4_iloc
*iloc
)
4987 if (IS_I_VERSION(inode
))
4988 inode_inc_iversion(inode
);
4990 /* the do_update_inode consumes one bh->b_count */
4993 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4994 err
= ext4_do_update_inode(handle
, inode
, iloc
);
5000 * On success, We end up with an outstanding reference count against
5001 * iloc->bh. This _must_ be cleaned up later.
5005 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
5006 struct ext4_iloc
*iloc
)
5010 err
= ext4_get_inode_loc(inode
, iloc
);
5012 BUFFER_TRACE(iloc
->bh
, "get_write_access");
5013 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
5019 ext4_std_error(inode
->i_sb
, err
);
5024 * Expand an inode by new_extra_isize bytes.
5025 * Returns 0 on success or negative error number on failure.
5027 static int ext4_expand_extra_isize(struct inode
*inode
,
5028 unsigned int new_extra_isize
,
5029 struct ext4_iloc iloc
,
5032 struct ext4_inode
*raw_inode
;
5033 struct ext4_xattr_ibody_header
*header
;
5035 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
5038 raw_inode
= ext4_raw_inode(&iloc
);
5040 header
= IHDR(inode
, raw_inode
);
5042 /* No extended attributes present */
5043 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
5044 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
5045 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
5047 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
5051 /* try to expand with EAs present */
5052 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
5057 * What we do here is to mark the in-core inode as clean with respect to inode
5058 * dirtiness (it may still be data-dirty).
5059 * This means that the in-core inode may be reaped by prune_icache
5060 * without having to perform any I/O. This is a very good thing,
5061 * because *any* task may call prune_icache - even ones which
5062 * have a transaction open against a different journal.
5064 * Is this cheating? Not really. Sure, we haven't written the
5065 * inode out, but prune_icache isn't a user-visible syncing function.
5066 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
5067 * we start and wait on commits.
5069 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
5071 struct ext4_iloc iloc
;
5072 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
5073 static unsigned int mnt_count
;
5077 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
5078 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
5079 if (ext4_handle_valid(handle
) &&
5080 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
5081 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
5083 * We need extra buffer credits since we may write into EA block
5084 * with this same handle. If journal_extend fails, then it will
5085 * only result in a minor loss of functionality for that inode.
5086 * If this is felt to be critical, then e2fsck should be run to
5087 * force a large enough s_min_extra_isize.
5089 if ((jbd2_journal_extend(handle
,
5090 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
5091 ret
= ext4_expand_extra_isize(inode
,
5092 sbi
->s_want_extra_isize
,
5095 ext4_set_inode_state(inode
,
5096 EXT4_STATE_NO_EXPAND
);
5098 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
5099 ext4_warning(inode
->i_sb
,
5100 "Unable to expand inode %lu. Delete"
5101 " some EAs or run e2fsck.",
5104 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
5110 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
5115 * ext4_dirty_inode() is called from __mark_inode_dirty()
5117 * We're really interested in the case where a file is being extended.
5118 * i_size has been changed by generic_commit_write() and we thus need
5119 * to include the updated inode in the current transaction.
5121 * Also, dquot_alloc_block() will always dirty the inode when blocks
5122 * are allocated to the file.
5124 * If the inode is marked synchronous, we don't honour that here - doing
5125 * so would cause a commit on atime updates, which we don't bother doing.
5126 * We handle synchronous inodes at the highest possible level.
5128 void ext4_dirty_inode(struct inode
*inode
, int flags
)
5132 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 2);
5136 ext4_mark_inode_dirty(handle
, inode
);
5138 ext4_journal_stop(handle
);
5145 * Bind an inode's backing buffer_head into this transaction, to prevent
5146 * it from being flushed to disk early. Unlike
5147 * ext4_reserve_inode_write, this leaves behind no bh reference and
5148 * returns no iloc structure, so the caller needs to repeat the iloc
5149 * lookup to mark the inode dirty later.
5151 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
5153 struct ext4_iloc iloc
;
5157 err
= ext4_get_inode_loc(inode
, &iloc
);
5159 BUFFER_TRACE(iloc
.bh
, "get_write_access");
5160 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
5162 err
= ext4_handle_dirty_metadata(handle
,
5168 ext4_std_error(inode
->i_sb
, err
);
5173 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
5180 * We have to be very careful here: changing a data block's
5181 * journaling status dynamically is dangerous. If we write a
5182 * data block to the journal, change the status and then delete
5183 * that block, we risk forgetting to revoke the old log record
5184 * from the journal and so a subsequent replay can corrupt data.
5185 * So, first we make sure that the journal is empty and that
5186 * nobody is changing anything.
5189 journal
= EXT4_JOURNAL(inode
);
5192 if (is_journal_aborted(journal
))
5194 /* We have to allocate physical blocks for delalloc blocks
5195 * before flushing journal. otherwise delalloc blocks can not
5196 * be allocated any more. even more truncate on delalloc blocks
5197 * could trigger BUG by flushing delalloc blocks in journal.
5198 * There is no delalloc block in non-journal data mode.
5200 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
5201 err
= ext4_alloc_da_blocks(inode
);
5206 /* Wait for all existing dio workers */
5207 ext4_inode_block_unlocked_dio(inode
);
5208 inode_dio_wait(inode
);
5210 jbd2_journal_lock_updates(journal
);
5213 * OK, there are no updates running now, and all cached data is
5214 * synced to disk. We are now in a completely consistent state
5215 * which doesn't have anything in the journal, and we know that
5216 * no filesystem updates are running, so it is safe to modify
5217 * the inode's in-core data-journaling state flag now.
5221 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5223 jbd2_journal_flush(journal
);
5224 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
5226 ext4_set_aops(inode
);
5228 jbd2_journal_unlock_updates(journal
);
5229 ext4_inode_resume_unlocked_dio(inode
);
5231 /* Finally we can mark the inode as dirty. */
5233 handle
= ext4_journal_start(inode
, EXT4_HT_INODE
, 1);
5235 return PTR_ERR(handle
);
5237 err
= ext4_mark_inode_dirty(handle
, inode
);
5238 ext4_handle_sync(handle
);
5239 ext4_journal_stop(handle
);
5240 ext4_std_error(inode
->i_sb
, err
);
5245 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
5247 return !buffer_mapped(bh
);
5250 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
5252 struct page
*page
= vmf
->page
;
5256 struct file
*file
= vma
->vm_file
;
5257 struct inode
*inode
= file_inode(file
);
5258 struct address_space
*mapping
= inode
->i_mapping
;
5260 get_block_t
*get_block
;
5263 sb_start_pagefault(inode
->i_sb
);
5264 file_update_time(vma
->vm_file
);
5265 /* Delalloc case is easy... */
5266 if (test_opt(inode
->i_sb
, DELALLOC
) &&
5267 !ext4_should_journal_data(inode
) &&
5268 !ext4_nonda_switch(inode
->i_sb
)) {
5270 ret
= __block_page_mkwrite(vma
, vmf
,
5271 ext4_da_get_block_prep
);
5272 } while (ret
== -ENOSPC
&&
5273 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
5278 size
= i_size_read(inode
);
5279 /* Page got truncated from under us? */
5280 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
5282 ret
= VM_FAULT_NOPAGE
;
5286 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
5287 len
= size
& ~PAGE_CACHE_MASK
;
5289 len
= PAGE_CACHE_SIZE
;
5291 * Return if we have all the buffers mapped. This avoids the need to do
5292 * journal_start/journal_stop which can block and take a long time
5294 if (page_has_buffers(page
)) {
5295 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
5297 ext4_bh_unmapped
)) {
5298 /* Wait so that we don't change page under IO */
5299 wait_for_stable_page(page
);
5300 ret
= VM_FAULT_LOCKED
;
5305 /* OK, we need to fill the hole... */
5306 if (ext4_should_dioread_nolock(inode
))
5307 get_block
= ext4_get_block_write
;
5309 get_block
= ext4_get_block
;
5311 handle
= ext4_journal_start(inode
, EXT4_HT_WRITE_PAGE
,
5312 ext4_writepage_trans_blocks(inode
));
5313 if (IS_ERR(handle
)) {
5314 ret
= VM_FAULT_SIGBUS
;
5317 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
5318 if (!ret
&& ext4_should_journal_data(inode
)) {
5319 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
5320 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
5322 ret
= VM_FAULT_SIGBUS
;
5323 ext4_journal_stop(handle
);
5326 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
5328 ext4_journal_stop(handle
);
5329 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
5332 ret
= block_page_mkwrite_return(ret
);
5334 sb_end_pagefault(inode
->i_sb
);