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>
41 #include "ext4_jbd2.h"
46 #include <trace/events/ext4.h>
48 #define MPAGE_DA_EXTENT_TAIL 0x01
50 static __u32
ext4_inode_csum(struct inode
*inode
, struct ext4_inode
*raw
,
51 struct ext4_inode_info
*ei
)
53 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
58 csum_lo
= raw
->i_checksum_lo
;
59 raw
->i_checksum_lo
= 0;
60 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
61 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
)) {
62 csum_hi
= raw
->i_checksum_hi
;
63 raw
->i_checksum_hi
= 0;
66 csum
= ext4_chksum(sbi
, ei
->i_csum_seed
, (__u8
*)raw
,
67 EXT4_INODE_SIZE(inode
->i_sb
));
69 raw
->i_checksum_lo
= csum_lo
;
70 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
71 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
72 raw
->i_checksum_hi
= csum_hi
;
77 static int ext4_inode_csum_verify(struct inode
*inode
, struct ext4_inode
*raw
,
78 struct ext4_inode_info
*ei
)
80 __u32 provided
, calculated
;
82 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
83 cpu_to_le32(EXT4_OS_LINUX
) ||
84 !EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
85 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
88 provided
= le16_to_cpu(raw
->i_checksum_lo
);
89 calculated
= ext4_inode_csum(inode
, raw
, ei
);
90 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
91 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
92 provided
|= ((__u32
)le16_to_cpu(raw
->i_checksum_hi
)) << 16;
96 return provided
== calculated
;
99 static void ext4_inode_csum_set(struct inode
*inode
, struct ext4_inode
*raw
,
100 struct ext4_inode_info
*ei
)
104 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
105 cpu_to_le32(EXT4_OS_LINUX
) ||
106 !EXT4_HAS_RO_COMPAT_FEATURE(inode
->i_sb
,
107 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
))
110 csum
= ext4_inode_csum(inode
, raw
, ei
);
111 raw
->i_checksum_lo
= cpu_to_le16(csum
& 0xFFFF);
112 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
&&
113 EXT4_FITS_IN_INODE(raw
, ei
, i_checksum_hi
))
114 raw
->i_checksum_hi
= cpu_to_le16(csum
>> 16);
117 static inline int ext4_begin_ordered_truncate(struct inode
*inode
,
120 trace_ext4_begin_ordered_truncate(inode
, new_size
);
122 * If jinode is zero, then we never opened the file for
123 * writing, so there's no need to call
124 * jbd2_journal_begin_ordered_truncate() since there's no
125 * outstanding writes we need to flush.
127 if (!EXT4_I(inode
)->jinode
)
129 return jbd2_journal_begin_ordered_truncate(EXT4_JOURNAL(inode
),
130 EXT4_I(inode
)->jinode
,
134 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
);
135 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
136 struct buffer_head
*bh_result
, int create
);
137 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
138 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
139 static int ext4_discard_partial_page_buffers_no_lock(handle_t
*handle
,
140 struct inode
*inode
, struct page
*page
, loff_t from
,
141 loff_t length
, int flags
);
144 * Test whether an inode is a fast symlink.
146 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
148 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
149 (inode
->i_sb
->s_blocksize
>> 9) : 0;
151 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
155 * Restart the transaction associated with *handle. This does a commit,
156 * so before we call here everything must be consistently dirtied against
159 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
165 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
166 * moment, get_block can be called only for blocks inside i_size since
167 * page cache has been already dropped and writes are blocked by
168 * i_mutex. So we can safely drop the i_data_sem here.
170 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
171 jbd_debug(2, "restarting handle %p\n", handle
);
172 up_write(&EXT4_I(inode
)->i_data_sem
);
173 ret
= ext4_journal_restart(handle
, nblocks
);
174 down_write(&EXT4_I(inode
)->i_data_sem
);
175 ext4_discard_preallocations(inode
);
181 * Called at the last iput() if i_nlink is zero.
183 void ext4_evict_inode(struct inode
*inode
)
188 trace_ext4_evict_inode(inode
);
190 ext4_ioend_wait(inode
);
192 if (inode
->i_nlink
) {
194 * When journalling data dirty buffers are tracked only in the
195 * journal. So although mm thinks everything is clean and
196 * ready for reaping the inode might still have some pages to
197 * write in the running transaction or waiting to be
198 * checkpointed. Thus calling jbd2_journal_invalidatepage()
199 * (via truncate_inode_pages()) to discard these buffers can
200 * cause data loss. Also even if we did not discard these
201 * buffers, we would have no way to find them after the inode
202 * is reaped and thus user could see stale data if he tries to
203 * read them before the transaction is checkpointed. So be
204 * careful and force everything to disk here... We use
205 * ei->i_datasync_tid to store the newest transaction
206 * containing inode's data.
208 * Note that directories do not have this problem because they
209 * don't use page cache.
211 if (ext4_should_journal_data(inode
) &&
212 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
))) {
213 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
214 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
216 jbd2_log_start_commit(journal
, commit_tid
);
217 jbd2_log_wait_commit(journal
, commit_tid
);
218 filemap_write_and_wait(&inode
->i_data
);
220 truncate_inode_pages(&inode
->i_data
, 0);
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);
231 if (is_bad_inode(inode
))
235 * Protect us against freezing - iput() caller didn't have to have any
236 * protection against it
238 sb_start_intwrite(inode
->i_sb
);
239 handle
= ext4_journal_start(inode
, ext4_blocks_for_truncate(inode
)+3);
240 if (IS_ERR(handle
)) {
241 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
243 * If we're going to skip the normal cleanup, we still need to
244 * make sure that the in-core orphan linked list is properly
247 ext4_orphan_del(NULL
, inode
);
248 sb_end_intwrite(inode
->i_sb
);
253 ext4_handle_sync(handle
);
255 err
= ext4_mark_inode_dirty(handle
, inode
);
257 ext4_warning(inode
->i_sb
,
258 "couldn't mark inode dirty (err %d)", err
);
262 ext4_truncate(inode
);
265 * ext4_ext_truncate() doesn't reserve any slop when it
266 * restarts journal transactions; therefore there may not be
267 * enough credits left in the handle to remove the inode from
268 * the orphan list and set the dtime field.
270 if (!ext4_handle_has_enough_credits(handle
, 3)) {
271 err
= ext4_journal_extend(handle
, 3);
273 err
= ext4_journal_restart(handle
, 3);
275 ext4_warning(inode
->i_sb
,
276 "couldn't extend journal (err %d)", err
);
278 ext4_journal_stop(handle
);
279 ext4_orphan_del(NULL
, inode
);
280 sb_end_intwrite(inode
->i_sb
);
286 * Kill off the orphan record which ext4_truncate created.
287 * AKPM: I think this can be inside the above `if'.
288 * Note that ext4_orphan_del() has to be able to cope with the
289 * deletion of a non-existent orphan - this is because we don't
290 * know if ext4_truncate() actually created an orphan record.
291 * (Well, we could do this if we need to, but heck - it works)
293 ext4_orphan_del(handle
, inode
);
294 EXT4_I(inode
)->i_dtime
= get_seconds();
297 * One subtle ordering requirement: if anything has gone wrong
298 * (transaction abort, IO errors, whatever), then we can still
299 * do these next steps (the fs will already have been marked as
300 * having errors), but we can't free the inode if the mark_dirty
303 if (ext4_mark_inode_dirty(handle
, inode
))
304 /* If that failed, just do the required in-core inode clear. */
305 ext4_clear_inode(inode
);
307 ext4_free_inode(handle
, inode
);
308 ext4_journal_stop(handle
);
309 sb_end_intwrite(inode
->i_sb
);
312 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
316 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
318 return &EXT4_I(inode
)->i_reserved_quota
;
323 * Calculate the number of metadata blocks need to reserve
324 * to allocate a block located at @lblock
326 static int ext4_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
328 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
329 return ext4_ext_calc_metadata_amount(inode
, lblock
);
331 return ext4_ind_calc_metadata_amount(inode
, lblock
);
335 * Called with i_data_sem down, which is important since we can call
336 * ext4_discard_preallocations() from here.
338 void ext4_da_update_reserve_space(struct inode
*inode
,
339 int used
, int quota_claim
)
341 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
342 struct ext4_inode_info
*ei
= EXT4_I(inode
);
344 spin_lock(&ei
->i_block_reservation_lock
);
345 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
346 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
347 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "%s: ino %lu, used %d "
348 "with only %d reserved data blocks",
349 __func__
, inode
->i_ino
, used
,
350 ei
->i_reserved_data_blocks
);
352 used
= ei
->i_reserved_data_blocks
;
355 if (unlikely(ei
->i_allocated_meta_blocks
> ei
->i_reserved_meta_blocks
)) {
356 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "%s: ino %lu, allocated %d "
357 "with only %d reserved metadata blocks\n", __func__
,
358 inode
->i_ino
, ei
->i_allocated_meta_blocks
,
359 ei
->i_reserved_meta_blocks
);
361 ei
->i_allocated_meta_blocks
= ei
->i_reserved_meta_blocks
;
364 /* Update per-inode reservations */
365 ei
->i_reserved_data_blocks
-= used
;
366 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
367 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
368 used
+ ei
->i_allocated_meta_blocks
);
369 ei
->i_allocated_meta_blocks
= 0;
371 if (ei
->i_reserved_data_blocks
== 0) {
373 * We can release all of the reserved metadata blocks
374 * only when we have written all of the delayed
377 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
378 ei
->i_reserved_meta_blocks
);
379 ei
->i_reserved_meta_blocks
= 0;
380 ei
->i_da_metadata_calc_len
= 0;
382 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
384 /* Update quota subsystem for data blocks */
386 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
389 * We did fallocate with an offset that is already delayed
390 * allocated. So on delayed allocated writeback we should
391 * not re-claim the quota for fallocated blocks.
393 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
397 * If we have done all the pending block allocations and if
398 * there aren't any writers on the inode, we can discard the
399 * inode's preallocations.
401 if ((ei
->i_reserved_data_blocks
== 0) &&
402 (atomic_read(&inode
->i_writecount
) == 0))
403 ext4_discard_preallocations(inode
);
406 static int __check_block_validity(struct inode
*inode
, const char *func
,
408 struct ext4_map_blocks
*map
)
410 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
412 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
413 "lblock %lu mapped to illegal pblock "
414 "(length %d)", (unsigned long) map
->m_lblk
,
421 #define check_block_validity(inode, map) \
422 __check_block_validity((inode), __func__, __LINE__, (map))
425 * Return the number of contiguous dirty pages in a given inode
426 * starting at page frame idx.
428 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
429 unsigned int max_pages
)
431 struct address_space
*mapping
= inode
->i_mapping
;
435 int i
, nr_pages
, done
= 0;
439 pagevec_init(&pvec
, 0);
442 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
444 (pgoff_t
)PAGEVEC_SIZE
);
447 for (i
= 0; i
< nr_pages
; i
++) {
448 struct page
*page
= pvec
.pages
[i
];
449 struct buffer_head
*bh
, *head
;
452 if (unlikely(page
->mapping
!= mapping
) ||
454 PageWriteback(page
) ||
455 page
->index
!= idx
) {
460 if (page_has_buffers(page
)) {
461 bh
= head
= page_buffers(page
);
463 if (!buffer_delay(bh
) &&
464 !buffer_unwritten(bh
))
466 bh
= bh
->b_this_page
;
467 } while (!done
&& (bh
!= head
));
474 if (num
>= max_pages
) {
479 pagevec_release(&pvec
);
485 * The ext4_map_blocks() function tries to look up the requested blocks,
486 * and returns if the blocks are already mapped.
488 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
489 * and store the allocated blocks in the result buffer head and mark it
492 * If file type is extents based, it will call ext4_ext_map_blocks(),
493 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
496 * On success, it returns the number of blocks being mapped or allocate.
497 * if create==0 and the blocks are pre-allocated and uninitialized block,
498 * the result buffer head is unmapped. If the create ==1, it will make sure
499 * the buffer head is mapped.
501 * It returns 0 if plain look up failed (blocks have not been allocated), in
502 * that case, buffer head is unmapped
504 * It returns the error in case of allocation failure.
506 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
507 struct ext4_map_blocks
*map
, int flags
)
512 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
513 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
514 (unsigned long) map
->m_lblk
);
516 * Try to see if we can get the block without requesting a new
519 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
520 down_read((&EXT4_I(inode
)->i_data_sem
));
521 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
522 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
523 EXT4_GET_BLOCKS_KEEP_SIZE
);
525 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
526 EXT4_GET_BLOCKS_KEEP_SIZE
);
528 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
529 up_read((&EXT4_I(inode
)->i_data_sem
));
531 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
533 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
534 /* delayed alloc may be allocated by fallocate and
535 * coverted to initialized by directIO.
536 * we need to handle delayed extent here.
538 down_write((&EXT4_I(inode
)->i_data_sem
));
541 ret
= check_block_validity(inode
, map
);
546 /* If it is only a block(s) look up */
547 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
551 * Returns if the blocks have already allocated
553 * Note that if blocks have been preallocated
554 * ext4_ext_get_block() returns the create = 0
555 * with buffer head unmapped.
557 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
561 * When we call get_blocks without the create flag, the
562 * BH_Unwritten flag could have gotten set if the blocks
563 * requested were part of a uninitialized extent. We need to
564 * clear this flag now that we are committed to convert all or
565 * part of the uninitialized extent to be an initialized
566 * extent. This is because we need to avoid the combination
567 * of BH_Unwritten and BH_Mapped flags being simultaneously
568 * set on the buffer_head.
570 map
->m_flags
&= ~EXT4_MAP_UNWRITTEN
;
573 * New blocks allocate and/or writing to uninitialized extent
574 * will possibly result in updating i_data, so we take
575 * the write lock of i_data_sem, and call get_blocks()
576 * with create == 1 flag.
578 down_write((&EXT4_I(inode
)->i_data_sem
));
581 * if the caller is from delayed allocation writeout path
582 * we have already reserved fs blocks for allocation
583 * let the underlying get_block() function know to
584 * avoid double accounting
586 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
587 ext4_set_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
589 * We need to check for EXT4 here because migrate
590 * could have changed the inode type in between
592 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
593 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
595 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
597 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
599 * We allocated new blocks which will result in
600 * i_data's format changing. Force the migrate
601 * to fail by clearing migrate flags
603 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
607 * Update reserved blocks/metadata blocks after successful
608 * block allocation which had been deferred till now. We don't
609 * support fallocate for non extent files. So we can update
610 * reserve space here.
613 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
614 ext4_da_update_reserve_space(inode
, retval
, 1);
616 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
617 ext4_clear_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
619 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
622 /* delayed allocation blocks has been allocated */
623 ret
= ext4_es_remove_extent(inode
, map
->m_lblk
,
630 up_write((&EXT4_I(inode
)->i_data_sem
));
631 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
632 int ret
= check_block_validity(inode
, map
);
639 /* Maximum number of blocks we map for direct IO at once. */
640 #define DIO_MAX_BLOCKS 4096
642 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
643 struct buffer_head
*bh
, int flags
)
645 handle_t
*handle
= ext4_journal_current_handle();
646 struct ext4_map_blocks map
;
647 int ret
= 0, started
= 0;
650 if (ext4_has_inline_data(inode
))
654 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
656 if (flags
&& !(flags
& EXT4_GET_BLOCKS_NO_LOCK
) && !handle
) {
657 /* Direct IO write... */
658 if (map
.m_len
> DIO_MAX_BLOCKS
)
659 map
.m_len
= DIO_MAX_BLOCKS
;
660 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
661 handle
= ext4_journal_start(inode
, dio_credits
);
662 if (IS_ERR(handle
)) {
663 ret
= PTR_ERR(handle
);
669 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
671 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
672 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
673 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
677 ext4_journal_stop(handle
);
681 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
682 struct buffer_head
*bh
, int create
)
684 return _ext4_get_block(inode
, iblock
, bh
,
685 create
? EXT4_GET_BLOCKS_CREATE
: 0);
689 * `handle' can be NULL if create is zero
691 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
692 ext4_lblk_t block
, int create
, int *errp
)
694 struct ext4_map_blocks map
;
695 struct buffer_head
*bh
;
698 J_ASSERT(handle
!= NULL
|| create
== 0);
702 err
= ext4_map_blocks(handle
, inode
, &map
,
703 create
? EXT4_GET_BLOCKS_CREATE
: 0);
705 /* ensure we send some value back into *errp */
713 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
718 if (map
.m_flags
& EXT4_MAP_NEW
) {
719 J_ASSERT(create
!= 0);
720 J_ASSERT(handle
!= NULL
);
723 * Now that we do not always journal data, we should
724 * keep in mind whether this should always journal the
725 * new buffer as metadata. For now, regular file
726 * writes use ext4_get_block instead, so it's not a
730 BUFFER_TRACE(bh
, "call get_create_access");
731 fatal
= ext4_journal_get_create_access(handle
, bh
);
732 if (!fatal
&& !buffer_uptodate(bh
)) {
733 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
734 set_buffer_uptodate(bh
);
737 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
738 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
742 BUFFER_TRACE(bh
, "not a new buffer");
752 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
753 ext4_lblk_t block
, int create
, int *err
)
755 struct buffer_head
*bh
;
757 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
760 if (buffer_uptodate(bh
))
762 ll_rw_block(READ
| REQ_META
| REQ_PRIO
, 1, &bh
);
764 if (buffer_uptodate(bh
))
771 int ext4_walk_page_buffers(handle_t
*handle
,
772 struct buffer_head
*head
,
776 int (*fn
)(handle_t
*handle
,
777 struct buffer_head
*bh
))
779 struct buffer_head
*bh
;
780 unsigned block_start
, block_end
;
781 unsigned blocksize
= head
->b_size
;
783 struct buffer_head
*next
;
785 for (bh
= head
, block_start
= 0;
786 ret
== 0 && (bh
!= head
|| !block_start
);
787 block_start
= block_end
, bh
= next
) {
788 next
= bh
->b_this_page
;
789 block_end
= block_start
+ blocksize
;
790 if (block_end
<= from
|| block_start
>= to
) {
791 if (partial
&& !buffer_uptodate(bh
))
795 err
= (*fn
)(handle
, bh
);
803 * To preserve ordering, it is essential that the hole instantiation and
804 * the data write be encapsulated in a single transaction. We cannot
805 * close off a transaction and start a new one between the ext4_get_block()
806 * and the commit_write(). So doing the jbd2_journal_start at the start of
807 * prepare_write() is the right place.
809 * Also, this function can nest inside ext4_writepage(). In that case, we
810 * *know* that ext4_writepage() has generated enough buffer credits to do the
811 * whole page. So we won't block on the journal in that case, which is good,
812 * because the caller may be PF_MEMALLOC.
814 * By accident, ext4 can be reentered when a transaction is open via
815 * quota file writes. If we were to commit the transaction while thus
816 * reentered, there can be a deadlock - we would be holding a quota
817 * lock, and the commit would never complete if another thread had a
818 * transaction open and was blocking on the quota lock - a ranking
821 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
822 * will _not_ run commit under these circumstances because handle->h_ref
823 * is elevated. We'll still have enough credits for the tiny quotafile
826 int do_journal_get_write_access(handle_t
*handle
,
827 struct buffer_head
*bh
)
829 int dirty
= buffer_dirty(bh
);
832 if (!buffer_mapped(bh
) || buffer_freed(bh
))
835 * __block_write_begin() could have dirtied some buffers. Clean
836 * the dirty bit as jbd2_journal_get_write_access() could complain
837 * otherwise about fs integrity issues. Setting of the dirty bit
838 * by __block_write_begin() isn't a real problem here as we clear
839 * the bit before releasing a page lock and thus writeback cannot
840 * ever write the buffer.
843 clear_buffer_dirty(bh
);
844 ret
= ext4_journal_get_write_access(handle
, bh
);
846 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
850 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
851 struct buffer_head
*bh_result
, int create
);
852 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
853 loff_t pos
, unsigned len
, unsigned flags
,
854 struct page
**pagep
, void **fsdata
)
856 struct inode
*inode
= mapping
->host
;
857 int ret
, needed_blocks
;
864 trace_ext4_write_begin(inode
, pos
, len
, flags
);
866 * Reserve one block more for addition to orphan list in case
867 * we allocate blocks but write fails for some reason
869 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
870 index
= pos
>> PAGE_CACHE_SHIFT
;
871 from
= pos
& (PAGE_CACHE_SIZE
- 1);
874 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
875 ret
= ext4_try_to_write_inline_data(mapping
, inode
, pos
, len
,
886 handle
= ext4_journal_start(inode
, needed_blocks
);
887 if (IS_ERR(handle
)) {
888 ret
= PTR_ERR(handle
);
892 /* We cannot recurse into the filesystem as the transaction is already
894 flags
|= AOP_FLAG_NOFS
;
896 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
898 ext4_journal_stop(handle
);
905 if (ext4_should_dioread_nolock(inode
))
906 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
908 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
910 if (!ret
&& ext4_should_journal_data(inode
)) {
911 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
),
913 do_journal_get_write_access
);
918 page_cache_release(page
);
920 * __block_write_begin may have instantiated a few blocks
921 * outside i_size. Trim these off again. Don't need
922 * i_size_read because we hold i_mutex.
924 * Add inode to orphan list in case we crash before
927 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
928 ext4_orphan_add(handle
, inode
);
930 ext4_journal_stop(handle
);
931 if (pos
+ len
> inode
->i_size
) {
932 ext4_truncate_failed_write(inode
);
934 * If truncate failed early the inode might
935 * still be on the orphan list; we need to
936 * make sure the inode is removed from the
937 * orphan list in that case.
940 ext4_orphan_del(NULL
, inode
);
944 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
950 /* For write_end() in data=journal mode */
951 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
953 if (!buffer_mapped(bh
) || buffer_freed(bh
))
955 set_buffer_uptodate(bh
);
956 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
959 static int ext4_generic_write_end(struct file
*file
,
960 struct address_space
*mapping
,
961 loff_t pos
, unsigned len
, unsigned copied
,
962 struct page
*page
, void *fsdata
)
964 int i_size_changed
= 0;
965 struct inode
*inode
= mapping
->host
;
966 handle_t
*handle
= ext4_journal_current_handle();
968 if (ext4_has_inline_data(inode
))
969 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
972 copied
= block_write_end(file
, mapping
, pos
,
973 len
, copied
, page
, fsdata
);
976 * No need to use i_size_read() here, the i_size
977 * cannot change under us because we hold i_mutex.
979 * But it's important to update i_size while still holding page lock:
980 * page writeout could otherwise come in and zero beyond i_size.
982 if (pos
+ copied
> inode
->i_size
) {
983 i_size_write(inode
, pos
+ copied
);
987 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
988 /* We need to mark inode dirty even if
989 * new_i_size is less that inode->i_size
990 * bu greater than i_disksize.(hint delalloc)
992 ext4_update_i_disksize(inode
, (pos
+ copied
));
996 page_cache_release(page
);
999 * Don't mark the inode dirty under page lock. First, it unnecessarily
1000 * makes the holding time of page lock longer. Second, it forces lock
1001 * ordering of page lock and transaction start for journaling
1005 ext4_mark_inode_dirty(handle
, inode
);
1011 * We need to pick up the new inode size which generic_commit_write gave us
1012 * `file' can be NULL - eg, when called from page_symlink().
1014 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1015 * buffers are managed internally.
1017 static int ext4_ordered_write_end(struct file
*file
,
1018 struct address_space
*mapping
,
1019 loff_t pos
, unsigned len
, unsigned copied
,
1020 struct page
*page
, void *fsdata
)
1022 handle_t
*handle
= ext4_journal_current_handle();
1023 struct inode
*inode
= mapping
->host
;
1026 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1027 ret
= ext4_jbd2_file_inode(handle
, inode
);
1030 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1033 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1034 /* if we have allocated more blocks and copied
1035 * less. We will have blocks allocated outside
1036 * inode->i_size. So truncate them
1038 ext4_orphan_add(handle
, inode
);
1043 page_cache_release(page
);
1046 ret2
= ext4_journal_stop(handle
);
1050 if (pos
+ len
> inode
->i_size
) {
1051 ext4_truncate_failed_write(inode
);
1053 * If truncate failed early the inode might still be
1054 * on the orphan list; we need to make sure the inode
1055 * is removed from the orphan list in that case.
1058 ext4_orphan_del(NULL
, inode
);
1062 return ret
? ret
: copied
;
1065 static int ext4_writeback_write_end(struct file
*file
,
1066 struct address_space
*mapping
,
1067 loff_t pos
, unsigned len
, unsigned copied
,
1068 struct page
*page
, void *fsdata
)
1070 handle_t
*handle
= ext4_journal_current_handle();
1071 struct inode
*inode
= mapping
->host
;
1074 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1075 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1078 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1079 /* if we have allocated more blocks and copied
1080 * less. We will have blocks allocated outside
1081 * inode->i_size. So truncate them
1083 ext4_orphan_add(handle
, inode
);
1088 ret2
= ext4_journal_stop(handle
);
1092 if (pos
+ len
> inode
->i_size
) {
1093 ext4_truncate_failed_write(inode
);
1095 * If truncate failed early the inode might still be
1096 * on the orphan list; we need to make sure the inode
1097 * is removed from the orphan list in that case.
1100 ext4_orphan_del(NULL
, inode
);
1103 return ret
? ret
: copied
;
1106 static int ext4_journalled_write_end(struct file
*file
,
1107 struct address_space
*mapping
,
1108 loff_t pos
, unsigned len
, unsigned copied
,
1109 struct page
*page
, void *fsdata
)
1111 handle_t
*handle
= ext4_journal_current_handle();
1112 struct inode
*inode
= mapping
->host
;
1118 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1119 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1122 BUG_ON(!ext4_handle_valid(handle
));
1124 if (ext4_has_inline_data(inode
))
1125 copied
= ext4_write_inline_data_end(inode
, pos
, len
,
1129 if (!PageUptodate(page
))
1131 page_zero_new_buffers(page
, from
+copied
, to
);
1134 ret
= ext4_walk_page_buffers(handle
, page_buffers(page
), from
,
1135 to
, &partial
, write_end_fn
);
1137 SetPageUptodate(page
);
1139 new_i_size
= pos
+ copied
;
1140 if (new_i_size
> inode
->i_size
)
1141 i_size_write(inode
, pos
+copied
);
1142 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1143 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1144 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1145 ext4_update_i_disksize(inode
, new_i_size
);
1146 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1152 page_cache_release(page
);
1153 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1154 /* if we have allocated more blocks and copied
1155 * less. We will have blocks allocated outside
1156 * inode->i_size. So truncate them
1158 ext4_orphan_add(handle
, inode
);
1160 ret2
= ext4_journal_stop(handle
);
1163 if (pos
+ len
> inode
->i_size
) {
1164 ext4_truncate_failed_write(inode
);
1166 * If truncate failed early the inode might still be
1167 * on the orphan list; we need to make sure the inode
1168 * is removed from the orphan list in that case.
1171 ext4_orphan_del(NULL
, inode
);
1174 return ret
? ret
: copied
;
1178 * Reserve a single cluster located at lblock
1180 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1183 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1184 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1185 unsigned int md_needed
;
1187 ext4_lblk_t save_last_lblock
;
1191 * We will charge metadata quota at writeout time; this saves
1192 * us from metadata over-estimation, though we may go over by
1193 * a small amount in the end. Here we just reserve for data.
1195 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1200 * recalculate the amount of metadata blocks to reserve
1201 * in order to allocate nrblocks
1202 * worse case is one extent per block
1205 spin_lock(&ei
->i_block_reservation_lock
);
1207 * ext4_calc_metadata_amount() has side effects, which we have
1208 * to be prepared undo if we fail to claim space.
1210 save_len
= ei
->i_da_metadata_calc_len
;
1211 save_last_lblock
= ei
->i_da_metadata_calc_last_lblock
;
1212 md_needed
= EXT4_NUM_B2C(sbi
,
1213 ext4_calc_metadata_amount(inode
, lblock
));
1214 trace_ext4_da_reserve_space(inode
, md_needed
);
1217 * We do still charge estimated metadata to the sb though;
1218 * we cannot afford to run out of free blocks.
1220 if (ext4_claim_free_clusters(sbi
, md_needed
+ 1, 0)) {
1221 ei
->i_da_metadata_calc_len
= save_len
;
1222 ei
->i_da_metadata_calc_last_lblock
= save_last_lblock
;
1223 spin_unlock(&ei
->i_block_reservation_lock
);
1224 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1228 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1231 ei
->i_reserved_data_blocks
++;
1232 ei
->i_reserved_meta_blocks
+= md_needed
;
1233 spin_unlock(&ei
->i_block_reservation_lock
);
1235 return 0; /* success */
1238 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1240 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1241 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1244 return; /* Nothing to release, exit */
1246 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1248 trace_ext4_da_release_space(inode
, to_free
);
1249 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1251 * if there aren't enough reserved blocks, then the
1252 * counter is messed up somewhere. Since this
1253 * function is called from invalidate page, it's
1254 * harmless to return without any action.
1256 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "ext4_da_release_space: "
1257 "ino %lu, to_free %d with only %d reserved "
1258 "data blocks", inode
->i_ino
, to_free
,
1259 ei
->i_reserved_data_blocks
);
1261 to_free
= ei
->i_reserved_data_blocks
;
1263 ei
->i_reserved_data_blocks
-= to_free
;
1265 if (ei
->i_reserved_data_blocks
== 0) {
1267 * We can release all of the reserved metadata blocks
1268 * only when we have written all of the delayed
1269 * allocation blocks.
1270 * Note that in case of bigalloc, i_reserved_meta_blocks,
1271 * i_reserved_data_blocks, etc. refer to number of clusters.
1273 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
1274 ei
->i_reserved_meta_blocks
);
1275 ei
->i_reserved_meta_blocks
= 0;
1276 ei
->i_da_metadata_calc_len
= 0;
1279 /* update fs dirty data blocks counter */
1280 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1282 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1284 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1287 static void ext4_da_page_release_reservation(struct page
*page
,
1288 unsigned long offset
)
1291 struct buffer_head
*head
, *bh
;
1292 unsigned int curr_off
= 0;
1293 struct inode
*inode
= page
->mapping
->host
;
1294 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1298 head
= page_buffers(page
);
1301 unsigned int next_off
= curr_off
+ bh
->b_size
;
1303 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1305 clear_buffer_delay(bh
);
1307 curr_off
= next_off
;
1308 } while ((bh
= bh
->b_this_page
) != head
);
1311 lblk
= page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1312 ext4_es_remove_extent(inode
, lblk
, to_release
);
1315 /* If we have released all the blocks belonging to a cluster, then we
1316 * need to release the reserved space for that cluster. */
1317 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1318 while (num_clusters
> 0) {
1319 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1320 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1321 if (sbi
->s_cluster_ratio
== 1 ||
1322 !ext4_find_delalloc_cluster(inode
, lblk
))
1323 ext4_da_release_space(inode
, 1);
1330 * Delayed allocation stuff
1334 * mpage_da_submit_io - walks through extent of pages and try to write
1335 * them with writepage() call back
1337 * @mpd->inode: inode
1338 * @mpd->first_page: first page of the extent
1339 * @mpd->next_page: page after the last page of the extent
1341 * By the time mpage_da_submit_io() is called we expect all blocks
1342 * to be allocated. this may be wrong if allocation failed.
1344 * As pages are already locked by write_cache_pages(), we can't use it
1346 static int mpage_da_submit_io(struct mpage_da_data
*mpd
,
1347 struct ext4_map_blocks
*map
)
1349 struct pagevec pvec
;
1350 unsigned long index
, end
;
1351 int ret
= 0, err
, nr_pages
, i
;
1352 struct inode
*inode
= mpd
->inode
;
1353 struct address_space
*mapping
= inode
->i_mapping
;
1354 loff_t size
= i_size_read(inode
);
1355 unsigned int len
, block_start
;
1356 struct buffer_head
*bh
, *page_bufs
= NULL
;
1357 sector_t pblock
= 0, cur_logical
= 0;
1358 struct ext4_io_submit io_submit
;
1360 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1361 memset(&io_submit
, 0, sizeof(io_submit
));
1363 * We need to start from the first_page to the next_page - 1
1364 * to make sure we also write the mapped dirty buffer_heads.
1365 * If we look at mpd->b_blocknr we would only be looking
1366 * at the currently mapped buffer_heads.
1368 index
= mpd
->first_page
;
1369 end
= mpd
->next_page
- 1;
1371 pagevec_init(&pvec
, 0);
1372 while (index
<= end
) {
1373 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1376 for (i
= 0; i
< nr_pages
; i
++) {
1377 int commit_write
= 0, skip_page
= 0;
1378 struct page
*page
= pvec
.pages
[i
];
1380 index
= page
->index
;
1384 if (index
== size
>> PAGE_CACHE_SHIFT
)
1385 len
= size
& ~PAGE_CACHE_MASK
;
1387 len
= PAGE_CACHE_SIZE
;
1389 cur_logical
= index
<< (PAGE_CACHE_SHIFT
-
1391 pblock
= map
->m_pblk
+ (cur_logical
-
1396 BUG_ON(!PageLocked(page
));
1397 BUG_ON(PageWriteback(page
));
1400 * If the page does not have buffers (for
1401 * whatever reason), try to create them using
1402 * __block_write_begin. If this fails,
1403 * skip the page and move on.
1405 if (!page_has_buffers(page
)) {
1406 if (__block_write_begin(page
, 0, len
,
1407 noalloc_get_block_write
)) {
1415 bh
= page_bufs
= page_buffers(page
);
1420 if (map
&& (cur_logical
>= map
->m_lblk
) &&
1421 (cur_logical
<= (map
->m_lblk
+
1422 (map
->m_len
- 1)))) {
1423 if (buffer_delay(bh
)) {
1424 clear_buffer_delay(bh
);
1425 bh
->b_blocknr
= pblock
;
1427 if (buffer_unwritten(bh
) ||
1429 BUG_ON(bh
->b_blocknr
!= pblock
);
1430 if (map
->m_flags
& EXT4_MAP_UNINIT
)
1431 set_buffer_uninit(bh
);
1432 clear_buffer_unwritten(bh
);
1436 * skip page if block allocation undone and
1439 if (ext4_bh_delay_or_unwritten(NULL
, bh
))
1441 bh
= bh
->b_this_page
;
1442 block_start
+= bh
->b_size
;
1445 } while (bh
!= page_bufs
);
1451 /* mark the buffer_heads as dirty & uptodate */
1452 block_commit_write(page
, 0, len
);
1454 clear_page_dirty_for_io(page
);
1455 err
= ext4_bio_write_page(&io_submit
, page
, len
,
1458 mpd
->pages_written
++;
1460 * In error case, we have to continue because
1461 * remaining pages are still locked
1466 pagevec_release(&pvec
);
1468 ext4_io_submit(&io_submit
);
1472 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
)
1476 struct pagevec pvec
;
1477 struct inode
*inode
= mpd
->inode
;
1478 struct address_space
*mapping
= inode
->i_mapping
;
1479 ext4_lblk_t start
, last
;
1481 index
= mpd
->first_page
;
1482 end
= mpd
->next_page
- 1;
1484 start
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1485 last
= end
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1486 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1488 pagevec_init(&pvec
, 0);
1489 while (index
<= end
) {
1490 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1493 for (i
= 0; i
< nr_pages
; i
++) {
1494 struct page
*page
= pvec
.pages
[i
];
1495 if (page
->index
> end
)
1497 BUG_ON(!PageLocked(page
));
1498 BUG_ON(PageWriteback(page
));
1499 block_invalidatepage(page
, 0);
1500 ClearPageUptodate(page
);
1503 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1504 pagevec_release(&pvec
);
1509 static void ext4_print_free_blocks(struct inode
*inode
)
1511 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1512 struct super_block
*sb
= inode
->i_sb
;
1514 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1515 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1516 ext4_count_free_clusters(inode
->i_sb
)));
1517 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1518 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1519 (long long) EXT4_C2B(EXT4_SB(inode
->i_sb
),
1520 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1521 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1522 (long long) EXT4_C2B(EXT4_SB(inode
->i_sb
),
1523 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1524 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1525 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1526 EXT4_I(inode
)->i_reserved_data_blocks
);
1527 ext4_msg(sb
, KERN_CRIT
, "i_reserved_meta_blocks=%u",
1528 EXT4_I(inode
)->i_reserved_meta_blocks
);
1533 * mpage_da_map_and_submit - go through given space, map them
1534 * if necessary, and then submit them for I/O
1536 * @mpd - bh describing space
1538 * The function skips space we know is already mapped to disk blocks.
1541 static void mpage_da_map_and_submit(struct mpage_da_data
*mpd
)
1543 int err
, blks
, get_blocks_flags
;
1544 struct ext4_map_blocks map
, *mapp
= NULL
;
1545 sector_t next
= mpd
->b_blocknr
;
1546 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1547 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
1548 handle_t
*handle
= NULL
;
1551 * If the blocks are mapped already, or we couldn't accumulate
1552 * any blocks, then proceed immediately to the submission stage.
1554 if ((mpd
->b_size
== 0) ||
1555 ((mpd
->b_state
& (1 << BH_Mapped
)) &&
1556 !(mpd
->b_state
& (1 << BH_Delay
)) &&
1557 !(mpd
->b_state
& (1 << BH_Unwritten
))))
1560 handle
= ext4_journal_current_handle();
1564 * Call ext4_map_blocks() to allocate any delayed allocation
1565 * blocks, or to convert an uninitialized extent to be
1566 * initialized (in the case where we have written into
1567 * one or more preallocated blocks).
1569 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1570 * indicate that we are on the delayed allocation path. This
1571 * affects functions in many different parts of the allocation
1572 * call path. This flag exists primarily because we don't
1573 * want to change *many* call functions, so ext4_map_blocks()
1574 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1575 * inode's allocation semaphore is taken.
1577 * If the blocks in questions were delalloc blocks, set
1578 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1579 * variables are updated after the blocks have been allocated.
1582 map
.m_len
= max_blocks
;
1583 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
1584 if (ext4_should_dioread_nolock(mpd
->inode
))
1585 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
1586 if (mpd
->b_state
& (1 << BH_Delay
))
1587 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
1589 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
1591 struct super_block
*sb
= mpd
->inode
->i_sb
;
1595 * If get block returns EAGAIN or ENOSPC and there
1596 * appears to be free blocks we will just let
1597 * mpage_da_submit_io() unlock all of the pages.
1602 if (err
== -ENOSPC
&& ext4_count_free_clusters(sb
)) {
1608 * get block failure will cause us to loop in
1609 * writepages, because a_ops->writepage won't be able
1610 * to make progress. The page will be redirtied by
1611 * writepage and writepages will again try to write
1614 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
1615 ext4_msg(sb
, KERN_CRIT
,
1616 "delayed block allocation failed for inode %lu "
1617 "at logical offset %llu with max blocks %zd "
1618 "with error %d", mpd
->inode
->i_ino
,
1619 (unsigned long long) next
,
1620 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1621 ext4_msg(sb
, KERN_CRIT
,
1622 "This should not happen!! Data will be lost\n");
1624 ext4_print_free_blocks(mpd
->inode
);
1626 /* invalidate all the pages */
1627 ext4_da_block_invalidatepages(mpd
);
1629 /* Mark this page range as having been completed */
1636 if (map
.m_flags
& EXT4_MAP_NEW
) {
1637 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
1640 for (i
= 0; i
< map
.m_len
; i
++)
1641 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
1645 * Update on-disk size along with block allocation.
1647 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
1648 if (disksize
> i_size_read(mpd
->inode
))
1649 disksize
= i_size_read(mpd
->inode
);
1650 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
1651 ext4_update_i_disksize(mpd
->inode
, disksize
);
1652 err
= ext4_mark_inode_dirty(handle
, mpd
->inode
);
1654 ext4_error(mpd
->inode
->i_sb
,
1655 "Failed to mark inode %lu dirty",
1660 mpage_da_submit_io(mpd
, mapp
);
1664 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1665 (1 << BH_Delay) | (1 << BH_Unwritten))
1668 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1670 * @mpd->lbh - extent of blocks
1671 * @logical - logical number of the block in the file
1672 * @bh - bh of the block (used to access block's state)
1674 * the function is used to collect contig. blocks in same state
1676 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1677 sector_t logical
, size_t b_size
,
1678 unsigned long b_state
)
1681 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1684 * XXX Don't go larger than mballoc is willing to allocate
1685 * This is a stopgap solution. We eventually need to fold
1686 * mpage_da_submit_io() into this function and then call
1687 * ext4_map_blocks() multiple times in a loop
1689 if (nrblocks
>= 8*1024*1024/mpd
->inode
->i_sb
->s_blocksize
)
1692 /* check if thereserved journal credits might overflow */
1693 if (!(ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
))) {
1694 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1696 * With non-extent format we are limited by the journal
1697 * credit available. Total credit needed to insert
1698 * nrblocks contiguous blocks is dependent on the
1699 * nrblocks. So limit nrblocks.
1702 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
1703 EXT4_MAX_TRANS_DATA
) {
1705 * Adding the new buffer_head would make it cross the
1706 * allowed limit for which we have journal credit
1707 * reserved. So limit the new bh->b_size
1709 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
1710 mpd
->inode
->i_blkbits
;
1711 /* we will do mpage_da_submit_io in the next loop */
1715 * First block in the extent
1717 if (mpd
->b_size
== 0) {
1718 mpd
->b_blocknr
= logical
;
1719 mpd
->b_size
= b_size
;
1720 mpd
->b_state
= b_state
& BH_FLAGS
;
1724 next
= mpd
->b_blocknr
+ nrblocks
;
1726 * Can we merge the block to our big extent?
1728 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
1729 mpd
->b_size
+= b_size
;
1735 * We couldn't merge the block to our extent, so we
1736 * need to flush current extent and start new one
1738 mpage_da_map_and_submit(mpd
);
1742 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1744 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1748 * This function is grabs code from the very beginning of
1749 * ext4_map_blocks, but assumes that the caller is from delayed write
1750 * time. This function looks up the requested blocks and sets the
1751 * buffer delay bit under the protection of i_data_sem.
1753 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1754 struct ext4_map_blocks
*map
,
1755 struct buffer_head
*bh
)
1758 sector_t invalid_block
= ~((sector_t
) 0xffff);
1760 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1764 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1765 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1766 (unsigned long) map
->m_lblk
);
1768 * Try to see if we can get the block without requesting a new
1769 * file system block.
1771 down_read((&EXT4_I(inode
)->i_data_sem
));
1772 if (ext4_has_inline_data(inode
)) {
1774 * We will soon create blocks for this page, and let
1775 * us pretend as if the blocks aren't allocated yet.
1776 * In case of clusters, we have to handle the work
1777 * of mapping from cluster so that the reserved space
1778 * is calculated properly.
1780 if ((EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) &&
1781 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
1782 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
1784 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1785 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1787 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1791 * XXX: __block_prepare_write() unmaps passed block,
1794 /* If the block was allocated from previously allocated cluster,
1795 * then we dont need to reserve it again. */
1796 if (!(map
->m_flags
& EXT4_MAP_FROM_CLUSTER
)) {
1797 retval
= ext4_da_reserve_space(inode
, iblock
);
1799 /* not enough space to reserve */
1803 retval
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
);
1807 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1808 * and it should not appear on the bh->b_state.
1810 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
1812 map_bh(bh
, inode
->i_sb
, invalid_block
);
1814 set_buffer_delay(bh
);
1818 up_read((&EXT4_I(inode
)->i_data_sem
));
1824 * This is a special get_blocks_t callback which is used by
1825 * ext4_da_write_begin(). It will either return mapped block or
1826 * reserve space for a single block.
1828 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1829 * We also have b_blocknr = -1 and b_bdev initialized properly
1831 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1832 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1833 * initialized properly.
1835 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1836 struct buffer_head
*bh
, int create
)
1838 struct ext4_map_blocks map
;
1841 BUG_ON(create
== 0);
1842 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1844 map
.m_lblk
= iblock
;
1848 * first, we need to know whether the block is allocated already
1849 * preallocated blocks are unmapped but should treated
1850 * the same as allocated blocks.
1852 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1856 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1857 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1859 if (buffer_unwritten(bh
)) {
1860 /* A delayed write to unwritten bh should be marked
1861 * new and mapped. Mapped ensures that we don't do
1862 * get_block multiple times when we write to the same
1863 * offset and new ensures that we do proper zero out
1864 * for partial write.
1867 set_buffer_mapped(bh
);
1873 * This function is used as a standard get_block_t calback function when there
1874 * is no desire to allocate any blocks. It is used as a callback function for
1875 * block_write_begin(). These functions should only try to map a single block
1878 * Since this function doesn't do block allocations even if the caller
1879 * requests it by passing in create=1, it is critically important that
1880 * any caller checks to make sure that any buffer heads are returned
1881 * by this function are either all already mapped or marked for
1882 * delayed allocation before calling ext4_bio_write_page(). Otherwise,
1883 * b_blocknr could be left unitialized, and the page write functions will
1884 * be taken by surprise.
1886 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
1887 struct buffer_head
*bh_result
, int create
)
1889 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
1890 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
1893 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1899 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1905 static int __ext4_journalled_writepage(struct page
*page
,
1908 struct address_space
*mapping
= page
->mapping
;
1909 struct inode
*inode
= mapping
->host
;
1910 struct buffer_head
*page_bufs
= NULL
;
1911 handle_t
*handle
= NULL
;
1912 int ret
= 0, err
= 0;
1913 int inline_data
= ext4_has_inline_data(inode
);
1914 struct buffer_head
*inode_bh
= NULL
;
1916 ClearPageChecked(page
);
1919 BUG_ON(page
->index
!= 0);
1920 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1921 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1922 if (inode_bh
== NULL
)
1925 page_bufs
= page_buffers(page
);
1930 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1933 /* As soon as we unlock the page, it can go away, but we have
1934 * references to buffers so we are safe */
1937 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
1938 if (IS_ERR(handle
)) {
1939 ret
= PTR_ERR(handle
);
1943 BUG_ON(!ext4_handle_valid(handle
));
1946 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1948 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1951 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1952 do_journal_get_write_access
);
1954 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1959 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1960 err
= ext4_journal_stop(handle
);
1964 if (!ext4_has_inline_data(inode
))
1965 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1967 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1974 * Note that we don't need to start a transaction unless we're journaling data
1975 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1976 * need to file the inode to the transaction's list in ordered mode because if
1977 * we are writing back data added by write(), the inode is already there and if
1978 * we are writing back data modified via mmap(), no one guarantees in which
1979 * transaction the data will hit the disk. In case we are journaling data, we
1980 * cannot start transaction directly because transaction start ranks above page
1981 * lock so we have to do some magic.
1983 * This function can get called via...
1984 * - ext4_da_writepages after taking page lock (have journal handle)
1985 * - journal_submit_inode_data_buffers (no journal handle)
1986 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1987 * - grab_page_cache when doing write_begin (have journal handle)
1989 * We don't do any block allocation in this function. If we have page with
1990 * multiple blocks we need to write those buffer_heads that are mapped. This
1991 * is important for mmaped based write. So if we do with blocksize 1K
1992 * truncate(f, 1024);
1993 * a = mmap(f, 0, 4096);
1995 * truncate(f, 4096);
1996 * we have in the page first buffer_head mapped via page_mkwrite call back
1997 * but other buffer_heads would be unmapped but dirty (dirty done via the
1998 * do_wp_page). So writepage should write the first block. If we modify
1999 * the mmap area beyond 1024 we will again get a page_fault and the
2000 * page_mkwrite callback will do the block allocation and mark the
2001 * buffer_heads mapped.
2003 * We redirty the page if we have any buffer_heads that is either delay or
2004 * unwritten in the page.
2006 * We can get recursively called as show below.
2008 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2011 * But since we don't do any block allocation we should not deadlock.
2012 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2014 static int ext4_writepage(struct page
*page
,
2015 struct writeback_control
*wbc
)
2017 int ret
= 0, commit_write
= 0;
2020 struct buffer_head
*page_bufs
= NULL
;
2021 struct inode
*inode
= page
->mapping
->host
;
2022 struct ext4_io_submit io_submit
;
2024 trace_ext4_writepage(page
);
2025 size
= i_size_read(inode
);
2026 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2027 len
= size
& ~PAGE_CACHE_MASK
;
2029 len
= PAGE_CACHE_SIZE
;
2032 * If the page does not have buffers (for whatever reason),
2033 * try to create them using __block_write_begin. If this
2034 * fails, redirty the page and move on.
2036 if (!page_has_buffers(page
)) {
2037 if (__block_write_begin(page
, 0, len
,
2038 noalloc_get_block_write
)) {
2040 redirty_page_for_writepage(wbc
, page
);
2046 page_bufs
= page_buffers(page
);
2047 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2048 ext4_bh_delay_or_unwritten
)) {
2050 * We don't want to do block allocation, so redirty
2051 * the page and return. We may reach here when we do
2052 * a journal commit via journal_submit_inode_data_buffers.
2053 * We can also reach here via shrink_page_list but it
2054 * should never be for direct reclaim so warn if that
2057 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
)) ==
2062 /* now mark the buffer_heads as dirty and uptodate */
2063 block_commit_write(page
, 0, len
);
2065 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2067 * It's mmapped pagecache. Add buffers and journal it. There
2068 * doesn't seem much point in redirtying the page here.
2070 return __ext4_journalled_writepage(page
, len
);
2072 memset(&io_submit
, 0, sizeof(io_submit
));
2073 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
);
2074 ext4_io_submit(&io_submit
);
2079 * This is called via ext4_da_writepages() to
2080 * calculate the total number of credits to reserve to fit
2081 * a single extent allocation into a single transaction,
2082 * ext4_da_writpeages() will loop calling this before
2083 * the block allocation.
2086 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2088 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2091 * With non-extent format the journal credit needed to
2092 * insert nrblocks contiguous block is dependent on
2093 * number of contiguous block. So we will limit
2094 * number of contiguous block to a sane value
2096 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
2097 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2098 max_blocks
= EXT4_MAX_TRANS_DATA
;
2100 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2104 * write_cache_pages_da - walk the list of dirty pages of the given
2105 * address space and accumulate pages that need writing, and call
2106 * mpage_da_map_and_submit to map a single contiguous memory region
2107 * and then write them.
2109 static int write_cache_pages_da(handle_t
*handle
,
2110 struct address_space
*mapping
,
2111 struct writeback_control
*wbc
,
2112 struct mpage_da_data
*mpd
,
2113 pgoff_t
*done_index
)
2115 struct buffer_head
*bh
, *head
;
2116 struct inode
*inode
= mapping
->host
;
2117 struct pagevec pvec
;
2118 unsigned int nr_pages
;
2121 long nr_to_write
= wbc
->nr_to_write
;
2122 int i
, tag
, ret
= 0;
2124 memset(mpd
, 0, sizeof(struct mpage_da_data
));
2127 pagevec_init(&pvec
, 0);
2128 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2129 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2131 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2132 tag
= PAGECACHE_TAG_TOWRITE
;
2134 tag
= PAGECACHE_TAG_DIRTY
;
2136 *done_index
= index
;
2137 while (index
<= end
) {
2138 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2139 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2143 for (i
= 0; i
< nr_pages
; i
++) {
2144 struct page
*page
= pvec
.pages
[i
];
2147 * At this point, the page may be truncated or
2148 * invalidated (changing page->mapping to NULL), or
2149 * even swizzled back from swapper_space to tmpfs file
2150 * mapping. However, page->index will not change
2151 * because we have a reference on the page.
2153 if (page
->index
> end
)
2156 *done_index
= page
->index
+ 1;
2159 * If we can't merge this page, and we have
2160 * accumulated an contiguous region, write it
2162 if ((mpd
->next_page
!= page
->index
) &&
2163 (mpd
->next_page
!= mpd
->first_page
)) {
2164 mpage_da_map_and_submit(mpd
);
2165 goto ret_extent_tail
;
2171 * If the page is no longer dirty, or its
2172 * mapping no longer corresponds to inode we
2173 * are writing (which means it has been
2174 * truncated or invalidated), or the page is
2175 * already under writeback and we are not
2176 * doing a data integrity writeback, skip the page
2178 if (!PageDirty(page
) ||
2179 (PageWriteback(page
) &&
2180 (wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2181 unlikely(page
->mapping
!= mapping
)) {
2186 wait_on_page_writeback(page
);
2187 BUG_ON(PageWriteback(page
));
2190 * If we have inline data and arrive here, it means that
2191 * we will soon create the block for the 1st page, so
2192 * we'd better clear the inline data here.
2194 if (ext4_has_inline_data(inode
)) {
2195 BUG_ON(ext4_test_inode_state(inode
,
2196 EXT4_STATE_MAY_INLINE_DATA
));
2197 ext4_destroy_inline_data(handle
, inode
);
2200 if (mpd
->next_page
!= page
->index
)
2201 mpd
->first_page
= page
->index
;
2202 mpd
->next_page
= page
->index
+ 1;
2203 logical
= (sector_t
) page
->index
<<
2204 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2206 if (!page_has_buffers(page
)) {
2207 mpage_add_bh_to_extent(mpd
, logical
,
2209 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2211 goto ret_extent_tail
;
2214 * Page with regular buffer heads,
2215 * just add all dirty ones
2217 head
= page_buffers(page
);
2220 BUG_ON(buffer_locked(bh
));
2222 * We need to try to allocate
2223 * unmapped blocks in the same page.
2224 * Otherwise we won't make progress
2225 * with the page in ext4_writepage
2227 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2228 mpage_add_bh_to_extent(mpd
, logical
,
2232 goto ret_extent_tail
;
2233 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2235 * mapped dirty buffer. We need
2236 * to update the b_state
2237 * because we look at b_state
2238 * in mpage_da_map_blocks. We
2239 * don't update b_size because
2240 * if we find an unmapped
2241 * buffer_head later we need to
2242 * use the b_state flag of that
2245 if (mpd
->b_size
== 0)
2246 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2249 } while ((bh
= bh
->b_this_page
) != head
);
2252 if (nr_to_write
> 0) {
2254 if (nr_to_write
== 0 &&
2255 wbc
->sync_mode
== WB_SYNC_NONE
)
2257 * We stop writing back only if we are
2258 * not doing integrity sync. In case of
2259 * integrity sync we have to keep going
2260 * because someone may be concurrently
2261 * dirtying pages, and we might have
2262 * synced a lot of newly appeared dirty
2263 * pages, but have not synced all of the
2269 pagevec_release(&pvec
);
2274 ret
= MPAGE_DA_EXTENT_TAIL
;
2276 pagevec_release(&pvec
);
2282 static int ext4_da_writepages(struct address_space
*mapping
,
2283 struct writeback_control
*wbc
)
2286 int range_whole
= 0;
2287 handle_t
*handle
= NULL
;
2288 struct mpage_da_data mpd
;
2289 struct inode
*inode
= mapping
->host
;
2290 int pages_written
= 0;
2291 unsigned int max_pages
;
2292 int range_cyclic
, cycled
= 1, io_done
= 0;
2293 int needed_blocks
, ret
= 0;
2294 long desired_nr_to_write
, nr_to_writebump
= 0;
2295 loff_t range_start
= wbc
->range_start
;
2296 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2297 pgoff_t done_index
= 0;
2299 struct blk_plug plug
;
2301 trace_ext4_da_writepages(inode
, wbc
);
2304 * No pages to write? This is mainly a kludge to avoid starting
2305 * a transaction for special inodes like journal inode on last iput()
2306 * because that could violate lock ordering on umount
2308 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2312 * If the filesystem has aborted, it is read-only, so return
2313 * right away instead of dumping stack traces later on that
2314 * will obscure the real source of the problem. We test
2315 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2316 * the latter could be true if the filesystem is mounted
2317 * read-only, and in that case, ext4_da_writepages should
2318 * *never* be called, so if that ever happens, we would want
2321 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2324 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2327 range_cyclic
= wbc
->range_cyclic
;
2328 if (wbc
->range_cyclic
) {
2329 index
= mapping
->writeback_index
;
2332 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2333 wbc
->range_end
= LLONG_MAX
;
2334 wbc
->range_cyclic
= 0;
2337 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2338 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2342 * This works around two forms of stupidity. The first is in
2343 * the writeback code, which caps the maximum number of pages
2344 * written to be 1024 pages. This is wrong on multiple
2345 * levels; different architectues have a different page size,
2346 * which changes the maximum amount of data which gets
2347 * written. Secondly, 4 megabytes is way too small. XFS
2348 * forces this value to be 16 megabytes by multiplying
2349 * nr_to_write parameter by four, and then relies on its
2350 * allocator to allocate larger extents to make them
2351 * contiguous. Unfortunately this brings us to the second
2352 * stupidity, which is that ext4's mballoc code only allocates
2353 * at most 2048 blocks. So we force contiguous writes up to
2354 * the number of dirty blocks in the inode, or
2355 * sbi->max_writeback_mb_bump whichever is smaller.
2357 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2358 if (!range_cyclic
&& range_whole
) {
2359 if (wbc
->nr_to_write
== LONG_MAX
)
2360 desired_nr_to_write
= wbc
->nr_to_write
;
2362 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2364 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2366 if (desired_nr_to_write
> max_pages
)
2367 desired_nr_to_write
= max_pages
;
2369 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2370 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2371 wbc
->nr_to_write
= desired_nr_to_write
;
2375 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2376 tag_pages_for_writeback(mapping
, index
, end
);
2378 blk_start_plug(&plug
);
2379 while (!ret
&& wbc
->nr_to_write
> 0) {
2382 * we insert one extent at a time. So we need
2383 * credit needed for single extent allocation.
2384 * journalled mode is currently not supported
2387 BUG_ON(ext4_should_journal_data(inode
));
2388 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2390 /* start a new transaction*/
2391 handle
= ext4_journal_start(inode
, needed_blocks
);
2392 if (IS_ERR(handle
)) {
2393 ret
= PTR_ERR(handle
);
2394 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2395 "%ld pages, ino %lu; err %d", __func__
,
2396 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2397 blk_finish_plug(&plug
);
2398 goto out_writepages
;
2402 * Now call write_cache_pages_da() to find the next
2403 * contiguous region of logical blocks that need
2404 * blocks to be allocated by ext4 and submit them.
2406 ret
= write_cache_pages_da(handle
, mapping
,
2407 wbc
, &mpd
, &done_index
);
2409 * If we have a contiguous extent of pages and we
2410 * haven't done the I/O yet, map the blocks and submit
2413 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2414 mpage_da_map_and_submit(&mpd
);
2415 ret
= MPAGE_DA_EXTENT_TAIL
;
2417 trace_ext4_da_write_pages(inode
, &mpd
);
2418 wbc
->nr_to_write
-= mpd
.pages_written
;
2420 ext4_journal_stop(handle
);
2422 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2423 /* commit the transaction which would
2424 * free blocks released in the transaction
2427 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2429 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2431 * Got one extent now try with rest of the pages.
2432 * If mpd.retval is set -EIO, journal is aborted.
2433 * So we don't need to write any more.
2435 pages_written
+= mpd
.pages_written
;
2438 } else if (wbc
->nr_to_write
)
2440 * There is no more writeout needed
2441 * or we requested for a noblocking writeout
2442 * and we found the device congested
2446 blk_finish_plug(&plug
);
2447 if (!io_done
&& !cycled
) {
2450 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2451 wbc
->range_end
= mapping
->writeback_index
- 1;
2456 wbc
->range_cyclic
= range_cyclic
;
2457 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2459 * set the writeback_index so that range_cyclic
2460 * mode will write it back later
2462 mapping
->writeback_index
= done_index
;
2465 wbc
->nr_to_write
-= nr_to_writebump
;
2466 wbc
->range_start
= range_start
;
2467 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2471 static int ext4_nonda_switch(struct super_block
*sb
)
2473 s64 free_blocks
, dirty_blocks
;
2474 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2477 * switch to non delalloc mode if we are running low
2478 * on free block. The free block accounting via percpu
2479 * counters can get slightly wrong with percpu_counter_batch getting
2480 * accumulated on each CPU without updating global counters
2481 * Delalloc need an accurate free block accounting. So switch
2482 * to non delalloc when we are near to error range.
2484 free_blocks
= EXT4_C2B(sbi
,
2485 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
));
2486 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2488 * Start pushing delalloc when 1/2 of free blocks are dirty.
2490 if (dirty_blocks
&& (free_blocks
< 2 * dirty_blocks
) &&
2491 !writeback_in_progress(sb
->s_bdi
) &&
2492 down_read_trylock(&sb
->s_umount
)) {
2493 writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2494 up_read(&sb
->s_umount
);
2497 if (2 * free_blocks
< 3 * dirty_blocks
||
2498 free_blocks
< (dirty_blocks
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2500 * free block count is less than 150% of dirty blocks
2501 * or free blocks is less than watermark
2508 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2509 loff_t pos
, unsigned len
, unsigned flags
,
2510 struct page
**pagep
, void **fsdata
)
2512 int ret
, retries
= 0;
2515 struct inode
*inode
= mapping
->host
;
2518 index
= pos
>> PAGE_CACHE_SHIFT
;
2520 if (ext4_nonda_switch(inode
->i_sb
)) {
2521 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2522 return ext4_write_begin(file
, mapping
, pos
,
2523 len
, flags
, pagep
, fsdata
);
2525 *fsdata
= (void *)0;
2526 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2528 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2529 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2542 * With delayed allocation, we don't log the i_disksize update
2543 * if there is delayed block allocation. But we still need
2544 * to journalling the i_disksize update if writes to the end
2545 * of file which has an already mapped buffer.
2547 handle
= ext4_journal_start(inode
, 1);
2548 if (IS_ERR(handle
)) {
2549 ret
= PTR_ERR(handle
);
2552 /* We cannot recurse into the filesystem as the transaction is already
2554 flags
|= AOP_FLAG_NOFS
;
2556 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2558 ext4_journal_stop(handle
);
2564 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2567 ext4_journal_stop(handle
);
2568 page_cache_release(page
);
2570 * block_write_begin may have instantiated a few blocks
2571 * outside i_size. Trim these off again. Don't need
2572 * i_size_read because we hold i_mutex.
2574 if (pos
+ len
> inode
->i_size
)
2575 ext4_truncate_failed_write(inode
);
2578 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2585 * Check if we should update i_disksize
2586 * when write to the end of file but not require block allocation
2588 static int ext4_da_should_update_i_disksize(struct page
*page
,
2589 unsigned long offset
)
2591 struct buffer_head
*bh
;
2592 struct inode
*inode
= page
->mapping
->host
;
2596 bh
= page_buffers(page
);
2597 idx
= offset
>> inode
->i_blkbits
;
2599 for (i
= 0; i
< idx
; i
++)
2600 bh
= bh
->b_this_page
;
2602 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2607 static int ext4_da_write_end(struct file
*file
,
2608 struct address_space
*mapping
,
2609 loff_t pos
, unsigned len
, unsigned copied
,
2610 struct page
*page
, void *fsdata
)
2612 struct inode
*inode
= mapping
->host
;
2614 handle_t
*handle
= ext4_journal_current_handle();
2616 unsigned long start
, end
;
2617 int write_mode
= (int)(unsigned long)fsdata
;
2619 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2620 switch (ext4_inode_journal_mode(inode
)) {
2621 case EXT4_INODE_ORDERED_DATA_MODE
:
2622 return ext4_ordered_write_end(file
, mapping
, pos
,
2623 len
, copied
, page
, fsdata
);
2624 case EXT4_INODE_WRITEBACK_DATA_MODE
:
2625 return ext4_writeback_write_end(file
, mapping
, pos
,
2626 len
, copied
, page
, fsdata
);
2632 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2633 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2634 end
= start
+ copied
- 1;
2637 * generic_write_end() will run mark_inode_dirty() if i_size
2638 * changes. So let's piggyback the i_disksize mark_inode_dirty
2641 new_i_size
= pos
+ copied
;
2642 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2643 if (ext4_has_inline_data(inode
) ||
2644 ext4_da_should_update_i_disksize(page
, end
)) {
2645 down_write(&EXT4_I(inode
)->i_data_sem
);
2646 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
2647 EXT4_I(inode
)->i_disksize
= new_i_size
;
2648 up_write(&EXT4_I(inode
)->i_data_sem
);
2649 /* We need to mark inode dirty even if
2650 * new_i_size is less that inode->i_size
2651 * bu greater than i_disksize.(hint delalloc)
2653 ext4_mark_inode_dirty(handle
, inode
);
2657 if (write_mode
!= CONVERT_INLINE_DATA
&&
2658 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
2659 ext4_has_inline_data(inode
))
2660 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
2663 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2669 ret2
= ext4_journal_stop(handle
);
2673 return ret
? ret
: copied
;
2676 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2679 * Drop reserved blocks
2681 BUG_ON(!PageLocked(page
));
2682 if (!page_has_buffers(page
))
2685 ext4_da_page_release_reservation(page
, offset
);
2688 ext4_invalidatepage(page
, offset
);
2694 * Force all delayed allocation blocks to be allocated for a given inode.
2696 int ext4_alloc_da_blocks(struct inode
*inode
)
2698 trace_ext4_alloc_da_blocks(inode
);
2700 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2701 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2705 * We do something simple for now. The filemap_flush() will
2706 * also start triggering a write of the data blocks, which is
2707 * not strictly speaking necessary (and for users of
2708 * laptop_mode, not even desirable). However, to do otherwise
2709 * would require replicating code paths in:
2711 * ext4_da_writepages() ->
2712 * write_cache_pages() ---> (via passed in callback function)
2713 * __mpage_da_writepage() -->
2714 * mpage_add_bh_to_extent()
2715 * mpage_da_map_blocks()
2717 * The problem is that write_cache_pages(), located in
2718 * mm/page-writeback.c, marks pages clean in preparation for
2719 * doing I/O, which is not desirable if we're not planning on
2722 * We could call write_cache_pages(), and then redirty all of
2723 * the pages by calling redirty_page_for_writepage() but that
2724 * would be ugly in the extreme. So instead we would need to
2725 * replicate parts of the code in the above functions,
2726 * simplifying them because we wouldn't actually intend to
2727 * write out the pages, but rather only collect contiguous
2728 * logical block extents, call the multi-block allocator, and
2729 * then update the buffer heads with the block allocations.
2731 * For now, though, we'll cheat by calling filemap_flush(),
2732 * which will map the blocks, and start the I/O, but not
2733 * actually wait for the I/O to complete.
2735 return filemap_flush(inode
->i_mapping
);
2739 * bmap() is special. It gets used by applications such as lilo and by
2740 * the swapper to find the on-disk block of a specific piece of data.
2742 * Naturally, this is dangerous if the block concerned is still in the
2743 * journal. If somebody makes a swapfile on an ext4 data-journaling
2744 * filesystem and enables swap, then they may get a nasty shock when the
2745 * data getting swapped to that swapfile suddenly gets overwritten by
2746 * the original zero's written out previously to the journal and
2747 * awaiting writeback in the kernel's buffer cache.
2749 * So, if we see any bmap calls here on a modified, data-journaled file,
2750 * take extra steps to flush any blocks which might be in the cache.
2752 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2754 struct inode
*inode
= mapping
->host
;
2759 * We can get here for an inline file via the FIBMAP ioctl
2761 if (ext4_has_inline_data(inode
))
2764 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2765 test_opt(inode
->i_sb
, DELALLOC
)) {
2767 * With delalloc we want to sync the file
2768 * so that we can make sure we allocate
2771 filemap_write_and_wait(mapping
);
2774 if (EXT4_JOURNAL(inode
) &&
2775 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2777 * This is a REALLY heavyweight approach, but the use of
2778 * bmap on dirty files is expected to be extremely rare:
2779 * only if we run lilo or swapon on a freshly made file
2780 * do we expect this to happen.
2782 * (bmap requires CAP_SYS_RAWIO so this does not
2783 * represent an unprivileged user DOS attack --- we'd be
2784 * in trouble if mortal users could trigger this path at
2787 * NB. EXT4_STATE_JDATA is not set on files other than
2788 * regular files. If somebody wants to bmap a directory
2789 * or symlink and gets confused because the buffer
2790 * hasn't yet been flushed to disk, they deserve
2791 * everything they get.
2794 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2795 journal
= EXT4_JOURNAL(inode
);
2796 jbd2_journal_lock_updates(journal
);
2797 err
= jbd2_journal_flush(journal
);
2798 jbd2_journal_unlock_updates(journal
);
2804 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2807 static int ext4_readpage(struct file
*file
, struct page
*page
)
2810 struct inode
*inode
= page
->mapping
->host
;
2812 trace_ext4_readpage(page
);
2814 if (ext4_has_inline_data(inode
))
2815 ret
= ext4_readpage_inline(inode
, page
);
2818 return mpage_readpage(page
, ext4_get_block
);
2824 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2825 struct list_head
*pages
, unsigned nr_pages
)
2827 struct inode
*inode
= mapping
->host
;
2829 /* If the file has inline data, no need to do readpages. */
2830 if (ext4_has_inline_data(inode
))
2833 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2836 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2838 trace_ext4_invalidatepage(page
, offset
);
2840 /* No journalling happens on data buffers when this function is used */
2841 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
2843 block_invalidatepage(page
, offset
);
2846 static int __ext4_journalled_invalidatepage(struct page
*page
,
2847 unsigned long offset
)
2849 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2851 trace_ext4_journalled_invalidatepage(page
, offset
);
2854 * If it's a full truncate we just forget about the pending dirtying
2857 ClearPageChecked(page
);
2859 return jbd2_journal_invalidatepage(journal
, page
, offset
);
2862 /* Wrapper for aops... */
2863 static void ext4_journalled_invalidatepage(struct page
*page
,
2864 unsigned long offset
)
2866 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
) < 0);
2869 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2871 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2873 trace_ext4_releasepage(page
);
2875 WARN_ON(PageChecked(page
));
2876 if (!page_has_buffers(page
))
2879 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2881 return try_to_free_buffers(page
);
2885 * ext4_get_block used when preparing for a DIO write or buffer write.
2886 * We allocate an uinitialized extent if blocks haven't been allocated.
2887 * The extent will be converted to initialized after the IO is complete.
2889 int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
2890 struct buffer_head
*bh_result
, int create
)
2892 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2893 inode
->i_ino
, create
);
2894 return _ext4_get_block(inode
, iblock
, bh_result
,
2895 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
2898 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
2899 struct buffer_head
*bh_result
, int create
)
2901 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2902 inode
->i_ino
, create
);
2903 return _ext4_get_block(inode
, iblock
, bh_result
,
2904 EXT4_GET_BLOCKS_NO_LOCK
);
2907 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
2908 ssize_t size
, void *private, int ret
,
2911 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
2912 ext4_io_end_t
*io_end
= iocb
->private;
2914 /* if not async direct IO or dio with 0 bytes write, just return */
2915 if (!io_end
|| !size
)
2918 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2919 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2920 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
2923 iocb
->private = NULL
;
2925 /* if not aio dio with unwritten extents, just free io and return */
2926 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
2927 ext4_free_io_end(io_end
);
2930 aio_complete(iocb
, ret
, 0);
2931 inode_dio_done(inode
);
2935 io_end
->offset
= offset
;
2936 io_end
->size
= size
;
2938 io_end
->iocb
= iocb
;
2939 io_end
->result
= ret
;
2942 ext4_add_complete_io(io_end
);
2946 * For ext4 extent files, ext4 will do direct-io write to holes,
2947 * preallocated extents, and those write extend the file, no need to
2948 * fall back to buffered IO.
2950 * For holes, we fallocate those blocks, mark them as uninitialized
2951 * If those blocks were preallocated, we mark sure they are split, but
2952 * still keep the range to write as uninitialized.
2954 * The unwritten extents will be converted to written when DIO is completed.
2955 * For async direct IO, since the IO may still pending when return, we
2956 * set up an end_io call back function, which will do the conversion
2957 * when async direct IO completed.
2959 * If the O_DIRECT write will extend the file then add this inode to the
2960 * orphan list. So recovery will truncate it back to the original size
2961 * if the machine crashes during the write.
2964 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
2965 const struct iovec
*iov
, loff_t offset
,
2966 unsigned long nr_segs
)
2968 struct file
*file
= iocb
->ki_filp
;
2969 struct inode
*inode
= file
->f_mapping
->host
;
2971 size_t count
= iov_length(iov
, nr_segs
);
2973 get_block_t
*get_block_func
= NULL
;
2975 loff_t final_size
= offset
+ count
;
2977 /* Use the old path for reads and writes beyond i_size. */
2978 if (rw
!= WRITE
|| final_size
> inode
->i_size
)
2979 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
2981 BUG_ON(iocb
->private == NULL
);
2983 /* If we do a overwrite dio, i_mutex locking can be released */
2984 overwrite
= *((int *)iocb
->private);
2987 atomic_inc(&inode
->i_dio_count
);
2988 down_read(&EXT4_I(inode
)->i_data_sem
);
2989 mutex_unlock(&inode
->i_mutex
);
2993 * We could direct write to holes and fallocate.
2995 * Allocated blocks to fill the hole are marked as
2996 * uninitialized to prevent parallel buffered read to expose
2997 * the stale data before DIO complete the data IO.
2999 * As to previously fallocated extents, ext4 get_block will
3000 * just simply mark the buffer mapped but still keep the
3001 * extents uninitialized.
3003 * For non AIO case, we will convert those unwritten extents
3004 * to written after return back from blockdev_direct_IO.
3006 * For async DIO, the conversion needs to be deferred when the
3007 * IO is completed. The ext4 end_io callback function will be
3008 * called to take care of the conversion work. Here for async
3009 * case, we allocate an io_end structure to hook to the iocb.
3011 iocb
->private = NULL
;
3012 ext4_inode_aio_set(inode
, NULL
);
3013 if (!is_sync_kiocb(iocb
)) {
3014 ext4_io_end_t
*io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
3019 io_end
->flag
|= EXT4_IO_END_DIRECT
;
3020 iocb
->private = io_end
;
3022 * we save the io structure for current async direct
3023 * IO, so that later ext4_map_blocks() could flag the
3024 * io structure whether there is a unwritten extents
3025 * needs to be converted when IO is completed.
3027 ext4_inode_aio_set(inode
, io_end
);
3031 get_block_func
= ext4_get_block_write_nolock
;
3033 get_block_func
= ext4_get_block_write
;
3034 dio_flags
= DIO_LOCKING
;
3036 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3037 inode
->i_sb
->s_bdev
, iov
,
3045 ext4_inode_aio_set(inode
, NULL
);
3047 * The io_end structure takes a reference to the inode, that
3048 * structure needs to be destroyed and the reference to the
3049 * inode need to be dropped, when IO is complete, even with 0
3050 * byte write, or failed.
3052 * In the successful AIO DIO case, the io_end structure will
3053 * be destroyed and the reference to the inode will be dropped
3054 * after the end_io call back function is called.
3056 * In the case there is 0 byte write, or error case, since VFS
3057 * direct IO won't invoke the end_io call back function, we
3058 * need to free the end_io structure here.
3060 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3061 ext4_free_io_end(iocb
->private);
3062 iocb
->private = NULL
;
3063 } else if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3064 EXT4_STATE_DIO_UNWRITTEN
)) {
3067 * for non AIO case, since the IO is already
3068 * completed, we could do the conversion right here
3070 err
= ext4_convert_unwritten_extents(inode
,
3074 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3078 /* take i_mutex locking again if we do a ovewrite dio */
3080 inode_dio_done(inode
);
3081 up_read(&EXT4_I(inode
)->i_data_sem
);
3082 mutex_lock(&inode
->i_mutex
);
3088 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3089 const struct iovec
*iov
, loff_t offset
,
3090 unsigned long nr_segs
)
3092 struct file
*file
= iocb
->ki_filp
;
3093 struct inode
*inode
= file
->f_mapping
->host
;
3097 * If we are doing data journalling we don't support O_DIRECT
3099 if (ext4_should_journal_data(inode
))
3102 /* Let buffer I/O handle the inline data case. */
3103 if (ext4_has_inline_data(inode
))
3106 trace_ext4_direct_IO_enter(inode
, offset
, iov_length(iov
, nr_segs
), rw
);
3107 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3108 ret
= ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3110 ret
= ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3111 trace_ext4_direct_IO_exit(inode
, offset
,
3112 iov_length(iov
, nr_segs
), rw
, ret
);
3117 * Pages can be marked dirty completely asynchronously from ext4's journalling
3118 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3119 * much here because ->set_page_dirty is called under VFS locks. The page is
3120 * not necessarily locked.
3122 * We cannot just dirty the page and leave attached buffers clean, because the
3123 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3124 * or jbddirty because all the journalling code will explode.
3126 * So what we do is to mark the page "pending dirty" and next time writepage
3127 * is called, propagate that into the buffers appropriately.
3129 static int ext4_journalled_set_page_dirty(struct page
*page
)
3131 SetPageChecked(page
);
3132 return __set_page_dirty_nobuffers(page
);
3135 static const struct address_space_operations ext4_ordered_aops
= {
3136 .readpage
= ext4_readpage
,
3137 .readpages
= ext4_readpages
,
3138 .writepage
= ext4_writepage
,
3139 .write_begin
= ext4_write_begin
,
3140 .write_end
= ext4_ordered_write_end
,
3142 .invalidatepage
= ext4_invalidatepage
,
3143 .releasepage
= ext4_releasepage
,
3144 .direct_IO
= ext4_direct_IO
,
3145 .migratepage
= buffer_migrate_page
,
3146 .is_partially_uptodate
= block_is_partially_uptodate
,
3147 .error_remove_page
= generic_error_remove_page
,
3150 static const struct address_space_operations ext4_writeback_aops
= {
3151 .readpage
= ext4_readpage
,
3152 .readpages
= ext4_readpages
,
3153 .writepage
= ext4_writepage
,
3154 .write_begin
= ext4_write_begin
,
3155 .write_end
= ext4_writeback_write_end
,
3157 .invalidatepage
= ext4_invalidatepage
,
3158 .releasepage
= ext4_releasepage
,
3159 .direct_IO
= ext4_direct_IO
,
3160 .migratepage
= buffer_migrate_page
,
3161 .is_partially_uptodate
= block_is_partially_uptodate
,
3162 .error_remove_page
= generic_error_remove_page
,
3165 static const struct address_space_operations ext4_journalled_aops
= {
3166 .readpage
= ext4_readpage
,
3167 .readpages
= ext4_readpages
,
3168 .writepage
= ext4_writepage
,
3169 .write_begin
= ext4_write_begin
,
3170 .write_end
= ext4_journalled_write_end
,
3171 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3173 .invalidatepage
= ext4_journalled_invalidatepage
,
3174 .releasepage
= ext4_releasepage
,
3175 .direct_IO
= ext4_direct_IO
,
3176 .is_partially_uptodate
= block_is_partially_uptodate
,
3177 .error_remove_page
= generic_error_remove_page
,
3180 static const struct address_space_operations ext4_da_aops
= {
3181 .readpage
= ext4_readpage
,
3182 .readpages
= ext4_readpages
,
3183 .writepage
= ext4_writepage
,
3184 .writepages
= ext4_da_writepages
,
3185 .write_begin
= ext4_da_write_begin
,
3186 .write_end
= ext4_da_write_end
,
3188 .invalidatepage
= ext4_da_invalidatepage
,
3189 .releasepage
= ext4_releasepage
,
3190 .direct_IO
= ext4_direct_IO
,
3191 .migratepage
= buffer_migrate_page
,
3192 .is_partially_uptodate
= block_is_partially_uptodate
,
3193 .error_remove_page
= generic_error_remove_page
,
3196 void ext4_set_aops(struct inode
*inode
)
3198 switch (ext4_inode_journal_mode(inode
)) {
3199 case EXT4_INODE_ORDERED_DATA_MODE
:
3200 if (test_opt(inode
->i_sb
, DELALLOC
))
3201 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3203 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3205 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3206 if (test_opt(inode
->i_sb
, DELALLOC
))
3207 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3209 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3211 case EXT4_INODE_JOURNAL_DATA_MODE
:
3212 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3221 * ext4_discard_partial_page_buffers()
3222 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3223 * This function finds and locks the page containing the offset
3224 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3225 * Calling functions that already have the page locked should call
3226 * ext4_discard_partial_page_buffers_no_lock directly.
3228 int ext4_discard_partial_page_buffers(handle_t
*handle
,
3229 struct address_space
*mapping
, loff_t from
,
3230 loff_t length
, int flags
)
3232 struct inode
*inode
= mapping
->host
;
3236 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3237 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3241 err
= ext4_discard_partial_page_buffers_no_lock(handle
, inode
, page
,
3242 from
, length
, flags
);
3245 page_cache_release(page
);
3250 * ext4_discard_partial_page_buffers_no_lock()
3251 * Zeros a page range of length 'length' starting from offset 'from'.
3252 * Buffer heads that correspond to the block aligned regions of the
3253 * zeroed range will be unmapped. Unblock aligned regions
3254 * will have the corresponding buffer head mapped if needed so that
3255 * that region of the page can be updated with the partial zero out.
3257 * This function assumes that the page has already been locked. The
3258 * The range to be discarded must be contained with in the given page.
3259 * If the specified range exceeds the end of the page it will be shortened
3260 * to the end of the page that corresponds to 'from'. This function is
3261 * appropriate for updating a page and it buffer heads to be unmapped and
3262 * zeroed for blocks that have been either released, or are going to be
3265 * handle: The journal handle
3266 * inode: The files inode
3267 * page: A locked page that contains the offset "from"
3268 * from: The starting byte offset (from the beginning of the file)
3269 * to begin discarding
3270 * len: The length of bytes to discard
3271 * flags: Optional flags that may be used:
3273 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3274 * Only zero the regions of the page whose buffer heads
3275 * have already been unmapped. This flag is appropriate
3276 * for updating the contents of a page whose blocks may
3277 * have already been released, and we only want to zero
3278 * out the regions that correspond to those released blocks.
3280 * Returns zero on success or negative on failure.
3282 static int ext4_discard_partial_page_buffers_no_lock(handle_t
*handle
,
3283 struct inode
*inode
, struct page
*page
, loff_t from
,
3284 loff_t length
, int flags
)
3286 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3287 unsigned int offset
= from
& (PAGE_CACHE_SIZE
-1);
3288 unsigned int blocksize
, max
, pos
;
3290 struct buffer_head
*bh
;
3293 blocksize
= inode
->i_sb
->s_blocksize
;
3294 max
= PAGE_CACHE_SIZE
- offset
;
3296 if (index
!= page
->index
)
3300 * correct length if it does not fall between
3301 * 'from' and the end of the page
3303 if (length
> max
|| length
< 0)
3306 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3308 if (!page_has_buffers(page
))
3309 create_empty_buffers(page
, blocksize
, 0);
3311 /* Find the buffer that contains "offset" */
3312 bh
= page_buffers(page
);
3314 while (offset
>= pos
) {
3315 bh
= bh
->b_this_page
;
3321 while (pos
< offset
+ length
) {
3322 unsigned int end_of_block
, range_to_discard
;
3326 /* The length of space left to zero and unmap */
3327 range_to_discard
= offset
+ length
- pos
;
3329 /* The length of space until the end of the block */
3330 end_of_block
= blocksize
- (pos
& (blocksize
-1));
3333 * Do not unmap or zero past end of block
3334 * for this buffer head
3336 if (range_to_discard
> end_of_block
)
3337 range_to_discard
= end_of_block
;
3341 * Skip this buffer head if we are only zeroing unampped
3342 * regions of the page
3344 if (flags
& EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
&&
3348 /* If the range is block aligned, unmap */
3349 if (range_to_discard
== blocksize
) {
3350 clear_buffer_dirty(bh
);
3352 clear_buffer_mapped(bh
);
3353 clear_buffer_req(bh
);
3354 clear_buffer_new(bh
);
3355 clear_buffer_delay(bh
);
3356 clear_buffer_unwritten(bh
);
3357 clear_buffer_uptodate(bh
);
3358 zero_user(page
, pos
, range_to_discard
);
3359 BUFFER_TRACE(bh
, "Buffer discarded");
3364 * If this block is not completely contained in the range
3365 * to be discarded, then it is not going to be released. Because
3366 * we need to keep this block, we need to make sure this part
3367 * of the page is uptodate before we modify it by writeing
3368 * partial zeros on it.
3370 if (!buffer_mapped(bh
)) {
3372 * Buffer head must be mapped before we can read
3375 BUFFER_TRACE(bh
, "unmapped");
3376 ext4_get_block(inode
, iblock
, bh
, 0);
3377 /* unmapped? It's a hole - nothing to do */
3378 if (!buffer_mapped(bh
)) {
3379 BUFFER_TRACE(bh
, "still unmapped");
3384 /* Ok, it's mapped. Make sure it's up-to-date */
3385 if (PageUptodate(page
))
3386 set_buffer_uptodate(bh
);
3388 if (!buffer_uptodate(bh
)) {
3390 ll_rw_block(READ
, 1, &bh
);
3392 /* Uhhuh. Read error. Complain and punt.*/
3393 if (!buffer_uptodate(bh
))
3397 if (ext4_should_journal_data(inode
)) {
3398 BUFFER_TRACE(bh
, "get write access");
3399 err
= ext4_journal_get_write_access(handle
, bh
);
3404 zero_user(page
, pos
, range_to_discard
);
3407 if (ext4_should_journal_data(inode
)) {
3408 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3410 mark_buffer_dirty(bh
);
3412 BUFFER_TRACE(bh
, "Partial buffer zeroed");
3414 bh
= bh
->b_this_page
;
3416 pos
+= range_to_discard
;
3422 int ext4_can_truncate(struct inode
*inode
)
3424 if (S_ISREG(inode
->i_mode
))
3426 if (S_ISDIR(inode
->i_mode
))
3428 if (S_ISLNK(inode
->i_mode
))
3429 return !ext4_inode_is_fast_symlink(inode
);
3434 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3435 * associated with the given offset and length
3437 * @inode: File inode
3438 * @offset: The offset where the hole will begin
3439 * @len: The length of the hole
3441 * Returns: 0 on success or negative on failure
3444 int ext4_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
3446 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
3447 if (!S_ISREG(inode
->i_mode
))
3450 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3451 return ext4_ind_punch_hole(file
, offset
, length
);
3453 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) {
3454 /* TODO: Add support for bigalloc file systems */
3458 trace_ext4_punch_hole(inode
, offset
, length
);
3460 return ext4_ext_punch_hole(file
, offset
, length
);
3466 * We block out ext4_get_block() block instantiations across the entire
3467 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3468 * simultaneously on behalf of the same inode.
3470 * As we work through the truncate and commit bits of it to the journal there
3471 * is one core, guiding principle: the file's tree must always be consistent on
3472 * disk. We must be able to restart the truncate after a crash.
3474 * The file's tree may be transiently inconsistent in memory (although it
3475 * probably isn't), but whenever we close off and commit a journal transaction,
3476 * the contents of (the filesystem + the journal) must be consistent and
3477 * restartable. It's pretty simple, really: bottom up, right to left (although
3478 * left-to-right works OK too).
3480 * Note that at recovery time, journal replay occurs *before* the restart of
3481 * truncate against the orphan inode list.
3483 * The committed inode has the new, desired i_size (which is the same as
3484 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3485 * that this inode's truncate did not complete and it will again call
3486 * ext4_truncate() to have another go. So there will be instantiated blocks
3487 * to the right of the truncation point in a crashed ext4 filesystem. But
3488 * that's fine - as long as they are linked from the inode, the post-crash
3489 * ext4_truncate() run will find them and release them.
3491 void ext4_truncate(struct inode
*inode
)
3493 trace_ext4_truncate_enter(inode
);
3495 if (!ext4_can_truncate(inode
))
3498 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3500 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3501 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3503 if (ext4_has_inline_data(inode
)) {
3506 ext4_inline_data_truncate(inode
, &has_inline
);
3511 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3512 ext4_ext_truncate(inode
);
3514 ext4_ind_truncate(inode
);
3516 trace_ext4_truncate_exit(inode
);
3520 * ext4_get_inode_loc returns with an extra refcount against the inode's
3521 * underlying buffer_head on success. If 'in_mem' is true, we have all
3522 * data in memory that is needed to recreate the on-disk version of this
3525 static int __ext4_get_inode_loc(struct inode
*inode
,
3526 struct ext4_iloc
*iloc
, int in_mem
)
3528 struct ext4_group_desc
*gdp
;
3529 struct buffer_head
*bh
;
3530 struct super_block
*sb
= inode
->i_sb
;
3532 int inodes_per_block
, inode_offset
;
3535 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3538 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3539 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3544 * Figure out the offset within the block group inode table
3546 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3547 inode_offset
= ((inode
->i_ino
- 1) %
3548 EXT4_INODES_PER_GROUP(sb
));
3549 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3550 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3552 bh
= sb_getblk(sb
, block
);
3555 if (!buffer_uptodate(bh
)) {
3559 * If the buffer has the write error flag, we have failed
3560 * to write out another inode in the same block. In this
3561 * case, we don't have to read the block because we may
3562 * read the old inode data successfully.
3564 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3565 set_buffer_uptodate(bh
);
3567 if (buffer_uptodate(bh
)) {
3568 /* someone brought it uptodate while we waited */
3574 * If we have all information of the inode in memory and this
3575 * is the only valid inode in the block, we need not read the
3579 struct buffer_head
*bitmap_bh
;
3582 start
= inode_offset
& ~(inodes_per_block
- 1);
3584 /* Is the inode bitmap in cache? */
3585 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3586 if (unlikely(!bitmap_bh
))
3590 * If the inode bitmap isn't in cache then the
3591 * optimisation may end up performing two reads instead
3592 * of one, so skip it.
3594 if (!buffer_uptodate(bitmap_bh
)) {
3598 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3599 if (i
== inode_offset
)
3601 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3605 if (i
== start
+ inodes_per_block
) {
3606 /* all other inodes are free, so skip I/O */
3607 memset(bh
->b_data
, 0, bh
->b_size
);
3608 set_buffer_uptodate(bh
);
3616 * If we need to do any I/O, try to pre-readahead extra
3617 * blocks from the inode table.
3619 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3620 ext4_fsblk_t b
, end
, table
;
3623 table
= ext4_inode_table(sb
, gdp
);
3624 /* s_inode_readahead_blks is always a power of 2 */
3625 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
3628 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
3629 num
= EXT4_INODES_PER_GROUP(sb
);
3630 if (ext4_has_group_desc_csum(sb
))
3631 num
-= ext4_itable_unused_count(sb
, gdp
);
3632 table
+= num
/ inodes_per_block
;
3636 sb_breadahead(sb
, b
++);
3640 * There are other valid inodes in the buffer, this inode
3641 * has in-inode xattrs, or we don't have this inode in memory.
3642 * Read the block from disk.
3644 trace_ext4_load_inode(inode
);
3646 bh
->b_end_io
= end_buffer_read_sync
;
3647 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
3649 if (!buffer_uptodate(bh
)) {
3650 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3651 "unable to read itable block");
3661 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3663 /* We have all inode data except xattrs in memory here. */
3664 return __ext4_get_inode_loc(inode
, iloc
,
3665 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
3668 void ext4_set_inode_flags(struct inode
*inode
)
3670 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3672 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3673 if (flags
& EXT4_SYNC_FL
)
3674 inode
->i_flags
|= S_SYNC
;
3675 if (flags
& EXT4_APPEND_FL
)
3676 inode
->i_flags
|= S_APPEND
;
3677 if (flags
& EXT4_IMMUTABLE_FL
)
3678 inode
->i_flags
|= S_IMMUTABLE
;
3679 if (flags
& EXT4_NOATIME_FL
)
3680 inode
->i_flags
|= S_NOATIME
;
3681 if (flags
& EXT4_DIRSYNC_FL
)
3682 inode
->i_flags
|= S_DIRSYNC
;
3685 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3686 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3688 unsigned int vfs_fl
;
3689 unsigned long old_fl
, new_fl
;
3692 vfs_fl
= ei
->vfs_inode
.i_flags
;
3693 old_fl
= ei
->i_flags
;
3694 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3695 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
3697 if (vfs_fl
& S_SYNC
)
3698 new_fl
|= EXT4_SYNC_FL
;
3699 if (vfs_fl
& S_APPEND
)
3700 new_fl
|= EXT4_APPEND_FL
;
3701 if (vfs_fl
& S_IMMUTABLE
)
3702 new_fl
|= EXT4_IMMUTABLE_FL
;
3703 if (vfs_fl
& S_NOATIME
)
3704 new_fl
|= EXT4_NOATIME_FL
;
3705 if (vfs_fl
& S_DIRSYNC
)
3706 new_fl
|= EXT4_DIRSYNC_FL
;
3707 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
3710 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3711 struct ext4_inode_info
*ei
)
3714 struct inode
*inode
= &(ei
->vfs_inode
);
3715 struct super_block
*sb
= inode
->i_sb
;
3717 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3718 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3719 /* we are using combined 48 bit field */
3720 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
3721 le32_to_cpu(raw_inode
->i_blocks_lo
);
3722 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
3723 /* i_blocks represent file system block size */
3724 return i_blocks
<< (inode
->i_blkbits
- 9);
3729 return le32_to_cpu(raw_inode
->i_blocks_lo
);
3733 static inline void ext4_iget_extra_inode(struct inode
*inode
,
3734 struct ext4_inode
*raw_inode
,
3735 struct ext4_inode_info
*ei
)
3737 __le32
*magic
= (void *)raw_inode
+
3738 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
3739 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
3740 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
3741 ext4_find_inline_data_nolock(inode
);
3743 EXT4_I(inode
)->i_inline_off
= 0;
3746 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
3748 struct ext4_iloc iloc
;
3749 struct ext4_inode
*raw_inode
;
3750 struct ext4_inode_info
*ei
;
3751 struct inode
*inode
;
3752 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
3758 inode
= iget_locked(sb
, ino
);
3760 return ERR_PTR(-ENOMEM
);
3761 if (!(inode
->i_state
& I_NEW
))
3767 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
3770 raw_inode
= ext4_raw_inode(&iloc
);
3772 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3773 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
3774 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
3775 EXT4_INODE_SIZE(inode
->i_sb
)) {
3776 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
3777 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
3778 EXT4_INODE_SIZE(inode
->i_sb
));
3783 ei
->i_extra_isize
= 0;
3785 /* Precompute checksum seed for inode metadata */
3786 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3787 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
)) {
3788 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3790 __le32 inum
= cpu_to_le32(inode
->i_ino
);
3791 __le32 gen
= raw_inode
->i_generation
;
3792 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
3794 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
3798 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
3799 EXT4_ERROR_INODE(inode
, "checksum invalid");
3804 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
3805 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
3806 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
3807 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
3808 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
3809 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
3811 i_uid_write(inode
, i_uid
);
3812 i_gid_write(inode
, i_gid
);
3813 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
3815 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
3816 ei
->i_inline_off
= 0;
3817 ei
->i_dir_start_lookup
= 0;
3818 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
3819 /* We now have enough fields to check if the inode was active or not.
3820 * This is needed because nfsd might try to access dead inodes
3821 * the test is that same one that e2fsck uses
3822 * NeilBrown 1999oct15
3824 if (inode
->i_nlink
== 0) {
3825 if (inode
->i_mode
== 0 ||
3826 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
3827 /* this inode is deleted */
3831 /* The only unlinked inodes we let through here have
3832 * valid i_mode and are being read by the orphan
3833 * recovery code: that's fine, we're about to complete
3834 * the process of deleting those. */
3836 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
3837 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
3838 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
3839 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
3841 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
3842 inode
->i_size
= ext4_isize(raw_inode
);
3843 ei
->i_disksize
= inode
->i_size
;
3845 ei
->i_reserved_quota
= 0;
3847 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
3848 ei
->i_block_group
= iloc
.block_group
;
3849 ei
->i_last_alloc_group
= ~0;
3851 * NOTE! The in-memory inode i_data array is in little-endian order
3852 * even on big-endian machines: we do NOT byteswap the block numbers!
3854 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
3855 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
3856 INIT_LIST_HEAD(&ei
->i_orphan
);
3859 * Set transaction id's of transactions that have to be committed
3860 * to finish f[data]sync. We set them to currently running transaction
3861 * as we cannot be sure that the inode or some of its metadata isn't
3862 * part of the transaction - the inode could have been reclaimed and
3863 * now it is reread from disk.
3866 transaction_t
*transaction
;
3869 read_lock(&journal
->j_state_lock
);
3870 if (journal
->j_running_transaction
)
3871 transaction
= journal
->j_running_transaction
;
3873 transaction
= journal
->j_committing_transaction
;
3875 tid
= transaction
->t_tid
;
3877 tid
= journal
->j_commit_sequence
;
3878 read_unlock(&journal
->j_state_lock
);
3879 ei
->i_sync_tid
= tid
;
3880 ei
->i_datasync_tid
= tid
;
3883 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3884 if (ei
->i_extra_isize
== 0) {
3885 /* The extra space is currently unused. Use it. */
3886 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
3887 EXT4_GOOD_OLD_INODE_SIZE
;
3889 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
3893 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
3894 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
3895 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
3896 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
3898 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
3899 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3900 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
3902 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
3906 if (ei
->i_file_acl
&&
3907 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
3908 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
3912 } else if (!ext4_has_inline_data(inode
)) {
3913 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
3914 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
3915 (S_ISLNK(inode
->i_mode
) &&
3916 !ext4_inode_is_fast_symlink(inode
))))
3917 /* Validate extent which is part of inode */
3918 ret
= ext4_ext_check_inode(inode
);
3919 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
3920 (S_ISLNK(inode
->i_mode
) &&
3921 !ext4_inode_is_fast_symlink(inode
))) {
3922 /* Validate block references which are part of inode */
3923 ret
= ext4_ind_check_inode(inode
);
3929 if (S_ISREG(inode
->i_mode
)) {
3930 inode
->i_op
= &ext4_file_inode_operations
;
3931 inode
->i_fop
= &ext4_file_operations
;
3932 ext4_set_aops(inode
);
3933 } else if (S_ISDIR(inode
->i_mode
)) {
3934 inode
->i_op
= &ext4_dir_inode_operations
;
3935 inode
->i_fop
= &ext4_dir_operations
;
3936 } else if (S_ISLNK(inode
->i_mode
)) {
3937 if (ext4_inode_is_fast_symlink(inode
)) {
3938 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
3939 nd_terminate_link(ei
->i_data
, inode
->i_size
,
3940 sizeof(ei
->i_data
) - 1);
3942 inode
->i_op
= &ext4_symlink_inode_operations
;
3943 ext4_set_aops(inode
);
3945 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
3946 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
3947 inode
->i_op
= &ext4_special_inode_operations
;
3948 if (raw_inode
->i_block
[0])
3949 init_special_inode(inode
, inode
->i_mode
,
3950 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
3952 init_special_inode(inode
, inode
->i_mode
,
3953 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
3956 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
3960 ext4_set_inode_flags(inode
);
3961 unlock_new_inode(inode
);
3967 return ERR_PTR(ret
);
3970 static int ext4_inode_blocks_set(handle_t
*handle
,
3971 struct ext4_inode
*raw_inode
,
3972 struct ext4_inode_info
*ei
)
3974 struct inode
*inode
= &(ei
->vfs_inode
);
3975 u64 i_blocks
= inode
->i_blocks
;
3976 struct super_block
*sb
= inode
->i_sb
;
3978 if (i_blocks
<= ~0U) {
3980 * i_blocks can be represented in a 32 bit variable
3981 * as multiple of 512 bytes
3983 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3984 raw_inode
->i_blocks_high
= 0;
3985 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
3988 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
3991 if (i_blocks
<= 0xffffffffffffULL
) {
3993 * i_blocks can be represented in a 48 bit variable
3994 * as multiple of 512 bytes
3996 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3997 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
3998 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4000 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4001 /* i_block is stored in file system block size */
4002 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4003 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4004 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4010 * Post the struct inode info into an on-disk inode location in the
4011 * buffer-cache. This gobbles the caller's reference to the
4012 * buffer_head in the inode location struct.
4014 * The caller must have write access to iloc->bh.
4016 static int ext4_do_update_inode(handle_t
*handle
,
4017 struct inode
*inode
,
4018 struct ext4_iloc
*iloc
)
4020 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4021 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4022 struct buffer_head
*bh
= iloc
->bh
;
4023 int err
= 0, rc
, block
;
4024 int need_datasync
= 0;
4028 /* For fields not not tracking in the in-memory inode,
4029 * initialise them to zero for new inodes. */
4030 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4031 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4033 ext4_get_inode_flags(ei
);
4034 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4035 i_uid
= i_uid_read(inode
);
4036 i_gid
= i_gid_read(inode
);
4037 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4038 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4039 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4041 * Fix up interoperability with old kernels. Otherwise, old inodes get
4042 * re-used with the upper 16 bits of the uid/gid intact
4045 raw_inode
->i_uid_high
=
4046 cpu_to_le16(high_16_bits(i_uid
));
4047 raw_inode
->i_gid_high
=
4048 cpu_to_le16(high_16_bits(i_gid
));
4050 raw_inode
->i_uid_high
= 0;
4051 raw_inode
->i_gid_high
= 0;
4054 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4055 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4056 raw_inode
->i_uid_high
= 0;
4057 raw_inode
->i_gid_high
= 0;
4059 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4061 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4062 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4063 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4064 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4066 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4068 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4069 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4070 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4071 cpu_to_le32(EXT4_OS_HURD
))
4072 raw_inode
->i_file_acl_high
=
4073 cpu_to_le16(ei
->i_file_acl
>> 32);
4074 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4075 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4076 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4079 if (ei
->i_disksize
> 0x7fffffffULL
) {
4080 struct super_block
*sb
= inode
->i_sb
;
4081 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4082 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4083 EXT4_SB(sb
)->s_es
->s_rev_level
==
4084 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4085 /* If this is the first large file
4086 * created, add a flag to the superblock.
4088 err
= ext4_journal_get_write_access(handle
,
4089 EXT4_SB(sb
)->s_sbh
);
4092 ext4_update_dynamic_rev(sb
);
4093 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4094 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4095 ext4_handle_sync(handle
);
4096 err
= ext4_handle_dirty_super(handle
, sb
);
4099 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4100 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4101 if (old_valid_dev(inode
->i_rdev
)) {
4102 raw_inode
->i_block
[0] =
4103 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4104 raw_inode
->i_block
[1] = 0;
4106 raw_inode
->i_block
[0] = 0;
4107 raw_inode
->i_block
[1] =
4108 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4109 raw_inode
->i_block
[2] = 0;
4111 } else if (!ext4_has_inline_data(inode
)) {
4112 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4113 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4116 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4117 if (ei
->i_extra_isize
) {
4118 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4119 raw_inode
->i_version_hi
=
4120 cpu_to_le32(inode
->i_version
>> 32);
4121 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4124 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4126 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4127 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4130 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4132 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4135 ext4_std_error(inode
->i_sb
, err
);
4140 * ext4_write_inode()
4142 * We are called from a few places:
4144 * - Within generic_file_write() for O_SYNC files.
4145 * Here, there will be no transaction running. We wait for any running
4146 * transaction to commit.
4148 * - Within sys_sync(), kupdate and such.
4149 * We wait on commit, if tol to.
4151 * - Within prune_icache() (PF_MEMALLOC == true)
4152 * Here we simply return. We can't afford to block kswapd on the
4155 * In all cases it is actually safe for us to return without doing anything,
4156 * because the inode has been copied into a raw inode buffer in
4157 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4160 * Note that we are absolutely dependent upon all inode dirtiers doing the
4161 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4162 * which we are interested.
4164 * It would be a bug for them to not do this. The code:
4166 * mark_inode_dirty(inode)
4168 * inode->i_size = expr;
4170 * is in error because a kswapd-driven write_inode() could occur while
4171 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4172 * will no longer be on the superblock's dirty inode list.
4174 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4178 if (current
->flags
& PF_MEMALLOC
)
4181 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4182 if (ext4_journal_current_handle()) {
4183 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4188 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
4191 err
= ext4_force_commit(inode
->i_sb
);
4193 struct ext4_iloc iloc
;
4195 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4198 if (wbc
->sync_mode
== WB_SYNC_ALL
)
4199 sync_dirty_buffer(iloc
.bh
);
4200 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4201 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4202 "IO error syncing inode");
4211 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4212 * buffers that are attached to a page stradding i_size and are undergoing
4213 * commit. In that case we have to wait for commit to finish and try again.
4215 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4219 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4220 tid_t commit_tid
= 0;
4223 offset
= inode
->i_size
& (PAGE_CACHE_SIZE
- 1);
4225 * All buffers in the last page remain valid? Then there's nothing to
4226 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4229 if (offset
> PAGE_CACHE_SIZE
- (1 << inode
->i_blkbits
))
4232 page
= find_lock_page(inode
->i_mapping
,
4233 inode
->i_size
>> PAGE_CACHE_SHIFT
);
4236 ret
= __ext4_journalled_invalidatepage(page
, offset
);
4238 page_cache_release(page
);
4242 read_lock(&journal
->j_state_lock
);
4243 if (journal
->j_committing_transaction
)
4244 commit_tid
= journal
->j_committing_transaction
->t_tid
;
4245 read_unlock(&journal
->j_state_lock
);
4247 jbd2_log_wait_commit(journal
, commit_tid
);
4254 * Called from notify_change.
4256 * We want to trap VFS attempts to truncate the file as soon as
4257 * possible. In particular, we want to make sure that when the VFS
4258 * shrinks i_size, we put the inode on the orphan list and modify
4259 * i_disksize immediately, so that during the subsequent flushing of
4260 * dirty pages and freeing of disk blocks, we can guarantee that any
4261 * commit will leave the blocks being flushed in an unused state on
4262 * disk. (On recovery, the inode will get truncated and the blocks will
4263 * be freed, so we have a strong guarantee that no future commit will
4264 * leave these blocks visible to the user.)
4266 * Another thing we have to assure is that if we are in ordered mode
4267 * and inode is still attached to the committing transaction, we must
4268 * we start writeout of all the dirty pages which are being truncated.
4269 * This way we are sure that all the data written in the previous
4270 * transaction are already on disk (truncate waits for pages under
4273 * Called with inode->i_mutex down.
4275 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4277 struct inode
*inode
= dentry
->d_inode
;
4280 const unsigned int ia_valid
= attr
->ia_valid
;
4282 error
= inode_change_ok(inode
, attr
);
4286 if (is_quota_modification(inode
, attr
))
4287 dquot_initialize(inode
);
4288 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4289 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4292 /* (user+group)*(old+new) structure, inode write (sb,
4293 * inode block, ? - but truncate inode update has it) */
4294 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
4295 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
4296 if (IS_ERR(handle
)) {
4297 error
= PTR_ERR(handle
);
4300 error
= dquot_transfer(inode
, attr
);
4302 ext4_journal_stop(handle
);
4305 /* Update corresponding info in inode so that everything is in
4306 * one transaction */
4307 if (attr
->ia_valid
& ATTR_UID
)
4308 inode
->i_uid
= attr
->ia_uid
;
4309 if (attr
->ia_valid
& ATTR_GID
)
4310 inode
->i_gid
= attr
->ia_gid
;
4311 error
= ext4_mark_inode_dirty(handle
, inode
);
4312 ext4_journal_stop(handle
);
4315 if (attr
->ia_valid
& ATTR_SIZE
) {
4317 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4318 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4320 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4325 if (S_ISREG(inode
->i_mode
) &&
4326 attr
->ia_valid
& ATTR_SIZE
&&
4327 (attr
->ia_size
< inode
->i_size
)) {
4330 handle
= ext4_journal_start(inode
, 3);
4331 if (IS_ERR(handle
)) {
4332 error
= PTR_ERR(handle
);
4335 if (ext4_handle_valid(handle
)) {
4336 error
= ext4_orphan_add(handle
, inode
);
4339 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4340 rc
= ext4_mark_inode_dirty(handle
, inode
);
4343 ext4_journal_stop(handle
);
4345 if (ext4_should_order_data(inode
)) {
4346 error
= ext4_begin_ordered_truncate(inode
,
4349 /* Do as much error cleanup as possible */
4350 handle
= ext4_journal_start(inode
, 3);
4351 if (IS_ERR(handle
)) {
4352 ext4_orphan_del(NULL
, inode
);
4355 ext4_orphan_del(handle
, inode
);
4357 ext4_journal_stop(handle
);
4363 if (attr
->ia_valid
& ATTR_SIZE
) {
4364 if (attr
->ia_size
!= inode
->i_size
) {
4365 loff_t oldsize
= inode
->i_size
;
4367 i_size_write(inode
, attr
->ia_size
);
4369 * Blocks are going to be removed from the inode. Wait
4370 * for dio in flight. Temporarily disable
4371 * dioread_nolock to prevent livelock.
4374 if (!ext4_should_journal_data(inode
)) {
4375 ext4_inode_block_unlocked_dio(inode
);
4376 inode_dio_wait(inode
);
4377 ext4_inode_resume_unlocked_dio(inode
);
4379 ext4_wait_for_tail_page_commit(inode
);
4382 * Truncate pagecache after we've waited for commit
4383 * in data=journal mode to make pages freeable.
4385 truncate_pagecache(inode
, oldsize
, inode
->i_size
);
4387 ext4_truncate(inode
);
4391 setattr_copy(inode
, attr
);
4392 mark_inode_dirty(inode
);
4396 * If the call to ext4_truncate failed to get a transaction handle at
4397 * all, we need to clean up the in-core orphan list manually.
4399 if (orphan
&& inode
->i_nlink
)
4400 ext4_orphan_del(NULL
, inode
);
4402 if (!rc
&& (ia_valid
& ATTR_MODE
))
4403 rc
= ext4_acl_chmod(inode
);
4406 ext4_std_error(inode
->i_sb
, error
);
4412 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4415 struct inode
*inode
;
4416 unsigned long delalloc_blocks
;
4418 inode
= dentry
->d_inode
;
4419 generic_fillattr(inode
, stat
);
4422 * We can't update i_blocks if the block allocation is delayed
4423 * otherwise in the case of system crash before the real block
4424 * allocation is done, we will have i_blocks inconsistent with
4425 * on-disk file blocks.
4426 * We always keep i_blocks updated together with real
4427 * allocation. But to not confuse with user, stat
4428 * will return the blocks that include the delayed allocation
4429 * blocks for this file.
4431 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
4432 EXT4_I(inode
)->i_reserved_data_blocks
);
4434 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4438 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4440 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4441 return ext4_ind_trans_blocks(inode
, nrblocks
, chunk
);
4442 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4446 * Account for index blocks, block groups bitmaps and block group
4447 * descriptor blocks if modify datablocks and index blocks
4448 * worse case, the indexs blocks spread over different block groups
4450 * If datablocks are discontiguous, they are possible to spread over
4451 * different block groups too. If they are contiguous, with flexbg,
4452 * they could still across block group boundary.
4454 * Also account for superblock, inode, quota and xattr blocks
4456 static int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4458 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4464 * How many index blocks need to touch to modify nrblocks?
4465 * The "Chunk" flag indicating whether the nrblocks is
4466 * physically contiguous on disk
4468 * For Direct IO and fallocate, they calls get_block to allocate
4469 * one single extent at a time, so they could set the "Chunk" flag
4471 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4476 * Now let's see how many group bitmaps and group descriptors need
4486 if (groups
> ngroups
)
4488 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4489 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4491 /* bitmaps and block group descriptor blocks */
4492 ret
+= groups
+ gdpblocks
;
4494 /* Blocks for super block, inode, quota and xattr blocks */
4495 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4501 * Calculate the total number of credits to reserve to fit
4502 * the modification of a single pages into a single transaction,
4503 * which may include multiple chunks of block allocations.
4505 * This could be called via ext4_write_begin()
4507 * We need to consider the worse case, when
4508 * one new block per extent.
4510 int ext4_writepage_trans_blocks(struct inode
*inode
)
4512 int bpp
= ext4_journal_blocks_per_page(inode
);
4515 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4517 /* Account for data blocks for journalled mode */
4518 if (ext4_should_journal_data(inode
))
4524 * Calculate the journal credits for a chunk of data modification.
4526 * This is called from DIO, fallocate or whoever calling
4527 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4529 * journal buffers for data blocks are not included here, as DIO
4530 * and fallocate do no need to journal data buffers.
4532 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4534 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4538 * The caller must have previously called ext4_reserve_inode_write().
4539 * Give this, we know that the caller already has write access to iloc->bh.
4541 int ext4_mark_iloc_dirty(handle_t
*handle
,
4542 struct inode
*inode
, struct ext4_iloc
*iloc
)
4546 if (IS_I_VERSION(inode
))
4547 inode_inc_iversion(inode
);
4549 /* the do_update_inode consumes one bh->b_count */
4552 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4553 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4559 * On success, We end up with an outstanding reference count against
4560 * iloc->bh. This _must_ be cleaned up later.
4564 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4565 struct ext4_iloc
*iloc
)
4569 err
= ext4_get_inode_loc(inode
, iloc
);
4571 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4572 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4578 ext4_std_error(inode
->i_sb
, err
);
4583 * Expand an inode by new_extra_isize bytes.
4584 * Returns 0 on success or negative error number on failure.
4586 static int ext4_expand_extra_isize(struct inode
*inode
,
4587 unsigned int new_extra_isize
,
4588 struct ext4_iloc iloc
,
4591 struct ext4_inode
*raw_inode
;
4592 struct ext4_xattr_ibody_header
*header
;
4594 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4597 raw_inode
= ext4_raw_inode(&iloc
);
4599 header
= IHDR(inode
, raw_inode
);
4601 /* No extended attributes present */
4602 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
4603 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4604 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4606 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4610 /* try to expand with EAs present */
4611 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4616 * What we do here is to mark the in-core inode as clean with respect to inode
4617 * dirtiness (it may still be data-dirty).
4618 * This means that the in-core inode may be reaped by prune_icache
4619 * without having to perform any I/O. This is a very good thing,
4620 * because *any* task may call prune_icache - even ones which
4621 * have a transaction open against a different journal.
4623 * Is this cheating? Not really. Sure, we haven't written the
4624 * inode out, but prune_icache isn't a user-visible syncing function.
4625 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4626 * we start and wait on commits.
4628 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4630 struct ext4_iloc iloc
;
4631 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4632 static unsigned int mnt_count
;
4636 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
4637 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4638 if (ext4_handle_valid(handle
) &&
4639 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4640 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
4642 * We need extra buffer credits since we may write into EA block
4643 * with this same handle. If journal_extend fails, then it will
4644 * only result in a minor loss of functionality for that inode.
4645 * If this is felt to be critical, then e2fsck should be run to
4646 * force a large enough s_min_extra_isize.
4648 if ((jbd2_journal_extend(handle
,
4649 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4650 ret
= ext4_expand_extra_isize(inode
,
4651 sbi
->s_want_extra_isize
,
4654 ext4_set_inode_state(inode
,
4655 EXT4_STATE_NO_EXPAND
);
4657 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4658 ext4_warning(inode
->i_sb
,
4659 "Unable to expand inode %lu. Delete"
4660 " some EAs or run e2fsck.",
4663 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4669 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4674 * ext4_dirty_inode() is called from __mark_inode_dirty()
4676 * We're really interested in the case where a file is being extended.
4677 * i_size has been changed by generic_commit_write() and we thus need
4678 * to include the updated inode in the current transaction.
4680 * Also, dquot_alloc_block() will always dirty the inode when blocks
4681 * are allocated to the file.
4683 * If the inode is marked synchronous, we don't honour that here - doing
4684 * so would cause a commit on atime updates, which we don't bother doing.
4685 * We handle synchronous inodes at the highest possible level.
4687 void ext4_dirty_inode(struct inode
*inode
, int flags
)
4691 handle
= ext4_journal_start(inode
, 2);
4695 ext4_mark_inode_dirty(handle
, inode
);
4697 ext4_journal_stop(handle
);
4704 * Bind an inode's backing buffer_head into this transaction, to prevent
4705 * it from being flushed to disk early. Unlike
4706 * ext4_reserve_inode_write, this leaves behind no bh reference and
4707 * returns no iloc structure, so the caller needs to repeat the iloc
4708 * lookup to mark the inode dirty later.
4710 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4712 struct ext4_iloc iloc
;
4716 err
= ext4_get_inode_loc(inode
, &iloc
);
4718 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4719 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4721 err
= ext4_handle_dirty_metadata(handle
,
4727 ext4_std_error(inode
->i_sb
, err
);
4732 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4739 * We have to be very careful here: changing a data block's
4740 * journaling status dynamically is dangerous. If we write a
4741 * data block to the journal, change the status and then delete
4742 * that block, we risk forgetting to revoke the old log record
4743 * from the journal and so a subsequent replay can corrupt data.
4744 * So, first we make sure that the journal is empty and that
4745 * nobody is changing anything.
4748 journal
= EXT4_JOURNAL(inode
);
4751 if (is_journal_aborted(journal
))
4753 /* We have to allocate physical blocks for delalloc blocks
4754 * before flushing journal. otherwise delalloc blocks can not
4755 * be allocated any more. even more truncate on delalloc blocks
4756 * could trigger BUG by flushing delalloc blocks in journal.
4757 * There is no delalloc block in non-journal data mode.
4759 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
4760 err
= ext4_alloc_da_blocks(inode
);
4765 /* Wait for all existing dio workers */
4766 ext4_inode_block_unlocked_dio(inode
);
4767 inode_dio_wait(inode
);
4769 jbd2_journal_lock_updates(journal
);
4772 * OK, there are no updates running now, and all cached data is
4773 * synced to disk. We are now in a completely consistent state
4774 * which doesn't have anything in the journal, and we know that
4775 * no filesystem updates are running, so it is safe to modify
4776 * the inode's in-core data-journaling state flag now.
4780 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4782 jbd2_journal_flush(journal
);
4783 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4785 ext4_set_aops(inode
);
4787 jbd2_journal_unlock_updates(journal
);
4788 ext4_inode_resume_unlocked_dio(inode
);
4790 /* Finally we can mark the inode as dirty. */
4792 handle
= ext4_journal_start(inode
, 1);
4794 return PTR_ERR(handle
);
4796 err
= ext4_mark_inode_dirty(handle
, inode
);
4797 ext4_handle_sync(handle
);
4798 ext4_journal_stop(handle
);
4799 ext4_std_error(inode
->i_sb
, err
);
4804 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4806 return !buffer_mapped(bh
);
4809 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
4811 struct page
*page
= vmf
->page
;
4815 struct file
*file
= vma
->vm_file
;
4816 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4817 struct address_space
*mapping
= inode
->i_mapping
;
4819 get_block_t
*get_block
;
4822 sb_start_pagefault(inode
->i_sb
);
4823 file_update_time(vma
->vm_file
);
4824 /* Delalloc case is easy... */
4825 if (test_opt(inode
->i_sb
, DELALLOC
) &&
4826 !ext4_should_journal_data(inode
) &&
4827 !ext4_nonda_switch(inode
->i_sb
)) {
4829 ret
= __block_page_mkwrite(vma
, vmf
,
4830 ext4_da_get_block_prep
);
4831 } while (ret
== -ENOSPC
&&
4832 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
4837 size
= i_size_read(inode
);
4838 /* Page got truncated from under us? */
4839 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
4841 ret
= VM_FAULT_NOPAGE
;
4845 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4846 len
= size
& ~PAGE_CACHE_MASK
;
4848 len
= PAGE_CACHE_SIZE
;
4850 * Return if we have all the buffers mapped. This avoids the need to do
4851 * journal_start/journal_stop which can block and take a long time
4853 if (page_has_buffers(page
)) {
4854 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
4856 ext4_bh_unmapped
)) {
4857 /* Wait so that we don't change page under IO */
4858 wait_on_page_writeback(page
);
4859 ret
= VM_FAULT_LOCKED
;
4864 /* OK, we need to fill the hole... */
4865 if (ext4_should_dioread_nolock(inode
))
4866 get_block
= ext4_get_block_write
;
4868 get_block
= ext4_get_block
;
4870 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
4871 if (IS_ERR(handle
)) {
4872 ret
= VM_FAULT_SIGBUS
;
4875 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
4876 if (!ret
&& ext4_should_journal_data(inode
)) {
4877 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
4878 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
4880 ret
= VM_FAULT_SIGBUS
;
4881 ext4_journal_stop(handle
);
4884 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
4886 ext4_journal_stop(handle
);
4887 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
4890 ret
= block_page_mkwrite_return(ret
);
4892 sb_end_pagefault(inode
->i_sb
);