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 int journal_data
= ext4_should_journal_data(inode
);
1358 sector_t pblock
= 0, cur_logical
= 0;
1359 struct ext4_io_submit io_submit
;
1361 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1362 memset(&io_submit
, 0, sizeof(io_submit
));
1364 * We need to start from the first_page to the next_page - 1
1365 * to make sure we also write the mapped dirty buffer_heads.
1366 * If we look at mpd->b_blocknr we would only be looking
1367 * at the currently mapped buffer_heads.
1369 index
= mpd
->first_page
;
1370 end
= mpd
->next_page
- 1;
1372 pagevec_init(&pvec
, 0);
1373 while (index
<= end
) {
1374 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1377 for (i
= 0; i
< nr_pages
; i
++) {
1378 int commit_write
= 0, skip_page
= 0;
1379 struct page
*page
= pvec
.pages
[i
];
1381 index
= page
->index
;
1385 if (index
== size
>> PAGE_CACHE_SHIFT
)
1386 len
= size
& ~PAGE_CACHE_MASK
;
1388 len
= PAGE_CACHE_SIZE
;
1390 cur_logical
= index
<< (PAGE_CACHE_SHIFT
-
1392 pblock
= map
->m_pblk
+ (cur_logical
-
1397 BUG_ON(!PageLocked(page
));
1398 BUG_ON(PageWriteback(page
));
1401 * If the page does not have buffers (for
1402 * whatever reason), try to create them using
1403 * __block_write_begin. If this fails,
1404 * skip the page and move on.
1406 if (!page_has_buffers(page
)) {
1407 if (__block_write_begin(page
, 0, len
,
1408 noalloc_get_block_write
)) {
1416 bh
= page_bufs
= page_buffers(page
);
1421 if (map
&& (cur_logical
>= map
->m_lblk
) &&
1422 (cur_logical
<= (map
->m_lblk
+
1423 (map
->m_len
- 1)))) {
1424 if (buffer_delay(bh
)) {
1425 clear_buffer_delay(bh
);
1426 bh
->b_blocknr
= pblock
;
1428 if (buffer_unwritten(bh
) ||
1430 BUG_ON(bh
->b_blocknr
!= pblock
);
1431 if (map
->m_flags
& EXT4_MAP_UNINIT
)
1432 set_buffer_uninit(bh
);
1433 clear_buffer_unwritten(bh
);
1437 * skip page if block allocation undone and
1440 if (ext4_bh_delay_or_unwritten(NULL
, bh
))
1442 bh
= bh
->b_this_page
;
1443 block_start
+= bh
->b_size
;
1446 } while (bh
!= page_bufs
);
1452 /* mark the buffer_heads as dirty & uptodate */
1453 block_commit_write(page
, 0, len
);
1455 clear_page_dirty_for_io(page
);
1457 * Delalloc doesn't support data journalling,
1458 * but eventually maybe we'll lift this
1461 if (unlikely(journal_data
&& PageChecked(page
)))
1462 err
= __ext4_journalled_writepage(page
, len
);
1464 err
= ext4_bio_write_page(&io_submit
, page
,
1467 mpd
->pages_written
++;
1469 * In error case, we have to continue because
1470 * remaining pages are still locked
1475 pagevec_release(&pvec
);
1477 ext4_io_submit(&io_submit
);
1481 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
)
1485 struct pagevec pvec
;
1486 struct inode
*inode
= mpd
->inode
;
1487 struct address_space
*mapping
= inode
->i_mapping
;
1488 ext4_lblk_t start
, last
;
1490 index
= mpd
->first_page
;
1491 end
= mpd
->next_page
- 1;
1493 start
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1494 last
= end
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
1495 ext4_es_remove_extent(inode
, start
, last
- start
+ 1);
1497 pagevec_init(&pvec
, 0);
1498 while (index
<= end
) {
1499 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1502 for (i
= 0; i
< nr_pages
; i
++) {
1503 struct page
*page
= pvec
.pages
[i
];
1504 if (page
->index
> end
)
1506 BUG_ON(!PageLocked(page
));
1507 BUG_ON(PageWriteback(page
));
1508 block_invalidatepage(page
, 0);
1509 ClearPageUptodate(page
);
1512 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1513 pagevec_release(&pvec
);
1518 static void ext4_print_free_blocks(struct inode
*inode
)
1520 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1521 struct super_block
*sb
= inode
->i_sb
;
1523 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1524 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1525 ext4_count_free_clusters(inode
->i_sb
)));
1526 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1527 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1528 (long long) EXT4_C2B(EXT4_SB(inode
->i_sb
),
1529 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1530 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1531 (long long) EXT4_C2B(EXT4_SB(inode
->i_sb
),
1532 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1533 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1534 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1535 EXT4_I(inode
)->i_reserved_data_blocks
);
1536 ext4_msg(sb
, KERN_CRIT
, "i_reserved_meta_blocks=%u",
1537 EXT4_I(inode
)->i_reserved_meta_blocks
);
1542 * mpage_da_map_and_submit - go through given space, map them
1543 * if necessary, and then submit them for I/O
1545 * @mpd - bh describing space
1547 * The function skips space we know is already mapped to disk blocks.
1550 static void mpage_da_map_and_submit(struct mpage_da_data
*mpd
)
1552 int err
, blks
, get_blocks_flags
;
1553 struct ext4_map_blocks map
, *mapp
= NULL
;
1554 sector_t next
= mpd
->b_blocknr
;
1555 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1556 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
1557 handle_t
*handle
= NULL
;
1560 * If the blocks are mapped already, or we couldn't accumulate
1561 * any blocks, then proceed immediately to the submission stage.
1563 if ((mpd
->b_size
== 0) ||
1564 ((mpd
->b_state
& (1 << BH_Mapped
)) &&
1565 !(mpd
->b_state
& (1 << BH_Delay
)) &&
1566 !(mpd
->b_state
& (1 << BH_Unwritten
))))
1569 handle
= ext4_journal_current_handle();
1573 * Call ext4_map_blocks() to allocate any delayed allocation
1574 * blocks, or to convert an uninitialized extent to be
1575 * initialized (in the case where we have written into
1576 * one or more preallocated blocks).
1578 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1579 * indicate that we are on the delayed allocation path. This
1580 * affects functions in many different parts of the allocation
1581 * call path. This flag exists primarily because we don't
1582 * want to change *many* call functions, so ext4_map_blocks()
1583 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1584 * inode's allocation semaphore is taken.
1586 * If the blocks in questions were delalloc blocks, set
1587 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1588 * variables are updated after the blocks have been allocated.
1591 map
.m_len
= max_blocks
;
1592 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
1593 if (ext4_should_dioread_nolock(mpd
->inode
))
1594 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
1595 if (mpd
->b_state
& (1 << BH_Delay
))
1596 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
1598 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
1600 struct super_block
*sb
= mpd
->inode
->i_sb
;
1604 * If get block returns EAGAIN or ENOSPC and there
1605 * appears to be free blocks we will just let
1606 * mpage_da_submit_io() unlock all of the pages.
1611 if (err
== -ENOSPC
&& ext4_count_free_clusters(sb
)) {
1617 * get block failure will cause us to loop in
1618 * writepages, because a_ops->writepage won't be able
1619 * to make progress. The page will be redirtied by
1620 * writepage and writepages will again try to write
1623 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
1624 ext4_msg(sb
, KERN_CRIT
,
1625 "delayed block allocation failed for inode %lu "
1626 "at logical offset %llu with max blocks %zd "
1627 "with error %d", mpd
->inode
->i_ino
,
1628 (unsigned long long) next
,
1629 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1630 ext4_msg(sb
, KERN_CRIT
,
1631 "This should not happen!! Data will be lost\n");
1633 ext4_print_free_blocks(mpd
->inode
);
1635 /* invalidate all the pages */
1636 ext4_da_block_invalidatepages(mpd
);
1638 /* Mark this page range as having been completed */
1645 if (map
.m_flags
& EXT4_MAP_NEW
) {
1646 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
1649 for (i
= 0; i
< map
.m_len
; i
++)
1650 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
1654 * Update on-disk size along with block allocation.
1656 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
1657 if (disksize
> i_size_read(mpd
->inode
))
1658 disksize
= i_size_read(mpd
->inode
);
1659 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
1660 ext4_update_i_disksize(mpd
->inode
, disksize
);
1661 err
= ext4_mark_inode_dirty(handle
, mpd
->inode
);
1663 ext4_error(mpd
->inode
->i_sb
,
1664 "Failed to mark inode %lu dirty",
1669 mpage_da_submit_io(mpd
, mapp
);
1673 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1674 (1 << BH_Delay) | (1 << BH_Unwritten))
1677 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1679 * @mpd->lbh - extent of blocks
1680 * @logical - logical number of the block in the file
1681 * @bh - bh of the block (used to access block's state)
1683 * the function is used to collect contig. blocks in same state
1685 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1686 sector_t logical
, size_t b_size
,
1687 unsigned long b_state
)
1690 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1693 * XXX Don't go larger than mballoc is willing to allocate
1694 * This is a stopgap solution. We eventually need to fold
1695 * mpage_da_submit_io() into this function and then call
1696 * ext4_map_blocks() multiple times in a loop
1698 if (nrblocks
>= 8*1024*1024/mpd
->inode
->i_sb
->s_blocksize
)
1701 /* check if thereserved journal credits might overflow */
1702 if (!(ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
))) {
1703 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1705 * With non-extent format we are limited by the journal
1706 * credit available. Total credit needed to insert
1707 * nrblocks contiguous blocks is dependent on the
1708 * nrblocks. So limit nrblocks.
1711 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
1712 EXT4_MAX_TRANS_DATA
) {
1714 * Adding the new buffer_head would make it cross the
1715 * allowed limit for which we have journal credit
1716 * reserved. So limit the new bh->b_size
1718 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
1719 mpd
->inode
->i_blkbits
;
1720 /* we will do mpage_da_submit_io in the next loop */
1724 * First block in the extent
1726 if (mpd
->b_size
== 0) {
1727 mpd
->b_blocknr
= logical
;
1728 mpd
->b_size
= b_size
;
1729 mpd
->b_state
= b_state
& BH_FLAGS
;
1733 next
= mpd
->b_blocknr
+ nrblocks
;
1735 * Can we merge the block to our big extent?
1737 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
1738 mpd
->b_size
+= b_size
;
1744 * We couldn't merge the block to our extent, so we
1745 * need to flush current extent and start new one
1747 mpage_da_map_and_submit(mpd
);
1751 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1753 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1757 * This function is grabs code from the very beginning of
1758 * ext4_map_blocks, but assumes that the caller is from delayed write
1759 * time. This function looks up the requested blocks and sets the
1760 * buffer delay bit under the protection of i_data_sem.
1762 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1763 struct ext4_map_blocks
*map
,
1764 struct buffer_head
*bh
)
1767 sector_t invalid_block
= ~((sector_t
) 0xffff);
1769 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1773 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1774 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1775 (unsigned long) map
->m_lblk
);
1777 * Try to see if we can get the block without requesting a new
1778 * file system block.
1780 down_read((&EXT4_I(inode
)->i_data_sem
));
1781 if (ext4_has_inline_data(inode
)) {
1783 * We will soon create blocks for this page, and let
1784 * us pretend as if the blocks aren't allocated yet.
1785 * In case of clusters, we have to handle the work
1786 * of mapping from cluster so that the reserved space
1787 * is calculated properly.
1789 if ((EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) &&
1790 ext4_find_delalloc_cluster(inode
, map
->m_lblk
))
1791 map
->m_flags
|= EXT4_MAP_FROM_CLUSTER
;
1793 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1794 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1796 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1800 * XXX: __block_prepare_write() unmaps passed block,
1803 /* If the block was allocated from previously allocated cluster,
1804 * then we dont need to reserve it again. */
1805 if (!(map
->m_flags
& EXT4_MAP_FROM_CLUSTER
)) {
1806 retval
= ext4_da_reserve_space(inode
, iblock
);
1808 /* not enough space to reserve */
1812 retval
= ext4_es_insert_extent(inode
, map
->m_lblk
, map
->m_len
);
1816 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1817 * and it should not appear on the bh->b_state.
1819 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
1821 map_bh(bh
, inode
->i_sb
, invalid_block
);
1823 set_buffer_delay(bh
);
1827 up_read((&EXT4_I(inode
)->i_data_sem
));
1833 * This is a special get_blocks_t callback which is used by
1834 * ext4_da_write_begin(). It will either return mapped block or
1835 * reserve space for a single block.
1837 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1838 * We also have b_blocknr = -1 and b_bdev initialized properly
1840 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1841 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1842 * initialized properly.
1844 int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1845 struct buffer_head
*bh
, int create
)
1847 struct ext4_map_blocks map
;
1850 BUG_ON(create
== 0);
1851 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1853 map
.m_lblk
= iblock
;
1857 * first, we need to know whether the block is allocated already
1858 * preallocated blocks are unmapped but should treated
1859 * the same as allocated blocks.
1861 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1865 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1866 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1868 if (buffer_unwritten(bh
)) {
1869 /* A delayed write to unwritten bh should be marked
1870 * new and mapped. Mapped ensures that we don't do
1871 * get_block multiple times when we write to the same
1872 * offset and new ensures that we do proper zero out
1873 * for partial write.
1876 set_buffer_mapped(bh
);
1882 * This function is used as a standard get_block_t calback function when there
1883 * is no desire to allocate any blocks. It is used as a callback function for
1884 * block_write_begin(). These functions should only try to map a single block
1887 * Since this function doesn't do block allocations even if the caller
1888 * requests it by passing in create=1, it is critically important that
1889 * any caller checks to make sure that any buffer heads are returned
1890 * by this function are either all already mapped or marked for
1891 * delayed allocation before calling ext4_bio_write_page(). Otherwise,
1892 * b_blocknr could be left unitialized, and the page write functions will
1893 * be taken by surprise.
1895 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
1896 struct buffer_head
*bh_result
, int create
)
1898 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
1899 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
1902 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1908 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1914 static int __ext4_journalled_writepage(struct page
*page
,
1917 struct address_space
*mapping
= page
->mapping
;
1918 struct inode
*inode
= mapping
->host
;
1919 struct buffer_head
*page_bufs
= NULL
;
1920 handle_t
*handle
= NULL
;
1921 int ret
= 0, err
= 0;
1922 int inline_data
= ext4_has_inline_data(inode
);
1923 struct buffer_head
*inode_bh
= NULL
;
1925 ClearPageChecked(page
);
1928 BUG_ON(page
->index
!= 0);
1929 BUG_ON(len
> ext4_get_max_inline_size(inode
));
1930 inode_bh
= ext4_journalled_write_inline_data(inode
, len
, page
);
1931 if (inode_bh
== NULL
)
1934 page_bufs
= page_buffers(page
);
1939 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1942 /* As soon as we unlock the page, it can go away, but we have
1943 * references to buffers so we are safe */
1946 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
1947 if (IS_ERR(handle
)) {
1948 ret
= PTR_ERR(handle
);
1952 BUG_ON(!ext4_handle_valid(handle
));
1955 ret
= ext4_journal_get_write_access(handle
, inode_bh
);
1957 err
= ext4_handle_dirty_metadata(handle
, inode
, inode_bh
);
1960 ret
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1961 do_journal_get_write_access
);
1963 err
= ext4_walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1968 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1969 err
= ext4_journal_stop(handle
);
1973 if (!ext4_has_inline_data(inode
))
1974 ext4_walk_page_buffers(handle
, page_bufs
, 0, len
,
1976 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1983 * Note that we don't need to start a transaction unless we're journaling data
1984 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1985 * need to file the inode to the transaction's list in ordered mode because if
1986 * we are writing back data added by write(), the inode is already there and if
1987 * we are writing back data modified via mmap(), no one guarantees in which
1988 * transaction the data will hit the disk. In case we are journaling data, we
1989 * cannot start transaction directly because transaction start ranks above page
1990 * lock so we have to do some magic.
1992 * This function can get called via...
1993 * - ext4_da_writepages after taking page lock (have journal handle)
1994 * - journal_submit_inode_data_buffers (no journal handle)
1995 * - shrink_page_list via the kswapd/direct reclaim (no journal handle)
1996 * - grab_page_cache when doing write_begin (have journal handle)
1998 * We don't do any block allocation in this function. If we have page with
1999 * multiple blocks we need to write those buffer_heads that are mapped. This
2000 * is important for mmaped based write. So if we do with blocksize 1K
2001 * truncate(f, 1024);
2002 * a = mmap(f, 0, 4096);
2004 * truncate(f, 4096);
2005 * we have in the page first buffer_head mapped via page_mkwrite call back
2006 * but other buffer_heads would be unmapped but dirty (dirty done via the
2007 * do_wp_page). So writepage should write the first block. If we modify
2008 * the mmap area beyond 1024 we will again get a page_fault and the
2009 * page_mkwrite callback will do the block allocation and mark the
2010 * buffer_heads mapped.
2012 * We redirty the page if we have any buffer_heads that is either delay or
2013 * unwritten in the page.
2015 * We can get recursively called as show below.
2017 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
2020 * But since we don't do any block allocation we should not deadlock.
2021 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2023 static int ext4_writepage(struct page
*page
,
2024 struct writeback_control
*wbc
)
2026 int ret
= 0, commit_write
= 0;
2029 struct buffer_head
*page_bufs
= NULL
;
2030 struct inode
*inode
= page
->mapping
->host
;
2031 struct ext4_io_submit io_submit
;
2033 trace_ext4_writepage(page
);
2034 size
= i_size_read(inode
);
2035 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2036 len
= size
& ~PAGE_CACHE_MASK
;
2038 len
= PAGE_CACHE_SIZE
;
2041 * If the page does not have buffers (for whatever reason),
2042 * try to create them using __block_write_begin. If this
2043 * fails, redirty the page and move on.
2045 if (!page_has_buffers(page
)) {
2046 if (__block_write_begin(page
, 0, len
,
2047 noalloc_get_block_write
)) {
2049 redirty_page_for_writepage(wbc
, page
);
2055 page_bufs
= page_buffers(page
);
2056 if (ext4_walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2057 ext4_bh_delay_or_unwritten
)) {
2059 * We don't want to do block allocation, so redirty
2060 * the page and return. We may reach here when we do
2061 * a journal commit via journal_submit_inode_data_buffers.
2062 * We can also reach here via shrink_page_list but it
2063 * should never be for direct reclaim so warn if that
2066 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
)) ==
2071 /* now mark the buffer_heads as dirty and uptodate */
2072 block_commit_write(page
, 0, len
);
2074 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2076 * It's mmapped pagecache. Add buffers and journal it. There
2077 * doesn't seem much point in redirtying the page here.
2079 return __ext4_journalled_writepage(page
, len
);
2081 memset(&io_submit
, 0, sizeof(io_submit
));
2082 ret
= ext4_bio_write_page(&io_submit
, page
, len
, wbc
);
2083 ext4_io_submit(&io_submit
);
2088 * This is called via ext4_da_writepages() to
2089 * calculate the total number of credits to reserve to fit
2090 * a single extent allocation into a single transaction,
2091 * ext4_da_writpeages() will loop calling this before
2092 * the block allocation.
2095 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2097 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2100 * With non-extent format the journal credit needed to
2101 * insert nrblocks contiguous block is dependent on
2102 * number of contiguous block. So we will limit
2103 * number of contiguous block to a sane value
2105 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
2106 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2107 max_blocks
= EXT4_MAX_TRANS_DATA
;
2109 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2113 * write_cache_pages_da - walk the list of dirty pages of the given
2114 * address space and accumulate pages that need writing, and call
2115 * mpage_da_map_and_submit to map a single contiguous memory region
2116 * and then write them.
2118 static int write_cache_pages_da(handle_t
*handle
,
2119 struct address_space
*mapping
,
2120 struct writeback_control
*wbc
,
2121 struct mpage_da_data
*mpd
,
2122 pgoff_t
*done_index
)
2124 struct buffer_head
*bh
, *head
;
2125 struct inode
*inode
= mapping
->host
;
2126 struct pagevec pvec
;
2127 unsigned int nr_pages
;
2130 long nr_to_write
= wbc
->nr_to_write
;
2131 int i
, tag
, ret
= 0;
2133 memset(mpd
, 0, sizeof(struct mpage_da_data
));
2136 pagevec_init(&pvec
, 0);
2137 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2138 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2140 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2141 tag
= PAGECACHE_TAG_TOWRITE
;
2143 tag
= PAGECACHE_TAG_DIRTY
;
2145 *done_index
= index
;
2146 while (index
<= end
) {
2147 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2148 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2152 for (i
= 0; i
< nr_pages
; i
++) {
2153 struct page
*page
= pvec
.pages
[i
];
2156 * At this point, the page may be truncated or
2157 * invalidated (changing page->mapping to NULL), or
2158 * even swizzled back from swapper_space to tmpfs file
2159 * mapping. However, page->index will not change
2160 * because we have a reference on the page.
2162 if (page
->index
> end
)
2165 *done_index
= page
->index
+ 1;
2168 * If we can't merge this page, and we have
2169 * accumulated an contiguous region, write it
2171 if ((mpd
->next_page
!= page
->index
) &&
2172 (mpd
->next_page
!= mpd
->first_page
)) {
2173 mpage_da_map_and_submit(mpd
);
2174 goto ret_extent_tail
;
2180 * If the page is no longer dirty, or its
2181 * mapping no longer corresponds to inode we
2182 * are writing (which means it has been
2183 * truncated or invalidated), or the page is
2184 * already under writeback and we are not
2185 * doing a data integrity writeback, skip the page
2187 if (!PageDirty(page
) ||
2188 (PageWriteback(page
) &&
2189 (wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2190 unlikely(page
->mapping
!= mapping
)) {
2195 wait_on_page_writeback(page
);
2196 BUG_ON(PageWriteback(page
));
2199 * If we have inline data and arrive here, it means that
2200 * we will soon create the block for the 1st page, so
2201 * we'd better clear the inline data here.
2203 if (ext4_has_inline_data(inode
)) {
2204 BUG_ON(ext4_test_inode_state(inode
,
2205 EXT4_STATE_MAY_INLINE_DATA
));
2206 ext4_destroy_inline_data(handle
, inode
);
2209 if (mpd
->next_page
!= page
->index
)
2210 mpd
->first_page
= page
->index
;
2211 mpd
->next_page
= page
->index
+ 1;
2212 logical
= (sector_t
) page
->index
<<
2213 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2215 if (!page_has_buffers(page
)) {
2216 mpage_add_bh_to_extent(mpd
, logical
,
2218 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2220 goto ret_extent_tail
;
2223 * Page with regular buffer heads,
2224 * just add all dirty ones
2226 head
= page_buffers(page
);
2229 BUG_ON(buffer_locked(bh
));
2231 * We need to try to allocate
2232 * unmapped blocks in the same page.
2233 * Otherwise we won't make progress
2234 * with the page in ext4_writepage
2236 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2237 mpage_add_bh_to_extent(mpd
, logical
,
2241 goto ret_extent_tail
;
2242 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2244 * mapped dirty buffer. We need
2245 * to update the b_state
2246 * because we look at b_state
2247 * in mpage_da_map_blocks. We
2248 * don't update b_size because
2249 * if we find an unmapped
2250 * buffer_head later we need to
2251 * use the b_state flag of that
2254 if (mpd
->b_size
== 0)
2255 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2258 } while ((bh
= bh
->b_this_page
) != head
);
2261 if (nr_to_write
> 0) {
2263 if (nr_to_write
== 0 &&
2264 wbc
->sync_mode
== WB_SYNC_NONE
)
2266 * We stop writing back only if we are
2267 * not doing integrity sync. In case of
2268 * integrity sync we have to keep going
2269 * because someone may be concurrently
2270 * dirtying pages, and we might have
2271 * synced a lot of newly appeared dirty
2272 * pages, but have not synced all of the
2278 pagevec_release(&pvec
);
2283 ret
= MPAGE_DA_EXTENT_TAIL
;
2285 pagevec_release(&pvec
);
2291 static int ext4_da_writepages(struct address_space
*mapping
,
2292 struct writeback_control
*wbc
)
2295 int range_whole
= 0;
2296 handle_t
*handle
= NULL
;
2297 struct mpage_da_data mpd
;
2298 struct inode
*inode
= mapping
->host
;
2299 int pages_written
= 0;
2300 unsigned int max_pages
;
2301 int range_cyclic
, cycled
= 1, io_done
= 0;
2302 int needed_blocks
, ret
= 0;
2303 long desired_nr_to_write
, nr_to_writebump
= 0;
2304 loff_t range_start
= wbc
->range_start
;
2305 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2306 pgoff_t done_index
= 0;
2308 struct blk_plug plug
;
2310 trace_ext4_da_writepages(inode
, wbc
);
2313 * No pages to write? This is mainly a kludge to avoid starting
2314 * a transaction for special inodes like journal inode on last iput()
2315 * because that could violate lock ordering on umount
2317 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2321 * If the filesystem has aborted, it is read-only, so return
2322 * right away instead of dumping stack traces later on that
2323 * will obscure the real source of the problem. We test
2324 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2325 * the latter could be true if the filesystem is mounted
2326 * read-only, and in that case, ext4_da_writepages should
2327 * *never* be called, so if that ever happens, we would want
2330 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2333 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2336 range_cyclic
= wbc
->range_cyclic
;
2337 if (wbc
->range_cyclic
) {
2338 index
= mapping
->writeback_index
;
2341 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2342 wbc
->range_end
= LLONG_MAX
;
2343 wbc
->range_cyclic
= 0;
2346 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2347 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2351 * This works around two forms of stupidity. The first is in
2352 * the writeback code, which caps the maximum number of pages
2353 * written to be 1024 pages. This is wrong on multiple
2354 * levels; different architectues have a different page size,
2355 * which changes the maximum amount of data which gets
2356 * written. Secondly, 4 megabytes is way too small. XFS
2357 * forces this value to be 16 megabytes by multiplying
2358 * nr_to_write parameter by four, and then relies on its
2359 * allocator to allocate larger extents to make them
2360 * contiguous. Unfortunately this brings us to the second
2361 * stupidity, which is that ext4's mballoc code only allocates
2362 * at most 2048 blocks. So we force contiguous writes up to
2363 * the number of dirty blocks in the inode, or
2364 * sbi->max_writeback_mb_bump whichever is smaller.
2366 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2367 if (!range_cyclic
&& range_whole
) {
2368 if (wbc
->nr_to_write
== LONG_MAX
)
2369 desired_nr_to_write
= wbc
->nr_to_write
;
2371 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2373 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2375 if (desired_nr_to_write
> max_pages
)
2376 desired_nr_to_write
= max_pages
;
2378 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2379 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2380 wbc
->nr_to_write
= desired_nr_to_write
;
2384 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2385 tag_pages_for_writeback(mapping
, index
, end
);
2387 blk_start_plug(&plug
);
2388 while (!ret
&& wbc
->nr_to_write
> 0) {
2391 * we insert one extent at a time. So we need
2392 * credit needed for single extent allocation.
2393 * journalled mode is currently not supported
2396 BUG_ON(ext4_should_journal_data(inode
));
2397 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2399 /* start a new transaction*/
2400 handle
= ext4_journal_start(inode
, needed_blocks
);
2401 if (IS_ERR(handle
)) {
2402 ret
= PTR_ERR(handle
);
2403 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2404 "%ld pages, ino %lu; err %d", __func__
,
2405 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2406 blk_finish_plug(&plug
);
2407 goto out_writepages
;
2411 * Now call write_cache_pages_da() to find the next
2412 * contiguous region of logical blocks that need
2413 * blocks to be allocated by ext4 and submit them.
2415 ret
= write_cache_pages_da(handle
, mapping
,
2416 wbc
, &mpd
, &done_index
);
2418 * If we have a contiguous extent of pages and we
2419 * haven't done the I/O yet, map the blocks and submit
2422 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2423 mpage_da_map_and_submit(&mpd
);
2424 ret
= MPAGE_DA_EXTENT_TAIL
;
2426 trace_ext4_da_write_pages(inode
, &mpd
);
2427 wbc
->nr_to_write
-= mpd
.pages_written
;
2429 ext4_journal_stop(handle
);
2431 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2432 /* commit the transaction which would
2433 * free blocks released in the transaction
2436 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2438 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2440 * Got one extent now try with rest of the pages.
2441 * If mpd.retval is set -EIO, journal is aborted.
2442 * So we don't need to write any more.
2444 pages_written
+= mpd
.pages_written
;
2447 } else if (wbc
->nr_to_write
)
2449 * There is no more writeout needed
2450 * or we requested for a noblocking writeout
2451 * and we found the device congested
2455 blk_finish_plug(&plug
);
2456 if (!io_done
&& !cycled
) {
2459 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2460 wbc
->range_end
= mapping
->writeback_index
- 1;
2465 wbc
->range_cyclic
= range_cyclic
;
2466 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2468 * set the writeback_index so that range_cyclic
2469 * mode will write it back later
2471 mapping
->writeback_index
= done_index
;
2474 wbc
->nr_to_write
-= nr_to_writebump
;
2475 wbc
->range_start
= range_start
;
2476 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2480 static int ext4_nonda_switch(struct super_block
*sb
)
2482 s64 free_blocks
, dirty_blocks
;
2483 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2486 * switch to non delalloc mode if we are running low
2487 * on free block. The free block accounting via percpu
2488 * counters can get slightly wrong with percpu_counter_batch getting
2489 * accumulated on each CPU without updating global counters
2490 * Delalloc need an accurate free block accounting. So switch
2491 * to non delalloc when we are near to error range.
2493 free_blocks
= EXT4_C2B(sbi
,
2494 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
));
2495 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2497 * Start pushing delalloc when 1/2 of free blocks are dirty.
2499 if (dirty_blocks
&& (free_blocks
< 2 * dirty_blocks
) &&
2500 !writeback_in_progress(sb
->s_bdi
) &&
2501 down_read_trylock(&sb
->s_umount
)) {
2502 writeback_inodes_sb(sb
, WB_REASON_FS_FREE_SPACE
);
2503 up_read(&sb
->s_umount
);
2506 if (2 * free_blocks
< 3 * dirty_blocks
||
2507 free_blocks
< (dirty_blocks
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2509 * free block count is less than 150% of dirty blocks
2510 * or free blocks is less than watermark
2517 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2518 loff_t pos
, unsigned len
, unsigned flags
,
2519 struct page
**pagep
, void **fsdata
)
2521 int ret
, retries
= 0;
2524 struct inode
*inode
= mapping
->host
;
2527 index
= pos
>> PAGE_CACHE_SHIFT
;
2529 if (ext4_nonda_switch(inode
->i_sb
)) {
2530 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2531 return ext4_write_begin(file
, mapping
, pos
,
2532 len
, flags
, pagep
, fsdata
);
2534 *fsdata
= (void *)0;
2535 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2537 if (ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
)) {
2538 ret
= ext4_da_write_inline_data_begin(mapping
, inode
,
2551 * With delayed allocation, we don't log the i_disksize update
2552 * if there is delayed block allocation. But we still need
2553 * to journalling the i_disksize update if writes to the end
2554 * of file which has an already mapped buffer.
2556 handle
= ext4_journal_start(inode
, 1);
2557 if (IS_ERR(handle
)) {
2558 ret
= PTR_ERR(handle
);
2561 /* We cannot recurse into the filesystem as the transaction is already
2563 flags
|= AOP_FLAG_NOFS
;
2565 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2567 ext4_journal_stop(handle
);
2573 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2576 ext4_journal_stop(handle
);
2577 page_cache_release(page
);
2579 * block_write_begin may have instantiated a few blocks
2580 * outside i_size. Trim these off again. Don't need
2581 * i_size_read because we hold i_mutex.
2583 if (pos
+ len
> inode
->i_size
)
2584 ext4_truncate_failed_write(inode
);
2587 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2594 * Check if we should update i_disksize
2595 * when write to the end of file but not require block allocation
2597 static int ext4_da_should_update_i_disksize(struct page
*page
,
2598 unsigned long offset
)
2600 struct buffer_head
*bh
;
2601 struct inode
*inode
= page
->mapping
->host
;
2605 bh
= page_buffers(page
);
2606 idx
= offset
>> inode
->i_blkbits
;
2608 for (i
= 0; i
< idx
; i
++)
2609 bh
= bh
->b_this_page
;
2611 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2616 static int ext4_da_write_end(struct file
*file
,
2617 struct address_space
*mapping
,
2618 loff_t pos
, unsigned len
, unsigned copied
,
2619 struct page
*page
, void *fsdata
)
2621 struct inode
*inode
= mapping
->host
;
2623 handle_t
*handle
= ext4_journal_current_handle();
2625 unsigned long start
, end
;
2626 int write_mode
= (int)(unsigned long)fsdata
;
2628 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2629 switch (ext4_inode_journal_mode(inode
)) {
2630 case EXT4_INODE_ORDERED_DATA_MODE
:
2631 return ext4_ordered_write_end(file
, mapping
, pos
,
2632 len
, copied
, page
, fsdata
);
2633 case EXT4_INODE_WRITEBACK_DATA_MODE
:
2634 return ext4_writeback_write_end(file
, mapping
, pos
,
2635 len
, copied
, page
, fsdata
);
2641 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2642 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2643 end
= start
+ copied
- 1;
2646 * generic_write_end() will run mark_inode_dirty() if i_size
2647 * changes. So let's piggyback the i_disksize mark_inode_dirty
2650 new_i_size
= pos
+ copied
;
2651 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2652 if (ext4_has_inline_data(inode
) ||
2653 ext4_da_should_update_i_disksize(page
, end
)) {
2654 down_write(&EXT4_I(inode
)->i_data_sem
);
2655 if (new_i_size
> EXT4_I(inode
)->i_disksize
)
2656 EXT4_I(inode
)->i_disksize
= new_i_size
;
2657 up_write(&EXT4_I(inode
)->i_data_sem
);
2658 /* We need to mark inode dirty even if
2659 * new_i_size is less that inode->i_size
2660 * bu greater than i_disksize.(hint delalloc)
2662 ext4_mark_inode_dirty(handle
, inode
);
2666 if (write_mode
!= CONVERT_INLINE_DATA
&&
2667 ext4_test_inode_state(inode
, EXT4_STATE_MAY_INLINE_DATA
) &&
2668 ext4_has_inline_data(inode
))
2669 ret2
= ext4_da_write_inline_data_end(inode
, pos
, len
, copied
,
2672 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2678 ret2
= ext4_journal_stop(handle
);
2682 return ret
? ret
: copied
;
2685 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2688 * Drop reserved blocks
2690 BUG_ON(!PageLocked(page
));
2691 if (!page_has_buffers(page
))
2694 ext4_da_page_release_reservation(page
, offset
);
2697 ext4_invalidatepage(page
, offset
);
2703 * Force all delayed allocation blocks to be allocated for a given inode.
2705 int ext4_alloc_da_blocks(struct inode
*inode
)
2707 trace_ext4_alloc_da_blocks(inode
);
2709 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2710 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2714 * We do something simple for now. The filemap_flush() will
2715 * also start triggering a write of the data blocks, which is
2716 * not strictly speaking necessary (and for users of
2717 * laptop_mode, not even desirable). However, to do otherwise
2718 * would require replicating code paths in:
2720 * ext4_da_writepages() ->
2721 * write_cache_pages() ---> (via passed in callback function)
2722 * __mpage_da_writepage() -->
2723 * mpage_add_bh_to_extent()
2724 * mpage_da_map_blocks()
2726 * The problem is that write_cache_pages(), located in
2727 * mm/page-writeback.c, marks pages clean in preparation for
2728 * doing I/O, which is not desirable if we're not planning on
2731 * We could call write_cache_pages(), and then redirty all of
2732 * the pages by calling redirty_page_for_writepage() but that
2733 * would be ugly in the extreme. So instead we would need to
2734 * replicate parts of the code in the above functions,
2735 * simplifying them because we wouldn't actually intend to
2736 * write out the pages, but rather only collect contiguous
2737 * logical block extents, call the multi-block allocator, and
2738 * then update the buffer heads with the block allocations.
2740 * For now, though, we'll cheat by calling filemap_flush(),
2741 * which will map the blocks, and start the I/O, but not
2742 * actually wait for the I/O to complete.
2744 return filemap_flush(inode
->i_mapping
);
2748 * bmap() is special. It gets used by applications such as lilo and by
2749 * the swapper to find the on-disk block of a specific piece of data.
2751 * Naturally, this is dangerous if the block concerned is still in the
2752 * journal. If somebody makes a swapfile on an ext4 data-journaling
2753 * filesystem and enables swap, then they may get a nasty shock when the
2754 * data getting swapped to that swapfile suddenly gets overwritten by
2755 * the original zero's written out previously to the journal and
2756 * awaiting writeback in the kernel's buffer cache.
2758 * So, if we see any bmap calls here on a modified, data-journaled file,
2759 * take extra steps to flush any blocks which might be in the cache.
2761 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2763 struct inode
*inode
= mapping
->host
;
2768 * We can get here for an inline file via the FIBMAP ioctl
2770 if (ext4_has_inline_data(inode
))
2773 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2774 test_opt(inode
->i_sb
, DELALLOC
)) {
2776 * With delalloc we want to sync the file
2777 * so that we can make sure we allocate
2780 filemap_write_and_wait(mapping
);
2783 if (EXT4_JOURNAL(inode
) &&
2784 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2786 * This is a REALLY heavyweight approach, but the use of
2787 * bmap on dirty files is expected to be extremely rare:
2788 * only if we run lilo or swapon on a freshly made file
2789 * do we expect this to happen.
2791 * (bmap requires CAP_SYS_RAWIO so this does not
2792 * represent an unprivileged user DOS attack --- we'd be
2793 * in trouble if mortal users could trigger this path at
2796 * NB. EXT4_STATE_JDATA is not set on files other than
2797 * regular files. If somebody wants to bmap a directory
2798 * or symlink and gets confused because the buffer
2799 * hasn't yet been flushed to disk, they deserve
2800 * everything they get.
2803 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2804 journal
= EXT4_JOURNAL(inode
);
2805 jbd2_journal_lock_updates(journal
);
2806 err
= jbd2_journal_flush(journal
);
2807 jbd2_journal_unlock_updates(journal
);
2813 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2816 static int ext4_readpage(struct file
*file
, struct page
*page
)
2819 struct inode
*inode
= page
->mapping
->host
;
2821 trace_ext4_readpage(page
);
2823 if (ext4_has_inline_data(inode
))
2824 ret
= ext4_readpage_inline(inode
, page
);
2827 return mpage_readpage(page
, ext4_get_block
);
2833 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2834 struct list_head
*pages
, unsigned nr_pages
)
2836 struct inode
*inode
= mapping
->host
;
2838 /* If the file has inline data, no need to do readpages. */
2839 if (ext4_has_inline_data(inode
))
2842 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2845 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2847 trace_ext4_invalidatepage(page
, offset
);
2849 /* No journalling happens on data buffers when this function is used */
2850 WARN_ON(page_has_buffers(page
) && buffer_jbd(page_buffers(page
)));
2852 block_invalidatepage(page
, offset
);
2855 static int __ext4_journalled_invalidatepage(struct page
*page
,
2856 unsigned long offset
)
2858 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2860 trace_ext4_journalled_invalidatepage(page
, offset
);
2863 * If it's a full truncate we just forget about the pending dirtying
2866 ClearPageChecked(page
);
2868 return jbd2_journal_invalidatepage(journal
, page
, offset
);
2871 /* Wrapper for aops... */
2872 static void ext4_journalled_invalidatepage(struct page
*page
,
2873 unsigned long offset
)
2875 WARN_ON(__ext4_journalled_invalidatepage(page
, offset
) < 0);
2878 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2880 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2882 trace_ext4_releasepage(page
);
2884 WARN_ON(PageChecked(page
));
2885 if (!page_has_buffers(page
))
2888 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2890 return try_to_free_buffers(page
);
2894 * ext4_get_block used when preparing for a DIO write or buffer write.
2895 * We allocate an uinitialized extent if blocks haven't been allocated.
2896 * The extent will be converted to initialized after the IO is complete.
2898 int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
2899 struct buffer_head
*bh_result
, int create
)
2901 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2902 inode
->i_ino
, create
);
2903 return _ext4_get_block(inode
, iblock
, bh_result
,
2904 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
2907 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
2908 struct buffer_head
*bh_result
, int create
)
2910 ext4_debug("ext4_get_block_write_nolock: inode %lu, create flag %d\n",
2911 inode
->i_ino
, create
);
2912 return _ext4_get_block(inode
, iblock
, bh_result
,
2913 EXT4_GET_BLOCKS_NO_LOCK
);
2916 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
2917 ssize_t size
, void *private, int ret
,
2920 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
2921 ext4_io_end_t
*io_end
= iocb
->private;
2923 /* if not async direct IO or dio with 0 bytes write, just return */
2924 if (!io_end
|| !size
)
2927 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2928 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2929 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
2932 iocb
->private = NULL
;
2934 /* if not aio dio with unwritten extents, just free io and return */
2935 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
2936 ext4_free_io_end(io_end
);
2939 aio_complete(iocb
, ret
, 0);
2940 inode_dio_done(inode
);
2944 io_end
->offset
= offset
;
2945 io_end
->size
= size
;
2947 io_end
->iocb
= iocb
;
2948 io_end
->result
= ret
;
2951 ext4_add_complete_io(io_end
);
2955 * For ext4 extent files, ext4 will do direct-io write to holes,
2956 * preallocated extents, and those write extend the file, no need to
2957 * fall back to buffered IO.
2959 * For holes, we fallocate those blocks, mark them as uninitialized
2960 * If those blocks were preallocated, we mark sure they are split, but
2961 * still keep the range to write as uninitialized.
2963 * The unwritten extents will be converted to written when DIO is completed.
2964 * For async direct IO, since the IO may still pending when return, we
2965 * set up an end_io call back function, which will do the conversion
2966 * when async direct IO completed.
2968 * If the O_DIRECT write will extend the file then add this inode to the
2969 * orphan list. So recovery will truncate it back to the original size
2970 * if the machine crashes during the write.
2973 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
2974 const struct iovec
*iov
, loff_t offset
,
2975 unsigned long nr_segs
)
2977 struct file
*file
= iocb
->ki_filp
;
2978 struct inode
*inode
= file
->f_mapping
->host
;
2980 size_t count
= iov_length(iov
, nr_segs
);
2982 get_block_t
*get_block_func
= NULL
;
2984 loff_t final_size
= offset
+ count
;
2986 /* Use the old path for reads and writes beyond i_size. */
2987 if (rw
!= WRITE
|| final_size
> inode
->i_size
)
2988 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
2990 BUG_ON(iocb
->private == NULL
);
2992 /* If we do a overwrite dio, i_mutex locking can be released */
2993 overwrite
= *((int *)iocb
->private);
2996 atomic_inc(&inode
->i_dio_count
);
2997 down_read(&EXT4_I(inode
)->i_data_sem
);
2998 mutex_unlock(&inode
->i_mutex
);
3002 * We could direct write to holes and fallocate.
3004 * Allocated blocks to fill the hole are marked as
3005 * uninitialized to prevent parallel buffered read to expose
3006 * the stale data before DIO complete the data IO.
3008 * As to previously fallocated extents, ext4 get_block will
3009 * just simply mark the buffer mapped but still keep the
3010 * extents uninitialized.
3012 * For non AIO case, we will convert those unwritten extents
3013 * to written after return back from blockdev_direct_IO.
3015 * For async DIO, the conversion needs to be deferred when the
3016 * IO is completed. The ext4 end_io callback function will be
3017 * called to take care of the conversion work. Here for async
3018 * case, we allocate an io_end structure to hook to the iocb.
3020 iocb
->private = NULL
;
3021 ext4_inode_aio_set(inode
, NULL
);
3022 if (!is_sync_kiocb(iocb
)) {
3023 ext4_io_end_t
*io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
3028 io_end
->flag
|= EXT4_IO_END_DIRECT
;
3029 iocb
->private = io_end
;
3031 * we save the io structure for current async direct
3032 * IO, so that later ext4_map_blocks() could flag the
3033 * io structure whether there is a unwritten extents
3034 * needs to be converted when IO is completed.
3036 ext4_inode_aio_set(inode
, io_end
);
3040 get_block_func
= ext4_get_block_write_nolock
;
3042 get_block_func
= ext4_get_block_write
;
3043 dio_flags
= DIO_LOCKING
;
3045 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3046 inode
->i_sb
->s_bdev
, iov
,
3054 ext4_inode_aio_set(inode
, NULL
);
3056 * The io_end structure takes a reference to the inode, that
3057 * structure needs to be destroyed and the reference to the
3058 * inode need to be dropped, when IO is complete, even with 0
3059 * byte write, or failed.
3061 * In the successful AIO DIO case, the io_end structure will
3062 * be destroyed and the reference to the inode will be dropped
3063 * after the end_io call back function is called.
3065 * In the case there is 0 byte write, or error case, since VFS
3066 * direct IO won't invoke the end_io call back function, we
3067 * need to free the end_io structure here.
3069 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3070 ext4_free_io_end(iocb
->private);
3071 iocb
->private = NULL
;
3072 } else if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3073 EXT4_STATE_DIO_UNWRITTEN
)) {
3076 * for non AIO case, since the IO is already
3077 * completed, we could do the conversion right here
3079 err
= ext4_convert_unwritten_extents(inode
,
3083 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3087 /* take i_mutex locking again if we do a ovewrite dio */
3089 inode_dio_done(inode
);
3090 up_read(&EXT4_I(inode
)->i_data_sem
);
3091 mutex_lock(&inode
->i_mutex
);
3097 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3098 const struct iovec
*iov
, loff_t offset
,
3099 unsigned long nr_segs
)
3101 struct file
*file
= iocb
->ki_filp
;
3102 struct inode
*inode
= file
->f_mapping
->host
;
3106 * If we are doing data journalling we don't support O_DIRECT
3108 if (ext4_should_journal_data(inode
))
3111 /* Let buffer I/O handle the inline data case. */
3112 if (ext4_has_inline_data(inode
))
3115 trace_ext4_direct_IO_enter(inode
, offset
, iov_length(iov
, nr_segs
), rw
);
3116 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3117 ret
= ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3119 ret
= ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3120 trace_ext4_direct_IO_exit(inode
, offset
,
3121 iov_length(iov
, nr_segs
), rw
, ret
);
3126 * Pages can be marked dirty completely asynchronously from ext4's journalling
3127 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3128 * much here because ->set_page_dirty is called under VFS locks. The page is
3129 * not necessarily locked.
3131 * We cannot just dirty the page and leave attached buffers clean, because the
3132 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3133 * or jbddirty because all the journalling code will explode.
3135 * So what we do is to mark the page "pending dirty" and next time writepage
3136 * is called, propagate that into the buffers appropriately.
3138 static int ext4_journalled_set_page_dirty(struct page
*page
)
3140 SetPageChecked(page
);
3141 return __set_page_dirty_nobuffers(page
);
3144 static const struct address_space_operations ext4_ordered_aops
= {
3145 .readpage
= ext4_readpage
,
3146 .readpages
= ext4_readpages
,
3147 .writepage
= ext4_writepage
,
3148 .write_begin
= ext4_write_begin
,
3149 .write_end
= ext4_ordered_write_end
,
3151 .invalidatepage
= ext4_invalidatepage
,
3152 .releasepage
= ext4_releasepage
,
3153 .direct_IO
= ext4_direct_IO
,
3154 .migratepage
= buffer_migrate_page
,
3155 .is_partially_uptodate
= block_is_partially_uptodate
,
3156 .error_remove_page
= generic_error_remove_page
,
3159 static const struct address_space_operations ext4_writeback_aops
= {
3160 .readpage
= ext4_readpage
,
3161 .readpages
= ext4_readpages
,
3162 .writepage
= ext4_writepage
,
3163 .write_begin
= ext4_write_begin
,
3164 .write_end
= ext4_writeback_write_end
,
3166 .invalidatepage
= ext4_invalidatepage
,
3167 .releasepage
= ext4_releasepage
,
3168 .direct_IO
= ext4_direct_IO
,
3169 .migratepage
= buffer_migrate_page
,
3170 .is_partially_uptodate
= block_is_partially_uptodate
,
3171 .error_remove_page
= generic_error_remove_page
,
3174 static const struct address_space_operations ext4_journalled_aops
= {
3175 .readpage
= ext4_readpage
,
3176 .readpages
= ext4_readpages
,
3177 .writepage
= ext4_writepage
,
3178 .write_begin
= ext4_write_begin
,
3179 .write_end
= ext4_journalled_write_end
,
3180 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3182 .invalidatepage
= ext4_journalled_invalidatepage
,
3183 .releasepage
= ext4_releasepage
,
3184 .direct_IO
= ext4_direct_IO
,
3185 .is_partially_uptodate
= block_is_partially_uptodate
,
3186 .error_remove_page
= generic_error_remove_page
,
3189 static const struct address_space_operations ext4_da_aops
= {
3190 .readpage
= ext4_readpage
,
3191 .readpages
= ext4_readpages
,
3192 .writepage
= ext4_writepage
,
3193 .writepages
= ext4_da_writepages
,
3194 .write_begin
= ext4_da_write_begin
,
3195 .write_end
= ext4_da_write_end
,
3197 .invalidatepage
= ext4_da_invalidatepage
,
3198 .releasepage
= ext4_releasepage
,
3199 .direct_IO
= ext4_direct_IO
,
3200 .migratepage
= buffer_migrate_page
,
3201 .is_partially_uptodate
= block_is_partially_uptodate
,
3202 .error_remove_page
= generic_error_remove_page
,
3205 void ext4_set_aops(struct inode
*inode
)
3207 switch (ext4_inode_journal_mode(inode
)) {
3208 case EXT4_INODE_ORDERED_DATA_MODE
:
3209 if (test_opt(inode
->i_sb
, DELALLOC
))
3210 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3212 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3214 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3215 if (test_opt(inode
->i_sb
, DELALLOC
))
3216 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3218 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3220 case EXT4_INODE_JOURNAL_DATA_MODE
:
3221 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3230 * ext4_discard_partial_page_buffers()
3231 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3232 * This function finds and locks the page containing the offset
3233 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3234 * Calling functions that already have the page locked should call
3235 * ext4_discard_partial_page_buffers_no_lock directly.
3237 int ext4_discard_partial_page_buffers(handle_t
*handle
,
3238 struct address_space
*mapping
, loff_t from
,
3239 loff_t length
, int flags
)
3241 struct inode
*inode
= mapping
->host
;
3245 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3246 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3250 err
= ext4_discard_partial_page_buffers_no_lock(handle
, inode
, page
,
3251 from
, length
, flags
);
3254 page_cache_release(page
);
3259 * ext4_discard_partial_page_buffers_no_lock()
3260 * Zeros a page range of length 'length' starting from offset 'from'.
3261 * Buffer heads that correspond to the block aligned regions of the
3262 * zeroed range will be unmapped. Unblock aligned regions
3263 * will have the corresponding buffer head mapped if needed so that
3264 * that region of the page can be updated with the partial zero out.
3266 * This function assumes that the page has already been locked. The
3267 * The range to be discarded must be contained with in the given page.
3268 * If the specified range exceeds the end of the page it will be shortened
3269 * to the end of the page that corresponds to 'from'. This function is
3270 * appropriate for updating a page and it buffer heads to be unmapped and
3271 * zeroed for blocks that have been either released, or are going to be
3274 * handle: The journal handle
3275 * inode: The files inode
3276 * page: A locked page that contains the offset "from"
3277 * from: The starting byte offset (from the beginning of the file)
3278 * to begin discarding
3279 * len: The length of bytes to discard
3280 * flags: Optional flags that may be used:
3282 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3283 * Only zero the regions of the page whose buffer heads
3284 * have already been unmapped. This flag is appropriate
3285 * for updating the contents of a page whose blocks may
3286 * have already been released, and we only want to zero
3287 * out the regions that correspond to those released blocks.
3289 * Returns zero on success or negative on failure.
3291 static int ext4_discard_partial_page_buffers_no_lock(handle_t
*handle
,
3292 struct inode
*inode
, struct page
*page
, loff_t from
,
3293 loff_t length
, int flags
)
3295 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3296 unsigned int offset
= from
& (PAGE_CACHE_SIZE
-1);
3297 unsigned int blocksize
, max
, pos
;
3299 struct buffer_head
*bh
;
3302 blocksize
= inode
->i_sb
->s_blocksize
;
3303 max
= PAGE_CACHE_SIZE
- offset
;
3305 if (index
!= page
->index
)
3309 * correct length if it does not fall between
3310 * 'from' and the end of the page
3312 if (length
> max
|| length
< 0)
3315 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3317 if (!page_has_buffers(page
))
3318 create_empty_buffers(page
, blocksize
, 0);
3320 /* Find the buffer that contains "offset" */
3321 bh
= page_buffers(page
);
3323 while (offset
>= pos
) {
3324 bh
= bh
->b_this_page
;
3330 while (pos
< offset
+ length
) {
3331 unsigned int end_of_block
, range_to_discard
;
3335 /* The length of space left to zero and unmap */
3336 range_to_discard
= offset
+ length
- pos
;
3338 /* The length of space until the end of the block */
3339 end_of_block
= blocksize
- (pos
& (blocksize
-1));
3342 * Do not unmap or zero past end of block
3343 * for this buffer head
3345 if (range_to_discard
> end_of_block
)
3346 range_to_discard
= end_of_block
;
3350 * Skip this buffer head if we are only zeroing unampped
3351 * regions of the page
3353 if (flags
& EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
&&
3357 /* If the range is block aligned, unmap */
3358 if (range_to_discard
== blocksize
) {
3359 clear_buffer_dirty(bh
);
3361 clear_buffer_mapped(bh
);
3362 clear_buffer_req(bh
);
3363 clear_buffer_new(bh
);
3364 clear_buffer_delay(bh
);
3365 clear_buffer_unwritten(bh
);
3366 clear_buffer_uptodate(bh
);
3367 zero_user(page
, pos
, range_to_discard
);
3368 BUFFER_TRACE(bh
, "Buffer discarded");
3373 * If this block is not completely contained in the range
3374 * to be discarded, then it is not going to be released. Because
3375 * we need to keep this block, we need to make sure this part
3376 * of the page is uptodate before we modify it by writeing
3377 * partial zeros on it.
3379 if (!buffer_mapped(bh
)) {
3381 * Buffer head must be mapped before we can read
3384 BUFFER_TRACE(bh
, "unmapped");
3385 ext4_get_block(inode
, iblock
, bh
, 0);
3386 /* unmapped? It's a hole - nothing to do */
3387 if (!buffer_mapped(bh
)) {
3388 BUFFER_TRACE(bh
, "still unmapped");
3393 /* Ok, it's mapped. Make sure it's up-to-date */
3394 if (PageUptodate(page
))
3395 set_buffer_uptodate(bh
);
3397 if (!buffer_uptodate(bh
)) {
3399 ll_rw_block(READ
, 1, &bh
);
3401 /* Uhhuh. Read error. Complain and punt.*/
3402 if (!buffer_uptodate(bh
))
3406 if (ext4_should_journal_data(inode
)) {
3407 BUFFER_TRACE(bh
, "get write access");
3408 err
= ext4_journal_get_write_access(handle
, bh
);
3413 zero_user(page
, pos
, range_to_discard
);
3416 if (ext4_should_journal_data(inode
)) {
3417 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3419 mark_buffer_dirty(bh
);
3421 BUFFER_TRACE(bh
, "Partial buffer zeroed");
3423 bh
= bh
->b_this_page
;
3425 pos
+= range_to_discard
;
3431 int ext4_can_truncate(struct inode
*inode
)
3433 if (S_ISREG(inode
->i_mode
))
3435 if (S_ISDIR(inode
->i_mode
))
3437 if (S_ISLNK(inode
->i_mode
))
3438 return !ext4_inode_is_fast_symlink(inode
);
3443 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3444 * associated with the given offset and length
3446 * @inode: File inode
3447 * @offset: The offset where the hole will begin
3448 * @len: The length of the hole
3450 * Returns: 0 on success or negative on failure
3453 int ext4_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
3455 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
3456 if (!S_ISREG(inode
->i_mode
))
3459 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3460 return ext4_ind_punch_hole(file
, offset
, length
);
3462 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) {
3463 /* TODO: Add support for bigalloc file systems */
3467 trace_ext4_punch_hole(inode
, offset
, length
);
3469 return ext4_ext_punch_hole(file
, offset
, length
);
3475 * We block out ext4_get_block() block instantiations across the entire
3476 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3477 * simultaneously on behalf of the same inode.
3479 * As we work through the truncate and commit bits of it to the journal there
3480 * is one core, guiding principle: the file's tree must always be consistent on
3481 * disk. We must be able to restart the truncate after a crash.
3483 * The file's tree may be transiently inconsistent in memory (although it
3484 * probably isn't), but whenever we close off and commit a journal transaction,
3485 * the contents of (the filesystem + the journal) must be consistent and
3486 * restartable. It's pretty simple, really: bottom up, right to left (although
3487 * left-to-right works OK too).
3489 * Note that at recovery time, journal replay occurs *before* the restart of
3490 * truncate against the orphan inode list.
3492 * The committed inode has the new, desired i_size (which is the same as
3493 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3494 * that this inode's truncate did not complete and it will again call
3495 * ext4_truncate() to have another go. So there will be instantiated blocks
3496 * to the right of the truncation point in a crashed ext4 filesystem. But
3497 * that's fine - as long as they are linked from the inode, the post-crash
3498 * ext4_truncate() run will find them and release them.
3500 void ext4_truncate(struct inode
*inode
)
3502 trace_ext4_truncate_enter(inode
);
3504 if (!ext4_can_truncate(inode
))
3507 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3509 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3510 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3512 if (ext4_has_inline_data(inode
)) {
3515 ext4_inline_data_truncate(inode
, &has_inline
);
3520 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3521 ext4_ext_truncate(inode
);
3523 ext4_ind_truncate(inode
);
3525 trace_ext4_truncate_exit(inode
);
3529 * ext4_get_inode_loc returns with an extra refcount against the inode's
3530 * underlying buffer_head on success. If 'in_mem' is true, we have all
3531 * data in memory that is needed to recreate the on-disk version of this
3534 static int __ext4_get_inode_loc(struct inode
*inode
,
3535 struct ext4_iloc
*iloc
, int in_mem
)
3537 struct ext4_group_desc
*gdp
;
3538 struct buffer_head
*bh
;
3539 struct super_block
*sb
= inode
->i_sb
;
3541 int inodes_per_block
, inode_offset
;
3544 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3547 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3548 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3553 * Figure out the offset within the block group inode table
3555 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3556 inode_offset
= ((inode
->i_ino
- 1) %
3557 EXT4_INODES_PER_GROUP(sb
));
3558 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3559 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3561 bh
= sb_getblk(sb
, block
);
3564 if (!buffer_uptodate(bh
)) {
3568 * If the buffer has the write error flag, we have failed
3569 * to write out another inode in the same block. In this
3570 * case, we don't have to read the block because we may
3571 * read the old inode data successfully.
3573 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3574 set_buffer_uptodate(bh
);
3576 if (buffer_uptodate(bh
)) {
3577 /* someone brought it uptodate while we waited */
3583 * If we have all information of the inode in memory and this
3584 * is the only valid inode in the block, we need not read the
3588 struct buffer_head
*bitmap_bh
;
3591 start
= inode_offset
& ~(inodes_per_block
- 1);
3593 /* Is the inode bitmap in cache? */
3594 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3595 if (unlikely(!bitmap_bh
))
3599 * If the inode bitmap isn't in cache then the
3600 * optimisation may end up performing two reads instead
3601 * of one, so skip it.
3603 if (!buffer_uptodate(bitmap_bh
)) {
3607 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3608 if (i
== inode_offset
)
3610 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3614 if (i
== start
+ inodes_per_block
) {
3615 /* all other inodes are free, so skip I/O */
3616 memset(bh
->b_data
, 0, bh
->b_size
);
3617 set_buffer_uptodate(bh
);
3625 * If we need to do any I/O, try to pre-readahead extra
3626 * blocks from the inode table.
3628 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3629 ext4_fsblk_t b
, end
, table
;
3632 table
= ext4_inode_table(sb
, gdp
);
3633 /* s_inode_readahead_blks is always a power of 2 */
3634 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
3637 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
3638 num
= EXT4_INODES_PER_GROUP(sb
);
3639 if (ext4_has_group_desc_csum(sb
))
3640 num
-= ext4_itable_unused_count(sb
, gdp
);
3641 table
+= num
/ inodes_per_block
;
3645 sb_breadahead(sb
, b
++);
3649 * There are other valid inodes in the buffer, this inode
3650 * has in-inode xattrs, or we don't have this inode in memory.
3651 * Read the block from disk.
3653 trace_ext4_load_inode(inode
);
3655 bh
->b_end_io
= end_buffer_read_sync
;
3656 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
3658 if (!buffer_uptodate(bh
)) {
3659 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3660 "unable to read itable block");
3670 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3672 /* We have all inode data except xattrs in memory here. */
3673 return __ext4_get_inode_loc(inode
, iloc
,
3674 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
3677 void ext4_set_inode_flags(struct inode
*inode
)
3679 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3681 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3682 if (flags
& EXT4_SYNC_FL
)
3683 inode
->i_flags
|= S_SYNC
;
3684 if (flags
& EXT4_APPEND_FL
)
3685 inode
->i_flags
|= S_APPEND
;
3686 if (flags
& EXT4_IMMUTABLE_FL
)
3687 inode
->i_flags
|= S_IMMUTABLE
;
3688 if (flags
& EXT4_NOATIME_FL
)
3689 inode
->i_flags
|= S_NOATIME
;
3690 if (flags
& EXT4_DIRSYNC_FL
)
3691 inode
->i_flags
|= S_DIRSYNC
;
3694 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3695 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3697 unsigned int vfs_fl
;
3698 unsigned long old_fl
, new_fl
;
3701 vfs_fl
= ei
->vfs_inode
.i_flags
;
3702 old_fl
= ei
->i_flags
;
3703 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3704 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
3706 if (vfs_fl
& S_SYNC
)
3707 new_fl
|= EXT4_SYNC_FL
;
3708 if (vfs_fl
& S_APPEND
)
3709 new_fl
|= EXT4_APPEND_FL
;
3710 if (vfs_fl
& S_IMMUTABLE
)
3711 new_fl
|= EXT4_IMMUTABLE_FL
;
3712 if (vfs_fl
& S_NOATIME
)
3713 new_fl
|= EXT4_NOATIME_FL
;
3714 if (vfs_fl
& S_DIRSYNC
)
3715 new_fl
|= EXT4_DIRSYNC_FL
;
3716 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
3719 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3720 struct ext4_inode_info
*ei
)
3723 struct inode
*inode
= &(ei
->vfs_inode
);
3724 struct super_block
*sb
= inode
->i_sb
;
3726 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3727 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3728 /* we are using combined 48 bit field */
3729 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
3730 le32_to_cpu(raw_inode
->i_blocks_lo
);
3731 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
3732 /* i_blocks represent file system block size */
3733 return i_blocks
<< (inode
->i_blkbits
- 9);
3738 return le32_to_cpu(raw_inode
->i_blocks_lo
);
3742 static inline void ext4_iget_extra_inode(struct inode
*inode
,
3743 struct ext4_inode
*raw_inode
,
3744 struct ext4_inode_info
*ei
)
3746 __le32
*magic
= (void *)raw_inode
+
3747 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
;
3748 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
)) {
3749 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
3750 ext4_find_inline_data_nolock(inode
);
3752 EXT4_I(inode
)->i_inline_off
= 0;
3755 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
3757 struct ext4_iloc iloc
;
3758 struct ext4_inode
*raw_inode
;
3759 struct ext4_inode_info
*ei
;
3760 struct inode
*inode
;
3761 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
3767 inode
= iget_locked(sb
, ino
);
3769 return ERR_PTR(-ENOMEM
);
3770 if (!(inode
->i_state
& I_NEW
))
3776 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
3779 raw_inode
= ext4_raw_inode(&iloc
);
3781 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3782 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
3783 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
3784 EXT4_INODE_SIZE(inode
->i_sb
)) {
3785 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
3786 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
3787 EXT4_INODE_SIZE(inode
->i_sb
));
3792 ei
->i_extra_isize
= 0;
3794 /* Precompute checksum seed for inode metadata */
3795 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3796 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
)) {
3797 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3799 __le32 inum
= cpu_to_le32(inode
->i_ino
);
3800 __le32 gen
= raw_inode
->i_generation
;
3801 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
3803 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
3807 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
3808 EXT4_ERROR_INODE(inode
, "checksum invalid");
3813 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
3814 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
3815 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
3816 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
3817 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
3818 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
3820 i_uid_write(inode
, i_uid
);
3821 i_gid_write(inode
, i_gid
);
3822 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
3824 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
3825 ei
->i_inline_off
= 0;
3826 ei
->i_dir_start_lookup
= 0;
3827 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
3828 /* We now have enough fields to check if the inode was active or not.
3829 * This is needed because nfsd might try to access dead inodes
3830 * the test is that same one that e2fsck uses
3831 * NeilBrown 1999oct15
3833 if (inode
->i_nlink
== 0) {
3834 if (inode
->i_mode
== 0 ||
3835 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
3836 /* this inode is deleted */
3840 /* The only unlinked inodes we let through here have
3841 * valid i_mode and are being read by the orphan
3842 * recovery code: that's fine, we're about to complete
3843 * the process of deleting those. */
3845 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
3846 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
3847 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
3848 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
3850 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
3851 inode
->i_size
= ext4_isize(raw_inode
);
3852 ei
->i_disksize
= inode
->i_size
;
3854 ei
->i_reserved_quota
= 0;
3856 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
3857 ei
->i_block_group
= iloc
.block_group
;
3858 ei
->i_last_alloc_group
= ~0;
3860 * NOTE! The in-memory inode i_data array is in little-endian order
3861 * even on big-endian machines: we do NOT byteswap the block numbers!
3863 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
3864 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
3865 INIT_LIST_HEAD(&ei
->i_orphan
);
3868 * Set transaction id's of transactions that have to be committed
3869 * to finish f[data]sync. We set them to currently running transaction
3870 * as we cannot be sure that the inode or some of its metadata isn't
3871 * part of the transaction - the inode could have been reclaimed and
3872 * now it is reread from disk.
3875 transaction_t
*transaction
;
3878 read_lock(&journal
->j_state_lock
);
3879 if (journal
->j_running_transaction
)
3880 transaction
= journal
->j_running_transaction
;
3882 transaction
= journal
->j_committing_transaction
;
3884 tid
= transaction
->t_tid
;
3886 tid
= journal
->j_commit_sequence
;
3887 read_unlock(&journal
->j_state_lock
);
3888 ei
->i_sync_tid
= tid
;
3889 ei
->i_datasync_tid
= tid
;
3892 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3893 if (ei
->i_extra_isize
== 0) {
3894 /* The extra space is currently unused. Use it. */
3895 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
3896 EXT4_GOOD_OLD_INODE_SIZE
;
3898 ext4_iget_extra_inode(inode
, raw_inode
, ei
);
3902 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
3903 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
3904 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
3905 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
3907 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
3908 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3909 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
3911 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
3915 if (ei
->i_file_acl
&&
3916 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
3917 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
3921 } else if (!ext4_has_inline_data(inode
)) {
3922 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
3923 if ((S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
3924 (S_ISLNK(inode
->i_mode
) &&
3925 !ext4_inode_is_fast_symlink(inode
))))
3926 /* Validate extent which is part of inode */
3927 ret
= ext4_ext_check_inode(inode
);
3928 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
3929 (S_ISLNK(inode
->i_mode
) &&
3930 !ext4_inode_is_fast_symlink(inode
))) {
3931 /* Validate block references which are part of inode */
3932 ret
= ext4_ind_check_inode(inode
);
3938 if (S_ISREG(inode
->i_mode
)) {
3939 inode
->i_op
= &ext4_file_inode_operations
;
3940 inode
->i_fop
= &ext4_file_operations
;
3941 ext4_set_aops(inode
);
3942 } else if (S_ISDIR(inode
->i_mode
)) {
3943 inode
->i_op
= &ext4_dir_inode_operations
;
3944 inode
->i_fop
= &ext4_dir_operations
;
3945 } else if (S_ISLNK(inode
->i_mode
)) {
3946 if (ext4_inode_is_fast_symlink(inode
)) {
3947 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
3948 nd_terminate_link(ei
->i_data
, inode
->i_size
,
3949 sizeof(ei
->i_data
) - 1);
3951 inode
->i_op
= &ext4_symlink_inode_operations
;
3952 ext4_set_aops(inode
);
3954 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
3955 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
3956 inode
->i_op
= &ext4_special_inode_operations
;
3957 if (raw_inode
->i_block
[0])
3958 init_special_inode(inode
, inode
->i_mode
,
3959 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
3961 init_special_inode(inode
, inode
->i_mode
,
3962 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
3965 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
3969 ext4_set_inode_flags(inode
);
3970 unlock_new_inode(inode
);
3976 return ERR_PTR(ret
);
3979 static int ext4_inode_blocks_set(handle_t
*handle
,
3980 struct ext4_inode
*raw_inode
,
3981 struct ext4_inode_info
*ei
)
3983 struct inode
*inode
= &(ei
->vfs_inode
);
3984 u64 i_blocks
= inode
->i_blocks
;
3985 struct super_block
*sb
= inode
->i_sb
;
3987 if (i_blocks
<= ~0U) {
3989 * i_blocks can be represented in a 32 bit variable
3990 * as multiple of 512 bytes
3992 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
3993 raw_inode
->i_blocks_high
= 0;
3994 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
3997 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4000 if (i_blocks
<= 0xffffffffffffULL
) {
4002 * i_blocks can be represented in a 48 bit variable
4003 * as multiple of 512 bytes
4005 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4006 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4007 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4009 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4010 /* i_block is stored in file system block size */
4011 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4012 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4013 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4019 * Post the struct inode info into an on-disk inode location in the
4020 * buffer-cache. This gobbles the caller's reference to the
4021 * buffer_head in the inode location struct.
4023 * The caller must have write access to iloc->bh.
4025 static int ext4_do_update_inode(handle_t
*handle
,
4026 struct inode
*inode
,
4027 struct ext4_iloc
*iloc
)
4029 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4030 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4031 struct buffer_head
*bh
= iloc
->bh
;
4032 int err
= 0, rc
, block
;
4033 int need_datasync
= 0;
4037 /* For fields not not tracking in the in-memory inode,
4038 * initialise them to zero for new inodes. */
4039 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4040 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4042 ext4_get_inode_flags(ei
);
4043 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4044 i_uid
= i_uid_read(inode
);
4045 i_gid
= i_gid_read(inode
);
4046 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4047 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4048 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4050 * Fix up interoperability with old kernels. Otherwise, old inodes get
4051 * re-used with the upper 16 bits of the uid/gid intact
4054 raw_inode
->i_uid_high
=
4055 cpu_to_le16(high_16_bits(i_uid
));
4056 raw_inode
->i_gid_high
=
4057 cpu_to_le16(high_16_bits(i_gid
));
4059 raw_inode
->i_uid_high
= 0;
4060 raw_inode
->i_gid_high
= 0;
4063 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4064 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4065 raw_inode
->i_uid_high
= 0;
4066 raw_inode
->i_gid_high
= 0;
4068 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4070 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4071 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4072 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4073 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4075 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4077 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4078 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4079 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4080 cpu_to_le32(EXT4_OS_HURD
))
4081 raw_inode
->i_file_acl_high
=
4082 cpu_to_le16(ei
->i_file_acl
>> 32);
4083 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4084 if (ei
->i_disksize
!= ext4_isize(raw_inode
)) {
4085 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4088 if (ei
->i_disksize
> 0x7fffffffULL
) {
4089 struct super_block
*sb
= inode
->i_sb
;
4090 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4091 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4092 EXT4_SB(sb
)->s_es
->s_rev_level
==
4093 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4094 /* If this is the first large file
4095 * created, add a flag to the superblock.
4097 err
= ext4_journal_get_write_access(handle
,
4098 EXT4_SB(sb
)->s_sbh
);
4101 ext4_update_dynamic_rev(sb
);
4102 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4103 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4104 ext4_handle_sync(handle
);
4105 err
= ext4_handle_dirty_super(handle
, sb
);
4108 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4109 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4110 if (old_valid_dev(inode
->i_rdev
)) {
4111 raw_inode
->i_block
[0] =
4112 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4113 raw_inode
->i_block
[1] = 0;
4115 raw_inode
->i_block
[0] = 0;
4116 raw_inode
->i_block
[1] =
4117 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4118 raw_inode
->i_block
[2] = 0;
4120 } else if (!ext4_has_inline_data(inode
)) {
4121 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4122 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4125 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4126 if (ei
->i_extra_isize
) {
4127 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4128 raw_inode
->i_version_hi
=
4129 cpu_to_le32(inode
->i_version
>> 32);
4130 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4133 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4135 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4136 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4139 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4141 ext4_update_inode_fsync_trans(handle
, inode
, need_datasync
);
4144 ext4_std_error(inode
->i_sb
, err
);
4149 * ext4_write_inode()
4151 * We are called from a few places:
4153 * - Within generic_file_write() for O_SYNC files.
4154 * Here, there will be no transaction running. We wait for any running
4155 * transaction to commit.
4157 * - Within sys_sync(), kupdate and such.
4158 * We wait on commit, if tol to.
4160 * - Within prune_icache() (PF_MEMALLOC == true)
4161 * Here we simply return. We can't afford to block kswapd on the
4164 * In all cases it is actually safe for us to return without doing anything,
4165 * because the inode has been copied into a raw inode buffer in
4166 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4169 * Note that we are absolutely dependent upon all inode dirtiers doing the
4170 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4171 * which we are interested.
4173 * It would be a bug for them to not do this. The code:
4175 * mark_inode_dirty(inode)
4177 * inode->i_size = expr;
4179 * is in error because a kswapd-driven write_inode() could occur while
4180 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4181 * will no longer be on the superblock's dirty inode list.
4183 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4187 if (current
->flags
& PF_MEMALLOC
)
4190 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4191 if (ext4_journal_current_handle()) {
4192 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4197 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
4200 err
= ext4_force_commit(inode
->i_sb
);
4202 struct ext4_iloc iloc
;
4204 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4207 if (wbc
->sync_mode
== WB_SYNC_ALL
)
4208 sync_dirty_buffer(iloc
.bh
);
4209 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4210 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4211 "IO error syncing inode");
4220 * In data=journal mode ext4_journalled_invalidatepage() may fail to invalidate
4221 * buffers that are attached to a page stradding i_size and are undergoing
4222 * commit. In that case we have to wait for commit to finish and try again.
4224 static void ext4_wait_for_tail_page_commit(struct inode
*inode
)
4228 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
4229 tid_t commit_tid
= 0;
4232 offset
= inode
->i_size
& (PAGE_CACHE_SIZE
- 1);
4234 * All buffers in the last page remain valid? Then there's nothing to
4235 * do. We do the check mainly to optimize the common PAGE_CACHE_SIZE ==
4238 if (offset
> PAGE_CACHE_SIZE
- (1 << inode
->i_blkbits
))
4241 page
= find_lock_page(inode
->i_mapping
,
4242 inode
->i_size
>> PAGE_CACHE_SHIFT
);
4245 ret
= __ext4_journalled_invalidatepage(page
, offset
);
4247 page_cache_release(page
);
4251 read_lock(&journal
->j_state_lock
);
4252 if (journal
->j_committing_transaction
)
4253 commit_tid
= journal
->j_committing_transaction
->t_tid
;
4254 read_unlock(&journal
->j_state_lock
);
4256 jbd2_log_wait_commit(journal
, commit_tid
);
4263 * Called from notify_change.
4265 * We want to trap VFS attempts to truncate the file as soon as
4266 * possible. In particular, we want to make sure that when the VFS
4267 * shrinks i_size, we put the inode on the orphan list and modify
4268 * i_disksize immediately, so that during the subsequent flushing of
4269 * dirty pages and freeing of disk blocks, we can guarantee that any
4270 * commit will leave the blocks being flushed in an unused state on
4271 * disk. (On recovery, the inode will get truncated and the blocks will
4272 * be freed, so we have a strong guarantee that no future commit will
4273 * leave these blocks visible to the user.)
4275 * Another thing we have to assure is that if we are in ordered mode
4276 * and inode is still attached to the committing transaction, we must
4277 * we start writeout of all the dirty pages which are being truncated.
4278 * This way we are sure that all the data written in the previous
4279 * transaction are already on disk (truncate waits for pages under
4282 * Called with inode->i_mutex down.
4284 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4286 struct inode
*inode
= dentry
->d_inode
;
4289 const unsigned int ia_valid
= attr
->ia_valid
;
4291 error
= inode_change_ok(inode
, attr
);
4295 if (is_quota_modification(inode
, attr
))
4296 dquot_initialize(inode
);
4297 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4298 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4301 /* (user+group)*(old+new) structure, inode write (sb,
4302 * inode block, ? - but truncate inode update has it) */
4303 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
4304 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
4305 if (IS_ERR(handle
)) {
4306 error
= PTR_ERR(handle
);
4309 error
= dquot_transfer(inode
, attr
);
4311 ext4_journal_stop(handle
);
4314 /* Update corresponding info in inode so that everything is in
4315 * one transaction */
4316 if (attr
->ia_valid
& ATTR_UID
)
4317 inode
->i_uid
= attr
->ia_uid
;
4318 if (attr
->ia_valid
& ATTR_GID
)
4319 inode
->i_gid
= attr
->ia_gid
;
4320 error
= ext4_mark_inode_dirty(handle
, inode
);
4321 ext4_journal_stop(handle
);
4324 if (attr
->ia_valid
& ATTR_SIZE
) {
4326 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4327 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4329 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4334 if (S_ISREG(inode
->i_mode
) &&
4335 attr
->ia_valid
& ATTR_SIZE
&&
4336 (attr
->ia_size
< inode
->i_size
)) {
4339 handle
= ext4_journal_start(inode
, 3);
4340 if (IS_ERR(handle
)) {
4341 error
= PTR_ERR(handle
);
4344 if (ext4_handle_valid(handle
)) {
4345 error
= ext4_orphan_add(handle
, inode
);
4348 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4349 rc
= ext4_mark_inode_dirty(handle
, inode
);
4352 ext4_journal_stop(handle
);
4354 if (ext4_should_order_data(inode
)) {
4355 error
= ext4_begin_ordered_truncate(inode
,
4358 /* Do as much error cleanup as possible */
4359 handle
= ext4_journal_start(inode
, 3);
4360 if (IS_ERR(handle
)) {
4361 ext4_orphan_del(NULL
, inode
);
4364 ext4_orphan_del(handle
, inode
);
4366 ext4_journal_stop(handle
);
4372 if (attr
->ia_valid
& ATTR_SIZE
) {
4373 if (attr
->ia_size
!= inode
->i_size
) {
4374 loff_t oldsize
= inode
->i_size
;
4376 i_size_write(inode
, attr
->ia_size
);
4378 * Blocks are going to be removed from the inode. Wait
4379 * for dio in flight. Temporarily disable
4380 * dioread_nolock to prevent livelock.
4383 if (!ext4_should_journal_data(inode
)) {
4384 ext4_inode_block_unlocked_dio(inode
);
4385 inode_dio_wait(inode
);
4386 ext4_inode_resume_unlocked_dio(inode
);
4388 ext4_wait_for_tail_page_commit(inode
);
4391 * Truncate pagecache after we've waited for commit
4392 * in data=journal mode to make pages freeable.
4394 truncate_pagecache(inode
, oldsize
, inode
->i_size
);
4396 ext4_truncate(inode
);
4400 setattr_copy(inode
, attr
);
4401 mark_inode_dirty(inode
);
4405 * If the call to ext4_truncate failed to get a transaction handle at
4406 * all, we need to clean up the in-core orphan list manually.
4408 if (orphan
&& inode
->i_nlink
)
4409 ext4_orphan_del(NULL
, inode
);
4411 if (!rc
&& (ia_valid
& ATTR_MODE
))
4412 rc
= ext4_acl_chmod(inode
);
4415 ext4_std_error(inode
->i_sb
, error
);
4421 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4424 struct inode
*inode
;
4425 unsigned long delalloc_blocks
;
4427 inode
= dentry
->d_inode
;
4428 generic_fillattr(inode
, stat
);
4431 * We can't update i_blocks if the block allocation is delayed
4432 * otherwise in the case of system crash before the real block
4433 * allocation is done, we will have i_blocks inconsistent with
4434 * on-disk file blocks.
4435 * We always keep i_blocks updated together with real
4436 * allocation. But to not confuse with user, stat
4437 * will return the blocks that include the delayed allocation
4438 * blocks for this file.
4440 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
4441 EXT4_I(inode
)->i_reserved_data_blocks
);
4443 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4447 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4449 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4450 return ext4_ind_trans_blocks(inode
, nrblocks
, chunk
);
4451 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4455 * Account for index blocks, block groups bitmaps and block group
4456 * descriptor blocks if modify datablocks and index blocks
4457 * worse case, the indexs blocks spread over different block groups
4459 * If datablocks are discontiguous, they are possible to spread over
4460 * different block groups too. If they are contiguous, with flexbg,
4461 * they could still across block group boundary.
4463 * Also account for superblock, inode, quota and xattr blocks
4465 static int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4467 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4473 * How many index blocks need to touch to modify nrblocks?
4474 * The "Chunk" flag indicating whether the nrblocks is
4475 * physically contiguous on disk
4477 * For Direct IO and fallocate, they calls get_block to allocate
4478 * one single extent at a time, so they could set the "Chunk" flag
4480 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4485 * Now let's see how many group bitmaps and group descriptors need
4495 if (groups
> ngroups
)
4497 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4498 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4500 /* bitmaps and block group descriptor blocks */
4501 ret
+= groups
+ gdpblocks
;
4503 /* Blocks for super block, inode, quota and xattr blocks */
4504 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4510 * Calculate the total number of credits to reserve to fit
4511 * the modification of a single pages into a single transaction,
4512 * which may include multiple chunks of block allocations.
4514 * This could be called via ext4_write_begin()
4516 * We need to consider the worse case, when
4517 * one new block per extent.
4519 int ext4_writepage_trans_blocks(struct inode
*inode
)
4521 int bpp
= ext4_journal_blocks_per_page(inode
);
4524 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4526 /* Account for data blocks for journalled mode */
4527 if (ext4_should_journal_data(inode
))
4533 * Calculate the journal credits for a chunk of data modification.
4535 * This is called from DIO, fallocate or whoever calling
4536 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4538 * journal buffers for data blocks are not included here, as DIO
4539 * and fallocate do no need to journal data buffers.
4541 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4543 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4547 * The caller must have previously called ext4_reserve_inode_write().
4548 * Give this, we know that the caller already has write access to iloc->bh.
4550 int ext4_mark_iloc_dirty(handle_t
*handle
,
4551 struct inode
*inode
, struct ext4_iloc
*iloc
)
4555 if (IS_I_VERSION(inode
))
4556 inode_inc_iversion(inode
);
4558 /* the do_update_inode consumes one bh->b_count */
4561 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4562 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4568 * On success, We end up with an outstanding reference count against
4569 * iloc->bh. This _must_ be cleaned up later.
4573 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4574 struct ext4_iloc
*iloc
)
4578 err
= ext4_get_inode_loc(inode
, iloc
);
4580 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4581 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4587 ext4_std_error(inode
->i_sb
, err
);
4592 * Expand an inode by new_extra_isize bytes.
4593 * Returns 0 on success or negative error number on failure.
4595 static int ext4_expand_extra_isize(struct inode
*inode
,
4596 unsigned int new_extra_isize
,
4597 struct ext4_iloc iloc
,
4600 struct ext4_inode
*raw_inode
;
4601 struct ext4_xattr_ibody_header
*header
;
4603 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4606 raw_inode
= ext4_raw_inode(&iloc
);
4608 header
= IHDR(inode
, raw_inode
);
4610 /* No extended attributes present */
4611 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
4612 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4613 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4615 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4619 /* try to expand with EAs present */
4620 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4625 * What we do here is to mark the in-core inode as clean with respect to inode
4626 * dirtiness (it may still be data-dirty).
4627 * This means that the in-core inode may be reaped by prune_icache
4628 * without having to perform any I/O. This is a very good thing,
4629 * because *any* task may call prune_icache - even ones which
4630 * have a transaction open against a different journal.
4632 * Is this cheating? Not really. Sure, we haven't written the
4633 * inode out, but prune_icache isn't a user-visible syncing function.
4634 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4635 * we start and wait on commits.
4637 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4639 struct ext4_iloc iloc
;
4640 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4641 static unsigned int mnt_count
;
4645 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
4646 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4647 if (ext4_handle_valid(handle
) &&
4648 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4649 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
4651 * We need extra buffer credits since we may write into EA block
4652 * with this same handle. If journal_extend fails, then it will
4653 * only result in a minor loss of functionality for that inode.
4654 * If this is felt to be critical, then e2fsck should be run to
4655 * force a large enough s_min_extra_isize.
4657 if ((jbd2_journal_extend(handle
,
4658 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4659 ret
= ext4_expand_extra_isize(inode
,
4660 sbi
->s_want_extra_isize
,
4663 ext4_set_inode_state(inode
,
4664 EXT4_STATE_NO_EXPAND
);
4666 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4667 ext4_warning(inode
->i_sb
,
4668 "Unable to expand inode %lu. Delete"
4669 " some EAs or run e2fsck.",
4672 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4678 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4683 * ext4_dirty_inode() is called from __mark_inode_dirty()
4685 * We're really interested in the case where a file is being extended.
4686 * i_size has been changed by generic_commit_write() and we thus need
4687 * to include the updated inode in the current transaction.
4689 * Also, dquot_alloc_block() will always dirty the inode when blocks
4690 * are allocated to the file.
4692 * If the inode is marked synchronous, we don't honour that here - doing
4693 * so would cause a commit on atime updates, which we don't bother doing.
4694 * We handle synchronous inodes at the highest possible level.
4696 void ext4_dirty_inode(struct inode
*inode
, int flags
)
4700 handle
= ext4_journal_start(inode
, 2);
4704 ext4_mark_inode_dirty(handle
, inode
);
4706 ext4_journal_stop(handle
);
4713 * Bind an inode's backing buffer_head into this transaction, to prevent
4714 * it from being flushed to disk early. Unlike
4715 * ext4_reserve_inode_write, this leaves behind no bh reference and
4716 * returns no iloc structure, so the caller needs to repeat the iloc
4717 * lookup to mark the inode dirty later.
4719 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4721 struct ext4_iloc iloc
;
4725 err
= ext4_get_inode_loc(inode
, &iloc
);
4727 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4728 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4730 err
= ext4_handle_dirty_metadata(handle
,
4736 ext4_std_error(inode
->i_sb
, err
);
4741 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4748 * We have to be very careful here: changing a data block's
4749 * journaling status dynamically is dangerous. If we write a
4750 * data block to the journal, change the status and then delete
4751 * that block, we risk forgetting to revoke the old log record
4752 * from the journal and so a subsequent replay can corrupt data.
4753 * So, first we make sure that the journal is empty and that
4754 * nobody is changing anything.
4757 journal
= EXT4_JOURNAL(inode
);
4760 if (is_journal_aborted(journal
))
4762 /* We have to allocate physical blocks for delalloc blocks
4763 * before flushing journal. otherwise delalloc blocks can not
4764 * be allocated any more. even more truncate on delalloc blocks
4765 * could trigger BUG by flushing delalloc blocks in journal.
4766 * There is no delalloc block in non-journal data mode.
4768 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
4769 err
= ext4_alloc_da_blocks(inode
);
4774 /* Wait for all existing dio workers */
4775 ext4_inode_block_unlocked_dio(inode
);
4776 inode_dio_wait(inode
);
4778 jbd2_journal_lock_updates(journal
);
4781 * OK, there are no updates running now, and all cached data is
4782 * synced to disk. We are now in a completely consistent state
4783 * which doesn't have anything in the journal, and we know that
4784 * no filesystem updates are running, so it is safe to modify
4785 * the inode's in-core data-journaling state flag now.
4789 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4791 jbd2_journal_flush(journal
);
4792 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4794 ext4_set_aops(inode
);
4796 jbd2_journal_unlock_updates(journal
);
4797 ext4_inode_resume_unlocked_dio(inode
);
4799 /* Finally we can mark the inode as dirty. */
4801 handle
= ext4_journal_start(inode
, 1);
4803 return PTR_ERR(handle
);
4805 err
= ext4_mark_inode_dirty(handle
, inode
);
4806 ext4_handle_sync(handle
);
4807 ext4_journal_stop(handle
);
4808 ext4_std_error(inode
->i_sb
, err
);
4813 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4815 return !buffer_mapped(bh
);
4818 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
4820 struct page
*page
= vmf
->page
;
4824 struct file
*file
= vma
->vm_file
;
4825 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4826 struct address_space
*mapping
= inode
->i_mapping
;
4828 get_block_t
*get_block
;
4831 sb_start_pagefault(inode
->i_sb
);
4832 file_update_time(vma
->vm_file
);
4833 /* Delalloc case is easy... */
4834 if (test_opt(inode
->i_sb
, DELALLOC
) &&
4835 !ext4_should_journal_data(inode
) &&
4836 !ext4_nonda_switch(inode
->i_sb
)) {
4838 ret
= __block_page_mkwrite(vma
, vmf
,
4839 ext4_da_get_block_prep
);
4840 } while (ret
== -ENOSPC
&&
4841 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
4846 size
= i_size_read(inode
);
4847 /* Page got truncated from under us? */
4848 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
4850 ret
= VM_FAULT_NOPAGE
;
4854 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4855 len
= size
& ~PAGE_CACHE_MASK
;
4857 len
= PAGE_CACHE_SIZE
;
4859 * Return if we have all the buffers mapped. This avoids the need to do
4860 * journal_start/journal_stop which can block and take a long time
4862 if (page_has_buffers(page
)) {
4863 if (!ext4_walk_page_buffers(NULL
, page_buffers(page
),
4865 ext4_bh_unmapped
)) {
4866 /* Wait so that we don't change page under IO */
4867 wait_on_page_writeback(page
);
4868 ret
= VM_FAULT_LOCKED
;
4873 /* OK, we need to fill the hole... */
4874 if (ext4_should_dioread_nolock(inode
))
4875 get_block
= ext4_get_block_write
;
4877 get_block
= ext4_get_block
;
4879 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
4880 if (IS_ERR(handle
)) {
4881 ret
= VM_FAULT_SIGBUS
;
4884 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
4885 if (!ret
&& ext4_should_journal_data(inode
)) {
4886 if (ext4_walk_page_buffers(handle
, page_buffers(page
), 0,
4887 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
4889 ret
= VM_FAULT_SIGBUS
;
4890 ext4_journal_stop(handle
);
4893 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
4895 ext4_journal_stop(handle
);
4896 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
4899 ret
= block_page_mkwrite_return(ret
);
4901 sb_end_pagefault(inode
->i_sb
);