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_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
138 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
139 static int __ext4_journalled_writepage(struct page
*page
, unsigned int len
);
140 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
);
141 static int ext4_discard_partial_page_buffers_no_lock(handle_t
*handle
,
142 struct inode
*inode
, struct page
*page
, loff_t from
,
143 loff_t length
, int flags
);
146 * Test whether an inode is a fast symlink.
148 static int ext4_inode_is_fast_symlink(struct inode
*inode
)
150 int ea_blocks
= EXT4_I(inode
)->i_file_acl
?
151 (inode
->i_sb
->s_blocksize
>> 9) : 0;
153 return (S_ISLNK(inode
->i_mode
) && inode
->i_blocks
- ea_blocks
== 0);
157 * Restart the transaction associated with *handle. This does a commit,
158 * so before we call here everything must be consistently dirtied against
161 int ext4_truncate_restart_trans(handle_t
*handle
, struct inode
*inode
,
167 * Drop i_data_sem to avoid deadlock with ext4_map_blocks. At this
168 * moment, get_block can be called only for blocks inside i_size since
169 * page cache has been already dropped and writes are blocked by
170 * i_mutex. So we can safely drop the i_data_sem here.
172 BUG_ON(EXT4_JOURNAL(inode
) == NULL
);
173 jbd_debug(2, "restarting handle %p\n", handle
);
174 up_write(&EXT4_I(inode
)->i_data_sem
);
175 ret
= ext4_journal_restart(handle
, nblocks
);
176 down_write(&EXT4_I(inode
)->i_data_sem
);
177 ext4_discard_preallocations(inode
);
183 * Called at the last iput() if i_nlink is zero.
185 void ext4_evict_inode(struct inode
*inode
)
190 trace_ext4_evict_inode(inode
);
192 ext4_ioend_wait(inode
);
194 if (inode
->i_nlink
) {
196 * When journalling data dirty buffers are tracked only in the
197 * journal. So although mm thinks everything is clean and
198 * ready for reaping the inode might still have some pages to
199 * write in the running transaction or waiting to be
200 * checkpointed. Thus calling jbd2_journal_invalidatepage()
201 * (via truncate_inode_pages()) to discard these buffers can
202 * cause data loss. Also even if we did not discard these
203 * buffers, we would have no way to find them after the inode
204 * is reaped and thus user could see stale data if he tries to
205 * read them before the transaction is checkpointed. So be
206 * careful and force everything to disk here... We use
207 * ei->i_datasync_tid to store the newest transaction
208 * containing inode's data.
210 * Note that directories do not have this problem because they
211 * don't use page cache.
213 if (ext4_should_journal_data(inode
) &&
214 (S_ISLNK(inode
->i_mode
) || S_ISREG(inode
->i_mode
))) {
215 journal_t
*journal
= EXT4_SB(inode
->i_sb
)->s_journal
;
216 tid_t commit_tid
= EXT4_I(inode
)->i_datasync_tid
;
218 jbd2_log_start_commit(journal
, commit_tid
);
219 jbd2_log_wait_commit(journal
, commit_tid
);
220 filemap_write_and_wait(&inode
->i_data
);
222 truncate_inode_pages(&inode
->i_data
, 0);
226 if (!is_bad_inode(inode
))
227 dquot_initialize(inode
);
229 if (ext4_should_order_data(inode
))
230 ext4_begin_ordered_truncate(inode
, 0);
231 truncate_inode_pages(&inode
->i_data
, 0);
233 if (is_bad_inode(inode
))
237 * Protect us against freezing - iput() caller didn't have to have any
238 * protection against it
240 sb_start_intwrite(inode
->i_sb
);
241 handle
= ext4_journal_start(inode
, ext4_blocks_for_truncate(inode
)+3);
242 if (IS_ERR(handle
)) {
243 ext4_std_error(inode
->i_sb
, PTR_ERR(handle
));
245 * If we're going to skip the normal cleanup, we still need to
246 * make sure that the in-core orphan linked list is properly
249 ext4_orphan_del(NULL
, inode
);
250 sb_end_intwrite(inode
->i_sb
);
255 ext4_handle_sync(handle
);
257 err
= ext4_mark_inode_dirty(handle
, inode
);
259 ext4_warning(inode
->i_sb
,
260 "couldn't mark inode dirty (err %d)", err
);
264 ext4_truncate(inode
);
267 * ext4_ext_truncate() doesn't reserve any slop when it
268 * restarts journal transactions; therefore there may not be
269 * enough credits left in the handle to remove the inode from
270 * the orphan list and set the dtime field.
272 if (!ext4_handle_has_enough_credits(handle
, 3)) {
273 err
= ext4_journal_extend(handle
, 3);
275 err
= ext4_journal_restart(handle
, 3);
277 ext4_warning(inode
->i_sb
,
278 "couldn't extend journal (err %d)", err
);
280 ext4_journal_stop(handle
);
281 ext4_orphan_del(NULL
, inode
);
282 sb_end_intwrite(inode
->i_sb
);
288 * Kill off the orphan record which ext4_truncate created.
289 * AKPM: I think this can be inside the above `if'.
290 * Note that ext4_orphan_del() has to be able to cope with the
291 * deletion of a non-existent orphan - this is because we don't
292 * know if ext4_truncate() actually created an orphan record.
293 * (Well, we could do this if we need to, but heck - it works)
295 ext4_orphan_del(handle
, inode
);
296 EXT4_I(inode
)->i_dtime
= get_seconds();
299 * One subtle ordering requirement: if anything has gone wrong
300 * (transaction abort, IO errors, whatever), then we can still
301 * do these next steps (the fs will already have been marked as
302 * having errors), but we can't free the inode if the mark_dirty
305 if (ext4_mark_inode_dirty(handle
, inode
))
306 /* If that failed, just do the required in-core inode clear. */
307 ext4_clear_inode(inode
);
309 ext4_free_inode(handle
, inode
);
310 ext4_journal_stop(handle
);
311 sb_end_intwrite(inode
->i_sb
);
314 ext4_clear_inode(inode
); /* We must guarantee clearing of inode... */
318 qsize_t
*ext4_get_reserved_space(struct inode
*inode
)
320 return &EXT4_I(inode
)->i_reserved_quota
;
325 * Calculate the number of metadata blocks need to reserve
326 * to allocate a block located at @lblock
328 static int ext4_calc_metadata_amount(struct inode
*inode
, ext4_lblk_t lblock
)
330 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
331 return ext4_ext_calc_metadata_amount(inode
, lblock
);
333 return ext4_ind_calc_metadata_amount(inode
, lblock
);
337 * Called with i_data_sem down, which is important since we can call
338 * ext4_discard_preallocations() from here.
340 void ext4_da_update_reserve_space(struct inode
*inode
,
341 int used
, int quota_claim
)
343 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
344 struct ext4_inode_info
*ei
= EXT4_I(inode
);
346 spin_lock(&ei
->i_block_reservation_lock
);
347 trace_ext4_da_update_reserve_space(inode
, used
, quota_claim
);
348 if (unlikely(used
> ei
->i_reserved_data_blocks
)) {
349 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "%s: ino %lu, used %d "
350 "with only %d reserved data blocks",
351 __func__
, inode
->i_ino
, used
,
352 ei
->i_reserved_data_blocks
);
354 used
= ei
->i_reserved_data_blocks
;
357 if (unlikely(ei
->i_allocated_meta_blocks
> ei
->i_reserved_meta_blocks
)) {
358 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "%s: ino %lu, allocated %d "
359 "with only %d reserved metadata blocks\n", __func__
,
360 inode
->i_ino
, ei
->i_allocated_meta_blocks
,
361 ei
->i_reserved_meta_blocks
);
363 ei
->i_allocated_meta_blocks
= ei
->i_reserved_meta_blocks
;
366 /* Update per-inode reservations */
367 ei
->i_reserved_data_blocks
-= used
;
368 ei
->i_reserved_meta_blocks
-= ei
->i_allocated_meta_blocks
;
369 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
370 used
+ ei
->i_allocated_meta_blocks
);
371 ei
->i_allocated_meta_blocks
= 0;
373 if (ei
->i_reserved_data_blocks
== 0) {
375 * We can release all of the reserved metadata blocks
376 * only when we have written all of the delayed
379 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
380 ei
->i_reserved_meta_blocks
);
381 ei
->i_reserved_meta_blocks
= 0;
382 ei
->i_da_metadata_calc_len
= 0;
384 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
386 /* Update quota subsystem for data blocks */
388 dquot_claim_block(inode
, EXT4_C2B(sbi
, used
));
391 * We did fallocate with an offset that is already delayed
392 * allocated. So on delayed allocated writeback we should
393 * not re-claim the quota for fallocated blocks.
395 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, used
));
399 * If we have done all the pending block allocations and if
400 * there aren't any writers on the inode, we can discard the
401 * inode's preallocations.
403 if ((ei
->i_reserved_data_blocks
== 0) &&
404 (atomic_read(&inode
->i_writecount
) == 0))
405 ext4_discard_preallocations(inode
);
408 static int __check_block_validity(struct inode
*inode
, const char *func
,
410 struct ext4_map_blocks
*map
)
412 if (!ext4_data_block_valid(EXT4_SB(inode
->i_sb
), map
->m_pblk
,
414 ext4_error_inode(inode
, func
, line
, map
->m_pblk
,
415 "lblock %lu mapped to illegal pblock "
416 "(length %d)", (unsigned long) map
->m_lblk
,
423 #define check_block_validity(inode, map) \
424 __check_block_validity((inode), __func__, __LINE__, (map))
427 * Return the number of contiguous dirty pages in a given inode
428 * starting at page frame idx.
430 static pgoff_t
ext4_num_dirty_pages(struct inode
*inode
, pgoff_t idx
,
431 unsigned int max_pages
)
433 struct address_space
*mapping
= inode
->i_mapping
;
437 int i
, nr_pages
, done
= 0;
441 pagevec_init(&pvec
, 0);
444 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
446 (pgoff_t
)PAGEVEC_SIZE
);
449 for (i
= 0; i
< nr_pages
; i
++) {
450 struct page
*page
= pvec
.pages
[i
];
451 struct buffer_head
*bh
, *head
;
454 if (unlikely(page
->mapping
!= mapping
) ||
456 PageWriteback(page
) ||
457 page
->index
!= idx
) {
462 if (page_has_buffers(page
)) {
463 bh
= head
= page_buffers(page
);
465 if (!buffer_delay(bh
) &&
466 !buffer_unwritten(bh
))
468 bh
= bh
->b_this_page
;
469 } while (!done
&& (bh
!= head
));
476 if (num
>= max_pages
) {
481 pagevec_release(&pvec
);
487 * Sets the BH_Da_Mapped bit on the buffer heads corresponding to the given map.
489 static void set_buffers_da_mapped(struct inode
*inode
,
490 struct ext4_map_blocks
*map
)
492 struct address_space
*mapping
= inode
->i_mapping
;
497 index
= map
->m_lblk
>> (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
498 end
= (map
->m_lblk
+ map
->m_len
- 1) >>
499 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
501 pagevec_init(&pvec
, 0);
502 while (index
<= end
) {
503 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
,
505 (pgoff_t
)PAGEVEC_SIZE
));
508 for (i
= 0; i
< nr_pages
; i
++) {
509 struct page
*page
= pvec
.pages
[i
];
510 struct buffer_head
*bh
, *head
;
512 if (unlikely(page
->mapping
!= mapping
) ||
516 if (page_has_buffers(page
)) {
517 bh
= head
= page_buffers(page
);
519 set_buffer_da_mapped(bh
);
520 bh
= bh
->b_this_page
;
521 } while (bh
!= head
);
525 pagevec_release(&pvec
);
530 * The ext4_map_blocks() function tries to look up the requested blocks,
531 * and returns if the blocks are already mapped.
533 * Otherwise it takes the write lock of the i_data_sem and allocate blocks
534 * and store the allocated blocks in the result buffer head and mark it
537 * If file type is extents based, it will call ext4_ext_map_blocks(),
538 * Otherwise, call with ext4_ind_map_blocks() to handle indirect mapping
541 * On success, it returns the number of blocks being mapped or allocate.
542 * if create==0 and the blocks are pre-allocated and uninitialized block,
543 * the result buffer head is unmapped. If the create ==1, it will make sure
544 * the buffer head is mapped.
546 * It returns 0 if plain look up failed (blocks have not been allocated), in
547 * that case, buffer head is unmapped
549 * It returns the error in case of allocation failure.
551 int ext4_map_blocks(handle_t
*handle
, struct inode
*inode
,
552 struct ext4_map_blocks
*map
, int flags
)
557 ext_debug("ext4_map_blocks(): inode %lu, flag %d, max_blocks %u,"
558 "logical block %lu\n", inode
->i_ino
, flags
, map
->m_len
,
559 (unsigned long) map
->m_lblk
);
561 * Try to see if we can get the block without requesting a new
564 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
565 down_read((&EXT4_I(inode
)->i_data_sem
));
566 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
567 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
&
568 EXT4_GET_BLOCKS_KEEP_SIZE
);
570 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
&
571 EXT4_GET_BLOCKS_KEEP_SIZE
);
573 if (!(flags
& EXT4_GET_BLOCKS_NO_LOCK
))
574 up_read((&EXT4_I(inode
)->i_data_sem
));
576 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
577 int ret
= check_block_validity(inode
, map
);
582 /* If it is only a block(s) look up */
583 if ((flags
& EXT4_GET_BLOCKS_CREATE
) == 0)
587 * Returns if the blocks have already allocated
589 * Note that if blocks have been preallocated
590 * ext4_ext_get_block() returns the create = 0
591 * with buffer head unmapped.
593 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
597 * When we call get_blocks without the create flag, the
598 * BH_Unwritten flag could have gotten set if the blocks
599 * requested were part of a uninitialized extent. We need to
600 * clear this flag now that we are committed to convert all or
601 * part of the uninitialized extent to be an initialized
602 * extent. This is because we need to avoid the combination
603 * of BH_Unwritten and BH_Mapped flags being simultaneously
604 * set on the buffer_head.
606 map
->m_flags
&= ~EXT4_MAP_UNWRITTEN
;
609 * New blocks allocate and/or writing to uninitialized extent
610 * will possibly result in updating i_data, so we take
611 * the write lock of i_data_sem, and call get_blocks()
612 * with create == 1 flag.
614 down_write((&EXT4_I(inode
)->i_data_sem
));
617 * if the caller is from delayed allocation writeout path
618 * we have already reserved fs blocks for allocation
619 * let the underlying get_block() function know to
620 * avoid double accounting
622 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
)
623 ext4_set_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
625 * We need to check for EXT4 here because migrate
626 * could have changed the inode type in between
628 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
629 retval
= ext4_ext_map_blocks(handle
, inode
, map
, flags
);
631 retval
= ext4_ind_map_blocks(handle
, inode
, map
, flags
);
633 if (retval
> 0 && map
->m_flags
& EXT4_MAP_NEW
) {
635 * We allocated new blocks which will result in
636 * i_data's format changing. Force the migrate
637 * to fail by clearing migrate flags
639 ext4_clear_inode_state(inode
, EXT4_STATE_EXT_MIGRATE
);
643 * Update reserved blocks/metadata blocks after successful
644 * block allocation which had been deferred till now. We don't
645 * support fallocate for non extent files. So we can update
646 * reserve space here.
649 (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
))
650 ext4_da_update_reserve_space(inode
, retval
, 1);
652 if (flags
& EXT4_GET_BLOCKS_DELALLOC_RESERVE
) {
653 ext4_clear_inode_state(inode
, EXT4_STATE_DELALLOC_RESERVED
);
655 /* If we have successfully mapped the delayed allocated blocks,
656 * set the BH_Da_Mapped bit on them. Its important to do this
657 * under the protection of i_data_sem.
659 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
)
660 set_buffers_da_mapped(inode
, map
);
663 up_write((&EXT4_I(inode
)->i_data_sem
));
664 if (retval
> 0 && map
->m_flags
& EXT4_MAP_MAPPED
) {
665 int ret
= check_block_validity(inode
, map
);
672 /* Maximum number of blocks we map for direct IO at once. */
673 #define DIO_MAX_BLOCKS 4096
675 static int _ext4_get_block(struct inode
*inode
, sector_t iblock
,
676 struct buffer_head
*bh
, int flags
)
678 handle_t
*handle
= ext4_journal_current_handle();
679 struct ext4_map_blocks map
;
680 int ret
= 0, started
= 0;
684 map
.m_len
= bh
->b_size
>> inode
->i_blkbits
;
686 if (flags
&& !handle
) {
687 /* Direct IO write... */
688 if (map
.m_len
> DIO_MAX_BLOCKS
)
689 map
.m_len
= DIO_MAX_BLOCKS
;
690 dio_credits
= ext4_chunk_trans_blocks(inode
, map
.m_len
);
691 handle
= ext4_journal_start(inode
, dio_credits
);
692 if (IS_ERR(handle
)) {
693 ret
= PTR_ERR(handle
);
699 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
701 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
702 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
703 bh
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
707 ext4_journal_stop(handle
);
711 int ext4_get_block(struct inode
*inode
, sector_t iblock
,
712 struct buffer_head
*bh
, int create
)
714 return _ext4_get_block(inode
, iblock
, bh
,
715 create
? EXT4_GET_BLOCKS_CREATE
: 0);
719 * `handle' can be NULL if create is zero
721 struct buffer_head
*ext4_getblk(handle_t
*handle
, struct inode
*inode
,
722 ext4_lblk_t block
, int create
, int *errp
)
724 struct ext4_map_blocks map
;
725 struct buffer_head
*bh
;
728 J_ASSERT(handle
!= NULL
|| create
== 0);
732 err
= ext4_map_blocks(handle
, inode
, &map
,
733 create
? EXT4_GET_BLOCKS_CREATE
: 0);
741 bh
= sb_getblk(inode
->i_sb
, map
.m_pblk
);
746 if (map
.m_flags
& EXT4_MAP_NEW
) {
747 J_ASSERT(create
!= 0);
748 J_ASSERT(handle
!= NULL
);
751 * Now that we do not always journal data, we should
752 * keep in mind whether this should always journal the
753 * new buffer as metadata. For now, regular file
754 * writes use ext4_get_block instead, so it's not a
758 BUFFER_TRACE(bh
, "call get_create_access");
759 fatal
= ext4_journal_get_create_access(handle
, bh
);
760 if (!fatal
&& !buffer_uptodate(bh
)) {
761 memset(bh
->b_data
, 0, inode
->i_sb
->s_blocksize
);
762 set_buffer_uptodate(bh
);
765 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
766 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
770 BUFFER_TRACE(bh
, "not a new buffer");
780 struct buffer_head
*ext4_bread(handle_t
*handle
, struct inode
*inode
,
781 ext4_lblk_t block
, int create
, int *err
)
783 struct buffer_head
*bh
;
785 bh
= ext4_getblk(handle
, inode
, block
, create
, err
);
788 if (buffer_uptodate(bh
))
790 ll_rw_block(READ
| REQ_META
| REQ_PRIO
, 1, &bh
);
792 if (buffer_uptodate(bh
))
799 static int walk_page_buffers(handle_t
*handle
,
800 struct buffer_head
*head
,
804 int (*fn
)(handle_t
*handle
,
805 struct buffer_head
*bh
))
807 struct buffer_head
*bh
;
808 unsigned block_start
, block_end
;
809 unsigned blocksize
= head
->b_size
;
811 struct buffer_head
*next
;
813 for (bh
= head
, block_start
= 0;
814 ret
== 0 && (bh
!= head
|| !block_start
);
815 block_start
= block_end
, bh
= next
) {
816 next
= bh
->b_this_page
;
817 block_end
= block_start
+ blocksize
;
818 if (block_end
<= from
|| block_start
>= to
) {
819 if (partial
&& !buffer_uptodate(bh
))
823 err
= (*fn
)(handle
, bh
);
831 * To preserve ordering, it is essential that the hole instantiation and
832 * the data write be encapsulated in a single transaction. We cannot
833 * close off a transaction and start a new one between the ext4_get_block()
834 * and the commit_write(). So doing the jbd2_journal_start at the start of
835 * prepare_write() is the right place.
837 * Also, this function can nest inside ext4_writepage() ->
838 * block_write_full_page(). In that case, we *know* that ext4_writepage()
839 * has generated enough buffer credits to do the whole page. So we won't
840 * block on the journal in that case, which is good, because the caller may
843 * By accident, ext4 can be reentered when a transaction is open via
844 * quota file writes. If we were to commit the transaction while thus
845 * reentered, there can be a deadlock - we would be holding a quota
846 * lock, and the commit would never complete if another thread had a
847 * transaction open and was blocking on the quota lock - a ranking
850 * So what we do is to rely on the fact that jbd2_journal_stop/journal_start
851 * will _not_ run commit under these circumstances because handle->h_ref
852 * is elevated. We'll still have enough credits for the tiny quotafile
855 static int do_journal_get_write_access(handle_t
*handle
,
856 struct buffer_head
*bh
)
858 int dirty
= buffer_dirty(bh
);
861 if (!buffer_mapped(bh
) || buffer_freed(bh
))
864 * __block_write_begin() could have dirtied some buffers. Clean
865 * the dirty bit as jbd2_journal_get_write_access() could complain
866 * otherwise about fs integrity issues. Setting of the dirty bit
867 * by __block_write_begin() isn't a real problem here as we clear
868 * the bit before releasing a page lock and thus writeback cannot
869 * ever write the buffer.
872 clear_buffer_dirty(bh
);
873 ret
= ext4_journal_get_write_access(handle
, bh
);
875 ret
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
879 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
880 struct buffer_head
*bh_result
, int create
);
881 static int ext4_write_begin(struct file
*file
, struct address_space
*mapping
,
882 loff_t pos
, unsigned len
, unsigned flags
,
883 struct page
**pagep
, void **fsdata
)
885 struct inode
*inode
= mapping
->host
;
886 int ret
, needed_blocks
;
893 trace_ext4_write_begin(inode
, pos
, len
, flags
);
895 * Reserve one block more for addition to orphan list in case
896 * we allocate blocks but write fails for some reason
898 needed_blocks
= ext4_writepage_trans_blocks(inode
) + 1;
899 index
= pos
>> PAGE_CACHE_SHIFT
;
900 from
= pos
& (PAGE_CACHE_SIZE
- 1);
904 handle
= ext4_journal_start(inode
, needed_blocks
);
905 if (IS_ERR(handle
)) {
906 ret
= PTR_ERR(handle
);
910 /* We cannot recurse into the filesystem as the transaction is already
912 flags
|= AOP_FLAG_NOFS
;
914 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
916 ext4_journal_stop(handle
);
922 if (ext4_should_dioread_nolock(inode
))
923 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block_write
);
925 ret
= __block_write_begin(page
, pos
, len
, ext4_get_block
);
927 if (!ret
&& ext4_should_journal_data(inode
)) {
928 ret
= walk_page_buffers(handle
, page_buffers(page
),
929 from
, to
, NULL
, do_journal_get_write_access
);
934 page_cache_release(page
);
936 * __block_write_begin may have instantiated a few blocks
937 * outside i_size. Trim these off again. Don't need
938 * i_size_read because we hold i_mutex.
940 * Add inode to orphan list in case we crash before
943 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
944 ext4_orphan_add(handle
, inode
);
946 ext4_journal_stop(handle
);
947 if (pos
+ len
> inode
->i_size
) {
948 ext4_truncate_failed_write(inode
);
950 * If truncate failed early the inode might
951 * still be on the orphan list; we need to
952 * make sure the inode is removed from the
953 * orphan list in that case.
956 ext4_orphan_del(NULL
, inode
);
960 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
966 /* For write_end() in data=journal mode */
967 static int write_end_fn(handle_t
*handle
, struct buffer_head
*bh
)
969 if (!buffer_mapped(bh
) || buffer_freed(bh
))
971 set_buffer_uptodate(bh
);
972 return ext4_handle_dirty_metadata(handle
, NULL
, bh
);
975 static int ext4_generic_write_end(struct file
*file
,
976 struct address_space
*mapping
,
977 loff_t pos
, unsigned len
, unsigned copied
,
978 struct page
*page
, void *fsdata
)
980 int i_size_changed
= 0;
981 struct inode
*inode
= mapping
->host
;
982 handle_t
*handle
= ext4_journal_current_handle();
984 copied
= block_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
987 * No need to use i_size_read() here, the i_size
988 * cannot change under us because we hold i_mutex.
990 * But it's important to update i_size while still holding page lock:
991 * page writeout could otherwise come in and zero beyond i_size.
993 if (pos
+ copied
> inode
->i_size
) {
994 i_size_write(inode
, pos
+ copied
);
998 if (pos
+ copied
> EXT4_I(inode
)->i_disksize
) {
999 /* We need to mark inode dirty even if
1000 * new_i_size is less that inode->i_size
1001 * bu greater than i_disksize.(hint delalloc)
1003 ext4_update_i_disksize(inode
, (pos
+ copied
));
1007 page_cache_release(page
);
1010 * Don't mark the inode dirty under page lock. First, it unnecessarily
1011 * makes the holding time of page lock longer. Second, it forces lock
1012 * ordering of page lock and transaction start for journaling
1016 ext4_mark_inode_dirty(handle
, inode
);
1022 * We need to pick up the new inode size which generic_commit_write gave us
1023 * `file' can be NULL - eg, when called from page_symlink().
1025 * ext4 never places buffers on inode->i_mapping->private_list. metadata
1026 * buffers are managed internally.
1028 static int ext4_ordered_write_end(struct file
*file
,
1029 struct address_space
*mapping
,
1030 loff_t pos
, unsigned len
, unsigned copied
,
1031 struct page
*page
, void *fsdata
)
1033 handle_t
*handle
= ext4_journal_current_handle();
1034 struct inode
*inode
= mapping
->host
;
1037 trace_ext4_ordered_write_end(inode
, pos
, len
, copied
);
1038 ret
= ext4_jbd2_file_inode(handle
, inode
);
1041 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1044 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1045 /* if we have allocated more blocks and copied
1046 * less. We will have blocks allocated outside
1047 * inode->i_size. So truncate them
1049 ext4_orphan_add(handle
, inode
);
1054 page_cache_release(page
);
1057 ret2
= ext4_journal_stop(handle
);
1061 if (pos
+ len
> inode
->i_size
) {
1062 ext4_truncate_failed_write(inode
);
1064 * If truncate failed early the inode might still be
1065 * on the orphan list; we need to make sure the inode
1066 * is removed from the orphan list in that case.
1069 ext4_orphan_del(NULL
, inode
);
1073 return ret
? ret
: copied
;
1076 static int ext4_writeback_write_end(struct file
*file
,
1077 struct address_space
*mapping
,
1078 loff_t pos
, unsigned len
, unsigned copied
,
1079 struct page
*page
, void *fsdata
)
1081 handle_t
*handle
= ext4_journal_current_handle();
1082 struct inode
*inode
= mapping
->host
;
1085 trace_ext4_writeback_write_end(inode
, pos
, len
, copied
);
1086 ret2
= ext4_generic_write_end(file
, mapping
, pos
, len
, copied
,
1089 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1090 /* if we have allocated more blocks and copied
1091 * less. We will have blocks allocated outside
1092 * inode->i_size. So truncate them
1094 ext4_orphan_add(handle
, inode
);
1099 ret2
= ext4_journal_stop(handle
);
1103 if (pos
+ len
> inode
->i_size
) {
1104 ext4_truncate_failed_write(inode
);
1106 * If truncate failed early the inode might still be
1107 * on the orphan list; we need to make sure the inode
1108 * is removed from the orphan list in that case.
1111 ext4_orphan_del(NULL
, inode
);
1114 return ret
? ret
: copied
;
1117 static int ext4_journalled_write_end(struct file
*file
,
1118 struct address_space
*mapping
,
1119 loff_t pos
, unsigned len
, unsigned copied
,
1120 struct page
*page
, void *fsdata
)
1122 handle_t
*handle
= ext4_journal_current_handle();
1123 struct inode
*inode
= mapping
->host
;
1129 trace_ext4_journalled_write_end(inode
, pos
, len
, copied
);
1130 from
= pos
& (PAGE_CACHE_SIZE
- 1);
1133 BUG_ON(!ext4_handle_valid(handle
));
1136 if (!PageUptodate(page
))
1138 page_zero_new_buffers(page
, from
+copied
, to
);
1141 ret
= walk_page_buffers(handle
, page_buffers(page
), from
,
1142 to
, &partial
, write_end_fn
);
1144 SetPageUptodate(page
);
1145 new_i_size
= pos
+ copied
;
1146 if (new_i_size
> inode
->i_size
)
1147 i_size_write(inode
, pos
+copied
);
1148 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1149 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1150 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
1151 ext4_update_i_disksize(inode
, new_i_size
);
1152 ret2
= ext4_mark_inode_dirty(handle
, inode
);
1158 page_cache_release(page
);
1159 if (pos
+ len
> inode
->i_size
&& ext4_can_truncate(inode
))
1160 /* if we have allocated more blocks and copied
1161 * less. We will have blocks allocated outside
1162 * inode->i_size. So truncate them
1164 ext4_orphan_add(handle
, inode
);
1166 ret2
= ext4_journal_stop(handle
);
1169 if (pos
+ len
> inode
->i_size
) {
1170 ext4_truncate_failed_write(inode
);
1172 * If truncate failed early the inode might still be
1173 * on the orphan list; we need to make sure the inode
1174 * is removed from the orphan list in that case.
1177 ext4_orphan_del(NULL
, inode
);
1180 return ret
? ret
: copied
;
1184 * Reserve a single cluster located at lblock
1186 static int ext4_da_reserve_space(struct inode
*inode
, ext4_lblk_t lblock
)
1189 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1190 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1191 unsigned int md_needed
;
1193 ext4_lblk_t save_last_lblock
;
1197 * We will charge metadata quota at writeout time; this saves
1198 * us from metadata over-estimation, though we may go over by
1199 * a small amount in the end. Here we just reserve for data.
1201 ret
= dquot_reserve_block(inode
, EXT4_C2B(sbi
, 1));
1206 * recalculate the amount of metadata blocks to reserve
1207 * in order to allocate nrblocks
1208 * worse case is one extent per block
1211 spin_lock(&ei
->i_block_reservation_lock
);
1213 * ext4_calc_metadata_amount() has side effects, which we have
1214 * to be prepared undo if we fail to claim space.
1216 save_len
= ei
->i_da_metadata_calc_len
;
1217 save_last_lblock
= ei
->i_da_metadata_calc_last_lblock
;
1218 md_needed
= EXT4_NUM_B2C(sbi
,
1219 ext4_calc_metadata_amount(inode
, lblock
));
1220 trace_ext4_da_reserve_space(inode
, md_needed
);
1223 * We do still charge estimated metadata to the sb though;
1224 * we cannot afford to run out of free blocks.
1226 if (ext4_claim_free_clusters(sbi
, md_needed
+ 1, 0)) {
1227 ei
->i_da_metadata_calc_len
= save_len
;
1228 ei
->i_da_metadata_calc_last_lblock
= save_last_lblock
;
1229 spin_unlock(&ei
->i_block_reservation_lock
);
1230 if (ext4_should_retry_alloc(inode
->i_sb
, &retries
)) {
1234 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, 1));
1237 ei
->i_reserved_data_blocks
++;
1238 ei
->i_reserved_meta_blocks
+= md_needed
;
1239 spin_unlock(&ei
->i_block_reservation_lock
);
1241 return 0; /* success */
1244 static void ext4_da_release_space(struct inode
*inode
, int to_free
)
1246 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1247 struct ext4_inode_info
*ei
= EXT4_I(inode
);
1250 return; /* Nothing to release, exit */
1252 spin_lock(&EXT4_I(inode
)->i_block_reservation_lock
);
1254 trace_ext4_da_release_space(inode
, to_free
);
1255 if (unlikely(to_free
> ei
->i_reserved_data_blocks
)) {
1257 * if there aren't enough reserved blocks, then the
1258 * counter is messed up somewhere. Since this
1259 * function is called from invalidate page, it's
1260 * harmless to return without any action.
1262 ext4_msg(inode
->i_sb
, KERN_NOTICE
, "ext4_da_release_space: "
1263 "ino %lu, to_free %d with only %d reserved "
1264 "data blocks", inode
->i_ino
, to_free
,
1265 ei
->i_reserved_data_blocks
);
1267 to_free
= ei
->i_reserved_data_blocks
;
1269 ei
->i_reserved_data_blocks
-= to_free
;
1271 if (ei
->i_reserved_data_blocks
== 0) {
1273 * We can release all of the reserved metadata blocks
1274 * only when we have written all of the delayed
1275 * allocation blocks.
1276 * Note that in case of bigalloc, i_reserved_meta_blocks,
1277 * i_reserved_data_blocks, etc. refer to number of clusters.
1279 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
,
1280 ei
->i_reserved_meta_blocks
);
1281 ei
->i_reserved_meta_blocks
= 0;
1282 ei
->i_da_metadata_calc_len
= 0;
1285 /* update fs dirty data blocks counter */
1286 percpu_counter_sub(&sbi
->s_dirtyclusters_counter
, to_free
);
1288 spin_unlock(&EXT4_I(inode
)->i_block_reservation_lock
);
1290 dquot_release_reservation_block(inode
, EXT4_C2B(sbi
, to_free
));
1293 static void ext4_da_page_release_reservation(struct page
*page
,
1294 unsigned long offset
)
1297 struct buffer_head
*head
, *bh
;
1298 unsigned int curr_off
= 0;
1299 struct inode
*inode
= page
->mapping
->host
;
1300 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1303 head
= page_buffers(page
);
1306 unsigned int next_off
= curr_off
+ bh
->b_size
;
1308 if ((offset
<= curr_off
) && (buffer_delay(bh
))) {
1310 clear_buffer_delay(bh
);
1311 clear_buffer_da_mapped(bh
);
1313 curr_off
= next_off
;
1314 } while ((bh
= bh
->b_this_page
) != head
);
1316 /* If we have released all the blocks belonging to a cluster, then we
1317 * need to release the reserved space for that cluster. */
1318 num_clusters
= EXT4_NUM_B2C(sbi
, to_release
);
1319 while (num_clusters
> 0) {
1321 lblk
= (page
->index
<< (PAGE_CACHE_SHIFT
- inode
->i_blkbits
)) +
1322 ((num_clusters
- 1) << sbi
->s_cluster_bits
);
1323 if (sbi
->s_cluster_ratio
== 1 ||
1324 !ext4_find_delalloc_cluster(inode
, lblk
, 1))
1325 ext4_da_release_space(inode
, 1);
1332 * Delayed allocation stuff
1336 * mpage_da_submit_io - walks through extent of pages and try to write
1337 * them with writepage() call back
1339 * @mpd->inode: inode
1340 * @mpd->first_page: first page of the extent
1341 * @mpd->next_page: page after the last page of the extent
1343 * By the time mpage_da_submit_io() is called we expect all blocks
1344 * to be allocated. this may be wrong if allocation failed.
1346 * As pages are already locked by write_cache_pages(), we can't use it
1348 static int mpage_da_submit_io(struct mpage_da_data
*mpd
,
1349 struct ext4_map_blocks
*map
)
1351 struct pagevec pvec
;
1352 unsigned long index
, end
;
1353 int ret
= 0, err
, nr_pages
, i
;
1354 struct inode
*inode
= mpd
->inode
;
1355 struct address_space
*mapping
= inode
->i_mapping
;
1356 loff_t size
= i_size_read(inode
);
1357 unsigned int len
, block_start
;
1358 struct buffer_head
*bh
, *page_bufs
= NULL
;
1359 int journal_data
= ext4_should_journal_data(inode
);
1360 sector_t pblock
= 0, cur_logical
= 0;
1361 struct ext4_io_submit io_submit
;
1363 BUG_ON(mpd
->next_page
<= mpd
->first_page
);
1364 memset(&io_submit
, 0, sizeof(io_submit
));
1366 * We need to start from the first_page to the next_page - 1
1367 * to make sure we also write the mapped dirty buffer_heads.
1368 * If we look at mpd->b_blocknr we would only be looking
1369 * at the currently mapped buffer_heads.
1371 index
= mpd
->first_page
;
1372 end
= mpd
->next_page
- 1;
1374 pagevec_init(&pvec
, 0);
1375 while (index
<= end
) {
1376 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1379 for (i
= 0; i
< nr_pages
; i
++) {
1380 int commit_write
= 0, skip_page
= 0;
1381 struct page
*page
= pvec
.pages
[i
];
1383 index
= page
->index
;
1387 if (index
== size
>> PAGE_CACHE_SHIFT
)
1388 len
= size
& ~PAGE_CACHE_MASK
;
1390 len
= PAGE_CACHE_SIZE
;
1392 cur_logical
= index
<< (PAGE_CACHE_SHIFT
-
1394 pblock
= map
->m_pblk
+ (cur_logical
-
1399 BUG_ON(!PageLocked(page
));
1400 BUG_ON(PageWriteback(page
));
1403 * If the page does not have buffers (for
1404 * whatever reason), try to create them using
1405 * __block_write_begin. If this fails,
1406 * skip the page and move on.
1408 if (!page_has_buffers(page
)) {
1409 if (__block_write_begin(page
, 0, len
,
1410 noalloc_get_block_write
)) {
1418 bh
= page_bufs
= page_buffers(page
);
1423 if (map
&& (cur_logical
>= map
->m_lblk
) &&
1424 (cur_logical
<= (map
->m_lblk
+
1425 (map
->m_len
- 1)))) {
1426 if (buffer_delay(bh
)) {
1427 clear_buffer_delay(bh
);
1428 bh
->b_blocknr
= pblock
;
1430 if (buffer_da_mapped(bh
))
1431 clear_buffer_da_mapped(bh
);
1432 if (buffer_unwritten(bh
) ||
1434 BUG_ON(bh
->b_blocknr
!= pblock
);
1435 if (map
->m_flags
& EXT4_MAP_UNINIT
)
1436 set_buffer_uninit(bh
);
1437 clear_buffer_unwritten(bh
);
1441 * skip page if block allocation undone and
1444 if (ext4_bh_delay_or_unwritten(NULL
, bh
))
1446 bh
= bh
->b_this_page
;
1447 block_start
+= bh
->b_size
;
1450 } while (bh
!= page_bufs
);
1456 /* mark the buffer_heads as dirty & uptodate */
1457 block_commit_write(page
, 0, len
);
1459 clear_page_dirty_for_io(page
);
1461 * Delalloc doesn't support data journalling,
1462 * but eventually maybe we'll lift this
1465 if (unlikely(journal_data
&& PageChecked(page
)))
1466 err
= __ext4_journalled_writepage(page
, len
);
1467 else if (test_opt(inode
->i_sb
, MBLK_IO_SUBMIT
))
1468 err
= ext4_bio_write_page(&io_submit
, page
,
1470 else if (buffer_uninit(page_bufs
)) {
1471 ext4_set_bh_endio(page_bufs
, inode
);
1472 err
= block_write_full_page_endio(page
,
1473 noalloc_get_block_write
,
1474 mpd
->wbc
, ext4_end_io_buffer_write
);
1476 err
= block_write_full_page(page
,
1477 noalloc_get_block_write
, mpd
->wbc
);
1480 mpd
->pages_written
++;
1482 * In error case, we have to continue because
1483 * remaining pages are still locked
1488 pagevec_release(&pvec
);
1490 ext4_io_submit(&io_submit
);
1494 static void ext4_da_block_invalidatepages(struct mpage_da_data
*mpd
)
1498 struct pagevec pvec
;
1499 struct inode
*inode
= mpd
->inode
;
1500 struct address_space
*mapping
= inode
->i_mapping
;
1502 index
= mpd
->first_page
;
1503 end
= mpd
->next_page
- 1;
1504 while (index
<= end
) {
1505 nr_pages
= pagevec_lookup(&pvec
, mapping
, index
, PAGEVEC_SIZE
);
1508 for (i
= 0; i
< nr_pages
; i
++) {
1509 struct page
*page
= pvec
.pages
[i
];
1510 if (page
->index
> end
)
1512 BUG_ON(!PageLocked(page
));
1513 BUG_ON(PageWriteback(page
));
1514 block_invalidatepage(page
, 0);
1515 ClearPageUptodate(page
);
1518 index
= pvec
.pages
[nr_pages
- 1]->index
+ 1;
1519 pagevec_release(&pvec
);
1524 static void ext4_print_free_blocks(struct inode
*inode
)
1526 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
1527 struct super_block
*sb
= inode
->i_sb
;
1529 ext4_msg(sb
, KERN_CRIT
, "Total free blocks count %lld",
1530 EXT4_C2B(EXT4_SB(inode
->i_sb
),
1531 ext4_count_free_clusters(inode
->i_sb
)));
1532 ext4_msg(sb
, KERN_CRIT
, "Free/Dirty block details");
1533 ext4_msg(sb
, KERN_CRIT
, "free_blocks=%lld",
1534 (long long) EXT4_C2B(EXT4_SB(inode
->i_sb
),
1535 percpu_counter_sum(&sbi
->s_freeclusters_counter
)));
1536 ext4_msg(sb
, KERN_CRIT
, "dirty_blocks=%lld",
1537 (long long) EXT4_C2B(EXT4_SB(inode
->i_sb
),
1538 percpu_counter_sum(&sbi
->s_dirtyclusters_counter
)));
1539 ext4_msg(sb
, KERN_CRIT
, "Block reservation details");
1540 ext4_msg(sb
, KERN_CRIT
, "i_reserved_data_blocks=%u",
1541 EXT4_I(inode
)->i_reserved_data_blocks
);
1542 ext4_msg(sb
, KERN_CRIT
, "i_reserved_meta_blocks=%u",
1543 EXT4_I(inode
)->i_reserved_meta_blocks
);
1548 * mpage_da_map_and_submit - go through given space, map them
1549 * if necessary, and then submit them for I/O
1551 * @mpd - bh describing space
1553 * The function skips space we know is already mapped to disk blocks.
1556 static void mpage_da_map_and_submit(struct mpage_da_data
*mpd
)
1558 int err
, blks
, get_blocks_flags
;
1559 struct ext4_map_blocks map
, *mapp
= NULL
;
1560 sector_t next
= mpd
->b_blocknr
;
1561 unsigned max_blocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1562 loff_t disksize
= EXT4_I(mpd
->inode
)->i_disksize
;
1563 handle_t
*handle
= NULL
;
1566 * If the blocks are mapped already, or we couldn't accumulate
1567 * any blocks, then proceed immediately to the submission stage.
1569 if ((mpd
->b_size
== 0) ||
1570 ((mpd
->b_state
& (1 << BH_Mapped
)) &&
1571 !(mpd
->b_state
& (1 << BH_Delay
)) &&
1572 !(mpd
->b_state
& (1 << BH_Unwritten
))))
1575 handle
= ext4_journal_current_handle();
1579 * Call ext4_map_blocks() to allocate any delayed allocation
1580 * blocks, or to convert an uninitialized extent to be
1581 * initialized (in the case where we have written into
1582 * one or more preallocated blocks).
1584 * We pass in the magic EXT4_GET_BLOCKS_DELALLOC_RESERVE to
1585 * indicate that we are on the delayed allocation path. This
1586 * affects functions in many different parts of the allocation
1587 * call path. This flag exists primarily because we don't
1588 * want to change *many* call functions, so ext4_map_blocks()
1589 * will set the EXT4_STATE_DELALLOC_RESERVED flag once the
1590 * inode's allocation semaphore is taken.
1592 * If the blocks in questions were delalloc blocks, set
1593 * EXT4_GET_BLOCKS_DELALLOC_RESERVE so the delalloc accounting
1594 * variables are updated after the blocks have been allocated.
1597 map
.m_len
= max_blocks
;
1598 get_blocks_flags
= EXT4_GET_BLOCKS_CREATE
;
1599 if (ext4_should_dioread_nolock(mpd
->inode
))
1600 get_blocks_flags
|= EXT4_GET_BLOCKS_IO_CREATE_EXT
;
1601 if (mpd
->b_state
& (1 << BH_Delay
))
1602 get_blocks_flags
|= EXT4_GET_BLOCKS_DELALLOC_RESERVE
;
1604 blks
= ext4_map_blocks(handle
, mpd
->inode
, &map
, get_blocks_flags
);
1606 struct super_block
*sb
= mpd
->inode
->i_sb
;
1610 * If get block returns EAGAIN or ENOSPC and there
1611 * appears to be free blocks we will just let
1612 * mpage_da_submit_io() unlock all of the pages.
1617 if (err
== -ENOSPC
&& ext4_count_free_clusters(sb
)) {
1623 * get block failure will cause us to loop in
1624 * writepages, because a_ops->writepage won't be able
1625 * to make progress. The page will be redirtied by
1626 * writepage and writepages will again try to write
1629 if (!(EXT4_SB(sb
)->s_mount_flags
& EXT4_MF_FS_ABORTED
)) {
1630 ext4_msg(sb
, KERN_CRIT
,
1631 "delayed block allocation failed for inode %lu "
1632 "at logical offset %llu with max blocks %zd "
1633 "with error %d", mpd
->inode
->i_ino
,
1634 (unsigned long long) next
,
1635 mpd
->b_size
>> mpd
->inode
->i_blkbits
, err
);
1636 ext4_msg(sb
, KERN_CRIT
,
1637 "This should not happen!! Data will be lost\n");
1639 ext4_print_free_blocks(mpd
->inode
);
1641 /* invalidate all the pages */
1642 ext4_da_block_invalidatepages(mpd
);
1644 /* Mark this page range as having been completed */
1651 if (map
.m_flags
& EXT4_MAP_NEW
) {
1652 struct block_device
*bdev
= mpd
->inode
->i_sb
->s_bdev
;
1655 for (i
= 0; i
< map
.m_len
; i
++)
1656 unmap_underlying_metadata(bdev
, map
.m_pblk
+ i
);
1658 if (ext4_should_order_data(mpd
->inode
)) {
1659 err
= ext4_jbd2_file_inode(handle
, mpd
->inode
);
1661 /* Only if the journal is aborted */
1669 * Update on-disk size along with block allocation.
1671 disksize
= ((loff_t
) next
+ blks
) << mpd
->inode
->i_blkbits
;
1672 if (disksize
> i_size_read(mpd
->inode
))
1673 disksize
= i_size_read(mpd
->inode
);
1674 if (disksize
> EXT4_I(mpd
->inode
)->i_disksize
) {
1675 ext4_update_i_disksize(mpd
->inode
, disksize
);
1676 err
= ext4_mark_inode_dirty(handle
, mpd
->inode
);
1678 ext4_error(mpd
->inode
->i_sb
,
1679 "Failed to mark inode %lu dirty",
1684 mpage_da_submit_io(mpd
, mapp
);
1688 #define BH_FLAGS ((1 << BH_Uptodate) | (1 << BH_Mapped) | \
1689 (1 << BH_Delay) | (1 << BH_Unwritten))
1692 * mpage_add_bh_to_extent - try to add one more block to extent of blocks
1694 * @mpd->lbh - extent of blocks
1695 * @logical - logical number of the block in the file
1696 * @bh - bh of the block (used to access block's state)
1698 * the function is used to collect contig. blocks in same state
1700 static void mpage_add_bh_to_extent(struct mpage_da_data
*mpd
,
1701 sector_t logical
, size_t b_size
,
1702 unsigned long b_state
)
1705 int nrblocks
= mpd
->b_size
>> mpd
->inode
->i_blkbits
;
1708 * XXX Don't go larger than mballoc is willing to allocate
1709 * This is a stopgap solution. We eventually need to fold
1710 * mpage_da_submit_io() into this function and then call
1711 * ext4_map_blocks() multiple times in a loop
1713 if (nrblocks
>= 8*1024*1024/mpd
->inode
->i_sb
->s_blocksize
)
1716 /* check if thereserved journal credits might overflow */
1717 if (!(ext4_test_inode_flag(mpd
->inode
, EXT4_INODE_EXTENTS
))) {
1718 if (nrblocks
>= EXT4_MAX_TRANS_DATA
) {
1720 * With non-extent format we are limited by the journal
1721 * credit available. Total credit needed to insert
1722 * nrblocks contiguous blocks is dependent on the
1723 * nrblocks. So limit nrblocks.
1726 } else if ((nrblocks
+ (b_size
>> mpd
->inode
->i_blkbits
)) >
1727 EXT4_MAX_TRANS_DATA
) {
1729 * Adding the new buffer_head would make it cross the
1730 * allowed limit for which we have journal credit
1731 * reserved. So limit the new bh->b_size
1733 b_size
= (EXT4_MAX_TRANS_DATA
- nrblocks
) <<
1734 mpd
->inode
->i_blkbits
;
1735 /* we will do mpage_da_submit_io in the next loop */
1739 * First block in the extent
1741 if (mpd
->b_size
== 0) {
1742 mpd
->b_blocknr
= logical
;
1743 mpd
->b_size
= b_size
;
1744 mpd
->b_state
= b_state
& BH_FLAGS
;
1748 next
= mpd
->b_blocknr
+ nrblocks
;
1750 * Can we merge the block to our big extent?
1752 if (logical
== next
&& (b_state
& BH_FLAGS
) == mpd
->b_state
) {
1753 mpd
->b_size
+= b_size
;
1759 * We couldn't merge the block to our extent, so we
1760 * need to flush current extent and start new one
1762 mpage_da_map_and_submit(mpd
);
1766 static int ext4_bh_delay_or_unwritten(handle_t
*handle
, struct buffer_head
*bh
)
1768 return (buffer_delay(bh
) || buffer_unwritten(bh
)) && buffer_dirty(bh
);
1772 * This function is grabs code from the very beginning of
1773 * ext4_map_blocks, but assumes that the caller is from delayed write
1774 * time. This function looks up the requested blocks and sets the
1775 * buffer delay bit under the protection of i_data_sem.
1777 static int ext4_da_map_blocks(struct inode
*inode
, sector_t iblock
,
1778 struct ext4_map_blocks
*map
,
1779 struct buffer_head
*bh
)
1782 sector_t invalid_block
= ~((sector_t
) 0xffff);
1784 if (invalid_block
< ext4_blocks_count(EXT4_SB(inode
->i_sb
)->s_es
))
1788 ext_debug("ext4_da_map_blocks(): inode %lu, max_blocks %u,"
1789 "logical block %lu\n", inode
->i_ino
, map
->m_len
,
1790 (unsigned long) map
->m_lblk
);
1792 * Try to see if we can get the block without requesting a new
1793 * file system block.
1795 down_read((&EXT4_I(inode
)->i_data_sem
));
1796 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
1797 retval
= ext4_ext_map_blocks(NULL
, inode
, map
, 0);
1799 retval
= ext4_ind_map_blocks(NULL
, inode
, map
, 0);
1803 * XXX: __block_prepare_write() unmaps passed block,
1806 /* If the block was allocated from previously allocated cluster,
1807 * then we dont need to reserve it again. */
1808 if (!(map
->m_flags
& EXT4_MAP_FROM_CLUSTER
)) {
1809 retval
= ext4_da_reserve_space(inode
, iblock
);
1811 /* not enough space to reserve */
1815 /* Clear EXT4_MAP_FROM_CLUSTER flag since its purpose is served
1816 * and it should not appear on the bh->b_state.
1818 map
->m_flags
&= ~EXT4_MAP_FROM_CLUSTER
;
1820 map_bh(bh
, inode
->i_sb
, invalid_block
);
1822 set_buffer_delay(bh
);
1826 up_read((&EXT4_I(inode
)->i_data_sem
));
1832 * This is a special get_blocks_t callback which is used by
1833 * ext4_da_write_begin(). It will either return mapped block or
1834 * reserve space for a single block.
1836 * For delayed buffer_head we have BH_Mapped, BH_New, BH_Delay set.
1837 * We also have b_blocknr = -1 and b_bdev initialized properly
1839 * For unwritten buffer_head we have BH_Mapped, BH_New, BH_Unwritten set.
1840 * We also have b_blocknr = physicalblock mapping unwritten extent and b_bdev
1841 * initialized properly.
1843 static int ext4_da_get_block_prep(struct inode
*inode
, sector_t iblock
,
1844 struct buffer_head
*bh
, int create
)
1846 struct ext4_map_blocks map
;
1849 BUG_ON(create
== 0);
1850 BUG_ON(bh
->b_size
!= inode
->i_sb
->s_blocksize
);
1852 map
.m_lblk
= iblock
;
1856 * first, we need to know whether the block is allocated already
1857 * preallocated blocks are unmapped but should treated
1858 * the same as allocated blocks.
1860 ret
= ext4_da_map_blocks(inode
, iblock
, &map
, bh
);
1864 map_bh(bh
, inode
->i_sb
, map
.m_pblk
);
1865 bh
->b_state
= (bh
->b_state
& ~EXT4_MAP_FLAGS
) | map
.m_flags
;
1867 if (buffer_unwritten(bh
)) {
1868 /* A delayed write to unwritten bh should be marked
1869 * new and mapped. Mapped ensures that we don't do
1870 * get_block multiple times when we write to the same
1871 * offset and new ensures that we do proper zero out
1872 * for partial write.
1875 set_buffer_mapped(bh
);
1881 * This function is used as a standard get_block_t calback function
1882 * when there is no desire to allocate any blocks. It is used as a
1883 * callback function for block_write_begin() and block_write_full_page().
1884 * These functions should only try to map a single block at a time.
1886 * Since this function doesn't do block allocations even if the caller
1887 * requests it by passing in create=1, it is critically important that
1888 * any caller checks to make sure that any buffer heads are returned
1889 * by this function are either all already mapped or marked for
1890 * delayed allocation before calling block_write_full_page(). Otherwise,
1891 * b_blocknr could be left unitialized, and the page write functions will
1892 * be taken by surprise.
1894 static int noalloc_get_block_write(struct inode
*inode
, sector_t iblock
,
1895 struct buffer_head
*bh_result
, int create
)
1897 BUG_ON(bh_result
->b_size
!= inode
->i_sb
->s_blocksize
);
1898 return _ext4_get_block(inode
, iblock
, bh_result
, 0);
1901 static int bget_one(handle_t
*handle
, struct buffer_head
*bh
)
1907 static int bput_one(handle_t
*handle
, struct buffer_head
*bh
)
1913 static int __ext4_journalled_writepage(struct page
*page
,
1916 struct address_space
*mapping
= page
->mapping
;
1917 struct inode
*inode
= mapping
->host
;
1918 struct buffer_head
*page_bufs
;
1919 handle_t
*handle
= NULL
;
1923 ClearPageChecked(page
);
1924 page_bufs
= page_buffers(page
);
1926 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bget_one
);
1927 /* As soon as we unlock the page, it can go away, but we have
1928 * references to buffers so we are safe */
1931 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
1932 if (IS_ERR(handle
)) {
1933 ret
= PTR_ERR(handle
);
1937 BUG_ON(!ext4_handle_valid(handle
));
1939 ret
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1940 do_journal_get_write_access
);
1942 err
= walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
,
1946 EXT4_I(inode
)->i_datasync_tid
= handle
->h_transaction
->t_tid
;
1947 err
= ext4_journal_stop(handle
);
1951 walk_page_buffers(handle
, page_bufs
, 0, len
, NULL
, bput_one
);
1952 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
1957 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
);
1958 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
);
1961 * Note that we don't need to start a transaction unless we're journaling data
1962 * because we should have holes filled from ext4_page_mkwrite(). We even don't
1963 * need to file the inode to the transaction's list in ordered mode because if
1964 * we are writing back data added by write(), the inode is already there and if
1965 * we are writing back data modified via mmap(), no one guarantees in which
1966 * transaction the data will hit the disk. In case we are journaling data, we
1967 * cannot start transaction directly because transaction start ranks above page
1968 * lock so we have to do some magic.
1970 * This function can get called via...
1971 * - ext4_da_writepages after taking page lock (have journal handle)
1972 * - journal_submit_inode_data_buffers (no journal handle)
1973 * - shrink_page_list via pdflush (no journal handle)
1974 * - grab_page_cache when doing write_begin (have journal handle)
1976 * We don't do any block allocation in this function. If we have page with
1977 * multiple blocks we need to write those buffer_heads that are mapped. This
1978 * is important for mmaped based write. So if we do with blocksize 1K
1979 * truncate(f, 1024);
1980 * a = mmap(f, 0, 4096);
1982 * truncate(f, 4096);
1983 * we have in the page first buffer_head mapped via page_mkwrite call back
1984 * but other buffer_heads would be unmapped but dirty (dirty done via the
1985 * do_wp_page). So writepage should write the first block. If we modify
1986 * the mmap area beyond 1024 we will again get a page_fault and the
1987 * page_mkwrite callback will do the block allocation and mark the
1988 * buffer_heads mapped.
1990 * We redirty the page if we have any buffer_heads that is either delay or
1991 * unwritten in the page.
1993 * We can get recursively called as show below.
1995 * ext4_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1998 * But since we don't do any block allocation we should not deadlock.
1999 * Page also have the dirty flag cleared so we don't get recurive page_lock.
2001 static int ext4_writepage(struct page
*page
,
2002 struct writeback_control
*wbc
)
2004 int ret
= 0, commit_write
= 0;
2007 struct buffer_head
*page_bufs
= NULL
;
2008 struct inode
*inode
= page
->mapping
->host
;
2010 trace_ext4_writepage(page
);
2011 size
= i_size_read(inode
);
2012 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
2013 len
= size
& ~PAGE_CACHE_MASK
;
2015 len
= PAGE_CACHE_SIZE
;
2018 * If the page does not have buffers (for whatever reason),
2019 * try to create them using __block_write_begin. If this
2020 * fails, redirty the page and move on.
2022 if (!page_has_buffers(page
)) {
2023 if (__block_write_begin(page
, 0, len
,
2024 noalloc_get_block_write
)) {
2026 redirty_page_for_writepage(wbc
, page
);
2032 page_bufs
= page_buffers(page
);
2033 if (walk_page_buffers(NULL
, page_bufs
, 0, len
, NULL
,
2034 ext4_bh_delay_or_unwritten
)) {
2036 * We don't want to do block allocation, so redirty
2037 * the page and return. We may reach here when we do
2038 * a journal commit via journal_submit_inode_data_buffers.
2039 * We can also reach here via shrink_page_list but it
2040 * should never be for direct reclaim so warn if that
2043 WARN_ON_ONCE((current
->flags
& (PF_MEMALLOC
|PF_KSWAPD
)) ==
2048 /* now mark the buffer_heads as dirty and uptodate */
2049 block_commit_write(page
, 0, len
);
2051 if (PageChecked(page
) && ext4_should_journal_data(inode
))
2053 * It's mmapped pagecache. Add buffers and journal it. There
2054 * doesn't seem much point in redirtying the page here.
2056 return __ext4_journalled_writepage(page
, len
);
2058 if (buffer_uninit(page_bufs
)) {
2059 ext4_set_bh_endio(page_bufs
, inode
);
2060 ret
= block_write_full_page_endio(page
, noalloc_get_block_write
,
2061 wbc
, ext4_end_io_buffer_write
);
2063 ret
= block_write_full_page(page
, noalloc_get_block_write
,
2070 * This is called via ext4_da_writepages() to
2071 * calculate the total number of credits to reserve to fit
2072 * a single extent allocation into a single transaction,
2073 * ext4_da_writpeages() will loop calling this before
2074 * the block allocation.
2077 static int ext4_da_writepages_trans_blocks(struct inode
*inode
)
2079 int max_blocks
= EXT4_I(inode
)->i_reserved_data_blocks
;
2082 * With non-extent format the journal credit needed to
2083 * insert nrblocks contiguous block is dependent on
2084 * number of contiguous block. So we will limit
2085 * number of contiguous block to a sane value
2087 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) &&
2088 (max_blocks
> EXT4_MAX_TRANS_DATA
))
2089 max_blocks
= EXT4_MAX_TRANS_DATA
;
2091 return ext4_chunk_trans_blocks(inode
, max_blocks
);
2095 * write_cache_pages_da - walk the list of dirty pages of the given
2096 * address space and accumulate pages that need writing, and call
2097 * mpage_da_map_and_submit to map a single contiguous memory region
2098 * and then write them.
2100 static int write_cache_pages_da(struct address_space
*mapping
,
2101 struct writeback_control
*wbc
,
2102 struct mpage_da_data
*mpd
,
2103 pgoff_t
*done_index
)
2105 struct buffer_head
*bh
, *head
;
2106 struct inode
*inode
= mapping
->host
;
2107 struct pagevec pvec
;
2108 unsigned int nr_pages
;
2111 long nr_to_write
= wbc
->nr_to_write
;
2112 int i
, tag
, ret
= 0;
2114 memset(mpd
, 0, sizeof(struct mpage_da_data
));
2117 pagevec_init(&pvec
, 0);
2118 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2119 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2121 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2122 tag
= PAGECACHE_TAG_TOWRITE
;
2124 tag
= PAGECACHE_TAG_DIRTY
;
2126 *done_index
= index
;
2127 while (index
<= end
) {
2128 nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
, tag
,
2129 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1);
2133 for (i
= 0; i
< nr_pages
; i
++) {
2134 struct page
*page
= pvec
.pages
[i
];
2137 * At this point, the page may be truncated or
2138 * invalidated (changing page->mapping to NULL), or
2139 * even swizzled back from swapper_space to tmpfs file
2140 * mapping. However, page->index will not change
2141 * because we have a reference on the page.
2143 if (page
->index
> end
)
2146 *done_index
= page
->index
+ 1;
2149 * If we can't merge this page, and we have
2150 * accumulated an contiguous region, write it
2152 if ((mpd
->next_page
!= page
->index
) &&
2153 (mpd
->next_page
!= mpd
->first_page
)) {
2154 mpage_da_map_and_submit(mpd
);
2155 goto ret_extent_tail
;
2161 * If the page is no longer dirty, or its
2162 * mapping no longer corresponds to inode we
2163 * are writing (which means it has been
2164 * truncated or invalidated), or the page is
2165 * already under writeback and we are not
2166 * doing a data integrity writeback, skip the page
2168 if (!PageDirty(page
) ||
2169 (PageWriteback(page
) &&
2170 (wbc
->sync_mode
== WB_SYNC_NONE
)) ||
2171 unlikely(page
->mapping
!= mapping
)) {
2176 wait_on_page_writeback(page
);
2177 BUG_ON(PageWriteback(page
));
2179 if (mpd
->next_page
!= page
->index
)
2180 mpd
->first_page
= page
->index
;
2181 mpd
->next_page
= page
->index
+ 1;
2182 logical
= (sector_t
) page
->index
<<
2183 (PAGE_CACHE_SHIFT
- inode
->i_blkbits
);
2185 if (!page_has_buffers(page
)) {
2186 mpage_add_bh_to_extent(mpd
, logical
,
2188 (1 << BH_Dirty
) | (1 << BH_Uptodate
));
2190 goto ret_extent_tail
;
2193 * Page with regular buffer heads,
2194 * just add all dirty ones
2196 head
= page_buffers(page
);
2199 BUG_ON(buffer_locked(bh
));
2201 * We need to try to allocate
2202 * unmapped blocks in the same page.
2203 * Otherwise we won't make progress
2204 * with the page in ext4_writepage
2206 if (ext4_bh_delay_or_unwritten(NULL
, bh
)) {
2207 mpage_add_bh_to_extent(mpd
, logical
,
2211 goto ret_extent_tail
;
2212 } else if (buffer_dirty(bh
) && (buffer_mapped(bh
))) {
2214 * mapped dirty buffer. We need
2215 * to update the b_state
2216 * because we look at b_state
2217 * in mpage_da_map_blocks. We
2218 * don't update b_size because
2219 * if we find an unmapped
2220 * buffer_head later we need to
2221 * use the b_state flag of that
2224 if (mpd
->b_size
== 0)
2225 mpd
->b_state
= bh
->b_state
& BH_FLAGS
;
2228 } while ((bh
= bh
->b_this_page
) != head
);
2231 if (nr_to_write
> 0) {
2233 if (nr_to_write
== 0 &&
2234 wbc
->sync_mode
== WB_SYNC_NONE
)
2236 * We stop writing back only if we are
2237 * not doing integrity sync. In case of
2238 * integrity sync we have to keep going
2239 * because someone may be concurrently
2240 * dirtying pages, and we might have
2241 * synced a lot of newly appeared dirty
2242 * pages, but have not synced all of the
2248 pagevec_release(&pvec
);
2253 ret
= MPAGE_DA_EXTENT_TAIL
;
2255 pagevec_release(&pvec
);
2261 static int ext4_da_writepages(struct address_space
*mapping
,
2262 struct writeback_control
*wbc
)
2265 int range_whole
= 0;
2266 handle_t
*handle
= NULL
;
2267 struct mpage_da_data mpd
;
2268 struct inode
*inode
= mapping
->host
;
2269 int pages_written
= 0;
2270 unsigned int max_pages
;
2271 int range_cyclic
, cycled
= 1, io_done
= 0;
2272 int needed_blocks
, ret
= 0;
2273 long desired_nr_to_write
, nr_to_writebump
= 0;
2274 loff_t range_start
= wbc
->range_start
;
2275 struct ext4_sb_info
*sbi
= EXT4_SB(mapping
->host
->i_sb
);
2276 pgoff_t done_index
= 0;
2278 struct blk_plug plug
;
2280 trace_ext4_da_writepages(inode
, wbc
);
2283 * No pages to write? This is mainly a kludge to avoid starting
2284 * a transaction for special inodes like journal inode on last iput()
2285 * because that could violate lock ordering on umount
2287 if (!mapping
->nrpages
|| !mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
2291 * If the filesystem has aborted, it is read-only, so return
2292 * right away instead of dumping stack traces later on that
2293 * will obscure the real source of the problem. We test
2294 * EXT4_MF_FS_ABORTED instead of sb->s_flag's MS_RDONLY because
2295 * the latter could be true if the filesystem is mounted
2296 * read-only, and in that case, ext4_da_writepages should
2297 * *never* be called, so if that ever happens, we would want
2300 if (unlikely(sbi
->s_mount_flags
& EXT4_MF_FS_ABORTED
))
2303 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
2306 range_cyclic
= wbc
->range_cyclic
;
2307 if (wbc
->range_cyclic
) {
2308 index
= mapping
->writeback_index
;
2311 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2312 wbc
->range_end
= LLONG_MAX
;
2313 wbc
->range_cyclic
= 0;
2316 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
2317 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
2321 * This works around two forms of stupidity. The first is in
2322 * the writeback code, which caps the maximum number of pages
2323 * written to be 1024 pages. This is wrong on multiple
2324 * levels; different architectues have a different page size,
2325 * which changes the maximum amount of data which gets
2326 * written. Secondly, 4 megabytes is way too small. XFS
2327 * forces this value to be 16 megabytes by multiplying
2328 * nr_to_write parameter by four, and then relies on its
2329 * allocator to allocate larger extents to make them
2330 * contiguous. Unfortunately this brings us to the second
2331 * stupidity, which is that ext4's mballoc code only allocates
2332 * at most 2048 blocks. So we force contiguous writes up to
2333 * the number of dirty blocks in the inode, or
2334 * sbi->max_writeback_mb_bump whichever is smaller.
2336 max_pages
= sbi
->s_max_writeback_mb_bump
<< (20 - PAGE_CACHE_SHIFT
);
2337 if (!range_cyclic
&& range_whole
) {
2338 if (wbc
->nr_to_write
== LONG_MAX
)
2339 desired_nr_to_write
= wbc
->nr_to_write
;
2341 desired_nr_to_write
= wbc
->nr_to_write
* 8;
2343 desired_nr_to_write
= ext4_num_dirty_pages(inode
, index
,
2345 if (desired_nr_to_write
> max_pages
)
2346 desired_nr_to_write
= max_pages
;
2348 if (wbc
->nr_to_write
< desired_nr_to_write
) {
2349 nr_to_writebump
= desired_nr_to_write
- wbc
->nr_to_write
;
2350 wbc
->nr_to_write
= desired_nr_to_write
;
2354 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
)
2355 tag_pages_for_writeback(mapping
, index
, end
);
2357 blk_start_plug(&plug
);
2358 while (!ret
&& wbc
->nr_to_write
> 0) {
2361 * we insert one extent at a time. So we need
2362 * credit needed for single extent allocation.
2363 * journalled mode is currently not supported
2366 BUG_ON(ext4_should_journal_data(inode
));
2367 needed_blocks
= ext4_da_writepages_trans_blocks(inode
);
2369 /* start a new transaction*/
2370 handle
= ext4_journal_start(inode
, needed_blocks
);
2371 if (IS_ERR(handle
)) {
2372 ret
= PTR_ERR(handle
);
2373 ext4_msg(inode
->i_sb
, KERN_CRIT
, "%s: jbd2_start: "
2374 "%ld pages, ino %lu; err %d", __func__
,
2375 wbc
->nr_to_write
, inode
->i_ino
, ret
);
2376 blk_finish_plug(&plug
);
2377 goto out_writepages
;
2381 * Now call write_cache_pages_da() to find the next
2382 * contiguous region of logical blocks that need
2383 * blocks to be allocated by ext4 and submit them.
2385 ret
= write_cache_pages_da(mapping
, wbc
, &mpd
, &done_index
);
2387 * If we have a contiguous extent of pages and we
2388 * haven't done the I/O yet, map the blocks and submit
2391 if (!mpd
.io_done
&& mpd
.next_page
!= mpd
.first_page
) {
2392 mpage_da_map_and_submit(&mpd
);
2393 ret
= MPAGE_DA_EXTENT_TAIL
;
2395 trace_ext4_da_write_pages(inode
, &mpd
);
2396 wbc
->nr_to_write
-= mpd
.pages_written
;
2398 ext4_journal_stop(handle
);
2400 if ((mpd
.retval
== -ENOSPC
) && sbi
->s_journal
) {
2401 /* commit the transaction which would
2402 * free blocks released in the transaction
2405 jbd2_journal_force_commit_nested(sbi
->s_journal
);
2407 } else if (ret
== MPAGE_DA_EXTENT_TAIL
) {
2409 * Got one extent now try with rest of the pages.
2410 * If mpd.retval is set -EIO, journal is aborted.
2411 * So we don't need to write any more.
2413 pages_written
+= mpd
.pages_written
;
2416 } else if (wbc
->nr_to_write
)
2418 * There is no more writeout needed
2419 * or we requested for a noblocking writeout
2420 * and we found the device congested
2424 blk_finish_plug(&plug
);
2425 if (!io_done
&& !cycled
) {
2428 wbc
->range_start
= index
<< PAGE_CACHE_SHIFT
;
2429 wbc
->range_end
= mapping
->writeback_index
- 1;
2434 wbc
->range_cyclic
= range_cyclic
;
2435 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
2437 * set the writeback_index so that range_cyclic
2438 * mode will write it back later
2440 mapping
->writeback_index
= done_index
;
2443 wbc
->nr_to_write
-= nr_to_writebump
;
2444 wbc
->range_start
= range_start
;
2445 trace_ext4_da_writepages_result(inode
, wbc
, ret
, pages_written
);
2449 #define FALL_BACK_TO_NONDELALLOC 1
2450 static int ext4_nonda_switch(struct super_block
*sb
)
2452 s64 free_blocks
, dirty_blocks
;
2453 struct ext4_sb_info
*sbi
= EXT4_SB(sb
);
2456 * switch to non delalloc mode if we are running low
2457 * on free block. The free block accounting via percpu
2458 * counters can get slightly wrong with percpu_counter_batch getting
2459 * accumulated on each CPU without updating global counters
2460 * Delalloc need an accurate free block accounting. So switch
2461 * to non delalloc when we are near to error range.
2463 free_blocks
= EXT4_C2B(sbi
,
2464 percpu_counter_read_positive(&sbi
->s_freeclusters_counter
));
2465 dirty_blocks
= percpu_counter_read_positive(&sbi
->s_dirtyclusters_counter
);
2466 if (2 * free_blocks
< 3 * dirty_blocks
||
2467 free_blocks
< (dirty_blocks
+ EXT4_FREECLUSTERS_WATERMARK
)) {
2469 * free block count is less than 150% of dirty blocks
2470 * or free blocks is less than watermark
2475 * Even if we don't switch but are nearing capacity,
2476 * start pushing delalloc when 1/2 of free blocks are dirty.
2478 if (free_blocks
< 2 * dirty_blocks
)
2479 writeback_inodes_sb_if_idle(sb
, WB_REASON_FS_FREE_SPACE
);
2484 static int ext4_da_write_begin(struct file
*file
, struct address_space
*mapping
,
2485 loff_t pos
, unsigned len
, unsigned flags
,
2486 struct page
**pagep
, void **fsdata
)
2488 int ret
, retries
= 0;
2491 struct inode
*inode
= mapping
->host
;
2494 index
= pos
>> PAGE_CACHE_SHIFT
;
2496 if (ext4_nonda_switch(inode
->i_sb
)) {
2497 *fsdata
= (void *)FALL_BACK_TO_NONDELALLOC
;
2498 return ext4_write_begin(file
, mapping
, pos
,
2499 len
, flags
, pagep
, fsdata
);
2501 *fsdata
= (void *)0;
2502 trace_ext4_da_write_begin(inode
, pos
, len
, flags
);
2505 * With delayed allocation, we don't log the i_disksize update
2506 * if there is delayed block allocation. But we still need
2507 * to journalling the i_disksize update if writes to the end
2508 * of file which has an already mapped buffer.
2510 handle
= ext4_journal_start(inode
, 1);
2511 if (IS_ERR(handle
)) {
2512 ret
= PTR_ERR(handle
);
2515 /* We cannot recurse into the filesystem as the transaction is already
2517 flags
|= AOP_FLAG_NOFS
;
2519 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
2521 ext4_journal_stop(handle
);
2527 ret
= __block_write_begin(page
, pos
, len
, ext4_da_get_block_prep
);
2530 ext4_journal_stop(handle
);
2531 page_cache_release(page
);
2533 * block_write_begin may have instantiated a few blocks
2534 * outside i_size. Trim these off again. Don't need
2535 * i_size_read because we hold i_mutex.
2537 if (pos
+ len
> inode
->i_size
)
2538 ext4_truncate_failed_write(inode
);
2541 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
2548 * Check if we should update i_disksize
2549 * when write to the end of file but not require block allocation
2551 static int ext4_da_should_update_i_disksize(struct page
*page
,
2552 unsigned long offset
)
2554 struct buffer_head
*bh
;
2555 struct inode
*inode
= page
->mapping
->host
;
2559 bh
= page_buffers(page
);
2560 idx
= offset
>> inode
->i_blkbits
;
2562 for (i
= 0; i
< idx
; i
++)
2563 bh
= bh
->b_this_page
;
2565 if (!buffer_mapped(bh
) || (buffer_delay(bh
)) || buffer_unwritten(bh
))
2570 static int ext4_da_write_end(struct file
*file
,
2571 struct address_space
*mapping
,
2572 loff_t pos
, unsigned len
, unsigned copied
,
2573 struct page
*page
, void *fsdata
)
2575 struct inode
*inode
= mapping
->host
;
2577 handle_t
*handle
= ext4_journal_current_handle();
2579 unsigned long start
, end
;
2580 int write_mode
= (int)(unsigned long)fsdata
;
2582 if (write_mode
== FALL_BACK_TO_NONDELALLOC
) {
2583 switch (ext4_inode_journal_mode(inode
)) {
2584 case EXT4_INODE_ORDERED_DATA_MODE
:
2585 return ext4_ordered_write_end(file
, mapping
, pos
,
2586 len
, copied
, page
, fsdata
);
2587 case EXT4_INODE_WRITEBACK_DATA_MODE
:
2588 return ext4_writeback_write_end(file
, mapping
, pos
,
2589 len
, copied
, page
, fsdata
);
2595 trace_ext4_da_write_end(inode
, pos
, len
, copied
);
2596 start
= pos
& (PAGE_CACHE_SIZE
- 1);
2597 end
= start
+ copied
- 1;
2600 * generic_write_end() will run mark_inode_dirty() if i_size
2601 * changes. So let's piggyback the i_disksize mark_inode_dirty
2605 new_i_size
= pos
+ copied
;
2606 if (copied
&& new_i_size
> EXT4_I(inode
)->i_disksize
) {
2607 if (ext4_da_should_update_i_disksize(page
, end
)) {
2608 down_write(&EXT4_I(inode
)->i_data_sem
);
2609 if (new_i_size
> EXT4_I(inode
)->i_disksize
) {
2611 * Updating i_disksize when extending file
2612 * without needing block allocation
2614 if (ext4_should_order_data(inode
))
2615 ret
= ext4_jbd2_file_inode(handle
,
2618 EXT4_I(inode
)->i_disksize
= new_i_size
;
2620 up_write(&EXT4_I(inode
)->i_data_sem
);
2621 /* We need to mark inode dirty even if
2622 * new_i_size is less that inode->i_size
2623 * bu greater than i_disksize.(hint delalloc)
2625 ext4_mark_inode_dirty(handle
, inode
);
2628 ret2
= generic_write_end(file
, mapping
, pos
, len
, copied
,
2633 ret2
= ext4_journal_stop(handle
);
2637 return ret
? ret
: copied
;
2640 static void ext4_da_invalidatepage(struct page
*page
, unsigned long offset
)
2643 * Drop reserved blocks
2645 BUG_ON(!PageLocked(page
));
2646 if (!page_has_buffers(page
))
2649 ext4_da_page_release_reservation(page
, offset
);
2652 ext4_invalidatepage(page
, offset
);
2658 * Force all delayed allocation blocks to be allocated for a given inode.
2660 int ext4_alloc_da_blocks(struct inode
*inode
)
2662 trace_ext4_alloc_da_blocks(inode
);
2664 if (!EXT4_I(inode
)->i_reserved_data_blocks
&&
2665 !EXT4_I(inode
)->i_reserved_meta_blocks
)
2669 * We do something simple for now. The filemap_flush() will
2670 * also start triggering a write of the data blocks, which is
2671 * not strictly speaking necessary (and for users of
2672 * laptop_mode, not even desirable). However, to do otherwise
2673 * would require replicating code paths in:
2675 * ext4_da_writepages() ->
2676 * write_cache_pages() ---> (via passed in callback function)
2677 * __mpage_da_writepage() -->
2678 * mpage_add_bh_to_extent()
2679 * mpage_da_map_blocks()
2681 * The problem is that write_cache_pages(), located in
2682 * mm/page-writeback.c, marks pages clean in preparation for
2683 * doing I/O, which is not desirable if we're not planning on
2686 * We could call write_cache_pages(), and then redirty all of
2687 * the pages by calling redirty_page_for_writepage() but that
2688 * would be ugly in the extreme. So instead we would need to
2689 * replicate parts of the code in the above functions,
2690 * simplifying them because we wouldn't actually intend to
2691 * write out the pages, but rather only collect contiguous
2692 * logical block extents, call the multi-block allocator, and
2693 * then update the buffer heads with the block allocations.
2695 * For now, though, we'll cheat by calling filemap_flush(),
2696 * which will map the blocks, and start the I/O, but not
2697 * actually wait for the I/O to complete.
2699 return filemap_flush(inode
->i_mapping
);
2703 * bmap() is special. It gets used by applications such as lilo and by
2704 * the swapper to find the on-disk block of a specific piece of data.
2706 * Naturally, this is dangerous if the block concerned is still in the
2707 * journal. If somebody makes a swapfile on an ext4 data-journaling
2708 * filesystem and enables swap, then they may get a nasty shock when the
2709 * data getting swapped to that swapfile suddenly gets overwritten by
2710 * the original zero's written out previously to the journal and
2711 * awaiting writeback in the kernel's buffer cache.
2713 * So, if we see any bmap calls here on a modified, data-journaled file,
2714 * take extra steps to flush any blocks which might be in the cache.
2716 static sector_t
ext4_bmap(struct address_space
*mapping
, sector_t block
)
2718 struct inode
*inode
= mapping
->host
;
2722 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
) &&
2723 test_opt(inode
->i_sb
, DELALLOC
)) {
2725 * With delalloc we want to sync the file
2726 * so that we can make sure we allocate
2729 filemap_write_and_wait(mapping
);
2732 if (EXT4_JOURNAL(inode
) &&
2733 ext4_test_inode_state(inode
, EXT4_STATE_JDATA
)) {
2735 * This is a REALLY heavyweight approach, but the use of
2736 * bmap on dirty files is expected to be extremely rare:
2737 * only if we run lilo or swapon on a freshly made file
2738 * do we expect this to happen.
2740 * (bmap requires CAP_SYS_RAWIO so this does not
2741 * represent an unprivileged user DOS attack --- we'd be
2742 * in trouble if mortal users could trigger this path at
2745 * NB. EXT4_STATE_JDATA is not set on files other than
2746 * regular files. If somebody wants to bmap a directory
2747 * or symlink and gets confused because the buffer
2748 * hasn't yet been flushed to disk, they deserve
2749 * everything they get.
2752 ext4_clear_inode_state(inode
, EXT4_STATE_JDATA
);
2753 journal
= EXT4_JOURNAL(inode
);
2754 jbd2_journal_lock_updates(journal
);
2755 err
= jbd2_journal_flush(journal
);
2756 jbd2_journal_unlock_updates(journal
);
2762 return generic_block_bmap(mapping
, block
, ext4_get_block
);
2765 static int ext4_readpage(struct file
*file
, struct page
*page
)
2767 trace_ext4_readpage(page
);
2768 return mpage_readpage(page
, ext4_get_block
);
2772 ext4_readpages(struct file
*file
, struct address_space
*mapping
,
2773 struct list_head
*pages
, unsigned nr_pages
)
2775 return mpage_readpages(mapping
, pages
, nr_pages
, ext4_get_block
);
2778 static void ext4_invalidatepage_free_endio(struct page
*page
, unsigned long offset
)
2780 struct buffer_head
*head
, *bh
;
2781 unsigned int curr_off
= 0;
2783 if (!page_has_buffers(page
))
2785 head
= bh
= page_buffers(page
);
2787 if (offset
<= curr_off
&& test_clear_buffer_uninit(bh
)
2789 ext4_free_io_end(bh
->b_private
);
2790 bh
->b_private
= NULL
;
2791 bh
->b_end_io
= NULL
;
2793 curr_off
= curr_off
+ bh
->b_size
;
2794 bh
= bh
->b_this_page
;
2795 } while (bh
!= head
);
2798 static void ext4_invalidatepage(struct page
*page
, unsigned long offset
)
2800 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2802 trace_ext4_invalidatepage(page
, offset
);
2805 * free any io_end structure allocated for buffers to be discarded
2807 if (ext4_should_dioread_nolock(page
->mapping
->host
))
2808 ext4_invalidatepage_free_endio(page
, offset
);
2810 * If it's a full truncate we just forget about the pending dirtying
2813 ClearPageChecked(page
);
2816 jbd2_journal_invalidatepage(journal
, page
, offset
);
2818 block_invalidatepage(page
, offset
);
2821 static int ext4_releasepage(struct page
*page
, gfp_t wait
)
2823 journal_t
*journal
= EXT4_JOURNAL(page
->mapping
->host
);
2825 trace_ext4_releasepage(page
);
2827 WARN_ON(PageChecked(page
));
2828 if (!page_has_buffers(page
))
2831 return jbd2_journal_try_to_free_buffers(journal
, page
, wait
);
2833 return try_to_free_buffers(page
);
2837 * ext4_get_block used when preparing for a DIO write or buffer write.
2838 * We allocate an uinitialized extent if blocks haven't been allocated.
2839 * The extent will be converted to initialized after the IO is complete.
2841 static int ext4_get_block_write(struct inode
*inode
, sector_t iblock
,
2842 struct buffer_head
*bh_result
, int create
)
2844 ext4_debug("ext4_get_block_write: inode %lu, create flag %d\n",
2845 inode
->i_ino
, create
);
2846 return _ext4_get_block(inode
, iblock
, bh_result
,
2847 EXT4_GET_BLOCKS_IO_CREATE_EXT
);
2850 static int ext4_get_block_write_nolock(struct inode
*inode
, sector_t iblock
,
2851 struct buffer_head
*bh_result
, int flags
)
2853 handle_t
*handle
= ext4_journal_current_handle();
2854 struct ext4_map_blocks map
;
2857 ext4_debug("ext4_get_block_write_nolock: inode %lu, flag %d\n",
2858 inode
->i_ino
, flags
);
2860 flags
= EXT4_GET_BLOCKS_NO_LOCK
;
2862 map
.m_lblk
= iblock
;
2863 map
.m_len
= bh_result
->b_size
>> inode
->i_blkbits
;
2865 ret
= ext4_map_blocks(handle
, inode
, &map
, flags
);
2867 map_bh(bh_result
, inode
->i_sb
, map
.m_pblk
);
2868 bh_result
->b_state
= (bh_result
->b_state
& ~EXT4_MAP_FLAGS
) |
2870 bh_result
->b_size
= inode
->i_sb
->s_blocksize
* map
.m_len
;
2876 static void ext4_end_io_dio(struct kiocb
*iocb
, loff_t offset
,
2877 ssize_t size
, void *private, int ret
,
2880 struct inode
*inode
= iocb
->ki_filp
->f_path
.dentry
->d_inode
;
2881 ext4_io_end_t
*io_end
= iocb
->private;
2882 struct workqueue_struct
*wq
;
2883 unsigned long flags
;
2884 struct ext4_inode_info
*ei
;
2886 /* if not async direct IO or dio with 0 bytes write, just return */
2887 if (!io_end
|| !size
)
2890 ext_debug("ext4_end_io_dio(): io_end 0x%p "
2891 "for inode %lu, iocb 0x%p, offset %llu, size %zd\n",
2892 iocb
->private, io_end
->inode
->i_ino
, iocb
, offset
,
2895 iocb
->private = NULL
;
2897 /* if not aio dio with unwritten extents, just free io and return */
2898 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
2899 ext4_free_io_end(io_end
);
2902 aio_complete(iocb
, ret
, 0);
2903 inode_dio_done(inode
);
2907 io_end
->offset
= offset
;
2908 io_end
->size
= size
;
2910 io_end
->iocb
= iocb
;
2911 io_end
->result
= ret
;
2913 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
2915 /* Add the io_end to per-inode completed aio dio list*/
2916 ei
= EXT4_I(io_end
->inode
);
2917 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
2918 list_add_tail(&io_end
->list
, &ei
->i_completed_io_list
);
2919 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
2921 /* queue the work to convert unwritten extents to written */
2922 queue_work(wq
, &io_end
->work
);
2925 static void ext4_end_io_buffer_write(struct buffer_head
*bh
, int uptodate
)
2927 ext4_io_end_t
*io_end
= bh
->b_private
;
2928 struct workqueue_struct
*wq
;
2929 struct inode
*inode
;
2930 unsigned long flags
;
2932 if (!test_clear_buffer_uninit(bh
) || !io_end
)
2935 if (!(io_end
->inode
->i_sb
->s_flags
& MS_ACTIVE
)) {
2936 ext4_msg(io_end
->inode
->i_sb
, KERN_INFO
,
2937 "sb umounted, discard end_io request for inode %lu",
2938 io_end
->inode
->i_ino
);
2939 ext4_free_io_end(io_end
);
2944 * It may be over-defensive here to check EXT4_IO_END_UNWRITTEN now,
2945 * but being more careful is always safe for the future change.
2947 inode
= io_end
->inode
;
2948 ext4_set_io_unwritten_flag(inode
, io_end
);
2950 /* Add the io_end to per-inode completed io list*/
2951 spin_lock_irqsave(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
2952 list_add_tail(&io_end
->list
, &EXT4_I(inode
)->i_completed_io_list
);
2953 spin_unlock_irqrestore(&EXT4_I(inode
)->i_completed_io_lock
, flags
);
2955 wq
= EXT4_SB(inode
->i_sb
)->dio_unwritten_wq
;
2956 /* queue the work to convert unwritten extents to written */
2957 queue_work(wq
, &io_end
->work
);
2959 bh
->b_private
= NULL
;
2960 bh
->b_end_io
= NULL
;
2961 clear_buffer_uninit(bh
);
2962 end_buffer_async_write(bh
, uptodate
);
2965 static int ext4_set_bh_endio(struct buffer_head
*bh
, struct inode
*inode
)
2967 ext4_io_end_t
*io_end
;
2968 struct page
*page
= bh
->b_page
;
2969 loff_t offset
= (sector_t
)page
->index
<< PAGE_CACHE_SHIFT
;
2970 size_t size
= bh
->b_size
;
2973 io_end
= ext4_init_io_end(inode
, GFP_ATOMIC
);
2975 pr_warn_ratelimited("%s: allocation fail\n", __func__
);
2979 io_end
->offset
= offset
;
2980 io_end
->size
= size
;
2982 * We need to hold a reference to the page to make sure it
2983 * doesn't get evicted before ext4_end_io_work() has a chance
2984 * to convert the extent from written to unwritten.
2986 io_end
->page
= page
;
2987 get_page(io_end
->page
);
2989 bh
->b_private
= io_end
;
2990 bh
->b_end_io
= ext4_end_io_buffer_write
;
2995 * For ext4 extent files, ext4 will do direct-io write to holes,
2996 * preallocated extents, and those write extend the file, no need to
2997 * fall back to buffered IO.
2999 * For holes, we fallocate those blocks, mark them as uninitialized
3000 * If those blocks were preallocated, we mark sure they are splited, but
3001 * still keep the range to write as uninitialized.
3003 * The unwrritten extents will be converted to written when DIO is completed.
3004 * For async direct IO, since the IO may still pending when return, we
3005 * set up an end_io call back function, which will do the conversion
3006 * when async direct IO completed.
3008 * If the O_DIRECT write will extend the file then add this inode to the
3009 * orphan list. So recovery will truncate it back to the original size
3010 * if the machine crashes during the write.
3013 static ssize_t
ext4_ext_direct_IO(int rw
, struct kiocb
*iocb
,
3014 const struct iovec
*iov
, loff_t offset
,
3015 unsigned long nr_segs
)
3017 struct file
*file
= iocb
->ki_filp
;
3018 struct inode
*inode
= file
->f_mapping
->host
;
3020 size_t count
= iov_length(iov
, nr_segs
);
3022 loff_t final_size
= offset
+ count
;
3023 if (rw
== WRITE
&& final_size
<= inode
->i_size
) {
3026 BUG_ON(iocb
->private == NULL
);
3028 /* If we do a overwrite dio, i_mutex locking can be released */
3029 overwrite
= *((int *)iocb
->private);
3032 down_read(&EXT4_I(inode
)->i_data_sem
);
3033 mutex_unlock(&inode
->i_mutex
);
3037 * We could direct write to holes and fallocate.
3039 * Allocated blocks to fill the hole are marked as uninitialized
3040 * to prevent parallel buffered read to expose the stale data
3041 * before DIO complete the data IO.
3043 * As to previously fallocated extents, ext4 get_block
3044 * will just simply mark the buffer mapped but still
3045 * keep the extents uninitialized.
3047 * for non AIO case, we will convert those unwritten extents
3048 * to written after return back from blockdev_direct_IO.
3050 * for async DIO, the conversion needs to be defered when
3051 * the IO is completed. The ext4 end_io callback function
3052 * will be called to take care of the conversion work.
3053 * Here for async case, we allocate an io_end structure to
3056 iocb
->private = NULL
;
3057 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3058 if (!is_sync_kiocb(iocb
)) {
3059 ext4_io_end_t
*io_end
=
3060 ext4_init_io_end(inode
, GFP_NOFS
);
3065 io_end
->flag
|= EXT4_IO_END_DIRECT
;
3066 iocb
->private = io_end
;
3068 * we save the io structure for current async
3069 * direct IO, so that later ext4_map_blocks()
3070 * could flag the io structure whether there
3071 * is a unwritten extents needs to be converted
3072 * when IO is completed.
3074 EXT4_I(inode
)->cur_aio_dio
= iocb
->private;
3078 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3079 inode
->i_sb
->s_bdev
, iov
,
3081 ext4_get_block_write_nolock
,
3086 ret
= __blockdev_direct_IO(rw
, iocb
, inode
,
3087 inode
->i_sb
->s_bdev
, iov
,
3089 ext4_get_block_write
,
3094 EXT4_I(inode
)->cur_aio_dio
= NULL
;
3096 * The io_end structure takes a reference to the inode,
3097 * that structure needs to be destroyed and the
3098 * reference to the inode need to be dropped, when IO is
3099 * complete, even with 0 byte write, or failed.
3101 * In the successful AIO DIO case, the io_end structure will be
3102 * desctroyed and the reference to the inode will be dropped
3103 * after the end_io call back function is called.
3105 * In the case there is 0 byte write, or error case, since
3106 * VFS direct IO won't invoke the end_io call back function,
3107 * we need to free the end_io structure here.
3109 if (ret
!= -EIOCBQUEUED
&& ret
<= 0 && iocb
->private) {
3110 ext4_free_io_end(iocb
->private);
3111 iocb
->private = NULL
;
3112 } else if (ret
> 0 && !overwrite
&& ext4_test_inode_state(inode
,
3113 EXT4_STATE_DIO_UNWRITTEN
)) {
3116 * for non AIO case, since the IO is already
3117 * completed, we could do the conversion right here
3119 err
= ext4_convert_unwritten_extents(inode
,
3123 ext4_clear_inode_state(inode
, EXT4_STATE_DIO_UNWRITTEN
);
3127 /* take i_mutex locking again if we do a ovewrite dio */
3129 up_read(&EXT4_I(inode
)->i_data_sem
);
3130 mutex_lock(&inode
->i_mutex
);
3136 /* for write the the end of file case, we fall back to old way */
3137 return ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3140 static ssize_t
ext4_direct_IO(int rw
, struct kiocb
*iocb
,
3141 const struct iovec
*iov
, loff_t offset
,
3142 unsigned long nr_segs
)
3144 struct file
*file
= iocb
->ki_filp
;
3145 struct inode
*inode
= file
->f_mapping
->host
;
3149 * If we are doing data journalling we don't support O_DIRECT
3151 if (ext4_should_journal_data(inode
))
3154 trace_ext4_direct_IO_enter(inode
, offset
, iov_length(iov
, nr_segs
), rw
);
3155 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3156 ret
= ext4_ext_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3158 ret
= ext4_ind_direct_IO(rw
, iocb
, iov
, offset
, nr_segs
);
3159 trace_ext4_direct_IO_exit(inode
, offset
,
3160 iov_length(iov
, nr_segs
), rw
, ret
);
3165 * Pages can be marked dirty completely asynchronously from ext4's journalling
3166 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
3167 * much here because ->set_page_dirty is called under VFS locks. The page is
3168 * not necessarily locked.
3170 * We cannot just dirty the page and leave attached buffers clean, because the
3171 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
3172 * or jbddirty because all the journalling code will explode.
3174 * So what we do is to mark the page "pending dirty" and next time writepage
3175 * is called, propagate that into the buffers appropriately.
3177 static int ext4_journalled_set_page_dirty(struct page
*page
)
3179 SetPageChecked(page
);
3180 return __set_page_dirty_nobuffers(page
);
3183 static const struct address_space_operations ext4_ordered_aops
= {
3184 .readpage
= ext4_readpage
,
3185 .readpages
= ext4_readpages
,
3186 .writepage
= ext4_writepage
,
3187 .write_begin
= ext4_write_begin
,
3188 .write_end
= ext4_ordered_write_end
,
3190 .invalidatepage
= ext4_invalidatepage
,
3191 .releasepage
= ext4_releasepage
,
3192 .direct_IO
= ext4_direct_IO
,
3193 .migratepage
= buffer_migrate_page
,
3194 .is_partially_uptodate
= block_is_partially_uptodate
,
3195 .error_remove_page
= generic_error_remove_page
,
3198 static const struct address_space_operations ext4_writeback_aops
= {
3199 .readpage
= ext4_readpage
,
3200 .readpages
= ext4_readpages
,
3201 .writepage
= ext4_writepage
,
3202 .write_begin
= ext4_write_begin
,
3203 .write_end
= ext4_writeback_write_end
,
3205 .invalidatepage
= ext4_invalidatepage
,
3206 .releasepage
= ext4_releasepage
,
3207 .direct_IO
= ext4_direct_IO
,
3208 .migratepage
= buffer_migrate_page
,
3209 .is_partially_uptodate
= block_is_partially_uptodate
,
3210 .error_remove_page
= generic_error_remove_page
,
3213 static const struct address_space_operations ext4_journalled_aops
= {
3214 .readpage
= ext4_readpage
,
3215 .readpages
= ext4_readpages
,
3216 .writepage
= ext4_writepage
,
3217 .write_begin
= ext4_write_begin
,
3218 .write_end
= ext4_journalled_write_end
,
3219 .set_page_dirty
= ext4_journalled_set_page_dirty
,
3221 .invalidatepage
= ext4_invalidatepage
,
3222 .releasepage
= ext4_releasepage
,
3223 .direct_IO
= ext4_direct_IO
,
3224 .is_partially_uptodate
= block_is_partially_uptodate
,
3225 .error_remove_page
= generic_error_remove_page
,
3228 static const struct address_space_operations ext4_da_aops
= {
3229 .readpage
= ext4_readpage
,
3230 .readpages
= ext4_readpages
,
3231 .writepage
= ext4_writepage
,
3232 .writepages
= ext4_da_writepages
,
3233 .write_begin
= ext4_da_write_begin
,
3234 .write_end
= ext4_da_write_end
,
3236 .invalidatepage
= ext4_da_invalidatepage
,
3237 .releasepage
= ext4_releasepage
,
3238 .direct_IO
= ext4_direct_IO
,
3239 .migratepage
= buffer_migrate_page
,
3240 .is_partially_uptodate
= block_is_partially_uptodate
,
3241 .error_remove_page
= generic_error_remove_page
,
3244 void ext4_set_aops(struct inode
*inode
)
3246 switch (ext4_inode_journal_mode(inode
)) {
3247 case EXT4_INODE_ORDERED_DATA_MODE
:
3248 if (test_opt(inode
->i_sb
, DELALLOC
))
3249 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3251 inode
->i_mapping
->a_ops
= &ext4_ordered_aops
;
3253 case EXT4_INODE_WRITEBACK_DATA_MODE
:
3254 if (test_opt(inode
->i_sb
, DELALLOC
))
3255 inode
->i_mapping
->a_ops
= &ext4_da_aops
;
3257 inode
->i_mapping
->a_ops
= &ext4_writeback_aops
;
3259 case EXT4_INODE_JOURNAL_DATA_MODE
:
3260 inode
->i_mapping
->a_ops
= &ext4_journalled_aops
;
3269 * ext4_discard_partial_page_buffers()
3270 * Wrapper function for ext4_discard_partial_page_buffers_no_lock.
3271 * This function finds and locks the page containing the offset
3272 * "from" and passes it to ext4_discard_partial_page_buffers_no_lock.
3273 * Calling functions that already have the page locked should call
3274 * ext4_discard_partial_page_buffers_no_lock directly.
3276 int ext4_discard_partial_page_buffers(handle_t
*handle
,
3277 struct address_space
*mapping
, loff_t from
,
3278 loff_t length
, int flags
)
3280 struct inode
*inode
= mapping
->host
;
3284 page
= find_or_create_page(mapping
, from
>> PAGE_CACHE_SHIFT
,
3285 mapping_gfp_mask(mapping
) & ~__GFP_FS
);
3289 err
= ext4_discard_partial_page_buffers_no_lock(handle
, inode
, page
,
3290 from
, length
, flags
);
3293 page_cache_release(page
);
3298 * ext4_discard_partial_page_buffers_no_lock()
3299 * Zeros a page range of length 'length' starting from offset 'from'.
3300 * Buffer heads that correspond to the block aligned regions of the
3301 * zeroed range will be unmapped. Unblock aligned regions
3302 * will have the corresponding buffer head mapped if needed so that
3303 * that region of the page can be updated with the partial zero out.
3305 * This function assumes that the page has already been locked. The
3306 * The range to be discarded must be contained with in the given page.
3307 * If the specified range exceeds the end of the page it will be shortened
3308 * to the end of the page that corresponds to 'from'. This function is
3309 * appropriate for updating a page and it buffer heads to be unmapped and
3310 * zeroed for blocks that have been either released, or are going to be
3313 * handle: The journal handle
3314 * inode: The files inode
3315 * page: A locked page that contains the offset "from"
3316 * from: The starting byte offset (from the begining of the file)
3317 * to begin discarding
3318 * len: The length of bytes to discard
3319 * flags: Optional flags that may be used:
3321 * EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
3322 * Only zero the regions of the page whose buffer heads
3323 * have already been unmapped. This flag is appropriate
3324 * for updateing the contents of a page whose blocks may
3325 * have already been released, and we only want to zero
3326 * out the regions that correspond to those released blocks.
3328 * Returns zero on sucess or negative on failure.
3330 static int ext4_discard_partial_page_buffers_no_lock(handle_t
*handle
,
3331 struct inode
*inode
, struct page
*page
, loff_t from
,
3332 loff_t length
, int flags
)
3334 ext4_fsblk_t index
= from
>> PAGE_CACHE_SHIFT
;
3335 unsigned int offset
= from
& (PAGE_CACHE_SIZE
-1);
3336 unsigned int blocksize
, max
, pos
;
3338 struct buffer_head
*bh
;
3341 blocksize
= inode
->i_sb
->s_blocksize
;
3342 max
= PAGE_CACHE_SIZE
- offset
;
3344 if (index
!= page
->index
)
3348 * correct length if it does not fall between
3349 * 'from' and the end of the page
3351 if (length
> max
|| length
< 0)
3354 iblock
= index
<< (PAGE_CACHE_SHIFT
- inode
->i_sb
->s_blocksize_bits
);
3356 if (!page_has_buffers(page
))
3357 create_empty_buffers(page
, blocksize
, 0);
3359 /* Find the buffer that contains "offset" */
3360 bh
= page_buffers(page
);
3362 while (offset
>= pos
) {
3363 bh
= bh
->b_this_page
;
3369 while (pos
< offset
+ length
) {
3370 unsigned int end_of_block
, range_to_discard
;
3374 /* The length of space left to zero and unmap */
3375 range_to_discard
= offset
+ length
- pos
;
3377 /* The length of space until the end of the block */
3378 end_of_block
= blocksize
- (pos
& (blocksize
-1));
3381 * Do not unmap or zero past end of block
3382 * for this buffer head
3384 if (range_to_discard
> end_of_block
)
3385 range_to_discard
= end_of_block
;
3389 * Skip this buffer head if we are only zeroing unampped
3390 * regions of the page
3392 if (flags
& EXT4_DISCARD_PARTIAL_PG_ZERO_UNMAPPED
&&
3396 /* If the range is block aligned, unmap */
3397 if (range_to_discard
== blocksize
) {
3398 clear_buffer_dirty(bh
);
3400 clear_buffer_mapped(bh
);
3401 clear_buffer_req(bh
);
3402 clear_buffer_new(bh
);
3403 clear_buffer_delay(bh
);
3404 clear_buffer_unwritten(bh
);
3405 clear_buffer_uptodate(bh
);
3406 zero_user(page
, pos
, range_to_discard
);
3407 BUFFER_TRACE(bh
, "Buffer discarded");
3412 * If this block is not completely contained in the range
3413 * to be discarded, then it is not going to be released. Because
3414 * we need to keep this block, we need to make sure this part
3415 * of the page is uptodate before we modify it by writeing
3416 * partial zeros on it.
3418 if (!buffer_mapped(bh
)) {
3420 * Buffer head must be mapped before we can read
3423 BUFFER_TRACE(bh
, "unmapped");
3424 ext4_get_block(inode
, iblock
, bh
, 0);
3425 /* unmapped? It's a hole - nothing to do */
3426 if (!buffer_mapped(bh
)) {
3427 BUFFER_TRACE(bh
, "still unmapped");
3432 /* Ok, it's mapped. Make sure it's up-to-date */
3433 if (PageUptodate(page
))
3434 set_buffer_uptodate(bh
);
3436 if (!buffer_uptodate(bh
)) {
3438 ll_rw_block(READ
, 1, &bh
);
3440 /* Uhhuh. Read error. Complain and punt.*/
3441 if (!buffer_uptodate(bh
))
3445 if (ext4_should_journal_data(inode
)) {
3446 BUFFER_TRACE(bh
, "get write access");
3447 err
= ext4_journal_get_write_access(handle
, bh
);
3452 zero_user(page
, pos
, range_to_discard
);
3455 if (ext4_should_journal_data(inode
)) {
3456 err
= ext4_handle_dirty_metadata(handle
, inode
, bh
);
3458 mark_buffer_dirty(bh
);
3460 BUFFER_TRACE(bh
, "Partial buffer zeroed");
3462 bh
= bh
->b_this_page
;
3464 pos
+= range_to_discard
;
3470 int ext4_can_truncate(struct inode
*inode
)
3472 if (S_ISREG(inode
->i_mode
))
3474 if (S_ISDIR(inode
->i_mode
))
3476 if (S_ISLNK(inode
->i_mode
))
3477 return !ext4_inode_is_fast_symlink(inode
);
3482 * ext4_punch_hole: punches a hole in a file by releaseing the blocks
3483 * associated with the given offset and length
3485 * @inode: File inode
3486 * @offset: The offset where the hole will begin
3487 * @len: The length of the hole
3489 * Returns: 0 on sucess or negative on failure
3492 int ext4_punch_hole(struct file
*file
, loff_t offset
, loff_t length
)
3494 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
3495 if (!S_ISREG(inode
->i_mode
))
3498 if (!ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
3499 /* TODO: Add support for non extent hole punching */
3503 if (EXT4_SB(inode
->i_sb
)->s_cluster_ratio
> 1) {
3504 /* TODO: Add support for bigalloc file systems */
3508 return ext4_ext_punch_hole(file
, offset
, length
);
3514 * We block out ext4_get_block() block instantiations across the entire
3515 * transaction, and VFS/VM ensures that ext4_truncate() cannot run
3516 * simultaneously on behalf of the same inode.
3518 * As we work through the truncate and commit bits of it to the journal there
3519 * is one core, guiding principle: the file's tree must always be consistent on
3520 * disk. We must be able to restart the truncate after a crash.
3522 * The file's tree may be transiently inconsistent in memory (although it
3523 * probably isn't), but whenever we close off and commit a journal transaction,
3524 * the contents of (the filesystem + the journal) must be consistent and
3525 * restartable. It's pretty simple, really: bottom up, right to left (although
3526 * left-to-right works OK too).
3528 * Note that at recovery time, journal replay occurs *before* the restart of
3529 * truncate against the orphan inode list.
3531 * The committed inode has the new, desired i_size (which is the same as
3532 * i_disksize in this case). After a crash, ext4_orphan_cleanup() will see
3533 * that this inode's truncate did not complete and it will again call
3534 * ext4_truncate() to have another go. So there will be instantiated blocks
3535 * to the right of the truncation point in a crashed ext4 filesystem. But
3536 * that's fine - as long as they are linked from the inode, the post-crash
3537 * ext4_truncate() run will find them and release them.
3539 void ext4_truncate(struct inode
*inode
)
3541 trace_ext4_truncate_enter(inode
);
3543 if (!ext4_can_truncate(inode
))
3546 ext4_clear_inode_flag(inode
, EXT4_INODE_EOFBLOCKS
);
3548 if (inode
->i_size
== 0 && !test_opt(inode
->i_sb
, NO_AUTO_DA_ALLOC
))
3549 ext4_set_inode_state(inode
, EXT4_STATE_DA_ALLOC_CLOSE
);
3551 if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))
3552 ext4_ext_truncate(inode
);
3554 ext4_ind_truncate(inode
);
3556 trace_ext4_truncate_exit(inode
);
3560 * ext4_get_inode_loc returns with an extra refcount against the inode's
3561 * underlying buffer_head on success. If 'in_mem' is true, we have all
3562 * data in memory that is needed to recreate the on-disk version of this
3565 static int __ext4_get_inode_loc(struct inode
*inode
,
3566 struct ext4_iloc
*iloc
, int in_mem
)
3568 struct ext4_group_desc
*gdp
;
3569 struct buffer_head
*bh
;
3570 struct super_block
*sb
= inode
->i_sb
;
3572 int inodes_per_block
, inode_offset
;
3575 if (!ext4_valid_inum(sb
, inode
->i_ino
))
3578 iloc
->block_group
= (inode
->i_ino
- 1) / EXT4_INODES_PER_GROUP(sb
);
3579 gdp
= ext4_get_group_desc(sb
, iloc
->block_group
, NULL
);
3584 * Figure out the offset within the block group inode table
3586 inodes_per_block
= EXT4_SB(sb
)->s_inodes_per_block
;
3587 inode_offset
= ((inode
->i_ino
- 1) %
3588 EXT4_INODES_PER_GROUP(sb
));
3589 block
= ext4_inode_table(sb
, gdp
) + (inode_offset
/ inodes_per_block
);
3590 iloc
->offset
= (inode_offset
% inodes_per_block
) * EXT4_INODE_SIZE(sb
);
3592 bh
= sb_getblk(sb
, block
);
3594 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3595 "unable to read itable block");
3598 if (!buffer_uptodate(bh
)) {
3602 * If the buffer has the write error flag, we have failed
3603 * to write out another inode in the same block. In this
3604 * case, we don't have to read the block because we may
3605 * read the old inode data successfully.
3607 if (buffer_write_io_error(bh
) && !buffer_uptodate(bh
))
3608 set_buffer_uptodate(bh
);
3610 if (buffer_uptodate(bh
)) {
3611 /* someone brought it uptodate while we waited */
3617 * If we have all information of the inode in memory and this
3618 * is the only valid inode in the block, we need not read the
3622 struct buffer_head
*bitmap_bh
;
3625 start
= inode_offset
& ~(inodes_per_block
- 1);
3627 /* Is the inode bitmap in cache? */
3628 bitmap_bh
= sb_getblk(sb
, ext4_inode_bitmap(sb
, gdp
));
3633 * If the inode bitmap isn't in cache then the
3634 * optimisation may end up performing two reads instead
3635 * of one, so skip it.
3637 if (!buffer_uptodate(bitmap_bh
)) {
3641 for (i
= start
; i
< start
+ inodes_per_block
; i
++) {
3642 if (i
== inode_offset
)
3644 if (ext4_test_bit(i
, bitmap_bh
->b_data
))
3648 if (i
== start
+ inodes_per_block
) {
3649 /* all other inodes are free, so skip I/O */
3650 memset(bh
->b_data
, 0, bh
->b_size
);
3651 set_buffer_uptodate(bh
);
3659 * If we need to do any I/O, try to pre-readahead extra
3660 * blocks from the inode table.
3662 if (EXT4_SB(sb
)->s_inode_readahead_blks
) {
3663 ext4_fsblk_t b
, end
, table
;
3666 table
= ext4_inode_table(sb
, gdp
);
3667 /* s_inode_readahead_blks is always a power of 2 */
3668 b
= block
& ~(EXT4_SB(sb
)->s_inode_readahead_blks
-1);
3671 end
= b
+ EXT4_SB(sb
)->s_inode_readahead_blks
;
3672 num
= EXT4_INODES_PER_GROUP(sb
);
3673 if (ext4_has_group_desc_csum(sb
))
3674 num
-= ext4_itable_unused_count(sb
, gdp
);
3675 table
+= num
/ inodes_per_block
;
3679 sb_breadahead(sb
, b
++);
3683 * There are other valid inodes in the buffer, this inode
3684 * has in-inode xattrs, or we don't have this inode in memory.
3685 * Read the block from disk.
3687 trace_ext4_load_inode(inode
);
3689 bh
->b_end_io
= end_buffer_read_sync
;
3690 submit_bh(READ
| REQ_META
| REQ_PRIO
, bh
);
3692 if (!buffer_uptodate(bh
)) {
3693 EXT4_ERROR_INODE_BLOCK(inode
, block
,
3694 "unable to read itable block");
3704 int ext4_get_inode_loc(struct inode
*inode
, struct ext4_iloc
*iloc
)
3706 /* We have all inode data except xattrs in memory here. */
3707 return __ext4_get_inode_loc(inode
, iloc
,
3708 !ext4_test_inode_state(inode
, EXT4_STATE_XATTR
));
3711 void ext4_set_inode_flags(struct inode
*inode
)
3713 unsigned int flags
= EXT4_I(inode
)->i_flags
;
3715 inode
->i_flags
&= ~(S_SYNC
|S_APPEND
|S_IMMUTABLE
|S_NOATIME
|S_DIRSYNC
);
3716 if (flags
& EXT4_SYNC_FL
)
3717 inode
->i_flags
|= S_SYNC
;
3718 if (flags
& EXT4_APPEND_FL
)
3719 inode
->i_flags
|= S_APPEND
;
3720 if (flags
& EXT4_IMMUTABLE_FL
)
3721 inode
->i_flags
|= S_IMMUTABLE
;
3722 if (flags
& EXT4_NOATIME_FL
)
3723 inode
->i_flags
|= S_NOATIME
;
3724 if (flags
& EXT4_DIRSYNC_FL
)
3725 inode
->i_flags
|= S_DIRSYNC
;
3728 /* Propagate flags from i_flags to EXT4_I(inode)->i_flags */
3729 void ext4_get_inode_flags(struct ext4_inode_info
*ei
)
3731 unsigned int vfs_fl
;
3732 unsigned long old_fl
, new_fl
;
3735 vfs_fl
= ei
->vfs_inode
.i_flags
;
3736 old_fl
= ei
->i_flags
;
3737 new_fl
= old_fl
& ~(EXT4_SYNC_FL
|EXT4_APPEND_FL
|
3738 EXT4_IMMUTABLE_FL
|EXT4_NOATIME_FL
|
3740 if (vfs_fl
& S_SYNC
)
3741 new_fl
|= EXT4_SYNC_FL
;
3742 if (vfs_fl
& S_APPEND
)
3743 new_fl
|= EXT4_APPEND_FL
;
3744 if (vfs_fl
& S_IMMUTABLE
)
3745 new_fl
|= EXT4_IMMUTABLE_FL
;
3746 if (vfs_fl
& S_NOATIME
)
3747 new_fl
|= EXT4_NOATIME_FL
;
3748 if (vfs_fl
& S_DIRSYNC
)
3749 new_fl
|= EXT4_DIRSYNC_FL
;
3750 } while (cmpxchg(&ei
->i_flags
, old_fl
, new_fl
) != old_fl
);
3753 static blkcnt_t
ext4_inode_blocks(struct ext4_inode
*raw_inode
,
3754 struct ext4_inode_info
*ei
)
3757 struct inode
*inode
= &(ei
->vfs_inode
);
3758 struct super_block
*sb
= inode
->i_sb
;
3760 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3761 EXT4_FEATURE_RO_COMPAT_HUGE_FILE
)) {
3762 /* we are using combined 48 bit field */
3763 i_blocks
= ((u64
)le16_to_cpu(raw_inode
->i_blocks_high
)) << 32 |
3764 le32_to_cpu(raw_inode
->i_blocks_lo
);
3765 if (ext4_test_inode_flag(inode
, EXT4_INODE_HUGE_FILE
)) {
3766 /* i_blocks represent file system block size */
3767 return i_blocks
<< (inode
->i_blkbits
- 9);
3772 return le32_to_cpu(raw_inode
->i_blocks_lo
);
3776 struct inode
*ext4_iget(struct super_block
*sb
, unsigned long ino
)
3778 struct ext4_iloc iloc
;
3779 struct ext4_inode
*raw_inode
;
3780 struct ext4_inode_info
*ei
;
3781 struct inode
*inode
;
3782 journal_t
*journal
= EXT4_SB(sb
)->s_journal
;
3788 inode
= iget_locked(sb
, ino
);
3790 return ERR_PTR(-ENOMEM
);
3791 if (!(inode
->i_state
& I_NEW
))
3797 ret
= __ext4_get_inode_loc(inode
, &iloc
, 0);
3800 raw_inode
= ext4_raw_inode(&iloc
);
3802 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3803 ei
->i_extra_isize
= le16_to_cpu(raw_inode
->i_extra_isize
);
3804 if (EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
>
3805 EXT4_INODE_SIZE(inode
->i_sb
)) {
3806 EXT4_ERROR_INODE(inode
, "bad extra_isize (%u != %u)",
3807 EXT4_GOOD_OLD_INODE_SIZE
+ ei
->i_extra_isize
,
3808 EXT4_INODE_SIZE(inode
->i_sb
));
3813 ei
->i_extra_isize
= 0;
3815 /* Precompute checksum seed for inode metadata */
3816 if (EXT4_HAS_RO_COMPAT_FEATURE(sb
,
3817 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM
)) {
3818 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
3820 __le32 inum
= cpu_to_le32(inode
->i_ino
);
3821 __le32 gen
= raw_inode
->i_generation
;
3822 csum
= ext4_chksum(sbi
, sbi
->s_csum_seed
, (__u8
*)&inum
,
3824 ei
->i_csum_seed
= ext4_chksum(sbi
, csum
, (__u8
*)&gen
,
3828 if (!ext4_inode_csum_verify(inode
, raw_inode
, ei
)) {
3829 EXT4_ERROR_INODE(inode
, "checksum invalid");
3834 inode
->i_mode
= le16_to_cpu(raw_inode
->i_mode
);
3835 i_uid
= (uid_t
)le16_to_cpu(raw_inode
->i_uid_low
);
3836 i_gid
= (gid_t
)le16_to_cpu(raw_inode
->i_gid_low
);
3837 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
3838 i_uid
|= le16_to_cpu(raw_inode
->i_uid_high
) << 16;
3839 i_gid
|= le16_to_cpu(raw_inode
->i_gid_high
) << 16;
3841 i_uid_write(inode
, i_uid
);
3842 i_gid_write(inode
, i_gid
);
3843 set_nlink(inode
, le16_to_cpu(raw_inode
->i_links_count
));
3845 ext4_clear_state_flags(ei
); /* Only relevant on 32-bit archs */
3846 ei
->i_dir_start_lookup
= 0;
3847 ei
->i_dtime
= le32_to_cpu(raw_inode
->i_dtime
);
3848 /* We now have enough fields to check if the inode was active or not.
3849 * This is needed because nfsd might try to access dead inodes
3850 * the test is that same one that e2fsck uses
3851 * NeilBrown 1999oct15
3853 if (inode
->i_nlink
== 0) {
3854 if (inode
->i_mode
== 0 ||
3855 !(EXT4_SB(inode
->i_sb
)->s_mount_state
& EXT4_ORPHAN_FS
)) {
3856 /* this inode is deleted */
3860 /* The only unlinked inodes we let through here have
3861 * valid i_mode and are being read by the orphan
3862 * recovery code: that's fine, we're about to complete
3863 * the process of deleting those. */
3865 ei
->i_flags
= le32_to_cpu(raw_inode
->i_flags
);
3866 inode
->i_blocks
= ext4_inode_blocks(raw_inode
, ei
);
3867 ei
->i_file_acl
= le32_to_cpu(raw_inode
->i_file_acl_lo
);
3868 if (EXT4_HAS_INCOMPAT_FEATURE(sb
, EXT4_FEATURE_INCOMPAT_64BIT
))
3870 ((__u64
)le16_to_cpu(raw_inode
->i_file_acl_high
)) << 32;
3871 inode
->i_size
= ext4_isize(raw_inode
);
3872 ei
->i_disksize
= inode
->i_size
;
3874 ei
->i_reserved_quota
= 0;
3876 inode
->i_generation
= le32_to_cpu(raw_inode
->i_generation
);
3877 ei
->i_block_group
= iloc
.block_group
;
3878 ei
->i_last_alloc_group
= ~0;
3880 * NOTE! The in-memory inode i_data array is in little-endian order
3881 * even on big-endian machines: we do NOT byteswap the block numbers!
3883 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
3884 ei
->i_data
[block
] = raw_inode
->i_block
[block
];
3885 INIT_LIST_HEAD(&ei
->i_orphan
);
3888 * Set transaction id's of transactions that have to be committed
3889 * to finish f[data]sync. We set them to currently running transaction
3890 * as we cannot be sure that the inode or some of its metadata isn't
3891 * part of the transaction - the inode could have been reclaimed and
3892 * now it is reread from disk.
3895 transaction_t
*transaction
;
3898 read_lock(&journal
->j_state_lock
);
3899 if (journal
->j_running_transaction
)
3900 transaction
= journal
->j_running_transaction
;
3902 transaction
= journal
->j_committing_transaction
;
3904 tid
= transaction
->t_tid
;
3906 tid
= journal
->j_commit_sequence
;
3907 read_unlock(&journal
->j_state_lock
);
3908 ei
->i_sync_tid
= tid
;
3909 ei
->i_datasync_tid
= tid
;
3912 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3913 if (ei
->i_extra_isize
== 0) {
3914 /* The extra space is currently unused. Use it. */
3915 ei
->i_extra_isize
= sizeof(struct ext4_inode
) -
3916 EXT4_GOOD_OLD_INODE_SIZE
;
3918 __le32
*magic
= (void *)raw_inode
+
3919 EXT4_GOOD_OLD_INODE_SIZE
+
3921 if (*magic
== cpu_to_le32(EXT4_XATTR_MAGIC
))
3922 ext4_set_inode_state(inode
, EXT4_STATE_XATTR
);
3926 EXT4_INODE_GET_XTIME(i_ctime
, inode
, raw_inode
);
3927 EXT4_INODE_GET_XTIME(i_mtime
, inode
, raw_inode
);
3928 EXT4_INODE_GET_XTIME(i_atime
, inode
, raw_inode
);
3929 EXT4_EINODE_GET_XTIME(i_crtime
, ei
, raw_inode
);
3931 inode
->i_version
= le32_to_cpu(raw_inode
->i_disk_version
);
3932 if (EXT4_INODE_SIZE(inode
->i_sb
) > EXT4_GOOD_OLD_INODE_SIZE
) {
3933 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
3935 (__u64
)(le32_to_cpu(raw_inode
->i_version_hi
)) << 32;
3939 if (ei
->i_file_acl
&&
3940 !ext4_data_block_valid(EXT4_SB(sb
), ei
->i_file_acl
, 1)) {
3941 EXT4_ERROR_INODE(inode
, "bad extended attribute block %llu",
3945 } else if (ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)) {
3946 if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
3947 (S_ISLNK(inode
->i_mode
) &&
3948 !ext4_inode_is_fast_symlink(inode
)))
3949 /* Validate extent which is part of inode */
3950 ret
= ext4_ext_check_inode(inode
);
3951 } else if (S_ISREG(inode
->i_mode
) || S_ISDIR(inode
->i_mode
) ||
3952 (S_ISLNK(inode
->i_mode
) &&
3953 !ext4_inode_is_fast_symlink(inode
))) {
3954 /* Validate block references which are part of inode */
3955 ret
= ext4_ind_check_inode(inode
);
3960 if (S_ISREG(inode
->i_mode
)) {
3961 inode
->i_op
= &ext4_file_inode_operations
;
3962 inode
->i_fop
= &ext4_file_operations
;
3963 ext4_set_aops(inode
);
3964 } else if (S_ISDIR(inode
->i_mode
)) {
3965 inode
->i_op
= &ext4_dir_inode_operations
;
3966 inode
->i_fop
= &ext4_dir_operations
;
3967 } else if (S_ISLNK(inode
->i_mode
)) {
3968 if (ext4_inode_is_fast_symlink(inode
)) {
3969 inode
->i_op
= &ext4_fast_symlink_inode_operations
;
3970 nd_terminate_link(ei
->i_data
, inode
->i_size
,
3971 sizeof(ei
->i_data
) - 1);
3973 inode
->i_op
= &ext4_symlink_inode_operations
;
3974 ext4_set_aops(inode
);
3976 } else if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
) ||
3977 S_ISFIFO(inode
->i_mode
) || S_ISSOCK(inode
->i_mode
)) {
3978 inode
->i_op
= &ext4_special_inode_operations
;
3979 if (raw_inode
->i_block
[0])
3980 init_special_inode(inode
, inode
->i_mode
,
3981 old_decode_dev(le32_to_cpu(raw_inode
->i_block
[0])));
3983 init_special_inode(inode
, inode
->i_mode
,
3984 new_decode_dev(le32_to_cpu(raw_inode
->i_block
[1])));
3987 EXT4_ERROR_INODE(inode
, "bogus i_mode (%o)", inode
->i_mode
);
3991 ext4_set_inode_flags(inode
);
3992 unlock_new_inode(inode
);
3998 return ERR_PTR(ret
);
4001 static int ext4_inode_blocks_set(handle_t
*handle
,
4002 struct ext4_inode
*raw_inode
,
4003 struct ext4_inode_info
*ei
)
4005 struct inode
*inode
= &(ei
->vfs_inode
);
4006 u64 i_blocks
= inode
->i_blocks
;
4007 struct super_block
*sb
= inode
->i_sb
;
4009 if (i_blocks
<= ~0U) {
4011 * i_blocks can be represnted in a 32 bit variable
4012 * as multiple of 512 bytes
4014 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4015 raw_inode
->i_blocks_high
= 0;
4016 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4019 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
, EXT4_FEATURE_RO_COMPAT_HUGE_FILE
))
4022 if (i_blocks
<= 0xffffffffffffULL
) {
4024 * i_blocks can be represented in a 48 bit variable
4025 * as multiple of 512 bytes
4027 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4028 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4029 ext4_clear_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4031 ext4_set_inode_flag(inode
, EXT4_INODE_HUGE_FILE
);
4032 /* i_block is stored in file system block size */
4033 i_blocks
= i_blocks
>> (inode
->i_blkbits
- 9);
4034 raw_inode
->i_blocks_lo
= cpu_to_le32(i_blocks
);
4035 raw_inode
->i_blocks_high
= cpu_to_le16(i_blocks
>> 32);
4041 * Post the struct inode info into an on-disk inode location in the
4042 * buffer-cache. This gobbles the caller's reference to the
4043 * buffer_head in the inode location struct.
4045 * The caller must have write access to iloc->bh.
4047 static int ext4_do_update_inode(handle_t
*handle
,
4048 struct inode
*inode
,
4049 struct ext4_iloc
*iloc
)
4051 struct ext4_inode
*raw_inode
= ext4_raw_inode(iloc
);
4052 struct ext4_inode_info
*ei
= EXT4_I(inode
);
4053 struct buffer_head
*bh
= iloc
->bh
;
4054 int err
= 0, rc
, block
;
4058 /* For fields not not tracking in the in-memory inode,
4059 * initialise them to zero for new inodes. */
4060 if (ext4_test_inode_state(inode
, EXT4_STATE_NEW
))
4061 memset(raw_inode
, 0, EXT4_SB(inode
->i_sb
)->s_inode_size
);
4063 ext4_get_inode_flags(ei
);
4064 raw_inode
->i_mode
= cpu_to_le16(inode
->i_mode
);
4065 i_uid
= i_uid_read(inode
);
4066 i_gid
= i_gid_read(inode
);
4067 if (!(test_opt(inode
->i_sb
, NO_UID32
))) {
4068 raw_inode
->i_uid_low
= cpu_to_le16(low_16_bits(i_uid
));
4069 raw_inode
->i_gid_low
= cpu_to_le16(low_16_bits(i_gid
));
4071 * Fix up interoperability with old kernels. Otherwise, old inodes get
4072 * re-used with the upper 16 bits of the uid/gid intact
4075 raw_inode
->i_uid_high
=
4076 cpu_to_le16(high_16_bits(i_uid
));
4077 raw_inode
->i_gid_high
=
4078 cpu_to_le16(high_16_bits(i_gid
));
4080 raw_inode
->i_uid_high
= 0;
4081 raw_inode
->i_gid_high
= 0;
4084 raw_inode
->i_uid_low
= cpu_to_le16(fs_high2lowuid(i_uid
));
4085 raw_inode
->i_gid_low
= cpu_to_le16(fs_high2lowgid(i_gid
));
4086 raw_inode
->i_uid_high
= 0;
4087 raw_inode
->i_gid_high
= 0;
4089 raw_inode
->i_links_count
= cpu_to_le16(inode
->i_nlink
);
4091 EXT4_INODE_SET_XTIME(i_ctime
, inode
, raw_inode
);
4092 EXT4_INODE_SET_XTIME(i_mtime
, inode
, raw_inode
);
4093 EXT4_INODE_SET_XTIME(i_atime
, inode
, raw_inode
);
4094 EXT4_EINODE_SET_XTIME(i_crtime
, ei
, raw_inode
);
4096 if (ext4_inode_blocks_set(handle
, raw_inode
, ei
))
4098 raw_inode
->i_dtime
= cpu_to_le32(ei
->i_dtime
);
4099 raw_inode
->i_flags
= cpu_to_le32(ei
->i_flags
& 0xFFFFFFFF);
4100 if (EXT4_SB(inode
->i_sb
)->s_es
->s_creator_os
!=
4101 cpu_to_le32(EXT4_OS_HURD
))
4102 raw_inode
->i_file_acl_high
=
4103 cpu_to_le16(ei
->i_file_acl
>> 32);
4104 raw_inode
->i_file_acl_lo
= cpu_to_le32(ei
->i_file_acl
);
4105 ext4_isize_set(raw_inode
, ei
->i_disksize
);
4106 if (ei
->i_disksize
> 0x7fffffffULL
) {
4107 struct super_block
*sb
= inode
->i_sb
;
4108 if (!EXT4_HAS_RO_COMPAT_FEATURE(sb
,
4109 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
) ||
4110 EXT4_SB(sb
)->s_es
->s_rev_level
==
4111 cpu_to_le32(EXT4_GOOD_OLD_REV
)) {
4112 /* If this is the first large file
4113 * created, add a flag to the superblock.
4115 err
= ext4_journal_get_write_access(handle
,
4116 EXT4_SB(sb
)->s_sbh
);
4119 ext4_update_dynamic_rev(sb
);
4120 EXT4_SET_RO_COMPAT_FEATURE(sb
,
4121 EXT4_FEATURE_RO_COMPAT_LARGE_FILE
);
4122 ext4_handle_sync(handle
);
4123 err
= ext4_handle_dirty_super(handle
, sb
);
4126 raw_inode
->i_generation
= cpu_to_le32(inode
->i_generation
);
4127 if (S_ISCHR(inode
->i_mode
) || S_ISBLK(inode
->i_mode
)) {
4128 if (old_valid_dev(inode
->i_rdev
)) {
4129 raw_inode
->i_block
[0] =
4130 cpu_to_le32(old_encode_dev(inode
->i_rdev
));
4131 raw_inode
->i_block
[1] = 0;
4133 raw_inode
->i_block
[0] = 0;
4134 raw_inode
->i_block
[1] =
4135 cpu_to_le32(new_encode_dev(inode
->i_rdev
));
4136 raw_inode
->i_block
[2] = 0;
4139 for (block
= 0; block
< EXT4_N_BLOCKS
; block
++)
4140 raw_inode
->i_block
[block
] = ei
->i_data
[block
];
4142 raw_inode
->i_disk_version
= cpu_to_le32(inode
->i_version
);
4143 if (ei
->i_extra_isize
) {
4144 if (EXT4_FITS_IN_INODE(raw_inode
, ei
, i_version_hi
))
4145 raw_inode
->i_version_hi
=
4146 cpu_to_le32(inode
->i_version
>> 32);
4147 raw_inode
->i_extra_isize
= cpu_to_le16(ei
->i_extra_isize
);
4150 ext4_inode_csum_set(inode
, raw_inode
, ei
);
4152 BUFFER_TRACE(bh
, "call ext4_handle_dirty_metadata");
4153 rc
= ext4_handle_dirty_metadata(handle
, NULL
, bh
);
4156 ext4_clear_inode_state(inode
, EXT4_STATE_NEW
);
4158 ext4_update_inode_fsync_trans(handle
, inode
, 0);
4161 ext4_std_error(inode
->i_sb
, err
);
4166 * ext4_write_inode()
4168 * We are called from a few places:
4170 * - Within generic_file_write() for O_SYNC files.
4171 * Here, there will be no transaction running. We wait for any running
4172 * trasnaction to commit.
4174 * - Within sys_sync(), kupdate and such.
4175 * We wait on commit, if tol to.
4177 * - Within prune_icache() (PF_MEMALLOC == true)
4178 * Here we simply return. We can't afford to block kswapd on the
4181 * In all cases it is actually safe for us to return without doing anything,
4182 * because the inode has been copied into a raw inode buffer in
4183 * ext4_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
4186 * Note that we are absolutely dependent upon all inode dirtiers doing the
4187 * right thing: they *must* call mark_inode_dirty() after dirtying info in
4188 * which we are interested.
4190 * It would be a bug for them to not do this. The code:
4192 * mark_inode_dirty(inode)
4194 * inode->i_size = expr;
4196 * is in error because a kswapd-driven write_inode() could occur while
4197 * `stuff()' is running, and the new i_size will be lost. Plus the inode
4198 * will no longer be on the superblock's dirty inode list.
4200 int ext4_write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
4204 if (current
->flags
& PF_MEMALLOC
)
4207 if (EXT4_SB(inode
->i_sb
)->s_journal
) {
4208 if (ext4_journal_current_handle()) {
4209 jbd_debug(1, "called recursively, non-PF_MEMALLOC!\n");
4214 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
4217 err
= ext4_force_commit(inode
->i_sb
);
4219 struct ext4_iloc iloc
;
4221 err
= __ext4_get_inode_loc(inode
, &iloc
, 0);
4224 if (wbc
->sync_mode
== WB_SYNC_ALL
)
4225 sync_dirty_buffer(iloc
.bh
);
4226 if (buffer_req(iloc
.bh
) && !buffer_uptodate(iloc
.bh
)) {
4227 EXT4_ERROR_INODE_BLOCK(inode
, iloc
.bh
->b_blocknr
,
4228 "IO error syncing inode");
4239 * Called from notify_change.
4241 * We want to trap VFS attempts to truncate the file as soon as
4242 * possible. In particular, we want to make sure that when the VFS
4243 * shrinks i_size, we put the inode on the orphan list and modify
4244 * i_disksize immediately, so that during the subsequent flushing of
4245 * dirty pages and freeing of disk blocks, we can guarantee that any
4246 * commit will leave the blocks being flushed in an unused state on
4247 * disk. (On recovery, the inode will get truncated and the blocks will
4248 * be freed, so we have a strong guarantee that no future commit will
4249 * leave these blocks visible to the user.)
4251 * Another thing we have to assure is that if we are in ordered mode
4252 * and inode is still attached to the committing transaction, we must
4253 * we start writeout of all the dirty pages which are being truncated.
4254 * This way we are sure that all the data written in the previous
4255 * transaction are already on disk (truncate waits for pages under
4258 * Called with inode->i_mutex down.
4260 int ext4_setattr(struct dentry
*dentry
, struct iattr
*attr
)
4262 struct inode
*inode
= dentry
->d_inode
;
4265 const unsigned int ia_valid
= attr
->ia_valid
;
4267 error
= inode_change_ok(inode
, attr
);
4271 if (is_quota_modification(inode
, attr
))
4272 dquot_initialize(inode
);
4273 if ((ia_valid
& ATTR_UID
&& !uid_eq(attr
->ia_uid
, inode
->i_uid
)) ||
4274 (ia_valid
& ATTR_GID
&& !gid_eq(attr
->ia_gid
, inode
->i_gid
))) {
4277 /* (user+group)*(old+new) structure, inode write (sb,
4278 * inode block, ? - but truncate inode update has it) */
4279 handle
= ext4_journal_start(inode
, (EXT4_MAXQUOTAS_INIT_BLOCKS(inode
->i_sb
)+
4280 EXT4_MAXQUOTAS_DEL_BLOCKS(inode
->i_sb
))+3);
4281 if (IS_ERR(handle
)) {
4282 error
= PTR_ERR(handle
);
4285 error
= dquot_transfer(inode
, attr
);
4287 ext4_journal_stop(handle
);
4290 /* Update corresponding info in inode so that everything is in
4291 * one transaction */
4292 if (attr
->ia_valid
& ATTR_UID
)
4293 inode
->i_uid
= attr
->ia_uid
;
4294 if (attr
->ia_valid
& ATTR_GID
)
4295 inode
->i_gid
= attr
->ia_gid
;
4296 error
= ext4_mark_inode_dirty(handle
, inode
);
4297 ext4_journal_stop(handle
);
4300 if (attr
->ia_valid
& ATTR_SIZE
) {
4301 inode_dio_wait(inode
);
4303 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
))) {
4304 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4306 if (attr
->ia_size
> sbi
->s_bitmap_maxbytes
)
4311 if (S_ISREG(inode
->i_mode
) &&
4312 attr
->ia_valid
& ATTR_SIZE
&&
4313 (attr
->ia_size
< inode
->i_size
)) {
4316 handle
= ext4_journal_start(inode
, 3);
4317 if (IS_ERR(handle
)) {
4318 error
= PTR_ERR(handle
);
4321 if (ext4_handle_valid(handle
)) {
4322 error
= ext4_orphan_add(handle
, inode
);
4325 EXT4_I(inode
)->i_disksize
= attr
->ia_size
;
4326 rc
= ext4_mark_inode_dirty(handle
, inode
);
4329 ext4_journal_stop(handle
);
4331 if (ext4_should_order_data(inode
)) {
4332 error
= ext4_begin_ordered_truncate(inode
,
4335 /* Do as much error cleanup as possible */
4336 handle
= ext4_journal_start(inode
, 3);
4337 if (IS_ERR(handle
)) {
4338 ext4_orphan_del(NULL
, inode
);
4341 ext4_orphan_del(handle
, inode
);
4343 ext4_journal_stop(handle
);
4349 if (attr
->ia_valid
& ATTR_SIZE
) {
4350 if (attr
->ia_size
!= i_size_read(inode
))
4351 truncate_setsize(inode
, attr
->ia_size
);
4352 ext4_truncate(inode
);
4356 setattr_copy(inode
, attr
);
4357 mark_inode_dirty(inode
);
4361 * If the call to ext4_truncate failed to get a transaction handle at
4362 * all, we need to clean up the in-core orphan list manually.
4364 if (orphan
&& inode
->i_nlink
)
4365 ext4_orphan_del(NULL
, inode
);
4367 if (!rc
&& (ia_valid
& ATTR_MODE
))
4368 rc
= ext4_acl_chmod(inode
);
4371 ext4_std_error(inode
->i_sb
, error
);
4377 int ext4_getattr(struct vfsmount
*mnt
, struct dentry
*dentry
,
4380 struct inode
*inode
;
4381 unsigned long delalloc_blocks
;
4383 inode
= dentry
->d_inode
;
4384 generic_fillattr(inode
, stat
);
4387 * We can't update i_blocks if the block allocation is delayed
4388 * otherwise in the case of system crash before the real block
4389 * allocation is done, we will have i_blocks inconsistent with
4390 * on-disk file blocks.
4391 * We always keep i_blocks updated together with real
4392 * allocation. But to not confuse with user, stat
4393 * will return the blocks that include the delayed allocation
4394 * blocks for this file.
4396 delalloc_blocks
= EXT4_C2B(EXT4_SB(inode
->i_sb
),
4397 EXT4_I(inode
)->i_reserved_data_blocks
);
4399 stat
->blocks
+= (delalloc_blocks
<< inode
->i_sb
->s_blocksize_bits
)>>9;
4403 static int ext4_index_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4405 if (!(ext4_test_inode_flag(inode
, EXT4_INODE_EXTENTS
)))
4406 return ext4_ind_trans_blocks(inode
, nrblocks
, chunk
);
4407 return ext4_ext_index_trans_blocks(inode
, nrblocks
, chunk
);
4411 * Account for index blocks, block groups bitmaps and block group
4412 * descriptor blocks if modify datablocks and index blocks
4413 * worse case, the indexs blocks spread over different block groups
4415 * If datablocks are discontiguous, they are possible to spread over
4416 * different block groups too. If they are contiuguous, with flexbg,
4417 * they could still across block group boundary.
4419 * Also account for superblock, inode, quota and xattr blocks
4421 static int ext4_meta_trans_blocks(struct inode
*inode
, int nrblocks
, int chunk
)
4423 ext4_group_t groups
, ngroups
= ext4_get_groups_count(inode
->i_sb
);
4429 * How many index blocks need to touch to modify nrblocks?
4430 * The "Chunk" flag indicating whether the nrblocks is
4431 * physically contiguous on disk
4433 * For Direct IO and fallocate, they calls get_block to allocate
4434 * one single extent at a time, so they could set the "Chunk" flag
4436 idxblocks
= ext4_index_trans_blocks(inode
, nrblocks
, chunk
);
4441 * Now let's see how many group bitmaps and group descriptors need
4451 if (groups
> ngroups
)
4453 if (groups
> EXT4_SB(inode
->i_sb
)->s_gdb_count
)
4454 gdpblocks
= EXT4_SB(inode
->i_sb
)->s_gdb_count
;
4456 /* bitmaps and block group descriptor blocks */
4457 ret
+= groups
+ gdpblocks
;
4459 /* Blocks for super block, inode, quota and xattr blocks */
4460 ret
+= EXT4_META_TRANS_BLOCKS(inode
->i_sb
);
4466 * Calculate the total number of credits to reserve to fit
4467 * the modification of a single pages into a single transaction,
4468 * which may include multiple chunks of block allocations.
4470 * This could be called via ext4_write_begin()
4472 * We need to consider the worse case, when
4473 * one new block per extent.
4475 int ext4_writepage_trans_blocks(struct inode
*inode
)
4477 int bpp
= ext4_journal_blocks_per_page(inode
);
4480 ret
= ext4_meta_trans_blocks(inode
, bpp
, 0);
4482 /* Account for data blocks for journalled mode */
4483 if (ext4_should_journal_data(inode
))
4489 * Calculate the journal credits for a chunk of data modification.
4491 * This is called from DIO, fallocate or whoever calling
4492 * ext4_map_blocks() to map/allocate a chunk of contiguous disk blocks.
4494 * journal buffers for data blocks are not included here, as DIO
4495 * and fallocate do no need to journal data buffers.
4497 int ext4_chunk_trans_blocks(struct inode
*inode
, int nrblocks
)
4499 return ext4_meta_trans_blocks(inode
, nrblocks
, 1);
4503 * The caller must have previously called ext4_reserve_inode_write().
4504 * Give this, we know that the caller already has write access to iloc->bh.
4506 int ext4_mark_iloc_dirty(handle_t
*handle
,
4507 struct inode
*inode
, struct ext4_iloc
*iloc
)
4511 if (IS_I_VERSION(inode
))
4512 inode_inc_iversion(inode
);
4514 /* the do_update_inode consumes one bh->b_count */
4517 /* ext4_do_update_inode() does jbd2_journal_dirty_metadata */
4518 err
= ext4_do_update_inode(handle
, inode
, iloc
);
4524 * On success, We end up with an outstanding reference count against
4525 * iloc->bh. This _must_ be cleaned up later.
4529 ext4_reserve_inode_write(handle_t
*handle
, struct inode
*inode
,
4530 struct ext4_iloc
*iloc
)
4534 err
= ext4_get_inode_loc(inode
, iloc
);
4536 BUFFER_TRACE(iloc
->bh
, "get_write_access");
4537 err
= ext4_journal_get_write_access(handle
, iloc
->bh
);
4543 ext4_std_error(inode
->i_sb
, err
);
4548 * Expand an inode by new_extra_isize bytes.
4549 * Returns 0 on success or negative error number on failure.
4551 static int ext4_expand_extra_isize(struct inode
*inode
,
4552 unsigned int new_extra_isize
,
4553 struct ext4_iloc iloc
,
4556 struct ext4_inode
*raw_inode
;
4557 struct ext4_xattr_ibody_header
*header
;
4559 if (EXT4_I(inode
)->i_extra_isize
>= new_extra_isize
)
4562 raw_inode
= ext4_raw_inode(&iloc
);
4564 header
= IHDR(inode
, raw_inode
);
4566 /* No extended attributes present */
4567 if (!ext4_test_inode_state(inode
, EXT4_STATE_XATTR
) ||
4568 header
->h_magic
!= cpu_to_le32(EXT4_XATTR_MAGIC
)) {
4569 memset((void *)raw_inode
+ EXT4_GOOD_OLD_INODE_SIZE
, 0,
4571 EXT4_I(inode
)->i_extra_isize
= new_extra_isize
;
4575 /* try to expand with EAs present */
4576 return ext4_expand_extra_isize_ea(inode
, new_extra_isize
,
4581 * What we do here is to mark the in-core inode as clean with respect to inode
4582 * dirtiness (it may still be data-dirty).
4583 * This means that the in-core inode may be reaped by prune_icache
4584 * without having to perform any I/O. This is a very good thing,
4585 * because *any* task may call prune_icache - even ones which
4586 * have a transaction open against a different journal.
4588 * Is this cheating? Not really. Sure, we haven't written the
4589 * inode out, but prune_icache isn't a user-visible syncing function.
4590 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
4591 * we start and wait on commits.
4593 * Is this efficient/effective? Well, we're being nice to the system
4594 * by cleaning up our inodes proactively so they can be reaped
4595 * without I/O. But we are potentially leaving up to five seconds'
4596 * worth of inodes floating about which prune_icache wants us to
4597 * write out. One way to fix that would be to get prune_icache()
4598 * to do a write_super() to free up some memory. It has the desired
4601 int ext4_mark_inode_dirty(handle_t
*handle
, struct inode
*inode
)
4603 struct ext4_iloc iloc
;
4604 struct ext4_sb_info
*sbi
= EXT4_SB(inode
->i_sb
);
4605 static unsigned int mnt_count
;
4609 trace_ext4_mark_inode_dirty(inode
, _RET_IP_
);
4610 err
= ext4_reserve_inode_write(handle
, inode
, &iloc
);
4611 if (ext4_handle_valid(handle
) &&
4612 EXT4_I(inode
)->i_extra_isize
< sbi
->s_want_extra_isize
&&
4613 !ext4_test_inode_state(inode
, EXT4_STATE_NO_EXPAND
)) {
4615 * We need extra buffer credits since we may write into EA block
4616 * with this same handle. If journal_extend fails, then it will
4617 * only result in a minor loss of functionality for that inode.
4618 * If this is felt to be critical, then e2fsck should be run to
4619 * force a large enough s_min_extra_isize.
4621 if ((jbd2_journal_extend(handle
,
4622 EXT4_DATA_TRANS_BLOCKS(inode
->i_sb
))) == 0) {
4623 ret
= ext4_expand_extra_isize(inode
,
4624 sbi
->s_want_extra_isize
,
4627 ext4_set_inode_state(inode
,
4628 EXT4_STATE_NO_EXPAND
);
4630 le16_to_cpu(sbi
->s_es
->s_mnt_count
)) {
4631 ext4_warning(inode
->i_sb
,
4632 "Unable to expand inode %lu. Delete"
4633 " some EAs or run e2fsck.",
4636 le16_to_cpu(sbi
->s_es
->s_mnt_count
);
4642 err
= ext4_mark_iloc_dirty(handle
, inode
, &iloc
);
4647 * ext4_dirty_inode() is called from __mark_inode_dirty()
4649 * We're really interested in the case where a file is being extended.
4650 * i_size has been changed by generic_commit_write() and we thus need
4651 * to include the updated inode in the current transaction.
4653 * Also, dquot_alloc_block() will always dirty the inode when blocks
4654 * are allocated to the file.
4656 * If the inode is marked synchronous, we don't honour that here - doing
4657 * so would cause a commit on atime updates, which we don't bother doing.
4658 * We handle synchronous inodes at the highest possible level.
4660 void ext4_dirty_inode(struct inode
*inode
, int flags
)
4664 handle
= ext4_journal_start(inode
, 2);
4668 ext4_mark_inode_dirty(handle
, inode
);
4670 ext4_journal_stop(handle
);
4677 * Bind an inode's backing buffer_head into this transaction, to prevent
4678 * it from being flushed to disk early. Unlike
4679 * ext4_reserve_inode_write, this leaves behind no bh reference and
4680 * returns no iloc structure, so the caller needs to repeat the iloc
4681 * lookup to mark the inode dirty later.
4683 static int ext4_pin_inode(handle_t
*handle
, struct inode
*inode
)
4685 struct ext4_iloc iloc
;
4689 err
= ext4_get_inode_loc(inode
, &iloc
);
4691 BUFFER_TRACE(iloc
.bh
, "get_write_access");
4692 err
= jbd2_journal_get_write_access(handle
, iloc
.bh
);
4694 err
= ext4_handle_dirty_metadata(handle
,
4700 ext4_std_error(inode
->i_sb
, err
);
4705 int ext4_change_inode_journal_flag(struct inode
*inode
, int val
)
4712 * We have to be very careful here: changing a data block's
4713 * journaling status dynamically is dangerous. If we write a
4714 * data block to the journal, change the status and then delete
4715 * that block, we risk forgetting to revoke the old log record
4716 * from the journal and so a subsequent replay can corrupt data.
4717 * So, first we make sure that the journal is empty and that
4718 * nobody is changing anything.
4721 journal
= EXT4_JOURNAL(inode
);
4724 if (is_journal_aborted(journal
))
4726 /* We have to allocate physical blocks for delalloc blocks
4727 * before flushing journal. otherwise delalloc blocks can not
4728 * be allocated any more. even more truncate on delalloc blocks
4729 * could trigger BUG by flushing delalloc blocks in journal.
4730 * There is no delalloc block in non-journal data mode.
4732 if (val
&& test_opt(inode
->i_sb
, DELALLOC
)) {
4733 err
= ext4_alloc_da_blocks(inode
);
4738 jbd2_journal_lock_updates(journal
);
4741 * OK, there are no updates running now, and all cached data is
4742 * synced to disk. We are now in a completely consistent state
4743 * which doesn't have anything in the journal, and we know that
4744 * no filesystem updates are running, so it is safe to modify
4745 * the inode's in-core data-journaling state flag now.
4749 ext4_set_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4751 jbd2_journal_flush(journal
);
4752 ext4_clear_inode_flag(inode
, EXT4_INODE_JOURNAL_DATA
);
4754 ext4_set_aops(inode
);
4756 jbd2_journal_unlock_updates(journal
);
4758 /* Finally we can mark the inode as dirty. */
4760 handle
= ext4_journal_start(inode
, 1);
4762 return PTR_ERR(handle
);
4764 err
= ext4_mark_inode_dirty(handle
, inode
);
4765 ext4_handle_sync(handle
);
4766 ext4_journal_stop(handle
);
4767 ext4_std_error(inode
->i_sb
, err
);
4772 static int ext4_bh_unmapped(handle_t
*handle
, struct buffer_head
*bh
)
4774 return !buffer_mapped(bh
);
4777 int ext4_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
4779 struct page
*page
= vmf
->page
;
4783 struct file
*file
= vma
->vm_file
;
4784 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
4785 struct address_space
*mapping
= inode
->i_mapping
;
4787 get_block_t
*get_block
;
4790 sb_start_pagefault(inode
->i_sb
);
4791 /* Delalloc case is easy... */
4792 if (test_opt(inode
->i_sb
, DELALLOC
) &&
4793 !ext4_should_journal_data(inode
) &&
4794 !ext4_nonda_switch(inode
->i_sb
)) {
4796 ret
= __block_page_mkwrite(vma
, vmf
,
4797 ext4_da_get_block_prep
);
4798 } while (ret
== -ENOSPC
&&
4799 ext4_should_retry_alloc(inode
->i_sb
, &retries
));
4804 size
= i_size_read(inode
);
4805 /* Page got truncated from under us? */
4806 if (page
->mapping
!= mapping
|| page_offset(page
) > size
) {
4808 ret
= VM_FAULT_NOPAGE
;
4812 if (page
->index
== size
>> PAGE_CACHE_SHIFT
)
4813 len
= size
& ~PAGE_CACHE_MASK
;
4815 len
= PAGE_CACHE_SIZE
;
4817 * Return if we have all the buffers mapped. This avoids the need to do
4818 * journal_start/journal_stop which can block and take a long time
4820 if (page_has_buffers(page
)) {
4821 if (!walk_page_buffers(NULL
, page_buffers(page
), 0, len
, NULL
,
4822 ext4_bh_unmapped
)) {
4823 /* Wait so that we don't change page under IO */
4824 wait_on_page_writeback(page
);
4825 ret
= VM_FAULT_LOCKED
;
4830 /* OK, we need to fill the hole... */
4831 if (ext4_should_dioread_nolock(inode
))
4832 get_block
= ext4_get_block_write
;
4834 get_block
= ext4_get_block
;
4836 handle
= ext4_journal_start(inode
, ext4_writepage_trans_blocks(inode
));
4837 if (IS_ERR(handle
)) {
4838 ret
= VM_FAULT_SIGBUS
;
4841 ret
= __block_page_mkwrite(vma
, vmf
, get_block
);
4842 if (!ret
&& ext4_should_journal_data(inode
)) {
4843 if (walk_page_buffers(handle
, page_buffers(page
), 0,
4844 PAGE_CACHE_SIZE
, NULL
, do_journal_get_write_access
)) {
4846 ret
= VM_FAULT_SIGBUS
;
4847 ext4_journal_stop(handle
);
4850 ext4_set_inode_state(inode
, EXT4_STATE_JDATA
);
4852 ext4_journal_stop(handle
);
4853 if (ret
== -ENOSPC
&& ext4_should_retry_alloc(inode
->i_sb
, &retries
))
4856 ret
= block_page_mkwrite_return(ret
);
4858 sb_end_pagefault(inode
->i_sb
);