2 * linux/fs/ext4/page-io.c
4 * This contains the new page_io functions for ext4
6 * Written by Theodore Ts'o, 2010.
10 #include <linux/time.h>
11 #include <linux/highuid.h>
12 #include <linux/pagemap.h>
13 #include <linux/quotaops.h>
14 #include <linux/string.h>
15 #include <linux/buffer_head.h>
16 #include <linux/writeback.h>
17 #include <linux/pagevec.h>
18 #include <linux/mpage.h>
19 #include <linux/namei.h>
20 #include <linux/uio.h>
21 #include <linux/bio.h>
22 #include <linux/workqueue.h>
23 #include <linux/kernel.h>
24 #include <linux/slab.h>
26 #include <linux/backing-dev.h>
28 #include "ext4_jbd2.h"
32 static struct kmem_cache
*io_end_cachep
;
34 int __init
ext4_init_pageio(void)
36 io_end_cachep
= KMEM_CACHE(ext4_io_end
, SLAB_RECLAIM_ACCOUNT
);
37 if (io_end_cachep
== NULL
)
42 void ext4_exit_pageio(void)
44 kmem_cache_destroy(io_end_cachep
);
48 * Print an buffer I/O error compatible with the fs/buffer.c. This
49 * provides compatibility with dmesg scrapers that look for a specific
50 * buffer I/O error message. We really need a unified error reporting
51 * structure to userspace ala Digital Unix's uerf system, but it's
52 * probably not going to happen in my lifetime, due to LKML politics...
54 static void buffer_io_error(struct buffer_head
*bh
)
56 char b
[BDEVNAME_SIZE
];
57 printk_ratelimited(KERN_ERR
"Buffer I/O error on device %s, logical block %llu\n",
58 bdevname(bh
->b_bdev
, b
),
59 (unsigned long long)bh
->b_blocknr
);
62 static void ext4_finish_bio(struct bio
*bio
)
67 bio_for_each_segment_all(bvec
, bio
, i
) {
68 struct page
*page
= bvec
->bv_page
;
69 #ifdef CONFIG_EXT4_FS_ENCRYPTION
70 struct page
*data_page
= NULL
;
71 struct ext4_crypto_ctx
*ctx
= NULL
;
73 struct buffer_head
*bh
, *head
;
74 unsigned bio_start
= bvec
->bv_offset
;
75 unsigned bio_end
= bio_start
+ bvec
->bv_len
;
76 unsigned under_io
= 0;
82 #ifdef CONFIG_EXT4_FS_ENCRYPTION
84 /* The bounce data pages are unmapped. */
86 ctx
= (struct ext4_crypto_ctx
*)page_private(data_page
);
87 page
= ctx
->w
.control_page
;
93 set_bit(AS_EIO
, &page
->mapping
->flags
);
95 bh
= head
= page_buffers(page
);
97 * We check all buffers in the page under BH_Uptodate_Lock
98 * to avoid races with other end io clearing async_write flags
100 local_irq_save(flags
);
101 bit_spin_lock(BH_Uptodate_Lock
, &head
->b_state
);
103 if (bh_offset(bh
) < bio_start
||
104 bh_offset(bh
) + bh
->b_size
> bio_end
) {
105 if (buffer_async_write(bh
))
109 clear_buffer_async_write(bh
);
112 } while ((bh
= bh
->b_this_page
) != head
);
113 bit_spin_unlock(BH_Uptodate_Lock
, &head
->b_state
);
114 local_irq_restore(flags
);
116 #ifdef CONFIG_EXT4_FS_ENCRYPTION
118 ext4_restore_control_page(data_page
);
120 end_page_writeback(page
);
125 static void ext4_release_io_end(ext4_io_end_t
*io_end
)
127 struct bio
*bio
, *next_bio
;
129 BUG_ON(!list_empty(&io_end
->list
));
130 BUG_ON(io_end
->flag
& EXT4_IO_END_UNWRITTEN
);
131 WARN_ON(io_end
->handle
);
133 if (atomic_dec_and_test(&EXT4_I(io_end
->inode
)->i_ioend_count
))
134 wake_up_all(ext4_ioend_wq(io_end
->inode
));
136 for (bio
= io_end
->bio
; bio
; bio
= next_bio
) {
137 next_bio
= bio
->bi_private
;
138 ext4_finish_bio(bio
);
141 kmem_cache_free(io_end_cachep
, io_end
);
144 static void ext4_clear_io_unwritten_flag(ext4_io_end_t
*io_end
)
146 struct inode
*inode
= io_end
->inode
;
148 io_end
->flag
&= ~EXT4_IO_END_UNWRITTEN
;
149 /* Wake up anyone waiting on unwritten extent conversion */
150 if (atomic_dec_and_test(&EXT4_I(inode
)->i_unwritten
))
151 wake_up_all(ext4_ioend_wq(inode
));
155 * Check a range of space and convert unwritten extents to written. Note that
156 * we are protected from truncate touching same part of extent tree by the
157 * fact that truncate code waits for all DIO to finish (thus exclusion from
158 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
159 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
160 * completed (happens from ext4_free_ioend()).
162 static int ext4_end_io(ext4_io_end_t
*io
)
164 struct inode
*inode
= io
->inode
;
165 loff_t offset
= io
->offset
;
166 ssize_t size
= io
->size
;
167 handle_t
*handle
= io
->handle
;
170 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
172 io
, inode
->i_ino
, io
->list
.next
, io
->list
.prev
);
174 io
->handle
= NULL
; /* Following call will use up the handle */
175 ret
= ext4_convert_unwritten_extents(handle
, inode
, offset
, size
);
177 ext4_msg(inode
->i_sb
, KERN_EMERG
,
178 "failed to convert unwritten extents to written "
179 "extents -- potential data loss! "
180 "(inode %lu, offset %llu, size %zd, error %d)",
181 inode
->i_ino
, offset
, size
, ret
);
183 ext4_clear_io_unwritten_flag(io
);
184 ext4_release_io_end(io
);
188 static void dump_completed_IO(struct inode
*inode
, struct list_head
*head
)
191 struct list_head
*cur
, *before
, *after
;
192 ext4_io_end_t
*io
, *io0
, *io1
;
194 if (list_empty(head
))
197 ext4_debug("Dump inode %lu completed io list\n", inode
->i_ino
);
198 list_for_each_entry(io
, head
, list
) {
201 io0
= container_of(before
, ext4_io_end_t
, list
);
203 io1
= container_of(after
, ext4_io_end_t
, list
);
205 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
206 io
, inode
->i_ino
, io0
, io1
);
211 /* Add the io_end to per-inode completed end_io list. */
212 static void ext4_add_complete_io(ext4_io_end_t
*io_end
)
214 struct ext4_inode_info
*ei
= EXT4_I(io_end
->inode
);
215 struct ext4_sb_info
*sbi
= EXT4_SB(io_end
->inode
->i_sb
);
216 struct workqueue_struct
*wq
;
219 /* Only reserved conversions from writeback should enter here */
220 WARN_ON(!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
));
221 WARN_ON(!io_end
->handle
&& sbi
->s_journal
);
222 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
223 wq
= sbi
->rsv_conversion_wq
;
224 if (list_empty(&ei
->i_rsv_conversion_list
))
225 queue_work(wq
, &ei
->i_rsv_conversion_work
);
226 list_add_tail(&io_end
->list
, &ei
->i_rsv_conversion_list
);
227 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
230 static int ext4_do_flush_completed_IO(struct inode
*inode
,
231 struct list_head
*head
)
234 struct list_head unwritten
;
236 struct ext4_inode_info
*ei
= EXT4_I(inode
);
239 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
240 dump_completed_IO(inode
, head
);
241 list_replace_init(head
, &unwritten
);
242 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
244 while (!list_empty(&unwritten
)) {
245 io
= list_entry(unwritten
.next
, ext4_io_end_t
, list
);
246 BUG_ON(!(io
->flag
& EXT4_IO_END_UNWRITTEN
));
247 list_del_init(&io
->list
);
249 err
= ext4_end_io(io
);
250 if (unlikely(!ret
&& err
))
257 * work on completed IO, to convert unwritten extents to extents
259 void ext4_end_io_rsv_work(struct work_struct
*work
)
261 struct ext4_inode_info
*ei
= container_of(work
, struct ext4_inode_info
,
262 i_rsv_conversion_work
);
263 ext4_do_flush_completed_IO(&ei
->vfs_inode
, &ei
->i_rsv_conversion_list
);
266 ext4_io_end_t
*ext4_init_io_end(struct inode
*inode
, gfp_t flags
)
268 ext4_io_end_t
*io
= kmem_cache_zalloc(io_end_cachep
, flags
);
270 atomic_inc(&EXT4_I(inode
)->i_ioend_count
);
272 INIT_LIST_HEAD(&io
->list
);
273 atomic_set(&io
->count
, 1);
278 void ext4_put_io_end_defer(ext4_io_end_t
*io_end
)
280 if (atomic_dec_and_test(&io_end
->count
)) {
281 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
) || !io_end
->size
) {
282 ext4_release_io_end(io_end
);
285 ext4_add_complete_io(io_end
);
289 int ext4_put_io_end(ext4_io_end_t
*io_end
)
293 if (atomic_dec_and_test(&io_end
->count
)) {
294 if (io_end
->flag
& EXT4_IO_END_UNWRITTEN
) {
295 err
= ext4_convert_unwritten_extents(io_end
->handle
,
296 io_end
->inode
, io_end
->offset
,
298 io_end
->handle
= NULL
;
299 ext4_clear_io_unwritten_flag(io_end
);
301 ext4_release_io_end(io_end
);
306 ext4_io_end_t
*ext4_get_io_end(ext4_io_end_t
*io_end
)
308 atomic_inc(&io_end
->count
);
312 /* BIO completion function for page writeback */
313 static void ext4_end_bio(struct bio
*bio
)
315 ext4_io_end_t
*io_end
= bio
->bi_private
;
316 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
319 bio
->bi_end_io
= NULL
;
322 struct inode
*inode
= io_end
->inode
;
324 ext4_warning(inode
->i_sb
, "I/O error %d writing to inode %lu "
325 "(offset %llu size %ld starting block %llu)",
326 bio
->bi_error
, inode
->i_ino
,
327 (unsigned long long) io_end
->offset
,
330 bi_sector
>> (inode
->i_blkbits
- 9));
331 mapping_set_error(inode
->i_mapping
, bio
->bi_error
);
334 if (io_end
->flag
& EXT4_IO_END_UNWRITTEN
) {
336 * Link bio into list hanging from io_end. We have to do it
337 * atomically as bio completions can be racing against each
340 bio
->bi_private
= xchg(&io_end
->bio
, bio
);
341 ext4_put_io_end_defer(io_end
);
344 * Drop io_end reference early. Inode can get freed once
347 ext4_put_io_end_defer(io_end
);
348 ext4_finish_bio(bio
);
353 void ext4_io_submit(struct ext4_io_submit
*io
)
355 struct bio
*bio
= io
->io_bio
;
358 int io_op
= io
->io_wbc
->sync_mode
== WB_SYNC_ALL
?
361 submit_bio(io_op
, io
->io_bio
);
367 void ext4_io_submit_init(struct ext4_io_submit
*io
,
368 struct writeback_control
*wbc
)
375 static int io_submit_init_bio(struct ext4_io_submit
*io
,
376 struct buffer_head
*bh
)
380 bio
= bio_alloc(GFP_NOIO
, BIO_MAX_PAGES
);
383 wbc_init_bio(io
->io_wbc
, bio
);
384 bio
->bi_iter
.bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
385 bio
->bi_bdev
= bh
->b_bdev
;
386 bio
->bi_end_io
= ext4_end_bio
;
387 bio
->bi_private
= ext4_get_io_end(io
->io_end
);
389 io
->io_next_block
= bh
->b_blocknr
;
393 static int io_submit_add_bh(struct ext4_io_submit
*io
,
396 struct buffer_head
*bh
)
400 if (io
->io_bio
&& bh
->b_blocknr
!= io
->io_next_block
) {
404 if (io
->io_bio
== NULL
) {
405 ret
= io_submit_init_bio(io
, bh
);
409 ret
= bio_add_page(io
->io_bio
, page
, bh
->b_size
, bh_offset(bh
));
410 if (ret
!= bh
->b_size
)
411 goto submit_and_retry
;
412 wbc_account_io(io
->io_wbc
, page
, bh
->b_size
);
417 int ext4_bio_write_page(struct ext4_io_submit
*io
,
420 struct writeback_control
*wbc
,
423 struct page
*data_page
= NULL
;
424 struct inode
*inode
= page
->mapping
->host
;
425 unsigned block_start
, blocksize
;
426 struct buffer_head
*bh
, *head
;
428 int nr_submitted
= 0;
429 int nr_to_submit
= 0;
431 blocksize
= 1 << inode
->i_blkbits
;
433 BUG_ON(!PageLocked(page
));
434 BUG_ON(PageWriteback(page
));
437 set_page_writeback_keepwrite(page
);
439 set_page_writeback(page
);
440 ClearPageError(page
);
443 * Comments copied from block_write_full_page:
445 * The page straddles i_size. It must be zeroed out on each and every
446 * writepage invocation because it may be mmapped. "A file is mapped
447 * in multiples of the page size. For a file that is not a multiple of
448 * the page size, the remaining memory is zeroed when mapped, and
449 * writes to that region are not written out to the file."
451 if (len
< PAGE_CACHE_SIZE
)
452 zero_user_segment(page
, len
, PAGE_CACHE_SIZE
);
454 * In the first loop we prepare and mark buffers to submit. We have to
455 * mark all buffers in the page before submitting so that
456 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
457 * on the first buffer finishes and we are still working on submitting
460 bh
= head
= page_buffers(page
);
462 block_start
= bh_offset(bh
);
463 if (block_start
>= len
) {
464 clear_buffer_dirty(bh
);
465 set_buffer_uptodate(bh
);
468 if (!buffer_dirty(bh
) || buffer_delay(bh
) ||
469 !buffer_mapped(bh
) || buffer_unwritten(bh
)) {
470 /* A hole? We can safely clear the dirty bit */
471 if (!buffer_mapped(bh
))
472 clear_buffer_dirty(bh
);
477 if (buffer_new(bh
)) {
478 clear_buffer_new(bh
);
479 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
481 set_buffer_async_write(bh
);
483 } while ((bh
= bh
->b_this_page
) != head
);
485 bh
= head
= page_buffers(page
);
487 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
) &&
489 gfp_t gfp_flags
= GFP_NOFS
;
492 data_page
= ext4_encrypt(inode
, page
, gfp_flags
);
493 if (IS_ERR(data_page
)) {
494 ret
= PTR_ERR(data_page
);
495 if (ret
== -ENOMEM
&& wbc
->sync_mode
== WB_SYNC_ALL
) {
498 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
500 gfp_flags
|= __GFP_NOFAIL
;
508 /* Now submit buffers to write */
510 if (!buffer_async_write(bh
))
512 ret
= io_submit_add_bh(io
, inode
,
513 data_page
? data_page
: page
, bh
);
516 * We only get here on ENOMEM. Not much else
517 * we can do but mark the page as dirty, and
518 * better luck next time.
523 clear_buffer_dirty(bh
);
524 } while ((bh
= bh
->b_this_page
) != head
);
526 /* Error stopped previous loop? Clean up buffers... */
530 ext4_restore_control_page(data_page
);
531 printk_ratelimited(KERN_ERR
"%s: ret = %d\n", __func__
, ret
);
532 redirty_page_for_writepage(wbc
, page
);
534 clear_buffer_async_write(bh
);
535 bh
= bh
->b_this_page
;
536 } while (bh
!= head
);
539 /* Nothing submitted - we have to end page writeback */
541 end_page_writeback(page
);