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/jbd2.h>
12 #include <linux/highuid.h>
13 #include <linux/pagemap.h>
14 #include <linux/quotaops.h>
15 #include <linux/string.h>
16 #include <linux/buffer_head.h>
17 #include <linux/writeback.h>
18 #include <linux/pagevec.h>
19 #include <linux/mpage.h>
20 #include <linux/namei.h>
21 #include <linux/aio.h>
22 #include <linux/uio.h>
23 #include <linux/bio.h>
24 #include <linux/workqueue.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
29 #include "ext4_jbd2.h"
33 static struct kmem_cache
*io_end_cachep
;
35 int __init
ext4_init_pageio(void)
37 io_end_cachep
= KMEM_CACHE(ext4_io_end
, SLAB_RECLAIM_ACCOUNT
);
38 if (io_end_cachep
== NULL
)
43 void ext4_exit_pageio(void)
45 kmem_cache_destroy(io_end_cachep
);
49 * This function is called by ext4_evict_inode() to make sure there is
50 * no more pending I/O completion work left to do.
52 void ext4_ioend_shutdown(struct inode
*inode
)
54 wait_queue_head_t
*wq
= ext4_ioend_wq(inode
);
56 wait_event(*wq
, (atomic_read(&EXT4_I(inode
)->i_ioend_count
) == 0));
58 * We need to make sure the work structure is finished being
59 * used before we let the inode get destroyed.
61 if (work_pending(&EXT4_I(inode
)->i_unwritten_work
))
62 cancel_work_sync(&EXT4_I(inode
)->i_unwritten_work
);
65 static void ext4_release_io_end(ext4_io_end_t
*io_end
)
67 BUG_ON(!list_empty(&io_end
->list
));
68 BUG_ON(io_end
->flag
& EXT4_IO_END_UNWRITTEN
);
69 WARN_ON(io_end
->handle
);
71 if (atomic_dec_and_test(&EXT4_I(io_end
->inode
)->i_ioend_count
))
72 wake_up_all(ext4_ioend_wq(io_end
->inode
));
73 if (io_end
->flag
& EXT4_IO_END_DIRECT
)
74 inode_dio_done(io_end
->inode
);
76 aio_complete(io_end
->iocb
, io_end
->result
, 0);
77 kmem_cache_free(io_end_cachep
, io_end
);
80 static void ext4_clear_io_unwritten_flag(ext4_io_end_t
*io_end
)
82 struct inode
*inode
= io_end
->inode
;
84 io_end
->flag
&= ~EXT4_IO_END_UNWRITTEN
;
85 /* Wake up anyone waiting on unwritten extent conversion */
86 if (atomic_dec_and_test(&EXT4_I(inode
)->i_unwritten
))
87 wake_up_all(ext4_ioend_wq(inode
));
90 /* check a range of space and convert unwritten extents to written. */
91 static int ext4_end_io(ext4_io_end_t
*io
)
93 struct inode
*inode
= io
->inode
;
94 loff_t offset
= io
->offset
;
95 ssize_t size
= io
->size
;
96 handle_t
*handle
= io
->handle
;
99 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
101 io
, inode
->i_ino
, io
->list
.next
, io
->list
.prev
);
103 io
->handle
= NULL
; /* Following call will use up the handle */
104 ret
= ext4_convert_unwritten_extents(handle
, inode
, offset
, size
);
106 ext4_msg(inode
->i_sb
, KERN_EMERG
,
107 "failed to convert unwritten extents to written "
108 "extents -- potential data loss! "
109 "(inode %lu, offset %llu, size %zd, error %d)",
110 inode
->i_ino
, offset
, size
, ret
);
112 ext4_clear_io_unwritten_flag(io
);
113 ext4_release_io_end(io
);
117 static void dump_completed_IO(struct inode
*inode
)
120 struct list_head
*cur
, *before
, *after
;
121 ext4_io_end_t
*io
, *io0
, *io1
;
123 if (list_empty(&EXT4_I(inode
)->i_completed_io_list
)) {
124 ext4_debug("inode %lu completed_io list is empty\n",
129 ext4_debug("Dump inode %lu completed_io list\n", inode
->i_ino
);
130 list_for_each_entry(io
, &EXT4_I(inode
)->i_completed_io_list
, list
) {
133 io0
= container_of(before
, ext4_io_end_t
, list
);
135 io1
= container_of(after
, ext4_io_end_t
, list
);
137 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
138 io
, inode
->i_ino
, io0
, io1
);
143 /* Add the io_end to per-inode completed end_io list. */
144 static void ext4_add_complete_io(ext4_io_end_t
*io_end
)
146 struct ext4_inode_info
*ei
= EXT4_I(io_end
->inode
);
147 struct workqueue_struct
*wq
;
150 BUG_ON(!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
));
151 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
153 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
154 if (list_empty(&ei
->i_completed_io_list
))
155 queue_work(wq
, &ei
->i_unwritten_work
);
156 list_add_tail(&io_end
->list
, &ei
->i_completed_io_list
);
157 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
160 static int ext4_do_flush_completed_IO(struct inode
*inode
)
163 struct list_head unwritten
;
165 struct ext4_inode_info
*ei
= EXT4_I(inode
);
168 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
169 dump_completed_IO(inode
);
170 list_replace_init(&ei
->i_completed_io_list
, &unwritten
);
171 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
173 while (!list_empty(&unwritten
)) {
174 io
= list_entry(unwritten
.next
, ext4_io_end_t
, list
);
175 BUG_ON(!(io
->flag
& EXT4_IO_END_UNWRITTEN
));
176 list_del_init(&io
->list
);
178 err
= ext4_end_io(io
);
179 if (unlikely(!ret
&& err
))
186 * work on completed aio dio IO, to convert unwritten extents to extents
188 void ext4_end_io_work(struct work_struct
*work
)
190 struct ext4_inode_info
*ei
= container_of(work
, struct ext4_inode_info
,
192 ext4_do_flush_completed_IO(&ei
->vfs_inode
);
195 int ext4_flush_unwritten_io(struct inode
*inode
)
198 WARN_ON_ONCE(!mutex_is_locked(&inode
->i_mutex
) &&
199 !(inode
->i_state
& I_FREEING
));
200 ret
= ext4_do_flush_completed_IO(inode
);
201 ext4_unwritten_wait(inode
);
205 ext4_io_end_t
*ext4_init_io_end(struct inode
*inode
, gfp_t flags
)
207 ext4_io_end_t
*io
= kmem_cache_zalloc(io_end_cachep
, flags
);
209 atomic_inc(&EXT4_I(inode
)->i_ioend_count
);
211 INIT_LIST_HEAD(&io
->list
);
212 atomic_set(&io
->count
, 1);
217 void ext4_put_io_end_defer(ext4_io_end_t
*io_end
)
219 if (atomic_dec_and_test(&io_end
->count
)) {
220 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
) || !io_end
->size
) {
221 ext4_release_io_end(io_end
);
224 ext4_add_complete_io(io_end
);
228 int ext4_put_io_end(ext4_io_end_t
*io_end
)
232 if (atomic_dec_and_test(&io_end
->count
)) {
233 if (io_end
->flag
& EXT4_IO_END_UNWRITTEN
) {
234 err
= ext4_convert_unwritten_extents(io_end
->handle
,
235 io_end
->inode
, io_end
->offset
,
237 io_end
->handle
= NULL
;
238 ext4_clear_io_unwritten_flag(io_end
);
240 ext4_release_io_end(io_end
);
245 ext4_io_end_t
*ext4_get_io_end(ext4_io_end_t
*io_end
)
247 atomic_inc(&io_end
->count
);
252 * Print an buffer I/O error compatible with the fs/buffer.c. This
253 * provides compatibility with dmesg scrapers that look for a specific
254 * buffer I/O error message. We really need a unified error reporting
255 * structure to userspace ala Digital Unix's uerf system, but it's
256 * probably not going to happen in my lifetime, due to LKML politics...
258 static void buffer_io_error(struct buffer_head
*bh
)
260 char b
[BDEVNAME_SIZE
];
261 printk(KERN_ERR
"Buffer I/O error on device %s, logical block %llu\n",
262 bdevname(bh
->b_bdev
, b
),
263 (unsigned long long)bh
->b_blocknr
);
266 static void ext4_end_bio(struct bio
*bio
, int error
)
268 ext4_io_end_t
*io_end
= bio
->bi_private
;
272 sector_t bi_sector
= bio
->bi_sector
;
275 inode
= io_end
->inode
;
276 blocksize
= 1 << inode
->i_blkbits
;
277 bio
->bi_private
= NULL
;
278 bio
->bi_end_io
= NULL
;
279 if (test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
281 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
282 struct bio_vec
*bvec
= &bio
->bi_io_vec
[i
];
283 struct page
*page
= bvec
->bv_page
;
284 struct buffer_head
*bh
, *head
;
285 unsigned bio_start
= bvec
->bv_offset
;
286 unsigned bio_end
= bio_start
+ bvec
->bv_len
;
287 unsigned under_io
= 0;
295 set_bit(AS_EIO
, &page
->mapping
->flags
);
297 bh
= head
= page_buffers(page
);
299 * We check all buffers in the page under BH_Uptodate_Lock
300 * to avoid races with other end io clearing async_write flags
302 local_irq_save(flags
);
303 bit_spin_lock(BH_Uptodate_Lock
, &head
->b_state
);
305 if (bh_offset(bh
) < bio_start
||
306 bh_offset(bh
) + blocksize
> bio_end
) {
307 if (buffer_async_write(bh
))
311 clear_buffer_async_write(bh
);
314 } while ((bh
= bh
->b_this_page
) != head
);
315 bit_spin_unlock(BH_Uptodate_Lock
, &head
->b_state
);
316 local_irq_restore(flags
);
318 end_page_writeback(page
);
323 io_end
->flag
|= EXT4_IO_END_ERROR
;
324 ext4_warning(inode
->i_sb
, "I/O error writing to inode %lu "
325 "(offset %llu size %ld starting block %llu)",
327 (unsigned long long) io_end
->offset
,
330 bi_sector
>> (inode
->i_blkbits
- 9));
333 ext4_put_io_end_defer(io_end
);
336 void ext4_io_submit(struct ext4_io_submit
*io
)
338 struct bio
*bio
= io
->io_bio
;
342 submit_bio(io
->io_op
, io
->io_bio
);
343 BUG_ON(bio_flagged(io
->io_bio
, BIO_EOPNOTSUPP
));
349 void ext4_io_submit_init(struct ext4_io_submit
*io
,
350 struct writeback_control
*wbc
)
352 io
->io_op
= (wbc
->sync_mode
== WB_SYNC_ALL
? WRITE_SYNC
: WRITE
);
357 static int io_submit_init_bio(struct ext4_io_submit
*io
,
358 struct buffer_head
*bh
)
360 int nvecs
= bio_get_nr_vecs(bh
->b_bdev
);
363 bio
= bio_alloc(GFP_NOIO
, min(nvecs
, BIO_MAX_PAGES
));
364 bio
->bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
365 bio
->bi_bdev
= bh
->b_bdev
;
366 bio
->bi_end_io
= ext4_end_bio
;
367 bio
->bi_private
= ext4_get_io_end(io
->io_end
);
369 io
->io_next_block
= bh
->b_blocknr
;
373 static int io_submit_add_bh(struct ext4_io_submit
*io
,
375 struct buffer_head
*bh
)
379 if (io
->io_bio
&& bh
->b_blocknr
!= io
->io_next_block
) {
383 if (io
->io_bio
== NULL
) {
384 ret
= io_submit_init_bio(io
, bh
);
388 ret
= bio_add_page(io
->io_bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
389 if (ret
!= bh
->b_size
)
390 goto submit_and_retry
;
395 int ext4_bio_write_page(struct ext4_io_submit
*io
,
398 struct writeback_control
*wbc
)
400 struct inode
*inode
= page
->mapping
->host
;
401 unsigned block_start
, blocksize
;
402 struct buffer_head
*bh
, *head
;
404 int nr_submitted
= 0;
406 blocksize
= 1 << inode
->i_blkbits
;
408 BUG_ON(!PageLocked(page
));
409 BUG_ON(PageWriteback(page
));
411 set_page_writeback(page
);
412 ClearPageError(page
);
415 * In the first loop we prepare and mark buffers to submit. We have to
416 * mark all buffers in the page before submitting so that
417 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
418 * on the first buffer finishes and we are still working on submitting
421 bh
= head
= page_buffers(page
);
423 block_start
= bh_offset(bh
);
424 if (block_start
>= len
) {
426 * Comments copied from block_write_full_page_endio:
428 * The page straddles i_size. It must be zeroed out on
429 * each and every writepage invocation because it may
430 * be mmapped. "A file is mapped in multiples of the
431 * page size. For a file that is not a multiple of
432 * the page size, the remaining memory is zeroed when
433 * mapped, and writes to that region are not written
436 zero_user_segment(page
, block_start
,
437 block_start
+ blocksize
);
438 clear_buffer_dirty(bh
);
439 set_buffer_uptodate(bh
);
442 if (!buffer_dirty(bh
) || buffer_delay(bh
) ||
443 !buffer_mapped(bh
) || buffer_unwritten(bh
)) {
444 /* A hole? We can safely clear the dirty bit */
445 if (!buffer_mapped(bh
))
446 clear_buffer_dirty(bh
);
451 if (buffer_new(bh
)) {
452 clear_buffer_new(bh
);
453 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
455 set_buffer_async_write(bh
);
456 } while ((bh
= bh
->b_this_page
) != head
);
458 /* Now submit buffers to write */
459 bh
= head
= page_buffers(page
);
461 if (!buffer_async_write(bh
))
463 ret
= io_submit_add_bh(io
, inode
, bh
);
466 * We only get here on ENOMEM. Not much else
467 * we can do but mark the page as dirty, and
468 * better luck next time.
470 redirty_page_for_writepage(wbc
, page
);
474 clear_buffer_dirty(bh
);
475 } while ((bh
= bh
->b_this_page
) != head
);
477 /* Error stopped previous loop? Clean up buffers... */
480 clear_buffer_async_write(bh
);
481 bh
= bh
->b_this_page
;
482 } while (bh
!= head
);
485 /* Nothing submitted - we have to end page writeback */
487 end_page_writeback(page
);
This page took 0.041388 seconds and 5 git commands to generate.