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/uio.h>
22 #include <linux/bio.h>
23 #include <linux/workqueue.h>
24 #include <linux/kernel.h>
25 #include <linux/slab.h>
28 #include "ext4_jbd2.h"
32 static struct kmem_cache
*io_page_cachep
, *io_end_cachep
;
34 int __init
ext4_init_pageio(void)
36 io_page_cachep
= KMEM_CACHE(ext4_io_page
, SLAB_RECLAIM_ACCOUNT
);
37 if (io_page_cachep
== NULL
)
39 io_end_cachep
= KMEM_CACHE(ext4_io_end
, SLAB_RECLAIM_ACCOUNT
);
40 if (io_end_cachep
== NULL
) {
41 kmem_cache_destroy(io_page_cachep
);
47 void ext4_exit_pageio(void)
49 kmem_cache_destroy(io_end_cachep
);
50 kmem_cache_destroy(io_page_cachep
);
53 void ext4_ioend_wait(struct inode
*inode
)
55 wait_queue_head_t
*wq
= ext4_ioend_wq(inode
);
57 wait_event(*wq
, (atomic_read(&EXT4_I(inode
)->i_ioend_count
) == 0));
60 static void put_io_page(struct ext4_io_page
*io_page
)
62 if (atomic_dec_and_test(&io_page
->p_count
)) {
63 end_page_writeback(io_page
->p_page
);
64 put_page(io_page
->p_page
);
65 kmem_cache_free(io_page_cachep
, io_page
);
69 void ext4_free_io_end(ext4_io_end_t
*io
)
74 BUG_ON(!list_empty(&io
->list
));
75 BUG_ON(io
->flag
& EXT4_IO_END_UNWRITTEN
);
77 for (i
= 0; i
< io
->num_io_pages
; i
++)
78 put_io_page(io
->pages
[i
]);
80 if (atomic_dec_and_test(&EXT4_I(io
->inode
)->i_ioend_count
))
81 wake_up_all(ext4_ioend_wq(io
->inode
));
82 kmem_cache_free(io_end_cachep
, io
);
85 /* check a range of space and convert unwritten extents to written. */
86 static int ext4_end_io(ext4_io_end_t
*io
)
88 struct inode
*inode
= io
->inode
;
89 loff_t offset
= io
->offset
;
90 ssize_t size
= io
->size
;
93 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
95 io
, inode
->i_ino
, io
->list
.next
, io
->list
.prev
);
97 ret
= ext4_convert_unwritten_extents(inode
, offset
, size
);
99 ext4_msg(inode
->i_sb
, KERN_EMERG
,
100 "failed to convert unwritten extents to written "
101 "extents -- potential data loss! "
102 "(inode %lu, offset %llu, size %zd, error %d)",
103 inode
->i_ino
, offset
, size
, ret
);
105 /* Wake up anyone waiting on unwritten extent conversion */
106 if (atomic_dec_and_test(&EXT4_I(inode
)->i_unwritten
))
107 wake_up_all(ext4_ioend_wq(inode
));
108 if (io
->flag
& EXT4_IO_END_DIRECT
)
109 inode_dio_done(inode
);
111 aio_complete(io
->iocb
, io
->result
, 0);
115 static void dump_completed_IO(struct inode
*inode
)
118 struct list_head
*cur
, *before
, *after
;
119 ext4_io_end_t
*io
, *io0
, *io1
;
121 if (list_empty(&EXT4_I(inode
)->i_completed_io_list
)) {
122 ext4_debug("inode %lu completed_io list is empty\n",
127 ext4_debug("Dump inode %lu completed_io list\n", inode
->i_ino
);
128 list_for_each_entry(io
, &EXT4_I(inode
)->i_completed_io_list
, list
) {
131 io0
= container_of(before
, ext4_io_end_t
, list
);
133 io1
= container_of(after
, ext4_io_end_t
, list
);
135 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
136 io
, inode
->i_ino
, io0
, io1
);
141 /* Add the io_end to per-inode completed end_io list. */
142 void ext4_add_complete_io(ext4_io_end_t
*io_end
)
144 struct ext4_inode_info
*ei
= EXT4_I(io_end
->inode
);
145 struct workqueue_struct
*wq
;
148 BUG_ON(!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
));
149 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
151 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
152 if (list_empty(&ei
->i_completed_io_list
))
153 queue_work(wq
, &ei
->i_unwritten_work
);
154 list_add_tail(&io_end
->list
, &ei
->i_completed_io_list
);
155 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
158 static int ext4_do_flush_completed_IO(struct inode
*inode
)
161 struct list_head unwritten
;
163 struct ext4_inode_info
*ei
= EXT4_I(inode
);
166 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
167 dump_completed_IO(inode
);
168 list_replace_init(&ei
->i_completed_io_list
, &unwritten
);
169 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
171 while (!list_empty(&unwritten
)) {
172 io
= list_entry(unwritten
.next
, ext4_io_end_t
, list
);
173 BUG_ON(!(io
->flag
& EXT4_IO_END_UNWRITTEN
));
174 list_del_init(&io
->list
);
176 err
= ext4_end_io(io
);
177 if (unlikely(!ret
&& err
))
179 io
->flag
&= ~EXT4_IO_END_UNWRITTEN
;
180 ext4_free_io_end(io
);
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
);
217 * Print an buffer I/O error compatible with the fs/buffer.c. This
218 * provides compatibility with dmesg scrapers that look for a specific
219 * buffer I/O error message. We really need a unified error reporting
220 * structure to userspace ala Digital Unix's uerf system, but it's
221 * probably not going to happen in my lifetime, due to LKML politics...
223 static void buffer_io_error(struct buffer_head
*bh
)
225 char b
[BDEVNAME_SIZE
];
226 printk(KERN_ERR
"Buffer I/O error on device %s, logical block %llu\n",
227 bdevname(bh
->b_bdev
, b
),
228 (unsigned long long)bh
->b_blocknr
);
231 static void ext4_end_bio(struct bio
*bio
, int error
)
233 ext4_io_end_t
*io_end
= bio
->bi_private
;
236 sector_t bi_sector
= bio
->bi_sector
;
239 bio
->bi_private
= NULL
;
240 bio
->bi_end_io
= NULL
;
241 if (test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
245 for (i
= 0; i
< io_end
->num_io_pages
; i
++) {
246 struct page
*page
= io_end
->pages
[i
]->p_page
;
247 struct buffer_head
*bh
, *head
;
249 loff_t io_end_offset
;
253 set_bit(AS_EIO
, &page
->mapping
->flags
);
254 head
= page_buffers(page
);
257 io_end_offset
= io_end
->offset
+ io_end
->size
;
259 offset
= (sector_t
) page
->index
<< PAGE_CACHE_SHIFT
;
262 if ((offset
>= io_end
->offset
) &&
263 (offset
+bh
->b_size
<= io_end_offset
))
266 offset
+= bh
->b_size
;
267 bh
= bh
->b_this_page
;
268 } while (bh
!= head
);
271 put_io_page(io_end
->pages
[i
]);
273 io_end
->num_io_pages
= 0;
274 inode
= io_end
->inode
;
277 io_end
->flag
|= EXT4_IO_END_ERROR
;
278 ext4_warning(inode
->i_sb
, "I/O error writing to inode %lu "
279 "(offset %llu size %ld starting block %llu)",
281 (unsigned long long) io_end
->offset
,
284 bi_sector
>> (inode
->i_blkbits
- 9));
287 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
288 ext4_free_io_end(io_end
);
292 ext4_add_complete_io(io_end
);
295 void ext4_io_submit(struct ext4_io_submit
*io
)
297 struct bio
*bio
= io
->io_bio
;
301 submit_bio(io
->io_op
, io
->io_bio
);
302 BUG_ON(bio_flagged(io
->io_bio
, BIO_EOPNOTSUPP
));
310 static int io_submit_init(struct ext4_io_submit
*io
,
312 struct writeback_control
*wbc
,
313 struct buffer_head
*bh
)
315 ext4_io_end_t
*io_end
;
316 struct page
*page
= bh
->b_page
;
317 int nvecs
= bio_get_nr_vecs(bh
->b_bdev
);
320 io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
323 bio
= bio_alloc(GFP_NOIO
, min(nvecs
, BIO_MAX_PAGES
));
324 bio
->bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
325 bio
->bi_bdev
= bh
->b_bdev
;
326 bio
->bi_private
= io
->io_end
= io_end
;
327 bio
->bi_end_io
= ext4_end_bio
;
329 io_end
->offset
= (page
->index
<< PAGE_CACHE_SHIFT
) + bh_offset(bh
);
332 io
->io_op
= (wbc
->sync_mode
== WB_SYNC_ALL
? WRITE_SYNC
: WRITE
);
333 io
->io_next_block
= bh
->b_blocknr
;
337 static int io_submit_add_bh(struct ext4_io_submit
*io
,
338 struct ext4_io_page
*io_page
,
340 struct writeback_control
*wbc
,
341 struct buffer_head
*bh
)
343 ext4_io_end_t
*io_end
;
346 if (buffer_new(bh
)) {
347 clear_buffer_new(bh
);
348 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
351 if (io
->io_bio
&& bh
->b_blocknr
!= io
->io_next_block
) {
355 if (io
->io_bio
== NULL
) {
356 ret
= io_submit_init(io
, inode
, wbc
, bh
);
361 if ((io_end
->num_io_pages
>= MAX_IO_PAGES
) &&
362 (io_end
->pages
[io_end
->num_io_pages
-1] != io_page
))
363 goto submit_and_retry
;
364 if (buffer_uninit(bh
))
365 ext4_set_io_unwritten_flag(inode
, io_end
);
366 io
->io_end
->size
+= bh
->b_size
;
368 ret
= bio_add_page(io
->io_bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
369 if (ret
!= bh
->b_size
)
370 goto submit_and_retry
;
371 if ((io_end
->num_io_pages
== 0) ||
372 (io_end
->pages
[io_end
->num_io_pages
-1] != io_page
)) {
373 io_end
->pages
[io_end
->num_io_pages
++] = io_page
;
374 atomic_inc(&io_page
->p_count
);
379 int ext4_bio_write_page(struct ext4_io_submit
*io
,
382 struct writeback_control
*wbc
)
384 struct inode
*inode
= page
->mapping
->host
;
385 unsigned block_start
, block_end
, blocksize
;
386 struct ext4_io_page
*io_page
;
387 struct buffer_head
*bh
, *head
;
390 blocksize
= 1 << inode
->i_blkbits
;
392 BUG_ON(!PageLocked(page
));
393 BUG_ON(PageWriteback(page
));
395 io_page
= kmem_cache_alloc(io_page_cachep
, GFP_NOFS
);
397 redirty_page_for_writepage(wbc
, page
);
401 io_page
->p_page
= page
;
402 atomic_set(&io_page
->p_count
, 1);
404 set_page_writeback(page
);
405 ClearPageError(page
);
407 for (bh
= head
= page_buffers(page
), block_start
= 0;
408 bh
!= head
|| !block_start
;
409 block_start
= block_end
, bh
= bh
->b_this_page
) {
411 block_end
= block_start
+ blocksize
;
412 if (block_start
>= len
) {
414 * Comments copied from block_write_full_page_endio:
416 * The page straddles i_size. It must be zeroed out on
417 * each and every writepage invocation because it may
418 * be mmapped. "A file is mapped in multiples of the
419 * page size. For a file that is not a multiple of
420 * the page size, the remaining memory is zeroed when
421 * mapped, and writes to that region are not written
424 zero_user_segment(page
, block_start
, block_end
);
425 clear_buffer_dirty(bh
);
426 set_buffer_uptodate(bh
);
429 if (!buffer_dirty(bh
) || buffer_delay(bh
) ||
430 !buffer_mapped(bh
) || buffer_unwritten(bh
)) {
431 /* A hole? We can safely clear the dirty bit */
432 if (!buffer_mapped(bh
))
433 clear_buffer_dirty(bh
);
438 ret
= io_submit_add_bh(io
, io_page
, inode
, wbc
, bh
);
441 * We only get here on ENOMEM. Not much else
442 * we can do but mark the page as dirty, and
443 * better luck next time.
445 redirty_page_for_writepage(wbc
, page
);
448 clear_buffer_dirty(bh
);
452 * If the page was truncated before we could do the writeback,
453 * or we had a memory allocation error while trying to write
454 * the first buffer head, we won't have submitted any pages for
455 * I/O. In that case we need to make sure we've cleared the
456 * PageWriteback bit from the page to prevent the system from
459 put_io_page(io_page
);