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
);
54 * This function is called by ext4_evict_inode() to make sure there is
55 * no more pending I/O completion work left to do.
57 void ext4_ioend_shutdown(struct inode
*inode
)
59 wait_queue_head_t
*wq
= ext4_ioend_wq(inode
);
61 wait_event(*wq
, (atomic_read(&EXT4_I(inode
)->i_ioend_count
) == 0));
63 * We need to make sure the work structure is finished being
64 * used before we let the inode get destroyed.
66 if (work_pending(&EXT4_I(inode
)->i_unwritten_work
))
67 cancel_work_sync(&EXT4_I(inode
)->i_unwritten_work
);
70 static void put_io_page(struct ext4_io_page
*io_page
)
72 if (atomic_dec_and_test(&io_page
->p_count
)) {
73 end_page_writeback(io_page
->p_page
);
74 put_page(io_page
->p_page
);
75 kmem_cache_free(io_page_cachep
, io_page
);
79 void ext4_free_io_end(ext4_io_end_t
*io
)
84 BUG_ON(!list_empty(&io
->list
));
85 BUG_ON(io
->flag
& EXT4_IO_END_UNWRITTEN
);
87 for (i
= 0; i
< io
->num_io_pages
; i
++)
88 put_io_page(io
->pages
[i
]);
90 if (atomic_dec_and_test(&EXT4_I(io
->inode
)->i_ioend_count
))
91 wake_up_all(ext4_ioend_wq(io
->inode
));
92 kmem_cache_free(io_end_cachep
, io
);
95 /* check a range of space and convert unwritten extents to written. */
96 static int ext4_end_io(ext4_io_end_t
*io
)
98 struct inode
*inode
= io
->inode
;
99 loff_t offset
= io
->offset
;
100 ssize_t size
= io
->size
;
103 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
105 io
, inode
->i_ino
, io
->list
.next
, io
->list
.prev
);
107 ret
= ext4_convert_unwritten_extents(inode
, offset
, size
);
109 ext4_msg(inode
->i_sb
, KERN_EMERG
,
110 "failed to convert unwritten extents to written "
111 "extents -- potential data loss! "
112 "(inode %lu, offset %llu, size %zd, error %d)",
113 inode
->i_ino
, offset
, size
, ret
);
115 /* Wake up anyone waiting on unwritten extent conversion */
116 if (atomic_dec_and_test(&EXT4_I(inode
)->i_unwritten
))
117 wake_up_all(ext4_ioend_wq(inode
));
118 if (io
->flag
& EXT4_IO_END_DIRECT
)
119 inode_dio_done(inode
);
121 aio_complete(io
->iocb
, io
->result
, 0);
125 static void dump_completed_IO(struct inode
*inode
)
128 struct list_head
*cur
, *before
, *after
;
129 ext4_io_end_t
*io
, *io0
, *io1
;
131 if (list_empty(&EXT4_I(inode
)->i_completed_io_list
)) {
132 ext4_debug("inode %lu completed_io list is empty\n",
137 ext4_debug("Dump inode %lu completed_io list\n", inode
->i_ino
);
138 list_for_each_entry(io
, &EXT4_I(inode
)->i_completed_io_list
, list
) {
141 io0
= container_of(before
, ext4_io_end_t
, list
);
143 io1
= container_of(after
, ext4_io_end_t
, list
);
145 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
146 io
, inode
->i_ino
, io0
, io1
);
151 /* Add the io_end to per-inode completed end_io list. */
152 void ext4_add_complete_io(ext4_io_end_t
*io_end
)
154 struct ext4_inode_info
*ei
= EXT4_I(io_end
->inode
);
155 struct workqueue_struct
*wq
;
158 BUG_ON(!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
));
159 wq
= EXT4_SB(io_end
->inode
->i_sb
)->dio_unwritten_wq
;
161 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
162 if (list_empty(&ei
->i_completed_io_list
))
163 queue_work(wq
, &ei
->i_unwritten_work
);
164 list_add_tail(&io_end
->list
, &ei
->i_completed_io_list
);
165 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
168 static int ext4_do_flush_completed_IO(struct inode
*inode
)
171 struct list_head unwritten
;
173 struct ext4_inode_info
*ei
= EXT4_I(inode
);
176 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
177 dump_completed_IO(inode
);
178 list_replace_init(&ei
->i_completed_io_list
, &unwritten
);
179 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
181 while (!list_empty(&unwritten
)) {
182 io
= list_entry(unwritten
.next
, ext4_io_end_t
, list
);
183 BUG_ON(!(io
->flag
& EXT4_IO_END_UNWRITTEN
));
184 list_del_init(&io
->list
);
186 err
= ext4_end_io(io
);
187 if (unlikely(!ret
&& err
))
189 io
->flag
&= ~EXT4_IO_END_UNWRITTEN
;
190 ext4_free_io_end(io
);
196 * work on completed aio dio IO, to convert unwritten extents to extents
198 void ext4_end_io_work(struct work_struct
*work
)
200 struct ext4_inode_info
*ei
= container_of(work
, struct ext4_inode_info
,
202 ext4_do_flush_completed_IO(&ei
->vfs_inode
);
205 int ext4_flush_unwritten_io(struct inode
*inode
)
208 WARN_ON_ONCE(!mutex_is_locked(&inode
->i_mutex
) &&
209 !(inode
->i_state
& I_FREEING
));
210 ret
= ext4_do_flush_completed_IO(inode
);
211 ext4_unwritten_wait(inode
);
215 ext4_io_end_t
*ext4_init_io_end(struct inode
*inode
, gfp_t flags
)
217 ext4_io_end_t
*io
= kmem_cache_zalloc(io_end_cachep
, flags
);
219 atomic_inc(&EXT4_I(inode
)->i_ioend_count
);
221 INIT_LIST_HEAD(&io
->list
);
227 * Print an buffer I/O error compatible with the fs/buffer.c. This
228 * provides compatibility with dmesg scrapers that look for a specific
229 * buffer I/O error message. We really need a unified error reporting
230 * structure to userspace ala Digital Unix's uerf system, but it's
231 * probably not going to happen in my lifetime, due to LKML politics...
233 static void buffer_io_error(struct buffer_head
*bh
)
235 char b
[BDEVNAME_SIZE
];
236 printk(KERN_ERR
"Buffer I/O error on device %s, logical block %llu\n",
237 bdevname(bh
->b_bdev
, b
),
238 (unsigned long long)bh
->b_blocknr
);
241 static void ext4_end_bio(struct bio
*bio
, int error
)
243 ext4_io_end_t
*io_end
= bio
->bi_private
;
246 sector_t bi_sector
= bio
->bi_sector
;
249 bio
->bi_private
= NULL
;
250 bio
->bi_end_io
= NULL
;
251 if (test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
255 for (i
= 0; i
< io_end
->num_io_pages
; i
++) {
256 struct page
*page
= io_end
->pages
[i
]->p_page
;
257 struct buffer_head
*bh
, *head
;
259 loff_t io_end_offset
;
263 set_bit(AS_EIO
, &page
->mapping
->flags
);
264 head
= page_buffers(page
);
267 io_end_offset
= io_end
->offset
+ io_end
->size
;
269 offset
= (sector_t
) page
->index
<< PAGE_CACHE_SHIFT
;
272 if ((offset
>= io_end
->offset
) &&
273 (offset
+bh
->b_size
<= io_end_offset
))
276 offset
+= bh
->b_size
;
277 bh
= bh
->b_this_page
;
278 } while (bh
!= head
);
281 put_io_page(io_end
->pages
[i
]);
283 io_end
->num_io_pages
= 0;
284 inode
= io_end
->inode
;
287 io_end
->flag
|= EXT4_IO_END_ERROR
;
288 ext4_warning(inode
->i_sb
, "I/O error writing to inode %lu "
289 "(offset %llu size %ld starting block %llu)",
291 (unsigned long long) io_end
->offset
,
294 bi_sector
>> (inode
->i_blkbits
- 9));
297 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
)) {
298 ext4_free_io_end(io_end
);
302 ext4_add_complete_io(io_end
);
305 void ext4_io_submit(struct ext4_io_submit
*io
)
307 struct bio
*bio
= io
->io_bio
;
311 submit_bio(io
->io_op
, io
->io_bio
);
312 BUG_ON(bio_flagged(io
->io_bio
, BIO_EOPNOTSUPP
));
320 static int io_submit_init(struct ext4_io_submit
*io
,
322 struct writeback_control
*wbc
,
323 struct buffer_head
*bh
)
325 ext4_io_end_t
*io_end
;
326 struct page
*page
= bh
->b_page
;
327 int nvecs
= bio_get_nr_vecs(bh
->b_bdev
);
330 io_end
= ext4_init_io_end(inode
, GFP_NOFS
);
333 bio
= bio_alloc(GFP_NOIO
, min(nvecs
, BIO_MAX_PAGES
));
334 bio
->bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
335 bio
->bi_bdev
= bh
->b_bdev
;
336 bio
->bi_private
= io
->io_end
= io_end
;
337 bio
->bi_end_io
= ext4_end_bio
;
339 io_end
->offset
= (page
->index
<< PAGE_CACHE_SHIFT
) + bh_offset(bh
);
342 io
->io_op
= (wbc
->sync_mode
== WB_SYNC_ALL
? WRITE_SYNC
: WRITE
);
343 io
->io_next_block
= bh
->b_blocknr
;
347 static int io_submit_add_bh(struct ext4_io_submit
*io
,
348 struct ext4_io_page
*io_page
,
350 struct writeback_control
*wbc
,
351 struct buffer_head
*bh
)
353 ext4_io_end_t
*io_end
;
356 if (buffer_new(bh
)) {
357 clear_buffer_new(bh
);
358 unmap_underlying_metadata(bh
->b_bdev
, bh
->b_blocknr
);
361 if (io
->io_bio
&& bh
->b_blocknr
!= io
->io_next_block
) {
365 if (io
->io_bio
== NULL
) {
366 ret
= io_submit_init(io
, inode
, wbc
, bh
);
371 if ((io_end
->num_io_pages
>= MAX_IO_PAGES
) &&
372 (io_end
->pages
[io_end
->num_io_pages
-1] != io_page
))
373 goto submit_and_retry
;
374 if (buffer_uninit(bh
))
375 ext4_set_io_unwritten_flag(inode
, io_end
);
376 io
->io_end
->size
+= bh
->b_size
;
378 ret
= bio_add_page(io
->io_bio
, bh
->b_page
, bh
->b_size
, bh_offset(bh
));
379 if (ret
!= bh
->b_size
)
380 goto submit_and_retry
;
381 if ((io_end
->num_io_pages
== 0) ||
382 (io_end
->pages
[io_end
->num_io_pages
-1] != io_page
)) {
383 io_end
->pages
[io_end
->num_io_pages
++] = io_page
;
384 atomic_inc(&io_page
->p_count
);
389 int ext4_bio_write_page(struct ext4_io_submit
*io
,
392 struct writeback_control
*wbc
)
394 struct inode
*inode
= page
->mapping
->host
;
395 unsigned block_start
, block_end
, blocksize
;
396 struct ext4_io_page
*io_page
;
397 struct buffer_head
*bh
, *head
;
400 blocksize
= 1 << inode
->i_blkbits
;
402 BUG_ON(!PageLocked(page
));
403 BUG_ON(PageWriteback(page
));
405 io_page
= kmem_cache_alloc(io_page_cachep
, GFP_NOFS
);
407 redirty_page_for_writepage(wbc
, page
);
411 io_page
->p_page
= page
;
412 atomic_set(&io_page
->p_count
, 1);
414 set_page_writeback(page
);
415 ClearPageError(page
);
417 for (bh
= head
= page_buffers(page
), block_start
= 0;
418 bh
!= head
|| !block_start
;
419 block_start
= block_end
, bh
= bh
->b_this_page
) {
421 block_end
= block_start
+ blocksize
;
422 if (block_start
>= len
) {
424 * Comments copied from block_write_full_page_endio:
426 * The page straddles i_size. It must be zeroed out on
427 * each and every writepage invocation because it may
428 * be mmapped. "A file is mapped in multiples of the
429 * page size. For a file that is not a multiple of
430 * the page size, the remaining memory is zeroed when
431 * mapped, and writes to that region are not written
434 zero_user_segment(page
, block_start
, block_end
);
435 clear_buffer_dirty(bh
);
436 set_buffer_uptodate(bh
);
439 if (!buffer_dirty(bh
) || buffer_delay(bh
) ||
440 !buffer_mapped(bh
) || buffer_unwritten(bh
)) {
441 /* A hole? We can safely clear the dirty bit */
442 if (!buffer_mapped(bh
))
443 clear_buffer_dirty(bh
);
448 ret
= io_submit_add_bh(io
, io_page
, inode
, wbc
, bh
);
451 * We only get here on ENOMEM. Not much else
452 * we can do but mark the page as dirty, and
453 * better luck next time.
455 redirty_page_for_writepage(wbc
, page
);
458 clear_buffer_dirty(bh
);
462 * If the page was truncated before we could do the writeback,
463 * or we had a memory allocation error while trying to write
464 * the first buffer head, we won't have submitted any pages for
465 * I/O. In that case we need to make sure we've cleared the
466 * PageWriteback bit from the page to prevent the system from
469 put_io_page(io_page
);