1 // SPDX-License-Identifier: GPL-2.0
3 * linux/fs/ext4/page-io.c
5 * This contains the new page_io functions for ext4
7 * Written by Theodore Ts'o, 2010.
11 #include <linux/time.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>
27 #include <linux/backing-dev.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 * Print an buffer I/O error compatible with the fs/buffer.c. This
50 * provides compatibility with dmesg scrapers that look for a specific
51 * buffer I/O error message. We really need a unified error reporting
52 * structure to userspace ala Digital Unix's uerf system, but it's
53 * probably not going to happen in my lifetime, due to LKML politics...
55 static void buffer_io_error(struct buffer_head
*bh
)
57 printk_ratelimited(KERN_ERR
"Buffer I/O error on device %pg, logical block %llu\n",
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
;
72 struct buffer_head
*bh
, *head
;
73 unsigned bio_start
= bvec
->bv_offset
;
74 unsigned bio_end
= bio_start
+ bvec
->bv_len
;
75 unsigned under_io
= 0;
81 #ifdef CONFIG_EXT4_FS_ENCRYPTION
83 /* The bounce data pages are unmapped. */
85 fscrypt_pullback_bio_page(&page
, false);
91 mapping_set_error(page
->mapping
, -EIO
);
93 bh
= head
= page_buffers(page
);
95 * We check all buffers in the page under BH_Uptodate_Lock
96 * to avoid races with other end io clearing async_write flags
98 local_irq_save(flags
);
99 bit_spin_lock(BH_Uptodate_Lock
, &head
->b_state
);
101 if (bh_offset(bh
) < bio_start
||
102 bh_offset(bh
) + bh
->b_size
> bio_end
) {
103 if (buffer_async_write(bh
))
107 clear_buffer_async_write(bh
);
110 } while ((bh
= bh
->b_this_page
) != head
);
111 bit_spin_unlock(BH_Uptodate_Lock
, &head
->b_state
);
112 local_irq_restore(flags
);
114 #ifdef CONFIG_EXT4_FS_ENCRYPTION
116 fscrypt_restore_control_page(data_page
);
118 end_page_writeback(page
);
123 static void ext4_release_io_end(ext4_io_end_t
*io_end
)
125 struct bio
*bio
, *next_bio
;
127 BUG_ON(!list_empty(&io_end
->list
));
128 BUG_ON(io_end
->flag
& EXT4_IO_END_UNWRITTEN
);
129 WARN_ON(io_end
->handle
);
131 for (bio
= io_end
->bio
; bio
; bio
= next_bio
) {
132 next_bio
= bio
->bi_private
;
133 ext4_finish_bio(bio
);
136 kmem_cache_free(io_end_cachep
, io_end
);
140 * Check a range of space and convert unwritten extents to written. Note that
141 * we are protected from truncate touching same part of extent tree by the
142 * fact that truncate code waits for all DIO to finish (thus exclusion from
143 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
144 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
145 * completed (happens from ext4_free_ioend()).
147 static int ext4_end_io(ext4_io_end_t
*io
)
149 struct inode
*inode
= io
->inode
;
150 loff_t offset
= io
->offset
;
151 ssize_t size
= io
->size
;
152 handle_t
*handle
= io
->handle
;
155 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
157 io
, inode
->i_ino
, io
->list
.next
, io
->list
.prev
);
159 io
->handle
= NULL
; /* Following call will use up the handle */
160 ret
= ext4_convert_unwritten_extents(handle
, inode
, offset
, size
);
161 if (ret
< 0 && !ext4_forced_shutdown(EXT4_SB(inode
->i_sb
))) {
162 ext4_msg(inode
->i_sb
, KERN_EMERG
,
163 "failed to convert unwritten extents to written "
164 "extents -- potential data loss! "
165 "(inode %lu, offset %llu, size %zd, error %d)",
166 inode
->i_ino
, offset
, size
, ret
);
168 ext4_clear_io_unwritten_flag(io
);
169 ext4_release_io_end(io
);
173 static void dump_completed_IO(struct inode
*inode
, struct list_head
*head
)
176 struct list_head
*cur
, *before
, *after
;
177 ext4_io_end_t
*io
, *io0
, *io1
;
179 if (list_empty(head
))
182 ext4_debug("Dump inode %lu completed io list\n", inode
->i_ino
);
183 list_for_each_entry(io
, head
, list
) {
186 io0
= container_of(before
, ext4_io_end_t
, list
);
188 io1
= container_of(after
, ext4_io_end_t
, list
);
190 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
191 io
, inode
->i_ino
, io0
, io1
);
196 /* Add the io_end to per-inode completed end_io list. */
197 static void ext4_add_complete_io(ext4_io_end_t
*io_end
)
199 struct ext4_inode_info
*ei
= EXT4_I(io_end
->inode
);
200 struct ext4_sb_info
*sbi
= EXT4_SB(io_end
->inode
->i_sb
);
201 struct workqueue_struct
*wq
;
204 /* Only reserved conversions from writeback should enter here */
205 WARN_ON(!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
));
206 WARN_ON(!io_end
->handle
&& sbi
->s_journal
);
207 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
208 wq
= sbi
->rsv_conversion_wq
;
209 if (list_empty(&ei
->i_rsv_conversion_list
))
210 queue_work(wq
, &ei
->i_rsv_conversion_work
);
211 list_add_tail(&io_end
->list
, &ei
->i_rsv_conversion_list
);
212 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
215 static int ext4_do_flush_completed_IO(struct inode
*inode
,
216 struct list_head
*head
)
219 struct list_head unwritten
;
221 struct ext4_inode_info
*ei
= EXT4_I(inode
);
224 spin_lock_irqsave(&ei
->i_completed_io_lock
, flags
);
225 dump_completed_IO(inode
, head
);
226 list_replace_init(head
, &unwritten
);
227 spin_unlock_irqrestore(&ei
->i_completed_io_lock
, flags
);
229 while (!list_empty(&unwritten
)) {
230 io
= list_entry(unwritten
.next
, ext4_io_end_t
, list
);
231 BUG_ON(!(io
->flag
& EXT4_IO_END_UNWRITTEN
));
232 list_del_init(&io
->list
);
234 err
= ext4_end_io(io
);
235 if (unlikely(!ret
&& err
))
242 * work on completed IO, to convert unwritten extents to extents
244 void ext4_end_io_rsv_work(struct work_struct
*work
)
246 struct ext4_inode_info
*ei
= container_of(work
, struct ext4_inode_info
,
247 i_rsv_conversion_work
);
248 ext4_do_flush_completed_IO(&ei
->vfs_inode
, &ei
->i_rsv_conversion_list
);
251 ext4_io_end_t
*ext4_init_io_end(struct inode
*inode
, gfp_t flags
)
253 ext4_io_end_t
*io
= kmem_cache_zalloc(io_end_cachep
, flags
);
256 INIT_LIST_HEAD(&io
->list
);
257 atomic_set(&io
->count
, 1);
262 void ext4_put_io_end_defer(ext4_io_end_t
*io_end
)
264 if (atomic_dec_and_test(&io_end
->count
)) {
265 if (!(io_end
->flag
& EXT4_IO_END_UNWRITTEN
) || !io_end
->size
) {
266 ext4_release_io_end(io_end
);
269 ext4_add_complete_io(io_end
);
273 int ext4_put_io_end(ext4_io_end_t
*io_end
)
277 if (atomic_dec_and_test(&io_end
->count
)) {
278 if (io_end
->flag
& EXT4_IO_END_UNWRITTEN
) {
279 err
= ext4_convert_unwritten_extents(io_end
->handle
,
280 io_end
->inode
, io_end
->offset
,
282 io_end
->handle
= NULL
;
283 ext4_clear_io_unwritten_flag(io_end
);
285 ext4_release_io_end(io_end
);
290 ext4_io_end_t
*ext4_get_io_end(ext4_io_end_t
*io_end
)
292 atomic_inc(&io_end
->count
);
296 /* BIO completion function for page writeback */
297 static void ext4_end_bio(struct bio
*bio
)
299 ext4_io_end_t
*io_end
= bio
->bi_private
;
300 sector_t bi_sector
= bio
->bi_iter
.bi_sector
;
301 char b
[BDEVNAME_SIZE
];
303 if (WARN_ONCE(!io_end
, "io_end is NULL: %s: sector %Lu len %u err %d\n",
305 (long long) bio
->bi_iter
.bi_sector
,
306 (unsigned) bio_sectors(bio
),
308 ext4_finish_bio(bio
);
312 bio
->bi_end_io
= NULL
;
314 if (bio
->bi_status
) {
315 struct inode
*inode
= io_end
->inode
;
317 ext4_warning(inode
->i_sb
, "I/O error %d writing to inode %lu "
318 "(offset %llu size %ld starting block %llu)",
319 bio
->bi_status
, inode
->i_ino
,
320 (unsigned long long) io_end
->offset
,
323 bi_sector
>> (inode
->i_blkbits
- 9));
324 mapping_set_error(inode
->i_mapping
,
325 blk_status_to_errno(bio
->bi_status
));
328 if (io_end
->flag
& EXT4_IO_END_UNWRITTEN
) {
330 * Link bio into list hanging from io_end. We have to do it
331 * atomically as bio completions can be racing against each
334 bio
->bi_private
= xchg(&io_end
->bio
, bio
);
335 ext4_put_io_end_defer(io_end
);
338 * Drop io_end reference early. Inode can get freed once
341 ext4_put_io_end_defer(io_end
);
342 ext4_finish_bio(bio
);
347 void ext4_io_submit(struct ext4_io_submit
*io
)
349 struct bio
*bio
= io
->io_bio
;
352 int io_op_flags
= io
->io_wbc
->sync_mode
== WB_SYNC_ALL
?
354 io
->io_bio
->bi_write_hint
= io
->io_end
->inode
->i_write_hint
;
355 bio_set_op_attrs(io
->io_bio
, REQ_OP_WRITE
, io_op_flags
);
356 submit_bio(io
->io_bio
);
361 void ext4_io_submit_init(struct ext4_io_submit
*io
,
362 struct writeback_control
*wbc
)
369 static int io_submit_init_bio(struct ext4_io_submit
*io
,
370 struct buffer_head
*bh
)
374 bio
= bio_alloc(GFP_NOIO
, BIO_MAX_PAGES
);
377 bio
->bi_iter
.bi_sector
= bh
->b_blocknr
* (bh
->b_size
>> 9);
378 bio_set_dev(bio
, bh
->b_bdev
);
379 bio
->bi_end_io
= ext4_end_bio
;
380 bio
->bi_private
= ext4_get_io_end(io
->io_end
);
382 io
->io_next_block
= bh
->b_blocknr
;
383 wbc_init_bio(io
->io_wbc
, bio
);
387 static int io_submit_add_bh(struct ext4_io_submit
*io
,
390 struct buffer_head
*bh
)
394 if (io
->io_bio
&& bh
->b_blocknr
!= io
->io_next_block
) {
398 if (io
->io_bio
== NULL
) {
399 ret
= io_submit_init_bio(io
, bh
);
402 io
->io_bio
->bi_write_hint
= inode
->i_write_hint
;
404 ret
= bio_add_page(io
->io_bio
, page
, bh
->b_size
, bh_offset(bh
));
405 if (ret
!= bh
->b_size
)
406 goto submit_and_retry
;
407 wbc_account_io(io
->io_wbc
, page
, bh
->b_size
);
412 int ext4_bio_write_page(struct ext4_io_submit
*io
,
415 struct writeback_control
*wbc
,
418 struct page
*data_page
= NULL
;
419 struct inode
*inode
= page
->mapping
->host
;
420 unsigned block_start
;
421 struct buffer_head
*bh
, *head
;
423 int nr_submitted
= 0;
424 int nr_to_submit
= 0;
426 BUG_ON(!PageLocked(page
));
427 BUG_ON(PageWriteback(page
));
430 set_page_writeback_keepwrite(page
);
432 set_page_writeback(page
);
433 ClearPageError(page
);
436 * Comments copied from block_write_full_page:
438 * The page straddles i_size. It must be zeroed out on each and every
439 * writepage invocation because it may be mmapped. "A file is mapped
440 * in multiples of the page size. For a file that is not a multiple of
441 * the page size, the remaining memory is zeroed when mapped, and
442 * writes to that region are not written out to the file."
445 zero_user_segment(page
, len
, PAGE_SIZE
);
447 * In the first loop we prepare and mark buffers to submit. We have to
448 * mark all buffers in the page before submitting so that
449 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
450 * on the first buffer finishes and we are still working on submitting
453 bh
= head
= page_buffers(page
);
455 block_start
= bh_offset(bh
);
456 if (block_start
>= len
) {
457 clear_buffer_dirty(bh
);
458 set_buffer_uptodate(bh
);
461 if (!buffer_dirty(bh
) || buffer_delay(bh
) ||
462 !buffer_mapped(bh
) || buffer_unwritten(bh
)) {
463 /* A hole? We can safely clear the dirty bit */
464 if (!buffer_mapped(bh
))
465 clear_buffer_dirty(bh
);
470 if (buffer_new(bh
)) {
471 clear_buffer_new(bh
);
472 clean_bdev_bh_alias(bh
);
474 set_buffer_async_write(bh
);
476 } while ((bh
= bh
->b_this_page
) != head
);
478 bh
= head
= page_buffers(page
);
480 if (ext4_encrypted_inode(inode
) && S_ISREG(inode
->i_mode
) &&
482 gfp_t gfp_flags
= GFP_NOFS
;
485 data_page
= fscrypt_encrypt_page(inode
, page
, PAGE_SIZE
, 0,
486 page
->index
, gfp_flags
);
487 if (IS_ERR(data_page
)) {
488 ret
= PTR_ERR(data_page
);
489 if (ret
== -ENOMEM
&& wbc
->sync_mode
== WB_SYNC_ALL
) {
492 congestion_wait(BLK_RW_ASYNC
, HZ
/50);
494 gfp_flags
|= __GFP_NOFAIL
;
502 /* Now submit buffers to write */
504 if (!buffer_async_write(bh
))
506 ret
= io_submit_add_bh(io
, inode
,
507 data_page
? data_page
: page
, bh
);
510 * We only get here on ENOMEM. Not much else
511 * we can do but mark the page as dirty, and
512 * better luck next time.
517 clear_buffer_dirty(bh
);
518 } while ((bh
= bh
->b_this_page
) != head
);
520 /* Error stopped previous loop? Clean up buffers... */
524 fscrypt_restore_control_page(data_page
);
525 printk_ratelimited(KERN_ERR
"%s: ret = %d\n", __func__
, ret
);
526 redirty_page_for_writepage(wbc
, page
);
528 clear_buffer_async_write(bh
);
529 bh
= bh
->b_this_page
;
530 } while (bh
!= head
);
533 /* Nothing submitted - we have to end page writeback */
535 end_page_writeback(page
);