Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/fs/ext4/page-io.c
4 *
5 * This contains the new page_io functions for ext4
6 *
7 * Written by Theodore Ts'o, 2010.
8 */
9
10#include <linux/fs.h>
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>
26#include <linux/mm.h>
27#include <linux/sched/mm.h>
28
29#include "ext4_jbd2.h"
30#include "xattr.h"
31#include "acl.h"
32
33static struct kmem_cache *io_end_cachep;
34static struct kmem_cache *io_end_vec_cachep;
35
36int __init ext4_init_pageio(void)
37{
38 io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
39 if (io_end_cachep == NULL)
40 return -ENOMEM;
41
42 io_end_vec_cachep = KMEM_CACHE(ext4_io_end_vec, 0);
43 if (io_end_vec_cachep == NULL) {
44 kmem_cache_destroy(io_end_cachep);
45 return -ENOMEM;
46 }
47 return 0;
48}
49
50void ext4_exit_pageio(void)
51{
52 kmem_cache_destroy(io_end_cachep);
53 kmem_cache_destroy(io_end_vec_cachep);
54}
55
56struct ext4_io_end_vec *ext4_alloc_io_end_vec(ext4_io_end_t *io_end)
57{
58 struct ext4_io_end_vec *io_end_vec;
59
60 io_end_vec = kmem_cache_zalloc(io_end_vec_cachep, GFP_NOFS);
61 if (!io_end_vec)
62 return ERR_PTR(-ENOMEM);
63 INIT_LIST_HEAD(&io_end_vec->list);
64 list_add_tail(&io_end_vec->list, &io_end->list_vec);
65 return io_end_vec;
66}
67
68static void ext4_free_io_end_vec(ext4_io_end_t *io_end)
69{
70 struct ext4_io_end_vec *io_end_vec, *tmp;
71
72 if (list_empty(&io_end->list_vec))
73 return;
74 list_for_each_entry_safe(io_end_vec, tmp, &io_end->list_vec, list) {
75 list_del(&io_end_vec->list);
76 kmem_cache_free(io_end_vec_cachep, io_end_vec);
77 }
78}
79
80struct ext4_io_end_vec *ext4_last_io_end_vec(ext4_io_end_t *io_end)
81{
82 BUG_ON(list_empty(&io_end->list_vec));
83 return list_last_entry(&io_end->list_vec, struct ext4_io_end_vec, list);
84}
85
86/*
87 * Print an buffer I/O error compatible with the fs/buffer.c. This
88 * provides compatibility with dmesg scrapers that look for a specific
89 * buffer I/O error message. We really need a unified error reporting
90 * structure to userspace ala Digital Unix's uerf system, but it's
91 * probably not going to happen in my lifetime, due to LKML politics...
92 */
93static void buffer_io_error(struct buffer_head *bh)
94{
95 printk_ratelimited(KERN_ERR "Buffer I/O error on device %pg, logical block %llu\n",
96 bh->b_bdev,
97 (unsigned long long)bh->b_blocknr);
98}
99
100static void ext4_finish_bio(struct bio *bio)
101{
102 struct bio_vec *bvec;
103 struct bvec_iter_all iter_all;
104
105 bio_for_each_segment_all(bvec, bio, iter_all) {
106 struct page *page = bvec->bv_page;
107 struct page *bounce_page = NULL;
108 struct buffer_head *bh, *head;
109 unsigned bio_start = bvec->bv_offset;
110 unsigned bio_end = bio_start + bvec->bv_len;
111 unsigned under_io = 0;
112 unsigned long flags;
113
114 if (fscrypt_is_bounce_page(page)) {
115 bounce_page = page;
116 page = fscrypt_pagecache_page(bounce_page);
117 }
118
119 if (bio->bi_status) {
120 SetPageError(page);
121 mapping_set_error(page->mapping, -EIO);
122 }
123 bh = head = page_buffers(page);
124 /*
125 * We check all buffers in the page under b_uptodate_lock
126 * to avoid races with other end io clearing async_write flags
127 */
128 spin_lock_irqsave(&head->b_uptodate_lock, flags);
129 do {
130 if (bh_offset(bh) < bio_start ||
131 bh_offset(bh) + bh->b_size > bio_end) {
132 if (buffer_async_write(bh))
133 under_io++;
134 continue;
135 }
136 clear_buffer_async_write(bh);
137 if (bio->bi_status) {
138 set_buffer_write_io_error(bh);
139 buffer_io_error(bh);
140 }
141 } while ((bh = bh->b_this_page) != head);
142 spin_unlock_irqrestore(&head->b_uptodate_lock, flags);
143 if (!under_io) {
144 fscrypt_free_bounce_page(bounce_page);
145 end_page_writeback(page);
146 }
147 }
148}
149
150static void ext4_release_io_end(ext4_io_end_t *io_end)
151{
152 struct bio *bio, *next_bio;
153
154 BUG_ON(!list_empty(&io_end->list));
155 BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
156 WARN_ON(io_end->handle);
157
158 for (bio = io_end->bio; bio; bio = next_bio) {
159 next_bio = bio->bi_private;
160 ext4_finish_bio(bio);
161 bio_put(bio);
162 }
163 ext4_free_io_end_vec(io_end);
164 kmem_cache_free(io_end_cachep, io_end);
165}
166
167/*
168 * Check a range of space and convert unwritten extents to written. Note that
169 * we are protected from truncate touching same part of extent tree by the
170 * fact that truncate code waits for all DIO to finish (thus exclusion from
171 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
172 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
173 * completed (happens from ext4_free_ioend()).
174 */
175static int ext4_end_io_end(ext4_io_end_t *io_end)
176{
177 struct inode *inode = io_end->inode;
178 handle_t *handle = io_end->handle;
179 int ret = 0;
180
181 ext4_debug("ext4_end_io_nolock: io_end 0x%p from inode %lu,list->next 0x%p,"
182 "list->prev 0x%p\n",
183 io_end, inode->i_ino, io_end->list.next, io_end->list.prev);
184
185 io_end->handle = NULL; /* Following call will use up the handle */
186 ret = ext4_convert_unwritten_io_end_vec(handle, io_end);
187 if (ret < 0 && !ext4_forced_shutdown(EXT4_SB(inode->i_sb))) {
188 ext4_msg(inode->i_sb, KERN_EMERG,
189 "failed to convert unwritten extents to written "
190 "extents -- potential data loss! "
191 "(inode %lu, error %d)", inode->i_ino, ret);
192 }
193 ext4_clear_io_unwritten_flag(io_end);
194 ext4_release_io_end(io_end);
195 return ret;
196}
197
198static void dump_completed_IO(struct inode *inode, struct list_head *head)
199{
200#ifdef EXT4FS_DEBUG
201 struct list_head *cur, *before, *after;
202 ext4_io_end_t *io_end, *io_end0, *io_end1;
203
204 if (list_empty(head))
205 return;
206
207 ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
208 list_for_each_entry(io_end, head, list) {
209 cur = &io_end->list;
210 before = cur->prev;
211 io_end0 = container_of(before, ext4_io_end_t, list);
212 after = cur->next;
213 io_end1 = container_of(after, ext4_io_end_t, list);
214
215 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
216 io_end, inode->i_ino, io_end0, io_end1);
217 }
218#endif
219}
220
221/* Add the io_end to per-inode completed end_io list. */
222static void ext4_add_complete_io(ext4_io_end_t *io_end)
223{
224 struct ext4_inode_info *ei = EXT4_I(io_end->inode);
225 struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
226 struct workqueue_struct *wq;
227 unsigned long flags;
228
229 /* Only reserved conversions from writeback should enter here */
230 WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
231 WARN_ON(!io_end->handle && sbi->s_journal);
232 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
233 wq = sbi->rsv_conversion_wq;
234 if (list_empty(&ei->i_rsv_conversion_list))
235 queue_work(wq, &ei->i_rsv_conversion_work);
236 list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
237 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
238}
239
240static int ext4_do_flush_completed_IO(struct inode *inode,
241 struct list_head *head)
242{
243 ext4_io_end_t *io_end;
244 struct list_head unwritten;
245 unsigned long flags;
246 struct ext4_inode_info *ei = EXT4_I(inode);
247 int err, ret = 0;
248
249 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
250 dump_completed_IO(inode, head);
251 list_replace_init(head, &unwritten);
252 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
253
254 while (!list_empty(&unwritten)) {
255 io_end = list_entry(unwritten.next, ext4_io_end_t, list);
256 BUG_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
257 list_del_init(&io_end->list);
258
259 err = ext4_end_io_end(io_end);
260 if (unlikely(!ret && err))
261 ret = err;
262 }
263 return ret;
264}
265
266/*
267 * work on completed IO, to convert unwritten extents to extents
268 */
269void ext4_end_io_rsv_work(struct work_struct *work)
270{
271 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
272 i_rsv_conversion_work);
273 ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
274}
275
276ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
277{
278 ext4_io_end_t *io_end = kmem_cache_zalloc(io_end_cachep, flags);
279
280 if (io_end) {
281 io_end->inode = inode;
282 INIT_LIST_HEAD(&io_end->list);
283 INIT_LIST_HEAD(&io_end->list_vec);
284 refcount_set(&io_end->count, 1);
285 }
286 return io_end;
287}
288
289void ext4_put_io_end_defer(ext4_io_end_t *io_end)
290{
291 if (refcount_dec_and_test(&io_end->count)) {
292 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) ||
293 list_empty(&io_end->list_vec)) {
294 ext4_release_io_end(io_end);
295 return;
296 }
297 ext4_add_complete_io(io_end);
298 }
299}
300
301int ext4_put_io_end(ext4_io_end_t *io_end)
302{
303 int err = 0;
304
305 if (refcount_dec_and_test(&io_end->count)) {
306 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
307 err = ext4_convert_unwritten_io_end_vec(io_end->handle,
308 io_end);
309 io_end->handle = NULL;
310 ext4_clear_io_unwritten_flag(io_end);
311 }
312 ext4_release_io_end(io_end);
313 }
314 return err;
315}
316
317ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
318{
319 refcount_inc(&io_end->count);
320 return io_end;
321}
322
323/* BIO completion function for page writeback */
324static void ext4_end_bio(struct bio *bio)
325{
326 ext4_io_end_t *io_end = bio->bi_private;
327 sector_t bi_sector = bio->bi_iter.bi_sector;
328
329 if (WARN_ONCE(!io_end, "io_end is NULL: %pg: sector %Lu len %u err %d\n",
330 bio->bi_bdev,
331 (long long) bio->bi_iter.bi_sector,
332 (unsigned) bio_sectors(bio),
333 bio->bi_status)) {
334 ext4_finish_bio(bio);
335 bio_put(bio);
336 return;
337 }
338 bio->bi_end_io = NULL;
339
340 if (bio->bi_status) {
341 struct inode *inode = io_end->inode;
342
343 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
344 "starting block %llu)",
345 bio->bi_status, inode->i_ino,
346 (unsigned long long)
347 bi_sector >> (inode->i_blkbits - 9));
348 mapping_set_error(inode->i_mapping,
349 blk_status_to_errno(bio->bi_status));
350 }
351
352 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
353 /*
354 * Link bio into list hanging from io_end. We have to do it
355 * atomically as bio completions can be racing against each
356 * other.
357 */
358 bio->bi_private = xchg(&io_end->bio, bio);
359 ext4_put_io_end_defer(io_end);
360 } else {
361 /*
362 * Drop io_end reference early. Inode can get freed once
363 * we finish the bio.
364 */
365 ext4_put_io_end_defer(io_end);
366 ext4_finish_bio(bio);
367 bio_put(bio);
368 }
369}
370
371void ext4_io_submit(struct ext4_io_submit *io)
372{
373 struct bio *bio = io->io_bio;
374
375 if (bio) {
376 if (io->io_wbc->sync_mode == WB_SYNC_ALL)
377 io->io_bio->bi_opf |= REQ_SYNC;
378 submit_bio(io->io_bio);
379 }
380 io->io_bio = NULL;
381}
382
383void ext4_io_submit_init(struct ext4_io_submit *io,
384 struct writeback_control *wbc)
385{
386 io->io_wbc = wbc;
387 io->io_bio = NULL;
388 io->io_end = NULL;
389}
390
391static void io_submit_init_bio(struct ext4_io_submit *io,
392 struct buffer_head *bh)
393{
394 struct bio *bio;
395
396 /*
397 * bio_alloc will _always_ be able to allocate a bio if
398 * __GFP_DIRECT_RECLAIM is set, see comments for bio_alloc_bioset().
399 */
400 bio = bio_alloc(bh->b_bdev, BIO_MAX_VECS, REQ_OP_WRITE, GFP_NOIO);
401 fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO);
402 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
403 bio->bi_end_io = ext4_end_bio;
404 bio->bi_private = ext4_get_io_end(io->io_end);
405 io->io_bio = bio;
406 io->io_next_block = bh->b_blocknr;
407 wbc_init_bio(io->io_wbc, bio);
408}
409
410static void io_submit_add_bh(struct ext4_io_submit *io,
411 struct inode *inode,
412 struct page *page,
413 struct buffer_head *bh)
414{
415 int ret;
416
417 if (io->io_bio && (bh->b_blocknr != io->io_next_block ||
418 !fscrypt_mergeable_bio_bh(io->io_bio, bh))) {
419submit_and_retry:
420 ext4_io_submit(io);
421 }
422 if (io->io_bio == NULL)
423 io_submit_init_bio(io, bh);
424 ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
425 if (ret != bh->b_size)
426 goto submit_and_retry;
427 wbc_account_cgroup_owner(io->io_wbc, page, bh->b_size);
428 io->io_next_block++;
429}
430
431int ext4_bio_write_page(struct ext4_io_submit *io,
432 struct page *page,
433 int len)
434{
435 struct page *bounce_page = NULL;
436 struct inode *inode = page->mapping->host;
437 unsigned block_start;
438 struct buffer_head *bh, *head;
439 int ret = 0;
440 int nr_to_submit = 0;
441 struct writeback_control *wbc = io->io_wbc;
442 bool keep_towrite = false;
443
444 BUG_ON(!PageLocked(page));
445 BUG_ON(PageWriteback(page));
446
447 ClearPageError(page);
448
449 /*
450 * Comments copied from block_write_full_page:
451 *
452 * The page straddles i_size. It must be zeroed out on each and every
453 * writepage invocation because it may be mmapped. "A file is mapped
454 * in multiples of the page size. For a file that is not a multiple of
455 * the page size, the remaining memory is zeroed when mapped, and
456 * writes to that region are not written out to the file."
457 */
458 if (len < PAGE_SIZE)
459 zero_user_segment(page, len, PAGE_SIZE);
460 /*
461 * In the first loop we prepare and mark buffers to submit. We have to
462 * mark all buffers in the page before submitting so that
463 * end_page_writeback() cannot be called from ext4_end_bio() when IO
464 * on the first buffer finishes and we are still working on submitting
465 * the second buffer.
466 */
467 bh = head = page_buffers(page);
468 do {
469 block_start = bh_offset(bh);
470 if (block_start >= len) {
471 clear_buffer_dirty(bh);
472 set_buffer_uptodate(bh);
473 continue;
474 }
475 if (!buffer_dirty(bh) || buffer_delay(bh) ||
476 !buffer_mapped(bh) || buffer_unwritten(bh)) {
477 /* A hole? We can safely clear the dirty bit */
478 if (!buffer_mapped(bh))
479 clear_buffer_dirty(bh);
480 /*
481 * Keeping dirty some buffer we cannot write? Make sure
482 * to redirty the page and keep TOWRITE tag so that
483 * racing WB_SYNC_ALL writeback does not skip the page.
484 * This happens e.g. when doing writeout for
485 * transaction commit.
486 */
487 if (buffer_dirty(bh)) {
488 if (!PageDirty(page))
489 redirty_page_for_writepage(wbc, page);
490 keep_towrite = true;
491 }
492 continue;
493 }
494 if (buffer_new(bh))
495 clear_buffer_new(bh);
496 set_buffer_async_write(bh);
497 clear_buffer_dirty(bh);
498 nr_to_submit++;
499 } while ((bh = bh->b_this_page) != head);
500
501 /* Nothing to submit? Just unlock the page... */
502 if (!nr_to_submit)
503 goto unlock;
504
505 bh = head = page_buffers(page);
506
507 /*
508 * If any blocks are being written to an encrypted file, encrypt them
509 * into a bounce page. For simplicity, just encrypt until the last
510 * block which might be needed. This may cause some unneeded blocks
511 * (e.g. holes) to be unnecessarily encrypted, but this is rare and
512 * can't happen in the common case of blocksize == PAGE_SIZE.
513 */
514 if (fscrypt_inode_uses_fs_layer_crypto(inode) && nr_to_submit) {
515 gfp_t gfp_flags = GFP_NOFS;
516 unsigned int enc_bytes = round_up(len, i_blocksize(inode));
517
518 /*
519 * Since bounce page allocation uses a mempool, we can only use
520 * a waiting mask (i.e. request guaranteed allocation) on the
521 * first page of the bio. Otherwise it can deadlock.
522 */
523 if (io->io_bio)
524 gfp_flags = GFP_NOWAIT | __GFP_NOWARN;
525 retry_encrypt:
526 bounce_page = fscrypt_encrypt_pagecache_blocks(page, enc_bytes,
527 0, gfp_flags);
528 if (IS_ERR(bounce_page)) {
529 ret = PTR_ERR(bounce_page);
530 if (ret == -ENOMEM &&
531 (io->io_bio || wbc->sync_mode == WB_SYNC_ALL)) {
532 gfp_t new_gfp_flags = GFP_NOFS;
533 if (io->io_bio)
534 ext4_io_submit(io);
535 else
536 new_gfp_flags |= __GFP_NOFAIL;
537 memalloc_retry_wait(gfp_flags);
538 gfp_flags = new_gfp_flags;
539 goto retry_encrypt;
540 }
541
542 printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
543 redirty_page_for_writepage(wbc, page);
544 do {
545 if (buffer_async_write(bh)) {
546 clear_buffer_async_write(bh);
547 set_buffer_dirty(bh);
548 }
549 bh = bh->b_this_page;
550 } while (bh != head);
551 goto unlock;
552 }
553 }
554
555 if (keep_towrite)
556 set_page_writeback_keepwrite(page);
557 else
558 set_page_writeback(page);
559
560 /* Now submit buffers to write */
561 do {
562 if (!buffer_async_write(bh))
563 continue;
564 io_submit_add_bh(io, inode,
565 bounce_page ? bounce_page : page, bh);
566 } while ((bh = bh->b_this_page) != head);
567unlock:
568 unlock_page(page);
569 return ret;
570}
1/*
2 * linux/fs/ext4/page-io.c
3 *
4 * This contains the new page_io functions for ext4
5 *
6 * Written by Theodore Ts'o, 2010.
7 */
8
9#include <linux/module.h>
10#include <linux/fs.h>
11#include <linux/time.h>
12#include <linux/jbd2.h>
13#include <linux/highuid.h>
14#include <linux/pagemap.h>
15#include <linux/quotaops.h>
16#include <linux/string.h>
17#include <linux/buffer_head.h>
18#include <linux/writeback.h>
19#include <linux/pagevec.h>
20#include <linux/mpage.h>
21#include <linux/namei.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>
27
28#include "ext4_jbd2.h"
29#include "xattr.h"
30#include "acl.h"
31#include "ext4_extents.h"
32
33static struct kmem_cache *io_page_cachep, *io_end_cachep;
34
35int __init ext4_init_pageio(void)
36{
37 io_page_cachep = KMEM_CACHE(ext4_io_page, SLAB_RECLAIM_ACCOUNT);
38 if (io_page_cachep == NULL)
39 return -ENOMEM;
40 io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
41 if (io_end_cachep == NULL) {
42 kmem_cache_destroy(io_page_cachep);
43 return -ENOMEM;
44 }
45 return 0;
46}
47
48void ext4_exit_pageio(void)
49{
50 kmem_cache_destroy(io_end_cachep);
51 kmem_cache_destroy(io_page_cachep);
52}
53
54void ext4_ioend_wait(struct inode *inode)
55{
56 wait_queue_head_t *wq = ext4_ioend_wq(inode);
57
58 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0));
59}
60
61static void put_io_page(struct ext4_io_page *io_page)
62{
63 if (atomic_dec_and_test(&io_page->p_count)) {
64 end_page_writeback(io_page->p_page);
65 put_page(io_page->p_page);
66 kmem_cache_free(io_page_cachep, io_page);
67 }
68}
69
70void ext4_free_io_end(ext4_io_end_t *io)
71{
72 int i;
73 wait_queue_head_t *wq;
74
75 BUG_ON(!io);
76 if (io->page)
77 put_page(io->page);
78 for (i = 0; i < io->num_io_pages; i++)
79 put_io_page(io->pages[i]);
80 io->num_io_pages = 0;
81 wq = ext4_ioend_wq(io->inode);
82 if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count) &&
83 waitqueue_active(wq))
84 wake_up_all(wq);
85 kmem_cache_free(io_end_cachep, io);
86}
87
88/*
89 * check a range of space and convert unwritten extents to written.
90 */
91int ext4_end_io_nolock(ext4_io_end_t *io)
92{
93 struct inode *inode = io->inode;
94 loff_t offset = io->offset;
95 ssize_t size = io->size;
96 wait_queue_head_t *wq;
97 int ret = 0;
98
99 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
100 "list->prev 0x%p\n",
101 io, inode->i_ino, io->list.next, io->list.prev);
102
103 if (list_empty(&io->list))
104 return ret;
105
106 if (!(io->flag & EXT4_IO_END_UNWRITTEN))
107 return ret;
108
109 ret = ext4_convert_unwritten_extents(inode, offset, size);
110 if (ret < 0) {
111 printk(KERN_EMERG "%s: failed to convert unwritten "
112 "extents to written extents, error is %d "
113 "io is still on inode %lu aio dio list\n",
114 __func__, ret, inode->i_ino);
115 return ret;
116 }
117
118 if (io->iocb)
119 aio_complete(io->iocb, io->result, 0);
120 /* clear the DIO AIO unwritten flag */
121 if (io->flag & EXT4_IO_END_UNWRITTEN) {
122 io->flag &= ~EXT4_IO_END_UNWRITTEN;
123 /* Wake up anyone waiting on unwritten extent conversion */
124 wq = ext4_ioend_wq(io->inode);
125 if (atomic_dec_and_test(&EXT4_I(inode)->i_aiodio_unwritten) &&
126 waitqueue_active(wq)) {
127 wake_up_all(wq);
128 }
129 }
130
131 return ret;
132}
133
134/*
135 * work on completed aio dio IO, to convert unwritten extents to extents
136 */
137static void ext4_end_io_work(struct work_struct *work)
138{
139 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
140 struct inode *inode = io->inode;
141 struct ext4_inode_info *ei = EXT4_I(inode);
142 unsigned long flags;
143 int ret;
144
145 if (!mutex_trylock(&inode->i_mutex)) {
146 /*
147 * Requeue the work instead of waiting so that the work
148 * items queued after this can be processed.
149 */
150 queue_work(EXT4_SB(inode->i_sb)->dio_unwritten_wq, &io->work);
151 /*
152 * To prevent the ext4-dio-unwritten thread from keeping
153 * requeueing end_io requests and occupying cpu for too long,
154 * yield the cpu if it sees an end_io request that has already
155 * been requeued.
156 */
157 if (io->flag & EXT4_IO_END_QUEUED)
158 yield();
159 io->flag |= EXT4_IO_END_QUEUED;
160 return;
161 }
162 ret = ext4_end_io_nolock(io);
163 if (ret < 0) {
164 mutex_unlock(&inode->i_mutex);
165 return;
166 }
167
168 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
169 if (!list_empty(&io->list))
170 list_del_init(&io->list);
171 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
172 mutex_unlock(&inode->i_mutex);
173 ext4_free_io_end(io);
174}
175
176ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
177{
178 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
179 if (io) {
180 atomic_inc(&EXT4_I(inode)->i_ioend_count);
181 io->inode = inode;
182 INIT_WORK(&io->work, ext4_end_io_work);
183 INIT_LIST_HEAD(&io->list);
184 }
185 return io;
186}
187
188/*
189 * Print an buffer I/O error compatible with the fs/buffer.c. This
190 * provides compatibility with dmesg scrapers that look for a specific
191 * buffer I/O error message. We really need a unified error reporting
192 * structure to userspace ala Digital Unix's uerf system, but it's
193 * probably not going to happen in my lifetime, due to LKML politics...
194 */
195static void buffer_io_error(struct buffer_head *bh)
196{
197 char b[BDEVNAME_SIZE];
198 printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
199 bdevname(bh->b_bdev, b),
200 (unsigned long long)bh->b_blocknr);
201}
202
203static void ext4_end_bio(struct bio *bio, int error)
204{
205 ext4_io_end_t *io_end = bio->bi_private;
206 struct workqueue_struct *wq;
207 struct inode *inode;
208 unsigned long flags;
209 int i;
210 sector_t bi_sector = bio->bi_sector;
211
212 BUG_ON(!io_end);
213 bio->bi_private = NULL;
214 bio->bi_end_io = NULL;
215 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
216 error = 0;
217 bio_put(bio);
218
219 for (i = 0; i < io_end->num_io_pages; i++) {
220 struct page *page = io_end->pages[i]->p_page;
221 struct buffer_head *bh, *head;
222 loff_t offset;
223 loff_t io_end_offset;
224
225 if (error) {
226 SetPageError(page);
227 set_bit(AS_EIO, &page->mapping->flags);
228 head = page_buffers(page);
229 BUG_ON(!head);
230
231 io_end_offset = io_end->offset + io_end->size;
232
233 offset = (sector_t) page->index << PAGE_CACHE_SHIFT;
234 bh = head;
235 do {
236 if ((offset >= io_end->offset) &&
237 (offset+bh->b_size <= io_end_offset))
238 buffer_io_error(bh);
239
240 offset += bh->b_size;
241 bh = bh->b_this_page;
242 } while (bh != head);
243 }
244
245 put_io_page(io_end->pages[i]);
246 }
247 io_end->num_io_pages = 0;
248 inode = io_end->inode;
249
250 if (error) {
251 io_end->flag |= EXT4_IO_END_ERROR;
252 ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
253 "(offset %llu size %ld starting block %llu)",
254 inode->i_ino,
255 (unsigned long long) io_end->offset,
256 (long) io_end->size,
257 (unsigned long long)
258 bi_sector >> (inode->i_blkbits - 9));
259 }
260
261 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
262 ext4_free_io_end(io_end);
263 return;
264 }
265
266 /* Add the io_end to per-inode completed io list*/
267 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
268 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
269 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
270
271 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
272 /* queue the work to convert unwritten extents to written */
273 queue_work(wq, &io_end->work);
274}
275
276void ext4_io_submit(struct ext4_io_submit *io)
277{
278 struct bio *bio = io->io_bio;
279
280 if (bio) {
281 bio_get(io->io_bio);
282 submit_bio(io->io_op, io->io_bio);
283 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
284 bio_put(io->io_bio);
285 }
286 io->io_bio = NULL;
287 io->io_op = 0;
288 io->io_end = NULL;
289}
290
291static int io_submit_init(struct ext4_io_submit *io,
292 struct inode *inode,
293 struct writeback_control *wbc,
294 struct buffer_head *bh)
295{
296 ext4_io_end_t *io_end;
297 struct page *page = bh->b_page;
298 int nvecs = bio_get_nr_vecs(bh->b_bdev);
299 struct bio *bio;
300
301 io_end = ext4_init_io_end(inode, GFP_NOFS);
302 if (!io_end)
303 return -ENOMEM;
304 bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
305 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
306 bio->bi_bdev = bh->b_bdev;
307 bio->bi_private = io->io_end = io_end;
308 bio->bi_end_io = ext4_end_bio;
309
310 io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh);
311
312 io->io_bio = bio;
313 io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
314 io->io_next_block = bh->b_blocknr;
315 return 0;
316}
317
318static int io_submit_add_bh(struct ext4_io_submit *io,
319 struct ext4_io_page *io_page,
320 struct inode *inode,
321 struct writeback_control *wbc,
322 struct buffer_head *bh)
323{
324 ext4_io_end_t *io_end;
325 int ret;
326
327 if (buffer_new(bh)) {
328 clear_buffer_new(bh);
329 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
330 }
331
332 if (!buffer_mapped(bh) || buffer_delay(bh)) {
333 if (!buffer_mapped(bh))
334 clear_buffer_dirty(bh);
335 if (io->io_bio)
336 ext4_io_submit(io);
337 return 0;
338 }
339
340 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
341submit_and_retry:
342 ext4_io_submit(io);
343 }
344 if (io->io_bio == NULL) {
345 ret = io_submit_init(io, inode, wbc, bh);
346 if (ret)
347 return ret;
348 }
349 io_end = io->io_end;
350 if ((io_end->num_io_pages >= MAX_IO_PAGES) &&
351 (io_end->pages[io_end->num_io_pages-1] != io_page))
352 goto submit_and_retry;
353 if (buffer_uninit(bh) && !(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
354 io_end->flag |= EXT4_IO_END_UNWRITTEN;
355 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
356 }
357 io->io_end->size += bh->b_size;
358 io->io_next_block++;
359 ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
360 if (ret != bh->b_size)
361 goto submit_and_retry;
362 if ((io_end->num_io_pages == 0) ||
363 (io_end->pages[io_end->num_io_pages-1] != io_page)) {
364 io_end->pages[io_end->num_io_pages++] = io_page;
365 atomic_inc(&io_page->p_count);
366 }
367 return 0;
368}
369
370int ext4_bio_write_page(struct ext4_io_submit *io,
371 struct page *page,
372 int len,
373 struct writeback_control *wbc)
374{
375 struct inode *inode = page->mapping->host;
376 unsigned block_start, block_end, blocksize;
377 struct ext4_io_page *io_page;
378 struct buffer_head *bh, *head;
379 int ret = 0;
380
381 blocksize = 1 << inode->i_blkbits;
382
383 BUG_ON(!PageLocked(page));
384 BUG_ON(PageWriteback(page));
385
386 io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS);
387 if (!io_page) {
388 set_page_dirty(page);
389 unlock_page(page);
390 return -ENOMEM;
391 }
392 io_page->p_page = page;
393 atomic_set(&io_page->p_count, 1);
394 get_page(page);
395 set_page_writeback(page);
396 ClearPageError(page);
397
398 for (bh = head = page_buffers(page), block_start = 0;
399 bh != head || !block_start;
400 block_start = block_end, bh = bh->b_this_page) {
401
402 block_end = block_start + blocksize;
403 if (block_start >= len) {
404 clear_buffer_dirty(bh);
405 set_buffer_uptodate(bh);
406 continue;
407 }
408 clear_buffer_dirty(bh);
409 ret = io_submit_add_bh(io, io_page, inode, wbc, bh);
410 if (ret) {
411 /*
412 * We only get here on ENOMEM. Not much else
413 * we can do but mark the page as dirty, and
414 * better luck next time.
415 */
416 set_page_dirty(page);
417 break;
418 }
419 }
420 unlock_page(page);
421 /*
422 * If the page was truncated before we could do the writeback,
423 * or we had a memory allocation error while trying to write
424 * the first buffer head, we won't have submitted any pages for
425 * I/O. In that case we need to make sure we've cleared the
426 * PageWriteback bit from the page to prevent the system from
427 * wedging later on.
428 */
429 put_io_page(io_page);
430 return ret;
431}