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(s: io_end_cachep);
45 return -ENOMEM;
46 }
47 return 0;
48}
49
50void ext4_exit_pageio(void)
51{
52 kmem_cache_destroy(s: io_end_cachep);
53 kmem_cache_destroy(s: 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(k: io_end_vec_cachep, GFP_NOFS);
61 if (!io_end_vec)
62 return ERR_PTR(error: -ENOMEM);
63 INIT_LIST_HEAD(list: &io_end_vec->list);
64 list_add_tail(new: &io_end_vec->list, head: &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(head: &io_end->list_vec))
73 return;
74 list_for_each_entry_safe(io_end_vec, tmp, &io_end->list_vec, list) {
75 list_del(entry: &io_end_vec->list);
76 kmem_cache_free(s: io_end_vec_cachep, objp: 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 folio_iter fi;
103
104 bio_for_each_folio_all(fi, bio) {
105 struct folio *folio = fi.folio;
106 struct folio *io_folio = NULL;
107 struct buffer_head *bh, *head;
108 size_t bio_start = fi.offset;
109 size_t bio_end = bio_start + fi.length;
110 unsigned under_io = 0;
111 unsigned long flags;
112
113 if (fscrypt_is_bounce_folio(folio)) {
114 io_folio = folio;
115 folio = fscrypt_pagecache_folio(bounce_folio: folio);
116 }
117
118 if (bio->bi_status) {
119 int err = blk_status_to_errno(status: bio->bi_status);
120 folio_set_error(folio);
121 mapping_set_error(mapping: folio->mapping, error: err);
122 }
123 bh = head = folio_buffers(folio);
124 /*
125 * We check all buffers in the folio 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(lock: &head->b_uptodate_lock, flags);
143 if (!under_io) {
144 fscrypt_free_bounce_page(bounce_page: &io_folio->page);
145 folio_end_writeback(folio);
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(s: io_end_cachep, objp: 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(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(inode: io_end->inode);
225 struct ext4_sb_info *sbi = EXT4_SB(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(head: &ei->i_rsv_conversion_list))
235 queue_work(wq, work: &ei->i_rsv_conversion_work);
236 list_add_tail(new: &io_end->list, head: &ei->i_rsv_conversion_list);
237 spin_unlock_irqrestore(lock: &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(old: head, new: &unwritten);
252 spin_unlock_irqrestore(lock: &ei->i_completed_io_lock, flags);
253
254 while (!list_empty(head: &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(entry: &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(inode: &ei->vfs_inode, head: &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(k: io_end_cachep, flags);
279
280 if (io_end) {
281 io_end->inode = inode;
282 INIT_LIST_HEAD(list: &io_end->list);
283 INIT_LIST_HEAD(list: &io_end->list_vec);
284 refcount_set(r: &io_end->count, n: 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(r: &io_end->count)) {
292 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) ||
293 list_empty(head: &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(r: &io_end->count)) {
306 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
307 err = ext4_convert_unwritten_io_end_vec(handle: 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(r: &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(mapping: inode->i_mapping,
349 error: blk_status_to_errno(status: 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(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(bdev: bh->b_bdev, BIO_MAX_VECS, opf: REQ_OP_WRITE, GFP_NOIO);
401 fscrypt_set_bio_crypt_ctx_bh(bio, first_bh: 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_end: io->io_end);
405 io->io_bio = bio;
406 io->io_next_block = bh->b_blocknr;
407 wbc_init_bio(wbc: io->io_wbc, bio);
408}
409
410static void io_submit_add_bh(struct ext4_io_submit *io,
411 struct inode *inode,
412 struct folio *folio,
413 struct folio *io_folio,
414 struct buffer_head *bh)
415{
416 if (io->io_bio && (bh->b_blocknr != io->io_next_block ||
417 !fscrypt_mergeable_bio_bh(bio: io->io_bio, next_bh: bh))) {
418submit_and_retry:
419 ext4_io_submit(io);
420 }
421 if (io->io_bio == NULL)
422 io_submit_init_bio(io, bh);
423 if (!bio_add_folio(bio: io->io_bio, folio: io_folio, len: bh->b_size, off: bh_offset(bh)))
424 goto submit_and_retry;
425 wbc_account_cgroup_owner(wbc: io->io_wbc, page: &folio->page, bytes: bh->b_size);
426 io->io_next_block++;
427}
428
429int ext4_bio_write_folio(struct ext4_io_submit *io, struct folio *folio,
430 size_t len)
431{
432 struct folio *io_folio = folio;
433 struct inode *inode = folio->mapping->host;
434 unsigned block_start;
435 struct buffer_head *bh, *head;
436 int ret = 0;
437 int nr_to_submit = 0;
438 struct writeback_control *wbc = io->io_wbc;
439 bool keep_towrite = false;
440
441 BUG_ON(!folio_test_locked(folio));
442 BUG_ON(folio_test_writeback(folio));
443
444 folio_clear_error(folio);
445
446 /*
447 * Comments copied from block_write_full_page:
448 *
449 * The folio straddles i_size. It must be zeroed out on each and every
450 * writepage invocation because it may be mmapped. "A file is mapped
451 * in multiples of the page size. For a file that is not a multiple of
452 * the page size, the remaining memory is zeroed when mapped, and
453 * writes to that region are not written out to the file."
454 */
455 if (len < folio_size(folio))
456 folio_zero_segment(folio, start: len, xend: folio_size(folio));
457 /*
458 * In the first loop we prepare and mark buffers to submit. We have to
459 * mark all buffers in the folio before submitting so that
460 * folio_end_writeback() cannot be called from ext4_end_bio() when IO
461 * on the first buffer finishes and we are still working on submitting
462 * the second buffer.
463 */
464 bh = head = folio_buffers(folio);
465 do {
466 block_start = bh_offset(bh);
467 if (block_start >= len) {
468 clear_buffer_dirty(bh);
469 set_buffer_uptodate(bh);
470 continue;
471 }
472 if (!buffer_dirty(bh) || buffer_delay(bh) ||
473 !buffer_mapped(bh) || buffer_unwritten(bh)) {
474 /* A hole? We can safely clear the dirty bit */
475 if (!buffer_mapped(bh))
476 clear_buffer_dirty(bh);
477 /*
478 * Keeping dirty some buffer we cannot write? Make sure
479 * to redirty the folio and keep TOWRITE tag so that
480 * racing WB_SYNC_ALL writeback does not skip the folio.
481 * This happens e.g. when doing writeout for
482 * transaction commit or when journalled data is not
483 * yet committed.
484 */
485 if (buffer_dirty(bh) ||
486 (buffer_jbd(bh) && buffer_jbddirty(bh))) {
487 if (!folio_test_dirty(folio))
488 folio_redirty_for_writepage(wbc, folio);
489 keep_towrite = true;
490 }
491 continue;
492 }
493 if (buffer_new(bh))
494 clear_buffer_new(bh);
495 set_buffer_async_write(bh);
496 clear_buffer_dirty(bh);
497 nr_to_submit++;
498 } while ((bh = bh->b_this_page) != head);
499
500 /* Nothing to submit? Just unlock the folio... */
501 if (!nr_to_submit)
502 return 0;
503
504 bh = head = folio_buffers(folio);
505
506 /*
507 * If any blocks are being written to an encrypted file, encrypt them
508 * into a bounce page. For simplicity, just encrypt until the last
509 * block which might be needed. This may cause some unneeded blocks
510 * (e.g. holes) to be unnecessarily encrypted, but this is rare and
511 * can't happen in the common case of blocksize == PAGE_SIZE.
512 */
513 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
514 gfp_t gfp_flags = GFP_NOFS;
515 unsigned int enc_bytes = round_up(len, i_blocksize(inode));
516 struct page *bounce_page;
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: &folio->page,
527 len: enc_bytes, offs: 0, gfp_flags);
528 if (IS_ERR(ptr: bounce_page)) {
529 ret = PTR_ERR(ptr: 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 folio_redirty_for_writepage(wbc, folio);
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
552 return ret;
553 }
554 io_folio = page_folio(bounce_page);
555 }
556
557 __folio_start_writeback(folio, keep_write: keep_towrite);
558
559 /* Now submit buffers to write */
560 do {
561 if (!buffer_async_write(bh))
562 continue;
563 io_submit_add_bh(io, inode, folio, io_folio, bh);
564 } while ((bh = bh->b_this_page) != head);
565
566 return 0;
567}
568

source code of linux/fs/ext4/page-io.c