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
33 static struct kmem_cache *io_end_cachep;
34 static struct kmem_cache *io_end_vec_cachep;
35
ext4_init_pageio(void)36 int __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
ext4_exit_pageio(void)50 void ext4_exit_pageio(void)
51 {
52 kmem_cache_destroy(io_end_cachep);
53 kmem_cache_destroy(io_end_vec_cachep);
54 }
55
ext4_alloc_io_end_vec(ext4_io_end_t * io_end)56 struct 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
ext4_free_io_end_vec(ext4_io_end_t * io_end)68 static 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
ext4_last_io_end_vec(ext4_io_end_t * io_end)80 struct 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 */
buffer_io_error(struct buffer_head * bh)93 static 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
ext4_finish_bio(struct bio * bio)100 static 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(folio);
116 }
117
118 if (bio->bi_status) {
119 int err = blk_status_to_errno(bio->bi_status);
120 mapping_set_error(folio->mapping, err);
121 }
122 bh = head = folio_buffers(folio);
123 /*
124 * We check all buffers in the folio under b_uptodate_lock
125 * to avoid races with other end io clearing async_write flags
126 */
127 spin_lock_irqsave(&head->b_uptodate_lock, flags);
128 do {
129 if (bh_offset(bh) < bio_start ||
130 bh_offset(bh) + bh->b_size > bio_end) {
131 if (buffer_async_write(bh))
132 under_io++;
133 continue;
134 }
135 clear_buffer_async_write(bh);
136 if (bio->bi_status) {
137 set_buffer_write_io_error(bh);
138 buffer_io_error(bh);
139 }
140 } while ((bh = bh->b_this_page) != head);
141 spin_unlock_irqrestore(&head->b_uptodate_lock, flags);
142 if (!under_io) {
143 fscrypt_free_bounce_page(&io_folio->page);
144 folio_end_writeback(folio);
145 }
146 }
147 }
148
ext4_release_io_end(ext4_io_end_t * io_end)149 static void ext4_release_io_end(ext4_io_end_t *io_end)
150 {
151 struct bio *bio, *next_bio;
152
153 BUG_ON(!list_empty(&io_end->list));
154 BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
155 WARN_ON(io_end->handle);
156
157 for (bio = io_end->bio; bio; bio = next_bio) {
158 next_bio = bio->bi_private;
159 ext4_finish_bio(bio);
160 bio_put(bio);
161 }
162 ext4_free_io_end_vec(io_end);
163 kmem_cache_free(io_end_cachep, io_end);
164 }
165
166 /*
167 * Check a range of space and convert unwritten extents to written. Note that
168 * we are protected from truncate touching same part of extent tree by the
169 * fact that truncate code waits for all DIO to finish (thus exclusion from
170 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
171 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
172 * completed (happens from ext4_free_ioend()).
173 */
ext4_end_io_end(ext4_io_end_t * io_end)174 static int ext4_end_io_end(ext4_io_end_t *io_end)
175 {
176 struct inode *inode = io_end->inode;
177 handle_t *handle = io_end->handle;
178 int ret = 0;
179
180 ext4_debug("ext4_end_io_nolock: io_end 0x%p from inode %lu,list->next 0x%p,"
181 "list->prev 0x%p\n",
182 io_end, inode->i_ino, io_end->list.next, io_end->list.prev);
183
184 io_end->handle = NULL; /* Following call will use up the handle */
185 ret = ext4_convert_unwritten_io_end_vec(handle, io_end);
186 if (ret < 0 && !ext4_forced_shutdown(inode->i_sb)) {
187 ext4_msg(inode->i_sb, KERN_EMERG,
188 "failed to convert unwritten extents to written "
189 "extents -- potential data loss! "
190 "(inode %lu, error %d)", inode->i_ino, ret);
191 }
192 ext4_clear_io_unwritten_flag(io_end);
193 ext4_release_io_end(io_end);
194 return ret;
195 }
196
dump_completed_IO(struct inode * inode,struct list_head * head)197 static void dump_completed_IO(struct inode *inode, struct list_head *head)
198 {
199 #ifdef EXT4FS_DEBUG
200 struct list_head *cur, *before, *after;
201 ext4_io_end_t *io_end, *io_end0, *io_end1;
202
203 if (list_empty(head))
204 return;
205
206 ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
207 list_for_each_entry(io_end, head, list) {
208 cur = &io_end->list;
209 before = cur->prev;
210 io_end0 = container_of(before, ext4_io_end_t, list);
211 after = cur->next;
212 io_end1 = container_of(after, ext4_io_end_t, list);
213
214 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
215 io_end, inode->i_ino, io_end0, io_end1);
216 }
217 #endif
218 }
219
220 /* Add the io_end to per-inode completed end_io list. */
ext4_add_complete_io(ext4_io_end_t * io_end)221 static void ext4_add_complete_io(ext4_io_end_t *io_end)
222 {
223 struct ext4_inode_info *ei = EXT4_I(io_end->inode);
224 struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
225 struct workqueue_struct *wq;
226 unsigned long flags;
227
228 /* Only reserved conversions from writeback should enter here */
229 WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
230 WARN_ON(!io_end->handle && sbi->s_journal);
231 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
232 wq = sbi->rsv_conversion_wq;
233 if (list_empty(&ei->i_rsv_conversion_list))
234 queue_work(wq, &ei->i_rsv_conversion_work);
235 list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
236 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
237 }
238
ext4_do_flush_completed_IO(struct inode * inode,struct list_head * head)239 static int ext4_do_flush_completed_IO(struct inode *inode,
240 struct list_head *head)
241 {
242 ext4_io_end_t *io_end;
243 struct list_head unwritten;
244 unsigned long flags;
245 struct ext4_inode_info *ei = EXT4_I(inode);
246 int err, ret = 0;
247
248 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
249 dump_completed_IO(inode, head);
250 list_replace_init(head, &unwritten);
251 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
252
253 while (!list_empty(&unwritten)) {
254 io_end = list_entry(unwritten.next, ext4_io_end_t, list);
255 BUG_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
256 list_del_init(&io_end->list);
257
258 err = ext4_end_io_end(io_end);
259 if (unlikely(!ret && err))
260 ret = err;
261 }
262 return ret;
263 }
264
265 /*
266 * work on completed IO, to convert unwritten extents to extents
267 */
ext4_end_io_rsv_work(struct work_struct * work)268 void ext4_end_io_rsv_work(struct work_struct *work)
269 {
270 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
271 i_rsv_conversion_work);
272 ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
273 }
274
ext4_init_io_end(struct inode * inode,gfp_t flags)275 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
276 {
277 ext4_io_end_t *io_end = kmem_cache_zalloc(io_end_cachep, flags);
278
279 if (io_end) {
280 io_end->inode = inode;
281 INIT_LIST_HEAD(&io_end->list);
282 INIT_LIST_HEAD(&io_end->list_vec);
283 refcount_set(&io_end->count, 1);
284 }
285 return io_end;
286 }
287
ext4_put_io_end_defer(ext4_io_end_t * io_end)288 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
289 {
290 if (refcount_dec_and_test(&io_end->count)) {
291 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) ||
292 list_empty(&io_end->list_vec)) {
293 ext4_release_io_end(io_end);
294 return;
295 }
296 ext4_add_complete_io(io_end);
297 }
298 }
299
ext4_put_io_end(ext4_io_end_t * io_end)300 int ext4_put_io_end(ext4_io_end_t *io_end)
301 {
302 int err = 0;
303
304 if (refcount_dec_and_test(&io_end->count)) {
305 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
306 err = ext4_convert_unwritten_io_end_vec(io_end->handle,
307 io_end);
308 io_end->handle = NULL;
309 ext4_clear_io_unwritten_flag(io_end);
310 }
311 ext4_release_io_end(io_end);
312 }
313 return err;
314 }
315
ext4_get_io_end(ext4_io_end_t * io_end)316 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
317 {
318 refcount_inc(&io_end->count);
319 return io_end;
320 }
321
322 /* BIO completion function for page writeback */
ext4_end_bio(struct bio * bio)323 static void ext4_end_bio(struct bio *bio)
324 {
325 ext4_io_end_t *io_end = bio->bi_private;
326 sector_t bi_sector = bio->bi_iter.bi_sector;
327
328 if (WARN_ONCE(!io_end, "io_end is NULL: %pg: sector %Lu len %u err %d\n",
329 bio->bi_bdev,
330 (long long) bio->bi_iter.bi_sector,
331 (unsigned) bio_sectors(bio),
332 bio->bi_status)) {
333 ext4_finish_bio(bio);
334 bio_put(bio);
335 return;
336 }
337 bio->bi_end_io = NULL;
338
339 if (bio->bi_status) {
340 struct inode *inode = io_end->inode;
341
342 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
343 "starting block %llu)",
344 bio->bi_status, inode->i_ino,
345 (unsigned long long)
346 bi_sector >> (inode->i_blkbits - 9));
347 mapping_set_error(inode->i_mapping,
348 blk_status_to_errno(bio->bi_status));
349 }
350
351 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
352 /*
353 * Link bio into list hanging from io_end. We have to do it
354 * atomically as bio completions can be racing against each
355 * other.
356 */
357 bio->bi_private = xchg(&io_end->bio, bio);
358 ext4_put_io_end_defer(io_end);
359 } else {
360 /*
361 * Drop io_end reference early. Inode can get freed once
362 * we finish the bio.
363 */
364 ext4_put_io_end_defer(io_end);
365 ext4_finish_bio(bio);
366 bio_put(bio);
367 }
368 }
369
ext4_io_submit(struct ext4_io_submit * io)370 void ext4_io_submit(struct ext4_io_submit *io)
371 {
372 struct bio *bio = io->io_bio;
373
374 if (bio) {
375 if (io->io_wbc->sync_mode == WB_SYNC_ALL)
376 io->io_bio->bi_opf |= REQ_SYNC;
377 submit_bio(io->io_bio);
378 }
379 io->io_bio = NULL;
380 }
381
ext4_io_submit_init(struct ext4_io_submit * io,struct writeback_control * wbc)382 void ext4_io_submit_init(struct ext4_io_submit *io,
383 struct writeback_control *wbc)
384 {
385 io->io_wbc = wbc;
386 io->io_bio = NULL;
387 io->io_end = NULL;
388 }
389
io_submit_init_bio(struct ext4_io_submit * io,struct buffer_head * bh)390 static void io_submit_init_bio(struct ext4_io_submit *io,
391 struct buffer_head *bh)
392 {
393 struct bio *bio;
394
395 /*
396 * bio_alloc will _always_ be able to allocate a bio if
397 * __GFP_DIRECT_RECLAIM is set, see comments for bio_alloc_bioset().
398 */
399 bio = bio_alloc(bh->b_bdev, BIO_MAX_VECS, REQ_OP_WRITE, GFP_NOIO);
400 fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO);
401 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
402 bio->bi_end_io = ext4_end_bio;
403 bio->bi_private = ext4_get_io_end(io->io_end);
404 io->io_bio = bio;
405 io->io_next_block = bh->b_blocknr;
406 wbc_init_bio(io->io_wbc, bio);
407 }
408
io_submit_add_bh(struct ext4_io_submit * io,struct inode * inode,struct folio * folio,struct folio * io_folio,struct buffer_head * bh)409 static void io_submit_add_bh(struct ext4_io_submit *io,
410 struct inode *inode,
411 struct folio *folio,
412 struct folio *io_folio,
413 struct buffer_head *bh)
414 {
415 if (io->io_bio && (bh->b_blocknr != io->io_next_block ||
416 !fscrypt_mergeable_bio_bh(io->io_bio, bh))) {
417 submit_and_retry:
418 ext4_io_submit(io);
419 }
420 if (io->io_bio == NULL)
421 io_submit_init_bio(io, bh);
422 if (!bio_add_folio(io->io_bio, io_folio, bh->b_size, bh_offset(bh)))
423 goto submit_and_retry;
424 wbc_account_cgroup_owner(io->io_wbc, &folio->page, bh->b_size);
425 io->io_next_block++;
426 }
427
ext4_bio_write_folio(struct ext4_io_submit * io,struct folio * folio,size_t len)428 int ext4_bio_write_folio(struct ext4_io_submit *io, struct folio *folio,
429 size_t len)
430 {
431 struct folio *io_folio = folio;
432 struct inode *inode = folio->mapping->host;
433 unsigned block_start;
434 struct buffer_head *bh, *head;
435 int ret = 0;
436 int nr_to_submit = 0;
437 struct writeback_control *wbc = io->io_wbc;
438 bool keep_towrite = false;
439
440 BUG_ON(!folio_test_locked(folio));
441 BUG_ON(folio_test_writeback(folio));
442
443 /*
444 * Comments copied from block_write_full_folio:
445 *
446 * The folio straddles i_size. It must be zeroed out on each and every
447 * writepage invocation because it may be mmapped. "A file is mapped
448 * in multiples of the page size. For a file that is not a multiple of
449 * the page size, the remaining memory is zeroed when mapped, and
450 * writes to that region are not written out to the file."
451 */
452 if (len < folio_size(folio))
453 folio_zero_segment(folio, len, folio_size(folio));
454 /*
455 * In the first loop we prepare and mark buffers to submit. We have to
456 * mark all buffers in the folio before submitting so that
457 * folio_end_writeback() cannot be called from ext4_end_bio() when IO
458 * on the first buffer finishes and we are still working on submitting
459 * the second buffer.
460 */
461 bh = head = folio_buffers(folio);
462 do {
463 block_start = bh_offset(bh);
464 if (block_start >= len) {
465 clear_buffer_dirty(bh);
466 set_buffer_uptodate(bh);
467 continue;
468 }
469 if (!buffer_dirty(bh) || buffer_delay(bh) ||
470 !buffer_mapped(bh) || buffer_unwritten(bh)) {
471 /* A hole? We can safely clear the dirty bit */
472 if (!buffer_mapped(bh))
473 clear_buffer_dirty(bh);
474 /*
475 * Keeping dirty some buffer we cannot write? Make sure
476 * to redirty the folio and keep TOWRITE tag so that
477 * racing WB_SYNC_ALL writeback does not skip the folio.
478 * This happens e.g. when doing writeout for
479 * transaction commit or when journalled data is not
480 * yet committed.
481 */
482 if (buffer_dirty(bh) ||
483 (buffer_jbd(bh) && buffer_jbddirty(bh))) {
484 if (!folio_test_dirty(folio))
485 folio_redirty_for_writepage(wbc, folio);
486 keep_towrite = true;
487 }
488 continue;
489 }
490 if (buffer_new(bh))
491 clear_buffer_new(bh);
492 set_buffer_async_write(bh);
493 clear_buffer_dirty(bh);
494 nr_to_submit++;
495 } while ((bh = bh->b_this_page) != head);
496
497 /* Nothing to submit? Just unlock the folio... */
498 if (!nr_to_submit)
499 return 0;
500
501 bh = head = folio_buffers(folio);
502
503 /*
504 * If any blocks are being written to an encrypted file, encrypt them
505 * into a bounce page. For simplicity, just encrypt until the last
506 * block which might be needed. This may cause some unneeded blocks
507 * (e.g. holes) to be unnecessarily encrypted, but this is rare and
508 * can't happen in the common case of blocksize == PAGE_SIZE.
509 */
510 if (fscrypt_inode_uses_fs_layer_crypto(inode)) {
511 gfp_t gfp_flags = GFP_NOFS;
512 unsigned int enc_bytes = round_up(len, i_blocksize(inode));
513 struct page *bounce_page;
514
515 /*
516 * Since bounce page allocation uses a mempool, we can only use
517 * a waiting mask (i.e. request guaranteed allocation) on the
518 * first page of the bio. Otherwise it can deadlock.
519 */
520 if (io->io_bio)
521 gfp_flags = GFP_NOWAIT | __GFP_NOWARN;
522 retry_encrypt:
523 bounce_page = fscrypt_encrypt_pagecache_blocks(&folio->page,
524 enc_bytes, 0, gfp_flags);
525 if (IS_ERR(bounce_page)) {
526 ret = PTR_ERR(bounce_page);
527 if (ret == -ENOMEM &&
528 (io->io_bio || wbc->sync_mode == WB_SYNC_ALL)) {
529 gfp_t new_gfp_flags = GFP_NOFS;
530 if (io->io_bio)
531 ext4_io_submit(io);
532 else
533 new_gfp_flags |= __GFP_NOFAIL;
534 memalloc_retry_wait(gfp_flags);
535 gfp_flags = new_gfp_flags;
536 goto retry_encrypt;
537 }
538
539 printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
540 folio_redirty_for_writepage(wbc, folio);
541 do {
542 if (buffer_async_write(bh)) {
543 clear_buffer_async_write(bh);
544 set_buffer_dirty(bh);
545 }
546 bh = bh->b_this_page;
547 } while (bh != head);
548
549 return ret;
550 }
551 io_folio = page_folio(bounce_page);
552 }
553
554 __folio_start_writeback(folio, keep_towrite);
555
556 /* Now submit buffers to write */
557 do {
558 if (!buffer_async_write(bh))
559 continue;
560 io_submit_add_bh(io, inode, folio, io_folio, bh);
561 } while ((bh = bh->b_this_page) != head);
562
563 return 0;
564 }
565