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 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
ext4_release_io_end(ext4_io_end_t * io_end)150 static 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 */
ext4_end_io_end(ext4_io_end_t * io_end)175 static 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
dump_completed_IO(struct inode * inode,struct list_head * head)198 static 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. */
ext4_add_complete_io(ext4_io_end_t * io_end)222 static 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
ext4_do_flush_completed_IO(struct inode * inode,struct list_head * head)240 static 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 */
ext4_end_io_rsv_work(struct work_struct * work)269 void 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
ext4_init_io_end(struct inode * inode,gfp_t flags)276 ext4_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
ext4_put_io_end_defer(ext4_io_end_t * io_end)289 void 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
ext4_put_io_end(ext4_io_end_t * io_end)301 int 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
ext4_get_io_end(ext4_io_end_t * io_end)317 ext4_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 */
ext4_end_bio(struct bio * bio)324 static 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
ext4_io_submit(struct ext4_io_submit * io)371 void 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
ext4_io_submit_init(struct ext4_io_submit * io,struct writeback_control * wbc)383 void 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
io_submit_init_bio(struct ext4_io_submit * io,struct buffer_head * bh)391 static 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
io_submit_add_bh(struct ext4_io_submit * io,struct inode * inode,struct page * page,struct buffer_head * bh)410 static 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))) {
419 submit_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
ext4_bio_write_page(struct ext4_io_submit * io,struct page * page,int len,bool keep_towrite)431 int ext4_bio_write_page(struct ext4_io_submit *io,
432 struct page *page,
433 int len,
434 bool keep_towrite)
435 {
436 struct page *bounce_page = NULL;
437 struct inode *inode = page->mapping->host;
438 unsigned block_start;
439 struct buffer_head *bh, *head;
440 int ret = 0;
441 int nr_submitted = 0;
442 int nr_to_submit = 0;
443 struct writeback_control *wbc = io->io_wbc;
444
445 BUG_ON(!PageLocked(page));
446 BUG_ON(PageWriteback(page));
447
448 if (keep_towrite)
449 set_page_writeback_keepwrite(page);
450 else
451 set_page_writeback(page);
452 ClearPageError(page);
453
454 /*
455 * Comments copied from block_write_full_page:
456 *
457 * The page straddles i_size. It must be zeroed out on each and every
458 * writepage invocation because it may be mmapped. "A file is mapped
459 * in multiples of the page size. For a file that is not a multiple of
460 * the page size, the remaining memory is zeroed when mapped, and
461 * writes to that region are not written out to the file."
462 */
463 if (len < PAGE_SIZE)
464 zero_user_segment(page, len, PAGE_SIZE);
465 /*
466 * In the first loop we prepare and mark buffers to submit. We have to
467 * mark all buffers in the page before submitting so that
468 * end_page_writeback() cannot be called from ext4_end_bio() when IO
469 * on the first buffer finishes and we are still working on submitting
470 * the second buffer.
471 */
472 bh = head = page_buffers(page);
473 do {
474 block_start = bh_offset(bh);
475 if (block_start >= len) {
476 clear_buffer_dirty(bh);
477 set_buffer_uptodate(bh);
478 continue;
479 }
480 if (!buffer_dirty(bh) || buffer_delay(bh) ||
481 !buffer_mapped(bh) || buffer_unwritten(bh)) {
482 /* A hole? We can safely clear the dirty bit */
483 if (!buffer_mapped(bh))
484 clear_buffer_dirty(bh);
485 if (io->io_bio)
486 ext4_io_submit(io);
487 continue;
488 }
489 if (buffer_new(bh))
490 clear_buffer_new(bh);
491 set_buffer_async_write(bh);
492 nr_to_submit++;
493 } while ((bh = bh->b_this_page) != head);
494
495 bh = head = page_buffers(page);
496
497 /*
498 * If any blocks are being written to an encrypted file, encrypt them
499 * into a bounce page. For simplicity, just encrypt until the last
500 * block which might be needed. This may cause some unneeded blocks
501 * (e.g. holes) to be unnecessarily encrypted, but this is rare and
502 * can't happen in the common case of blocksize == PAGE_SIZE.
503 */
504 if (fscrypt_inode_uses_fs_layer_crypto(inode) && nr_to_submit) {
505 gfp_t gfp_flags = GFP_NOFS;
506 unsigned int enc_bytes = round_up(len, i_blocksize(inode));
507
508 /*
509 * Since bounce page allocation uses a mempool, we can only use
510 * a waiting mask (i.e. request guaranteed allocation) on the
511 * first page of the bio. Otherwise it can deadlock.
512 */
513 if (io->io_bio)
514 gfp_flags = GFP_NOWAIT | __GFP_NOWARN;
515 retry_encrypt:
516 bounce_page = fscrypt_encrypt_pagecache_blocks(page, enc_bytes,
517 0, gfp_flags);
518 if (IS_ERR(bounce_page)) {
519 ret = PTR_ERR(bounce_page);
520 if (ret == -ENOMEM &&
521 (io->io_bio || wbc->sync_mode == WB_SYNC_ALL)) {
522 gfp_t new_gfp_flags = GFP_NOFS;
523 if (io->io_bio)
524 ext4_io_submit(io);
525 else
526 new_gfp_flags |= __GFP_NOFAIL;
527 memalloc_retry_wait(gfp_flags);
528 gfp_flags = new_gfp_flags;
529 goto retry_encrypt;
530 }
531
532 printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
533 redirty_page_for_writepage(wbc, page);
534 do {
535 clear_buffer_async_write(bh);
536 bh = bh->b_this_page;
537 } while (bh != head);
538 goto unlock;
539 }
540 }
541
542 /* Now submit buffers to write */
543 do {
544 if (!buffer_async_write(bh))
545 continue;
546 io_submit_add_bh(io, inode,
547 bounce_page ? bounce_page : page, bh);
548 nr_submitted++;
549 clear_buffer_dirty(bh);
550 } while ((bh = bh->b_this_page) != head);
551
552 unlock:
553 unlock_page(page);
554 /* Nothing submitted - we have to end page writeback */
555 if (!nr_submitted)
556 end_page_writeback(page);
557 return ret;
558 }
559