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;
34 static struct kmem_cache *io_end_vec_cachep;
36 int __init ext4_init_pageio(void)
38 io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
39 if (io_end_cachep == NULL)
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);
50 void ext4_exit_pageio(void)
52 kmem_cache_destroy(io_end_cachep);
53 kmem_cache_destroy(io_end_vec_cachep);
56 struct ext4_io_end_vec *ext4_alloc_io_end_vec(ext4_io_end_t *io_end)
58 struct ext4_io_end_vec *io_end_vec;
60 io_end_vec = kmem_cache_zalloc(io_end_vec_cachep, GFP_NOFS);
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);
68 static void ext4_free_io_end_vec(ext4_io_end_t *io_end)
70 struct ext4_io_end_vec *io_end_vec, *tmp;
72 if (list_empty(&io_end->list_vec))
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);
80 struct ext4_io_end_vec *ext4_last_io_end_vec(ext4_io_end_t *io_end)
82 BUG_ON(list_empty(&io_end->list_vec));
83 return list_last_entry(&io_end->list_vec, struct ext4_io_end_vec, list);
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...
93 static void buffer_io_error(struct buffer_head *bh)
95 printk_ratelimited(KERN_ERR "Buffer I/O error on device %pg, logical block %llu\n",
97 (unsigned long long)bh->b_blocknr);
100 static void ext4_finish_bio(struct bio *bio)
102 struct bio_vec *bvec;
103 struct bvec_iter_all iter_all;
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;
117 if (fscrypt_is_bounce_page(page)) {
119 page = fscrypt_pagecache_page(bounce_page);
122 if (bio->bi_status) {
124 mapping_set_error(page->mapping, -EIO);
126 bh = head = page_buffers(page);
128 * We check all buffers in the page under b_uptodate_lock
129 * to avoid races with other end io clearing async_write flags
131 spin_lock_irqsave(&head->b_uptodate_lock, flags);
133 if (bh_offset(bh) < bio_start ||
134 bh_offset(bh) + bh->b_size > bio_end) {
135 if (buffer_async_write(bh))
139 clear_buffer_async_write(bh);
142 } while ((bh = bh->b_this_page) != head);
143 spin_unlock_irqrestore(&head->b_uptodate_lock, flags);
145 fscrypt_free_bounce_page(bounce_page);
146 end_page_writeback(page);
151 static void ext4_release_io_end(ext4_io_end_t *io_end)
153 struct bio *bio, *next_bio;
155 BUG_ON(!list_empty(&io_end->list));
156 BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
157 WARN_ON(io_end->handle);
159 for (bio = io_end->bio; bio; bio = next_bio) {
160 next_bio = bio->bi_private;
161 ext4_finish_bio(bio);
164 ext4_free_io_end_vec(io_end);
165 kmem_cache_free(io_end_cachep, io_end);
169 * Check a range of space and convert unwritten extents to written. Note that
170 * we are protected from truncate touching same part of extent tree by the
171 * fact that truncate code waits for all DIO to finish (thus exclusion from
172 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
173 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
174 * completed (happens from ext4_free_ioend()).
176 static int ext4_end_io_end(ext4_io_end_t *io_end)
178 struct inode *inode = io_end->inode;
179 handle_t *handle = io_end->handle;
182 ext4_debug("ext4_end_io_nolock: io_end 0x%p from inode %lu,list->next 0x%p,"
184 io_end, inode->i_ino, io_end->list.next, io_end->list.prev);
186 io_end->handle = NULL; /* Following call will use up the handle */
187 ret = ext4_convert_unwritten_io_end_vec(handle, io_end);
188 if (ret < 0 && !ext4_forced_shutdown(EXT4_SB(inode->i_sb))) {
189 ext4_msg(inode->i_sb, KERN_EMERG,
190 "failed to convert unwritten extents to written "
191 "extents -- potential data loss! "
192 "(inode %lu, error %d)", inode->i_ino, ret);
194 ext4_clear_io_unwritten_flag(io_end);
195 ext4_release_io_end(io_end);
199 static void dump_completed_IO(struct inode *inode, struct list_head *head)
202 struct list_head *cur, *before, *after;
203 ext4_io_end_t *io_end, *io_end0, *io_end1;
205 if (list_empty(head))
208 ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
209 list_for_each_entry(io_end, head, list) {
212 io_end0 = container_of(before, ext4_io_end_t, list);
214 io_end1 = container_of(after, ext4_io_end_t, list);
216 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
217 io_end, inode->i_ino, io_end0, io_end1);
222 /* Add the io_end to per-inode completed end_io list. */
223 static void ext4_add_complete_io(ext4_io_end_t *io_end)
225 struct ext4_inode_info *ei = EXT4_I(io_end->inode);
226 struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
227 struct workqueue_struct *wq;
230 /* Only reserved conversions from writeback should enter here */
231 WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
232 WARN_ON(!io_end->handle && sbi->s_journal);
233 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
234 wq = sbi->rsv_conversion_wq;
235 if (list_empty(&ei->i_rsv_conversion_list))
236 queue_work(wq, &ei->i_rsv_conversion_work);
237 list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
238 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
241 static int ext4_do_flush_completed_IO(struct inode *inode,
242 struct list_head *head)
244 ext4_io_end_t *io_end;
245 struct list_head unwritten;
247 struct ext4_inode_info *ei = EXT4_I(inode);
250 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
251 dump_completed_IO(inode, head);
252 list_replace_init(head, &unwritten);
253 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
255 while (!list_empty(&unwritten)) {
256 io_end = list_entry(unwritten.next, ext4_io_end_t, list);
257 BUG_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
258 list_del_init(&io_end->list);
260 err = ext4_end_io_end(io_end);
261 if (unlikely(!ret && err))
268 * work on completed IO, to convert unwritten extents to extents
270 void ext4_end_io_rsv_work(struct work_struct *work)
272 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
273 i_rsv_conversion_work);
274 ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
277 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
279 ext4_io_end_t *io_end = kmem_cache_zalloc(io_end_cachep, flags);
282 io_end->inode = inode;
283 INIT_LIST_HEAD(&io_end->list);
284 INIT_LIST_HEAD(&io_end->list_vec);
285 atomic_set(&io_end->count, 1);
290 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
292 if (atomic_dec_and_test(&io_end->count)) {
293 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) ||
294 list_empty(&io_end->list_vec)) {
295 ext4_release_io_end(io_end);
298 ext4_add_complete_io(io_end);
302 int ext4_put_io_end(ext4_io_end_t *io_end)
306 if (atomic_dec_and_test(&io_end->count)) {
307 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
308 err = ext4_convert_unwritten_io_end_vec(io_end->handle,
310 io_end->handle = NULL;
311 ext4_clear_io_unwritten_flag(io_end);
313 ext4_release_io_end(io_end);
318 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
320 atomic_inc(&io_end->count);
324 /* BIO completion function for page writeback */
325 static void ext4_end_bio(struct bio *bio)
327 ext4_io_end_t *io_end = bio->bi_private;
328 sector_t bi_sector = bio->bi_iter.bi_sector;
329 char b[BDEVNAME_SIZE];
331 if (WARN_ONCE(!io_end, "io_end is NULL: %s: sector %Lu len %u err %d\n",
333 (long long) bio->bi_iter.bi_sector,
334 (unsigned) bio_sectors(bio),
336 ext4_finish_bio(bio);
340 bio->bi_end_io = NULL;
342 if (bio->bi_status) {
343 struct inode *inode = io_end->inode;
345 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
346 "starting block %llu)",
347 bio->bi_status, inode->i_ino,
349 bi_sector >> (inode->i_blkbits - 9));
350 mapping_set_error(inode->i_mapping,
351 blk_status_to_errno(bio->bi_status));
354 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
356 * Link bio into list hanging from io_end. We have to do it
357 * atomically as bio completions can be racing against each
360 bio->bi_private = xchg(&io_end->bio, bio);
361 ext4_put_io_end_defer(io_end);
364 * Drop io_end reference early. Inode can get freed once
367 ext4_put_io_end_defer(io_end);
368 ext4_finish_bio(bio);
373 void ext4_io_submit(struct ext4_io_submit *io)
375 struct bio *bio = io->io_bio;
378 int io_op_flags = io->io_wbc->sync_mode == WB_SYNC_ALL ?
380 io->io_bio->bi_write_hint = io->io_end->inode->i_write_hint;
381 bio_set_op_attrs(io->io_bio, REQ_OP_WRITE, io_op_flags);
382 submit_bio(io->io_bio);
387 void ext4_io_submit_init(struct ext4_io_submit *io,
388 struct writeback_control *wbc)
395 static void io_submit_init_bio(struct ext4_io_submit *io,
396 struct buffer_head *bh)
401 * bio_alloc will _always_ be able to allocate a bio if
402 * __GFP_DIRECT_RECLAIM is set, see comments for bio_alloc_bioset().
404 bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
405 fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO);
406 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
407 bio_set_dev(bio, bh->b_bdev);
408 bio->bi_end_io = ext4_end_bio;
409 bio->bi_private = ext4_get_io_end(io->io_end);
411 io->io_next_block = bh->b_blocknr;
412 wbc_init_bio(io->io_wbc, bio);
415 static void io_submit_add_bh(struct ext4_io_submit *io,
418 struct buffer_head *bh)
422 if (io->io_bio && (bh->b_blocknr != io->io_next_block ||
423 !fscrypt_mergeable_bio_bh(io->io_bio, bh))) {
427 if (io->io_bio == NULL) {
428 io_submit_init_bio(io, bh);
429 io->io_bio->bi_write_hint = inode->i_write_hint;
431 ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
432 if (ret != bh->b_size)
433 goto submit_and_retry;
434 wbc_account_cgroup_owner(io->io_wbc, page, bh->b_size);
438 int ext4_bio_write_page(struct ext4_io_submit *io,
441 struct writeback_control *wbc,
444 struct page *bounce_page = NULL;
445 struct inode *inode = page->mapping->host;
446 unsigned block_start;
447 struct buffer_head *bh, *head;
449 int nr_submitted = 0;
450 int nr_to_submit = 0;
452 BUG_ON(!PageLocked(page));
453 BUG_ON(PageWriteback(page));
456 set_page_writeback_keepwrite(page);
458 set_page_writeback(page);
459 ClearPageError(page);
462 * Comments copied from block_write_full_page:
464 * The page straddles i_size. It must be zeroed out on each and every
465 * writepage invocation because it may be mmapped. "A file is mapped
466 * in multiples of the page size. For a file that is not a multiple of
467 * the page size, the remaining memory is zeroed when mapped, and
468 * writes to that region are not written out to the file."
471 zero_user_segment(page, len, PAGE_SIZE);
473 * In the first loop we prepare and mark buffers to submit. We have to
474 * mark all buffers in the page before submitting so that
475 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
476 * on the first buffer finishes and we are still working on submitting
479 bh = head = page_buffers(page);
481 block_start = bh_offset(bh);
482 if (block_start >= len) {
483 clear_buffer_dirty(bh);
484 set_buffer_uptodate(bh);
487 if (!buffer_dirty(bh) || buffer_delay(bh) ||
488 !buffer_mapped(bh) || buffer_unwritten(bh)) {
489 /* A hole? We can safely clear the dirty bit */
490 if (!buffer_mapped(bh))
491 clear_buffer_dirty(bh);
497 clear_buffer_new(bh);
498 set_buffer_async_write(bh);
500 } while ((bh = bh->b_this_page) != head);
502 bh = head = page_buffers(page);
505 * If any blocks are being written to an encrypted file, encrypt them
506 * into a bounce page. For simplicity, just encrypt until the last
507 * block which might be needed. This may cause some unneeded blocks
508 * (e.g. holes) to be unnecessarily encrypted, but this is rare and
509 * can't happen in the common case of blocksize == PAGE_SIZE.
511 if (fscrypt_inode_uses_fs_layer_crypto(inode) && nr_to_submit) {
512 gfp_t gfp_flags = GFP_NOFS;
513 unsigned int enc_bytes = round_up(len, i_blocksize(inode));
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.
521 gfp_flags = GFP_NOWAIT | __GFP_NOWARN;
523 bounce_page = fscrypt_encrypt_pagecache_blocks(page, enc_bytes,
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_flags = GFP_NOFS;
533 gfp_flags |= __GFP_NOFAIL;
534 congestion_wait(BLK_RW_ASYNC, HZ/50);
538 printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
539 redirty_page_for_writepage(wbc, page);
541 clear_buffer_async_write(bh);
542 bh = bh->b_this_page;
543 } while (bh != head);
548 /* Now submit buffers to write */
550 if (!buffer_async_write(bh))
552 io_submit_add_bh(io, inode,
553 bounce_page ? bounce_page : page, bh);
555 clear_buffer_dirty(bh);
556 } while ((bh = bh->b_this_page) != head);
560 /* Nothing submitted - we have to end page writeback */
562 end_page_writeback(page);