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;
35 int __init ext4_init_pageio(void)
37 io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
38 if (io_end_cachep == NULL)
43 void ext4_exit_pageio(void)
45 kmem_cache_destroy(io_end_cachep);
49 * Print an buffer I/O error compatible with the fs/buffer.c. This
50 * provides compatibility with dmesg scrapers that look for a specific
51 * buffer I/O error message. We really need a unified error reporting
52 * structure to userspace ala Digital Unix's uerf system, but it's
53 * probably not going to happen in my lifetime, due to LKML politics...
55 static void buffer_io_error(struct buffer_head *bh)
57 printk_ratelimited(KERN_ERR "Buffer I/O error on device %pg, logical block %llu\n",
59 (unsigned long long)bh->b_blocknr);
62 static void ext4_finish_bio(struct bio *bio)
67 bio_for_each_segment_all(bvec, bio, i) {
68 struct page *page = bvec->bv_page;
69 #ifdef CONFIG_EXT4_FS_ENCRYPTION
70 struct page *data_page = NULL;
72 struct buffer_head *bh, *head;
73 unsigned bio_start = bvec->bv_offset;
74 unsigned bio_end = bio_start + bvec->bv_len;
75 unsigned under_io = 0;
81 #ifdef CONFIG_EXT4_FS_ENCRYPTION
83 /* The bounce data pages are unmapped. */
85 fscrypt_pullback_bio_page(&page, false);
91 mapping_set_error(page->mapping, -EIO);
93 bh = head = page_buffers(page);
95 * We check all buffers in the page under BH_Uptodate_Lock
96 * to avoid races with other end io clearing async_write flags
98 local_irq_save(flags);
99 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
101 if (bh_offset(bh) < bio_start ||
102 bh_offset(bh) + bh->b_size > bio_end) {
103 if (buffer_async_write(bh))
107 clear_buffer_async_write(bh);
108 if (bio->bi_status) {
109 set_buffer_write_io_error(bh);
112 } while ((bh = bh->b_this_page) != head);
113 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
114 local_irq_restore(flags);
116 #ifdef CONFIG_EXT4_FS_ENCRYPTION
118 fscrypt_restore_control_page(data_page);
120 end_page_writeback(page);
125 static void ext4_release_io_end(ext4_io_end_t *io_end)
127 struct bio *bio, *next_bio;
129 BUG_ON(!list_empty(&io_end->list));
130 BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
131 WARN_ON(io_end->handle);
133 for (bio = io_end->bio; bio; bio = next_bio) {
134 next_bio = bio->bi_private;
135 ext4_finish_bio(bio);
138 kmem_cache_free(io_end_cachep, io_end);
142 * Check a range of space and convert unwritten extents to written. Note that
143 * we are protected from truncate touching same part of extent tree by the
144 * fact that truncate code waits for all DIO to finish (thus exclusion from
145 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
146 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
147 * completed (happens from ext4_free_ioend()).
149 static int ext4_end_io(ext4_io_end_t *io)
151 struct inode *inode = io->inode;
152 loff_t offset = io->offset;
153 ssize_t size = io->size;
154 handle_t *handle = io->handle;
157 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
159 io, inode->i_ino, io->list.next, io->list.prev);
161 io->handle = NULL; /* Following call will use up the handle */
162 ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
163 if (ret < 0 && !ext4_forced_shutdown(EXT4_SB(inode->i_sb))) {
164 ext4_msg(inode->i_sb, KERN_EMERG,
165 "failed to convert unwritten extents to written "
166 "extents -- potential data loss! "
167 "(inode %lu, offset %llu, size %zd, error %d)",
168 inode->i_ino, offset, size, ret);
170 ext4_clear_io_unwritten_flag(io);
171 ext4_release_io_end(io);
175 static void dump_completed_IO(struct inode *inode, struct list_head *head)
178 struct list_head *cur, *before, *after;
179 ext4_io_end_t *io, *io0, *io1;
181 if (list_empty(head))
184 ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
185 list_for_each_entry(io, head, list) {
188 io0 = container_of(before, ext4_io_end_t, list);
190 io1 = container_of(after, ext4_io_end_t, list);
192 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
193 io, inode->i_ino, io0, io1);
198 /* Add the io_end to per-inode completed end_io list. */
199 static void ext4_add_complete_io(ext4_io_end_t *io_end)
201 struct ext4_inode_info *ei = EXT4_I(io_end->inode);
202 struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
203 struct workqueue_struct *wq;
206 /* Only reserved conversions from writeback should enter here */
207 WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
208 WARN_ON(!io_end->handle && sbi->s_journal);
209 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
210 wq = sbi->rsv_conversion_wq;
211 if (list_empty(&ei->i_rsv_conversion_list))
212 queue_work(wq, &ei->i_rsv_conversion_work);
213 list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
214 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
217 static int ext4_do_flush_completed_IO(struct inode *inode,
218 struct list_head *head)
221 struct list_head unwritten;
223 struct ext4_inode_info *ei = EXT4_I(inode);
226 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
227 dump_completed_IO(inode, head);
228 list_replace_init(head, &unwritten);
229 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
231 while (!list_empty(&unwritten)) {
232 io = list_entry(unwritten.next, ext4_io_end_t, list);
233 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
234 list_del_init(&io->list);
236 err = ext4_end_io(io);
237 if (unlikely(!ret && err))
244 * work on completed IO, to convert unwritten extents to extents
246 void ext4_end_io_rsv_work(struct work_struct *work)
248 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
249 i_rsv_conversion_work);
250 ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
253 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
255 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
258 INIT_LIST_HEAD(&io->list);
259 atomic_set(&io->count, 1);
264 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
266 if (atomic_dec_and_test(&io_end->count)) {
267 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
268 ext4_release_io_end(io_end);
271 ext4_add_complete_io(io_end);
275 int ext4_put_io_end(ext4_io_end_t *io_end)
279 if (atomic_dec_and_test(&io_end->count)) {
280 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
281 err = ext4_convert_unwritten_extents(io_end->handle,
282 io_end->inode, io_end->offset,
284 io_end->handle = NULL;
285 ext4_clear_io_unwritten_flag(io_end);
287 ext4_release_io_end(io_end);
292 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
294 atomic_inc(&io_end->count);
298 /* BIO completion function for page writeback */
299 static void ext4_end_bio(struct bio *bio)
301 ext4_io_end_t *io_end = bio->bi_private;
302 sector_t bi_sector = bio->bi_iter.bi_sector;
303 char b[BDEVNAME_SIZE];
305 if (WARN_ONCE(!io_end, "io_end is NULL: %s: sector %Lu len %u err %d\n",
307 (long long) bio->bi_iter.bi_sector,
308 (unsigned) bio_sectors(bio),
310 ext4_finish_bio(bio);
314 bio->bi_end_io = NULL;
316 if (bio->bi_status) {
317 struct inode *inode = io_end->inode;
319 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
320 "(offset %llu size %ld starting block %llu)",
321 bio->bi_status, inode->i_ino,
322 (unsigned long long) io_end->offset,
325 bi_sector >> (inode->i_blkbits - 9));
326 mapping_set_error(inode->i_mapping,
327 blk_status_to_errno(bio->bi_status));
330 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
332 * Link bio into list hanging from io_end. We have to do it
333 * atomically as bio completions can be racing against each
336 bio->bi_private = xchg(&io_end->bio, bio);
337 ext4_put_io_end_defer(io_end);
340 * Drop io_end reference early. Inode can get freed once
343 ext4_put_io_end_defer(io_end);
344 ext4_finish_bio(bio);
349 void ext4_io_submit(struct ext4_io_submit *io)
351 struct bio *bio = io->io_bio;
354 int io_op_flags = io->io_wbc->sync_mode == WB_SYNC_ALL ?
356 io->io_bio->bi_write_hint = io->io_end->inode->i_write_hint;
357 bio_set_op_attrs(io->io_bio, REQ_OP_WRITE, io_op_flags);
358 submit_bio(io->io_bio);
363 void ext4_io_submit_init(struct ext4_io_submit *io,
364 struct writeback_control *wbc)
371 static int io_submit_init_bio(struct ext4_io_submit *io,
372 struct buffer_head *bh)
376 bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
379 wbc_init_bio(io->io_wbc, bio);
380 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
381 bio_set_dev(bio, bh->b_bdev);
382 bio->bi_end_io = ext4_end_bio;
383 bio->bi_private = ext4_get_io_end(io->io_end);
385 io->io_next_block = bh->b_blocknr;
389 static int io_submit_add_bh(struct ext4_io_submit *io,
391 struct page *pagecache_page,
392 struct page *bounce_page,
393 struct buffer_head *bh)
397 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
401 if (io->io_bio == NULL) {
402 ret = io_submit_init_bio(io, bh);
405 io->io_bio->bi_write_hint = inode->i_write_hint;
407 ret = bio_add_page(io->io_bio, bounce_page ?: pagecache_page,
408 bh->b_size, bh_offset(bh));
409 if (ret != bh->b_size)
410 goto submit_and_retry;
411 wbc_account_io(io->io_wbc, pagecache_page, bh->b_size);
416 int ext4_bio_write_page(struct ext4_io_submit *io,
419 struct writeback_control *wbc,
422 struct page *data_page = NULL;
423 struct inode *inode = page->mapping->host;
424 unsigned block_start;
425 struct buffer_head *bh, *head;
427 int nr_submitted = 0;
428 int nr_to_submit = 0;
430 BUG_ON(!PageLocked(page));
431 BUG_ON(PageWriteback(page));
434 set_page_writeback_keepwrite(page);
436 set_page_writeback(page);
437 ClearPageError(page);
440 * Comments copied from block_write_full_page:
442 * The page straddles i_size. It must be zeroed out on each and every
443 * writepage invocation because it may be mmapped. "A file is mapped
444 * in multiples of the page size. For a file that is not a multiple of
445 * the page size, the remaining memory is zeroed when mapped, and
446 * writes to that region are not written out to the file."
449 zero_user_segment(page, len, PAGE_SIZE);
451 * In the first loop we prepare and mark buffers to submit. We have to
452 * mark all buffers in the page before submitting so that
453 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
454 * on the first buffer finishes and we are still working on submitting
457 bh = head = page_buffers(page);
459 block_start = bh_offset(bh);
460 if (block_start >= len) {
461 clear_buffer_dirty(bh);
462 set_buffer_uptodate(bh);
465 if (!buffer_dirty(bh) || buffer_delay(bh) ||
466 !buffer_mapped(bh) || buffer_unwritten(bh)) {
467 /* A hole? We can safely clear the dirty bit */
468 if (!buffer_mapped(bh))
469 clear_buffer_dirty(bh);
474 if (buffer_new(bh)) {
475 clear_buffer_new(bh);
476 clean_bdev_bh_alias(bh);
478 set_buffer_async_write(bh);
480 } while ((bh = bh->b_this_page) != head);
482 bh = head = page_buffers(page);
484 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode) &&
486 gfp_t gfp_flags = GFP_NOFS;
489 * Since bounce page allocation uses a mempool, we can only use
490 * a waiting mask (i.e. request guaranteed allocation) on the
491 * first page of the bio. Otherwise it can deadlock.
494 gfp_flags = GFP_NOWAIT | __GFP_NOWARN;
496 data_page = fscrypt_encrypt_page(inode, page, PAGE_SIZE, 0,
497 page->index, gfp_flags);
498 if (IS_ERR(data_page)) {
499 ret = PTR_ERR(data_page);
500 if (ret == -ENOMEM &&
501 (io->io_bio || wbc->sync_mode == WB_SYNC_ALL)) {
502 gfp_flags = GFP_NOFS;
506 gfp_flags |= __GFP_NOFAIL;
507 congestion_wait(BLK_RW_ASYNC, HZ/50);
515 /* Now submit buffers to write */
517 if (!buffer_async_write(bh))
519 ret = io_submit_add_bh(io, inode, page, data_page, bh);
522 * We only get here on ENOMEM. Not much else
523 * we can do but mark the page as dirty, and
524 * better luck next time.
529 clear_buffer_dirty(bh);
530 } while ((bh = bh->b_this_page) != head);
532 /* Error stopped previous loop? Clean up buffers... */
536 fscrypt_restore_control_page(data_page);
537 printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
538 redirty_page_for_writepage(wbc, page);
540 clear_buffer_async_write(bh);
541 bh = bh->b_this_page;
542 } while (bh != head);
545 /* Nothing submitted - we have to end page writeback */
547 end_page_writeback(page);