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)
66 struct bvec_iter_all iter_all;
68 bio_for_each_segment_all(bvec, bio, i, iter_all) {
69 struct page *page = bvec->bv_page;
70 #ifdef CONFIG_FS_ENCRYPTION
71 struct page *data_page = NULL;
73 struct buffer_head *bh, *head;
74 unsigned bio_start = bvec->bv_offset;
75 unsigned bio_end = bio_start + bvec->bv_len;
76 unsigned under_io = 0;
82 #ifdef CONFIG_FS_ENCRYPTION
84 /* The bounce data pages are unmapped. */
86 fscrypt_pullback_bio_page(&page, false);
92 mapping_set_error(page->mapping, -EIO);
94 bh = head = page_buffers(page);
96 * We check all buffers in the page under BH_Uptodate_Lock
97 * to avoid races with other end io clearing async_write flags
99 local_irq_save(flags);
100 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
102 if (bh_offset(bh) < bio_start ||
103 bh_offset(bh) + bh->b_size > bio_end) {
104 if (buffer_async_write(bh))
108 clear_buffer_async_write(bh);
111 } while ((bh = bh->b_this_page) != head);
112 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
113 local_irq_restore(flags);
115 #ifdef CONFIG_FS_ENCRYPTION
117 fscrypt_restore_control_page(data_page);
119 end_page_writeback(page);
124 static void ext4_release_io_end(ext4_io_end_t *io_end)
126 struct bio *bio, *next_bio;
128 BUG_ON(!list_empty(&io_end->list));
129 BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
130 WARN_ON(io_end->handle);
132 for (bio = io_end->bio; bio; bio = next_bio) {
133 next_bio = bio->bi_private;
134 ext4_finish_bio(bio);
137 kmem_cache_free(io_end_cachep, io_end);
141 * Check a range of space and convert unwritten extents to written. Note that
142 * we are protected from truncate touching same part of extent tree by the
143 * fact that truncate code waits for all DIO to finish (thus exclusion from
144 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
145 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
146 * completed (happens from ext4_free_ioend()).
148 static int ext4_end_io(ext4_io_end_t *io)
150 struct inode *inode = io->inode;
151 loff_t offset = io->offset;
152 ssize_t size = io->size;
153 handle_t *handle = io->handle;
156 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
158 io, inode->i_ino, io->list.next, io->list.prev);
160 io->handle = NULL; /* Following call will use up the handle */
161 ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
162 if (ret < 0 && !ext4_forced_shutdown(EXT4_SB(inode->i_sb))) {
163 ext4_msg(inode->i_sb, KERN_EMERG,
164 "failed to convert unwritten extents to written "
165 "extents -- potential data loss! "
166 "(inode %lu, offset %llu, size %zd, error %d)",
167 inode->i_ino, offset, size, ret);
169 ext4_clear_io_unwritten_flag(io);
170 ext4_release_io_end(io);
174 static void dump_completed_IO(struct inode *inode, struct list_head *head)
177 struct list_head *cur, *before, *after;
178 ext4_io_end_t *io, *io0, *io1;
180 if (list_empty(head))
183 ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
184 list_for_each_entry(io, head, list) {
187 io0 = container_of(before, ext4_io_end_t, list);
189 io1 = container_of(after, ext4_io_end_t, list);
191 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
192 io, inode->i_ino, io0, io1);
197 /* Add the io_end to per-inode completed end_io list. */
198 static void ext4_add_complete_io(ext4_io_end_t *io_end)
200 struct ext4_inode_info *ei = EXT4_I(io_end->inode);
201 struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
202 struct workqueue_struct *wq;
205 /* Only reserved conversions from writeback should enter here */
206 WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
207 WARN_ON(!io_end->handle && sbi->s_journal);
208 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
209 wq = sbi->rsv_conversion_wq;
210 if (list_empty(&ei->i_rsv_conversion_list))
211 queue_work(wq, &ei->i_rsv_conversion_work);
212 list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
213 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
216 static int ext4_do_flush_completed_IO(struct inode *inode,
217 struct list_head *head)
220 struct list_head unwritten;
222 struct ext4_inode_info *ei = EXT4_I(inode);
225 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
226 dump_completed_IO(inode, head);
227 list_replace_init(head, &unwritten);
228 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
230 while (!list_empty(&unwritten)) {
231 io = list_entry(unwritten.next, ext4_io_end_t, list);
232 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
233 list_del_init(&io->list);
235 err = ext4_end_io(io);
236 if (unlikely(!ret && err))
243 * work on completed IO, to convert unwritten extents to extents
245 void ext4_end_io_rsv_work(struct work_struct *work)
247 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
248 i_rsv_conversion_work);
249 ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
252 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
254 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
257 INIT_LIST_HEAD(&io->list);
258 atomic_set(&io->count, 1);
263 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
265 if (atomic_dec_and_test(&io_end->count)) {
266 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
267 ext4_release_io_end(io_end);
270 ext4_add_complete_io(io_end);
274 int ext4_put_io_end(ext4_io_end_t *io_end)
278 if (atomic_dec_and_test(&io_end->count)) {
279 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
280 err = ext4_convert_unwritten_extents(io_end->handle,
281 io_end->inode, io_end->offset,
283 io_end->handle = NULL;
284 ext4_clear_io_unwritten_flag(io_end);
286 ext4_release_io_end(io_end);
291 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
293 atomic_inc(&io_end->count);
297 /* BIO completion function for page writeback */
298 static void ext4_end_bio(struct bio *bio)
300 ext4_io_end_t *io_end = bio->bi_private;
301 sector_t bi_sector = bio->bi_iter.bi_sector;
302 char b[BDEVNAME_SIZE];
304 if (WARN_ONCE(!io_end, "io_end is NULL: %s: sector %Lu len %u err %d\n",
306 (long long) bio->bi_iter.bi_sector,
307 (unsigned) bio_sectors(bio),
309 ext4_finish_bio(bio);
313 bio->bi_end_io = NULL;
315 if (bio->bi_status) {
316 struct inode *inode = io_end->inode;
318 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
319 "(offset %llu size %ld starting block %llu)",
320 bio->bi_status, inode->i_ino,
321 (unsigned long long) io_end->offset,
324 bi_sector >> (inode->i_blkbits - 9));
325 mapping_set_error(inode->i_mapping,
326 blk_status_to_errno(bio->bi_status));
329 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
331 * Link bio into list hanging from io_end. We have to do it
332 * atomically as bio completions can be racing against each
335 bio->bi_private = xchg(&io_end->bio, bio);
336 ext4_put_io_end_defer(io_end);
339 * Drop io_end reference early. Inode can get freed once
342 ext4_put_io_end_defer(io_end);
343 ext4_finish_bio(bio);
348 void ext4_io_submit(struct ext4_io_submit *io)
350 struct bio *bio = io->io_bio;
353 int io_op_flags = io->io_wbc->sync_mode == WB_SYNC_ALL ?
355 io->io_bio->bi_write_hint = io->io_end->inode->i_write_hint;
356 bio_set_op_attrs(io->io_bio, REQ_OP_WRITE, io_op_flags);
357 submit_bio(io->io_bio);
362 void ext4_io_submit_init(struct ext4_io_submit *io,
363 struct writeback_control *wbc)
370 static int io_submit_init_bio(struct ext4_io_submit *io,
371 struct buffer_head *bh)
375 bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
378 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
379 bio_set_dev(bio, bh->b_bdev);
380 bio->bi_end_io = ext4_end_bio;
381 bio->bi_private = ext4_get_io_end(io->io_end);
383 io->io_next_block = bh->b_blocknr;
384 wbc_init_bio(io->io_wbc, bio);
388 static int io_submit_add_bh(struct ext4_io_submit *io,
391 struct buffer_head *bh)
395 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
399 if (io->io_bio == NULL) {
400 ret = io_submit_init_bio(io, bh);
403 io->io_bio->bi_write_hint = inode->i_write_hint;
405 ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
406 if (ret != bh->b_size)
407 goto submit_and_retry;
408 wbc_account_io(io->io_wbc, page, bh->b_size);
413 int ext4_bio_write_page(struct ext4_io_submit *io,
416 struct writeback_control *wbc,
419 struct page *data_page = NULL;
420 struct inode *inode = page->mapping->host;
421 unsigned block_start;
422 struct buffer_head *bh, *head;
424 int nr_submitted = 0;
425 int nr_to_submit = 0;
427 BUG_ON(!PageLocked(page));
428 BUG_ON(PageWriteback(page));
431 set_page_writeback_keepwrite(page);
433 set_page_writeback(page);
434 ClearPageError(page);
437 * Comments copied from block_write_full_page:
439 * The page straddles i_size. It must be zeroed out on each and every
440 * writepage invocation because it may be mmapped. "A file is mapped
441 * in multiples of the page size. For a file that is not a multiple of
442 * the page size, the remaining memory is zeroed when mapped, and
443 * writes to that region are not written out to the file."
446 zero_user_segment(page, len, PAGE_SIZE);
448 * In the first loop we prepare and mark buffers to submit. We have to
449 * mark all buffers in the page before submitting so that
450 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
451 * on the first buffer finishes and we are still working on submitting
454 bh = head = page_buffers(page);
456 block_start = bh_offset(bh);
457 if (block_start >= len) {
458 clear_buffer_dirty(bh);
459 set_buffer_uptodate(bh);
462 if (!buffer_dirty(bh) || buffer_delay(bh) ||
463 !buffer_mapped(bh) || buffer_unwritten(bh)) {
464 /* A hole? We can safely clear the dirty bit */
465 if (!buffer_mapped(bh))
466 clear_buffer_dirty(bh);
472 clear_buffer_new(bh);
473 set_buffer_async_write(bh);
475 } while ((bh = bh->b_this_page) != head);
477 bh = head = page_buffers(page);
479 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode) && nr_to_submit) {
480 gfp_t gfp_flags = GFP_NOFS;
483 data_page = fscrypt_encrypt_page(inode, page, PAGE_SIZE, 0,
484 page->index, gfp_flags);
485 if (IS_ERR(data_page)) {
486 ret = PTR_ERR(data_page);
487 if (ret == -ENOMEM && wbc->sync_mode == WB_SYNC_ALL) {
490 congestion_wait(BLK_RW_ASYNC, HZ/50);
492 gfp_flags |= __GFP_NOFAIL;
500 /* Now submit buffers to write */
502 if (!buffer_async_write(bh))
504 ret = io_submit_add_bh(io, inode,
505 data_page ? data_page : page, bh);
508 * We only get here on ENOMEM. Not much else
509 * we can do but mark the page as dirty, and
510 * better luck next time.
515 clear_buffer_dirty(bh);
516 } while ((bh = bh->b_this_page) != head);
518 /* Error stopped previous loop? Clean up buffers... */
522 fscrypt_restore_control_page(data_page);
523 printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
524 redirty_page_for_writepage(wbc, page);
526 clear_buffer_async_write(bh);
527 bh = bh->b_this_page;
528 } while (bh != head);
531 /* Nothing submitted - we have to end page writeback */
533 end_page_writeback(page);