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)
65 struct bvec_iter_all iter_all;
67 bio_for_each_segment_all(bvec, bio, iter_all) {
68 struct page *page = bvec->bv_page;
69 struct page *bounce_page = NULL;
70 struct buffer_head *bh, *head;
71 unsigned bio_start = bvec->bv_offset;
72 unsigned bio_end = bio_start + bvec->bv_len;
73 unsigned under_io = 0;
79 if (fscrypt_is_bounce_page(page)) {
81 page = fscrypt_pagecache_page(bounce_page);
86 mapping_set_error(page->mapping, -EIO);
88 bh = head = page_buffers(page);
90 * We check all buffers in the page under BH_Uptodate_Lock
91 * to avoid races with other end io clearing async_write flags
93 local_irq_save(flags);
94 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
96 if (bh_offset(bh) < bio_start ||
97 bh_offset(bh) + bh->b_size > bio_end) {
98 if (buffer_async_write(bh))
102 clear_buffer_async_write(bh);
103 if (bio->bi_status) {
104 set_buffer_write_io_error(bh);
107 } while ((bh = bh->b_this_page) != head);
108 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
109 local_irq_restore(flags);
111 fscrypt_free_bounce_page(bounce_page);
112 end_page_writeback(page);
117 static void ext4_release_io_end(ext4_io_end_t *io_end)
119 struct bio *bio, *next_bio;
121 BUG_ON(!list_empty(&io_end->list));
122 BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
123 WARN_ON(io_end->handle);
125 for (bio = io_end->bio; bio; bio = next_bio) {
126 next_bio = bio->bi_private;
127 ext4_finish_bio(bio);
130 kmem_cache_free(io_end_cachep, io_end);
134 * Check a range of space and convert unwritten extents to written. Note that
135 * we are protected from truncate touching same part of extent tree by the
136 * fact that truncate code waits for all DIO to finish (thus exclusion from
137 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
138 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
139 * completed (happens from ext4_free_ioend()).
141 static int ext4_end_io(ext4_io_end_t *io)
143 struct inode *inode = io->inode;
144 loff_t offset = io->offset;
145 ssize_t size = io->size;
146 handle_t *handle = io->handle;
149 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
151 io, inode->i_ino, io->list.next, io->list.prev);
153 io->handle = NULL; /* Following call will use up the handle */
154 ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
155 if (ret < 0 && !ext4_forced_shutdown(EXT4_SB(inode->i_sb))) {
156 ext4_msg(inode->i_sb, KERN_EMERG,
157 "failed to convert unwritten extents to written "
158 "extents -- potential data loss! "
159 "(inode %lu, offset %llu, size %zd, error %d)",
160 inode->i_ino, offset, size, ret);
162 ext4_clear_io_unwritten_flag(io);
163 ext4_release_io_end(io);
167 static void dump_completed_IO(struct inode *inode, struct list_head *head)
170 struct list_head *cur, *before, *after;
171 ext4_io_end_t *io, *io0, *io1;
173 if (list_empty(head))
176 ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
177 list_for_each_entry(io, head, list) {
180 io0 = container_of(before, ext4_io_end_t, list);
182 io1 = container_of(after, ext4_io_end_t, list);
184 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
185 io, inode->i_ino, io0, io1);
190 /* Add the io_end to per-inode completed end_io list. */
191 static void ext4_add_complete_io(ext4_io_end_t *io_end)
193 struct ext4_inode_info *ei = EXT4_I(io_end->inode);
194 struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
195 struct workqueue_struct *wq;
198 /* Only reserved conversions from writeback should enter here */
199 WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
200 WARN_ON(!io_end->handle && sbi->s_journal);
201 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
202 wq = sbi->rsv_conversion_wq;
203 if (list_empty(&ei->i_rsv_conversion_list))
204 queue_work(wq, &ei->i_rsv_conversion_work);
205 list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
206 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
209 static int ext4_do_flush_completed_IO(struct inode *inode,
210 struct list_head *head)
213 struct list_head unwritten;
215 struct ext4_inode_info *ei = EXT4_I(inode);
218 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
219 dump_completed_IO(inode, head);
220 list_replace_init(head, &unwritten);
221 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
223 while (!list_empty(&unwritten)) {
224 io = list_entry(unwritten.next, ext4_io_end_t, list);
225 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
226 list_del_init(&io->list);
228 err = ext4_end_io(io);
229 if (unlikely(!ret && err))
236 * work on completed IO, to convert unwritten extents to extents
238 void ext4_end_io_rsv_work(struct work_struct *work)
240 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
241 i_rsv_conversion_work);
242 ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
245 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
247 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
250 INIT_LIST_HEAD(&io->list);
251 atomic_set(&io->count, 1);
256 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
258 if (atomic_dec_and_test(&io_end->count)) {
259 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
260 ext4_release_io_end(io_end);
263 ext4_add_complete_io(io_end);
267 int ext4_put_io_end(ext4_io_end_t *io_end)
271 if (atomic_dec_and_test(&io_end->count)) {
272 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
273 err = ext4_convert_unwritten_extents(io_end->handle,
274 io_end->inode, io_end->offset,
276 io_end->handle = NULL;
277 ext4_clear_io_unwritten_flag(io_end);
279 ext4_release_io_end(io_end);
284 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
286 atomic_inc(&io_end->count);
290 /* BIO completion function for page writeback */
291 static void ext4_end_bio(struct bio *bio)
293 ext4_io_end_t *io_end = bio->bi_private;
294 sector_t bi_sector = bio->bi_iter.bi_sector;
295 char b[BDEVNAME_SIZE];
297 if (WARN_ONCE(!io_end, "io_end is NULL: %s: sector %Lu len %u err %d\n",
299 (long long) bio->bi_iter.bi_sector,
300 (unsigned) bio_sectors(bio),
302 ext4_finish_bio(bio);
306 bio->bi_end_io = NULL;
308 if (bio->bi_status) {
309 struct inode *inode = io_end->inode;
311 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
312 "(offset %llu size %ld starting block %llu)",
313 bio->bi_status, inode->i_ino,
314 (unsigned long long) io_end->offset,
317 bi_sector >> (inode->i_blkbits - 9));
318 mapping_set_error(inode->i_mapping,
319 blk_status_to_errno(bio->bi_status));
322 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
324 * Link bio into list hanging from io_end. We have to do it
325 * atomically as bio completions can be racing against each
328 bio->bi_private = xchg(&io_end->bio, bio);
329 ext4_put_io_end_defer(io_end);
332 * Drop io_end reference early. Inode can get freed once
335 ext4_put_io_end_defer(io_end);
336 ext4_finish_bio(bio);
341 void ext4_io_submit(struct ext4_io_submit *io)
343 struct bio *bio = io->io_bio;
346 int io_op_flags = io->io_wbc->sync_mode == WB_SYNC_ALL ?
348 io->io_bio->bi_write_hint = io->io_end->inode->i_write_hint;
349 bio_set_op_attrs(io->io_bio, REQ_OP_WRITE, io_op_flags);
350 submit_bio(io->io_bio);
355 void ext4_io_submit_init(struct ext4_io_submit *io,
356 struct writeback_control *wbc)
363 static int io_submit_init_bio(struct ext4_io_submit *io,
364 struct buffer_head *bh)
368 bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
371 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
372 bio_set_dev(bio, bh->b_bdev);
373 bio->bi_end_io = ext4_end_bio;
374 bio->bi_private = ext4_get_io_end(io->io_end);
376 io->io_next_block = bh->b_blocknr;
377 wbc_init_bio(io->io_wbc, bio);
381 static int io_submit_add_bh(struct ext4_io_submit *io,
383 struct page *pagecache_page,
384 struct page *bounce_page,
385 struct buffer_head *bh)
389 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
393 if (io->io_bio == NULL) {
394 ret = io_submit_init_bio(io, bh);
397 io->io_bio->bi_write_hint = inode->i_write_hint;
399 ret = bio_add_page(io->io_bio, bounce_page ?: pagecache_page,
400 bh->b_size, bh_offset(bh));
401 if (ret != bh->b_size)
402 goto submit_and_retry;
403 wbc_account_cgroup_owner(io->io_wbc, pagecache_page, bh->b_size);
408 int ext4_bio_write_page(struct ext4_io_submit *io,
411 struct writeback_control *wbc,
414 struct page *bounce_page = NULL;
415 struct inode *inode = page->mapping->host;
416 unsigned block_start;
417 struct buffer_head *bh, *head;
419 int nr_submitted = 0;
420 int nr_to_submit = 0;
422 BUG_ON(!PageLocked(page));
423 BUG_ON(PageWriteback(page));
426 set_page_writeback_keepwrite(page);
428 set_page_writeback(page);
429 ClearPageError(page);
432 * Comments copied from block_write_full_page:
434 * The page straddles i_size. It must be zeroed out on each and every
435 * writepage invocation because it may be mmapped. "A file is mapped
436 * in multiples of the page size. For a file that is not a multiple of
437 * the page size, the remaining memory is zeroed when mapped, and
438 * writes to that region are not written out to the file."
441 zero_user_segment(page, len, PAGE_SIZE);
443 * In the first loop we prepare and mark buffers to submit. We have to
444 * mark all buffers in the page before submitting so that
445 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
446 * on the first buffer finishes and we are still working on submitting
449 bh = head = page_buffers(page);
451 block_start = bh_offset(bh);
452 if (block_start >= len) {
453 clear_buffer_dirty(bh);
454 set_buffer_uptodate(bh);
457 if (!buffer_dirty(bh) || buffer_delay(bh) ||
458 !buffer_mapped(bh) || buffer_unwritten(bh)) {
459 /* A hole? We can safely clear the dirty bit */
460 if (!buffer_mapped(bh))
461 clear_buffer_dirty(bh);
467 clear_buffer_new(bh);
468 set_buffer_async_write(bh);
470 } while ((bh = bh->b_this_page) != head);
472 bh = head = page_buffers(page);
475 * If any blocks are being written to an encrypted file, encrypt them
476 * into a bounce page. For simplicity, just encrypt until the last
477 * block which might be needed. This may cause some unneeded blocks
478 * (e.g. holes) to be unnecessarily encrypted, but this is rare and
479 * can't happen in the common case of blocksize == PAGE_SIZE.
481 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode) && nr_to_submit) {
482 gfp_t gfp_flags = GFP_NOFS;
483 unsigned int enc_bytes = round_up(len, i_blocksize(inode));
486 * Since bounce page allocation uses a mempool, we can only use
487 * a waiting mask (i.e. request guaranteed allocation) on the
488 * first page of the bio. Otherwise it can deadlock.
491 gfp_flags = GFP_NOWAIT | __GFP_NOWARN;
493 bounce_page = fscrypt_encrypt_pagecache_blocks(page, enc_bytes,
495 if (IS_ERR(bounce_page)) {
496 ret = PTR_ERR(bounce_page);
497 if (ret == -ENOMEM &&
498 (io->io_bio || wbc->sync_mode == WB_SYNC_ALL)) {
499 gfp_flags = GFP_NOFS;
503 gfp_flags |= __GFP_NOFAIL;
504 congestion_wait(BLK_RW_ASYNC, HZ/50);
512 /* Now submit buffers to write */
514 if (!buffer_async_write(bh))
516 ret = io_submit_add_bh(io, inode, page, bounce_page, bh);
519 * We only get here on ENOMEM. Not much else
520 * we can do but mark the page as dirty, and
521 * better luck next time.
526 clear_buffer_dirty(bh);
527 } while ((bh = bh->b_this_page) != head);
529 /* Error stopped previous loop? Clean up buffers... */
532 fscrypt_free_bounce_page(bounce_page);
533 printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
534 redirty_page_for_writepage(wbc, page);
536 clear_buffer_async_write(bh);
537 bh = bh->b_this_page;
538 } while (bh != head);
541 /* Nothing submitted - we have to end page writeback */
543 end_page_writeback(page);