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);
105 } while ((bh = bh->b_this_page) != head);
106 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
107 local_irq_restore(flags);
109 fscrypt_free_bounce_page(bounce_page);
110 end_page_writeback(page);
115 static void ext4_release_io_end(ext4_io_end_t *io_end)
117 struct bio *bio, *next_bio;
119 BUG_ON(!list_empty(&io_end->list));
120 BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
121 WARN_ON(io_end->handle);
123 for (bio = io_end->bio; bio; bio = next_bio) {
124 next_bio = bio->bi_private;
125 ext4_finish_bio(bio);
128 kmem_cache_free(io_end_cachep, io_end);
132 * Check a range of space and convert unwritten extents to written. Note that
133 * we are protected from truncate touching same part of extent tree by the
134 * fact that truncate code waits for all DIO to finish (thus exclusion from
135 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
136 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
137 * completed (happens from ext4_free_ioend()).
139 static int ext4_end_io(ext4_io_end_t *io)
141 struct inode *inode = io->inode;
142 loff_t offset = io->offset;
143 ssize_t size = io->size;
144 handle_t *handle = io->handle;
147 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
149 io, inode->i_ino, io->list.next, io->list.prev);
151 io->handle = NULL; /* Following call will use up the handle */
152 ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
153 if (ret < 0 && !ext4_forced_shutdown(EXT4_SB(inode->i_sb))) {
154 ext4_msg(inode->i_sb, KERN_EMERG,
155 "failed to convert unwritten extents to written "
156 "extents -- potential data loss! "
157 "(inode %lu, offset %llu, size %zd, error %d)",
158 inode->i_ino, offset, size, ret);
160 ext4_clear_io_unwritten_flag(io);
161 ext4_release_io_end(io);
165 static void dump_completed_IO(struct inode *inode, struct list_head *head)
168 struct list_head *cur, *before, *after;
169 ext4_io_end_t *io, *io0, *io1;
171 if (list_empty(head))
174 ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
175 list_for_each_entry(io, head, list) {
178 io0 = container_of(before, ext4_io_end_t, list);
180 io1 = container_of(after, ext4_io_end_t, list);
182 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
183 io, inode->i_ino, io0, io1);
188 /* Add the io_end to per-inode completed end_io list. */
189 static void ext4_add_complete_io(ext4_io_end_t *io_end)
191 struct ext4_inode_info *ei = EXT4_I(io_end->inode);
192 struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
193 struct workqueue_struct *wq;
196 /* Only reserved conversions from writeback should enter here */
197 WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
198 WARN_ON(!io_end->handle && sbi->s_journal);
199 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
200 wq = sbi->rsv_conversion_wq;
201 if (list_empty(&ei->i_rsv_conversion_list))
202 queue_work(wq, &ei->i_rsv_conversion_work);
203 list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
204 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
207 static int ext4_do_flush_completed_IO(struct inode *inode,
208 struct list_head *head)
211 struct list_head unwritten;
213 struct ext4_inode_info *ei = EXT4_I(inode);
216 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
217 dump_completed_IO(inode, head);
218 list_replace_init(head, &unwritten);
219 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
221 while (!list_empty(&unwritten)) {
222 io = list_entry(unwritten.next, ext4_io_end_t, list);
223 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
224 list_del_init(&io->list);
226 err = ext4_end_io(io);
227 if (unlikely(!ret && err))
234 * work on completed IO, to convert unwritten extents to extents
236 void ext4_end_io_rsv_work(struct work_struct *work)
238 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
239 i_rsv_conversion_work);
240 ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
243 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
245 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
248 INIT_LIST_HEAD(&io->list);
249 atomic_set(&io->count, 1);
254 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
256 if (atomic_dec_and_test(&io_end->count)) {
257 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
258 ext4_release_io_end(io_end);
261 ext4_add_complete_io(io_end);
265 int ext4_put_io_end(ext4_io_end_t *io_end)
269 if (atomic_dec_and_test(&io_end->count)) {
270 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
271 err = ext4_convert_unwritten_extents(io_end->handle,
272 io_end->inode, io_end->offset,
274 io_end->handle = NULL;
275 ext4_clear_io_unwritten_flag(io_end);
277 ext4_release_io_end(io_end);
282 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
284 atomic_inc(&io_end->count);
288 /* BIO completion function for page writeback */
289 static void ext4_end_bio(struct bio *bio)
291 ext4_io_end_t *io_end = bio->bi_private;
292 sector_t bi_sector = bio->bi_iter.bi_sector;
293 char b[BDEVNAME_SIZE];
295 if (WARN_ONCE(!io_end, "io_end is NULL: %s: sector %Lu len %u err %d\n",
297 (long long) bio->bi_iter.bi_sector,
298 (unsigned) bio_sectors(bio),
300 ext4_finish_bio(bio);
304 bio->bi_end_io = NULL;
306 if (bio->bi_status) {
307 struct inode *inode = io_end->inode;
309 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
310 "(offset %llu size %ld starting block %llu)",
311 bio->bi_status, inode->i_ino,
312 (unsigned long long) io_end->offset,
315 bi_sector >> (inode->i_blkbits - 9));
316 mapping_set_error(inode->i_mapping,
317 blk_status_to_errno(bio->bi_status));
320 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
322 * Link bio into list hanging from io_end. We have to do it
323 * atomically as bio completions can be racing against each
326 bio->bi_private = xchg(&io_end->bio, bio);
327 ext4_put_io_end_defer(io_end);
330 * Drop io_end reference early. Inode can get freed once
333 ext4_put_io_end_defer(io_end);
334 ext4_finish_bio(bio);
339 void ext4_io_submit(struct ext4_io_submit *io)
341 struct bio *bio = io->io_bio;
344 int io_op_flags = io->io_wbc->sync_mode == WB_SYNC_ALL ?
346 io->io_bio->bi_write_hint = io->io_end->inode->i_write_hint;
347 bio_set_op_attrs(io->io_bio, REQ_OP_WRITE, io_op_flags);
348 submit_bio(io->io_bio);
353 void ext4_io_submit_init(struct ext4_io_submit *io,
354 struct writeback_control *wbc)
361 static int io_submit_init_bio(struct ext4_io_submit *io,
362 struct buffer_head *bh)
366 bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
369 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
370 bio_set_dev(bio, bh->b_bdev);
371 bio->bi_end_io = ext4_end_bio;
372 bio->bi_private = ext4_get_io_end(io->io_end);
374 io->io_next_block = bh->b_blocknr;
375 wbc_init_bio(io->io_wbc, bio);
379 static int io_submit_add_bh(struct ext4_io_submit *io,
382 struct buffer_head *bh)
386 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
390 if (io->io_bio == NULL) {
391 ret = io_submit_init_bio(io, bh);
394 io->io_bio->bi_write_hint = inode->i_write_hint;
396 ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
397 if (ret != bh->b_size)
398 goto submit_and_retry;
399 wbc_account_cgroup_owner(io->io_wbc, page, bh->b_size);
404 int ext4_bio_write_page(struct ext4_io_submit *io,
407 struct writeback_control *wbc,
410 struct page *bounce_page = NULL;
411 struct inode *inode = page->mapping->host;
412 unsigned block_start;
413 struct buffer_head *bh, *head;
415 int nr_submitted = 0;
416 int nr_to_submit = 0;
418 BUG_ON(!PageLocked(page));
419 BUG_ON(PageWriteback(page));
422 set_page_writeback_keepwrite(page);
424 set_page_writeback(page);
425 ClearPageError(page);
428 * Comments copied from block_write_full_page:
430 * The page straddles i_size. It must be zeroed out on each and every
431 * writepage invocation because it may be mmapped. "A file is mapped
432 * in multiples of the page size. For a file that is not a multiple of
433 * the page size, the remaining memory is zeroed when mapped, and
434 * writes to that region are not written out to the file."
437 zero_user_segment(page, len, PAGE_SIZE);
439 * In the first loop we prepare and mark buffers to submit. We have to
440 * mark all buffers in the page before submitting so that
441 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
442 * on the first buffer finishes and we are still working on submitting
445 bh = head = page_buffers(page);
447 block_start = bh_offset(bh);
448 if (block_start >= len) {
449 clear_buffer_dirty(bh);
450 set_buffer_uptodate(bh);
453 if (!buffer_dirty(bh) || buffer_delay(bh) ||
454 !buffer_mapped(bh) || buffer_unwritten(bh)) {
455 /* A hole? We can safely clear the dirty bit */
456 if (!buffer_mapped(bh))
457 clear_buffer_dirty(bh);
463 clear_buffer_new(bh);
464 set_buffer_async_write(bh);
466 } while ((bh = bh->b_this_page) != head);
468 bh = head = page_buffers(page);
471 * If any blocks are being written to an encrypted file, encrypt them
472 * into a bounce page. For simplicity, just encrypt until the last
473 * block which might be needed. This may cause some unneeded blocks
474 * (e.g. holes) to be unnecessarily encrypted, but this is rare and
475 * can't happen in the common case of blocksize == PAGE_SIZE.
477 if (IS_ENCRYPTED(inode) && S_ISREG(inode->i_mode) && nr_to_submit) {
478 gfp_t gfp_flags = GFP_NOFS;
479 unsigned int enc_bytes = round_up(len, i_blocksize(inode));
482 bounce_page = fscrypt_encrypt_pagecache_blocks(page, enc_bytes,
484 if (IS_ERR(bounce_page)) {
485 ret = PTR_ERR(bounce_page);
486 if (ret == -ENOMEM && wbc->sync_mode == WB_SYNC_ALL) {
489 congestion_wait(BLK_RW_ASYNC, HZ/50);
491 gfp_flags |= __GFP_NOFAIL;
499 /* Now submit buffers to write */
501 if (!buffer_async_write(bh))
503 ret = io_submit_add_bh(io, inode, bounce_page ?: page, bh);
506 * We only get here on ENOMEM. Not much else
507 * we can do but mark the page as dirty, and
508 * better luck next time.
513 clear_buffer_dirty(bh);
514 } while ((bh = bh->b_this_page) != head);
516 /* Error stopped previous loop? Clean up buffers... */
519 fscrypt_free_bounce_page(bounce_page);
520 printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
521 redirty_page_for_writepage(wbc, page);
523 clear_buffer_async_write(bh);
524 bh = bh->b_this_page;
525 } while (bh != head);
528 /* Nothing submitted - we have to end page writeback */
530 end_page_writeback(page);