2 * linux/fs/ext4/page-io.c
4 * This contains the new page_io functions for ext4
6 * Written by Theodore Ts'o, 2010.
10 #include <linux/time.h>
11 #include <linux/highuid.h>
12 #include <linux/pagemap.h>
13 #include <linux/quotaops.h>
14 #include <linux/string.h>
15 #include <linux/buffer_head.h>
16 #include <linux/writeback.h>
17 #include <linux/pagevec.h>
18 #include <linux/mpage.h>
19 #include <linux/namei.h>
20 #include <linux/uio.h>
21 #include <linux/bio.h>
22 #include <linux/workqueue.h>
23 #include <linux/kernel.h>
24 #include <linux/slab.h>
26 #include <linux/backing-dev.h>
28 #include "ext4_jbd2.h"
32 static struct kmem_cache *io_end_cachep;
34 int __init ext4_init_pageio(void)
36 io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
37 if (io_end_cachep == NULL)
42 void ext4_exit_pageio(void)
44 kmem_cache_destroy(io_end_cachep);
48 * Print an buffer I/O error compatible with the fs/buffer.c. This
49 * provides compatibility with dmesg scrapers that look for a specific
50 * buffer I/O error message. We really need a unified error reporting
51 * structure to userspace ala Digital Unix's uerf system, but it's
52 * probably not going to happen in my lifetime, due to LKML politics...
54 static void buffer_io_error(struct buffer_head *bh)
56 printk_ratelimited(KERN_ERR "Buffer I/O error on device %pg, logical block %llu\n",
58 (unsigned long long)bh->b_blocknr);
61 static void ext4_finish_bio(struct bio *bio)
66 bio_for_each_segment_all(bvec, bio, i) {
67 struct page *page = bvec->bv_page;
68 #ifdef CONFIG_EXT4_FS_ENCRYPTION
69 struct page *data_page = NULL;
71 struct buffer_head *bh, *head;
72 unsigned bio_start = bvec->bv_offset;
73 unsigned bio_end = bio_start + bvec->bv_len;
74 unsigned under_io = 0;
80 #ifdef CONFIG_EXT4_FS_ENCRYPTION
82 /* The bounce data pages are unmapped. */
84 fscrypt_pullback_bio_page(&page, false);
90 mapping_set_error(page->mapping, -EIO);
92 bh = head = page_buffers(page);
94 * We check all buffers in the page under BH_Uptodate_Lock
95 * to avoid races with other end io clearing async_write flags
97 local_irq_save(flags);
98 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
100 if (bh_offset(bh) < bio_start ||
101 bh_offset(bh) + bh->b_size > bio_end) {
102 if (buffer_async_write(bh))
106 clear_buffer_async_write(bh);
109 } while ((bh = bh->b_this_page) != head);
110 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
111 local_irq_restore(flags);
113 #ifdef CONFIG_EXT4_FS_ENCRYPTION
115 fscrypt_restore_control_page(data_page);
117 end_page_writeback(page);
122 static void ext4_release_io_end(ext4_io_end_t *io_end)
124 struct bio *bio, *next_bio;
126 BUG_ON(!list_empty(&io_end->list));
127 BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
128 WARN_ON(io_end->handle);
130 for (bio = io_end->bio; bio; bio = next_bio) {
131 next_bio = bio->bi_private;
132 ext4_finish_bio(bio);
135 kmem_cache_free(io_end_cachep, io_end);
139 * Check a range of space and convert unwritten extents to written. Note that
140 * we are protected from truncate touching same part of extent tree by the
141 * fact that truncate code waits for all DIO to finish (thus exclusion from
142 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
143 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
144 * completed (happens from ext4_free_ioend()).
146 static int ext4_end_io(ext4_io_end_t *io)
148 struct inode *inode = io->inode;
149 loff_t offset = io->offset;
150 ssize_t size = io->size;
151 handle_t *handle = io->handle;
154 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
156 io, inode->i_ino, io->list.next, io->list.prev);
158 io->handle = NULL; /* Following call will use up the handle */
159 ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
160 if (ret < 0 && !ext4_forced_shutdown(EXT4_SB(inode->i_sb))) {
161 ext4_msg(inode->i_sb, KERN_EMERG,
162 "failed to convert unwritten extents to written "
163 "extents -- potential data loss! "
164 "(inode %lu, offset %llu, size %zd, error %d)",
165 inode->i_ino, offset, size, ret);
167 ext4_clear_io_unwritten_flag(io);
168 ext4_release_io_end(io);
172 static void dump_completed_IO(struct inode *inode, struct list_head *head)
175 struct list_head *cur, *before, *after;
176 ext4_io_end_t *io, *io0, *io1;
178 if (list_empty(head))
181 ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
182 list_for_each_entry(io, head, list) {
185 io0 = container_of(before, ext4_io_end_t, list);
187 io1 = container_of(after, ext4_io_end_t, list);
189 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
190 io, inode->i_ino, io0, io1);
195 /* Add the io_end to per-inode completed end_io list. */
196 static void ext4_add_complete_io(ext4_io_end_t *io_end)
198 struct ext4_inode_info *ei = EXT4_I(io_end->inode);
199 struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb);
200 struct workqueue_struct *wq;
203 /* Only reserved conversions from writeback should enter here */
204 WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
205 WARN_ON(!io_end->handle && sbi->s_journal);
206 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
207 wq = sbi->rsv_conversion_wq;
208 if (list_empty(&ei->i_rsv_conversion_list))
209 queue_work(wq, &ei->i_rsv_conversion_work);
210 list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
211 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
214 static int ext4_do_flush_completed_IO(struct inode *inode,
215 struct list_head *head)
218 struct list_head unwritten;
220 struct ext4_inode_info *ei = EXT4_I(inode);
223 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
224 dump_completed_IO(inode, head);
225 list_replace_init(head, &unwritten);
226 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
228 while (!list_empty(&unwritten)) {
229 io = list_entry(unwritten.next, ext4_io_end_t, list);
230 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
231 list_del_init(&io->list);
233 err = ext4_end_io(io);
234 if (unlikely(!ret && err))
241 * work on completed IO, to convert unwritten extents to extents
243 void ext4_end_io_rsv_work(struct work_struct *work)
245 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
246 i_rsv_conversion_work);
247 ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
250 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
252 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
255 INIT_LIST_HEAD(&io->list);
256 atomic_set(&io->count, 1);
261 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
263 if (atomic_dec_and_test(&io_end->count)) {
264 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
265 ext4_release_io_end(io_end);
268 ext4_add_complete_io(io_end);
272 int ext4_put_io_end(ext4_io_end_t *io_end)
276 if (atomic_dec_and_test(&io_end->count)) {
277 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
278 err = ext4_convert_unwritten_extents(io_end->handle,
279 io_end->inode, io_end->offset,
281 io_end->handle = NULL;
282 ext4_clear_io_unwritten_flag(io_end);
284 ext4_release_io_end(io_end);
289 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
291 atomic_inc(&io_end->count);
295 /* BIO completion function for page writeback */
296 static void ext4_end_bio(struct bio *bio)
298 ext4_io_end_t *io_end = bio->bi_private;
299 sector_t bi_sector = bio->bi_iter.bi_sector;
300 char b[BDEVNAME_SIZE];
302 if (WARN_ONCE(!io_end, "io_end is NULL: %s: sector %Lu len %u err %d\n",
304 (long long) bio->bi_iter.bi_sector,
305 (unsigned) bio_sectors(bio),
307 ext4_finish_bio(bio);
311 bio->bi_end_io = NULL;
313 if (bio->bi_status) {
314 struct inode *inode = io_end->inode;
316 ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu "
317 "(offset %llu size %ld starting block %llu)",
318 bio->bi_status, inode->i_ino,
319 (unsigned long long) io_end->offset,
322 bi_sector >> (inode->i_blkbits - 9));
323 mapping_set_error(inode->i_mapping,
324 blk_status_to_errno(bio->bi_status));
327 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
329 * Link bio into list hanging from io_end. We have to do it
330 * atomically as bio completions can be racing against each
333 bio->bi_private = xchg(&io_end->bio, bio);
334 ext4_put_io_end_defer(io_end);
337 * Drop io_end reference early. Inode can get freed once
340 ext4_put_io_end_defer(io_end);
341 ext4_finish_bio(bio);
346 void ext4_io_submit(struct ext4_io_submit *io)
348 struct bio *bio = io->io_bio;
351 int io_op_flags = io->io_wbc->sync_mode == WB_SYNC_ALL ?
353 io->io_bio->bi_write_hint = io->io_end->inode->i_write_hint;
354 bio_set_op_attrs(io->io_bio, REQ_OP_WRITE, io_op_flags);
355 submit_bio(io->io_bio);
360 void ext4_io_submit_init(struct ext4_io_submit *io,
361 struct writeback_control *wbc)
368 static int io_submit_init_bio(struct ext4_io_submit *io,
369 struct buffer_head *bh)
373 bio = bio_alloc(GFP_NOIO, BIO_MAX_PAGES);
376 wbc_init_bio(io->io_wbc, bio);
377 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
378 bio_set_dev(bio, bh->b_bdev);
379 bio->bi_end_io = ext4_end_bio;
380 bio->bi_private = ext4_get_io_end(io->io_end);
382 io->io_next_block = bh->b_blocknr;
386 static int io_submit_add_bh(struct ext4_io_submit *io,
389 struct buffer_head *bh)
393 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
397 if (io->io_bio == NULL) {
398 ret = io_submit_init_bio(io, bh);
401 io->io_bio->bi_write_hint = inode->i_write_hint;
403 ret = bio_add_page(io->io_bio, page, bh->b_size, bh_offset(bh));
404 if (ret != bh->b_size)
405 goto submit_and_retry;
406 wbc_account_io(io->io_wbc, page, bh->b_size);
411 int ext4_bio_write_page(struct ext4_io_submit *io,
414 struct writeback_control *wbc,
417 struct page *data_page = NULL;
418 struct inode *inode = page->mapping->host;
419 unsigned block_start;
420 struct buffer_head *bh, *head;
422 int nr_submitted = 0;
423 int nr_to_submit = 0;
425 BUG_ON(!PageLocked(page));
426 BUG_ON(PageWriteback(page));
429 set_page_writeback_keepwrite(page);
431 set_page_writeback(page);
432 ClearPageError(page);
435 * Comments copied from block_write_full_page:
437 * The page straddles i_size. It must be zeroed out on each and every
438 * writepage invocation because it may be mmapped. "A file is mapped
439 * in multiples of the page size. For a file that is not a multiple of
440 * the page size, the remaining memory is zeroed when mapped, and
441 * writes to that region are not written out to the file."
444 zero_user_segment(page, len, PAGE_SIZE);
446 * In the first loop we prepare and mark buffers to submit. We have to
447 * mark all buffers in the page before submitting so that
448 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
449 * on the first buffer finishes and we are still working on submitting
452 bh = head = page_buffers(page);
454 block_start = bh_offset(bh);
455 if (block_start >= len) {
456 clear_buffer_dirty(bh);
457 set_buffer_uptodate(bh);
460 if (!buffer_dirty(bh) || buffer_delay(bh) ||
461 !buffer_mapped(bh) || buffer_unwritten(bh)) {
462 /* A hole? We can safely clear the dirty bit */
463 if (!buffer_mapped(bh))
464 clear_buffer_dirty(bh);
469 if (buffer_new(bh)) {
470 clear_buffer_new(bh);
471 clean_bdev_bh_alias(bh);
473 set_buffer_async_write(bh);
475 } while ((bh = bh->b_this_page) != head);
477 bh = head = page_buffers(page);
479 if (ext4_encrypted_inode(inode) && S_ISREG(inode->i_mode) &&
481 gfp_t gfp_flags = GFP_NOFS;
484 data_page = fscrypt_encrypt_page(inode, page, PAGE_SIZE, 0,
485 page->index, gfp_flags);
486 if (IS_ERR(data_page)) {
487 ret = PTR_ERR(data_page);
488 if (ret == -ENOMEM && wbc->sync_mode == WB_SYNC_ALL) {
491 congestion_wait(BLK_RW_ASYNC, HZ/50);
493 gfp_flags |= __GFP_NOFAIL;
501 /* Now submit buffers to write */
503 if (!buffer_async_write(bh))
505 ret = io_submit_add_bh(io, inode,
506 data_page ? data_page : page, bh);
509 * We only get here on ENOMEM. Not much else
510 * we can do but mark the page as dirty, and
511 * better luck next time.
516 clear_buffer_dirty(bh);
517 } while ((bh = bh->b_this_page) != head);
519 /* Error stopped previous loop? Clean up buffers... */
523 fscrypt_restore_control_page(data_page);
524 printk_ratelimited(KERN_ERR "%s: ret = %d\n", __func__, ret);
525 redirty_page_for_writepage(wbc, page);
527 clear_buffer_async_write(bh);
528 bh = bh->b_this_page;
529 } while (bh != head);
532 /* Nothing submitted - we have to end page writeback */
534 end_page_writeback(page);