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/jbd2.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/aio.h>
22 #include <linux/uio.h>
23 #include <linux/bio.h>
24 #include <linux/workqueue.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.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 char b[BDEVNAME_SIZE];
58 printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
59 bdevname(bh->b_bdev, b),
60 (unsigned long long)bh->b_blocknr);
63 static void ext4_finish_bio(struct bio *bio)
66 int error = !test_bit(BIO_UPTODATE, &bio->bi_flags);
68 for (i = 0; i < bio->bi_vcnt; i++) {
69 struct bio_vec *bvec = &bio->bi_io_vec[i];
70 struct page *page = bvec->bv_page;
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;
82 set_bit(AS_EIO, &page->mapping->flags);
84 bh = head = page_buffers(page);
86 * We check all buffers in the page under BH_Uptodate_Lock
87 * to avoid races with other end io clearing async_write flags
89 local_irq_save(flags);
90 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
92 if (bh_offset(bh) < bio_start ||
93 bh_offset(bh) + bh->b_size > bio_end) {
94 if (buffer_async_write(bh))
98 clear_buffer_async_write(bh);
101 } while ((bh = bh->b_this_page) != head);
102 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
103 local_irq_restore(flags);
105 end_page_writeback(page);
109 static void ext4_release_io_end(ext4_io_end_t *io_end)
111 struct bio *bio, *next_bio;
113 BUG_ON(!list_empty(&io_end->list));
114 BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
115 WARN_ON(io_end->handle);
117 if (atomic_dec_and_test(&EXT4_I(io_end->inode)->i_ioend_count))
118 wake_up_all(ext4_ioend_wq(io_end->inode));
120 for (bio = io_end->bio; bio; bio = next_bio) {
121 next_bio = bio->bi_private;
122 ext4_finish_bio(bio);
125 if (io_end->flag & EXT4_IO_END_DIRECT)
126 inode_dio_done(io_end->inode);
128 aio_complete(io_end->iocb, io_end->result, 0);
129 kmem_cache_free(io_end_cachep, io_end);
132 static void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end)
134 struct inode *inode = io_end->inode;
136 io_end->flag &= ~EXT4_IO_END_UNWRITTEN;
137 /* Wake up anyone waiting on unwritten extent conversion */
138 if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten))
139 wake_up_all(ext4_ioend_wq(inode));
143 * Check a range of space and convert unwritten extents to written. Note that
144 * we are protected from truncate touching same part of extent tree by the
145 * fact that truncate code waits for all DIO to finish (thus exclusion from
146 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
147 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
148 * completed (happens from ext4_free_ioend()).
150 static int ext4_end_io(ext4_io_end_t *io)
152 struct inode *inode = io->inode;
153 loff_t offset = io->offset;
154 ssize_t size = io->size;
155 handle_t *handle = io->handle;
158 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
160 io, inode->i_ino, io->list.next, io->list.prev);
162 io->handle = NULL; /* Following call will use up the handle */
163 ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
165 ext4_msg(inode->i_sb, KERN_EMERG,
166 "failed to convert unwritten extents to written "
167 "extents -- potential data loss! "
168 "(inode %lu, offset %llu, size %zd, error %d)",
169 inode->i_ino, offset, size, ret);
171 ext4_clear_io_unwritten_flag(io);
172 ext4_release_io_end(io);
176 static void dump_completed_IO(struct inode *inode, struct list_head *head)
179 struct list_head *cur, *before, *after;
180 ext4_io_end_t *io, *io0, *io1;
182 if (list_empty(head))
185 ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
186 list_for_each_entry(io, head, list) {
189 io0 = container_of(before, ext4_io_end_t, list);
191 io1 = container_of(after, ext4_io_end_t, list);
193 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
194 io, inode->i_ino, io0, io1);
199 /* Add the io_end to per-inode completed end_io list. */
200 static void ext4_add_complete_io(ext4_io_end_t *io_end)
202 struct ext4_inode_info *ei = EXT4_I(io_end->inode);
203 struct workqueue_struct *wq;
206 BUG_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
207 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
208 if (io_end->handle) {
209 wq = EXT4_SB(io_end->inode->i_sb)->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);
214 wq = EXT4_SB(io_end->inode->i_sb)->unrsv_conversion_wq;
215 if (list_empty(&ei->i_unrsv_conversion_list))
216 queue_work(wq, &ei->i_unrsv_conversion_work);
217 list_add_tail(&io_end->list, &ei->i_unrsv_conversion_list);
219 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
222 static int ext4_do_flush_completed_IO(struct inode *inode,
223 struct list_head *head)
226 struct list_head unwritten;
228 struct ext4_inode_info *ei = EXT4_I(inode);
231 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
232 dump_completed_IO(inode, head);
233 list_replace_init(head, &unwritten);
234 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
236 while (!list_empty(&unwritten)) {
237 io = list_entry(unwritten.next, ext4_io_end_t, list);
238 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
239 list_del_init(&io->list);
241 err = ext4_end_io(io);
242 if (unlikely(!ret && err))
249 * work on completed IO, to convert unwritten extents to extents
251 void ext4_end_io_rsv_work(struct work_struct *work)
253 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
254 i_rsv_conversion_work);
255 ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
258 void ext4_end_io_unrsv_work(struct work_struct *work)
260 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
261 i_unrsv_conversion_work);
262 ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_unrsv_conversion_list);
265 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
267 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
269 atomic_inc(&EXT4_I(inode)->i_ioend_count);
271 INIT_LIST_HEAD(&io->list);
272 atomic_set(&io->count, 1);
277 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
279 if (atomic_dec_and_test(&io_end->count)) {
280 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
281 ext4_release_io_end(io_end);
284 ext4_add_complete_io(io_end);
288 int ext4_put_io_end(ext4_io_end_t *io_end)
292 if (atomic_dec_and_test(&io_end->count)) {
293 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
294 err = ext4_convert_unwritten_extents(io_end->handle,
295 io_end->inode, io_end->offset,
297 io_end->handle = NULL;
298 ext4_clear_io_unwritten_flag(io_end);
300 ext4_release_io_end(io_end);
305 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
307 atomic_inc(&io_end->count);
311 static void ext4_end_bio(struct bio *bio, int error)
313 ext4_io_end_t *io_end = bio->bi_private;
314 sector_t bi_sector = bio->bi_sector;
317 bio->bi_end_io = NULL;
318 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
321 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
323 * Link bio into list hanging from io_end. We have to do it
324 * atomically as bio completions can be racing against each
327 bio->bi_private = xchg(&io_end->bio, bio);
329 ext4_finish_bio(bio);
334 struct inode *inode = io_end->inode;
336 ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
337 "(offset %llu size %ld starting block %llu)",
339 (unsigned long long) io_end->offset,
342 bi_sector >> (inode->i_blkbits - 9));
344 ext4_put_io_end_defer(io_end);
347 void ext4_io_submit(struct ext4_io_submit *io)
349 struct bio *bio = io->io_bio;
353 submit_bio(io->io_op, io->io_bio);
354 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
360 void ext4_io_submit_init(struct ext4_io_submit *io,
361 struct writeback_control *wbc)
363 io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
368 static int io_submit_init_bio(struct ext4_io_submit *io,
369 struct buffer_head *bh)
371 int nvecs = bio_get_nr_vecs(bh->b_bdev);
374 bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
377 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
378 bio->bi_bdev = 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,
388 struct buffer_head *bh)
392 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
396 if (io->io_bio == NULL) {
397 ret = io_submit_init_bio(io, bh);
401 ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
402 if (ret != bh->b_size)
403 goto submit_and_retry;
408 int ext4_bio_write_page(struct ext4_io_submit *io,
411 struct writeback_control *wbc)
413 struct inode *inode = page->mapping->host;
414 unsigned block_start, blocksize;
415 struct buffer_head *bh, *head;
417 int nr_submitted = 0;
419 blocksize = 1 << inode->i_blkbits;
421 BUG_ON(!PageLocked(page));
422 BUG_ON(PageWriteback(page));
424 set_page_writeback(page);
425 ClearPageError(page);
428 * In the first loop we prepare and mark buffers to submit. We have to
429 * mark all buffers in the page before submitting so that
430 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
431 * on the first buffer finishes and we are still working on submitting
434 bh = head = page_buffers(page);
436 block_start = bh_offset(bh);
437 if (block_start >= len) {
439 * Comments copied from block_write_full_page_endio:
441 * The page straddles i_size. It must be zeroed out on
442 * each and every writepage invocation because it may
443 * be mmapped. "A file is mapped in multiples of the
444 * page size. For a file that is not a multiple of
445 * the page size, the remaining memory is zeroed when
446 * mapped, and writes to that region are not written
449 zero_user_segment(page, block_start,
450 block_start + blocksize);
451 clear_buffer_dirty(bh);
452 set_buffer_uptodate(bh);
455 if (!buffer_dirty(bh) || buffer_delay(bh) ||
456 !buffer_mapped(bh) || buffer_unwritten(bh)) {
457 /* A hole? We can safely clear the dirty bit */
458 if (!buffer_mapped(bh))
459 clear_buffer_dirty(bh);
464 if (buffer_new(bh)) {
465 clear_buffer_new(bh);
466 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
468 set_buffer_async_write(bh);
469 } while ((bh = bh->b_this_page) != head);
471 /* Now submit buffers to write */
472 bh = head = page_buffers(page);
474 if (!buffer_async_write(bh))
476 ret = io_submit_add_bh(io, inode, bh);
479 * We only get here on ENOMEM. Not much else
480 * we can do but mark the page as dirty, and
481 * better luck next time.
483 redirty_page_for_writepage(wbc, page);
487 clear_buffer_dirty(bh);
488 } while ((bh = bh->b_this_page) != head);
490 /* Error stopped previous loop? Clean up buffers... */
493 clear_buffer_async_write(bh);
494 bh = bh->b_this_page;
495 } while (bh != head);
498 /* Nothing submitted - we have to end page writeback */
500 end_page_writeback(page);