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 * This function is called by ext4_evict_inode() to make sure there is
50 * no more pending I/O completion work left to do.
52 void ext4_ioend_shutdown(struct inode *inode)
54 wait_queue_head_t *wq = ext4_ioend_wq(inode);
56 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0));
58 * We need to make sure the work structure is finished being
59 * used before we let the inode get destroyed.
61 if (work_pending(&EXT4_I(inode)->i_rsv_conversion_work))
62 cancel_work_sync(&EXT4_I(inode)->i_rsv_conversion_work);
63 if (work_pending(&EXT4_I(inode)->i_unrsv_conversion_work))
64 cancel_work_sync(&EXT4_I(inode)->i_unrsv_conversion_work);
68 * Print an buffer I/O error compatible with the fs/buffer.c. This
69 * provides compatibility with dmesg scrapers that look for a specific
70 * buffer I/O error message. We really need a unified error reporting
71 * structure to userspace ala Digital Unix's uerf system, but it's
72 * probably not going to happen in my lifetime, due to LKML politics...
74 static void buffer_io_error(struct buffer_head *bh)
76 char b[BDEVNAME_SIZE];
77 printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
78 bdevname(bh->b_bdev, b),
79 (unsigned long long)bh->b_blocknr);
82 static void ext4_finish_bio(struct bio *bio)
85 int error = !test_bit(BIO_UPTODATE, &bio->bi_flags);
87 for (i = 0; i < bio->bi_vcnt; i++) {
88 struct bio_vec *bvec = &bio->bi_io_vec[i];
89 struct page *page = bvec->bv_page;
90 struct buffer_head *bh, *head;
91 unsigned bio_start = bvec->bv_offset;
92 unsigned bio_end = bio_start + bvec->bv_len;
93 unsigned under_io = 0;
101 set_bit(AS_EIO, &page->mapping->flags);
103 bh = head = page_buffers(page);
105 * We check all buffers in the page under BH_Uptodate_Lock
106 * to avoid races with other end io clearing async_write flags
108 local_irq_save(flags);
109 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
111 if (bh_offset(bh) < bio_start ||
112 bh_offset(bh) + bh->b_size > bio_end) {
113 if (buffer_async_write(bh))
117 clear_buffer_async_write(bh);
120 } while ((bh = bh->b_this_page) != head);
121 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
122 local_irq_restore(flags);
124 end_page_writeback(page);
128 static void ext4_release_io_end(ext4_io_end_t *io_end)
130 struct bio *bio, *next_bio;
132 BUG_ON(!list_empty(&io_end->list));
133 BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
134 WARN_ON(io_end->handle);
136 if (atomic_dec_and_test(&EXT4_I(io_end->inode)->i_ioend_count))
137 wake_up_all(ext4_ioend_wq(io_end->inode));
139 for (bio = io_end->bio; bio; bio = next_bio) {
140 next_bio = bio->bi_private;
141 ext4_finish_bio(bio);
144 if (io_end->flag & EXT4_IO_END_DIRECT)
145 inode_dio_done(io_end->inode);
147 aio_complete(io_end->iocb, io_end->result, 0);
148 kmem_cache_free(io_end_cachep, io_end);
151 static void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end)
153 struct inode *inode = io_end->inode;
155 io_end->flag &= ~EXT4_IO_END_UNWRITTEN;
156 /* Wake up anyone waiting on unwritten extent conversion */
157 if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten))
158 wake_up_all(ext4_ioend_wq(inode));
162 * Check a range of space and convert unwritten extents to written. Note that
163 * we are protected from truncate touching same part of extent tree by the
164 * fact that truncate code waits for all DIO to finish (thus exclusion from
165 * direct IO is achieved) and also waits for PageWriteback bits. Thus we
166 * cannot get to ext4_ext_truncate() before all IOs overlapping that range are
167 * completed (happens from ext4_free_ioend()).
169 static int ext4_end_io(ext4_io_end_t *io)
171 struct inode *inode = io->inode;
172 loff_t offset = io->offset;
173 ssize_t size = io->size;
174 handle_t *handle = io->handle;
177 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
179 io, inode->i_ino, io->list.next, io->list.prev);
181 io->handle = NULL; /* Following call will use up the handle */
182 ret = ext4_convert_unwritten_extents(handle, inode, offset, size);
184 ext4_msg(inode->i_sb, KERN_EMERG,
185 "failed to convert unwritten extents to written "
186 "extents -- potential data loss! "
187 "(inode %lu, offset %llu, size %zd, error %d)",
188 inode->i_ino, offset, size, ret);
190 ext4_clear_io_unwritten_flag(io);
191 ext4_release_io_end(io);
195 static void dump_completed_IO(struct inode *inode, struct list_head *head)
198 struct list_head *cur, *before, *after;
199 ext4_io_end_t *io, *io0, *io1;
201 if (list_empty(head))
204 ext4_debug("Dump inode %lu completed io list\n", inode->i_ino);
205 list_for_each_entry(io, head, list) {
208 io0 = container_of(before, ext4_io_end_t, list);
210 io1 = container_of(after, ext4_io_end_t, list);
212 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
213 io, inode->i_ino, io0, io1);
218 /* Add the io_end to per-inode completed end_io list. */
219 static void ext4_add_complete_io(ext4_io_end_t *io_end)
221 struct ext4_inode_info *ei = EXT4_I(io_end->inode);
222 struct workqueue_struct *wq;
225 BUG_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
226 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
227 if (io_end->handle) {
228 wq = EXT4_SB(io_end->inode->i_sb)->rsv_conversion_wq;
229 if (list_empty(&ei->i_rsv_conversion_list))
230 queue_work(wq, &ei->i_rsv_conversion_work);
231 list_add_tail(&io_end->list, &ei->i_rsv_conversion_list);
233 wq = EXT4_SB(io_end->inode->i_sb)->unrsv_conversion_wq;
234 if (list_empty(&ei->i_unrsv_conversion_list))
235 queue_work(wq, &ei->i_unrsv_conversion_work);
236 list_add_tail(&io_end->list, &ei->i_unrsv_conversion_list);
238 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
241 static int ext4_do_flush_completed_IO(struct inode *inode,
242 struct list_head *head)
245 struct list_head unwritten;
247 struct ext4_inode_info *ei = EXT4_I(inode);
250 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
251 dump_completed_IO(inode, head);
252 list_replace_init(head, &unwritten);
253 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
255 while (!list_empty(&unwritten)) {
256 io = list_entry(unwritten.next, ext4_io_end_t, list);
257 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
258 list_del_init(&io->list);
260 err = ext4_end_io(io);
261 if (unlikely(!ret && err))
268 * work on completed IO, to convert unwritten extents to extents
270 void ext4_end_io_rsv_work(struct work_struct *work)
272 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
273 i_rsv_conversion_work);
274 ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list);
277 void ext4_end_io_unrsv_work(struct work_struct *work)
279 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
280 i_unrsv_conversion_work);
281 ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_unrsv_conversion_list);
284 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
286 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
288 atomic_inc(&EXT4_I(inode)->i_ioend_count);
290 INIT_LIST_HEAD(&io->list);
291 atomic_set(&io->count, 1);
296 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
298 if (atomic_dec_and_test(&io_end->count)) {
299 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
300 ext4_release_io_end(io_end);
303 ext4_add_complete_io(io_end);
307 int ext4_put_io_end(ext4_io_end_t *io_end)
311 if (atomic_dec_and_test(&io_end->count)) {
312 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
313 err = ext4_convert_unwritten_extents(io_end->handle,
314 io_end->inode, io_end->offset,
316 io_end->handle = NULL;
317 ext4_clear_io_unwritten_flag(io_end);
319 ext4_release_io_end(io_end);
324 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
326 atomic_inc(&io_end->count);
330 static void ext4_end_bio(struct bio *bio, int error)
332 ext4_io_end_t *io_end = bio->bi_private;
333 sector_t bi_sector = bio->bi_sector;
336 bio->bi_end_io = NULL;
337 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
340 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
342 * Link bio into list hanging from io_end. We have to do it
343 * atomically as bio completions can be racing against each
346 bio->bi_private = xchg(&io_end->bio, bio);
348 ext4_finish_bio(bio);
353 struct inode *inode = io_end->inode;
355 ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
356 "(offset %llu size %ld starting block %llu)",
358 (unsigned long long) io_end->offset,
361 bi_sector >> (inode->i_blkbits - 9));
363 ext4_put_io_end_defer(io_end);
366 void ext4_io_submit(struct ext4_io_submit *io)
368 struct bio *bio = io->io_bio;
372 submit_bio(io->io_op, io->io_bio);
373 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
379 void ext4_io_submit_init(struct ext4_io_submit *io,
380 struct writeback_control *wbc)
382 io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
387 static int io_submit_init_bio(struct ext4_io_submit *io,
388 struct buffer_head *bh)
390 int nvecs = bio_get_nr_vecs(bh->b_bdev);
393 bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
394 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
395 bio->bi_bdev = bh->b_bdev;
396 bio->bi_end_io = ext4_end_bio;
397 bio->bi_private = ext4_get_io_end(io->io_end);
399 io->io_next_block = bh->b_blocknr;
403 static int io_submit_add_bh(struct ext4_io_submit *io,
405 struct buffer_head *bh)
409 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
413 if (io->io_bio == NULL) {
414 ret = io_submit_init_bio(io, bh);
418 ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
419 if (ret != bh->b_size)
420 goto submit_and_retry;
425 int ext4_bio_write_page(struct ext4_io_submit *io,
428 struct writeback_control *wbc)
430 struct inode *inode = page->mapping->host;
431 unsigned block_start, blocksize;
432 struct buffer_head *bh, *head;
434 int nr_submitted = 0;
436 blocksize = 1 << inode->i_blkbits;
438 BUG_ON(!PageLocked(page));
439 BUG_ON(PageWriteback(page));
441 set_page_writeback(page);
442 ClearPageError(page);
445 * In the first loop we prepare and mark buffers to submit. We have to
446 * mark all buffers in the page before submitting so that
447 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
448 * on the first buffer finishes and we are still working on submitting
451 bh = head = page_buffers(page);
453 block_start = bh_offset(bh);
454 if (block_start >= len) {
456 * Comments copied from block_write_full_page_endio:
458 * The page straddles i_size. It must be zeroed out on
459 * each and every writepage invocation because it may
460 * be mmapped. "A file is mapped in multiples of the
461 * page size. For a file that is not a multiple of
462 * the page size, the remaining memory is zeroed when
463 * mapped, and writes to that region are not written
466 zero_user_segment(page, block_start,
467 block_start + blocksize);
468 clear_buffer_dirty(bh);
469 set_buffer_uptodate(bh);
472 if (!buffer_dirty(bh) || buffer_delay(bh) ||
473 !buffer_mapped(bh) || buffer_unwritten(bh)) {
474 /* A hole? We can safely clear the dirty bit */
475 if (!buffer_mapped(bh))
476 clear_buffer_dirty(bh);
481 if (buffer_new(bh)) {
482 clear_buffer_new(bh);
483 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
485 set_buffer_async_write(bh);
486 } while ((bh = bh->b_this_page) != head);
488 /* Now submit buffers to write */
489 bh = head = page_buffers(page);
491 if (!buffer_async_write(bh))
493 ret = io_submit_add_bh(io, inode, bh);
496 * We only get here on ENOMEM. Not much else
497 * we can do but mark the page as dirty, and
498 * better luck next time.
500 redirty_page_for_writepage(wbc, page);
504 clear_buffer_dirty(bh);
505 } while ((bh = bh->b_this_page) != head);
507 /* Error stopped previous loop? Clean up buffers... */
510 clear_buffer_async_write(bh);
511 bh = bh->b_this_page;
512 } while (bh != head);
515 /* Nothing submitted - we have to end page writeback */
517 end_page_writeback(page);