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_unwritten_work))
62 cancel_work_sync(&EXT4_I(inode)->i_unwritten_work);
65 static void ext4_release_io_end(ext4_io_end_t *io_end)
67 BUG_ON(!list_empty(&io_end->list));
68 BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN);
70 if (atomic_dec_and_test(&EXT4_I(io_end->inode)->i_ioend_count))
71 wake_up_all(ext4_ioend_wq(io_end->inode));
72 if (io_end->flag & EXT4_IO_END_DIRECT)
73 inode_dio_done(io_end->inode);
75 aio_complete(io_end->iocb, io_end->result, 0);
76 kmem_cache_free(io_end_cachep, io_end);
79 static void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end)
81 struct inode *inode = io_end->inode;
83 io_end->flag &= ~EXT4_IO_END_UNWRITTEN;
84 /* Wake up anyone waiting on unwritten extent conversion */
85 if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten))
86 wake_up_all(ext4_ioend_wq(inode));
89 /* check a range of space and convert unwritten extents to written. */
90 static int ext4_end_io(ext4_io_end_t *io)
92 struct inode *inode = io->inode;
93 loff_t offset = io->offset;
94 ssize_t size = io->size;
97 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
99 io, inode->i_ino, io->list.next, io->list.prev);
101 ret = ext4_convert_unwritten_extents(inode, offset, size);
103 ext4_msg(inode->i_sb, KERN_EMERG,
104 "failed to convert unwritten extents to written "
105 "extents -- potential data loss! "
106 "(inode %lu, offset %llu, size %zd, error %d)",
107 inode->i_ino, offset, size, ret);
109 ext4_clear_io_unwritten_flag(io);
110 ext4_release_io_end(io);
114 static void dump_completed_IO(struct inode *inode)
117 struct list_head *cur, *before, *after;
118 ext4_io_end_t *io, *io0, *io1;
120 if (list_empty(&EXT4_I(inode)->i_completed_io_list)) {
121 ext4_debug("inode %lu completed_io list is empty\n",
126 ext4_debug("Dump inode %lu completed_io list\n", inode->i_ino);
127 list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list) {
130 io0 = container_of(before, ext4_io_end_t, list);
132 io1 = container_of(after, ext4_io_end_t, list);
134 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
135 io, inode->i_ino, io0, io1);
140 /* Add the io_end to per-inode completed end_io list. */
141 static void ext4_add_complete_io(ext4_io_end_t *io_end)
143 struct ext4_inode_info *ei = EXT4_I(io_end->inode);
144 struct workqueue_struct *wq;
147 BUG_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN));
148 wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq;
150 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
151 if (list_empty(&ei->i_completed_io_list))
152 queue_work(wq, &ei->i_unwritten_work);
153 list_add_tail(&io_end->list, &ei->i_completed_io_list);
154 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
157 static int ext4_do_flush_completed_IO(struct inode *inode)
160 struct list_head unwritten;
162 struct ext4_inode_info *ei = EXT4_I(inode);
165 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
166 dump_completed_IO(inode);
167 list_replace_init(&ei->i_completed_io_list, &unwritten);
168 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
170 while (!list_empty(&unwritten)) {
171 io = list_entry(unwritten.next, ext4_io_end_t, list);
172 BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN));
173 list_del_init(&io->list);
175 err = ext4_end_io(io);
176 if (unlikely(!ret && err))
183 * work on completed aio dio IO, to convert unwritten extents to extents
185 void ext4_end_io_work(struct work_struct *work)
187 struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info,
189 ext4_do_flush_completed_IO(&ei->vfs_inode);
192 int ext4_flush_unwritten_io(struct inode *inode)
195 WARN_ON_ONCE(!mutex_is_locked(&inode->i_mutex) &&
196 !(inode->i_state & I_FREEING));
197 ret = ext4_do_flush_completed_IO(inode);
198 ext4_unwritten_wait(inode);
202 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
204 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
206 atomic_inc(&EXT4_I(inode)->i_ioend_count);
208 INIT_LIST_HEAD(&io->list);
209 atomic_set(&io->count, 1);
214 void ext4_put_io_end_defer(ext4_io_end_t *io_end)
216 if (atomic_dec_and_test(&io_end->count)) {
217 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) {
218 ext4_release_io_end(io_end);
221 ext4_add_complete_io(io_end);
225 int ext4_put_io_end(ext4_io_end_t *io_end)
229 if (atomic_dec_and_test(&io_end->count)) {
230 if (io_end->flag & EXT4_IO_END_UNWRITTEN) {
231 err = ext4_convert_unwritten_extents(io_end->inode,
232 io_end->offset, io_end->size);
233 ext4_clear_io_unwritten_flag(io_end);
235 ext4_release_io_end(io_end);
240 ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end)
242 atomic_inc(&io_end->count);
247 * Print an buffer I/O error compatible with the fs/buffer.c. This
248 * provides compatibility with dmesg scrapers that look for a specific
249 * buffer I/O error message. We really need a unified error reporting
250 * structure to userspace ala Digital Unix's uerf system, but it's
251 * probably not going to happen in my lifetime, due to LKML politics...
253 static void buffer_io_error(struct buffer_head *bh)
255 char b[BDEVNAME_SIZE];
256 printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
257 bdevname(bh->b_bdev, b),
258 (unsigned long long)bh->b_blocknr);
261 static void ext4_end_bio(struct bio *bio, int error)
263 ext4_io_end_t *io_end = bio->bi_private;
267 sector_t bi_sector = bio->bi_sector;
270 inode = io_end->inode;
271 blocksize = 1 << inode->i_blkbits;
272 bio->bi_private = NULL;
273 bio->bi_end_io = NULL;
274 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
276 for (i = 0; i < bio->bi_vcnt; i++) {
277 struct bio_vec *bvec = &bio->bi_io_vec[i];
278 struct page *page = bvec->bv_page;
279 struct buffer_head *bh, *head;
280 unsigned bio_start = bvec->bv_offset;
281 unsigned bio_end = bio_start + bvec->bv_len;
282 unsigned under_io = 0;
290 set_bit(AS_EIO, &page->mapping->flags);
292 bh = head = page_buffers(page);
294 * We check all buffers in the page under BH_Uptodate_Lock
295 * to avoid races with other end io clearing async_write flags
297 local_irq_save(flags);
298 bit_spin_lock(BH_Uptodate_Lock, &head->b_state);
300 if (bh_offset(bh) < bio_start ||
301 bh_offset(bh) + blocksize > bio_end) {
302 if (buffer_async_write(bh))
306 clear_buffer_async_write(bh);
309 } while ((bh = bh->b_this_page) != head);
310 bit_spin_unlock(BH_Uptodate_Lock, &head->b_state);
311 local_irq_restore(flags);
313 end_page_writeback(page);
318 io_end->flag |= EXT4_IO_END_ERROR;
319 ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
320 "(offset %llu size %ld starting block %llu)",
322 (unsigned long long) io_end->offset,
325 bi_sector >> (inode->i_blkbits - 9));
328 ext4_put_io_end_defer(io_end);
331 void ext4_io_submit(struct ext4_io_submit *io)
333 struct bio *bio = io->io_bio;
337 submit_bio(io->io_op, io->io_bio);
338 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
344 void ext4_io_submit_init(struct ext4_io_submit *io,
345 struct writeback_control *wbc)
347 io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
352 static int io_submit_init_bio(struct ext4_io_submit *io,
353 struct buffer_head *bh)
355 int nvecs = bio_get_nr_vecs(bh->b_bdev);
358 bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
359 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
360 bio->bi_bdev = bh->b_bdev;
361 bio->bi_end_io = ext4_end_bio;
362 bio->bi_private = ext4_get_io_end(io->io_end);
363 if (!io->io_end->size)
364 io->io_end->offset = (bh->b_page->index << PAGE_CACHE_SHIFT)
367 io->io_next_block = bh->b_blocknr;
371 static int io_submit_add_bh(struct ext4_io_submit *io,
373 struct buffer_head *bh)
375 ext4_io_end_t *io_end;
378 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
382 if (io->io_bio == NULL) {
383 ret = io_submit_init_bio(io, bh);
387 ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
388 if (ret != bh->b_size)
389 goto submit_and_retry;
391 if (test_clear_buffer_uninit(bh))
392 ext4_set_io_unwritten_flag(inode, io_end);
393 io_end->size += bh->b_size;
398 int ext4_bio_write_page(struct ext4_io_submit *io,
401 struct writeback_control *wbc)
403 struct inode *inode = page->mapping->host;
404 unsigned block_start, blocksize;
405 struct buffer_head *bh, *head;
407 int nr_submitted = 0;
409 blocksize = 1 << inode->i_blkbits;
411 BUG_ON(!PageLocked(page));
412 BUG_ON(PageWriteback(page));
414 set_page_writeback(page);
415 ClearPageError(page);
418 * In the first loop we prepare and mark buffers to submit. We have to
419 * mark all buffers in the page before submitting so that
420 * end_page_writeback() cannot be called from ext4_bio_end_io() when IO
421 * on the first buffer finishes and we are still working on submitting
424 bh = head = page_buffers(page);
426 block_start = bh_offset(bh);
427 if (block_start >= len) {
429 * Comments copied from block_write_full_page_endio:
431 * The page straddles i_size. It must be zeroed out on
432 * each and every writepage invocation because it may
433 * be mmapped. "A file is mapped in multiples of the
434 * page size. For a file that is not a multiple of
435 * the page size, the remaining memory is zeroed when
436 * mapped, and writes to that region are not written
439 zero_user_segment(page, block_start,
440 block_start + blocksize);
441 clear_buffer_dirty(bh);
442 set_buffer_uptodate(bh);
445 if (!buffer_dirty(bh) || buffer_delay(bh) ||
446 !buffer_mapped(bh) || buffer_unwritten(bh)) {
447 /* A hole? We can safely clear the dirty bit */
448 if (!buffer_mapped(bh))
449 clear_buffer_dirty(bh);
454 if (buffer_new(bh)) {
455 clear_buffer_new(bh);
456 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
458 set_buffer_async_write(bh);
459 } while ((bh = bh->b_this_page) != head);
461 /* Now submit buffers to write */
462 bh = head = page_buffers(page);
464 if (!buffer_async_write(bh))
466 ret = io_submit_add_bh(io, inode, bh);
469 * We only get here on ENOMEM. Not much else
470 * we can do but mark the page as dirty, and
471 * better luck next time.
473 redirty_page_for_writepage(wbc, page);
477 clear_buffer_dirty(bh);
478 } while ((bh = bh->b_this_page) != head);
480 /* Error stopped previous loop? Clean up buffers... */
483 clear_buffer_async_write(bh);
484 bh = bh->b_this_page;
485 } while (bh != head);
488 /* Nothing submitted - we have to end page writeback */
490 end_page_writeback(page);