2 * linux/fs/ext4/page-io.c
4 * This contains the new page_io functions for ext4
6 * Written by Theodore Ts'o, 2010.
9 #include <linux/module.h>
11 #include <linux/time.h>
12 #include <linux/jbd2.h>
13 #include <linux/highuid.h>
14 #include <linux/pagemap.h>
15 #include <linux/quotaops.h>
16 #include <linux/string.h>
17 #include <linux/buffer_head.h>
18 #include <linux/writeback.h>
19 #include <linux/pagevec.h>
20 #include <linux/mpage.h>
21 #include <linux/namei.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>
28 #include "ext4_jbd2.h"
31 #include "ext4_extents.h"
33 static struct kmem_cache *io_page_cachep, *io_end_cachep;
35 int __init ext4_init_pageio(void)
37 io_page_cachep = KMEM_CACHE(ext4_io_page, SLAB_RECLAIM_ACCOUNT);
38 if (io_page_cachep == NULL)
40 io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
41 if (io_end_cachep == NULL) {
42 kmem_cache_destroy(io_page_cachep);
48 void ext4_exit_pageio(void)
50 kmem_cache_destroy(io_end_cachep);
51 kmem_cache_destroy(io_page_cachep);
54 void ext4_ioend_wait(struct inode *inode)
56 wait_queue_head_t *wq = ext4_ioend_wq(inode);
58 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0));
61 static void put_io_page(struct ext4_io_page *io_page)
63 if (atomic_dec_and_test(&io_page->p_count)) {
64 end_page_writeback(io_page->p_page);
65 put_page(io_page->p_page);
66 kmem_cache_free(io_page_cachep, io_page);
70 void ext4_free_io_end(ext4_io_end_t *io)
77 for (i = 0; i < io->num_io_pages; i++)
78 put_io_page(io->pages[i]);
80 if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count))
81 wake_up_all(ext4_ioend_wq(io->inode));
82 kmem_cache_free(io_end_cachep, io);
86 * check a range of space and convert unwritten extents to written.
88 * Called with inode->i_mutex; we depend on this when we manipulate
89 * io->flag, since we could otherwise race with ext4_flush_completed_IO()
91 int ext4_end_io_nolock(ext4_io_end_t *io)
93 struct inode *inode = io->inode;
94 loff_t offset = io->offset;
95 ssize_t size = io->size;
98 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
100 io, inode->i_ino, io->list.next, io->list.prev);
102 ret = ext4_convert_unwritten_extents(inode, offset, size);
104 ext4_msg(inode->i_sb, KERN_EMERG,
105 "failed to convert unwritten extents to written "
106 "extents -- potential data loss! "
107 "(inode %lu, offset %llu, size %zd, error %d)",
108 inode->i_ino, offset, size, ret);
112 aio_complete(io->iocb, io->result, 0);
114 /* Wake up anyone waiting on unwritten extent conversion */
115 if (atomic_dec_and_test(&EXT4_I(inode)->i_aiodio_unwritten))
116 wake_up_all(ext4_ioend_wq(io->inode));
121 * work on completed aio dio IO, to convert unwritten extents to extents
123 static void ext4_end_io_work(struct work_struct *work)
125 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
126 struct inode *inode = io->inode;
127 struct ext4_inode_info *ei = EXT4_I(inode);
130 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
131 if (list_empty(&io->list)) {
132 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
136 if (!mutex_trylock(&inode->i_mutex)) {
137 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
139 * Requeue the work instead of waiting so that the work
140 * items queued after this can be processed.
142 queue_work(EXT4_SB(inode->i_sb)->dio_unwritten_wq, &io->work);
144 * To prevent the ext4-dio-unwritten thread from keeping
145 * requeueing end_io requests and occupying cpu for too long,
146 * yield the cpu if it sees an end_io request that has already
149 if (io->flag & EXT4_IO_END_QUEUED)
151 io->flag |= EXT4_IO_END_QUEUED;
154 list_del_init(&io->list);
155 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
156 (void) ext4_end_io_nolock(io);
157 mutex_unlock(&inode->i_mutex);
159 ext4_free_io_end(io);
162 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
164 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
166 atomic_inc(&EXT4_I(inode)->i_ioend_count);
168 INIT_WORK(&io->work, ext4_end_io_work);
169 INIT_LIST_HEAD(&io->list);
175 * Print an buffer I/O error compatible with the fs/buffer.c. This
176 * provides compatibility with dmesg scrapers that look for a specific
177 * buffer I/O error message. We really need a unified error reporting
178 * structure to userspace ala Digital Unix's uerf system, but it's
179 * probably not going to happen in my lifetime, due to LKML politics...
181 static void buffer_io_error(struct buffer_head *bh)
183 char b[BDEVNAME_SIZE];
184 printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
185 bdevname(bh->b_bdev, b),
186 (unsigned long long)bh->b_blocknr);
189 static void ext4_end_bio(struct bio *bio, int error)
191 ext4_io_end_t *io_end = bio->bi_private;
192 struct workqueue_struct *wq;
196 sector_t bi_sector = bio->bi_sector;
199 bio->bi_private = NULL;
200 bio->bi_end_io = NULL;
201 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
205 for (i = 0; i < io_end->num_io_pages; i++) {
206 struct page *page = io_end->pages[i]->p_page;
207 struct buffer_head *bh, *head;
209 loff_t io_end_offset;
213 set_bit(AS_EIO, &page->mapping->flags);
214 head = page_buffers(page);
217 io_end_offset = io_end->offset + io_end->size;
219 offset = (sector_t) page->index << PAGE_CACHE_SHIFT;
222 if ((offset >= io_end->offset) &&
223 (offset+bh->b_size <= io_end_offset))
226 offset += bh->b_size;
227 bh = bh->b_this_page;
228 } while (bh != head);
231 put_io_page(io_end->pages[i]);
233 io_end->num_io_pages = 0;
234 inode = io_end->inode;
237 io_end->flag |= EXT4_IO_END_ERROR;
238 ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
239 "(offset %llu size %ld starting block %llu)",
241 (unsigned long long) io_end->offset,
244 bi_sector >> (inode->i_blkbits - 9));
247 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
248 ext4_free_io_end(io_end);
252 /* Add the io_end to per-inode completed io list*/
253 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
254 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
255 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
257 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
258 /* queue the work to convert unwritten extents to written */
259 queue_work(wq, &io_end->work);
262 void ext4_io_submit(struct ext4_io_submit *io)
264 struct bio *bio = io->io_bio;
268 submit_bio(io->io_op, io->io_bio);
269 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
277 static int io_submit_init(struct ext4_io_submit *io,
279 struct writeback_control *wbc,
280 struct buffer_head *bh)
282 ext4_io_end_t *io_end;
283 struct page *page = bh->b_page;
284 int nvecs = bio_get_nr_vecs(bh->b_bdev);
287 io_end = ext4_init_io_end(inode, GFP_NOFS);
290 bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES));
291 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
292 bio->bi_bdev = bh->b_bdev;
293 bio->bi_private = io->io_end = io_end;
294 bio->bi_end_io = ext4_end_bio;
296 io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh);
299 io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
300 io->io_next_block = bh->b_blocknr;
304 static int io_submit_add_bh(struct ext4_io_submit *io,
305 struct ext4_io_page *io_page,
307 struct writeback_control *wbc,
308 struct buffer_head *bh)
310 ext4_io_end_t *io_end;
313 if (buffer_new(bh)) {
314 clear_buffer_new(bh);
315 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
318 if (!buffer_mapped(bh) || buffer_delay(bh)) {
319 if (!buffer_mapped(bh))
320 clear_buffer_dirty(bh);
326 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
330 if (io->io_bio == NULL) {
331 ret = io_submit_init(io, inode, wbc, bh);
336 if ((io_end->num_io_pages >= MAX_IO_PAGES) &&
337 (io_end->pages[io_end->num_io_pages-1] != io_page))
338 goto submit_and_retry;
339 if (buffer_uninit(bh))
340 ext4_set_io_unwritten_flag(inode, io_end);
341 io->io_end->size += bh->b_size;
343 ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
344 if (ret != bh->b_size)
345 goto submit_and_retry;
346 if ((io_end->num_io_pages == 0) ||
347 (io_end->pages[io_end->num_io_pages-1] != io_page)) {
348 io_end->pages[io_end->num_io_pages++] = io_page;
349 atomic_inc(&io_page->p_count);
354 int ext4_bio_write_page(struct ext4_io_submit *io,
357 struct writeback_control *wbc)
359 struct inode *inode = page->mapping->host;
360 unsigned block_start, block_end, blocksize;
361 struct ext4_io_page *io_page;
362 struct buffer_head *bh, *head;
365 blocksize = 1 << inode->i_blkbits;
367 BUG_ON(!PageLocked(page));
368 BUG_ON(PageWriteback(page));
370 io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS);
372 set_page_dirty(page);
376 io_page->p_page = page;
377 atomic_set(&io_page->p_count, 1);
379 set_page_writeback(page);
380 ClearPageError(page);
382 for (bh = head = page_buffers(page), block_start = 0;
383 bh != head || !block_start;
384 block_start = block_end, bh = bh->b_this_page) {
386 block_end = block_start + blocksize;
387 if (block_start >= len) {
389 * Comments copied from block_write_full_page_endio:
391 * The page straddles i_size. It must be zeroed out on
392 * each and every writepage invocation because it may
393 * be mmapped. "A file is mapped in multiples of the
394 * page size. For a file that is not a multiple of
395 * the page size, the remaining memory is zeroed when
396 * mapped, and writes to that region are not written
399 zero_user_segment(page, block_start, block_end);
400 clear_buffer_dirty(bh);
401 set_buffer_uptodate(bh);
404 clear_buffer_dirty(bh);
405 ret = io_submit_add_bh(io, io_page, inode, wbc, bh);
408 * We only get here on ENOMEM. Not much else
409 * we can do but mark the page as dirty, and
410 * better luck next time.
412 set_page_dirty(page);
418 * If the page was truncated before we could do the writeback,
419 * or we had a memory allocation error while trying to write
420 * the first buffer head, we won't have submitted any pages for
421 * I/O. In that case we need to make sure we've cleared the
422 * PageWriteback bit from the page to prevent the system from
425 put_io_page(io_page);