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;
36 #define to_ioend_wq(v) (&ioend_wq[((unsigned long)v) % WQ_HASH_SZ])
37 static wait_queue_head_t ioend_wq[WQ_HASH_SZ];
39 int __init ext4_init_pageio(void)
43 io_page_cachep = KMEM_CACHE(ext4_io_page, SLAB_RECLAIM_ACCOUNT);
44 if (io_page_cachep == NULL)
46 io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
47 if (io_page_cachep == NULL) {
48 kmem_cache_destroy(io_page_cachep);
51 for (i = 0; i < WQ_HASH_SZ; i++)
52 init_waitqueue_head(&ioend_wq[i]);
57 void ext4_exit_pageio(void)
59 kmem_cache_destroy(io_end_cachep);
60 kmem_cache_destroy(io_page_cachep);
63 void ext4_ioend_wait(struct inode *inode)
65 wait_queue_head_t *wq = to_ioend_wq(inode);
67 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0));
70 static void put_io_page(struct ext4_io_page *io_page)
72 if (atomic_dec_and_test(&io_page->p_count)) {
73 end_page_writeback(io_page->p_page);
74 put_page(io_page->p_page);
75 kmem_cache_free(io_page_cachep, io_page);
79 void ext4_free_io_end(ext4_io_end_t *io)
82 wait_queue_head_t *wq;
87 for (i = 0; i < io->num_io_pages; i++)
88 put_io_page(io->pages[i]);
90 wq = to_ioend_wq(io->inode);
91 if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count) &&
94 kmem_cache_free(io_end_cachep, io);
98 * check a range of space and convert unwritten extents to written.
100 int ext4_end_io_nolock(ext4_io_end_t *io)
102 struct inode *inode = io->inode;
103 loff_t offset = io->offset;
104 ssize_t size = io->size;
107 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
109 io, inode->i_ino, io->list.next, io->list.prev);
111 if (list_empty(&io->list))
114 if (!(io->flag & EXT4_IO_END_UNWRITTEN))
117 ret = ext4_convert_unwritten_extents(inode, offset, size);
119 printk(KERN_EMERG "%s: failed to convert unwritten "
120 "extents to written extents, error is %d "
121 "io is still on inode %lu aio dio list\n",
122 __func__, ret, inode->i_ino);
127 aio_complete(io->iocb, io->result, 0);
128 /* clear the DIO AIO unwritten flag */
129 io->flag &= ~EXT4_IO_END_UNWRITTEN;
134 * work on completed aio dio IO, to convert unwritten extents to extents
136 static void ext4_end_io_work(struct work_struct *work)
138 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
139 struct inode *inode = io->inode;
140 struct ext4_inode_info *ei = EXT4_I(inode);
144 mutex_lock(&inode->i_mutex);
145 ret = ext4_end_io_nolock(io);
147 mutex_unlock(&inode->i_mutex);
151 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
152 if (!list_empty(&io->list))
153 list_del_init(&io->list);
154 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
155 mutex_unlock(&inode->i_mutex);
156 ext4_free_io_end(io);
159 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
161 ext4_io_end_t *io = NULL;
163 io = kmem_cache_alloc(io_end_cachep, flags);
165 memset(io, 0, sizeof(*io));
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;
198 bio->bi_private = NULL;
199 bio->bi_end_io = NULL;
200 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
204 for (i = 0; i < io_end->num_io_pages; i++) {
205 struct page *page = io_end->pages[i]->p_page;
206 struct buffer_head *bh, *head;
207 int partial_write = 0;
209 head = page_buffers(page);
213 if (head->b_size == PAGE_CACHE_SIZE)
214 clear_buffer_dirty(head);
217 loff_t 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)) {
227 clear_buffer_dirty(bh);
229 if (buffer_delay(bh))
231 else if (!buffer_mapped(bh))
232 clear_buffer_dirty(bh);
233 else if (buffer_dirty(bh))
235 offset += bh->b_size;
236 bh = bh->b_this_page;
237 } while (bh != head);
241 * If this is a partial write which happened to make
242 * all buffers uptodate then we can optimize away a
243 * bogus readpage() for the next read(). Here we
244 * 'discover' whether the page went uptodate as a
245 * result of this (potentially partial) write.
248 SetPageUptodate(page);
250 put_io_page(io_end->pages[i]);
252 io_end->num_io_pages = 0;
253 inode = io_end->inode;
256 io_end->flag |= EXT4_IO_END_ERROR;
257 ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
258 "(offset %llu size %ld starting block %llu)",
260 (unsigned long long) io_end->offset,
263 bio->bi_sector >> (inode->i_blkbits - 9));
266 /* Add the io_end to per-inode completed io list*/
267 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
268 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
269 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
271 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
272 /* queue the work to convert unwritten extents to written */
273 queue_work(wq, &io_end->work);
276 void ext4_io_submit(struct ext4_io_submit *io)
278 struct bio *bio = io->io_bio;
282 submit_bio(io->io_op, io->io_bio);
283 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
291 static int io_submit_init(struct ext4_io_submit *io,
293 struct writeback_control *wbc,
294 struct buffer_head *bh)
296 ext4_io_end_t *io_end;
297 struct page *page = bh->b_page;
298 int nvecs = bio_get_nr_vecs(bh->b_bdev);
301 io_end = ext4_init_io_end(inode, GFP_NOFS);
305 bio = bio_alloc(GFP_NOIO, nvecs);
307 } while (bio == NULL);
309 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
310 bio->bi_bdev = bh->b_bdev;
311 bio->bi_private = io->io_end = io_end;
312 bio->bi_end_io = ext4_end_bio;
314 io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh);
317 io->io_op = (wbc->sync_mode == WB_SYNC_ALL ?
318 WRITE_SYNC_PLUG : WRITE);
319 io->io_next_block = bh->b_blocknr;
323 static int io_submit_add_bh(struct ext4_io_submit *io,
324 struct ext4_io_page *io_page,
326 struct writeback_control *wbc,
327 struct buffer_head *bh)
329 ext4_io_end_t *io_end;
332 if (buffer_new(bh)) {
333 clear_buffer_new(bh);
334 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
337 if (!buffer_mapped(bh) || buffer_delay(bh)) {
338 if (!buffer_mapped(bh))
339 clear_buffer_dirty(bh);
345 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
349 if (io->io_bio == NULL) {
350 ret = io_submit_init(io, inode, wbc, bh);
355 if ((io_end->num_io_pages >= MAX_IO_PAGES) &&
356 (io_end->pages[io_end->num_io_pages-1] != io_page))
357 goto submit_and_retry;
358 if (buffer_uninit(bh))
359 io->io_end->flag |= EXT4_IO_END_UNWRITTEN;
360 io->io_end->size += bh->b_size;
362 ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
363 if (ret != bh->b_size)
364 goto submit_and_retry;
365 if ((io_end->num_io_pages == 0) ||
366 (io_end->pages[io_end->num_io_pages-1] != io_page)) {
367 io_end->pages[io_end->num_io_pages++] = io_page;
368 atomic_inc(&io_page->p_count);
373 int ext4_bio_write_page(struct ext4_io_submit *io,
376 struct writeback_control *wbc)
378 struct inode *inode = page->mapping->host;
379 unsigned block_start, block_end, blocksize;
380 struct ext4_io_page *io_page;
381 struct buffer_head *bh, *head;
384 blocksize = 1 << inode->i_blkbits;
386 BUG_ON(PageWriteback(page));
387 set_page_writeback(page);
388 ClearPageError(page);
390 io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS);
392 set_page_dirty(page);
396 io_page->p_page = page;
397 atomic_set(&io_page->p_count, 1);
400 for (bh = head = page_buffers(page), block_start = 0;
401 bh != head || !block_start;
402 block_start = block_end, bh = bh->b_this_page) {
403 block_end = block_start + blocksize;
404 if (block_start >= len) {
405 clear_buffer_dirty(bh);
406 set_buffer_uptodate(bh);
409 ret = io_submit_add_bh(io, io_page, inode, wbc, bh);
412 * We only get here on ENOMEM. Not much else
413 * we can do but mark the page as dirty, and
414 * better luck next time.
416 set_page_dirty(page);
422 * If the page was truncated before we could do the writeback,
423 * or we had a memory allocation error while trying to write
424 * the first buffer head, we won't have submitted any pages for
425 * I/O. In that case we need to make sure we've cleared the
426 * PageWriteback bit from the page to prevent the system from
429 put_io_page(io_page);