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)
73 wait_queue_head_t *wq;
78 for (i = 0; i < io->num_io_pages; i++)
79 put_io_page(io->pages[i]);
81 wq = ext4_ioend_wq(io->inode);
82 if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count) &&
85 kmem_cache_free(io_end_cachep, io);
89 * check a range of space and convert unwritten extents to written.
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;
96 wait_queue_head_t *wq;
99 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
101 io, inode->i_ino, io->list.next, io->list.prev);
103 if (list_empty(&io->list))
106 if (!(io->flag & EXT4_IO_END_UNWRITTEN))
109 ret = ext4_convert_unwritten_extents(inode, offset, size);
111 printk(KERN_EMERG "%s: failed to convert unwritten "
112 "extents to written extents, error is %d "
113 "io is still on inode %lu aio dio list\n",
114 __func__, ret, inode->i_ino);
119 aio_complete(io->iocb, io->result, 0);
120 /* clear the DIO AIO unwritten flag */
121 if (io->flag & EXT4_IO_END_UNWRITTEN) {
122 io->flag &= ~EXT4_IO_END_UNWRITTEN;
123 /* Wake up anyone waiting on unwritten extent conversion */
124 wq = ext4_ioend_wq(io->inode);
125 if (atomic_dec_and_test(&EXT4_I(inode)->i_aiodio_unwritten) &&
126 waitqueue_active(wq)) {
135 * work on completed aio dio IO, to convert unwritten extents to extents
137 static void ext4_end_io_work(struct work_struct *work)
139 ext4_io_end_t *io = container_of(work, ext4_io_end_t, work);
140 struct inode *inode = io->inode;
141 struct ext4_inode_info *ei = EXT4_I(inode);
145 mutex_lock(&inode->i_mutex);
146 ret = ext4_end_io_nolock(io);
148 mutex_unlock(&inode->i_mutex);
152 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
153 if (!list_empty(&io->list))
154 list_del_init(&io->list);
155 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
156 mutex_unlock(&inode->i_mutex);
157 ext4_free_io_end(io);
160 ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
162 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
164 atomic_inc(&EXT4_I(inode)->i_ioend_count);
166 INIT_WORK(&io->work, ext4_end_io_work);
167 INIT_LIST_HEAD(&io->list);
173 * Print an buffer I/O error compatible with the fs/buffer.c. This
174 * provides compatibility with dmesg scrapers that look for a specific
175 * buffer I/O error message. We really need a unified error reporting
176 * structure to userspace ala Digital Unix's uerf system, but it's
177 * probably not going to happen in my lifetime, due to LKML politics...
179 static void buffer_io_error(struct buffer_head *bh)
181 char b[BDEVNAME_SIZE];
182 printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n",
183 bdevname(bh->b_bdev, b),
184 (unsigned long long)bh->b_blocknr);
187 static void ext4_end_bio(struct bio *bio, int error)
189 ext4_io_end_t *io_end = bio->bi_private;
190 struct workqueue_struct *wq;
194 sector_t bi_sector = bio->bi_sector;
197 bio->bi_private = NULL;
198 bio->bi_end_io = NULL;
199 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
203 for (i = 0; i < io_end->num_io_pages; i++) {
204 struct page *page = io_end->pages[i]->p_page;
205 struct buffer_head *bh, *head;
206 int partial_write = 0;
208 head = page_buffers(page);
212 if (head->b_size != PAGE_CACHE_SIZE) {
214 loff_t io_end_offset = io_end->offset + io_end->size;
216 offset = (sector_t) page->index << PAGE_CACHE_SHIFT;
219 if ((offset >= io_end->offset) &&
220 (offset+bh->b_size <= io_end_offset)) {
225 if (buffer_delay(bh))
227 else if (!buffer_mapped(bh))
228 clear_buffer_dirty(bh);
229 else if (buffer_dirty(bh))
231 offset += bh->b_size;
232 bh = bh->b_this_page;
233 } while (bh != head);
237 * If this is a partial write which happened to make
238 * all buffers uptodate then we can optimize away a
239 * bogus readpage() for the next read(). Here we
240 * 'discover' whether the page went uptodate as a
241 * result of this (potentially partial) write.
244 SetPageUptodate(page);
246 put_io_page(io_end->pages[i]);
248 io_end->num_io_pages = 0;
249 inode = io_end->inode;
252 io_end->flag |= EXT4_IO_END_ERROR;
253 ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
254 "(offset %llu size %ld starting block %llu)",
256 (unsigned long long) io_end->offset,
259 bi_sector >> (inode->i_blkbits - 9));
262 /* Add the io_end to per-inode completed io list*/
263 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
264 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
265 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
267 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
268 /* queue the work to convert unwritten extents to written */
269 queue_work(wq, &io_end->work);
272 void ext4_io_submit(struct ext4_io_submit *io)
274 struct bio *bio = io->io_bio;
278 submit_bio(io->io_op, io->io_bio);
279 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
287 static int io_submit_init(struct ext4_io_submit *io,
289 struct writeback_control *wbc,
290 struct buffer_head *bh)
292 ext4_io_end_t *io_end;
293 struct page *page = bh->b_page;
294 int nvecs = bio_get_nr_vecs(bh->b_bdev);
297 io_end = ext4_init_io_end(inode, GFP_NOFS);
301 bio = bio_alloc(GFP_NOIO, nvecs);
303 } while (bio == NULL);
305 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
306 bio->bi_bdev = bh->b_bdev;
307 bio->bi_private = io->io_end = io_end;
308 bio->bi_end_io = ext4_end_bio;
310 io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh);
313 io->io_op = (wbc->sync_mode == WB_SYNC_ALL ?
314 WRITE_SYNC_PLUG : WRITE);
315 io->io_next_block = bh->b_blocknr;
319 static int io_submit_add_bh(struct ext4_io_submit *io,
320 struct ext4_io_page *io_page,
322 struct writeback_control *wbc,
323 struct buffer_head *bh)
325 ext4_io_end_t *io_end;
328 if (buffer_new(bh)) {
329 clear_buffer_new(bh);
330 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
333 if (!buffer_mapped(bh) || buffer_delay(bh)) {
334 if (!buffer_mapped(bh))
335 clear_buffer_dirty(bh);
341 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
345 if (io->io_bio == NULL) {
346 ret = io_submit_init(io, inode, wbc, bh);
351 if ((io_end->num_io_pages >= MAX_IO_PAGES) &&
352 (io_end->pages[io_end->num_io_pages-1] != io_page))
353 goto submit_and_retry;
354 if (buffer_uninit(bh))
355 io->io_end->flag |= EXT4_IO_END_UNWRITTEN;
356 io->io_end->size += bh->b_size;
358 ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
359 if (ret != bh->b_size)
360 goto submit_and_retry;
361 if ((io_end->num_io_pages == 0) ||
362 (io_end->pages[io_end->num_io_pages-1] != io_page)) {
363 io_end->pages[io_end->num_io_pages++] = io_page;
364 atomic_inc(&io_page->p_count);
369 int ext4_bio_write_page(struct ext4_io_submit *io,
372 struct writeback_control *wbc)
374 struct inode *inode = page->mapping->host;
375 unsigned block_start, block_end, blocksize;
376 struct ext4_io_page *io_page;
377 struct buffer_head *bh, *head;
380 blocksize = 1 << inode->i_blkbits;
382 BUG_ON(!PageLocked(page));
383 BUG_ON(PageWriteback(page));
384 set_page_writeback(page);
385 ClearPageError(page);
387 io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS);
389 set_page_dirty(page);
393 io_page->p_page = page;
394 atomic_set(&io_page->p_count, 1);
397 for (bh = head = page_buffers(page), block_start = 0;
398 bh != head || !block_start;
399 block_start = block_end, bh = bh->b_this_page) {
401 block_end = block_start + blocksize;
402 if (block_start >= len) {
403 clear_buffer_dirty(bh);
404 set_buffer_uptodate(bh);
407 clear_buffer_dirty(bh);
408 ret = io_submit_add_bh(io, io_page, inode, wbc, bh);
411 * We only get here on ENOMEM. Not much else
412 * we can do but mark the page as dirty, and
413 * better luck next time.
415 set_page_dirty(page);
421 * If the page was truncated before we could do the writeback,
422 * or we had a memory allocation error while trying to write
423 * the first buffer head, we won't have submitted any pages for
424 * I/O. In that case we need to make sure we've cleared the
425 * PageWriteback bit from the page to prevent the system from
428 put_io_page(io_page);