blob: 97e5e98fd42a291f79c4c314b21c5b8427ac3e63 [file] [log] [blame]
Theodore Ts'obd2d0212010-10-27 21:30:10 -04001/*
2 * linux/fs/ext4/page-io.c
3 *
4 * This contains the new page_io functions for ext4
5 *
6 * Written by Theodore Ts'o, 2010.
7 */
8
9#include <linux/module.h>
10#include <linux/fs.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>
27
28#include "ext4_jbd2.h"
29#include "xattr.h"
30#include "acl.h"
31#include "ext4_extents.h"
32
33static struct kmem_cache *io_page_cachep, *io_end_cachep;
34
Theodore Ts'o5dabfc72010-10-27 21:30:14 -040035int __init ext4_init_pageio(void)
Theodore Ts'obd2d0212010-10-27 21:30:10 -040036{
37 io_page_cachep = KMEM_CACHE(ext4_io_page, SLAB_RECLAIM_ACCOUNT);
38 if (io_page_cachep == NULL)
39 return -ENOMEM;
40 io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT);
Dan Carpenter13195182011-01-10 12:10:44 -050041 if (io_end_cachep == NULL) {
Theodore Ts'obd2d0212010-10-27 21:30:10 -040042 kmem_cache_destroy(io_page_cachep);
43 return -ENOMEM;
44 }
Theodore Ts'obd2d0212010-10-27 21:30:10 -040045 return 0;
46}
47
Theodore Ts'o5dabfc72010-10-27 21:30:14 -040048void ext4_exit_pageio(void)
Theodore Ts'obd2d0212010-10-27 21:30:10 -040049{
50 kmem_cache_destroy(io_end_cachep);
51 kmem_cache_destroy(io_page_cachep);
52}
53
Theodore Ts'of7ad6d22010-11-08 13:43:33 -050054void ext4_ioend_wait(struct inode *inode)
55{
Eric Sandeene9e3bce2011-02-12 08:17:34 -050056 wait_queue_head_t *wq = ext4_ioend_wq(inode);
Theodore Ts'of7ad6d22010-11-08 13:43:33 -050057
58 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0));
59}
60
Theodore Ts'o83668e72010-11-08 13:45:33 -050061static void put_io_page(struct ext4_io_page *io_page)
62{
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);
67 }
68}
69
Theodore Ts'obd2d0212010-10-27 21:30:10 -040070void ext4_free_io_end(ext4_io_end_t *io)
71{
72 int i;
Theodore Ts'of7ad6d22010-11-08 13:43:33 -050073 wait_queue_head_t *wq;
Theodore Ts'obd2d0212010-10-27 21:30:10 -040074
75 BUG_ON(!io);
76 if (io->page)
77 put_page(io->page);
Theodore Ts'o83668e72010-11-08 13:45:33 -050078 for (i = 0; i < io->num_io_pages; i++)
79 put_io_page(io->pages[i]);
Theodore Ts'obd2d0212010-10-27 21:30:10 -040080 io->num_io_pages = 0;
Eric Sandeene9e3bce2011-02-12 08:17:34 -050081 wq = ext4_ioend_wq(io->inode);
Theodore Ts'of7ad6d22010-11-08 13:43:33 -050082 if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count) &&
83 waitqueue_active(wq))
84 wake_up_all(wq);
Theodore Ts'obd2d0212010-10-27 21:30:10 -040085 kmem_cache_free(io_end_cachep, io);
86}
87
88/*
89 * check a range of space and convert unwritten extents to written.
90 */
91int ext4_end_io_nolock(ext4_io_end_t *io)
92{
93 struct inode *inode = io->inode;
94 loff_t offset = io->offset;
95 ssize_t size = io->size;
Eric Sandeene9e3bce2011-02-12 08:17:34 -050096 wait_queue_head_t *wq;
Theodore Ts'obd2d0212010-10-27 21:30:10 -040097 int ret = 0;
98
99 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
100 "list->prev 0x%p\n",
101 io, inode->i_ino, io->list.next, io->list.prev);
102
103 if (list_empty(&io->list))
104 return ret;
105
106 if (!(io->flag & EXT4_IO_END_UNWRITTEN))
107 return ret;
108
109 ret = ext4_convert_unwritten_extents(inode, offset, size);
110 if (ret < 0) {
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);
115 return ret;
116 }
117
118 if (io->iocb)
119 aio_complete(io->iocb, io->result, 0);
120 /* clear the DIO AIO unwritten flag */
Eric Sandeene9e3bce2011-02-12 08:17:34 -0500121 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)) {
127 wake_up_all(wq);
128 }
129 }
130
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400131 return ret;
132}
133
134/*
135 * work on completed aio dio IO, to convert unwritten extents to extents
136 */
137static void ext4_end_io_work(struct work_struct *work)
138{
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);
142 unsigned long flags;
143 int ret;
144
145 mutex_lock(&inode->i_mutex);
146 ret = ext4_end_io_nolock(io);
147 if (ret < 0) {
148 mutex_unlock(&inode->i_mutex);
149 return;
150 }
151
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);
158}
159
160ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
161{
Jesper Juhlb17b35e2010-12-19 21:41:55 -0500162 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400163 if (io) {
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500164 atomic_inc(&EXT4_I(inode)->i_ioend_count);
165 io->inode = inode;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400166 INIT_WORK(&io->work, ext4_end_io_work);
167 INIT_LIST_HEAD(&io->list);
168 }
169 return io;
170}
171
172/*
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...
178 */
179static void buffer_io_error(struct buffer_head *bh)
180{
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);
185}
186
187static void ext4_end_bio(struct bio *bio, int error)
188{
189 ext4_io_end_t *io_end = bio->bi_private;
190 struct workqueue_struct *wq;
191 struct inode *inode;
192 unsigned long flags;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400193 int i;
Curt Wohlgemuthd50bdd52011-02-07 12:46:14 -0500194 sector_t bi_sector = bio->bi_sector;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400195
196 BUG_ON(!io_end);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400197 bio->bi_private = NULL;
198 bio->bi_end_io = NULL;
199 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
200 error = 0;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400201 bio_put(bio);
202
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400203 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;
Curt Wohlgemuth39db00f2011-04-30 13:26:26 -0400206 loff_t offset;
207 loff_t io_end_offset;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400208
Curt Wohlgemuth39db00f2011-04-30 13:26:26 -0400209 if (error) {
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400210 SetPageError(page);
Curt Wohlgemuth39db00f2011-04-30 13:26:26 -0400211 set_bit(AS_EIO, &page->mapping->flags);
212 head = page_buffers(page);
213 BUG_ON(!head);
214
215 io_end_offset = io_end->offset + io_end->size;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400216
217 offset = (sector_t) page->index << PAGE_CACHE_SHIFT;
218 bh = head;
219 do {
220 if ((offset >= io_end->offset) &&
Curt Wohlgemuth39db00f2011-04-30 13:26:26 -0400221 (offset+bh->b_size <= io_end_offset))
222 buffer_io_error(bh);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400223
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400224 offset += bh->b_size;
225 bh = bh->b_this_page;
226 } while (bh != head);
227 }
228
Markus Trippelsdorf08da1192010-11-17 21:46:06 -0500229 put_io_page(io_end->pages[i]);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400230 }
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400231 io_end->num_io_pages = 0;
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500232 inode = io_end->inode;
233
234 if (error) {
235 io_end->flag |= EXT4_IO_END_ERROR;
236 ext4_warning(inode->i_sb, "I/O error writing to inode %lu "
237 "(offset %llu size %ld starting block %llu)",
238 inode->i_ino,
239 (unsigned long long) io_end->offset,
240 (long) io_end->size,
241 (unsigned long long)
Curt Wohlgemuthd50bdd52011-02-07 12:46:14 -0500242 bi_sector >> (inode->i_blkbits - 9));
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500243 }
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400244
Theodore Ts'ob6168442011-02-28 13:12:38 -0500245 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
246 ext4_free_io_end(io_end);
247 return;
248 }
249
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400250 /* Add the io_end to per-inode completed io list*/
251 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
252 list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list);
253 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
254
255 wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq;
256 /* queue the work to convert unwritten extents to written */
257 queue_work(wq, &io_end->work);
258}
259
260void ext4_io_submit(struct ext4_io_submit *io)
261{
262 struct bio *bio = io->io_bio;
263
264 if (bio) {
265 bio_get(io->io_bio);
266 submit_bio(io->io_op, io->io_bio);
267 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
268 bio_put(io->io_bio);
269 }
Peter Huewe7dc57612011-02-21 21:01:42 -0500270 io->io_bio = NULL;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400271 io->io_op = 0;
Peter Huewe7dc57612011-02-21 21:01:42 -0500272 io->io_end = NULL;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400273}
274
275static int io_submit_init(struct ext4_io_submit *io,
276 struct inode *inode,
277 struct writeback_control *wbc,
278 struct buffer_head *bh)
279{
280 ext4_io_end_t *io_end;
281 struct page *page = bh->b_page;
282 int nvecs = bio_get_nr_vecs(bh->b_bdev);
283 struct bio *bio;
284
285 io_end = ext4_init_io_end(inode, GFP_NOFS);
286 if (!io_end)
287 return -ENOMEM;
288 do {
289 bio = bio_alloc(GFP_NOIO, nvecs);
290 nvecs >>= 1;
291 } while (bio == NULL);
292
293 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
294 bio->bi_bdev = bh->b_bdev;
295 bio->bi_private = io->io_end = io_end;
296 bio->bi_end_io = ext4_end_bio;
297
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400298 io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh);
299
300 io->io_bio = bio;
Jens Axboe721a9602011-03-09 11:56:30 +0100301 io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400302 io->io_next_block = bh->b_blocknr;
303 return 0;
304}
305
306static int io_submit_add_bh(struct ext4_io_submit *io,
307 struct ext4_io_page *io_page,
308 struct inode *inode,
309 struct writeback_control *wbc,
310 struct buffer_head *bh)
311{
312 ext4_io_end_t *io_end;
313 int ret;
314
315 if (buffer_new(bh)) {
316 clear_buffer_new(bh);
317 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
318 }
319
320 if (!buffer_mapped(bh) || buffer_delay(bh)) {
321 if (!buffer_mapped(bh))
322 clear_buffer_dirty(bh);
323 if (io->io_bio)
324 ext4_io_submit(io);
325 return 0;
326 }
327
328 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
329submit_and_retry:
330 ext4_io_submit(io);
331 }
332 if (io->io_bio == NULL) {
333 ret = io_submit_init(io, inode, wbc, bh);
334 if (ret)
335 return ret;
336 }
337 io_end = io->io_end;
338 if ((io_end->num_io_pages >= MAX_IO_PAGES) &&
339 (io_end->pages[io_end->num_io_pages-1] != io_page))
340 goto submit_and_retry;
Tao Ma4eddd2a2011-08-13 12:30:59 -0400341 if (buffer_uninit(bh) && !(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
342 io_end->flag |= EXT4_IO_END_UNWRITTEN;
343 atomic_inc(&EXT4_I(inode)->i_aiodio_unwritten);
344 }
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400345 io->io_end->size += bh->b_size;
346 io->io_next_block++;
347 ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
348 if (ret != bh->b_size)
349 goto submit_and_retry;
350 if ((io_end->num_io_pages == 0) ||
351 (io_end->pages[io_end->num_io_pages-1] != io_page)) {
352 io_end->pages[io_end->num_io_pages++] = io_page;
Theodore Ts'o83668e72010-11-08 13:45:33 -0500353 atomic_inc(&io_page->p_count);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400354 }
355 return 0;
356}
357
358int ext4_bio_write_page(struct ext4_io_submit *io,
359 struct page *page,
360 int len,
361 struct writeback_control *wbc)
362{
363 struct inode *inode = page->mapping->host;
364 unsigned block_start, block_end, blocksize;
365 struct ext4_io_page *io_page;
366 struct buffer_head *bh, *head;
367 int ret = 0;
368
369 blocksize = 1 << inode->i_blkbits;
370
Curt Wohlgemuthd50bdd52011-02-07 12:46:14 -0500371 BUG_ON(!PageLocked(page));
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400372 BUG_ON(PageWriteback(page));
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400373
374 io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS);
375 if (!io_page) {
376 set_page_dirty(page);
377 unlock_page(page);
378 return -ENOMEM;
379 }
380 io_page->p_page = page;
Theodore Ts'o83668e72010-11-08 13:45:33 -0500381 atomic_set(&io_page->p_count, 1);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400382 get_page(page);
Theodore Ts'oa54aa762011-02-27 16:43:24 -0500383 set_page_writeback(page);
384 ClearPageError(page);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400385
386 for (bh = head = page_buffers(page), block_start = 0;
387 bh != head || !block_start;
388 block_start = block_end, bh = bh->b_this_page) {
Curt Wohlgemuthd50bdd52011-02-07 12:46:14 -0500389
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400390 block_end = block_start + blocksize;
391 if (block_start >= len) {
392 clear_buffer_dirty(bh);
393 set_buffer_uptodate(bh);
394 continue;
395 }
Curt Wohlgemuthd50bdd52011-02-07 12:46:14 -0500396 clear_buffer_dirty(bh);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400397 ret = io_submit_add_bh(io, io_page, inode, wbc, bh);
398 if (ret) {
399 /*
400 * We only get here on ENOMEM. Not much else
401 * we can do but mark the page as dirty, and
402 * better luck next time.
403 */
404 set_page_dirty(page);
405 break;
406 }
407 }
408 unlock_page(page);
409 /*
410 * If the page was truncated before we could do the writeback,
411 * or we had a memory allocation error while trying to write
412 * the first buffer head, we won't have submitted any pages for
413 * I/O. In that case we need to make sure we've cleared the
414 * PageWriteback bit from the page to prevent the system from
415 * wedging later on.
416 */
Theodore Ts'o83668e72010-11-08 13:45:33 -0500417 put_io_page(io_page);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400418 return ret;
419}