blob: beacce11ac506ff3dd9958d1a3f82db095a36468 [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'of7ad6d22010-11-08 13:43:33 -050035#define WQ_HASH_SZ 37
36#define to_ioend_wq(v) (&ioend_wq[((unsigned long)v) % WQ_HASH_SZ])
37static wait_queue_head_t ioend_wq[WQ_HASH_SZ];
38
Theodore Ts'o5dabfc72010-10-27 21:30:14 -040039int __init ext4_init_pageio(void)
Theodore Ts'obd2d0212010-10-27 21:30:10 -040040{
Theodore Ts'of7ad6d22010-11-08 13:43:33 -050041 int i;
42
Theodore Ts'obd2d0212010-10-27 21:30:10 -040043 io_page_cachep = KMEM_CACHE(ext4_io_page, SLAB_RECLAIM_ACCOUNT);
44 if (io_page_cachep == NULL)
45 return -ENOMEM;
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);
49 return -ENOMEM;
50 }
Theodore Ts'of7ad6d22010-11-08 13:43:33 -050051 for (i = 0; i < WQ_HASH_SZ; i++)
52 init_waitqueue_head(&ioend_wq[i]);
Theodore Ts'obd2d0212010-10-27 21:30:10 -040053
54 return 0;
55}
56
Theodore Ts'o5dabfc72010-10-27 21:30:14 -040057void ext4_exit_pageio(void)
Theodore Ts'obd2d0212010-10-27 21:30:10 -040058{
59 kmem_cache_destroy(io_end_cachep);
60 kmem_cache_destroy(io_page_cachep);
61}
62
Theodore Ts'of7ad6d22010-11-08 13:43:33 -050063void ext4_ioend_wait(struct inode *inode)
64{
65 wait_queue_head_t *wq = to_ioend_wq(inode);
66
67 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0));
68}
69
Theodore Ts'o83668e72010-11-08 13:45:33 -050070static void put_io_page(struct ext4_io_page *io_page)
71{
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);
76 }
77}
78
Theodore Ts'obd2d0212010-10-27 21:30:10 -040079void ext4_free_io_end(ext4_io_end_t *io)
80{
81 int i;
Theodore Ts'of7ad6d22010-11-08 13:43:33 -050082 wait_queue_head_t *wq;
Theodore Ts'obd2d0212010-10-27 21:30:10 -040083
84 BUG_ON(!io);
85 if (io->page)
86 put_page(io->page);
Theodore Ts'o83668e72010-11-08 13:45:33 -050087 for (i = 0; i < io->num_io_pages; i++)
88 put_io_page(io->pages[i]);
Theodore Ts'obd2d0212010-10-27 21:30:10 -040089 io->num_io_pages = 0;
Theodore Ts'of7ad6d22010-11-08 13:43:33 -050090 wq = to_ioend_wq(io->inode);
91 if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count) &&
92 waitqueue_active(wq))
93 wake_up_all(wq);
Theodore Ts'obd2d0212010-10-27 21:30:10 -040094 kmem_cache_free(io_end_cachep, io);
95}
96
97/*
98 * check a range of space and convert unwritten extents to written.
99 */
100int ext4_end_io_nolock(ext4_io_end_t *io)
101{
102 struct inode *inode = io->inode;
103 loff_t offset = io->offset;
104 ssize_t size = io->size;
105 int ret = 0;
106
107 ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p,"
108 "list->prev 0x%p\n",
109 io, inode->i_ino, io->list.next, io->list.prev);
110
111 if (list_empty(&io->list))
112 return ret;
113
114 if (!(io->flag & EXT4_IO_END_UNWRITTEN))
115 return ret;
116
117 ret = ext4_convert_unwritten_extents(inode, offset, size);
118 if (ret < 0) {
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);
123 return ret;
124 }
125
126 if (io->iocb)
127 aio_complete(io->iocb, io->result, 0);
128 /* clear the DIO AIO unwritten flag */
129 io->flag &= ~EXT4_IO_END_UNWRITTEN;
130 return ret;
131}
132
133/*
134 * work on completed aio dio IO, to convert unwritten extents to extents
135 */
136static void ext4_end_io_work(struct work_struct *work)
137{
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);
141 unsigned long flags;
142 int ret;
143
144 mutex_lock(&inode->i_mutex);
145 ret = ext4_end_io_nolock(io);
146 if (ret < 0) {
147 mutex_unlock(&inode->i_mutex);
148 return;
149 }
150
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);
157}
158
159ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags)
160{
161 ext4_io_end_t *io = NULL;
162
163 io = kmem_cache_alloc(io_end_cachep, flags);
164 if (io) {
165 memset(io, 0, sizeof(*io));
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500166 atomic_inc(&EXT4_I(inode)->i_ioend_count);
167 io->inode = inode;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400168 INIT_WORK(&io->work, ext4_end_io_work);
169 INIT_LIST_HEAD(&io->list);
170 }
171 return io;
172}
173
174/*
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...
180 */
181static void buffer_io_error(struct buffer_head *bh)
182{
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);
187}
188
189static void ext4_end_bio(struct bio *bio, int error)
190{
191 ext4_io_end_t *io_end = bio->bi_private;
192 struct workqueue_struct *wq;
193 struct inode *inode;
194 unsigned long flags;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400195 int i;
196
197 BUG_ON(!io_end);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400198 bio->bi_private = NULL;
199 bio->bi_end_io = NULL;
200 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
201 error = 0;
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400202 bio_put(bio);
203
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400204 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;
208
209 head = page_buffers(page);
210 if (error)
211 SetPageError(page);
212 BUG_ON(!head);
213 if (head->b_size == PAGE_CACHE_SIZE)
214 clear_buffer_dirty(head);
215 else {
216 loff_t offset;
217 loff_t io_end_offset = io_end->offset + io_end->size;
218
219 offset = (sector_t) page->index << PAGE_CACHE_SHIFT;
220 bh = head;
221 do {
222 if ((offset >= io_end->offset) &&
223 (offset+bh->b_size <= io_end_offset)) {
224 if (error)
225 buffer_io_error(bh);
226
227 clear_buffer_dirty(bh);
228 }
229 if (buffer_delay(bh))
230 partial_write = 1;
231 else if (!buffer_mapped(bh))
232 clear_buffer_dirty(bh);
233 else if (buffer_dirty(bh))
234 partial_write = 1;
235 offset += bh->b_size;
236 bh = bh->b_this_page;
237 } while (bh != head);
238 }
239
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400240 /*
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.
246 */
247 if (!partial_write)
248 SetPageUptodate(page);
Markus Trippelsdorf08da1192010-11-17 21:46:06 -0500249
250 put_io_page(io_end->pages[i]);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400251 }
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400252 io_end->num_io_pages = 0;
Theodore Ts'of7ad6d22010-11-08 13:43:33 -0500253 inode = io_end->inode;
254
255 if (error) {
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)",
259 inode->i_ino,
260 (unsigned long long) io_end->offset,
261 (long) io_end->size,
262 (unsigned long long)
263 bio->bi_sector >> (inode->i_blkbits - 9));
264 }
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400265
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);
270
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);
274}
275
276void ext4_io_submit(struct ext4_io_submit *io)
277{
278 struct bio *bio = io->io_bio;
279
280 if (bio) {
281 bio_get(io->io_bio);
282 submit_bio(io->io_op, io->io_bio);
283 BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP));
284 bio_put(io->io_bio);
285 }
286 io->io_bio = 0;
287 io->io_op = 0;
288 io->io_end = 0;
289}
290
291static int io_submit_init(struct ext4_io_submit *io,
292 struct inode *inode,
293 struct writeback_control *wbc,
294 struct buffer_head *bh)
295{
296 ext4_io_end_t *io_end;
297 struct page *page = bh->b_page;
298 int nvecs = bio_get_nr_vecs(bh->b_bdev);
299 struct bio *bio;
300
301 io_end = ext4_init_io_end(inode, GFP_NOFS);
302 if (!io_end)
303 return -ENOMEM;
304 do {
305 bio = bio_alloc(GFP_NOIO, nvecs);
306 nvecs >>= 1;
307 } while (bio == NULL);
308
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;
313
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400314 io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh);
315
316 io->io_bio = bio;
317 io->io_op = (wbc->sync_mode == WB_SYNC_ALL ?
318 WRITE_SYNC_PLUG : WRITE);
319 io->io_next_block = bh->b_blocknr;
320 return 0;
321}
322
323static int io_submit_add_bh(struct ext4_io_submit *io,
324 struct ext4_io_page *io_page,
325 struct inode *inode,
326 struct writeback_control *wbc,
327 struct buffer_head *bh)
328{
329 ext4_io_end_t *io_end;
330 int ret;
331
332 if (buffer_new(bh)) {
333 clear_buffer_new(bh);
334 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
335 }
336
337 if (!buffer_mapped(bh) || buffer_delay(bh)) {
338 if (!buffer_mapped(bh))
339 clear_buffer_dirty(bh);
340 if (io->io_bio)
341 ext4_io_submit(io);
342 return 0;
343 }
344
345 if (io->io_bio && bh->b_blocknr != io->io_next_block) {
346submit_and_retry:
347 ext4_io_submit(io);
348 }
349 if (io->io_bio == NULL) {
350 ret = io_submit_init(io, inode, wbc, bh);
351 if (ret)
352 return ret;
353 }
354 io_end = io->io_end;
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;
361 io->io_next_block++;
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;
Theodore Ts'o83668e72010-11-08 13:45:33 -0500368 atomic_inc(&io_page->p_count);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400369 }
370 return 0;
371}
372
373int ext4_bio_write_page(struct ext4_io_submit *io,
374 struct page *page,
375 int len,
376 struct writeback_control *wbc)
377{
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;
382 int ret = 0;
383
384 blocksize = 1 << inode->i_blkbits;
385
386 BUG_ON(PageWriteback(page));
387 set_page_writeback(page);
388 ClearPageError(page);
389
390 io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS);
391 if (!io_page) {
392 set_page_dirty(page);
393 unlock_page(page);
394 return -ENOMEM;
395 }
396 io_page->p_page = page;
Theodore Ts'o83668e72010-11-08 13:45:33 -0500397 atomic_set(&io_page->p_count, 1);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400398 get_page(page);
399
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);
407 continue;
408 }
409 ret = io_submit_add_bh(io, io_page, inode, wbc, bh);
410 if (ret) {
411 /*
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.
415 */
416 set_page_dirty(page);
417 break;
418 }
419 }
420 unlock_page(page);
421 /*
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
427 * wedging later on.
428 */
Theodore Ts'o83668e72010-11-08 13:45:33 -0500429 put_io_page(io_page);
Theodore Ts'obd2d0212010-10-27 21:30:10 -0400430 return ret;
431}