1
2
3
4
5
6
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
35int __init ext4_init_pageio(void)
36{
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);
41 if (io_end_cachep == NULL) {
42 kmem_cache_destroy(io_page_cachep);
43 return -ENOMEM;
44 }
45 return 0;
46}
47
48void ext4_exit_pageio(void)
49{
50 kmem_cache_destroy(io_end_cachep);
51 kmem_cache_destroy(io_page_cachep);
52}
53
54void ext4_ioend_wait(struct inode *inode)
55{
56 wait_queue_head_t *wq = ext4_ioend_wq(inode);
57
58 wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0));
59}
60
61static 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
70void ext4_free_io_end(ext4_io_end_t *io)
71{
72 int i;
73 wait_queue_head_t *wq;
74
75 BUG_ON(!io);
76 if (io->page)
77 put_page(io->page);
78 for (i = 0; i < io->num_io_pages; i++)
79 put_io_page(io->pages[i]);
80 io->num_io_pages = 0;
81 wq = ext4_ioend_wq(io->inode);
82 if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count) &&
83 waitqueue_active(wq))
84 wake_up_all(wq);
85 kmem_cache_free(io_end_cachep, io);
86}
87
88
89
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;
96 wait_queue_head_t *wq;
97 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
121 if (io->flag & EXT4_IO_END_UNWRITTEN) {
122 io->flag &= ~EXT4_IO_END_UNWRITTEN;
123
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
131 return ret;
132}
133
134
135
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{
162 ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags);
163 if (io) {
164 atomic_inc(&EXT4_I(inode)->i_ioend_count);
165 io->inode = inode;
166 INIT_WORK(&io->work, ext4_end_io_work);
167 INIT_LIST_HEAD(&io->list);
168 }
169 return io;
170}
171
172
173
174
175
176
177
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;
193 int i;
194 sector_t bi_sector = bio->bi_sector;
195
196 BUG_ON(!io_end);
197 bio->bi_private = NULL;
198 bio->bi_end_io = NULL;
199 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
200 error = 0;
201 bio_put(bio);
202
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 loff_t offset;
207 loff_t io_end_offset;
208
209 if (error) {
210 SetPageError(page);
211 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;
216
217 offset = (sector_t) page->index << PAGE_CACHE_SHIFT;
218 bh = head;
219 do {
220 if ((offset >= io_end->offset) &&
221 (offset+bh->b_size <= io_end_offset))
222 buffer_io_error(bh);
223
224 offset += bh->b_size;
225 bh = bh->b_this_page;
226 } while (bh != head);
227 }
228
229 put_io_page(io_end->pages[i]);
230 }
231 io_end->num_io_pages = 0;
232 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)
242 bi_sector >> (inode->i_blkbits - 9));
243 }
244
245 if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) {
246 ext4_free_io_end(io_end);
247 return;
248 }
249
250
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
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 }
270 io->io_bio = NULL;
271 io->io_op = 0;
272 io->io_end = NULL;
273}
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
298 io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh);
299
300 io->io_bio = bio;
301 io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
302 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;
341 if (buffer_uninit(bh))
342 io->io_end->flag |= EXT4_IO_END_UNWRITTEN;
343 io->io_end->size += bh->b_size;
344 io->io_next_block++;
345 ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh));
346 if (ret != bh->b_size)
347 goto submit_and_retry;
348 if ((io_end->num_io_pages == 0) ||
349 (io_end->pages[io_end->num_io_pages-1] != io_page)) {
350 io_end->pages[io_end->num_io_pages++] = io_page;
351 atomic_inc(&io_page->p_count);
352 }
353 return 0;
354}
355
356int ext4_bio_write_page(struct ext4_io_submit *io,
357 struct page *page,
358 int len,
359 struct writeback_control *wbc)
360{
361 struct inode *inode = page->mapping->host;
362 unsigned block_start, block_end, blocksize;
363 struct ext4_io_page *io_page;
364 struct buffer_head *bh, *head;
365 int ret = 0;
366
367 blocksize = 1 << inode->i_blkbits;
368
369 BUG_ON(!PageLocked(page));
370 BUG_ON(PageWriteback(page));
371
372 io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS);
373 if (!io_page) {
374 set_page_dirty(page);
375 unlock_page(page);
376 return -ENOMEM;
377 }
378 io_page->p_page = page;
379 atomic_set(&io_page->p_count, 1);
380 get_page(page);
381 set_page_writeback(page);
382 ClearPageError(page);
383
384 for (bh = head = page_buffers(page), block_start = 0;
385 bh != head || !block_start;
386 block_start = block_end, bh = bh->b_this_page) {
387
388 block_end = block_start + blocksize;
389 if (block_start >= len) {
390 clear_buffer_dirty(bh);
391 set_buffer_uptodate(bh);
392 continue;
393 }
394 clear_buffer_dirty(bh);
395 ret = io_submit_add_bh(io, io_page, inode, wbc, bh);
396 if (ret) {
397
398
399
400
401
402 set_page_dirty(page);
403 break;
404 }
405 }
406 unlock_page(page);
407
408
409
410
411
412
413
414
415 put_io_page(io_page);
416 return ret;
417}
418