Commit | Line | Data |
---|---|---|
bd2d0210 TT |
1 | /* |
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 | ||
33 | static struct kmem_cache *io_page_cachep, *io_end_cachep; | |
34 | ||
5dabfc78 | 35 | int __init ext4_init_pageio(void) |
bd2d0210 TT |
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); | |
13195184 | 41 | if (io_end_cachep == NULL) { |
bd2d0210 TT |
42 | kmem_cache_destroy(io_page_cachep); |
43 | return -ENOMEM; | |
44 | } | |
bd2d0210 TT |
45 | return 0; |
46 | } | |
47 | ||
5dabfc78 | 48 | void ext4_exit_pageio(void) |
bd2d0210 TT |
49 | { |
50 | kmem_cache_destroy(io_end_cachep); | |
51 | kmem_cache_destroy(io_page_cachep); | |
52 | } | |
53 | ||
f7ad6d2e TT |
54 | void ext4_ioend_wait(struct inode *inode) |
55 | { | |
e9e3bcec | 56 | wait_queue_head_t *wq = ext4_ioend_wq(inode); |
f7ad6d2e TT |
57 | |
58 | wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0)); | |
59 | } | |
60 | ||
83668e71 TT |
61 | static 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 | ||
bd2d0210 TT |
70 | void ext4_free_io_end(ext4_io_end_t *io) |
71 | { | |
72 | int i; | |
f7ad6d2e | 73 | wait_queue_head_t *wq; |
bd2d0210 TT |
74 | |
75 | BUG_ON(!io); | |
76 | if (io->page) | |
77 | put_page(io->page); | |
83668e71 TT |
78 | for (i = 0; i < io->num_io_pages; i++) |
79 | put_io_page(io->pages[i]); | |
bd2d0210 | 80 | io->num_io_pages = 0; |
e9e3bcec | 81 | wq = ext4_ioend_wq(io->inode); |
f7ad6d2e TT |
82 | if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count) && |
83 | waitqueue_active(wq)) | |
84 | wake_up_all(wq); | |
bd2d0210 TT |
85 | kmem_cache_free(io_end_cachep, io); |
86 | } | |
87 | ||
88 | /* | |
89 | * check a range of space and convert unwritten extents to written. | |
90 | */ | |
91 | int 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; | |
e9e3bcec | 96 | wait_queue_head_t *wq; |
bd2d0210 TT |
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 | /* clear the DIO AIO unwritten flag */ | |
e9e3bcec ES |
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)) { | |
127 | wake_up_all(wq); | |
128 | } | |
129 | } | |
130 | ||
bd2d0210 TT |
131 | return ret; |
132 | } | |
133 | ||
134 | /* | |
135 | * work on completed aio dio IO, to convert unwritten extents to extents | |
136 | */ | |
137 | static 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 | ||
160 | ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags) | |
161 | { | |
b17b35ec | 162 | ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags); |
bd2d0210 | 163 | if (io) { |
f7ad6d2e TT |
164 | atomic_inc(&EXT4_I(inode)->i_ioend_count); |
165 | io->inode = inode; | |
bd2d0210 TT |
166 | 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 | */ | |
179 | static 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 | ||
187 | static 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; | |
bd2d0210 | 193 | int i; |
d50bdd5a | 194 | sector_t bi_sector = bio->bi_sector; |
bd2d0210 TT |
195 | |
196 | BUG_ON(!io_end); | |
bd2d0210 TT |
197 | bio->bi_private = NULL; |
198 | bio->bi_end_io = NULL; | |
199 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
200 | error = 0; | |
bd2d0210 TT |
201 | bio_put(bio); |
202 | ||
bd2d0210 TT |
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; | |
207 | ||
208 | head = page_buffers(page); | |
209 | if (error) | |
210 | SetPageError(page); | |
211 | BUG_ON(!head); | |
d50bdd5a | 212 | if (head->b_size != PAGE_CACHE_SIZE) { |
bd2d0210 TT |
213 | loff_t offset; |
214 | loff_t io_end_offset = io_end->offset + io_end->size; | |
215 | ||
216 | offset = (sector_t) page->index << PAGE_CACHE_SHIFT; | |
217 | bh = head; | |
218 | do { | |
219 | if ((offset >= io_end->offset) && | |
220 | (offset+bh->b_size <= io_end_offset)) { | |
221 | if (error) | |
222 | buffer_io_error(bh); | |
223 | ||
bd2d0210 TT |
224 | } |
225 | if (buffer_delay(bh)) | |
226 | partial_write = 1; | |
227 | else if (!buffer_mapped(bh)) | |
228 | clear_buffer_dirty(bh); | |
229 | else if (buffer_dirty(bh)) | |
230 | partial_write = 1; | |
231 | offset += bh->b_size; | |
232 | bh = bh->b_this_page; | |
233 | } while (bh != head); | |
234 | } | |
235 | ||
bd2d0210 TT |
236 | /* |
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. | |
242 | */ | |
243 | if (!partial_write) | |
244 | SetPageUptodate(page); | |
08da1193 MT |
245 | |
246 | put_io_page(io_end->pages[i]); | |
bd2d0210 | 247 | } |
bd2d0210 | 248 | io_end->num_io_pages = 0; |
f7ad6d2e TT |
249 | inode = io_end->inode; |
250 | ||
251 | if (error) { | |
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)", | |
255 | inode->i_ino, | |
256 | (unsigned long long) io_end->offset, | |
257 | (long) io_end->size, | |
258 | (unsigned long long) | |
d50bdd5a | 259 | bi_sector >> (inode->i_blkbits - 9)); |
f7ad6d2e | 260 | } |
bd2d0210 TT |
261 | |
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); | |
266 | ||
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); | |
270 | } | |
271 | ||
272 | void ext4_io_submit(struct ext4_io_submit *io) | |
273 | { | |
274 | struct bio *bio = io->io_bio; | |
275 | ||
276 | if (bio) { | |
277 | bio_get(io->io_bio); | |
278 | submit_bio(io->io_op, io->io_bio); | |
279 | BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP)); | |
280 | bio_put(io->io_bio); | |
281 | } | |
7dc57615 | 282 | io->io_bio = NULL; |
bd2d0210 | 283 | io->io_op = 0; |
7dc57615 | 284 | io->io_end = NULL; |
bd2d0210 TT |
285 | } |
286 | ||
287 | static int io_submit_init(struct ext4_io_submit *io, | |
288 | struct inode *inode, | |
289 | struct writeback_control *wbc, | |
290 | struct buffer_head *bh) | |
291 | { | |
292 | ext4_io_end_t *io_end; | |
293 | struct page *page = bh->b_page; | |
294 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
295 | struct bio *bio; | |
296 | ||
297 | io_end = ext4_init_io_end(inode, GFP_NOFS); | |
298 | if (!io_end) | |
299 | return -ENOMEM; | |
300 | do { | |
301 | bio = bio_alloc(GFP_NOIO, nvecs); | |
302 | nvecs >>= 1; | |
303 | } while (bio == NULL); | |
304 | ||
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; | |
309 | ||
bd2d0210 TT |
310 | io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh); |
311 | ||
312 | io->io_bio = bio; | |
313 | io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? | |
314 | WRITE_SYNC_PLUG : WRITE); | |
315 | io->io_next_block = bh->b_blocknr; | |
316 | return 0; | |
317 | } | |
318 | ||
319 | static int io_submit_add_bh(struct ext4_io_submit *io, | |
320 | struct ext4_io_page *io_page, | |
321 | struct inode *inode, | |
322 | struct writeback_control *wbc, | |
323 | struct buffer_head *bh) | |
324 | { | |
325 | ext4_io_end_t *io_end; | |
326 | int ret; | |
327 | ||
328 | if (buffer_new(bh)) { | |
329 | clear_buffer_new(bh); | |
330 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | |
331 | } | |
332 | ||
333 | if (!buffer_mapped(bh) || buffer_delay(bh)) { | |
334 | if (!buffer_mapped(bh)) | |
335 | clear_buffer_dirty(bh); | |
336 | if (io->io_bio) | |
337 | ext4_io_submit(io); | |
338 | return 0; | |
339 | } | |
340 | ||
341 | if (io->io_bio && bh->b_blocknr != io->io_next_block) { | |
342 | submit_and_retry: | |
343 | ext4_io_submit(io); | |
344 | } | |
345 | if (io->io_bio == NULL) { | |
346 | ret = io_submit_init(io, inode, wbc, bh); | |
347 | if (ret) | |
348 | return ret; | |
349 | } | |
350 | io_end = io->io_end; | |
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; | |
357 | io->io_next_block++; | |
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; | |
83668e71 | 364 | atomic_inc(&io_page->p_count); |
bd2d0210 TT |
365 | } |
366 | return 0; | |
367 | } | |
368 | ||
369 | int ext4_bio_write_page(struct ext4_io_submit *io, | |
370 | struct page *page, | |
371 | int len, | |
372 | struct writeback_control *wbc) | |
373 | { | |
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; | |
378 | int ret = 0; | |
379 | ||
380 | blocksize = 1 << inode->i_blkbits; | |
381 | ||
d50bdd5a | 382 | BUG_ON(!PageLocked(page)); |
bd2d0210 TT |
383 | BUG_ON(PageWriteback(page)); |
384 | set_page_writeback(page); | |
385 | ClearPageError(page); | |
386 | ||
387 | io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS); | |
388 | if (!io_page) { | |
389 | set_page_dirty(page); | |
390 | unlock_page(page); | |
391 | return -ENOMEM; | |
392 | } | |
393 | io_page->p_page = page; | |
83668e71 | 394 | atomic_set(&io_page->p_count, 1); |
bd2d0210 TT |
395 | get_page(page); |
396 | ||
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) { | |
d50bdd5a | 400 | |
bd2d0210 TT |
401 | block_end = block_start + blocksize; |
402 | if (block_start >= len) { | |
403 | clear_buffer_dirty(bh); | |
404 | set_buffer_uptodate(bh); | |
405 | continue; | |
406 | } | |
d50bdd5a | 407 | clear_buffer_dirty(bh); |
bd2d0210 TT |
408 | ret = io_submit_add_bh(io, io_page, inode, wbc, bh); |
409 | if (ret) { | |
410 | /* | |
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. | |
414 | */ | |
415 | set_page_dirty(page); | |
416 | break; | |
417 | } | |
418 | } | |
419 | unlock_page(page); | |
420 | /* | |
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 | |
426 | * wedging later on. | |
427 | */ | |
83668e71 | 428 | put_io_page(io_page); |
bd2d0210 TT |
429 | return ret; |
430 | } |