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 | ||
f7ad6d2e TT |
35 | #define WQ_HASH_SZ 37 |
36 | #define to_ioend_wq(v) (&ioend_wq[((unsigned long)v) % WQ_HASH_SZ]) | |
37 | static wait_queue_head_t ioend_wq[WQ_HASH_SZ]; | |
38 | ||
5dabfc78 | 39 | int __init ext4_init_pageio(void) |
bd2d0210 | 40 | { |
f7ad6d2e TT |
41 | int i; |
42 | ||
bd2d0210 TT |
43 | 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); | |
13195184 | 47 | if (io_end_cachep == NULL) { |
bd2d0210 TT |
48 | kmem_cache_destroy(io_page_cachep); |
49 | return -ENOMEM; | |
50 | } | |
f7ad6d2e TT |
51 | for (i = 0; i < WQ_HASH_SZ; i++) |
52 | init_waitqueue_head(&ioend_wq[i]); | |
bd2d0210 TT |
53 | |
54 | return 0; | |
55 | } | |
56 | ||
5dabfc78 | 57 | void ext4_exit_pageio(void) |
bd2d0210 TT |
58 | { |
59 | kmem_cache_destroy(io_end_cachep); | |
60 | kmem_cache_destroy(io_page_cachep); | |
61 | } | |
62 | ||
f7ad6d2e TT |
63 | void 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 | ||
83668e71 TT |
70 | static 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 | ||
bd2d0210 TT |
79 | void ext4_free_io_end(ext4_io_end_t *io) |
80 | { | |
81 | int i; | |
f7ad6d2e | 82 | wait_queue_head_t *wq; |
bd2d0210 TT |
83 | |
84 | BUG_ON(!io); | |
85 | if (io->page) | |
86 | put_page(io->page); | |
83668e71 TT |
87 | for (i = 0; i < io->num_io_pages; i++) |
88 | put_io_page(io->pages[i]); | |
bd2d0210 | 89 | io->num_io_pages = 0; |
f7ad6d2e TT |
90 | 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); | |
bd2d0210 TT |
94 | kmem_cache_free(io_end_cachep, io); |
95 | } | |
96 | ||
97 | /* | |
98 | * check a range of space and convert unwritten extents to written. | |
99 | */ | |
100 | int 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 | */ | |
136 | static 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 | ||
159 | ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags) | |
160 | { | |
b17b35ec | 161 | ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags); |
bd2d0210 | 162 | if (io) { |
f7ad6d2e TT |
163 | atomic_inc(&EXT4_I(inode)->i_ioend_count); |
164 | io->inode = inode; | |
bd2d0210 TT |
165 | INIT_WORK(&io->work, ext4_end_io_work); |
166 | INIT_LIST_HEAD(&io->list); | |
167 | } | |
168 | return io; | |
169 | } | |
170 | ||
171 | /* | |
172 | * Print an buffer I/O error compatible with the fs/buffer.c. This | |
173 | * provides compatibility with dmesg scrapers that look for a specific | |
174 | * buffer I/O error message. We really need a unified error reporting | |
175 | * structure to userspace ala Digital Unix's uerf system, but it's | |
176 | * probably not going to happen in my lifetime, due to LKML politics... | |
177 | */ | |
178 | static void buffer_io_error(struct buffer_head *bh) | |
179 | { | |
180 | char b[BDEVNAME_SIZE]; | |
181 | printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n", | |
182 | bdevname(bh->b_bdev, b), | |
183 | (unsigned long long)bh->b_blocknr); | |
184 | } | |
185 | ||
186 | static void ext4_end_bio(struct bio *bio, int error) | |
187 | { | |
188 | ext4_io_end_t *io_end = bio->bi_private; | |
189 | struct workqueue_struct *wq; | |
190 | struct inode *inode; | |
191 | unsigned long flags; | |
bd2d0210 | 192 | int i; |
d50bdd5a | 193 | sector_t bi_sector = bio->bi_sector; |
bd2d0210 TT |
194 | |
195 | BUG_ON(!io_end); | |
bd2d0210 TT |
196 | bio->bi_private = NULL; |
197 | bio->bi_end_io = NULL; | |
198 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
199 | error = 0; | |
bd2d0210 TT |
200 | bio_put(bio); |
201 | ||
bd2d0210 TT |
202 | for (i = 0; i < io_end->num_io_pages; i++) { |
203 | struct page *page = io_end->pages[i]->p_page; | |
204 | struct buffer_head *bh, *head; | |
205 | int partial_write = 0; | |
206 | ||
207 | head = page_buffers(page); | |
208 | if (error) | |
209 | SetPageError(page); | |
210 | BUG_ON(!head); | |
d50bdd5a | 211 | if (head->b_size != PAGE_CACHE_SIZE) { |
bd2d0210 TT |
212 | loff_t offset; |
213 | loff_t io_end_offset = io_end->offset + io_end->size; | |
214 | ||
215 | offset = (sector_t) page->index << PAGE_CACHE_SHIFT; | |
216 | bh = head; | |
217 | do { | |
218 | if ((offset >= io_end->offset) && | |
219 | (offset+bh->b_size <= io_end_offset)) { | |
220 | if (error) | |
221 | buffer_io_error(bh); | |
222 | ||
bd2d0210 TT |
223 | } |
224 | if (buffer_delay(bh)) | |
225 | partial_write = 1; | |
226 | else if (!buffer_mapped(bh)) | |
227 | clear_buffer_dirty(bh); | |
228 | else if (buffer_dirty(bh)) | |
229 | partial_write = 1; | |
230 | offset += bh->b_size; | |
231 | bh = bh->b_this_page; | |
232 | } while (bh != head); | |
233 | } | |
234 | ||
bd2d0210 TT |
235 | /* |
236 | * If this is a partial write which happened to make | |
237 | * all buffers uptodate then we can optimize away a | |
238 | * bogus readpage() for the next read(). Here we | |
239 | * 'discover' whether the page went uptodate as a | |
240 | * result of this (potentially partial) write. | |
241 | */ | |
242 | if (!partial_write) | |
243 | SetPageUptodate(page); | |
08da1193 MT |
244 | |
245 | put_io_page(io_end->pages[i]); | |
bd2d0210 | 246 | } |
bd2d0210 | 247 | io_end->num_io_pages = 0; |
f7ad6d2e TT |
248 | inode = io_end->inode; |
249 | ||
250 | if (error) { | |
251 | io_end->flag |= EXT4_IO_END_ERROR; | |
252 | ext4_warning(inode->i_sb, "I/O error writing to inode %lu " | |
253 | "(offset %llu size %ld starting block %llu)", | |
254 | inode->i_ino, | |
255 | (unsigned long long) io_end->offset, | |
256 | (long) io_end->size, | |
257 | (unsigned long long) | |
d50bdd5a | 258 | bi_sector >> (inode->i_blkbits - 9)); |
f7ad6d2e | 259 | } |
bd2d0210 TT |
260 | |
261 | /* Add the io_end to per-inode completed io list*/ | |
262 | spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags); | |
263 | list_add_tail(&io_end->list, &EXT4_I(inode)->i_completed_io_list); | |
264 | spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags); | |
265 | ||
266 | wq = EXT4_SB(inode->i_sb)->dio_unwritten_wq; | |
267 | /* queue the work to convert unwritten extents to written */ | |
268 | queue_work(wq, &io_end->work); | |
269 | } | |
270 | ||
271 | void ext4_io_submit(struct ext4_io_submit *io) | |
272 | { | |
273 | struct bio *bio = io->io_bio; | |
274 | ||
275 | if (bio) { | |
276 | bio_get(io->io_bio); | |
277 | submit_bio(io->io_op, io->io_bio); | |
278 | BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP)); | |
279 | bio_put(io->io_bio); | |
280 | } | |
281 | io->io_bio = 0; | |
282 | io->io_op = 0; | |
283 | io->io_end = 0; | |
284 | } | |
285 | ||
286 | static int io_submit_init(struct ext4_io_submit *io, | |
287 | struct inode *inode, | |
288 | struct writeback_control *wbc, | |
289 | struct buffer_head *bh) | |
290 | { | |
291 | ext4_io_end_t *io_end; | |
292 | struct page *page = bh->b_page; | |
293 | int nvecs = bio_get_nr_vecs(bh->b_bdev); | |
294 | struct bio *bio; | |
295 | ||
296 | io_end = ext4_init_io_end(inode, GFP_NOFS); | |
297 | if (!io_end) | |
298 | return -ENOMEM; | |
299 | do { | |
300 | bio = bio_alloc(GFP_NOIO, nvecs); | |
301 | nvecs >>= 1; | |
302 | } while (bio == NULL); | |
303 | ||
304 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); | |
305 | bio->bi_bdev = bh->b_bdev; | |
306 | bio->bi_private = io->io_end = io_end; | |
307 | bio->bi_end_io = ext4_end_bio; | |
308 | ||
bd2d0210 TT |
309 | io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh); |
310 | ||
311 | io->io_bio = bio; | |
312 | io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? | |
313 | WRITE_SYNC_PLUG : WRITE); | |
314 | io->io_next_block = bh->b_blocknr; | |
315 | return 0; | |
316 | } | |
317 | ||
318 | static int io_submit_add_bh(struct ext4_io_submit *io, | |
319 | struct ext4_io_page *io_page, | |
320 | struct inode *inode, | |
321 | struct writeback_control *wbc, | |
322 | struct buffer_head *bh) | |
323 | { | |
324 | ext4_io_end_t *io_end; | |
325 | int ret; | |
326 | ||
327 | if (buffer_new(bh)) { | |
328 | clear_buffer_new(bh); | |
329 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | |
330 | } | |
331 | ||
332 | if (!buffer_mapped(bh) || buffer_delay(bh)) { | |
333 | if (!buffer_mapped(bh)) | |
334 | clear_buffer_dirty(bh); | |
335 | if (io->io_bio) | |
336 | ext4_io_submit(io); | |
337 | return 0; | |
338 | } | |
339 | ||
340 | if (io->io_bio && bh->b_blocknr != io->io_next_block) { | |
341 | submit_and_retry: | |
342 | ext4_io_submit(io); | |
343 | } | |
344 | if (io->io_bio == NULL) { | |
345 | ret = io_submit_init(io, inode, wbc, bh); | |
346 | if (ret) | |
347 | return ret; | |
348 | } | |
349 | io_end = io->io_end; | |
350 | if ((io_end->num_io_pages >= MAX_IO_PAGES) && | |
351 | (io_end->pages[io_end->num_io_pages-1] != io_page)) | |
352 | goto submit_and_retry; | |
353 | if (buffer_uninit(bh)) | |
354 | io->io_end->flag |= EXT4_IO_END_UNWRITTEN; | |
355 | io->io_end->size += bh->b_size; | |
356 | io->io_next_block++; | |
357 | ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh)); | |
358 | if (ret != bh->b_size) | |
359 | goto submit_and_retry; | |
360 | if ((io_end->num_io_pages == 0) || | |
361 | (io_end->pages[io_end->num_io_pages-1] != io_page)) { | |
362 | io_end->pages[io_end->num_io_pages++] = io_page; | |
83668e71 | 363 | atomic_inc(&io_page->p_count); |
bd2d0210 TT |
364 | } |
365 | return 0; | |
366 | } | |
367 | ||
368 | int ext4_bio_write_page(struct ext4_io_submit *io, | |
369 | struct page *page, | |
370 | int len, | |
371 | struct writeback_control *wbc) | |
372 | { | |
373 | struct inode *inode = page->mapping->host; | |
374 | unsigned block_start, block_end, blocksize; | |
375 | struct ext4_io_page *io_page; | |
376 | struct buffer_head *bh, *head; | |
377 | int ret = 0; | |
378 | ||
379 | blocksize = 1 << inode->i_blkbits; | |
380 | ||
d50bdd5a | 381 | BUG_ON(!PageLocked(page)); |
bd2d0210 TT |
382 | BUG_ON(PageWriteback(page)); |
383 | set_page_writeback(page); | |
384 | ClearPageError(page); | |
385 | ||
386 | io_page = kmem_cache_alloc(io_page_cachep, GFP_NOFS); | |
387 | if (!io_page) { | |
388 | set_page_dirty(page); | |
389 | unlock_page(page); | |
390 | return -ENOMEM; | |
391 | } | |
392 | io_page->p_page = page; | |
83668e71 | 393 | atomic_set(&io_page->p_count, 1); |
bd2d0210 TT |
394 | get_page(page); |
395 | ||
396 | for (bh = head = page_buffers(page), block_start = 0; | |
397 | bh != head || !block_start; | |
398 | block_start = block_end, bh = bh->b_this_page) { | |
d50bdd5a | 399 | |
bd2d0210 TT |
400 | block_end = block_start + blocksize; |
401 | if (block_start >= len) { | |
402 | clear_buffer_dirty(bh); | |
403 | set_buffer_uptodate(bh); | |
404 | continue; | |
405 | } | |
d50bdd5a | 406 | clear_buffer_dirty(bh); |
bd2d0210 TT |
407 | ret = io_submit_add_bh(io, io_page, inode, wbc, bh); |
408 | if (ret) { | |
409 | /* | |
410 | * We only get here on ENOMEM. Not much else | |
411 | * we can do but mark the page as dirty, and | |
412 | * better luck next time. | |
413 | */ | |
414 | set_page_dirty(page); | |
415 | break; | |
416 | } | |
417 | } | |
418 | unlock_page(page); | |
419 | /* | |
420 | * If the page was truncated before we could do the writeback, | |
421 | * or we had a memory allocation error while trying to write | |
422 | * the first buffer head, we won't have submitted any pages for | |
423 | * I/O. In that case we need to make sure we've cleared the | |
424 | * PageWriteback bit from the page to prevent the system from | |
425 | * wedging later on. | |
426 | */ | |
83668e71 | 427 | put_io_page(io_page); |
bd2d0210 TT |
428 | return ret; |
429 | } |