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 | ||
bd2d0210 TT |
9 | #include <linux/fs.h> |
10 | #include <linux/time.h> | |
11 | #include <linux/jbd2.h> | |
12 | #include <linux/highuid.h> | |
13 | #include <linux/pagemap.h> | |
14 | #include <linux/quotaops.h> | |
15 | #include <linux/string.h> | |
16 | #include <linux/buffer_head.h> | |
17 | #include <linux/writeback.h> | |
18 | #include <linux/pagevec.h> | |
19 | #include <linux/mpage.h> | |
20 | #include <linux/namei.h> | |
a27bb332 | 21 | #include <linux/aio.h> |
bd2d0210 TT |
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> | |
1ae48a63 | 27 | #include <linux/mm.h> |
bd2d0210 TT |
28 | |
29 | #include "ext4_jbd2.h" | |
30 | #include "xattr.h" | |
31 | #include "acl.h" | |
bd2d0210 | 32 | |
0058f965 | 33 | static struct kmem_cache *io_end_cachep; |
bd2d0210 | 34 | |
5dabfc78 | 35 | int __init ext4_init_pageio(void) |
bd2d0210 | 36 | { |
bd2d0210 | 37 | io_end_cachep = KMEM_CACHE(ext4_io_end, SLAB_RECLAIM_ACCOUNT); |
0058f965 | 38 | if (io_end_cachep == NULL) |
bd2d0210 | 39 | return -ENOMEM; |
bd2d0210 TT |
40 | return 0; |
41 | } | |
42 | ||
5dabfc78 | 43 | void ext4_exit_pageio(void) |
bd2d0210 TT |
44 | { |
45 | kmem_cache_destroy(io_end_cachep); | |
bd2d0210 TT |
46 | } |
47 | ||
1ada47d9 TT |
48 | /* |
49 | * This function is called by ext4_evict_inode() to make sure there is | |
50 | * no more pending I/O completion work left to do. | |
51 | */ | |
52 | void ext4_ioend_shutdown(struct inode *inode) | |
f7ad6d2e | 53 | { |
e9e3bcec | 54 | wait_queue_head_t *wq = ext4_ioend_wq(inode); |
f7ad6d2e TT |
55 | |
56 | wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_ioend_count) == 0)); | |
1ada47d9 TT |
57 | /* |
58 | * We need to make sure the work structure is finished being | |
59 | * used before we let the inode get destroyed. | |
60 | */ | |
61 | if (work_pending(&EXT4_I(inode)->i_unwritten_work)) | |
62 | cancel_work_sync(&EXT4_I(inode)->i_unwritten_work); | |
f7ad6d2e TT |
63 | } |
64 | ||
4eec708d | 65 | static void ext4_release_io_end(ext4_io_end_t *io_end) |
bd2d0210 | 66 | { |
4eec708d JK |
67 | BUG_ON(!list_empty(&io_end->list)); |
68 | BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN); | |
69 | ||
70 | if (atomic_dec_and_test(&EXT4_I(io_end->inode)->i_ioend_count)) | |
71 | wake_up_all(ext4_ioend_wq(io_end->inode)); | |
72 | if (io_end->flag & EXT4_IO_END_DIRECT) | |
73 | inode_dio_done(io_end->inode); | |
74 | if (io_end->iocb) | |
75 | aio_complete(io_end->iocb, io_end->result, 0); | |
76 | kmem_cache_free(io_end_cachep, io_end); | |
77 | } | |
bd2d0210 | 78 | |
4eec708d JK |
79 | static void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end) |
80 | { | |
81 | struct inode *inode = io_end->inode; | |
82e54229 | 82 | |
4eec708d JK |
83 | io_end->flag &= ~EXT4_IO_END_UNWRITTEN; |
84 | /* Wake up anyone waiting on unwritten extent conversion */ | |
85 | if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten)) | |
86 | wake_up_all(ext4_ioend_wq(inode)); | |
bd2d0210 TT |
87 | } |
88 | ||
28a535f9 DM |
89 | /* check a range of space and convert unwritten extents to written. */ |
90 | static int ext4_end_io(ext4_io_end_t *io) | |
bd2d0210 TT |
91 | { |
92 | struct inode *inode = io->inode; | |
93 | loff_t offset = io->offset; | |
94 | ssize_t size = io->size; | |
95 | int ret = 0; | |
96 | ||
97 | ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p," | |
98 | "list->prev 0x%p\n", | |
99 | io, inode->i_ino, io->list.next, io->list.prev); | |
100 | ||
bd2d0210 TT |
101 | ret = ext4_convert_unwritten_extents(inode, offset, size); |
102 | if (ret < 0) { | |
b82e384c TT |
103 | ext4_msg(inode->i_sb, KERN_EMERG, |
104 | "failed to convert unwritten extents to written " | |
105 | "extents -- potential data loss! " | |
106 | "(inode %lu, offset %llu, size %zd, error %d)", | |
107 | inode->i_ino, offset, size, ret); | |
bd2d0210 | 108 | } |
4eec708d JK |
109 | ext4_clear_io_unwritten_flag(io); |
110 | ext4_release_io_end(io); | |
bd2d0210 TT |
111 | return ret; |
112 | } | |
113 | ||
28a535f9 DM |
114 | static void dump_completed_IO(struct inode *inode) |
115 | { | |
116 | #ifdef EXT4FS_DEBUG | |
117 | struct list_head *cur, *before, *after; | |
118 | ext4_io_end_t *io, *io0, *io1; | |
28a535f9 DM |
119 | |
120 | if (list_empty(&EXT4_I(inode)->i_completed_io_list)) { | |
121 | ext4_debug("inode %lu completed_io list is empty\n", | |
122 | inode->i_ino); | |
123 | return; | |
124 | } | |
125 | ||
126 | ext4_debug("Dump inode %lu completed_io list\n", inode->i_ino); | |
127 | list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list) { | |
128 | cur = &io->list; | |
129 | before = cur->prev; | |
130 | io0 = container_of(before, ext4_io_end_t, list); | |
131 | after = cur->next; | |
132 | io1 = container_of(after, ext4_io_end_t, list); | |
133 | ||
134 | ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n", | |
135 | io, inode->i_ino, io0, io1); | |
136 | } | |
137 | #endif | |
138 | } | |
139 | ||
140 | /* Add the io_end to per-inode completed end_io list. */ | |
4eec708d | 141 | static void ext4_add_complete_io(ext4_io_end_t *io_end) |
bd2d0210 | 142 | { |
28a535f9 DM |
143 | struct ext4_inode_info *ei = EXT4_I(io_end->inode); |
144 | struct workqueue_struct *wq; | |
145 | unsigned long flags; | |
146 | ||
147 | BUG_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN)); | |
148 | wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq; | |
bd2d0210 | 149 | |
d73d5046 | 150 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
84c17543 JK |
151 | if (list_empty(&ei->i_completed_io_list)) |
152 | queue_work(wq, &ei->i_unwritten_work); | |
28a535f9 DM |
153 | list_add_tail(&io_end->list, &ei->i_completed_io_list); |
154 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); | |
155 | } | |
d73d5046 | 156 | |
84c17543 | 157 | static int ext4_do_flush_completed_IO(struct inode *inode) |
28a535f9 DM |
158 | { |
159 | ext4_io_end_t *io; | |
002bd7fa | 160 | struct list_head unwritten; |
28a535f9 DM |
161 | unsigned long flags; |
162 | struct ext4_inode_info *ei = EXT4_I(inode); | |
163 | int err, ret = 0; | |
164 | ||
28a535f9 DM |
165 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
166 | dump_completed_IO(inode); | |
167 | list_replace_init(&ei->i_completed_io_list, &unwritten); | |
168 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); | |
169 | ||
170 | while (!list_empty(&unwritten)) { | |
171 | io = list_entry(unwritten.next, ext4_io_end_t, list); | |
172 | BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN)); | |
173 | list_del_init(&io->list); | |
174 | ||
175 | err = ext4_end_io(io); | |
176 | if (unlikely(!ret && err)) | |
177 | ret = err; | |
28a535f9 DM |
178 | } |
179 | return ret; | |
180 | } | |
181 | ||
182 | /* | |
183 | * work on completed aio dio IO, to convert unwritten extents to extents | |
184 | */ | |
84c17543 | 185 | void ext4_end_io_work(struct work_struct *work) |
28a535f9 | 186 | { |
84c17543 JK |
187 | struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info, |
188 | i_unwritten_work); | |
189 | ext4_do_flush_completed_IO(&ei->vfs_inode); | |
28a535f9 DM |
190 | } |
191 | ||
c278531d | 192 | int ext4_flush_unwritten_io(struct inode *inode) |
28a535f9 | 193 | { |
c278531d DM |
194 | int ret; |
195 | WARN_ON_ONCE(!mutex_is_locked(&inode->i_mutex) && | |
196 | !(inode->i_state & I_FREEING)); | |
84c17543 | 197 | ret = ext4_do_flush_completed_IO(inode); |
c278531d DM |
198 | ext4_unwritten_wait(inode); |
199 | return ret; | |
bd2d0210 TT |
200 | } |
201 | ||
202 | ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags) | |
203 | { | |
b17b35ec | 204 | ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags); |
bd2d0210 | 205 | if (io) { |
f7ad6d2e TT |
206 | atomic_inc(&EXT4_I(inode)->i_ioend_count); |
207 | io->inode = inode; | |
bd2d0210 | 208 | INIT_LIST_HEAD(&io->list); |
4eec708d | 209 | atomic_set(&io->count, 1); |
bd2d0210 TT |
210 | } |
211 | return io; | |
212 | } | |
213 | ||
4eec708d JK |
214 | void ext4_put_io_end_defer(ext4_io_end_t *io_end) |
215 | { | |
216 | if (atomic_dec_and_test(&io_end->count)) { | |
217 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) { | |
218 | ext4_release_io_end(io_end); | |
219 | return; | |
220 | } | |
221 | ext4_add_complete_io(io_end); | |
222 | } | |
223 | } | |
224 | ||
225 | int ext4_put_io_end(ext4_io_end_t *io_end) | |
226 | { | |
227 | int err = 0; | |
228 | ||
229 | if (atomic_dec_and_test(&io_end->count)) { | |
230 | if (io_end->flag & EXT4_IO_END_UNWRITTEN) { | |
231 | err = ext4_convert_unwritten_extents(io_end->inode, | |
232 | io_end->offset, io_end->size); | |
233 | ext4_clear_io_unwritten_flag(io_end); | |
234 | } | |
235 | ext4_release_io_end(io_end); | |
236 | } | |
237 | return err; | |
238 | } | |
239 | ||
240 | ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end) | |
241 | { | |
242 | atomic_inc(&io_end->count); | |
243 | return io_end; | |
244 | } | |
245 | ||
bd2d0210 TT |
246 | /* |
247 | * Print an buffer I/O error compatible with the fs/buffer.c. This | |
248 | * provides compatibility with dmesg scrapers that look for a specific | |
249 | * buffer I/O error message. We really need a unified error reporting | |
250 | * structure to userspace ala Digital Unix's uerf system, but it's | |
251 | * probably not going to happen in my lifetime, due to LKML politics... | |
252 | */ | |
253 | static void buffer_io_error(struct buffer_head *bh) | |
254 | { | |
255 | char b[BDEVNAME_SIZE]; | |
256 | printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n", | |
257 | bdevname(bh->b_bdev, b), | |
258 | (unsigned long long)bh->b_blocknr); | |
259 | } | |
260 | ||
261 | static void ext4_end_bio(struct bio *bio, int error) | |
262 | { | |
263 | ext4_io_end_t *io_end = bio->bi_private; | |
bd2d0210 | 264 | struct inode *inode; |
bd2d0210 | 265 | int i; |
0058f965 | 266 | int blocksize; |
d50bdd5a | 267 | sector_t bi_sector = bio->bi_sector; |
bd2d0210 TT |
268 | |
269 | BUG_ON(!io_end); | |
0058f965 JK |
270 | inode = io_end->inode; |
271 | blocksize = 1 << inode->i_blkbits; | |
bd2d0210 TT |
272 | bio->bi_private = NULL; |
273 | bio->bi_end_io = NULL; | |
274 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
275 | error = 0; | |
0058f965 JK |
276 | for (i = 0; i < bio->bi_vcnt; i++) { |
277 | struct bio_vec *bvec = &bio->bi_io_vec[i]; | |
278 | struct page *page = bvec->bv_page; | |
bd2d0210 | 279 | struct buffer_head *bh, *head; |
0058f965 JK |
280 | unsigned bio_start = bvec->bv_offset; |
281 | unsigned bio_end = bio_start + bvec->bv_len; | |
282 | unsigned under_io = 0; | |
283 | unsigned long flags; | |
284 | ||
285 | if (!page) | |
286 | continue; | |
bd2d0210 | 287 | |
39db00f1 | 288 | if (error) { |
bd2d0210 | 289 | SetPageError(page); |
39db00f1 | 290 | set_bit(AS_EIO, &page->mapping->flags); |
bd2d0210 | 291 | } |
0058f965 JK |
292 | bh = head = page_buffers(page); |
293 | /* | |
294 | * We check all buffers in the page under BH_Uptodate_Lock | |
295 | * to avoid races with other end io clearing async_write flags | |
296 | */ | |
297 | local_irq_save(flags); | |
298 | bit_spin_lock(BH_Uptodate_Lock, &head->b_state); | |
299 | do { | |
300 | if (bh_offset(bh) < bio_start || | |
301 | bh_offset(bh) + blocksize > bio_end) { | |
302 | if (buffer_async_write(bh)) | |
303 | under_io++; | |
304 | continue; | |
305 | } | |
306 | clear_buffer_async_write(bh); | |
307 | if (error) | |
308 | buffer_io_error(bh); | |
309 | } while ((bh = bh->b_this_page) != head); | |
310 | bit_spin_unlock(BH_Uptodate_Lock, &head->b_state); | |
311 | local_irq_restore(flags); | |
312 | if (!under_io) | |
313 | end_page_writeback(page); | |
bd2d0210 | 314 | } |
0058f965 | 315 | bio_put(bio); |
f7ad6d2e TT |
316 | |
317 | if (error) { | |
318 | io_end->flag |= EXT4_IO_END_ERROR; | |
319 | ext4_warning(inode->i_sb, "I/O error writing to inode %lu " | |
320 | "(offset %llu size %ld starting block %llu)", | |
321 | inode->i_ino, | |
322 | (unsigned long long) io_end->offset, | |
323 | (long) io_end->size, | |
324 | (unsigned long long) | |
d50bdd5a | 325 | bi_sector >> (inode->i_blkbits - 9)); |
f7ad6d2e | 326 | } |
bd2d0210 | 327 | |
4eec708d | 328 | ext4_put_io_end_defer(io_end); |
bd2d0210 TT |
329 | } |
330 | ||
331 | void ext4_io_submit(struct ext4_io_submit *io) | |
332 | { | |
333 | struct bio *bio = io->io_bio; | |
334 | ||
335 | if (bio) { | |
336 | bio_get(io->io_bio); | |
337 | submit_bio(io->io_op, io->io_bio); | |
338 | BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP)); | |
339 | bio_put(io->io_bio); | |
340 | } | |
7dc57615 | 341 | io->io_bio = NULL; |
4eec708d JK |
342 | } |
343 | ||
344 | void ext4_io_submit_init(struct ext4_io_submit *io, | |
345 | struct writeback_control *wbc) | |
346 | { | |
347 | io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE); | |
348 | io->io_bio = NULL; | |
7dc57615 | 349 | io->io_end = NULL; |
bd2d0210 TT |
350 | } |
351 | ||
4eec708d JK |
352 | static int io_submit_init_bio(struct ext4_io_submit *io, |
353 | struct buffer_head *bh) | |
bd2d0210 | 354 | { |
bd2d0210 TT |
355 | int nvecs = bio_get_nr_vecs(bh->b_bdev); |
356 | struct bio *bio; | |
357 | ||
275d3ba6 | 358 | bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES)); |
bd2d0210 TT |
359 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
360 | bio->bi_bdev = bh->b_bdev; | |
bd2d0210 | 361 | bio->bi_end_io = ext4_end_bio; |
4eec708d JK |
362 | bio->bi_private = ext4_get_io_end(io->io_end); |
363 | if (!io->io_end->size) | |
364 | io->io_end->offset = (bh->b_page->index << PAGE_CACHE_SHIFT) | |
365 | + bh_offset(bh); | |
bd2d0210 | 366 | io->io_bio = bio; |
bd2d0210 TT |
367 | io->io_next_block = bh->b_blocknr; |
368 | return 0; | |
369 | } | |
370 | ||
371 | static int io_submit_add_bh(struct ext4_io_submit *io, | |
bd2d0210 | 372 | struct inode *inode, |
bd2d0210 TT |
373 | struct buffer_head *bh) |
374 | { | |
375 | ext4_io_end_t *io_end; | |
376 | int ret; | |
377 | ||
bd2d0210 TT |
378 | if (io->io_bio && bh->b_blocknr != io->io_next_block) { |
379 | submit_and_retry: | |
380 | ext4_io_submit(io); | |
381 | } | |
382 | if (io->io_bio == NULL) { | |
4eec708d | 383 | ret = io_submit_init_bio(io, bh); |
bd2d0210 TT |
384 | if (ret) |
385 | return ret; | |
386 | } | |
4eec708d JK |
387 | ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh)); |
388 | if (ret != bh->b_size) | |
389 | goto submit_and_retry; | |
bd2d0210 | 390 | io_end = io->io_end; |
7b001d6a | 391 | if (test_clear_buffer_uninit(bh)) |
0edeb71d | 392 | ext4_set_io_unwritten_flag(inode, io_end); |
4eec708d | 393 | io_end->size += bh->b_size; |
bd2d0210 | 394 | io->io_next_block++; |
bd2d0210 TT |
395 | return 0; |
396 | } | |
397 | ||
398 | int ext4_bio_write_page(struct ext4_io_submit *io, | |
399 | struct page *page, | |
400 | int len, | |
401 | struct writeback_control *wbc) | |
402 | { | |
403 | struct inode *inode = page->mapping->host; | |
0058f965 | 404 | unsigned block_start, blocksize; |
bd2d0210 TT |
405 | struct buffer_head *bh, *head; |
406 | int ret = 0; | |
0058f965 | 407 | int nr_submitted = 0; |
bd2d0210 TT |
408 | |
409 | blocksize = 1 << inode->i_blkbits; | |
410 | ||
d50bdd5a | 411 | BUG_ON(!PageLocked(page)); |
bd2d0210 | 412 | BUG_ON(PageWriteback(page)); |
bd2d0210 | 413 | |
a54aa761 TT |
414 | set_page_writeback(page); |
415 | ClearPageError(page); | |
bd2d0210 | 416 | |
0058f965 JK |
417 | /* |
418 | * In the first loop we prepare and mark buffers to submit. We have to | |
419 | * mark all buffers in the page before submitting so that | |
420 | * end_page_writeback() cannot be called from ext4_bio_end_io() when IO | |
421 | * on the first buffer finishes and we are still working on submitting | |
422 | * the second buffer. | |
423 | */ | |
424 | bh = head = page_buffers(page); | |
425 | do { | |
426 | block_start = bh_offset(bh); | |
bd2d0210 | 427 | if (block_start >= len) { |
5a0dc736 YY |
428 | /* |
429 | * Comments copied from block_write_full_page_endio: | |
430 | * | |
431 | * The page straddles i_size. It must be zeroed out on | |
432 | * each and every writepage invocation because it may | |
433 | * be mmapped. "A file is mapped in multiples of the | |
434 | * page size. For a file that is not a multiple of | |
435 | * the page size, the remaining memory is zeroed when | |
436 | * mapped, and writes to that region are not written | |
437 | * out to the file." | |
438 | */ | |
0058f965 JK |
439 | zero_user_segment(page, block_start, |
440 | block_start + blocksize); | |
bd2d0210 TT |
441 | clear_buffer_dirty(bh); |
442 | set_buffer_uptodate(bh); | |
443 | continue; | |
444 | } | |
8a850c3f JK |
445 | if (!buffer_dirty(bh) || buffer_delay(bh) || |
446 | !buffer_mapped(bh) || buffer_unwritten(bh)) { | |
447 | /* A hole? We can safely clear the dirty bit */ | |
448 | if (!buffer_mapped(bh)) | |
449 | clear_buffer_dirty(bh); | |
450 | if (io->io_bio) | |
451 | ext4_io_submit(io); | |
452 | continue; | |
453 | } | |
0058f965 JK |
454 | if (buffer_new(bh)) { |
455 | clear_buffer_new(bh); | |
456 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | |
457 | } | |
458 | set_buffer_async_write(bh); | |
459 | } while ((bh = bh->b_this_page) != head); | |
460 | ||
461 | /* Now submit buffers to write */ | |
462 | bh = head = page_buffers(page); | |
463 | do { | |
464 | if (!buffer_async_write(bh)) | |
465 | continue; | |
4eec708d | 466 | ret = io_submit_add_bh(io, inode, bh); |
bd2d0210 TT |
467 | if (ret) { |
468 | /* | |
469 | * We only get here on ENOMEM. Not much else | |
470 | * we can do but mark the page as dirty, and | |
471 | * better luck next time. | |
472 | */ | |
1ae48a63 | 473 | redirty_page_for_writepage(wbc, page); |
bd2d0210 TT |
474 | break; |
475 | } | |
0058f965 | 476 | nr_submitted++; |
1ae48a63 | 477 | clear_buffer_dirty(bh); |
0058f965 JK |
478 | } while ((bh = bh->b_this_page) != head); |
479 | ||
480 | /* Error stopped previous loop? Clean up buffers... */ | |
481 | if (ret) { | |
482 | do { | |
483 | clear_buffer_async_write(bh); | |
484 | bh = bh->b_this_page; | |
485 | } while (bh != head); | |
bd2d0210 TT |
486 | } |
487 | unlock_page(page); | |
0058f965 JK |
488 | /* Nothing submitted - we have to end page writeback */ |
489 | if (!nr_submitted) | |
490 | end_page_writeback(page); | |
bd2d0210 TT |
491 | return ret; |
492 | } |