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 | ||
97a851ed | 65 | static void ext4_release_io_end(ext4_io_end_t *io_end) |
bd2d0210 | 66 | { |
97a851ed JK |
67 | BUG_ON(!list_empty(&io_end->list)); |
68 | BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN); | |
6b523df4 | 69 | WARN_ON(io_end->handle); |
97a851ed JK |
70 | |
71 | if (atomic_dec_and_test(&EXT4_I(io_end->inode)->i_ioend_count)) | |
72 | wake_up_all(ext4_ioend_wq(io_end->inode)); | |
73 | if (io_end->flag & EXT4_IO_END_DIRECT) | |
74 | inode_dio_done(io_end->inode); | |
75 | if (io_end->iocb) | |
76 | aio_complete(io_end->iocb, io_end->result, 0); | |
77 | kmem_cache_free(io_end_cachep, io_end); | |
78 | } | |
79 | ||
80 | static void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end) | |
81 | { | |
82 | struct inode *inode = io_end->inode; | |
82e54229 | 83 | |
97a851ed JK |
84 | io_end->flag &= ~EXT4_IO_END_UNWRITTEN; |
85 | /* Wake up anyone waiting on unwritten extent conversion */ | |
86 | if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten)) | |
87 | wake_up_all(ext4_ioend_wq(inode)); | |
bd2d0210 TT |
88 | } |
89 | ||
28a535f9 DM |
90 | /* check a range of space and convert unwritten extents to written. */ |
91 | static int ext4_end_io(ext4_io_end_t *io) | |
bd2d0210 TT |
92 | { |
93 | struct inode *inode = io->inode; | |
94 | loff_t offset = io->offset; | |
95 | ssize_t size = io->size; | |
6b523df4 | 96 | handle_t *handle = io->handle; |
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 | ||
6b523df4 JK |
103 | io->handle = NULL; /* Following call will use up the handle */ |
104 | ret = ext4_convert_unwritten_extents(handle, inode, offset, size); | |
bd2d0210 | 105 | if (ret < 0) { |
b82e384c TT |
106 | ext4_msg(inode->i_sb, KERN_EMERG, |
107 | "failed to convert unwritten extents to written " | |
108 | "extents -- potential data loss! " | |
109 | "(inode %lu, offset %llu, size %zd, error %d)", | |
110 | inode->i_ino, offset, size, ret); | |
bd2d0210 | 111 | } |
97a851ed JK |
112 | ext4_clear_io_unwritten_flag(io); |
113 | ext4_release_io_end(io); | |
bd2d0210 TT |
114 | return ret; |
115 | } | |
116 | ||
28a535f9 DM |
117 | static void dump_completed_IO(struct inode *inode) |
118 | { | |
119 | #ifdef EXT4FS_DEBUG | |
120 | struct list_head *cur, *before, *after; | |
121 | ext4_io_end_t *io, *io0, *io1; | |
28a535f9 DM |
122 | |
123 | if (list_empty(&EXT4_I(inode)->i_completed_io_list)) { | |
124 | ext4_debug("inode %lu completed_io list is empty\n", | |
125 | inode->i_ino); | |
126 | return; | |
127 | } | |
128 | ||
129 | ext4_debug("Dump inode %lu completed_io list\n", inode->i_ino); | |
130 | list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list) { | |
131 | cur = &io->list; | |
132 | before = cur->prev; | |
133 | io0 = container_of(before, ext4_io_end_t, list); | |
134 | after = cur->next; | |
135 | io1 = container_of(after, ext4_io_end_t, list); | |
136 | ||
137 | ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n", | |
138 | io, inode->i_ino, io0, io1); | |
139 | } | |
140 | #endif | |
141 | } | |
142 | ||
143 | /* Add the io_end to per-inode completed end_io list. */ | |
97a851ed | 144 | static void ext4_add_complete_io(ext4_io_end_t *io_end) |
bd2d0210 | 145 | { |
28a535f9 DM |
146 | struct ext4_inode_info *ei = EXT4_I(io_end->inode); |
147 | struct workqueue_struct *wq; | |
148 | unsigned long flags; | |
149 | ||
150 | BUG_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN)); | |
151 | wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq; | |
bd2d0210 | 152 | |
d73d5046 | 153 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
84c17543 JK |
154 | if (list_empty(&ei->i_completed_io_list)) |
155 | queue_work(wq, &ei->i_unwritten_work); | |
28a535f9 DM |
156 | list_add_tail(&io_end->list, &ei->i_completed_io_list); |
157 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); | |
158 | } | |
d73d5046 | 159 | |
84c17543 | 160 | static int ext4_do_flush_completed_IO(struct inode *inode) |
28a535f9 DM |
161 | { |
162 | ext4_io_end_t *io; | |
002bd7fa | 163 | struct list_head unwritten; |
28a535f9 DM |
164 | unsigned long flags; |
165 | struct ext4_inode_info *ei = EXT4_I(inode); | |
166 | int err, ret = 0; | |
167 | ||
28a535f9 DM |
168 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
169 | dump_completed_IO(inode); | |
170 | list_replace_init(&ei->i_completed_io_list, &unwritten); | |
171 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); | |
172 | ||
173 | while (!list_empty(&unwritten)) { | |
174 | io = list_entry(unwritten.next, ext4_io_end_t, list); | |
175 | BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN)); | |
176 | list_del_init(&io->list); | |
177 | ||
178 | err = ext4_end_io(io); | |
179 | if (unlikely(!ret && err)) | |
180 | ret = err; | |
28a535f9 DM |
181 | } |
182 | return ret; | |
183 | } | |
184 | ||
185 | /* | |
186 | * work on completed aio dio IO, to convert unwritten extents to extents | |
187 | */ | |
84c17543 | 188 | void ext4_end_io_work(struct work_struct *work) |
28a535f9 | 189 | { |
84c17543 JK |
190 | struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info, |
191 | i_unwritten_work); | |
192 | ext4_do_flush_completed_IO(&ei->vfs_inode); | |
28a535f9 DM |
193 | } |
194 | ||
c278531d | 195 | int ext4_flush_unwritten_io(struct inode *inode) |
28a535f9 | 196 | { |
c278531d DM |
197 | int ret; |
198 | WARN_ON_ONCE(!mutex_is_locked(&inode->i_mutex) && | |
199 | !(inode->i_state & I_FREEING)); | |
84c17543 | 200 | ret = ext4_do_flush_completed_IO(inode); |
c278531d DM |
201 | ext4_unwritten_wait(inode); |
202 | return ret; | |
bd2d0210 TT |
203 | } |
204 | ||
205 | ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags) | |
206 | { | |
b17b35ec | 207 | ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags); |
bd2d0210 | 208 | if (io) { |
f7ad6d2e TT |
209 | atomic_inc(&EXT4_I(inode)->i_ioend_count); |
210 | io->inode = inode; | |
bd2d0210 | 211 | INIT_LIST_HEAD(&io->list); |
97a851ed | 212 | atomic_set(&io->count, 1); |
bd2d0210 TT |
213 | } |
214 | return io; | |
215 | } | |
216 | ||
97a851ed JK |
217 | void ext4_put_io_end_defer(ext4_io_end_t *io_end) |
218 | { | |
219 | if (atomic_dec_and_test(&io_end->count)) { | |
220 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) { | |
221 | ext4_release_io_end(io_end); | |
222 | return; | |
223 | } | |
224 | ext4_add_complete_io(io_end); | |
225 | } | |
226 | } | |
227 | ||
228 | int ext4_put_io_end(ext4_io_end_t *io_end) | |
229 | { | |
230 | int err = 0; | |
231 | ||
232 | if (atomic_dec_and_test(&io_end->count)) { | |
233 | if (io_end->flag & EXT4_IO_END_UNWRITTEN) { | |
6b523df4 JK |
234 | err = ext4_convert_unwritten_extents(io_end->handle, |
235 | io_end->inode, io_end->offset, | |
236 | io_end->size); | |
237 | io_end->handle = NULL; | |
97a851ed JK |
238 | ext4_clear_io_unwritten_flag(io_end); |
239 | } | |
240 | ext4_release_io_end(io_end); | |
241 | } | |
242 | return err; | |
243 | } | |
244 | ||
245 | ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end) | |
246 | { | |
247 | atomic_inc(&io_end->count); | |
248 | return io_end; | |
249 | } | |
250 | ||
bd2d0210 TT |
251 | /* |
252 | * Print an buffer I/O error compatible with the fs/buffer.c. This | |
253 | * provides compatibility with dmesg scrapers that look for a specific | |
254 | * buffer I/O error message. We really need a unified error reporting | |
255 | * structure to userspace ala Digital Unix's uerf system, but it's | |
256 | * probably not going to happen in my lifetime, due to LKML politics... | |
257 | */ | |
258 | static void buffer_io_error(struct buffer_head *bh) | |
259 | { | |
260 | char b[BDEVNAME_SIZE]; | |
261 | printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n", | |
262 | bdevname(bh->b_bdev, b), | |
263 | (unsigned long long)bh->b_blocknr); | |
264 | } | |
265 | ||
266 | static void ext4_end_bio(struct bio *bio, int error) | |
267 | { | |
268 | ext4_io_end_t *io_end = bio->bi_private; | |
bd2d0210 | 269 | struct inode *inode; |
bd2d0210 | 270 | int i; |
0058f965 | 271 | int blocksize; |
d50bdd5a | 272 | sector_t bi_sector = bio->bi_sector; |
bd2d0210 TT |
273 | |
274 | BUG_ON(!io_end); | |
0058f965 JK |
275 | inode = io_end->inode; |
276 | blocksize = 1 << inode->i_blkbits; | |
bd2d0210 TT |
277 | bio->bi_private = NULL; |
278 | bio->bi_end_io = NULL; | |
279 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
280 | error = 0; | |
0058f965 JK |
281 | for (i = 0; i < bio->bi_vcnt; i++) { |
282 | struct bio_vec *bvec = &bio->bi_io_vec[i]; | |
283 | struct page *page = bvec->bv_page; | |
bd2d0210 | 284 | struct buffer_head *bh, *head; |
0058f965 JK |
285 | unsigned bio_start = bvec->bv_offset; |
286 | unsigned bio_end = bio_start + bvec->bv_len; | |
287 | unsigned under_io = 0; | |
288 | unsigned long flags; | |
289 | ||
290 | if (!page) | |
291 | continue; | |
bd2d0210 | 292 | |
39db00f1 | 293 | if (error) { |
bd2d0210 | 294 | SetPageError(page); |
39db00f1 | 295 | set_bit(AS_EIO, &page->mapping->flags); |
bd2d0210 | 296 | } |
0058f965 JK |
297 | bh = head = page_buffers(page); |
298 | /* | |
299 | * We check all buffers in the page under BH_Uptodate_Lock | |
300 | * to avoid races with other end io clearing async_write flags | |
301 | */ | |
302 | local_irq_save(flags); | |
303 | bit_spin_lock(BH_Uptodate_Lock, &head->b_state); | |
304 | do { | |
305 | if (bh_offset(bh) < bio_start || | |
306 | bh_offset(bh) + blocksize > bio_end) { | |
307 | if (buffer_async_write(bh)) | |
308 | under_io++; | |
309 | continue; | |
310 | } | |
311 | clear_buffer_async_write(bh); | |
312 | if (error) | |
313 | buffer_io_error(bh); | |
314 | } while ((bh = bh->b_this_page) != head); | |
315 | bit_spin_unlock(BH_Uptodate_Lock, &head->b_state); | |
316 | local_irq_restore(flags); | |
317 | if (!under_io) | |
318 | end_page_writeback(page); | |
bd2d0210 | 319 | } |
0058f965 | 320 | bio_put(bio); |
f7ad6d2e TT |
321 | |
322 | if (error) { | |
323 | io_end->flag |= EXT4_IO_END_ERROR; | |
324 | ext4_warning(inode->i_sb, "I/O error writing to inode %lu " | |
325 | "(offset %llu size %ld starting block %llu)", | |
326 | inode->i_ino, | |
327 | (unsigned long long) io_end->offset, | |
328 | (long) io_end->size, | |
329 | (unsigned long long) | |
d50bdd5a | 330 | bi_sector >> (inode->i_blkbits - 9)); |
f7ad6d2e | 331 | } |
bd2d0210 | 332 | |
97a851ed | 333 | ext4_put_io_end_defer(io_end); |
bd2d0210 TT |
334 | } |
335 | ||
336 | void ext4_io_submit(struct ext4_io_submit *io) | |
337 | { | |
338 | struct bio *bio = io->io_bio; | |
339 | ||
340 | if (bio) { | |
341 | bio_get(io->io_bio); | |
342 | submit_bio(io->io_op, io->io_bio); | |
343 | BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP)); | |
344 | bio_put(io->io_bio); | |
345 | } | |
7dc57615 | 346 | io->io_bio = NULL; |
97a851ed JK |
347 | } |
348 | ||
349 | void ext4_io_submit_init(struct ext4_io_submit *io, | |
350 | struct writeback_control *wbc) | |
351 | { | |
352 | io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE); | |
353 | io->io_bio = NULL; | |
7dc57615 | 354 | io->io_end = NULL; |
bd2d0210 TT |
355 | } |
356 | ||
97a851ed JK |
357 | static int io_submit_init_bio(struct ext4_io_submit *io, |
358 | struct buffer_head *bh) | |
bd2d0210 | 359 | { |
bd2d0210 TT |
360 | int nvecs = bio_get_nr_vecs(bh->b_bdev); |
361 | struct bio *bio; | |
362 | ||
275d3ba6 | 363 | bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES)); |
bd2d0210 TT |
364 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
365 | bio->bi_bdev = bh->b_bdev; | |
bd2d0210 | 366 | bio->bi_end_io = ext4_end_bio; |
97a851ed | 367 | bio->bi_private = ext4_get_io_end(io->io_end); |
bd2d0210 | 368 | io->io_bio = bio; |
bd2d0210 TT |
369 | io->io_next_block = bh->b_blocknr; |
370 | return 0; | |
371 | } | |
372 | ||
373 | static int io_submit_add_bh(struct ext4_io_submit *io, | |
bd2d0210 | 374 | struct inode *inode, |
bd2d0210 TT |
375 | struct buffer_head *bh) |
376 | { | |
bd2d0210 TT |
377 | int ret; |
378 | ||
bd2d0210 TT |
379 | if (io->io_bio && bh->b_blocknr != io->io_next_block) { |
380 | submit_and_retry: | |
381 | ext4_io_submit(io); | |
382 | } | |
383 | if (io->io_bio == NULL) { | |
97a851ed | 384 | ret = io_submit_init_bio(io, bh); |
bd2d0210 TT |
385 | if (ret) |
386 | return ret; | |
387 | } | |
97a851ed JK |
388 | ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh)); |
389 | if (ret != bh->b_size) | |
390 | goto submit_and_retry; | |
bd2d0210 | 391 | io->io_next_block++; |
bd2d0210 TT |
392 | return 0; |
393 | } | |
394 | ||
395 | int ext4_bio_write_page(struct ext4_io_submit *io, | |
396 | struct page *page, | |
397 | int len, | |
398 | struct writeback_control *wbc) | |
399 | { | |
400 | struct inode *inode = page->mapping->host; | |
0058f965 | 401 | unsigned block_start, blocksize; |
bd2d0210 TT |
402 | struct buffer_head *bh, *head; |
403 | int ret = 0; | |
0058f965 | 404 | int nr_submitted = 0; |
bd2d0210 TT |
405 | |
406 | blocksize = 1 << inode->i_blkbits; | |
407 | ||
d50bdd5a | 408 | BUG_ON(!PageLocked(page)); |
bd2d0210 | 409 | BUG_ON(PageWriteback(page)); |
bd2d0210 | 410 | |
a54aa761 TT |
411 | set_page_writeback(page); |
412 | ClearPageError(page); | |
bd2d0210 | 413 | |
0058f965 JK |
414 | /* |
415 | * In the first loop we prepare and mark buffers to submit. We have to | |
416 | * mark all buffers in the page before submitting so that | |
417 | * end_page_writeback() cannot be called from ext4_bio_end_io() when IO | |
418 | * on the first buffer finishes and we are still working on submitting | |
419 | * the second buffer. | |
420 | */ | |
421 | bh = head = page_buffers(page); | |
422 | do { | |
423 | block_start = bh_offset(bh); | |
bd2d0210 | 424 | if (block_start >= len) { |
5a0dc736 YY |
425 | /* |
426 | * Comments copied from block_write_full_page_endio: | |
427 | * | |
428 | * The page straddles i_size. It must be zeroed out on | |
429 | * each and every writepage invocation because it may | |
430 | * be mmapped. "A file is mapped in multiples of the | |
431 | * page size. For a file that is not a multiple of | |
432 | * the page size, the remaining memory is zeroed when | |
433 | * mapped, and writes to that region are not written | |
434 | * out to the file." | |
435 | */ | |
0058f965 JK |
436 | zero_user_segment(page, block_start, |
437 | block_start + blocksize); | |
bd2d0210 TT |
438 | clear_buffer_dirty(bh); |
439 | set_buffer_uptodate(bh); | |
440 | continue; | |
441 | } | |
8a850c3f JK |
442 | if (!buffer_dirty(bh) || buffer_delay(bh) || |
443 | !buffer_mapped(bh) || buffer_unwritten(bh)) { | |
444 | /* A hole? We can safely clear the dirty bit */ | |
445 | if (!buffer_mapped(bh)) | |
446 | clear_buffer_dirty(bh); | |
447 | if (io->io_bio) | |
448 | ext4_io_submit(io); | |
449 | continue; | |
450 | } | |
0058f965 JK |
451 | if (buffer_new(bh)) { |
452 | clear_buffer_new(bh); | |
453 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | |
454 | } | |
455 | set_buffer_async_write(bh); | |
456 | } while ((bh = bh->b_this_page) != head); | |
457 | ||
458 | /* Now submit buffers to write */ | |
459 | bh = head = page_buffers(page); | |
460 | do { | |
461 | if (!buffer_async_write(bh)) | |
462 | continue; | |
97a851ed | 463 | ret = io_submit_add_bh(io, inode, bh); |
bd2d0210 TT |
464 | if (ret) { |
465 | /* | |
466 | * We only get here on ENOMEM. Not much else | |
467 | * we can do but mark the page as dirty, and | |
468 | * better luck next time. | |
469 | */ | |
1ae48a63 | 470 | redirty_page_for_writepage(wbc, page); |
bd2d0210 TT |
471 | break; |
472 | } | |
0058f965 | 473 | nr_submitted++; |
1ae48a63 | 474 | clear_buffer_dirty(bh); |
0058f965 JK |
475 | } while ((bh = bh->b_this_page) != head); |
476 | ||
477 | /* Error stopped previous loop? Clean up buffers... */ | |
478 | if (ret) { | |
479 | do { | |
480 | clear_buffer_async_write(bh); | |
481 | bh = bh->b_this_page; | |
482 | } while (bh != head); | |
bd2d0210 TT |
483 | } |
484 | unlock_page(page); | |
0058f965 JK |
485 | /* Nothing submitted - we have to end page writeback */ |
486 | if (!nr_submitted) | |
487 | end_page_writeback(page); | |
bd2d0210 TT |
488 | return ret; |
489 | } |