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 | ||
a549984b | 65 | void ext4_free_io_end(ext4_io_end_t *io) |
bd2d0210 | 66 | { |
a549984b TT |
67 | BUG_ON(!io); |
68 | BUG_ON(!list_empty(&io->list)); | |
69 | BUG_ON(io->flag & EXT4_IO_END_UNWRITTEN); | |
82e54229 | 70 | |
a549984b TT |
71 | if (atomic_dec_and_test(&EXT4_I(io->inode)->i_ioend_count)) |
72 | wake_up_all(ext4_ioend_wq(io->inode)); | |
73 | kmem_cache_free(io_end_cachep, io); | |
bd2d0210 TT |
74 | } |
75 | ||
28a535f9 DM |
76 | /* check a range of space and convert unwritten extents to written. */ |
77 | static int ext4_end_io(ext4_io_end_t *io) | |
bd2d0210 TT |
78 | { |
79 | struct inode *inode = io->inode; | |
80 | loff_t offset = io->offset; | |
81 | ssize_t size = io->size; | |
82 | int ret = 0; | |
83 | ||
84 | ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p," | |
85 | "list->prev 0x%p\n", | |
86 | io, inode->i_ino, io->list.next, io->list.prev); | |
87 | ||
bd2d0210 TT |
88 | ret = ext4_convert_unwritten_extents(inode, offset, size); |
89 | if (ret < 0) { | |
b82e384c TT |
90 | ext4_msg(inode->i_sb, KERN_EMERG, |
91 | "failed to convert unwritten extents to written " | |
92 | "extents -- potential data loss! " | |
93 | "(inode %lu, offset %llu, size %zd, error %d)", | |
94 | inode->i_ino, offset, size, ret); | |
bd2d0210 | 95 | } |
a549984b TT |
96 | /* Wake up anyone waiting on unwritten extent conversion */ |
97 | if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten)) | |
98 | wake_up_all(ext4_ioend_wq(inode)); | |
99 | if (io->flag & EXT4_IO_END_DIRECT) | |
100 | inode_dio_done(inode); | |
101 | if (io->iocb) | |
102 | aio_complete(io->iocb, io->result, 0); | |
bd2d0210 TT |
103 | return ret; |
104 | } | |
105 | ||
28a535f9 DM |
106 | static void dump_completed_IO(struct inode *inode) |
107 | { | |
108 | #ifdef EXT4FS_DEBUG | |
109 | struct list_head *cur, *before, *after; | |
110 | ext4_io_end_t *io, *io0, *io1; | |
28a535f9 DM |
111 | |
112 | if (list_empty(&EXT4_I(inode)->i_completed_io_list)) { | |
113 | ext4_debug("inode %lu completed_io list is empty\n", | |
114 | inode->i_ino); | |
115 | return; | |
116 | } | |
117 | ||
118 | ext4_debug("Dump inode %lu completed_io list\n", inode->i_ino); | |
119 | list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list) { | |
120 | cur = &io->list; | |
121 | before = cur->prev; | |
122 | io0 = container_of(before, ext4_io_end_t, list); | |
123 | after = cur->next; | |
124 | io1 = container_of(after, ext4_io_end_t, list); | |
125 | ||
126 | ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n", | |
127 | io, inode->i_ino, io0, io1); | |
128 | } | |
129 | #endif | |
130 | } | |
131 | ||
132 | /* Add the io_end to per-inode completed end_io list. */ | |
a549984b | 133 | void ext4_add_complete_io(ext4_io_end_t *io_end) |
bd2d0210 | 134 | { |
28a535f9 DM |
135 | struct ext4_inode_info *ei = EXT4_I(io_end->inode); |
136 | struct workqueue_struct *wq; | |
137 | unsigned long flags; | |
138 | ||
139 | BUG_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN)); | |
140 | wq = EXT4_SB(io_end->inode->i_sb)->dio_unwritten_wq; | |
bd2d0210 | 141 | |
d73d5046 | 142 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
84c17543 JK |
143 | if (list_empty(&ei->i_completed_io_list)) |
144 | queue_work(wq, &ei->i_unwritten_work); | |
28a535f9 DM |
145 | list_add_tail(&io_end->list, &ei->i_completed_io_list); |
146 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); | |
147 | } | |
d73d5046 | 148 | |
84c17543 | 149 | static int ext4_do_flush_completed_IO(struct inode *inode) |
28a535f9 DM |
150 | { |
151 | ext4_io_end_t *io; | |
002bd7fa | 152 | struct list_head unwritten; |
28a535f9 DM |
153 | unsigned long flags; |
154 | struct ext4_inode_info *ei = EXT4_I(inode); | |
155 | int err, ret = 0; | |
156 | ||
28a535f9 DM |
157 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
158 | dump_completed_IO(inode); | |
159 | list_replace_init(&ei->i_completed_io_list, &unwritten); | |
160 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); | |
161 | ||
162 | while (!list_empty(&unwritten)) { | |
163 | io = list_entry(unwritten.next, ext4_io_end_t, list); | |
164 | BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN)); | |
165 | list_del_init(&io->list); | |
166 | ||
167 | err = ext4_end_io(io); | |
168 | if (unlikely(!ret && err)) | |
169 | ret = err; | |
a549984b TT |
170 | io->flag &= ~EXT4_IO_END_UNWRITTEN; |
171 | ext4_free_io_end(io); | |
28a535f9 DM |
172 | } |
173 | return ret; | |
174 | } | |
175 | ||
176 | /* | |
177 | * work on completed aio dio IO, to convert unwritten extents to extents | |
178 | */ | |
84c17543 | 179 | void ext4_end_io_work(struct work_struct *work) |
28a535f9 | 180 | { |
84c17543 JK |
181 | struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info, |
182 | i_unwritten_work); | |
183 | ext4_do_flush_completed_IO(&ei->vfs_inode); | |
28a535f9 DM |
184 | } |
185 | ||
c278531d | 186 | int ext4_flush_unwritten_io(struct inode *inode) |
28a535f9 | 187 | { |
c278531d DM |
188 | int ret; |
189 | WARN_ON_ONCE(!mutex_is_locked(&inode->i_mutex) && | |
190 | !(inode->i_state & I_FREEING)); | |
84c17543 | 191 | ret = ext4_do_flush_completed_IO(inode); |
c278531d DM |
192 | ext4_unwritten_wait(inode); |
193 | return ret; | |
bd2d0210 TT |
194 | } |
195 | ||
196 | ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags) | |
197 | { | |
b17b35ec | 198 | ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags); |
bd2d0210 | 199 | if (io) { |
f7ad6d2e TT |
200 | atomic_inc(&EXT4_I(inode)->i_ioend_count); |
201 | io->inode = inode; | |
bd2d0210 TT |
202 | INIT_LIST_HEAD(&io->list); |
203 | } | |
204 | return io; | |
205 | } | |
206 | ||
207 | /* | |
208 | * Print an buffer I/O error compatible with the fs/buffer.c. This | |
209 | * provides compatibility with dmesg scrapers that look for a specific | |
210 | * buffer I/O error message. We really need a unified error reporting | |
211 | * structure to userspace ala Digital Unix's uerf system, but it's | |
212 | * probably not going to happen in my lifetime, due to LKML politics... | |
213 | */ | |
214 | static void buffer_io_error(struct buffer_head *bh) | |
215 | { | |
216 | char b[BDEVNAME_SIZE]; | |
217 | printk(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n", | |
218 | bdevname(bh->b_bdev, b), | |
219 | (unsigned long long)bh->b_blocknr); | |
220 | } | |
221 | ||
222 | static void ext4_end_bio(struct bio *bio, int error) | |
223 | { | |
224 | ext4_io_end_t *io_end = bio->bi_private; | |
bd2d0210 | 225 | struct inode *inode; |
bd2d0210 | 226 | int i; |
0058f965 | 227 | int blocksize; |
d50bdd5a | 228 | sector_t bi_sector = bio->bi_sector; |
bd2d0210 TT |
229 | |
230 | BUG_ON(!io_end); | |
0058f965 JK |
231 | inode = io_end->inode; |
232 | blocksize = 1 << inode->i_blkbits; | |
bd2d0210 TT |
233 | bio->bi_private = NULL; |
234 | bio->bi_end_io = NULL; | |
235 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
236 | error = 0; | |
0058f965 JK |
237 | for (i = 0; i < bio->bi_vcnt; i++) { |
238 | struct bio_vec *bvec = &bio->bi_io_vec[i]; | |
239 | struct page *page = bvec->bv_page; | |
bd2d0210 | 240 | struct buffer_head *bh, *head; |
0058f965 JK |
241 | unsigned bio_start = bvec->bv_offset; |
242 | unsigned bio_end = bio_start + bvec->bv_len; | |
243 | unsigned under_io = 0; | |
244 | unsigned long flags; | |
245 | ||
246 | if (!page) | |
247 | continue; | |
bd2d0210 | 248 | |
39db00f1 | 249 | if (error) { |
bd2d0210 | 250 | SetPageError(page); |
39db00f1 | 251 | set_bit(AS_EIO, &page->mapping->flags); |
bd2d0210 | 252 | } |
0058f965 JK |
253 | bh = head = page_buffers(page); |
254 | /* | |
255 | * We check all buffers in the page under BH_Uptodate_Lock | |
256 | * to avoid races with other end io clearing async_write flags | |
257 | */ | |
258 | local_irq_save(flags); | |
259 | bit_spin_lock(BH_Uptodate_Lock, &head->b_state); | |
260 | do { | |
261 | if (bh_offset(bh) < bio_start || | |
262 | bh_offset(bh) + blocksize > bio_end) { | |
263 | if (buffer_async_write(bh)) | |
264 | under_io++; | |
265 | continue; | |
266 | } | |
267 | clear_buffer_async_write(bh); | |
268 | if (error) | |
269 | buffer_io_error(bh); | |
270 | } while ((bh = bh->b_this_page) != head); | |
271 | bit_spin_unlock(BH_Uptodate_Lock, &head->b_state); | |
272 | local_irq_restore(flags); | |
273 | if (!under_io) | |
274 | end_page_writeback(page); | |
bd2d0210 | 275 | } |
0058f965 | 276 | bio_put(bio); |
f7ad6d2e TT |
277 | |
278 | if (error) { | |
279 | io_end->flag |= EXT4_IO_END_ERROR; | |
280 | ext4_warning(inode->i_sb, "I/O error writing to inode %lu " | |
281 | "(offset %llu size %ld starting block %llu)", | |
282 | inode->i_ino, | |
283 | (unsigned long long) io_end->offset, | |
284 | (long) io_end->size, | |
285 | (unsigned long long) | |
d50bdd5a | 286 | bi_sector >> (inode->i_blkbits - 9)); |
f7ad6d2e | 287 | } |
bd2d0210 | 288 | |
a549984b TT |
289 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN)) { |
290 | ext4_free_io_end(io_end); | |
291 | return; | |
292 | } | |
293 | ||
294 | ext4_add_complete_io(io_end); | |
bd2d0210 TT |
295 | } |
296 | ||
297 | void ext4_io_submit(struct ext4_io_submit *io) | |
298 | { | |
299 | struct bio *bio = io->io_bio; | |
300 | ||
301 | if (bio) { | |
302 | bio_get(io->io_bio); | |
303 | submit_bio(io->io_op, io->io_bio); | |
304 | BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP)); | |
305 | bio_put(io->io_bio); | |
306 | } | |
7dc57615 | 307 | io->io_bio = NULL; |
a549984b | 308 | io->io_op = 0; |
7dc57615 | 309 | io->io_end = NULL; |
bd2d0210 TT |
310 | } |
311 | ||
a549984b TT |
312 | static int io_submit_init(struct ext4_io_submit *io, |
313 | struct inode *inode, | |
314 | struct writeback_control *wbc, | |
315 | struct buffer_head *bh) | |
bd2d0210 | 316 | { |
a549984b TT |
317 | ext4_io_end_t *io_end; |
318 | struct page *page = bh->b_page; | |
bd2d0210 TT |
319 | int nvecs = bio_get_nr_vecs(bh->b_bdev); |
320 | struct bio *bio; | |
321 | ||
a549984b TT |
322 | io_end = ext4_init_io_end(inode, GFP_NOFS); |
323 | if (!io_end) | |
324 | return -ENOMEM; | |
275d3ba6 | 325 | bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES)); |
bd2d0210 TT |
326 | bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
327 | bio->bi_bdev = bh->b_bdev; | |
a549984b | 328 | bio->bi_private = io->io_end = io_end; |
bd2d0210 | 329 | bio->bi_end_io = ext4_end_bio; |
a549984b TT |
330 | |
331 | io_end->offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(bh); | |
332 | ||
bd2d0210 | 333 | io->io_bio = bio; |
a549984b | 334 | io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE); |
bd2d0210 TT |
335 | io->io_next_block = bh->b_blocknr; |
336 | return 0; | |
337 | } | |
338 | ||
339 | static int io_submit_add_bh(struct ext4_io_submit *io, | |
bd2d0210 | 340 | struct inode *inode, |
a549984b | 341 | struct writeback_control *wbc, |
bd2d0210 TT |
342 | struct buffer_head *bh) |
343 | { | |
344 | ext4_io_end_t *io_end; | |
345 | int ret; | |
346 | ||
bd2d0210 TT |
347 | if (io->io_bio && bh->b_blocknr != io->io_next_block) { |
348 | submit_and_retry: | |
349 | ext4_io_submit(io); | |
350 | } | |
351 | if (io->io_bio == NULL) { | |
a549984b | 352 | ret = io_submit_init(io, inode, wbc, bh); |
bd2d0210 TT |
353 | if (ret) |
354 | return ret; | |
355 | } | |
356 | io_end = io->io_end; | |
7b001d6a | 357 | if (test_clear_buffer_uninit(bh)) |
0edeb71d | 358 | ext4_set_io_unwritten_flag(inode, io_end); |
a549984b | 359 | io->io_end->size += bh->b_size; |
bd2d0210 | 360 | io->io_next_block++; |
a549984b TT |
361 | ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh)); |
362 | if (ret != bh->b_size) | |
363 | goto submit_and_retry; | |
bd2d0210 TT |
364 | return 0; |
365 | } | |
366 | ||
367 | int ext4_bio_write_page(struct ext4_io_submit *io, | |
368 | struct page *page, | |
369 | int len, | |
370 | struct writeback_control *wbc) | |
371 | { | |
372 | struct inode *inode = page->mapping->host; | |
0058f965 | 373 | unsigned block_start, blocksize; |
bd2d0210 TT |
374 | struct buffer_head *bh, *head; |
375 | int ret = 0; | |
0058f965 | 376 | int nr_submitted = 0; |
bd2d0210 TT |
377 | |
378 | blocksize = 1 << inode->i_blkbits; | |
379 | ||
d50bdd5a | 380 | BUG_ON(!PageLocked(page)); |
bd2d0210 | 381 | BUG_ON(PageWriteback(page)); |
bd2d0210 | 382 | |
a54aa761 TT |
383 | set_page_writeback(page); |
384 | ClearPageError(page); | |
bd2d0210 | 385 | |
0058f965 JK |
386 | /* |
387 | * In the first loop we prepare and mark buffers to submit. We have to | |
388 | * mark all buffers in the page before submitting so that | |
389 | * end_page_writeback() cannot be called from ext4_bio_end_io() when IO | |
390 | * on the first buffer finishes and we are still working on submitting | |
391 | * the second buffer. | |
392 | */ | |
393 | bh = head = page_buffers(page); | |
394 | do { | |
395 | block_start = bh_offset(bh); | |
bd2d0210 | 396 | if (block_start >= len) { |
5a0dc736 YY |
397 | /* |
398 | * Comments copied from block_write_full_page_endio: | |
399 | * | |
400 | * The page straddles i_size. It must be zeroed out on | |
401 | * each and every writepage invocation because it may | |
402 | * be mmapped. "A file is mapped in multiples of the | |
403 | * page size. For a file that is not a multiple of | |
404 | * the page size, the remaining memory is zeroed when | |
405 | * mapped, and writes to that region are not written | |
406 | * out to the file." | |
407 | */ | |
0058f965 JK |
408 | zero_user_segment(page, block_start, |
409 | block_start + blocksize); | |
bd2d0210 TT |
410 | clear_buffer_dirty(bh); |
411 | set_buffer_uptodate(bh); | |
412 | continue; | |
413 | } | |
8a850c3f JK |
414 | if (!buffer_dirty(bh) || buffer_delay(bh) || |
415 | !buffer_mapped(bh) || buffer_unwritten(bh)) { | |
416 | /* A hole? We can safely clear the dirty bit */ | |
417 | if (!buffer_mapped(bh)) | |
418 | clear_buffer_dirty(bh); | |
419 | if (io->io_bio) | |
420 | ext4_io_submit(io); | |
421 | continue; | |
422 | } | |
0058f965 JK |
423 | if (buffer_new(bh)) { |
424 | clear_buffer_new(bh); | |
425 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | |
426 | } | |
427 | set_buffer_async_write(bh); | |
428 | } while ((bh = bh->b_this_page) != head); | |
429 | ||
430 | /* Now submit buffers to write */ | |
431 | bh = head = page_buffers(page); | |
432 | do { | |
433 | if (!buffer_async_write(bh)) | |
434 | continue; | |
a549984b | 435 | ret = io_submit_add_bh(io, inode, wbc, bh); |
bd2d0210 TT |
436 | if (ret) { |
437 | /* | |
438 | * We only get here on ENOMEM. Not much else | |
439 | * we can do but mark the page as dirty, and | |
440 | * better luck next time. | |
441 | */ | |
1ae48a63 | 442 | redirty_page_for_writepage(wbc, page); |
bd2d0210 TT |
443 | break; |
444 | } | |
0058f965 | 445 | nr_submitted++; |
1ae48a63 | 446 | clear_buffer_dirty(bh); |
0058f965 JK |
447 | } while ((bh = bh->b_this_page) != head); |
448 | ||
449 | /* Error stopped previous loop? Clean up buffers... */ | |
450 | if (ret) { | |
451 | do { | |
452 | clear_buffer_async_write(bh); | |
453 | bh = bh->b_this_page; | |
454 | } while (bh != head); | |
bd2d0210 TT |
455 | } |
456 | unlock_page(page); | |
0058f965 JK |
457 | /* Nothing submitted - we have to end page writeback */ |
458 | if (!nr_submitted) | |
459 | end_page_writeback(page); | |
bd2d0210 TT |
460 | return ret; |
461 | } |