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> | |
21 | #include <linux/uio.h> | |
22 | #include <linux/bio.h> | |
23 | #include <linux/workqueue.h> | |
24 | #include <linux/kernel.h> | |
25 | #include <linux/slab.h> | |
1ae48a63 | 26 | #include <linux/mm.h> |
e8974c39 | 27 | #include <linux/ratelimit.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 | ||
b0857d30 JK |
48 | /* |
49 | * Print an buffer I/O error compatible with the fs/buffer.c. This | |
50 | * provides compatibility with dmesg scrapers that look for a specific | |
51 | * buffer I/O error message. We really need a unified error reporting | |
52 | * structure to userspace ala Digital Unix's uerf system, but it's | |
53 | * probably not going to happen in my lifetime, due to LKML politics... | |
54 | */ | |
55 | static void buffer_io_error(struct buffer_head *bh) | |
56 | { | |
57 | char b[BDEVNAME_SIZE]; | |
e8974c39 | 58 | printk_ratelimited(KERN_ERR "Buffer I/O error on device %s, logical block %llu\n", |
b0857d30 JK |
59 | bdevname(bh->b_bdev, b), |
60 | (unsigned long long)bh->b_blocknr); | |
61 | } | |
62 | ||
63 | static void ext4_finish_bio(struct bio *bio) | |
64 | { | |
65 | int i; | |
66 | int error = !test_bit(BIO_UPTODATE, &bio->bi_flags); | |
2c30c71b | 67 | struct bio_vec *bvec; |
b0857d30 | 68 | |
2c30c71b | 69 | bio_for_each_segment_all(bvec, bio, i) { |
b0857d30 JK |
70 | struct page *page = bvec->bv_page; |
71 | struct buffer_head *bh, *head; | |
72 | unsigned bio_start = bvec->bv_offset; | |
73 | unsigned bio_end = bio_start + bvec->bv_len; | |
74 | unsigned under_io = 0; | |
75 | unsigned long flags; | |
76 | ||
77 | if (!page) | |
78 | continue; | |
79 | ||
80 | if (error) { | |
81 | SetPageError(page); | |
82 | set_bit(AS_EIO, &page->mapping->flags); | |
83 | } | |
84 | bh = head = page_buffers(page); | |
85 | /* | |
86 | * We check all buffers in the page under BH_Uptodate_Lock | |
87 | * to avoid races with other end io clearing async_write flags | |
88 | */ | |
89 | local_irq_save(flags); | |
90 | bit_spin_lock(BH_Uptodate_Lock, &head->b_state); | |
91 | do { | |
92 | if (bh_offset(bh) < bio_start || | |
93 | bh_offset(bh) + bh->b_size > bio_end) { | |
94 | if (buffer_async_write(bh)) | |
95 | under_io++; | |
96 | continue; | |
97 | } | |
98 | clear_buffer_async_write(bh); | |
99 | if (error) | |
100 | buffer_io_error(bh); | |
101 | } while ((bh = bh->b_this_page) != head); | |
102 | bit_spin_unlock(BH_Uptodate_Lock, &head->b_state); | |
103 | local_irq_restore(flags); | |
104 | if (!under_io) | |
105 | end_page_writeback(page); | |
106 | } | |
107 | } | |
108 | ||
97a851ed | 109 | static void ext4_release_io_end(ext4_io_end_t *io_end) |
bd2d0210 | 110 | { |
b0857d30 JK |
111 | struct bio *bio, *next_bio; |
112 | ||
97a851ed JK |
113 | BUG_ON(!list_empty(&io_end->list)); |
114 | BUG_ON(io_end->flag & EXT4_IO_END_UNWRITTEN); | |
6b523df4 | 115 | WARN_ON(io_end->handle); |
97a851ed JK |
116 | |
117 | if (atomic_dec_and_test(&EXT4_I(io_end->inode)->i_ioend_count)) | |
118 | wake_up_all(ext4_ioend_wq(io_end->inode)); | |
b0857d30 JK |
119 | |
120 | for (bio = io_end->bio; bio; bio = next_bio) { | |
121 | next_bio = bio->bi_private; | |
122 | ext4_finish_bio(bio); | |
123 | bio_put(bio); | |
124 | } | |
97a851ed JK |
125 | kmem_cache_free(io_end_cachep, io_end); |
126 | } | |
127 | ||
128 | static void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end) | |
129 | { | |
130 | struct inode *inode = io_end->inode; | |
82e54229 | 131 | |
97a851ed JK |
132 | io_end->flag &= ~EXT4_IO_END_UNWRITTEN; |
133 | /* Wake up anyone waiting on unwritten extent conversion */ | |
134 | if (atomic_dec_and_test(&EXT4_I(inode)->i_unwritten)) | |
135 | wake_up_all(ext4_ioend_wq(inode)); | |
bd2d0210 TT |
136 | } |
137 | ||
a115f749 JK |
138 | /* |
139 | * Check a range of space and convert unwritten extents to written. Note that | |
140 | * we are protected from truncate touching same part of extent tree by the | |
141 | * fact that truncate code waits for all DIO to finish (thus exclusion from | |
142 | * direct IO is achieved) and also waits for PageWriteback bits. Thus we | |
143 | * cannot get to ext4_ext_truncate() before all IOs overlapping that range are | |
144 | * completed (happens from ext4_free_ioend()). | |
145 | */ | |
28a535f9 | 146 | static int ext4_end_io(ext4_io_end_t *io) |
bd2d0210 TT |
147 | { |
148 | struct inode *inode = io->inode; | |
149 | loff_t offset = io->offset; | |
150 | ssize_t size = io->size; | |
6b523df4 | 151 | handle_t *handle = io->handle; |
bd2d0210 TT |
152 | int ret = 0; |
153 | ||
154 | ext4_debug("ext4_end_io_nolock: io 0x%p from inode %lu,list->next 0x%p," | |
155 | "list->prev 0x%p\n", | |
156 | io, inode->i_ino, io->list.next, io->list.prev); | |
157 | ||
6b523df4 JK |
158 | io->handle = NULL; /* Following call will use up the handle */ |
159 | ret = ext4_convert_unwritten_extents(handle, inode, offset, size); | |
bd2d0210 | 160 | if (ret < 0) { |
b82e384c TT |
161 | ext4_msg(inode->i_sb, KERN_EMERG, |
162 | "failed to convert unwritten extents to written " | |
163 | "extents -- potential data loss! " | |
164 | "(inode %lu, offset %llu, size %zd, error %d)", | |
165 | inode->i_ino, offset, size, ret); | |
bd2d0210 | 166 | } |
97a851ed JK |
167 | ext4_clear_io_unwritten_flag(io); |
168 | ext4_release_io_end(io); | |
bd2d0210 TT |
169 | return ret; |
170 | } | |
171 | ||
2e8fa54e | 172 | static void dump_completed_IO(struct inode *inode, struct list_head *head) |
28a535f9 DM |
173 | { |
174 | #ifdef EXT4FS_DEBUG | |
175 | struct list_head *cur, *before, *after; | |
176 | ext4_io_end_t *io, *io0, *io1; | |
28a535f9 | 177 | |
2e8fa54e | 178 | if (list_empty(head)) |
28a535f9 | 179 | return; |
28a535f9 | 180 | |
2e8fa54e JK |
181 | ext4_debug("Dump inode %lu completed io list\n", inode->i_ino); |
182 | list_for_each_entry(io, head, list) { | |
28a535f9 DM |
183 | cur = &io->list; |
184 | before = cur->prev; | |
185 | io0 = container_of(before, ext4_io_end_t, list); | |
186 | after = cur->next; | |
187 | io1 = container_of(after, ext4_io_end_t, list); | |
188 | ||
189 | ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n", | |
190 | io, inode->i_ino, io0, io1); | |
191 | } | |
192 | #endif | |
193 | } | |
194 | ||
195 | /* Add the io_end to per-inode completed end_io list. */ | |
97a851ed | 196 | static void ext4_add_complete_io(ext4_io_end_t *io_end) |
bd2d0210 | 197 | { |
28a535f9 | 198 | struct ext4_inode_info *ei = EXT4_I(io_end->inode); |
78371a45 | 199 | struct ext4_sb_info *sbi = EXT4_SB(io_end->inode->i_sb); |
28a535f9 DM |
200 | struct workqueue_struct *wq; |
201 | unsigned long flags; | |
202 | ||
7b7a8665 CH |
203 | /* Only reserved conversions from writeback should enter here */ |
204 | WARN_ON(!(io_end->flag & EXT4_IO_END_UNWRITTEN)); | |
78371a45 | 205 | WARN_ON(!io_end->handle && sbi->s_journal); |
d73d5046 | 206 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
78371a45 | 207 | wq = sbi->rsv_conversion_wq; |
7b7a8665 CH |
208 | if (list_empty(&ei->i_rsv_conversion_list)) |
209 | queue_work(wq, &ei->i_rsv_conversion_work); | |
210 | list_add_tail(&io_end->list, &ei->i_rsv_conversion_list); | |
28a535f9 DM |
211 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); |
212 | } | |
d73d5046 | 213 | |
2e8fa54e JK |
214 | static int ext4_do_flush_completed_IO(struct inode *inode, |
215 | struct list_head *head) | |
28a535f9 DM |
216 | { |
217 | ext4_io_end_t *io; | |
002bd7fa | 218 | struct list_head unwritten; |
28a535f9 DM |
219 | unsigned long flags; |
220 | struct ext4_inode_info *ei = EXT4_I(inode); | |
221 | int err, ret = 0; | |
222 | ||
28a535f9 | 223 | spin_lock_irqsave(&ei->i_completed_io_lock, flags); |
2e8fa54e JK |
224 | dump_completed_IO(inode, head); |
225 | list_replace_init(head, &unwritten); | |
28a535f9 DM |
226 | spin_unlock_irqrestore(&ei->i_completed_io_lock, flags); |
227 | ||
228 | while (!list_empty(&unwritten)) { | |
229 | io = list_entry(unwritten.next, ext4_io_end_t, list); | |
230 | BUG_ON(!(io->flag & EXT4_IO_END_UNWRITTEN)); | |
231 | list_del_init(&io->list); | |
232 | ||
233 | err = ext4_end_io(io); | |
234 | if (unlikely(!ret && err)) | |
235 | ret = err; | |
28a535f9 DM |
236 | } |
237 | return ret; | |
238 | } | |
239 | ||
240 | /* | |
2e8fa54e | 241 | * work on completed IO, to convert unwritten extents to extents |
28a535f9 | 242 | */ |
2e8fa54e JK |
243 | void ext4_end_io_rsv_work(struct work_struct *work) |
244 | { | |
245 | struct ext4_inode_info *ei = container_of(work, struct ext4_inode_info, | |
246 | i_rsv_conversion_work); | |
247 | ext4_do_flush_completed_IO(&ei->vfs_inode, &ei->i_rsv_conversion_list); | |
248 | } | |
249 | ||
bd2d0210 TT |
250 | ext4_io_end_t *ext4_init_io_end(struct inode *inode, gfp_t flags) |
251 | { | |
b17b35ec | 252 | ext4_io_end_t *io = kmem_cache_zalloc(io_end_cachep, flags); |
bd2d0210 | 253 | if (io) { |
f7ad6d2e TT |
254 | atomic_inc(&EXT4_I(inode)->i_ioend_count); |
255 | io->inode = inode; | |
bd2d0210 | 256 | INIT_LIST_HEAD(&io->list); |
97a851ed | 257 | atomic_set(&io->count, 1); |
bd2d0210 TT |
258 | } |
259 | return io; | |
260 | } | |
261 | ||
97a851ed JK |
262 | void ext4_put_io_end_defer(ext4_io_end_t *io_end) |
263 | { | |
264 | if (atomic_dec_and_test(&io_end->count)) { | |
265 | if (!(io_end->flag & EXT4_IO_END_UNWRITTEN) || !io_end->size) { | |
266 | ext4_release_io_end(io_end); | |
267 | return; | |
268 | } | |
269 | ext4_add_complete_io(io_end); | |
270 | } | |
271 | } | |
272 | ||
273 | int ext4_put_io_end(ext4_io_end_t *io_end) | |
274 | { | |
275 | int err = 0; | |
276 | ||
277 | if (atomic_dec_and_test(&io_end->count)) { | |
278 | if (io_end->flag & EXT4_IO_END_UNWRITTEN) { | |
6b523df4 JK |
279 | err = ext4_convert_unwritten_extents(io_end->handle, |
280 | io_end->inode, io_end->offset, | |
281 | io_end->size); | |
282 | io_end->handle = NULL; | |
97a851ed JK |
283 | ext4_clear_io_unwritten_flag(io_end); |
284 | } | |
285 | ext4_release_io_end(io_end); | |
286 | } | |
287 | return err; | |
288 | } | |
289 | ||
290 | ext4_io_end_t *ext4_get_io_end(ext4_io_end_t *io_end) | |
291 | { | |
292 | atomic_inc(&io_end->count); | |
293 | return io_end; | |
294 | } | |
295 | ||
822dbba3 | 296 | /* BIO completion function for page writeback */ |
bd2d0210 TT |
297 | static void ext4_end_bio(struct bio *bio, int error) |
298 | { | |
299 | ext4_io_end_t *io_end = bio->bi_private; | |
4f024f37 | 300 | sector_t bi_sector = bio->bi_iter.bi_sector; |
bd2d0210 TT |
301 | |
302 | BUG_ON(!io_end); | |
bd2d0210 TT |
303 | bio->bi_end_io = NULL; |
304 | if (test_bit(BIO_UPTODATE, &bio->bi_flags)) | |
305 | error = 0; | |
0058f965 | 306 | |
f7ad6d2e | 307 | if (error) { |
b0857d30 JK |
308 | struct inode *inode = io_end->inode; |
309 | ||
9503c67c | 310 | ext4_warning(inode->i_sb, "I/O error %d writing to inode %lu " |
f7ad6d2e | 311 | "(offset %llu size %ld starting block %llu)", |
9503c67c | 312 | error, inode->i_ino, |
f7ad6d2e TT |
313 | (unsigned long long) io_end->offset, |
314 | (long) io_end->size, | |
315 | (unsigned long long) | |
d50bdd5a | 316 | bi_sector >> (inode->i_blkbits - 9)); |
9503c67c | 317 | mapping_set_error(inode->i_mapping, error); |
f7ad6d2e | 318 | } |
822dbba3 JK |
319 | |
320 | if (io_end->flag & EXT4_IO_END_UNWRITTEN) { | |
321 | /* | |
322 | * Link bio into list hanging from io_end. We have to do it | |
323 | * atomically as bio completions can be racing against each | |
324 | * other. | |
325 | */ | |
326 | bio->bi_private = xchg(&io_end->bio, bio); | |
327 | ext4_put_io_end_defer(io_end); | |
328 | } else { | |
329 | /* | |
330 | * Drop io_end reference early. Inode can get freed once | |
331 | * we finish the bio. | |
332 | */ | |
333 | ext4_put_io_end_defer(io_end); | |
334 | ext4_finish_bio(bio); | |
335 | bio_put(bio); | |
336 | } | |
bd2d0210 TT |
337 | } |
338 | ||
339 | void ext4_io_submit(struct ext4_io_submit *io) | |
340 | { | |
341 | struct bio *bio = io->io_bio; | |
342 | ||
343 | if (bio) { | |
344 | bio_get(io->io_bio); | |
345 | submit_bio(io->io_op, io->io_bio); | |
346 | BUG_ON(bio_flagged(io->io_bio, BIO_EOPNOTSUPP)); | |
347 | bio_put(io->io_bio); | |
348 | } | |
7dc57615 | 349 | io->io_bio = NULL; |
97a851ed JK |
350 | } |
351 | ||
352 | void ext4_io_submit_init(struct ext4_io_submit *io, | |
353 | struct writeback_control *wbc) | |
354 | { | |
355 | io->io_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE); | |
356 | io->io_bio = NULL; | |
7dc57615 | 357 | io->io_end = NULL; |
bd2d0210 TT |
358 | } |
359 | ||
97a851ed JK |
360 | static int io_submit_init_bio(struct ext4_io_submit *io, |
361 | struct buffer_head *bh) | |
bd2d0210 | 362 | { |
bd2d0210 TT |
363 | int nvecs = bio_get_nr_vecs(bh->b_bdev); |
364 | struct bio *bio; | |
365 | ||
275d3ba6 | 366 | bio = bio_alloc(GFP_NOIO, min(nvecs, BIO_MAX_PAGES)); |
a1d8d9a7 TT |
367 | if (!bio) |
368 | return -ENOMEM; | |
4f024f37 | 369 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
bd2d0210 | 370 | bio->bi_bdev = bh->b_bdev; |
bd2d0210 | 371 | bio->bi_end_io = ext4_end_bio; |
97a851ed | 372 | bio->bi_private = ext4_get_io_end(io->io_end); |
bd2d0210 | 373 | io->io_bio = bio; |
bd2d0210 TT |
374 | io->io_next_block = bh->b_blocknr; |
375 | return 0; | |
376 | } | |
377 | ||
378 | static int io_submit_add_bh(struct ext4_io_submit *io, | |
bd2d0210 | 379 | struct inode *inode, |
bd2d0210 TT |
380 | struct buffer_head *bh) |
381 | { | |
bd2d0210 TT |
382 | int ret; |
383 | ||
bd2d0210 TT |
384 | if (io->io_bio && bh->b_blocknr != io->io_next_block) { |
385 | submit_and_retry: | |
386 | ext4_io_submit(io); | |
387 | } | |
388 | if (io->io_bio == NULL) { | |
97a851ed | 389 | ret = io_submit_init_bio(io, bh); |
bd2d0210 TT |
390 | if (ret) |
391 | return ret; | |
392 | } | |
97a851ed JK |
393 | ret = bio_add_page(io->io_bio, bh->b_page, bh->b_size, bh_offset(bh)); |
394 | if (ret != bh->b_size) | |
395 | goto submit_and_retry; | |
bd2d0210 | 396 | io->io_next_block++; |
bd2d0210 TT |
397 | return 0; |
398 | } | |
399 | ||
400 | int ext4_bio_write_page(struct ext4_io_submit *io, | |
401 | struct page *page, | |
402 | int len, | |
1c8349a1 NJ |
403 | struct writeback_control *wbc, |
404 | bool keep_towrite) | |
bd2d0210 TT |
405 | { |
406 | struct inode *inode = page->mapping->host; | |
0058f965 | 407 | unsigned block_start, blocksize; |
bd2d0210 TT |
408 | struct buffer_head *bh, *head; |
409 | int ret = 0; | |
0058f965 | 410 | int nr_submitted = 0; |
bd2d0210 TT |
411 | |
412 | blocksize = 1 << inode->i_blkbits; | |
413 | ||
d50bdd5a | 414 | BUG_ON(!PageLocked(page)); |
bd2d0210 | 415 | BUG_ON(PageWriteback(page)); |
bd2d0210 | 416 | |
1c8349a1 NJ |
417 | if (keep_towrite) |
418 | set_page_writeback_keepwrite(page); | |
419 | else | |
420 | set_page_writeback(page); | |
a54aa761 | 421 | ClearPageError(page); |
bd2d0210 | 422 | |
eeece469 | 423 | /* |
f8409abd | 424 | * Comments copied from block_write_full_page: |
eeece469 JK |
425 | * |
426 | * The page straddles i_size. It must be zeroed out on each and every | |
427 | * writepage invocation because it may be mmapped. "A file is mapped | |
428 | * in multiples of the page size. For a file that is not a multiple of | |
429 | * the page size, the remaining memory is zeroed when mapped, and | |
430 | * writes to that region are not written out to the file." | |
431 | */ | |
432 | if (len < PAGE_CACHE_SIZE) | |
433 | zero_user_segment(page, len, PAGE_CACHE_SIZE); | |
0058f965 JK |
434 | /* |
435 | * In the first loop we prepare and mark buffers to submit. We have to | |
436 | * mark all buffers in the page before submitting so that | |
437 | * end_page_writeback() cannot be called from ext4_bio_end_io() when IO | |
438 | * on the first buffer finishes and we are still working on submitting | |
439 | * the second buffer. | |
440 | */ | |
441 | bh = head = page_buffers(page); | |
442 | do { | |
443 | block_start = bh_offset(bh); | |
bd2d0210 TT |
444 | if (block_start >= len) { |
445 | clear_buffer_dirty(bh); | |
446 | set_buffer_uptodate(bh); | |
447 | continue; | |
448 | } | |
8a850c3f JK |
449 | if (!buffer_dirty(bh) || buffer_delay(bh) || |
450 | !buffer_mapped(bh) || buffer_unwritten(bh)) { | |
451 | /* A hole? We can safely clear the dirty bit */ | |
452 | if (!buffer_mapped(bh)) | |
453 | clear_buffer_dirty(bh); | |
454 | if (io->io_bio) | |
455 | ext4_io_submit(io); | |
456 | continue; | |
457 | } | |
0058f965 JK |
458 | if (buffer_new(bh)) { |
459 | clear_buffer_new(bh); | |
460 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | |
461 | } | |
462 | set_buffer_async_write(bh); | |
463 | } while ((bh = bh->b_this_page) != head); | |
464 | ||
465 | /* Now submit buffers to write */ | |
466 | bh = head = page_buffers(page); | |
467 | do { | |
468 | if (!buffer_async_write(bh)) | |
469 | continue; | |
97a851ed | 470 | ret = io_submit_add_bh(io, inode, bh); |
bd2d0210 TT |
471 | if (ret) { |
472 | /* | |
473 | * We only get here on ENOMEM. Not much else | |
474 | * we can do but mark the page as dirty, and | |
475 | * better luck next time. | |
476 | */ | |
1ae48a63 | 477 | redirty_page_for_writepage(wbc, page); |
bd2d0210 TT |
478 | break; |
479 | } | |
0058f965 | 480 | nr_submitted++; |
1ae48a63 | 481 | clear_buffer_dirty(bh); |
0058f965 JK |
482 | } while ((bh = bh->b_this_page) != head); |
483 | ||
484 | /* Error stopped previous loop? Clean up buffers... */ | |
485 | if (ret) { | |
486 | do { | |
487 | clear_buffer_async_write(bh); | |
488 | bh = bh->b_this_page; | |
489 | } while (bh != head); | |
bd2d0210 TT |
490 | } |
491 | unlock_page(page); | |
0058f965 JK |
492 | /* Nothing submitted - we have to end page writeback */ |
493 | if (!nr_submitted) | |
494 | end_page_writeback(page); | |
bd2d0210 TT |
495 | return ret; |
496 | } |