Commit | Line | Data |
---|---|---|
0e9cebe7 JB |
1 | /* |
2 | * Copyright (C) 2014 Facebook. All rights reserved. | |
3 | * | |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | ||
7 | #include <linux/device-mapper.h> | |
8 | ||
9 | #include <linux/module.h> | |
10 | #include <linux/init.h> | |
11 | #include <linux/blkdev.h> | |
12 | #include <linux/bio.h> | |
98d82f48 | 13 | #include <linux/dax.h> |
0e9cebe7 JB |
14 | #include <linux/slab.h> |
15 | #include <linux/kthread.h> | |
16 | #include <linux/freezer.h> | |
98d82f48 | 17 | #include <linux/uio.h> |
0e9cebe7 JB |
18 | |
19 | #define DM_MSG_PREFIX "log-writes" | |
20 | ||
21 | /* | |
22 | * This target will sequentially log all writes to the target device onto the | |
23 | * log device. This is helpful for replaying writes to check for fs consistency | |
24 | * at all times. This target provides a mechanism to mark specific events to | |
25 | * check data at a later time. So for example you would: | |
26 | * | |
27 | * write data | |
28 | * fsync | |
29 | * dmsetup message /dev/whatever mark mymark | |
30 | * unmount /mnt/test | |
31 | * | |
32 | * Then replay the log up to mymark and check the contents of the replay to | |
33 | * verify it matches what was written. | |
34 | * | |
35 | * We log writes only after they have been flushed, this makes the log describe | |
36 | * close to the order in which the data hits the actual disk, not its cache. So | |
37 | * for example the following sequence (W means write, C means complete) | |
38 | * | |
39 | * Wa,Wb,Wc,Cc,Ca,FLUSH,FUAd,Cb,CFLUSH,CFUAd | |
40 | * | |
41 | * Would result in the log looking like this: | |
42 | * | |
43 | * c,a,flush,fuad,b,<other writes>,<next flush> | |
44 | * | |
45 | * This is meant to help expose problems where file systems do not properly wait | |
46 | * on data being written before invoking a FLUSH. FUA bypasses cache so once it | |
47 | * completes it is added to the log as it should be on disk. | |
48 | * | |
49 | * We treat DISCARDs as if they don't bypass cache so that they are logged in | |
50 | * order of completion along with the normal writes. If we didn't do it this | |
51 | * way we would process all the discards first and then write all the data, when | |
52 | * in fact we want to do the data and the discard in the order that they | |
53 | * completed. | |
54 | */ | |
e5c4cb9b QW |
55 | #define LOG_FLUSH_FLAG (1 << 0) |
56 | #define LOG_FUA_FLAG (1 << 1) | |
57 | #define LOG_DISCARD_FLAG (1 << 2) | |
58 | #define LOG_MARK_FLAG (1 << 3) | |
59 | #define LOG_METADATA_FLAG (1 << 4) | |
0e9cebe7 | 60 | |
f4ad317a GU |
61 | #define WRITE_LOG_VERSION 1ULL |
62 | #define WRITE_LOG_MAGIC 0x6a736677736872ULL | |
0e9cebe7 JB |
63 | |
64 | /* | |
65 | * The disk format for this is braindead simple. | |
66 | * | |
67 | * At byte 0 we have our super, followed by the following sequence for | |
68 | * nr_entries: | |
69 | * | |
70 | * [ 1 sector ][ entry->nr_sectors ] | |
71 | * [log_write_entry][ data written ] | |
72 | * | |
73 | * The log_write_entry takes up a full sector so we can have arbitrary length | |
74 | * marks and it leaves us room for extra content in the future. | |
75 | */ | |
76 | ||
77 | /* | |
78 | * Basic info about the log for userspace. | |
79 | */ | |
80 | struct log_write_super { | |
81 | __le64 magic; | |
82 | __le64 version; | |
83 | __le64 nr_entries; | |
84 | __le32 sectorsize; | |
85 | }; | |
86 | ||
87 | /* | |
88 | * sector - the sector we wrote. | |
89 | * nr_sectors - the number of sectors we wrote. | |
90 | * flags - flags for this log entry. | |
91 | * data_len - the size of the data in this log entry, this is for private log | |
92 | * entry stuff, the MARK data provided by userspace for example. | |
93 | */ | |
94 | struct log_write_entry { | |
95 | __le64 sector; | |
96 | __le64 nr_sectors; | |
97 | __le64 flags; | |
98 | __le64 data_len; | |
99 | }; | |
100 | ||
101 | struct log_writes_c { | |
102 | struct dm_dev *dev; | |
103 | struct dm_dev *logdev; | |
104 | u64 logged_entries; | |
105 | u32 sectorsize; | |
228bb5b2 | 106 | u32 sectorshift; |
0e9cebe7 JB |
107 | atomic_t io_blocks; |
108 | atomic_t pending_blocks; | |
109 | sector_t next_sector; | |
110 | sector_t end_sector; | |
111 | bool logging_enabled; | |
112 | bool device_supports_discard; | |
113 | spinlock_t blocks_lock; | |
114 | struct list_head unflushed_blocks; | |
115 | struct list_head logging_blocks; | |
116 | wait_queue_head_t wait; | |
117 | struct task_struct *log_kthread; | |
118 | }; | |
119 | ||
120 | struct pending_block { | |
121 | int vec_cnt; | |
122 | u64 flags; | |
123 | sector_t sector; | |
124 | sector_t nr_sectors; | |
125 | char *data; | |
126 | u32 datalen; | |
127 | struct list_head list; | |
128 | struct bio_vec vecs[0]; | |
129 | }; | |
130 | ||
131 | struct per_bio_data { | |
132 | struct pending_block *block; | |
133 | }; | |
134 | ||
228bb5b2 JB |
135 | static inline sector_t bio_to_dev_sectors(struct log_writes_c *lc, |
136 | sector_t sectors) | |
137 | { | |
138 | return sectors >> (lc->sectorshift - SECTOR_SHIFT); | |
139 | } | |
140 | ||
141 | static inline sector_t dev_to_bio_sectors(struct log_writes_c *lc, | |
142 | sector_t sectors) | |
143 | { | |
144 | return sectors << (lc->sectorshift - SECTOR_SHIFT); | |
145 | } | |
146 | ||
0e9cebe7 JB |
147 | static void put_pending_block(struct log_writes_c *lc) |
148 | { | |
149 | if (atomic_dec_and_test(&lc->pending_blocks)) { | |
150 | smp_mb__after_atomic(); | |
151 | if (waitqueue_active(&lc->wait)) | |
152 | wake_up(&lc->wait); | |
153 | } | |
154 | } | |
155 | ||
156 | static void put_io_block(struct log_writes_c *lc) | |
157 | { | |
158 | if (atomic_dec_and_test(&lc->io_blocks)) { | |
159 | smp_mb__after_atomic(); | |
160 | if (waitqueue_active(&lc->wait)) | |
161 | wake_up(&lc->wait); | |
162 | } | |
163 | } | |
164 | ||
4246a0b6 | 165 | static void log_end_io(struct bio *bio) |
0e9cebe7 JB |
166 | { |
167 | struct log_writes_c *lc = bio->bi_private; | |
0e9cebe7 | 168 | |
4e4cbee9 | 169 | if (bio->bi_status) { |
0e9cebe7 JB |
170 | unsigned long flags; |
171 | ||
4e4cbee9 | 172 | DMERR("Error writing log block, error=%d", bio->bi_status); |
0e9cebe7 JB |
173 | spin_lock_irqsave(&lc->blocks_lock, flags); |
174 | lc->logging_enabled = false; | |
175 | spin_unlock_irqrestore(&lc->blocks_lock, flags); | |
176 | } | |
177 | ||
491221f8 | 178 | bio_free_pages(bio); |
0e9cebe7 JB |
179 | put_io_block(lc); |
180 | bio_put(bio); | |
181 | } | |
182 | ||
183 | /* | |
184 | * Meant to be called if there is an error, it will free all the pages | |
185 | * associated with the block. | |
186 | */ | |
187 | static void free_pending_block(struct log_writes_c *lc, | |
188 | struct pending_block *block) | |
189 | { | |
190 | int i; | |
191 | ||
192 | for (i = 0; i < block->vec_cnt; i++) { | |
193 | if (block->vecs[i].bv_page) | |
194 | __free_page(block->vecs[i].bv_page); | |
195 | } | |
196 | kfree(block->data); | |
197 | kfree(block); | |
198 | put_pending_block(lc); | |
199 | } | |
200 | ||
201 | static int write_metadata(struct log_writes_c *lc, void *entry, | |
202 | size_t entrylen, void *data, size_t datalen, | |
203 | sector_t sector) | |
204 | { | |
205 | struct bio *bio; | |
206 | struct page *page; | |
207 | void *ptr; | |
208 | size_t ret; | |
209 | ||
210 | bio = bio_alloc(GFP_KERNEL, 1); | |
211 | if (!bio) { | |
212 | DMERR("Couldn't alloc log bio"); | |
213 | goto error; | |
214 | } | |
215 | bio->bi_iter.bi_size = 0; | |
216 | bio->bi_iter.bi_sector = sector; | |
74d46992 | 217 | bio_set_dev(bio, lc->logdev->bdev); |
0e9cebe7 JB |
218 | bio->bi_end_io = log_end_io; |
219 | bio->bi_private = lc; | |
e6047149 | 220 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
0e9cebe7 JB |
221 | |
222 | page = alloc_page(GFP_KERNEL); | |
223 | if (!page) { | |
224 | DMERR("Couldn't alloc log page"); | |
225 | bio_put(bio); | |
226 | goto error; | |
227 | } | |
228 | ||
229 | ptr = kmap_atomic(page); | |
230 | memcpy(ptr, entry, entrylen); | |
231 | if (datalen) | |
232 | memcpy(ptr + entrylen, data, datalen); | |
233 | memset(ptr + entrylen + datalen, 0, | |
234 | lc->sectorsize - entrylen - datalen); | |
235 | kunmap_atomic(ptr); | |
236 | ||
237 | ret = bio_add_page(bio, page, lc->sectorsize, 0); | |
238 | if (ret != lc->sectorsize) { | |
239 | DMERR("Couldn't add page to the log block"); | |
240 | goto error_bio; | |
241 | } | |
4e49ea4a | 242 | submit_bio(bio); |
0e9cebe7 JB |
243 | return 0; |
244 | error_bio: | |
245 | bio_put(bio); | |
246 | __free_page(page); | |
247 | error: | |
248 | put_io_block(lc); | |
249 | return -1; | |
250 | } | |
251 | ||
e5a20660 RZ |
252 | static int write_inline_data(struct log_writes_c *lc, void *entry, |
253 | size_t entrylen, void *data, size_t datalen, | |
254 | sector_t sector) | |
255 | { | |
256 | int num_pages, bio_pages, pg_datalen, pg_sectorlen, i; | |
257 | struct page *page; | |
258 | struct bio *bio; | |
259 | size_t ret; | |
260 | void *ptr; | |
261 | ||
262 | while (datalen) { | |
263 | num_pages = ALIGN(datalen, PAGE_SIZE) >> PAGE_SHIFT; | |
264 | bio_pages = min(num_pages, BIO_MAX_PAGES); | |
265 | ||
266 | atomic_inc(&lc->io_blocks); | |
267 | ||
268 | bio = bio_alloc(GFP_KERNEL, bio_pages); | |
269 | if (!bio) { | |
270 | DMERR("Couldn't alloc inline data bio"); | |
271 | goto error; | |
272 | } | |
273 | ||
274 | bio->bi_iter.bi_size = 0; | |
275 | bio->bi_iter.bi_sector = sector; | |
276 | bio_set_dev(bio, lc->logdev->bdev); | |
277 | bio->bi_end_io = log_end_io; | |
278 | bio->bi_private = lc; | |
279 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); | |
280 | ||
281 | for (i = 0; i < bio_pages; i++) { | |
282 | pg_datalen = min_t(int, datalen, PAGE_SIZE); | |
283 | pg_sectorlen = ALIGN(pg_datalen, lc->sectorsize); | |
284 | ||
285 | page = alloc_page(GFP_KERNEL); | |
286 | if (!page) { | |
287 | DMERR("Couldn't alloc inline data page"); | |
288 | goto error_bio; | |
289 | } | |
290 | ||
291 | ptr = kmap_atomic(page); | |
292 | memcpy(ptr, data, pg_datalen); | |
293 | if (pg_sectorlen > pg_datalen) | |
294 | memset(ptr + pg_datalen, 0, pg_sectorlen - pg_datalen); | |
295 | kunmap_atomic(ptr); | |
296 | ||
297 | ret = bio_add_page(bio, page, pg_sectorlen, 0); | |
298 | if (ret != pg_sectorlen) { | |
299 | DMERR("Couldn't add page of inline data"); | |
300 | __free_page(page); | |
301 | goto error_bio; | |
302 | } | |
303 | ||
304 | datalen -= pg_datalen; | |
305 | data += pg_datalen; | |
306 | } | |
307 | submit_bio(bio); | |
308 | ||
309 | sector += bio_pages * PAGE_SECTORS; | |
310 | } | |
311 | return 0; | |
312 | error_bio: | |
313 | bio_free_pages(bio); | |
314 | bio_put(bio); | |
315 | error: | |
316 | put_io_block(lc); | |
317 | return -1; | |
318 | } | |
319 | ||
0e9cebe7 JB |
320 | static int log_one_block(struct log_writes_c *lc, |
321 | struct pending_block *block, sector_t sector) | |
322 | { | |
323 | struct bio *bio; | |
324 | struct log_write_entry entry; | |
e5a20660 | 325 | size_t metadatalen, ret; |
0e9cebe7 JB |
326 | int i; |
327 | ||
328 | entry.sector = cpu_to_le64(block->sector); | |
329 | entry.nr_sectors = cpu_to_le64(block->nr_sectors); | |
330 | entry.flags = cpu_to_le64(block->flags); | |
331 | entry.data_len = cpu_to_le64(block->datalen); | |
e5a20660 RZ |
332 | |
333 | metadatalen = (block->flags & LOG_MARK_FLAG) ? block->datalen : 0; | |
0e9cebe7 | 334 | if (write_metadata(lc, &entry, sizeof(entry), block->data, |
e5a20660 | 335 | metadatalen, sector)) { |
0e9cebe7 JB |
336 | free_pending_block(lc, block); |
337 | return -1; | |
338 | } | |
339 | ||
e5a20660 RZ |
340 | sector += dev_to_bio_sectors(lc, 1); |
341 | ||
342 | if (block->datalen && metadatalen == 0) { | |
343 | if (write_inline_data(lc, &entry, sizeof(entry), block->data, | |
344 | block->datalen, sector)) { | |
345 | free_pending_block(lc, block); | |
346 | return -1; | |
347 | } | |
348 | /* we don't support both inline data & bio data */ | |
349 | goto out; | |
350 | } | |
351 | ||
0e9cebe7 JB |
352 | if (!block->vec_cnt) |
353 | goto out; | |
0e9cebe7 | 354 | |
a5d60783 | 355 | atomic_inc(&lc->io_blocks); |
7efb3673 | 356 | bio = bio_alloc(GFP_KERNEL, min(block->vec_cnt, BIO_MAX_PAGES)); |
0e9cebe7 JB |
357 | if (!bio) { |
358 | DMERR("Couldn't alloc log bio"); | |
359 | goto error; | |
360 | } | |
0e9cebe7 JB |
361 | bio->bi_iter.bi_size = 0; |
362 | bio->bi_iter.bi_sector = sector; | |
74d46992 | 363 | bio_set_dev(bio, lc->logdev->bdev); |
0e9cebe7 JB |
364 | bio->bi_end_io = log_end_io; |
365 | bio->bi_private = lc; | |
e6047149 | 366 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
0e9cebe7 JB |
367 | |
368 | for (i = 0; i < block->vec_cnt; i++) { | |
369 | /* | |
370 | * The page offset is always 0 because we allocate a new page | |
371 | * for every bvec in the original bio for simplicity sake. | |
372 | */ | |
373 | ret = bio_add_page(bio, block->vecs[i].bv_page, | |
374 | block->vecs[i].bv_len, 0); | |
375 | if (ret != block->vecs[i].bv_len) { | |
376 | atomic_inc(&lc->io_blocks); | |
4e49ea4a | 377 | submit_bio(bio); |
7efb3673 | 378 | bio = bio_alloc(GFP_KERNEL, min(block->vec_cnt - i, BIO_MAX_PAGES)); |
0e9cebe7 JB |
379 | if (!bio) { |
380 | DMERR("Couldn't alloc log bio"); | |
381 | goto error; | |
382 | } | |
383 | bio->bi_iter.bi_size = 0; | |
384 | bio->bi_iter.bi_sector = sector; | |
74d46992 | 385 | bio_set_dev(bio, lc->logdev->bdev); |
0e9cebe7 JB |
386 | bio->bi_end_io = log_end_io; |
387 | bio->bi_private = lc; | |
e6047149 | 388 | bio_set_op_attrs(bio, REQ_OP_WRITE, 0); |
0e9cebe7 JB |
389 | |
390 | ret = bio_add_page(bio, block->vecs[i].bv_page, | |
391 | block->vecs[i].bv_len, 0); | |
392 | if (ret != block->vecs[i].bv_len) { | |
393 | DMERR("Couldn't add page on new bio?"); | |
394 | bio_put(bio); | |
395 | goto error; | |
396 | } | |
397 | } | |
398 | sector += block->vecs[i].bv_len >> SECTOR_SHIFT; | |
399 | } | |
4e49ea4a | 400 | submit_bio(bio); |
0e9cebe7 JB |
401 | out: |
402 | kfree(block->data); | |
403 | kfree(block); | |
404 | put_pending_block(lc); | |
405 | return 0; | |
406 | error: | |
407 | free_pending_block(lc, block); | |
408 | put_io_block(lc); | |
409 | return -1; | |
410 | } | |
411 | ||
412 | static int log_super(struct log_writes_c *lc) | |
413 | { | |
414 | struct log_write_super super; | |
415 | ||
416 | super.magic = cpu_to_le64(WRITE_LOG_MAGIC); | |
417 | super.version = cpu_to_le64(WRITE_LOG_VERSION); | |
418 | super.nr_entries = cpu_to_le64(lc->logged_entries); | |
419 | super.sectorsize = cpu_to_le32(lc->sectorsize); | |
420 | ||
421 | if (write_metadata(lc, &super, sizeof(super), NULL, 0, 0)) { | |
422 | DMERR("Couldn't write super"); | |
423 | return -1; | |
424 | } | |
425 | ||
426 | return 0; | |
427 | } | |
428 | ||
429 | static inline sector_t logdev_last_sector(struct log_writes_c *lc) | |
430 | { | |
431 | return i_size_read(lc->logdev->bdev->bd_inode) >> SECTOR_SHIFT; | |
432 | } | |
433 | ||
434 | static int log_writes_kthread(void *arg) | |
435 | { | |
436 | struct log_writes_c *lc = (struct log_writes_c *)arg; | |
437 | sector_t sector = 0; | |
438 | ||
439 | while (!kthread_should_stop()) { | |
440 | bool super = false; | |
441 | bool logging_enabled; | |
442 | struct pending_block *block = NULL; | |
443 | int ret; | |
444 | ||
445 | spin_lock_irq(&lc->blocks_lock); | |
446 | if (!list_empty(&lc->logging_blocks)) { | |
447 | block = list_first_entry(&lc->logging_blocks, | |
448 | struct pending_block, list); | |
449 | list_del_init(&block->list); | |
450 | if (!lc->logging_enabled) | |
451 | goto next; | |
452 | ||
453 | sector = lc->next_sector; | |
228bb5b2 JB |
454 | if (!(block->flags & LOG_DISCARD_FLAG)) |
455 | lc->next_sector += dev_to_bio_sectors(lc, block->nr_sectors); | |
456 | lc->next_sector += dev_to_bio_sectors(lc, 1); | |
0e9cebe7 JB |
457 | |
458 | /* | |
459 | * Apparently the size of the device may not be known | |
460 | * right away, so handle this properly. | |
461 | */ | |
462 | if (!lc->end_sector) | |
463 | lc->end_sector = logdev_last_sector(lc); | |
464 | if (lc->end_sector && | |
465 | lc->next_sector >= lc->end_sector) { | |
466 | DMERR("Ran out of space on the logdev"); | |
467 | lc->logging_enabled = false; | |
468 | goto next; | |
469 | } | |
470 | lc->logged_entries++; | |
471 | atomic_inc(&lc->io_blocks); | |
472 | ||
473 | super = (block->flags & (LOG_FUA_FLAG | LOG_MARK_FLAG)); | |
474 | if (super) | |
475 | atomic_inc(&lc->io_blocks); | |
476 | } | |
477 | next: | |
478 | logging_enabled = lc->logging_enabled; | |
479 | spin_unlock_irq(&lc->blocks_lock); | |
480 | if (block) { | |
481 | if (logging_enabled) { | |
482 | ret = log_one_block(lc, block, sector); | |
483 | if (!ret && super) | |
484 | ret = log_super(lc); | |
485 | if (ret) { | |
486 | spin_lock_irq(&lc->blocks_lock); | |
487 | lc->logging_enabled = false; | |
488 | spin_unlock_irq(&lc->blocks_lock); | |
489 | } | |
490 | } else | |
491 | free_pending_block(lc, block); | |
492 | continue; | |
493 | } | |
494 | ||
495 | if (!try_to_freeze()) { | |
496 | set_current_state(TASK_INTERRUPTIBLE); | |
497 | if (!kthread_should_stop() && | |
0c79c620 | 498 | list_empty(&lc->logging_blocks)) |
0e9cebe7 JB |
499 | schedule(); |
500 | __set_current_state(TASK_RUNNING); | |
501 | } | |
502 | } | |
503 | return 0; | |
504 | } | |
505 | ||
506 | /* | |
507 | * Construct a log-writes mapping: | |
508 | * log-writes <dev_path> <log_dev_path> | |
509 | */ | |
510 | static int log_writes_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
511 | { | |
512 | struct log_writes_c *lc; | |
513 | struct dm_arg_set as; | |
514 | const char *devname, *logdevname; | |
e80d1c80 | 515 | int ret; |
0e9cebe7 JB |
516 | |
517 | as.argc = argc; | |
518 | as.argv = argv; | |
519 | ||
520 | if (argc < 2) { | |
521 | ti->error = "Invalid argument count"; | |
522 | return -EINVAL; | |
523 | } | |
524 | ||
525 | lc = kzalloc(sizeof(struct log_writes_c), GFP_KERNEL); | |
526 | if (!lc) { | |
527 | ti->error = "Cannot allocate context"; | |
528 | return -ENOMEM; | |
529 | } | |
530 | spin_lock_init(&lc->blocks_lock); | |
531 | INIT_LIST_HEAD(&lc->unflushed_blocks); | |
532 | INIT_LIST_HEAD(&lc->logging_blocks); | |
533 | init_waitqueue_head(&lc->wait); | |
0e9cebe7 JB |
534 | atomic_set(&lc->io_blocks, 0); |
535 | atomic_set(&lc->pending_blocks, 0); | |
536 | ||
537 | devname = dm_shift_arg(&as); | |
e80d1c80 VG |
538 | ret = dm_get_device(ti, devname, dm_table_get_mode(ti->table), &lc->dev); |
539 | if (ret) { | |
0e9cebe7 JB |
540 | ti->error = "Device lookup failed"; |
541 | goto bad; | |
542 | } | |
543 | ||
544 | logdevname = dm_shift_arg(&as); | |
e80d1c80 VG |
545 | ret = dm_get_device(ti, logdevname, dm_table_get_mode(ti->table), |
546 | &lc->logdev); | |
547 | if (ret) { | |
0e9cebe7 JB |
548 | ti->error = "Log device lookup failed"; |
549 | dm_put_device(ti, lc->dev); | |
550 | goto bad; | |
551 | } | |
552 | ||
228bb5b2 JB |
553 | lc->sectorsize = bdev_logical_block_size(lc->dev->bdev); |
554 | lc->sectorshift = ilog2(lc->sectorsize); | |
0e9cebe7 | 555 | lc->log_kthread = kthread_run(log_writes_kthread, lc, "log-write"); |
91e630d9 VZ |
556 | if (IS_ERR(lc->log_kthread)) { |
557 | ret = PTR_ERR(lc->log_kthread); | |
0e9cebe7 JB |
558 | ti->error = "Couldn't alloc kthread"; |
559 | dm_put_device(ti, lc->dev); | |
560 | dm_put_device(ti, lc->logdev); | |
561 | goto bad; | |
562 | } | |
563 | ||
228bb5b2 JB |
564 | /* |
565 | * next_sector is in 512b sectors to correspond to what bi_sector expects. | |
566 | * The super starts at sector 0, and the next_sector is the next logical | |
567 | * one based on the sectorsize of the device. | |
568 | */ | |
569 | lc->next_sector = lc->sectorsize >> SECTOR_SHIFT; | |
0e9cebe7 JB |
570 | lc->logging_enabled = true; |
571 | lc->end_sector = logdev_last_sector(lc); | |
572 | lc->device_supports_discard = true; | |
573 | ||
574 | ti->num_flush_bios = 1; | |
575 | ti->flush_supported = true; | |
576 | ti->num_discard_bios = 1; | |
577 | ti->discards_supported = true; | |
30187e1d | 578 | ti->per_io_data_size = sizeof(struct per_bio_data); |
0e9cebe7 JB |
579 | ti->private = lc; |
580 | return 0; | |
581 | ||
582 | bad: | |
583 | kfree(lc); | |
e80d1c80 | 584 | return ret; |
0e9cebe7 JB |
585 | } |
586 | ||
587 | static int log_mark(struct log_writes_c *lc, char *data) | |
588 | { | |
589 | struct pending_block *block; | |
590 | size_t maxsize = lc->sectorsize - sizeof(struct log_write_entry); | |
591 | ||
592 | block = kzalloc(sizeof(struct pending_block), GFP_KERNEL); | |
593 | if (!block) { | |
594 | DMERR("Error allocating pending block"); | |
595 | return -ENOMEM; | |
596 | } | |
597 | ||
4b259fc4 | 598 | block->data = kstrndup(data, maxsize - 1, GFP_KERNEL); |
0e9cebe7 JB |
599 | if (!block->data) { |
600 | DMERR("Error copying mark data"); | |
601 | kfree(block); | |
602 | return -ENOMEM; | |
603 | } | |
604 | atomic_inc(&lc->pending_blocks); | |
605 | block->datalen = strlen(block->data); | |
606 | block->flags |= LOG_MARK_FLAG; | |
607 | spin_lock_irq(&lc->blocks_lock); | |
608 | list_add_tail(&block->list, &lc->logging_blocks); | |
609 | spin_unlock_irq(&lc->blocks_lock); | |
610 | wake_up_process(lc->log_kthread); | |
611 | return 0; | |
612 | } | |
613 | ||
614 | static void log_writes_dtr(struct dm_target *ti) | |
615 | { | |
616 | struct log_writes_c *lc = ti->private; | |
617 | ||
618 | spin_lock_irq(&lc->blocks_lock); | |
619 | list_splice_init(&lc->unflushed_blocks, &lc->logging_blocks); | |
620 | spin_unlock_irq(&lc->blocks_lock); | |
621 | ||
622 | /* | |
623 | * This is just nice to have since it'll update the super to include the | |
624 | * unflushed blocks, if it fails we don't really care. | |
625 | */ | |
626 | log_mark(lc, "dm-log-writes-end"); | |
627 | wake_up_process(lc->log_kthread); | |
628 | wait_event(lc->wait, !atomic_read(&lc->io_blocks) && | |
629 | !atomic_read(&lc->pending_blocks)); | |
630 | kthread_stop(lc->log_kthread); | |
631 | ||
632 | WARN_ON(!list_empty(&lc->logging_blocks)); | |
633 | WARN_ON(!list_empty(&lc->unflushed_blocks)); | |
634 | dm_put_device(ti, lc->dev); | |
635 | dm_put_device(ti, lc->logdev); | |
636 | kfree(lc); | |
637 | } | |
638 | ||
639 | static void normal_map_bio(struct dm_target *ti, struct bio *bio) | |
640 | { | |
641 | struct log_writes_c *lc = ti->private; | |
642 | ||
74d46992 | 643 | bio_set_dev(bio, lc->dev->bdev); |
0e9cebe7 JB |
644 | } |
645 | ||
646 | static int log_writes_map(struct dm_target *ti, struct bio *bio) | |
647 | { | |
648 | struct log_writes_c *lc = ti->private; | |
649 | struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data)); | |
650 | struct pending_block *block; | |
651 | struct bvec_iter iter; | |
652 | struct bio_vec bv; | |
653 | size_t alloc_size; | |
654 | int i = 0; | |
1eff9d32 JA |
655 | bool flush_bio = (bio->bi_opf & REQ_PREFLUSH); |
656 | bool fua_bio = (bio->bi_opf & REQ_FUA); | |
e6047149 | 657 | bool discard_bio = (bio_op(bio) == REQ_OP_DISCARD); |
e5c4cb9b | 658 | bool meta_bio = (bio->bi_opf & REQ_META); |
0e9cebe7 JB |
659 | |
660 | pb->block = NULL; | |
661 | ||
662 | /* Don't bother doing anything if logging has been disabled */ | |
663 | if (!lc->logging_enabled) | |
664 | goto map_bio; | |
665 | ||
666 | /* | |
667 | * Map reads as normal. | |
668 | */ | |
669 | if (bio_data_dir(bio) == READ) | |
670 | goto map_bio; | |
671 | ||
672 | /* No sectors and not a flush? Don't care */ | |
673 | if (!bio_sectors(bio) && !flush_bio) | |
674 | goto map_bio; | |
675 | ||
676 | /* | |
677 | * Discards will have bi_size set but there's no actual data, so just | |
678 | * allocate the size of the pending block. | |
679 | */ | |
680 | if (discard_bio) | |
681 | alloc_size = sizeof(struct pending_block); | |
682 | else | |
683 | alloc_size = sizeof(struct pending_block) + sizeof(struct bio_vec) * bio_segments(bio); | |
684 | ||
685 | block = kzalloc(alloc_size, GFP_NOIO); | |
686 | if (!block) { | |
687 | DMERR("Error allocating pending block"); | |
688 | spin_lock_irq(&lc->blocks_lock); | |
689 | lc->logging_enabled = false; | |
690 | spin_unlock_irq(&lc->blocks_lock); | |
846785e6 | 691 | return DM_MAPIO_KILL; |
0e9cebe7 JB |
692 | } |
693 | INIT_LIST_HEAD(&block->list); | |
694 | pb->block = block; | |
695 | atomic_inc(&lc->pending_blocks); | |
696 | ||
697 | if (flush_bio) | |
698 | block->flags |= LOG_FLUSH_FLAG; | |
699 | if (fua_bio) | |
700 | block->flags |= LOG_FUA_FLAG; | |
701 | if (discard_bio) | |
702 | block->flags |= LOG_DISCARD_FLAG; | |
e5c4cb9b QW |
703 | if (meta_bio) |
704 | block->flags |= LOG_METADATA_FLAG; | |
0e9cebe7 | 705 | |
228bb5b2 JB |
706 | block->sector = bio_to_dev_sectors(lc, bio->bi_iter.bi_sector); |
707 | block->nr_sectors = bio_to_dev_sectors(lc, bio_sectors(bio)); | |
0e9cebe7 JB |
708 | |
709 | /* We don't need the data, just submit */ | |
710 | if (discard_bio) { | |
711 | WARN_ON(flush_bio || fua_bio); | |
712 | if (lc->device_supports_discard) | |
713 | goto map_bio; | |
4246a0b6 | 714 | bio_endio(bio); |
0e9cebe7 JB |
715 | return DM_MAPIO_SUBMITTED; |
716 | } | |
717 | ||
718 | /* Flush bio, splice the unflushed blocks onto this list and submit */ | |
719 | if (flush_bio && !bio_sectors(bio)) { | |
720 | spin_lock_irq(&lc->blocks_lock); | |
721 | list_splice_init(&lc->unflushed_blocks, &block->list); | |
722 | spin_unlock_irq(&lc->blocks_lock); | |
723 | goto map_bio; | |
724 | } | |
725 | ||
726 | /* | |
727 | * We will write this bio somewhere else way later so we need to copy | |
728 | * the actual contents into new pages so we know the data will always be | |
729 | * there. | |
730 | * | |
731 | * We do this because this could be a bio from O_DIRECT in which case we | |
732 | * can't just hold onto the page until some later point, we have to | |
733 | * manually copy the contents. | |
734 | */ | |
735 | bio_for_each_segment(bv, bio, iter) { | |
736 | struct page *page; | |
737 | void *src, *dst; | |
738 | ||
739 | page = alloc_page(GFP_NOIO); | |
740 | if (!page) { | |
741 | DMERR("Error allocing page"); | |
742 | free_pending_block(lc, block); | |
743 | spin_lock_irq(&lc->blocks_lock); | |
744 | lc->logging_enabled = false; | |
745 | spin_unlock_irq(&lc->blocks_lock); | |
846785e6 | 746 | return DM_MAPIO_KILL; |
0e9cebe7 JB |
747 | } |
748 | ||
749 | src = kmap_atomic(bv.bv_page); | |
750 | dst = kmap_atomic(page); | |
751 | memcpy(dst, src + bv.bv_offset, bv.bv_len); | |
752 | kunmap_atomic(dst); | |
753 | kunmap_atomic(src); | |
754 | block->vecs[i].bv_page = page; | |
755 | block->vecs[i].bv_len = bv.bv_len; | |
756 | block->vec_cnt++; | |
757 | i++; | |
758 | } | |
759 | ||
760 | /* Had a flush with data in it, weird */ | |
761 | if (flush_bio) { | |
762 | spin_lock_irq(&lc->blocks_lock); | |
763 | list_splice_init(&lc->unflushed_blocks, &block->list); | |
764 | spin_unlock_irq(&lc->blocks_lock); | |
765 | } | |
766 | map_bio: | |
767 | normal_map_bio(ti, bio); | |
768 | return DM_MAPIO_REMAPPED; | |
769 | } | |
770 | ||
4e4cbee9 CH |
771 | static int normal_end_io(struct dm_target *ti, struct bio *bio, |
772 | blk_status_t *error) | |
0e9cebe7 JB |
773 | { |
774 | struct log_writes_c *lc = ti->private; | |
775 | struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data)); | |
776 | ||
777 | if (bio_data_dir(bio) == WRITE && pb->block) { | |
778 | struct pending_block *block = pb->block; | |
779 | unsigned long flags; | |
780 | ||
781 | spin_lock_irqsave(&lc->blocks_lock, flags); | |
782 | if (block->flags & LOG_FLUSH_FLAG) { | |
783 | list_splice_tail_init(&block->list, &lc->logging_blocks); | |
784 | list_add_tail(&block->list, &lc->logging_blocks); | |
785 | wake_up_process(lc->log_kthread); | |
786 | } else if (block->flags & LOG_FUA_FLAG) { | |
787 | list_add_tail(&block->list, &lc->logging_blocks); | |
788 | wake_up_process(lc->log_kthread); | |
789 | } else | |
790 | list_add_tail(&block->list, &lc->unflushed_blocks); | |
791 | spin_unlock_irqrestore(&lc->blocks_lock, flags); | |
792 | } | |
793 | ||
1be56909 | 794 | return DM_ENDIO_DONE; |
0e9cebe7 JB |
795 | } |
796 | ||
797 | /* | |
798 | * INFO format: <logged entries> <highest allocated sector> | |
799 | */ | |
800 | static void log_writes_status(struct dm_target *ti, status_type_t type, | |
801 | unsigned status_flags, char *result, | |
802 | unsigned maxlen) | |
803 | { | |
804 | unsigned sz = 0; | |
805 | struct log_writes_c *lc = ti->private; | |
806 | ||
807 | switch (type) { | |
808 | case STATUSTYPE_INFO: | |
809 | DMEMIT("%llu %llu", lc->logged_entries, | |
810 | (unsigned long long)lc->next_sector - 1); | |
811 | if (!lc->logging_enabled) | |
812 | DMEMIT(" logging_disabled"); | |
813 | break; | |
814 | ||
815 | case STATUSTYPE_TABLE: | |
816 | DMEMIT("%s %s", lc->dev->name, lc->logdev->name); | |
817 | break; | |
818 | } | |
819 | } | |
820 | ||
e56f81e0 | 821 | static int log_writes_prepare_ioctl(struct dm_target *ti, |
5bd5e8d8 | 822 | struct block_device **bdev) |
0e9cebe7 JB |
823 | { |
824 | struct log_writes_c *lc = ti->private; | |
825 | struct dm_dev *dev = lc->dev; | |
0e9cebe7 | 826 | |
e56f81e0 | 827 | *bdev = dev->bdev; |
0e9cebe7 JB |
828 | /* |
829 | * Only pass ioctls through if the device sizes match exactly. | |
830 | */ | |
831 | if (ti->len != i_size_read(dev->bdev->bd_inode) >> SECTOR_SHIFT) | |
e56f81e0 CH |
832 | return 1; |
833 | return 0; | |
0e9cebe7 JB |
834 | } |
835 | ||
0e9cebe7 JB |
836 | static int log_writes_iterate_devices(struct dm_target *ti, |
837 | iterate_devices_callout_fn fn, | |
838 | void *data) | |
839 | { | |
840 | struct log_writes_c *lc = ti->private; | |
841 | ||
842 | return fn(ti, lc->dev, 0, ti->len, data); | |
843 | } | |
844 | ||
845 | /* | |
846 | * Messages supported: | |
847 | * mark <mark data> - specify the marked data. | |
848 | */ | |
1eb5fa84 MS |
849 | static int log_writes_message(struct dm_target *ti, unsigned argc, char **argv, |
850 | char *result, unsigned maxlen) | |
0e9cebe7 JB |
851 | { |
852 | int r = -EINVAL; | |
853 | struct log_writes_c *lc = ti->private; | |
854 | ||
855 | if (argc != 2) { | |
856 | DMWARN("Invalid log-writes message arguments, expect 2 arguments, got %d", argc); | |
857 | return r; | |
858 | } | |
859 | ||
860 | if (!strcasecmp(argv[0], "mark")) | |
861 | r = log_mark(lc, argv[1]); | |
862 | else | |
863 | DMWARN("Unrecognised log writes target message received: %s", argv[0]); | |
864 | ||
865 | return r; | |
866 | } | |
867 | ||
868 | static void log_writes_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
869 | { | |
870 | struct log_writes_c *lc = ti->private; | |
871 | struct request_queue *q = bdev_get_queue(lc->dev->bdev); | |
872 | ||
873 | if (!q || !blk_queue_discard(q)) { | |
874 | lc->device_supports_discard = false; | |
228bb5b2 | 875 | limits->discard_granularity = lc->sectorsize; |
0e9cebe7 JB |
876 | limits->max_discard_sectors = (UINT_MAX >> SECTOR_SHIFT); |
877 | } | |
228bb5b2 JB |
878 | limits->logical_block_size = bdev_logical_block_size(lc->dev->bdev); |
879 | limits->physical_block_size = bdev_physical_block_size(lc->dev->bdev); | |
880 | limits->io_min = limits->physical_block_size; | |
0e9cebe7 JB |
881 | } |
882 | ||
976431b0 DW |
883 | #if IS_ENABLED(CONFIG_DAX_DRIVER) |
884 | static int log_dax(struct log_writes_c *lc, sector_t sector, size_t bytes, | |
885 | struct iov_iter *i) | |
886 | { | |
887 | struct pending_block *block; | |
888 | ||
889 | if (!bytes) | |
890 | return 0; | |
891 | ||
892 | block = kzalloc(sizeof(struct pending_block), GFP_KERNEL); | |
893 | if (!block) { | |
894 | DMERR("Error allocating dax pending block"); | |
895 | return -ENOMEM; | |
896 | } | |
897 | ||
898 | block->data = kzalloc(bytes, GFP_KERNEL); | |
899 | if (!block->data) { | |
900 | DMERR("Error allocating dax data space"); | |
901 | kfree(block); | |
902 | return -ENOMEM; | |
903 | } | |
904 | ||
905 | /* write data provided via the iterator */ | |
906 | if (!copy_from_iter(block->data, bytes, i)) { | |
907 | DMERR("Error copying dax data"); | |
908 | kfree(block->data); | |
909 | kfree(block); | |
910 | return -EIO; | |
911 | } | |
912 | ||
913 | /* rewind the iterator so that the block driver can use it */ | |
914 | iov_iter_revert(i, bytes); | |
915 | ||
916 | block->datalen = bytes; | |
917 | block->sector = bio_to_dev_sectors(lc, sector); | |
918 | block->nr_sectors = ALIGN(bytes, lc->sectorsize) >> lc->sectorshift; | |
919 | ||
920 | atomic_inc(&lc->pending_blocks); | |
921 | spin_lock_irq(&lc->blocks_lock); | |
922 | list_add_tail(&block->list, &lc->unflushed_blocks); | |
923 | spin_unlock_irq(&lc->blocks_lock); | |
924 | wake_up_process(lc->log_kthread); | |
925 | ||
926 | return 0; | |
927 | } | |
928 | ||
98d82f48 RZ |
929 | static long log_writes_dax_direct_access(struct dm_target *ti, pgoff_t pgoff, |
930 | long nr_pages, void **kaddr, pfn_t *pfn) | |
931 | { | |
932 | struct log_writes_c *lc = ti->private; | |
933 | sector_t sector = pgoff * PAGE_SECTORS; | |
934 | int ret; | |
935 | ||
936 | ret = bdev_dax_pgoff(lc->dev->bdev, sector, nr_pages * PAGE_SIZE, &pgoff); | |
937 | if (ret) | |
938 | return ret; | |
939 | return dax_direct_access(lc->dev->dax_dev, pgoff, nr_pages, kaddr, pfn); | |
940 | } | |
941 | ||
942 | static size_t log_writes_dax_copy_from_iter(struct dm_target *ti, | |
943 | pgoff_t pgoff, void *addr, size_t bytes, | |
944 | struct iov_iter *i) | |
945 | { | |
946 | struct log_writes_c *lc = ti->private; | |
947 | sector_t sector = pgoff * PAGE_SECTORS; | |
948 | int err; | |
949 | ||
950 | if (bdev_dax_pgoff(lc->dev->bdev, sector, ALIGN(bytes, PAGE_SIZE), &pgoff)) | |
951 | return 0; | |
952 | ||
953 | /* Don't bother doing anything if logging has been disabled */ | |
954 | if (!lc->logging_enabled) | |
955 | goto dax_copy; | |
956 | ||
957 | err = log_dax(lc, sector, bytes, i); | |
958 | if (err) { | |
959 | DMWARN("Error %d logging DAX write", err); | |
960 | return 0; | |
961 | } | |
962 | dax_copy: | |
963 | return dax_copy_from_iter(lc->dev->dax_dev, pgoff, addr, bytes, i); | |
964 | } | |
b3a9a0c3 DW |
965 | |
966 | static size_t log_writes_dax_copy_to_iter(struct dm_target *ti, | |
967 | pgoff_t pgoff, void *addr, size_t bytes, | |
968 | struct iov_iter *i) | |
969 | { | |
970 | struct log_writes_c *lc = ti->private; | |
971 | sector_t sector = pgoff * PAGE_SECTORS; | |
972 | ||
973 | if (bdev_dax_pgoff(lc->dev->bdev, sector, ALIGN(bytes, PAGE_SIZE), &pgoff)) | |
974 | return 0; | |
975 | return dax_copy_to_iter(lc->dev->dax_dev, pgoff, addr, bytes, i); | |
976 | } | |
977 | ||
976431b0 DW |
978 | #else |
979 | #define log_writes_dax_direct_access NULL | |
980 | #define log_writes_dax_copy_from_iter NULL | |
b3a9a0c3 | 981 | #define log_writes_dax_copy_to_iter NULL |
976431b0 | 982 | #endif |
98d82f48 | 983 | |
0e9cebe7 JB |
984 | static struct target_type log_writes_target = { |
985 | .name = "log-writes", | |
98d82f48 | 986 | .version = {1, 1, 0}, |
0e9cebe7 JB |
987 | .module = THIS_MODULE, |
988 | .ctr = log_writes_ctr, | |
989 | .dtr = log_writes_dtr, | |
990 | .map = log_writes_map, | |
991 | .end_io = normal_end_io, | |
992 | .status = log_writes_status, | |
e56f81e0 | 993 | .prepare_ioctl = log_writes_prepare_ioctl, |
0e9cebe7 JB |
994 | .message = log_writes_message, |
995 | .iterate_devices = log_writes_iterate_devices, | |
996 | .io_hints = log_writes_io_hints, | |
98d82f48 RZ |
997 | .direct_access = log_writes_dax_direct_access, |
998 | .dax_copy_from_iter = log_writes_dax_copy_from_iter, | |
b3a9a0c3 | 999 | .dax_copy_to_iter = log_writes_dax_copy_to_iter, |
0e9cebe7 JB |
1000 | }; |
1001 | ||
1002 | static int __init dm_log_writes_init(void) | |
1003 | { | |
1004 | int r = dm_register_target(&log_writes_target); | |
1005 | ||
1006 | if (r < 0) | |
1007 | DMERR("register failed %d", r); | |
1008 | ||
1009 | return r; | |
1010 | } | |
1011 | ||
1012 | static void __exit dm_log_writes_exit(void) | |
1013 | { | |
1014 | dm_unregister_target(&log_writes_target); | |
1015 | } | |
1016 | ||
1017 | module_init(dm_log_writes_init); | |
1018 | module_exit(dm_log_writes_exit); | |
1019 | ||
1020 | MODULE_DESCRIPTION(DM_NAME " log writes target"); | |
1021 | MODULE_AUTHOR("Josef Bacik <jbacik@fb.com>"); | |
1022 | MODULE_LICENSE("GPL"); |