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 | * | |
7537dad7 | 43 | * c,a,b,flush,fuad,<other writes>,<next flush> |
0e9cebe7 JB |
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 | |
211ad4b7 | 63 | #define WRITE_LOG_SUPER_SECTOR 0 |
0e9cebe7 JB |
64 | |
65 | /* | |
66 | * The disk format for this is braindead simple. | |
67 | * | |
68 | * At byte 0 we have our super, followed by the following sequence for | |
69 | * nr_entries: | |
70 | * | |
71 | * [ 1 sector ][ entry->nr_sectors ] | |
72 | * [log_write_entry][ data written ] | |
73 | * | |
74 | * The log_write_entry takes up a full sector so we can have arbitrary length | |
75 | * marks and it leaves us room for extra content in the future. | |
76 | */ | |
77 | ||
78 | /* | |
79 | * Basic info about the log for userspace. | |
80 | */ | |
81 | struct log_write_super { | |
82 | __le64 magic; | |
83 | __le64 version; | |
84 | __le64 nr_entries; | |
85 | __le32 sectorsize; | |
86 | }; | |
87 | ||
88 | /* | |
89 | * sector - the sector we wrote. | |
90 | * nr_sectors - the number of sectors we wrote. | |
91 | * flags - flags for this log entry. | |
92 | * data_len - the size of the data in this log entry, this is for private log | |
93 | * entry stuff, the MARK data provided by userspace for example. | |
94 | */ | |
95 | struct log_write_entry { | |
96 | __le64 sector; | |
97 | __le64 nr_sectors; | |
98 | __le64 flags; | |
99 | __le64 data_len; | |
100 | }; | |
101 | ||
102 | struct log_writes_c { | |
103 | struct dm_dev *dev; | |
104 | struct dm_dev *logdev; | |
105 | u64 logged_entries; | |
106 | u32 sectorsize; | |
228bb5b2 | 107 | u32 sectorshift; |
0e9cebe7 JB |
108 | atomic_t io_blocks; |
109 | atomic_t pending_blocks; | |
110 | sector_t next_sector; | |
111 | sector_t end_sector; | |
112 | bool logging_enabled; | |
113 | bool device_supports_discard; | |
114 | spinlock_t blocks_lock; | |
115 | struct list_head unflushed_blocks; | |
116 | struct list_head logging_blocks; | |
117 | wait_queue_head_t wait; | |
118 | struct task_struct *log_kthread; | |
211ad4b7 | 119 | struct completion super_done; |
0e9cebe7 JB |
120 | }; |
121 | ||
122 | struct pending_block { | |
123 | int vec_cnt; | |
124 | u64 flags; | |
125 | sector_t sector; | |
126 | sector_t nr_sectors; | |
127 | char *data; | |
128 | u32 datalen; | |
129 | struct list_head list; | |
b18ae8dd | 130 | struct bio_vec vecs[]; |
0e9cebe7 JB |
131 | }; |
132 | ||
133 | struct per_bio_data { | |
134 | struct pending_block *block; | |
135 | }; | |
136 | ||
228bb5b2 JB |
137 | static inline sector_t bio_to_dev_sectors(struct log_writes_c *lc, |
138 | sector_t sectors) | |
139 | { | |
140 | return sectors >> (lc->sectorshift - SECTOR_SHIFT); | |
141 | } | |
142 | ||
143 | static inline sector_t dev_to_bio_sectors(struct log_writes_c *lc, | |
144 | sector_t sectors) | |
145 | { | |
146 | return sectors << (lc->sectorshift - SECTOR_SHIFT); | |
147 | } | |
148 | ||
0e9cebe7 JB |
149 | static void put_pending_block(struct log_writes_c *lc) |
150 | { | |
151 | if (atomic_dec_and_test(&lc->pending_blocks)) { | |
152 | smp_mb__after_atomic(); | |
153 | if (waitqueue_active(&lc->wait)) | |
154 | wake_up(&lc->wait); | |
155 | } | |
156 | } | |
157 | ||
158 | static void put_io_block(struct log_writes_c *lc) | |
159 | { | |
160 | if (atomic_dec_and_test(&lc->io_blocks)) { | |
161 | smp_mb__after_atomic(); | |
162 | if (waitqueue_active(&lc->wait)) | |
163 | wake_up(&lc->wait); | |
164 | } | |
165 | } | |
166 | ||
4246a0b6 | 167 | static void log_end_io(struct bio *bio) |
0e9cebe7 JB |
168 | { |
169 | struct log_writes_c *lc = bio->bi_private; | |
0e9cebe7 | 170 | |
4e4cbee9 | 171 | if (bio->bi_status) { |
0e9cebe7 JB |
172 | unsigned long flags; |
173 | ||
4e4cbee9 | 174 | DMERR("Error writing log block, error=%d", bio->bi_status); |
0e9cebe7 JB |
175 | spin_lock_irqsave(&lc->blocks_lock, flags); |
176 | lc->logging_enabled = false; | |
177 | spin_unlock_irqrestore(&lc->blocks_lock, flags); | |
178 | } | |
179 | ||
491221f8 | 180 | bio_free_pages(bio); |
0e9cebe7 JB |
181 | put_io_block(lc); |
182 | bio_put(bio); | |
183 | } | |
184 | ||
211ad4b7 | 185 | static void log_end_super(struct bio *bio) |
186 | { | |
187 | struct log_writes_c *lc = bio->bi_private; | |
188 | ||
189 | complete(&lc->super_done); | |
190 | log_end_io(bio); | |
191 | } | |
192 | ||
0e9cebe7 JB |
193 | /* |
194 | * Meant to be called if there is an error, it will free all the pages | |
195 | * associated with the block. | |
196 | */ | |
197 | static void free_pending_block(struct log_writes_c *lc, | |
198 | struct pending_block *block) | |
199 | { | |
200 | int i; | |
201 | ||
202 | for (i = 0; i < block->vec_cnt; i++) { | |
203 | if (block->vecs[i].bv_page) | |
204 | __free_page(block->vecs[i].bv_page); | |
205 | } | |
206 | kfree(block->data); | |
207 | kfree(block); | |
208 | put_pending_block(lc); | |
209 | } | |
210 | ||
211 | static int write_metadata(struct log_writes_c *lc, void *entry, | |
212 | size_t entrylen, void *data, size_t datalen, | |
213 | sector_t sector) | |
214 | { | |
215 | struct bio *bio; | |
216 | struct page *page; | |
217 | void *ptr; | |
218 | size_t ret; | |
219 | ||
07888c66 | 220 | bio = bio_alloc(lc->logdev->bdev, 1, REQ_OP_WRITE, GFP_KERNEL); |
0e9cebe7 JB |
221 | bio->bi_iter.bi_size = 0; |
222 | bio->bi_iter.bi_sector = sector; | |
211ad4b7 | 223 | bio->bi_end_io = (sector == WRITE_LOG_SUPER_SECTOR) ? |
224 | log_end_super : log_end_io; | |
0e9cebe7 | 225 | bio->bi_private = lc; |
0e9cebe7 JB |
226 | |
227 | page = alloc_page(GFP_KERNEL); | |
228 | if (!page) { | |
229 | DMERR("Couldn't alloc log page"); | |
230 | bio_put(bio); | |
231 | goto error; | |
232 | } | |
233 | ||
234 | ptr = kmap_atomic(page); | |
235 | memcpy(ptr, entry, entrylen); | |
236 | if (datalen) | |
237 | memcpy(ptr + entrylen, data, datalen); | |
238 | memset(ptr + entrylen + datalen, 0, | |
239 | lc->sectorsize - entrylen - datalen); | |
240 | kunmap_atomic(ptr); | |
241 | ||
242 | ret = bio_add_page(bio, page, lc->sectorsize, 0); | |
243 | if (ret != lc->sectorsize) { | |
244 | DMERR("Couldn't add page to the log block"); | |
245 | goto error_bio; | |
246 | } | |
4e49ea4a | 247 | submit_bio(bio); |
0e9cebe7 JB |
248 | return 0; |
249 | error_bio: | |
250 | bio_put(bio); | |
251 | __free_page(page); | |
252 | error: | |
253 | put_io_block(lc); | |
254 | return -1; | |
255 | } | |
256 | ||
e5a20660 RZ |
257 | static int write_inline_data(struct log_writes_c *lc, void *entry, |
258 | size_t entrylen, void *data, size_t datalen, | |
259 | sector_t sector) | |
260 | { | |
5f7136db | 261 | int bio_pages, pg_datalen, pg_sectorlen, i; |
e5a20660 RZ |
262 | struct page *page; |
263 | struct bio *bio; | |
264 | size_t ret; | |
265 | void *ptr; | |
266 | ||
267 | while (datalen) { | |
5f7136db | 268 | bio_pages = bio_max_segs(DIV_ROUND_UP(datalen, PAGE_SIZE)); |
e5a20660 RZ |
269 | |
270 | atomic_inc(&lc->io_blocks); | |
271 | ||
07888c66 CH |
272 | bio = bio_alloc(lc->logdev->bdev, bio_pages, REQ_OP_WRITE, |
273 | GFP_KERNEL); | |
e5a20660 RZ |
274 | bio->bi_iter.bi_size = 0; |
275 | bio->bi_iter.bi_sector = sector; | |
e5a20660 RZ |
276 | bio->bi_end_io = log_end_io; |
277 | bio->bi_private = lc; | |
e5a20660 RZ |
278 | |
279 | for (i = 0; i < bio_pages; i++) { | |
280 | pg_datalen = min_t(int, datalen, PAGE_SIZE); | |
281 | pg_sectorlen = ALIGN(pg_datalen, lc->sectorsize); | |
282 | ||
283 | page = alloc_page(GFP_KERNEL); | |
284 | if (!page) { | |
285 | DMERR("Couldn't alloc inline data page"); | |
286 | goto error_bio; | |
287 | } | |
288 | ||
289 | ptr = kmap_atomic(page); | |
290 | memcpy(ptr, data, pg_datalen); | |
291 | if (pg_sectorlen > pg_datalen) | |
292 | memset(ptr + pg_datalen, 0, pg_sectorlen - pg_datalen); | |
293 | kunmap_atomic(ptr); | |
294 | ||
295 | ret = bio_add_page(bio, page, pg_sectorlen, 0); | |
296 | if (ret != pg_sectorlen) { | |
297 | DMERR("Couldn't add page of inline data"); | |
298 | __free_page(page); | |
299 | goto error_bio; | |
300 | } | |
301 | ||
302 | datalen -= pg_datalen; | |
303 | data += pg_datalen; | |
304 | } | |
305 | submit_bio(bio); | |
306 | ||
307 | sector += bio_pages * PAGE_SECTORS; | |
308 | } | |
309 | return 0; | |
310 | error_bio: | |
311 | bio_free_pages(bio); | |
312 | bio_put(bio); | |
e5a20660 RZ |
313 | put_io_block(lc); |
314 | return -1; | |
315 | } | |
316 | ||
0e9cebe7 JB |
317 | static int log_one_block(struct log_writes_c *lc, |
318 | struct pending_block *block, sector_t sector) | |
319 | { | |
320 | struct bio *bio; | |
321 | struct log_write_entry entry; | |
e5a20660 | 322 | size_t metadatalen, ret; |
0e9cebe7 JB |
323 | int i; |
324 | ||
325 | entry.sector = cpu_to_le64(block->sector); | |
326 | entry.nr_sectors = cpu_to_le64(block->nr_sectors); | |
327 | entry.flags = cpu_to_le64(block->flags); | |
328 | entry.data_len = cpu_to_le64(block->datalen); | |
e5a20660 RZ |
329 | |
330 | metadatalen = (block->flags & LOG_MARK_FLAG) ? block->datalen : 0; | |
0e9cebe7 | 331 | if (write_metadata(lc, &entry, sizeof(entry), block->data, |
e5a20660 | 332 | metadatalen, sector)) { |
0e9cebe7 JB |
333 | free_pending_block(lc, block); |
334 | return -1; | |
335 | } | |
336 | ||
e5a20660 RZ |
337 | sector += dev_to_bio_sectors(lc, 1); |
338 | ||
339 | if (block->datalen && metadatalen == 0) { | |
340 | if (write_inline_data(lc, &entry, sizeof(entry), block->data, | |
341 | block->datalen, sector)) { | |
342 | free_pending_block(lc, block); | |
343 | return -1; | |
344 | } | |
345 | /* we don't support both inline data & bio data */ | |
346 | goto out; | |
347 | } | |
348 | ||
0e9cebe7 JB |
349 | if (!block->vec_cnt) |
350 | goto out; | |
0e9cebe7 | 351 | |
a5d60783 | 352 | atomic_inc(&lc->io_blocks); |
07888c66 CH |
353 | bio = bio_alloc(lc->logdev->bdev, bio_max_segs(block->vec_cnt), |
354 | REQ_OP_WRITE, GFP_KERNEL); | |
0e9cebe7 JB |
355 | bio->bi_iter.bi_size = 0; |
356 | bio->bi_iter.bi_sector = sector; | |
0e9cebe7 JB |
357 | bio->bi_end_io = log_end_io; |
358 | bio->bi_private = lc; | |
0e9cebe7 JB |
359 | |
360 | for (i = 0; i < block->vec_cnt; i++) { | |
361 | /* | |
362 | * The page offset is always 0 because we allocate a new page | |
363 | * for every bvec in the original bio for simplicity sake. | |
364 | */ | |
365 | ret = bio_add_page(bio, block->vecs[i].bv_page, | |
366 | block->vecs[i].bv_len, 0); | |
367 | if (ret != block->vecs[i].bv_len) { | |
368 | atomic_inc(&lc->io_blocks); | |
4e49ea4a | 369 | submit_bio(bio); |
07888c66 CH |
370 | bio = bio_alloc(lc->logdev->bdev, |
371 | bio_max_segs(block->vec_cnt - i), | |
372 | REQ_OP_WRITE, GFP_KERNEL); | |
0e9cebe7 JB |
373 | bio->bi_iter.bi_size = 0; |
374 | bio->bi_iter.bi_sector = sector; | |
0e9cebe7 JB |
375 | bio->bi_end_io = log_end_io; |
376 | bio->bi_private = lc; | |
0e9cebe7 JB |
377 | |
378 | ret = bio_add_page(bio, block->vecs[i].bv_page, | |
379 | block->vecs[i].bv_len, 0); | |
380 | if (ret != block->vecs[i].bv_len) { | |
381 | DMERR("Couldn't add page on new bio?"); | |
382 | bio_put(bio); | |
383 | goto error; | |
384 | } | |
385 | } | |
386 | sector += block->vecs[i].bv_len >> SECTOR_SHIFT; | |
387 | } | |
4e49ea4a | 388 | submit_bio(bio); |
0e9cebe7 JB |
389 | out: |
390 | kfree(block->data); | |
391 | kfree(block); | |
392 | put_pending_block(lc); | |
393 | return 0; | |
394 | error: | |
395 | free_pending_block(lc, block); | |
396 | put_io_block(lc); | |
397 | return -1; | |
398 | } | |
399 | ||
400 | static int log_super(struct log_writes_c *lc) | |
401 | { | |
402 | struct log_write_super super; | |
403 | ||
404 | super.magic = cpu_to_le64(WRITE_LOG_MAGIC); | |
405 | super.version = cpu_to_le64(WRITE_LOG_VERSION); | |
406 | super.nr_entries = cpu_to_le64(lc->logged_entries); | |
407 | super.sectorsize = cpu_to_le32(lc->sectorsize); | |
408 | ||
211ad4b7 | 409 | if (write_metadata(lc, &super, sizeof(super), NULL, 0, |
410 | WRITE_LOG_SUPER_SECTOR)) { | |
0e9cebe7 JB |
411 | DMERR("Couldn't write super"); |
412 | return -1; | |
413 | } | |
414 | ||
211ad4b7 | 415 | /* |
416 | * Super sector should be writen in-order, otherwise the | |
417 | * nr_entries could be rewritten incorrectly by an old bio. | |
418 | */ | |
419 | wait_for_completion_io(&lc->super_done); | |
420 | ||
0e9cebe7 JB |
421 | return 0; |
422 | } | |
423 | ||
424 | static inline sector_t logdev_last_sector(struct log_writes_c *lc) | |
425 | { | |
6dcbb52c | 426 | return bdev_nr_sectors(lc->logdev->bdev); |
0e9cebe7 JB |
427 | } |
428 | ||
429 | static int log_writes_kthread(void *arg) | |
430 | { | |
431 | struct log_writes_c *lc = (struct log_writes_c *)arg; | |
432 | sector_t sector = 0; | |
433 | ||
434 | while (!kthread_should_stop()) { | |
435 | bool super = false; | |
436 | bool logging_enabled; | |
437 | struct pending_block *block = NULL; | |
438 | int ret; | |
439 | ||
440 | spin_lock_irq(&lc->blocks_lock); | |
441 | if (!list_empty(&lc->logging_blocks)) { | |
442 | block = list_first_entry(&lc->logging_blocks, | |
443 | struct pending_block, list); | |
444 | list_del_init(&block->list); | |
445 | if (!lc->logging_enabled) | |
446 | goto next; | |
447 | ||
448 | sector = lc->next_sector; | |
228bb5b2 JB |
449 | if (!(block->flags & LOG_DISCARD_FLAG)) |
450 | lc->next_sector += dev_to_bio_sectors(lc, block->nr_sectors); | |
451 | lc->next_sector += dev_to_bio_sectors(lc, 1); | |
0e9cebe7 JB |
452 | |
453 | /* | |
454 | * Apparently the size of the device may not be known | |
455 | * right away, so handle this properly. | |
456 | */ | |
457 | if (!lc->end_sector) | |
458 | lc->end_sector = logdev_last_sector(lc); | |
459 | if (lc->end_sector && | |
460 | lc->next_sector >= lc->end_sector) { | |
461 | DMERR("Ran out of space on the logdev"); | |
462 | lc->logging_enabled = false; | |
463 | goto next; | |
464 | } | |
465 | lc->logged_entries++; | |
466 | atomic_inc(&lc->io_blocks); | |
467 | ||
468 | super = (block->flags & (LOG_FUA_FLAG | LOG_MARK_FLAG)); | |
469 | if (super) | |
470 | atomic_inc(&lc->io_blocks); | |
471 | } | |
472 | next: | |
473 | logging_enabled = lc->logging_enabled; | |
474 | spin_unlock_irq(&lc->blocks_lock); | |
475 | if (block) { | |
476 | if (logging_enabled) { | |
477 | ret = log_one_block(lc, block, sector); | |
478 | if (!ret && super) | |
479 | ret = log_super(lc); | |
480 | if (ret) { | |
481 | spin_lock_irq(&lc->blocks_lock); | |
482 | lc->logging_enabled = false; | |
483 | spin_unlock_irq(&lc->blocks_lock); | |
484 | } | |
485 | } else | |
486 | free_pending_block(lc, block); | |
487 | continue; | |
488 | } | |
489 | ||
490 | if (!try_to_freeze()) { | |
491 | set_current_state(TASK_INTERRUPTIBLE); | |
492 | if (!kthread_should_stop() && | |
0c79c620 | 493 | list_empty(&lc->logging_blocks)) |
0e9cebe7 JB |
494 | schedule(); |
495 | __set_current_state(TASK_RUNNING); | |
496 | } | |
497 | } | |
498 | return 0; | |
499 | } | |
500 | ||
501 | /* | |
502 | * Construct a log-writes mapping: | |
503 | * log-writes <dev_path> <log_dev_path> | |
504 | */ | |
505 | static int log_writes_ctr(struct dm_target *ti, unsigned int argc, char **argv) | |
506 | { | |
507 | struct log_writes_c *lc; | |
508 | struct dm_arg_set as; | |
509 | const char *devname, *logdevname; | |
e80d1c80 | 510 | int ret; |
0e9cebe7 JB |
511 | |
512 | as.argc = argc; | |
513 | as.argv = argv; | |
514 | ||
515 | if (argc < 2) { | |
516 | ti->error = "Invalid argument count"; | |
517 | return -EINVAL; | |
518 | } | |
519 | ||
520 | lc = kzalloc(sizeof(struct log_writes_c), GFP_KERNEL); | |
521 | if (!lc) { | |
522 | ti->error = "Cannot allocate context"; | |
523 | return -ENOMEM; | |
524 | } | |
525 | spin_lock_init(&lc->blocks_lock); | |
526 | INIT_LIST_HEAD(&lc->unflushed_blocks); | |
527 | INIT_LIST_HEAD(&lc->logging_blocks); | |
528 | init_waitqueue_head(&lc->wait); | |
211ad4b7 | 529 | init_completion(&lc->super_done); |
0e9cebe7 JB |
530 | atomic_set(&lc->io_blocks, 0); |
531 | atomic_set(&lc->pending_blocks, 0); | |
532 | ||
533 | devname = dm_shift_arg(&as); | |
e80d1c80 VG |
534 | ret = dm_get_device(ti, devname, dm_table_get_mode(ti->table), &lc->dev); |
535 | if (ret) { | |
0e9cebe7 JB |
536 | ti->error = "Device lookup failed"; |
537 | goto bad; | |
538 | } | |
539 | ||
540 | logdevname = dm_shift_arg(&as); | |
e80d1c80 VG |
541 | ret = dm_get_device(ti, logdevname, dm_table_get_mode(ti->table), |
542 | &lc->logdev); | |
543 | if (ret) { | |
0e9cebe7 JB |
544 | ti->error = "Log device lookup failed"; |
545 | dm_put_device(ti, lc->dev); | |
546 | goto bad; | |
547 | } | |
548 | ||
228bb5b2 JB |
549 | lc->sectorsize = bdev_logical_block_size(lc->dev->bdev); |
550 | lc->sectorshift = ilog2(lc->sectorsize); | |
0e9cebe7 | 551 | lc->log_kthread = kthread_run(log_writes_kthread, lc, "log-write"); |
91e630d9 VZ |
552 | if (IS_ERR(lc->log_kthread)) { |
553 | ret = PTR_ERR(lc->log_kthread); | |
0e9cebe7 JB |
554 | ti->error = "Couldn't alloc kthread"; |
555 | dm_put_device(ti, lc->dev); | |
556 | dm_put_device(ti, lc->logdev); | |
557 | goto bad; | |
558 | } | |
559 | ||
228bb5b2 JB |
560 | /* |
561 | * next_sector is in 512b sectors to correspond to what bi_sector expects. | |
562 | * The super starts at sector 0, and the next_sector is the next logical | |
563 | * one based on the sectorsize of the device. | |
564 | */ | |
565 | lc->next_sector = lc->sectorsize >> SECTOR_SHIFT; | |
0e9cebe7 JB |
566 | lc->logging_enabled = true; |
567 | lc->end_sector = logdev_last_sector(lc); | |
568 | lc->device_supports_discard = true; | |
569 | ||
570 | ti->num_flush_bios = 1; | |
571 | ti->flush_supported = true; | |
572 | ti->num_discard_bios = 1; | |
573 | ti->discards_supported = true; | |
30187e1d | 574 | ti->per_io_data_size = sizeof(struct per_bio_data); |
0e9cebe7 JB |
575 | ti->private = lc; |
576 | return 0; | |
577 | ||
578 | bad: | |
579 | kfree(lc); | |
e80d1c80 | 580 | return ret; |
0e9cebe7 JB |
581 | } |
582 | ||
583 | static int log_mark(struct log_writes_c *lc, char *data) | |
584 | { | |
585 | struct pending_block *block; | |
586 | size_t maxsize = lc->sectorsize - sizeof(struct log_write_entry); | |
587 | ||
588 | block = kzalloc(sizeof(struct pending_block), GFP_KERNEL); | |
589 | if (!block) { | |
590 | DMERR("Error allocating pending block"); | |
591 | return -ENOMEM; | |
592 | } | |
593 | ||
4b259fc4 | 594 | block->data = kstrndup(data, maxsize - 1, GFP_KERNEL); |
0e9cebe7 JB |
595 | if (!block->data) { |
596 | DMERR("Error copying mark data"); | |
597 | kfree(block); | |
598 | return -ENOMEM; | |
599 | } | |
600 | atomic_inc(&lc->pending_blocks); | |
601 | block->datalen = strlen(block->data); | |
602 | block->flags |= LOG_MARK_FLAG; | |
603 | spin_lock_irq(&lc->blocks_lock); | |
604 | list_add_tail(&block->list, &lc->logging_blocks); | |
605 | spin_unlock_irq(&lc->blocks_lock); | |
606 | wake_up_process(lc->log_kthread); | |
607 | return 0; | |
608 | } | |
609 | ||
610 | static void log_writes_dtr(struct dm_target *ti) | |
611 | { | |
612 | struct log_writes_c *lc = ti->private; | |
613 | ||
614 | spin_lock_irq(&lc->blocks_lock); | |
615 | list_splice_init(&lc->unflushed_blocks, &lc->logging_blocks); | |
616 | spin_unlock_irq(&lc->blocks_lock); | |
617 | ||
618 | /* | |
619 | * This is just nice to have since it'll update the super to include the | |
620 | * unflushed blocks, if it fails we don't really care. | |
621 | */ | |
622 | log_mark(lc, "dm-log-writes-end"); | |
623 | wake_up_process(lc->log_kthread); | |
624 | wait_event(lc->wait, !atomic_read(&lc->io_blocks) && | |
625 | !atomic_read(&lc->pending_blocks)); | |
626 | kthread_stop(lc->log_kthread); | |
627 | ||
628 | WARN_ON(!list_empty(&lc->logging_blocks)); | |
629 | WARN_ON(!list_empty(&lc->unflushed_blocks)); | |
630 | dm_put_device(ti, lc->dev); | |
631 | dm_put_device(ti, lc->logdev); | |
632 | kfree(lc); | |
633 | } | |
634 | ||
635 | static void normal_map_bio(struct dm_target *ti, struct bio *bio) | |
636 | { | |
637 | struct log_writes_c *lc = ti->private; | |
638 | ||
74d46992 | 639 | bio_set_dev(bio, lc->dev->bdev); |
0e9cebe7 JB |
640 | } |
641 | ||
642 | static int log_writes_map(struct dm_target *ti, struct bio *bio) | |
643 | { | |
644 | struct log_writes_c *lc = ti->private; | |
645 | struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data)); | |
646 | struct pending_block *block; | |
647 | struct bvec_iter iter; | |
648 | struct bio_vec bv; | |
649 | size_t alloc_size; | |
650 | int i = 0; | |
1eff9d32 JA |
651 | bool flush_bio = (bio->bi_opf & REQ_PREFLUSH); |
652 | bool fua_bio = (bio->bi_opf & REQ_FUA); | |
e6047149 | 653 | bool discard_bio = (bio_op(bio) == REQ_OP_DISCARD); |
e5c4cb9b | 654 | bool meta_bio = (bio->bi_opf & REQ_META); |
0e9cebe7 JB |
655 | |
656 | pb->block = NULL; | |
657 | ||
658 | /* Don't bother doing anything if logging has been disabled */ | |
659 | if (!lc->logging_enabled) | |
660 | goto map_bio; | |
661 | ||
662 | /* | |
663 | * Map reads as normal. | |
664 | */ | |
665 | if (bio_data_dir(bio) == READ) | |
666 | goto map_bio; | |
667 | ||
668 | /* No sectors and not a flush? Don't care */ | |
669 | if (!bio_sectors(bio) && !flush_bio) | |
670 | goto map_bio; | |
671 | ||
672 | /* | |
673 | * Discards will have bi_size set but there's no actual data, so just | |
674 | * allocate the size of the pending block. | |
675 | */ | |
676 | if (discard_bio) | |
677 | alloc_size = sizeof(struct pending_block); | |
678 | else | |
d4e6e836 | 679 | alloc_size = struct_size(block, vecs, bio_segments(bio)); |
0e9cebe7 JB |
680 | |
681 | block = kzalloc(alloc_size, GFP_NOIO); | |
682 | if (!block) { | |
683 | DMERR("Error allocating pending block"); | |
684 | spin_lock_irq(&lc->blocks_lock); | |
685 | lc->logging_enabled = false; | |
686 | spin_unlock_irq(&lc->blocks_lock); | |
846785e6 | 687 | return DM_MAPIO_KILL; |
0e9cebe7 JB |
688 | } |
689 | INIT_LIST_HEAD(&block->list); | |
690 | pb->block = block; | |
691 | atomic_inc(&lc->pending_blocks); | |
692 | ||
693 | if (flush_bio) | |
694 | block->flags |= LOG_FLUSH_FLAG; | |
695 | if (fua_bio) | |
696 | block->flags |= LOG_FUA_FLAG; | |
697 | if (discard_bio) | |
698 | block->flags |= LOG_DISCARD_FLAG; | |
e5c4cb9b QW |
699 | if (meta_bio) |
700 | block->flags |= LOG_METADATA_FLAG; | |
0e9cebe7 | 701 | |
228bb5b2 JB |
702 | block->sector = bio_to_dev_sectors(lc, bio->bi_iter.bi_sector); |
703 | block->nr_sectors = bio_to_dev_sectors(lc, bio_sectors(bio)); | |
0e9cebe7 JB |
704 | |
705 | /* We don't need the data, just submit */ | |
706 | if (discard_bio) { | |
707 | WARN_ON(flush_bio || fua_bio); | |
708 | if (lc->device_supports_discard) | |
709 | goto map_bio; | |
4246a0b6 | 710 | bio_endio(bio); |
0e9cebe7 JB |
711 | return DM_MAPIO_SUBMITTED; |
712 | } | |
713 | ||
714 | /* Flush bio, splice the unflushed blocks onto this list and submit */ | |
715 | if (flush_bio && !bio_sectors(bio)) { | |
716 | spin_lock_irq(&lc->blocks_lock); | |
717 | list_splice_init(&lc->unflushed_blocks, &block->list); | |
718 | spin_unlock_irq(&lc->blocks_lock); | |
719 | goto map_bio; | |
720 | } | |
721 | ||
722 | /* | |
723 | * We will write this bio somewhere else way later so we need to copy | |
724 | * the actual contents into new pages so we know the data will always be | |
725 | * there. | |
726 | * | |
727 | * We do this because this could be a bio from O_DIRECT in which case we | |
728 | * can't just hold onto the page until some later point, we have to | |
729 | * manually copy the contents. | |
730 | */ | |
731 | bio_for_each_segment(bv, bio, iter) { | |
732 | struct page *page; | |
27db2717 | 733 | void *dst; |
0e9cebe7 JB |
734 | |
735 | page = alloc_page(GFP_NOIO); | |
736 | if (!page) { | |
737 | DMERR("Error allocing page"); | |
738 | free_pending_block(lc, block); | |
739 | spin_lock_irq(&lc->blocks_lock); | |
740 | lc->logging_enabled = false; | |
741 | spin_unlock_irq(&lc->blocks_lock); | |
846785e6 | 742 | return DM_MAPIO_KILL; |
0e9cebe7 JB |
743 | } |
744 | ||
0e9cebe7 | 745 | dst = kmap_atomic(page); |
27db2717 | 746 | memcpy_from_bvec(dst, &bv); |
0e9cebe7 | 747 | kunmap_atomic(dst); |
0e9cebe7 JB |
748 | block->vecs[i].bv_page = page; |
749 | block->vecs[i].bv_len = bv.bv_len; | |
750 | block->vec_cnt++; | |
751 | i++; | |
752 | } | |
753 | ||
754 | /* Had a flush with data in it, weird */ | |
755 | if (flush_bio) { | |
756 | spin_lock_irq(&lc->blocks_lock); | |
757 | list_splice_init(&lc->unflushed_blocks, &block->list); | |
758 | spin_unlock_irq(&lc->blocks_lock); | |
759 | } | |
760 | map_bio: | |
761 | normal_map_bio(ti, bio); | |
762 | return DM_MAPIO_REMAPPED; | |
763 | } | |
764 | ||
4e4cbee9 CH |
765 | static int normal_end_io(struct dm_target *ti, struct bio *bio, |
766 | blk_status_t *error) | |
0e9cebe7 JB |
767 | { |
768 | struct log_writes_c *lc = ti->private; | |
769 | struct per_bio_data *pb = dm_per_bio_data(bio, sizeof(struct per_bio_data)); | |
770 | ||
771 | if (bio_data_dir(bio) == WRITE && pb->block) { | |
772 | struct pending_block *block = pb->block; | |
773 | unsigned long flags; | |
774 | ||
775 | spin_lock_irqsave(&lc->blocks_lock, flags); | |
776 | if (block->flags & LOG_FLUSH_FLAG) { | |
777 | list_splice_tail_init(&block->list, &lc->logging_blocks); | |
778 | list_add_tail(&block->list, &lc->logging_blocks); | |
779 | wake_up_process(lc->log_kthread); | |
780 | } else if (block->flags & LOG_FUA_FLAG) { | |
781 | list_add_tail(&block->list, &lc->logging_blocks); | |
782 | wake_up_process(lc->log_kthread); | |
783 | } else | |
784 | list_add_tail(&block->list, &lc->unflushed_blocks); | |
785 | spin_unlock_irqrestore(&lc->blocks_lock, flags); | |
786 | } | |
787 | ||
1be56909 | 788 | return DM_ENDIO_DONE; |
0e9cebe7 JB |
789 | } |
790 | ||
791 | /* | |
792 | * INFO format: <logged entries> <highest allocated sector> | |
793 | */ | |
794 | static void log_writes_status(struct dm_target *ti, status_type_t type, | |
795 | unsigned status_flags, char *result, | |
796 | unsigned maxlen) | |
797 | { | |
798 | unsigned sz = 0; | |
799 | struct log_writes_c *lc = ti->private; | |
800 | ||
801 | switch (type) { | |
802 | case STATUSTYPE_INFO: | |
803 | DMEMIT("%llu %llu", lc->logged_entries, | |
804 | (unsigned long long)lc->next_sector - 1); | |
805 | if (!lc->logging_enabled) | |
806 | DMEMIT(" logging_disabled"); | |
807 | break; | |
808 | ||
809 | case STATUSTYPE_TABLE: | |
810 | DMEMIT("%s %s", lc->dev->name, lc->logdev->name); | |
811 | break; | |
8ec45662 TS |
812 | |
813 | case STATUSTYPE_IMA: | |
814 | *result = '\0'; | |
815 | break; | |
0e9cebe7 JB |
816 | } |
817 | } | |
818 | ||
e56f81e0 | 819 | static int log_writes_prepare_ioctl(struct dm_target *ti, |
5bd5e8d8 | 820 | struct block_device **bdev) |
0e9cebe7 JB |
821 | { |
822 | struct log_writes_c *lc = ti->private; | |
823 | struct dm_dev *dev = lc->dev; | |
0e9cebe7 | 824 | |
e56f81e0 | 825 | *bdev = dev->bdev; |
0e9cebe7 JB |
826 | /* |
827 | * Only pass ioctls through if the device sizes match exactly. | |
828 | */ | |
6dcbb52c | 829 | if (ti->len != bdev_nr_sectors(dev->bdev)) |
e56f81e0 CH |
830 | return 1; |
831 | return 0; | |
0e9cebe7 JB |
832 | } |
833 | ||
0e9cebe7 JB |
834 | static int log_writes_iterate_devices(struct dm_target *ti, |
835 | iterate_devices_callout_fn fn, | |
836 | void *data) | |
837 | { | |
838 | struct log_writes_c *lc = ti->private; | |
839 | ||
840 | return fn(ti, lc->dev, 0, ti->len, data); | |
841 | } | |
842 | ||
843 | /* | |
844 | * Messages supported: | |
845 | * mark <mark data> - specify the marked data. | |
846 | */ | |
1eb5fa84 MS |
847 | static int log_writes_message(struct dm_target *ti, unsigned argc, char **argv, |
848 | char *result, unsigned maxlen) | |
0e9cebe7 JB |
849 | { |
850 | int r = -EINVAL; | |
851 | struct log_writes_c *lc = ti->private; | |
852 | ||
853 | if (argc != 2) { | |
854 | DMWARN("Invalid log-writes message arguments, expect 2 arguments, got %d", argc); | |
855 | return r; | |
856 | } | |
857 | ||
858 | if (!strcasecmp(argv[0], "mark")) | |
859 | r = log_mark(lc, argv[1]); | |
860 | else | |
861 | DMWARN("Unrecognised log writes target message received: %s", argv[0]); | |
862 | ||
863 | return r; | |
864 | } | |
865 | ||
866 | static void log_writes_io_hints(struct dm_target *ti, struct queue_limits *limits) | |
867 | { | |
868 | struct log_writes_c *lc = ti->private; | |
0e9cebe7 | 869 | |
70200574 | 870 | if (!bdev_max_discard_sectors(lc->dev->bdev)) { |
0e9cebe7 | 871 | lc->device_supports_discard = false; |
228bb5b2 | 872 | limits->discard_granularity = lc->sectorsize; |
0e9cebe7 JB |
873 | limits->max_discard_sectors = (UINT_MAX >> SECTOR_SHIFT); |
874 | } | |
228bb5b2 JB |
875 | limits->logical_block_size = bdev_logical_block_size(lc->dev->bdev); |
876 | limits->physical_block_size = bdev_physical_block_size(lc->dev->bdev); | |
877 | limits->io_min = limits->physical_block_size; | |
0e9cebe7 JB |
878 | } |
879 | ||
5d2a228b | 880 | #if IS_ENABLED(CONFIG_FS_DAX) |
d19bd675 CH |
881 | static struct dax_device *log_writes_dax_pgoff(struct dm_target *ti, |
882 | pgoff_t *pgoff) | |
883 | { | |
884 | struct log_writes_c *lc = ti->private; | |
885 | ||
886 | *pgoff += (get_start_sect(lc->dev->bdev) >> PAGE_SECTORS_SHIFT); | |
887 | return lc->dev->dax_dev; | |
888 | } | |
889 | ||
98d82f48 | 890 | static long log_writes_dax_direct_access(struct dm_target *ti, pgoff_t pgoff, |
e511c4a3 JC |
891 | long nr_pages, enum dax_access_mode mode, void **kaddr, |
892 | pfn_t *pfn) | |
98d82f48 | 893 | { |
d19bd675 | 894 | struct dax_device *dax_dev = log_writes_dax_pgoff(ti, &pgoff); |
98d82f48 | 895 | |
e511c4a3 | 896 | return dax_direct_access(dax_dev, pgoff, nr_pages, mode, kaddr, pfn); |
98d82f48 RZ |
897 | } |
898 | ||
cdf6cdcd VG |
899 | static int log_writes_dax_zero_page_range(struct dm_target *ti, pgoff_t pgoff, |
900 | size_t nr_pages) | |
901 | { | |
d19bd675 | 902 | struct dax_device *dax_dev = log_writes_dax_pgoff(ti, &pgoff); |
cdf6cdcd | 903 | |
d19bd675 | 904 | return dax_zero_page_range(dax_dev, pgoff, nr_pages << PAGE_SHIFT); |
cdf6cdcd VG |
905 | } |
906 | ||
047218ec JC |
907 | static size_t log_writes_dax_recovery_write(struct dm_target *ti, |
908 | pgoff_t pgoff, void *addr, size_t bytes, struct iov_iter *i) | |
909 | { | |
910 | struct dax_device *dax_dev = log_writes_dax_pgoff(ti, &pgoff); | |
911 | ||
912 | return dax_recovery_write(dax_dev, pgoff, addr, bytes, i); | |
913 | } | |
914 | ||
976431b0 DW |
915 | #else |
916 | #define log_writes_dax_direct_access NULL | |
cdf6cdcd | 917 | #define log_writes_dax_zero_page_range NULL |
047218ec | 918 | #define log_writes_dax_recovery_write NULL |
976431b0 | 919 | #endif |
98d82f48 | 920 | |
0e9cebe7 JB |
921 | static struct target_type log_writes_target = { |
922 | .name = "log-writes", | |
98d82f48 | 923 | .version = {1, 1, 0}, |
0e9cebe7 JB |
924 | .module = THIS_MODULE, |
925 | .ctr = log_writes_ctr, | |
926 | .dtr = log_writes_dtr, | |
927 | .map = log_writes_map, | |
928 | .end_io = normal_end_io, | |
929 | .status = log_writes_status, | |
e56f81e0 | 930 | .prepare_ioctl = log_writes_prepare_ioctl, |
0e9cebe7 JB |
931 | .message = log_writes_message, |
932 | .iterate_devices = log_writes_iterate_devices, | |
933 | .io_hints = log_writes_io_hints, | |
98d82f48 | 934 | .direct_access = log_writes_dax_direct_access, |
cdf6cdcd | 935 | .dax_zero_page_range = log_writes_dax_zero_page_range, |
047218ec | 936 | .dax_recovery_write = log_writes_dax_recovery_write, |
0e9cebe7 JB |
937 | }; |
938 | ||
939 | static int __init dm_log_writes_init(void) | |
940 | { | |
941 | int r = dm_register_target(&log_writes_target); | |
942 | ||
943 | if (r < 0) | |
944 | DMERR("register failed %d", r); | |
945 | ||
946 | return r; | |
947 | } | |
948 | ||
949 | static void __exit dm_log_writes_exit(void) | |
950 | { | |
951 | dm_unregister_target(&log_writes_target); | |
952 | } | |
953 | ||
954 | module_init(dm_log_writes_init); | |
955 | module_exit(dm_log_writes_exit); | |
956 | ||
957 | MODULE_DESCRIPTION(DM_NAME " log writes target"); | |
958 | MODULE_AUTHOR("Josef Bacik <jbacik@fb.com>"); | |
959 | MODULE_LICENSE("GPL"); |