Commit | Line | Data |
---|---|---|
cafe5635 KO |
1 | /* |
2 | * Main bcache entry point - handle a read or a write request and decide what to | |
3 | * do with it; the make_request functions are called by the block layer. | |
4 | * | |
5 | * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com> | |
6 | * Copyright 2012 Google, Inc. | |
7 | */ | |
8 | ||
9 | #include "bcache.h" | |
10 | #include "btree.h" | |
11 | #include "debug.h" | |
12 | #include "request.h" | |
279afbad | 13 | #include "writeback.h" |
cafe5635 | 14 | |
cafe5635 KO |
15 | #include <linux/module.h> |
16 | #include <linux/hash.h> | |
17 | #include <linux/random.h> | |
66114cad | 18 | #include <linux/backing-dev.h> |
cafe5635 KO |
19 | |
20 | #include <trace/events/bcache.h> | |
21 | ||
22 | #define CUTOFF_CACHE_ADD 95 | |
23 | #define CUTOFF_CACHE_READA 90 | |
cafe5635 KO |
24 | |
25 | struct kmem_cache *bch_search_cache; | |
26 | ||
a34a8bfd KO |
27 | static void bch_data_insert_start(struct closure *); |
28 | ||
cafe5635 KO |
29 | static unsigned cache_mode(struct cached_dev *dc, struct bio *bio) |
30 | { | |
cafe5635 KO |
31 | return BDEV_CACHE_MODE(&dc->sb); |
32 | } | |
33 | ||
34 | static bool verify(struct cached_dev *dc, struct bio *bio) | |
35 | { | |
cafe5635 KO |
36 | return dc->verify; |
37 | } | |
38 | ||
39 | static void bio_csum(struct bio *bio, struct bkey *k) | |
40 | { | |
7988613b KO |
41 | struct bio_vec bv; |
42 | struct bvec_iter iter; | |
cafe5635 | 43 | uint64_t csum = 0; |
cafe5635 | 44 | |
7988613b KO |
45 | bio_for_each_segment(bv, bio, iter) { |
46 | void *d = kmap(bv.bv_page) + bv.bv_offset; | |
47 | csum = bch_crc64_update(csum, d, bv.bv_len); | |
48 | kunmap(bv.bv_page); | |
cafe5635 KO |
49 | } |
50 | ||
51 | k->ptr[KEY_PTRS(k)] = csum & (~0ULL >> 1); | |
52 | } | |
53 | ||
54 | /* Insert data into cache */ | |
55 | ||
a34a8bfd | 56 | static void bch_data_insert_keys(struct closure *cl) |
cafe5635 | 57 | { |
220bb38c | 58 | struct data_insert_op *op = container_of(cl, struct data_insert_op, cl); |
c18536a7 | 59 | atomic_t *journal_ref = NULL; |
220bb38c | 60 | struct bkey *replace_key = op->replace ? &op->replace_key : NULL; |
6054c6d4 | 61 | int ret; |
cafe5635 | 62 | |
a34a8bfd KO |
63 | /* |
64 | * If we're looping, might already be waiting on | |
65 | * another journal write - can't wait on more than one journal write at | |
66 | * a time | |
67 | * | |
68 | * XXX: this looks wrong | |
69 | */ | |
70 | #if 0 | |
71 | while (atomic_read(&s->cl.remaining) & CLOSURE_WAITING) | |
72 | closure_sync(&s->cl); | |
73 | #endif | |
cafe5635 | 74 | |
220bb38c KO |
75 | if (!op->replace) |
76 | journal_ref = bch_journal(op->c, &op->insert_keys, | |
77 | op->flush_journal ? cl : NULL); | |
cafe5635 | 78 | |
220bb38c | 79 | ret = bch_btree_insert(op->c, &op->insert_keys, |
6054c6d4 KO |
80 | journal_ref, replace_key); |
81 | if (ret == -ESRCH) { | |
220bb38c | 82 | op->replace_collision = true; |
6054c6d4 | 83 | } else if (ret) { |
220bb38c KO |
84 | op->error = -ENOMEM; |
85 | op->insert_data_done = true; | |
a34a8bfd | 86 | } |
cafe5635 | 87 | |
c18536a7 KO |
88 | if (journal_ref) |
89 | atomic_dec_bug(journal_ref); | |
cafe5635 | 90 | |
77b5a084 | 91 | if (!op->insert_data_done) { |
da415a09 | 92 | continue_at(cl, bch_data_insert_start, op->wq); |
77b5a084 JA |
93 | return; |
94 | } | |
cafe5635 | 95 | |
220bb38c | 96 | bch_keylist_free(&op->insert_keys); |
a34a8bfd | 97 | closure_return(cl); |
cafe5635 KO |
98 | } |
99 | ||
085d2a3d KO |
100 | static int bch_keylist_realloc(struct keylist *l, unsigned u64s, |
101 | struct cache_set *c) | |
102 | { | |
103 | size_t oldsize = bch_keylist_nkeys(l); | |
104 | size_t newsize = oldsize + u64s; | |
105 | ||
106 | /* | |
107 | * The journalling code doesn't handle the case where the keys to insert | |
108 | * is bigger than an empty write: If we just return -ENOMEM here, | |
109 | * bio_insert() and bio_invalidate() will insert the keys created so far | |
110 | * and finish the rest when the keylist is empty. | |
111 | */ | |
112 | if (newsize * sizeof(uint64_t) > block_bytes(c) - sizeof(struct jset)) | |
113 | return -ENOMEM; | |
114 | ||
115 | return __bch_keylist_realloc(l, u64s); | |
116 | } | |
117 | ||
a34a8bfd KO |
118 | static void bch_data_invalidate(struct closure *cl) |
119 | { | |
220bb38c KO |
120 | struct data_insert_op *op = container_of(cl, struct data_insert_op, cl); |
121 | struct bio *bio = op->bio; | |
a34a8bfd KO |
122 | |
123 | pr_debug("invalidating %i sectors from %llu", | |
4f024f37 | 124 | bio_sectors(bio), (uint64_t) bio->bi_iter.bi_sector); |
a34a8bfd KO |
125 | |
126 | while (bio_sectors(bio)) { | |
81ab4190 KO |
127 | unsigned sectors = min(bio_sectors(bio), |
128 | 1U << (KEY_SIZE_BITS - 1)); | |
a34a8bfd | 129 | |
085d2a3d | 130 | if (bch_keylist_realloc(&op->insert_keys, 2, op->c)) |
a34a8bfd KO |
131 | goto out; |
132 | ||
4f024f37 KO |
133 | bio->bi_iter.bi_sector += sectors; |
134 | bio->bi_iter.bi_size -= sectors << 9; | |
a34a8bfd | 135 | |
220bb38c | 136 | bch_keylist_add(&op->insert_keys, |
4f024f37 | 137 | &KEY(op->inode, bio->bi_iter.bi_sector, sectors)); |
a34a8bfd KO |
138 | } |
139 | ||
220bb38c | 140 | op->insert_data_done = true; |
a34a8bfd KO |
141 | bio_put(bio); |
142 | out: | |
da415a09 | 143 | continue_at(cl, bch_data_insert_keys, op->wq); |
a34a8bfd KO |
144 | } |
145 | ||
146 | static void bch_data_insert_error(struct closure *cl) | |
cafe5635 | 147 | { |
220bb38c | 148 | struct data_insert_op *op = container_of(cl, struct data_insert_op, cl); |
cafe5635 KO |
149 | |
150 | /* | |
151 | * Our data write just errored, which means we've got a bunch of keys to | |
152 | * insert that point to data that wasn't succesfully written. | |
153 | * | |
154 | * We don't have to insert those keys but we still have to invalidate | |
155 | * that region of the cache - so, if we just strip off all the pointers | |
156 | * from the keys we'll accomplish just that. | |
157 | */ | |
158 | ||
220bb38c | 159 | struct bkey *src = op->insert_keys.keys, *dst = op->insert_keys.keys; |
cafe5635 | 160 | |
220bb38c | 161 | while (src != op->insert_keys.top) { |
cafe5635 KO |
162 | struct bkey *n = bkey_next(src); |
163 | ||
164 | SET_KEY_PTRS(src, 0); | |
c2f95ae2 | 165 | memmove(dst, src, bkey_bytes(src)); |
cafe5635 KO |
166 | |
167 | dst = bkey_next(dst); | |
168 | src = n; | |
169 | } | |
170 | ||
220bb38c | 171 | op->insert_keys.top = dst; |
cafe5635 | 172 | |
a34a8bfd | 173 | bch_data_insert_keys(cl); |
cafe5635 KO |
174 | } |
175 | ||
4246a0b6 | 176 | static void bch_data_insert_endio(struct bio *bio) |
cafe5635 KO |
177 | { |
178 | struct closure *cl = bio->bi_private; | |
220bb38c | 179 | struct data_insert_op *op = container_of(cl, struct data_insert_op, cl); |
cafe5635 | 180 | |
4246a0b6 | 181 | if (bio->bi_error) { |
cafe5635 | 182 | /* TODO: We could try to recover from this. */ |
220bb38c | 183 | if (op->writeback) |
4246a0b6 | 184 | op->error = bio->bi_error; |
220bb38c | 185 | else if (!op->replace) |
da415a09 | 186 | set_closure_fn(cl, bch_data_insert_error, op->wq); |
cafe5635 KO |
187 | else |
188 | set_closure_fn(cl, NULL, NULL); | |
189 | } | |
190 | ||
4246a0b6 | 191 | bch_bbio_endio(op->c, bio, bio->bi_error, "writing data to cache"); |
cafe5635 KO |
192 | } |
193 | ||
a34a8bfd | 194 | static void bch_data_insert_start(struct closure *cl) |
cafe5635 | 195 | { |
220bb38c KO |
196 | struct data_insert_op *op = container_of(cl, struct data_insert_op, cl); |
197 | struct bio *bio = op->bio, *n; | |
cafe5635 | 198 | |
220bb38c KO |
199 | if (atomic_sub_return(bio_sectors(bio), &op->c->sectors_to_gc) < 0) { |
200 | set_gc_sectors(op->c); | |
201 | wake_up_gc(op->c); | |
cafe5635 KO |
202 | } |
203 | ||
e3b4825b NS |
204 | if (op->bypass) |
205 | return bch_data_invalidate(cl); | |
206 | ||
54d12f2b KO |
207 | /* |
208 | * Journal writes are marked REQ_FLUSH; if the original write was a | |
209 | * flush, it'll wait on the journal write. | |
210 | */ | |
211 | bio->bi_rw &= ~(REQ_FLUSH|REQ_FUA); | |
212 | ||
cafe5635 KO |
213 | do { |
214 | unsigned i; | |
215 | struct bkey *k; | |
220bb38c | 216 | struct bio_set *split = op->c->bio_split; |
cafe5635 KO |
217 | |
218 | /* 1 for the device pointer and 1 for the chksum */ | |
220bb38c | 219 | if (bch_keylist_realloc(&op->insert_keys, |
085d2a3d | 220 | 3 + (op->csum ? 1 : 0), |
77b5a084 | 221 | op->c)) { |
da415a09 | 222 | continue_at(cl, bch_data_insert_keys, op->wq); |
77b5a084 JA |
223 | return; |
224 | } | |
cafe5635 | 225 | |
220bb38c | 226 | k = op->insert_keys.top; |
cafe5635 | 227 | bkey_init(k); |
220bb38c | 228 | SET_KEY_INODE(k, op->inode); |
4f024f37 | 229 | SET_KEY_OFFSET(k, bio->bi_iter.bi_sector); |
cafe5635 | 230 | |
2599b53b KO |
231 | if (!bch_alloc_sectors(op->c, k, bio_sectors(bio), |
232 | op->write_point, op->write_prio, | |
233 | op->writeback)) | |
cafe5635 KO |
234 | goto err; |
235 | ||
20d0189b | 236 | n = bio_next_split(bio, KEY_SIZE(k), GFP_NOIO, split); |
cafe5635 | 237 | |
a34a8bfd | 238 | n->bi_end_io = bch_data_insert_endio; |
cafe5635 KO |
239 | n->bi_private = cl; |
240 | ||
220bb38c | 241 | if (op->writeback) { |
cafe5635 KO |
242 | SET_KEY_DIRTY(k, true); |
243 | ||
244 | for (i = 0; i < KEY_PTRS(k); i++) | |
220bb38c | 245 | SET_GC_MARK(PTR_BUCKET(op->c, k, i), |
cafe5635 KO |
246 | GC_MARK_DIRTY); |
247 | } | |
248 | ||
220bb38c | 249 | SET_KEY_CSUM(k, op->csum); |
cafe5635 KO |
250 | if (KEY_CSUM(k)) |
251 | bio_csum(n, k); | |
252 | ||
c37511b8 | 253 | trace_bcache_cache_insert(k); |
220bb38c | 254 | bch_keylist_push(&op->insert_keys); |
cafe5635 | 255 | |
cafe5635 | 256 | n->bi_rw |= REQ_WRITE; |
220bb38c | 257 | bch_submit_bbio(n, op->c, k, 0); |
cafe5635 KO |
258 | } while (n != bio); |
259 | ||
220bb38c | 260 | op->insert_data_done = true; |
da415a09 | 261 | continue_at(cl, bch_data_insert_keys, op->wq); |
77b5a084 | 262 | return; |
cafe5635 KO |
263 | err: |
264 | /* bch_alloc_sectors() blocks if s->writeback = true */ | |
220bb38c | 265 | BUG_ON(op->writeback); |
cafe5635 KO |
266 | |
267 | /* | |
268 | * But if it's not a writeback write we'd rather just bail out if | |
269 | * there aren't any buckets ready to write to - it might take awhile and | |
270 | * we might be starving btree writes for gc or something. | |
271 | */ | |
272 | ||
220bb38c | 273 | if (!op->replace) { |
cafe5635 KO |
274 | /* |
275 | * Writethrough write: We can't complete the write until we've | |
276 | * updated the index. But we don't want to delay the write while | |
277 | * we wait for buckets to be freed up, so just invalidate the | |
278 | * rest of the write. | |
279 | */ | |
220bb38c | 280 | op->bypass = true; |
a34a8bfd | 281 | return bch_data_invalidate(cl); |
cafe5635 KO |
282 | } else { |
283 | /* | |
284 | * From a cache miss, we can just insert the keys for the data | |
285 | * we have written or bail out if we didn't do anything. | |
286 | */ | |
220bb38c | 287 | op->insert_data_done = true; |
cafe5635 KO |
288 | bio_put(bio); |
289 | ||
220bb38c | 290 | if (!bch_keylist_empty(&op->insert_keys)) |
da415a09 | 291 | continue_at(cl, bch_data_insert_keys, op->wq); |
cafe5635 KO |
292 | else |
293 | closure_return(cl); | |
294 | } | |
295 | } | |
296 | ||
297 | /** | |
a34a8bfd | 298 | * bch_data_insert - stick some data in the cache |
cafe5635 KO |
299 | * |
300 | * This is the starting point for any data to end up in a cache device; it could | |
301 | * be from a normal write, or a writeback write, or a write to a flash only | |
302 | * volume - it's also used by the moving garbage collector to compact data in | |
303 | * mostly empty buckets. | |
304 | * | |
305 | * It first writes the data to the cache, creating a list of keys to be inserted | |
306 | * (if the data had to be fragmented there will be multiple keys); after the | |
307 | * data is written it calls bch_journal, and after the keys have been added to | |
308 | * the next journal write they're inserted into the btree. | |
309 | * | |
c18536a7 | 310 | * It inserts the data in s->cache_bio; bi_sector is used for the key offset, |
cafe5635 KO |
311 | * and op->inode is used for the key inode. |
312 | * | |
c18536a7 KO |
313 | * If s->bypass is true, instead of inserting the data it invalidates the |
314 | * region of the cache represented by s->cache_bio and op->inode. | |
cafe5635 | 315 | */ |
a34a8bfd | 316 | void bch_data_insert(struct closure *cl) |
cafe5635 | 317 | { |
220bb38c | 318 | struct data_insert_op *op = container_of(cl, struct data_insert_op, cl); |
cafe5635 | 319 | |
60ae81ee SP |
320 | trace_bcache_write(op->c, op->inode, op->bio, |
321 | op->writeback, op->bypass); | |
220bb38c KO |
322 | |
323 | bch_keylist_init(&op->insert_keys); | |
324 | bio_get(op->bio); | |
a34a8bfd | 325 | bch_data_insert_start(cl); |
cafe5635 KO |
326 | } |
327 | ||
220bb38c KO |
328 | /* Congested? */ |
329 | ||
330 | unsigned bch_get_congested(struct cache_set *c) | |
331 | { | |
332 | int i; | |
333 | long rand; | |
334 | ||
335 | if (!c->congested_read_threshold_us && | |
336 | !c->congested_write_threshold_us) | |
337 | return 0; | |
338 | ||
339 | i = (local_clock_us() - c->congested_last_us) / 1024; | |
340 | if (i < 0) | |
341 | return 0; | |
342 | ||
343 | i += atomic_read(&c->congested); | |
344 | if (i >= 0) | |
345 | return 0; | |
346 | ||
347 | i += CONGESTED_MAX; | |
348 | ||
349 | if (i > 0) | |
350 | i = fract_exp_two(i, 6); | |
351 | ||
352 | rand = get_random_int(); | |
353 | i -= bitmap_weight(&rand, BITS_PER_LONG); | |
354 | ||
355 | return i > 0 ? i : 1; | |
356 | } | |
357 | ||
358 | static void add_sequential(struct task_struct *t) | |
359 | { | |
360 | ewma_add(t->sequential_io_avg, | |
361 | t->sequential_io, 8, 0); | |
362 | ||
363 | t->sequential_io = 0; | |
364 | } | |
365 | ||
366 | static struct hlist_head *iohash(struct cached_dev *dc, uint64_t k) | |
367 | { | |
368 | return &dc->io_hash[hash_64(k, RECENT_IO_BITS)]; | |
369 | } | |
370 | ||
371 | static bool check_should_bypass(struct cached_dev *dc, struct bio *bio) | |
372 | { | |
373 | struct cache_set *c = dc->disk.c; | |
374 | unsigned mode = cache_mode(dc, bio); | |
375 | unsigned sectors, congested = bch_get_congested(c); | |
376 | struct task_struct *task = current; | |
8aee1220 | 377 | struct io *i; |
220bb38c | 378 | |
c4d951dd | 379 | if (test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags) || |
220bb38c KO |
380 | c->gc_stats.in_use > CUTOFF_CACHE_ADD || |
381 | (bio->bi_rw & REQ_DISCARD)) | |
382 | goto skip; | |
383 | ||
384 | if (mode == CACHE_MODE_NONE || | |
385 | (mode == CACHE_MODE_WRITEAROUND && | |
386 | (bio->bi_rw & REQ_WRITE))) | |
387 | goto skip; | |
388 | ||
4f024f37 | 389 | if (bio->bi_iter.bi_sector & (c->sb.block_size - 1) || |
220bb38c KO |
390 | bio_sectors(bio) & (c->sb.block_size - 1)) { |
391 | pr_debug("skipping unaligned io"); | |
392 | goto skip; | |
393 | } | |
394 | ||
5ceaaad7 KO |
395 | if (bypass_torture_test(dc)) { |
396 | if ((get_random_int() & 3) == 3) | |
397 | goto skip; | |
398 | else | |
399 | goto rescale; | |
400 | } | |
401 | ||
220bb38c KO |
402 | if (!congested && !dc->sequential_cutoff) |
403 | goto rescale; | |
404 | ||
405 | if (!congested && | |
406 | mode == CACHE_MODE_WRITEBACK && | |
407 | (bio->bi_rw & REQ_WRITE) && | |
408 | (bio->bi_rw & REQ_SYNC)) | |
409 | goto rescale; | |
410 | ||
8aee1220 | 411 | spin_lock(&dc->io_lock); |
220bb38c | 412 | |
4f024f37 KO |
413 | hlist_for_each_entry(i, iohash(dc, bio->bi_iter.bi_sector), hash) |
414 | if (i->last == bio->bi_iter.bi_sector && | |
8aee1220 KO |
415 | time_before(jiffies, i->jiffies)) |
416 | goto found; | |
220bb38c | 417 | |
8aee1220 | 418 | i = list_first_entry(&dc->io_lru, struct io, lru); |
220bb38c | 419 | |
8aee1220 KO |
420 | add_sequential(task); |
421 | i->sequential = 0; | |
220bb38c | 422 | found: |
4f024f37 KO |
423 | if (i->sequential + bio->bi_iter.bi_size > i->sequential) |
424 | i->sequential += bio->bi_iter.bi_size; | |
220bb38c | 425 | |
8aee1220 KO |
426 | i->last = bio_end_sector(bio); |
427 | i->jiffies = jiffies + msecs_to_jiffies(5000); | |
428 | task->sequential_io = i->sequential; | |
220bb38c | 429 | |
8aee1220 KO |
430 | hlist_del(&i->hash); |
431 | hlist_add_head(&i->hash, iohash(dc, i->last)); | |
432 | list_move_tail(&i->lru, &dc->io_lru); | |
220bb38c | 433 | |
8aee1220 | 434 | spin_unlock(&dc->io_lock); |
220bb38c KO |
435 | |
436 | sectors = max(task->sequential_io, | |
437 | task->sequential_io_avg) >> 9; | |
438 | ||
439 | if (dc->sequential_cutoff && | |
440 | sectors >= dc->sequential_cutoff >> 9) { | |
441 | trace_bcache_bypass_sequential(bio); | |
442 | goto skip; | |
443 | } | |
444 | ||
445 | if (congested && sectors >= congested) { | |
446 | trace_bcache_bypass_congested(bio); | |
447 | goto skip; | |
448 | } | |
449 | ||
450 | rescale: | |
451 | bch_rescale_priorities(c, bio_sectors(bio)); | |
452 | return false; | |
453 | skip: | |
454 | bch_mark_sectors_bypassed(c, dc, bio_sectors(bio)); | |
455 | return true; | |
456 | } | |
457 | ||
2c1953e2 | 458 | /* Cache lookup */ |
cafe5635 | 459 | |
220bb38c KO |
460 | struct search { |
461 | /* Stack frame for bio_complete */ | |
462 | struct closure cl; | |
463 | ||
220bb38c KO |
464 | struct bbio bio; |
465 | struct bio *orig_bio; | |
466 | struct bio *cache_miss; | |
a5ae4300 | 467 | struct bcache_device *d; |
220bb38c KO |
468 | |
469 | unsigned insert_bio_sectors; | |
220bb38c | 470 | unsigned recoverable:1; |
220bb38c | 471 | unsigned write:1; |
5ceaaad7 | 472 | unsigned read_dirty_data:1; |
220bb38c KO |
473 | |
474 | unsigned long start_time; | |
475 | ||
476 | struct btree_op op; | |
477 | struct data_insert_op iop; | |
478 | }; | |
479 | ||
4246a0b6 | 480 | static void bch_cache_read_endio(struct bio *bio) |
cafe5635 KO |
481 | { |
482 | struct bbio *b = container_of(bio, struct bbio, bio); | |
483 | struct closure *cl = bio->bi_private; | |
484 | struct search *s = container_of(cl, struct search, cl); | |
485 | ||
486 | /* | |
487 | * If the bucket was reused while our bio was in flight, we might have | |
488 | * read the wrong data. Set s->error but not error so it doesn't get | |
489 | * counted against the cache device, but we'll still reread the data | |
490 | * from the backing device. | |
491 | */ | |
492 | ||
4246a0b6 CH |
493 | if (bio->bi_error) |
494 | s->iop.error = bio->bi_error; | |
d56d000a KO |
495 | else if (!KEY_DIRTY(&b->key) && |
496 | ptr_stale(s->iop.c, &b->key, 0)) { | |
220bb38c KO |
497 | atomic_long_inc(&s->iop.c->cache_read_races); |
498 | s->iop.error = -EINTR; | |
cafe5635 KO |
499 | } |
500 | ||
4246a0b6 | 501 | bch_bbio_endio(s->iop.c, bio, bio->bi_error, "reading from cache"); |
cafe5635 KO |
502 | } |
503 | ||
2c1953e2 KO |
504 | /* |
505 | * Read from a single key, handling the initial cache miss if the key starts in | |
506 | * the middle of the bio | |
507 | */ | |
cc231966 | 508 | static int cache_lookup_fn(struct btree_op *op, struct btree *b, struct bkey *k) |
2c1953e2 KO |
509 | { |
510 | struct search *s = container_of(op, struct search, op); | |
cc231966 KO |
511 | struct bio *n, *bio = &s->bio.bio; |
512 | struct bkey *bio_key; | |
2c1953e2 | 513 | unsigned ptr; |
2c1953e2 | 514 | |
4f024f37 | 515 | if (bkey_cmp(k, &KEY(s->iop.inode, bio->bi_iter.bi_sector, 0)) <= 0) |
cc231966 KO |
516 | return MAP_CONTINUE; |
517 | ||
220bb38c | 518 | if (KEY_INODE(k) != s->iop.inode || |
4f024f37 | 519 | KEY_START(k) > bio->bi_iter.bi_sector) { |
cc231966 | 520 | unsigned bio_sectors = bio_sectors(bio); |
220bb38c | 521 | unsigned sectors = KEY_INODE(k) == s->iop.inode |
cc231966 | 522 | ? min_t(uint64_t, INT_MAX, |
4f024f37 | 523 | KEY_START(k) - bio->bi_iter.bi_sector) |
cc231966 KO |
524 | : INT_MAX; |
525 | ||
526 | int ret = s->d->cache_miss(b, s, bio, sectors); | |
527 | if (ret != MAP_CONTINUE) | |
528 | return ret; | |
529 | ||
530 | /* if this was a complete miss we shouldn't get here */ | |
531 | BUG_ON(bio_sectors <= sectors); | |
532 | } | |
533 | ||
534 | if (!KEY_SIZE(k)) | |
535 | return MAP_CONTINUE; | |
2c1953e2 KO |
536 | |
537 | /* XXX: figure out best pointer - for multiple cache devices */ | |
538 | ptr = 0; | |
539 | ||
540 | PTR_BUCKET(b->c, k, ptr)->prio = INITIAL_PRIO; | |
541 | ||
5ceaaad7 KO |
542 | if (KEY_DIRTY(k)) |
543 | s->read_dirty_data = true; | |
544 | ||
20d0189b KO |
545 | n = bio_next_split(bio, min_t(uint64_t, INT_MAX, |
546 | KEY_OFFSET(k) - bio->bi_iter.bi_sector), | |
547 | GFP_NOIO, s->d->bio_split); | |
2c1953e2 | 548 | |
cc231966 KO |
549 | bio_key = &container_of(n, struct bbio, bio)->key; |
550 | bch_bkey_copy_single_ptr(bio_key, k, ptr); | |
2c1953e2 | 551 | |
4f024f37 | 552 | bch_cut_front(&KEY(s->iop.inode, n->bi_iter.bi_sector, 0), bio_key); |
220bb38c | 553 | bch_cut_back(&KEY(s->iop.inode, bio_end_sector(n), 0), bio_key); |
2c1953e2 | 554 | |
cc231966 KO |
555 | n->bi_end_io = bch_cache_read_endio; |
556 | n->bi_private = &s->cl; | |
2c1953e2 | 557 | |
cc231966 KO |
558 | /* |
559 | * The bucket we're reading from might be reused while our bio | |
560 | * is in flight, and we could then end up reading the wrong | |
561 | * data. | |
562 | * | |
563 | * We guard against this by checking (in cache_read_endio()) if | |
564 | * the pointer is stale again; if so, we treat it as an error | |
565 | * and reread from the backing device (but we don't pass that | |
566 | * error up anywhere). | |
567 | */ | |
2c1953e2 | 568 | |
cc231966 KO |
569 | __bch_submit_bbio(n, b->c); |
570 | return n == bio ? MAP_DONE : MAP_CONTINUE; | |
2c1953e2 KO |
571 | } |
572 | ||
573 | static void cache_lookup(struct closure *cl) | |
574 | { | |
220bb38c | 575 | struct search *s = container_of(cl, struct search, iop.cl); |
2c1953e2 | 576 | struct bio *bio = &s->bio.bio; |
a5ae4300 | 577 | int ret; |
2c1953e2 | 578 | |
a5ae4300 | 579 | bch_btree_op_init(&s->op, -1); |
2c1953e2 | 580 | |
a5ae4300 KO |
581 | ret = bch_btree_map_keys(&s->op, s->iop.c, |
582 | &KEY(s->iop.inode, bio->bi_iter.bi_sector, 0), | |
583 | cache_lookup_fn, MAP_END_KEY); | |
77b5a084 | 584 | if (ret == -EAGAIN) { |
2c1953e2 | 585 | continue_at(cl, cache_lookup, bcache_wq); |
77b5a084 JA |
586 | return; |
587 | } | |
2c1953e2 KO |
588 | |
589 | closure_return(cl); | |
590 | } | |
591 | ||
592 | /* Common code for the make_request functions */ | |
593 | ||
4246a0b6 | 594 | static void request_endio(struct bio *bio) |
2c1953e2 KO |
595 | { |
596 | struct closure *cl = bio->bi_private; | |
597 | ||
4246a0b6 | 598 | if (bio->bi_error) { |
2c1953e2 | 599 | struct search *s = container_of(cl, struct search, cl); |
4246a0b6 | 600 | s->iop.error = bio->bi_error; |
2c1953e2 KO |
601 | /* Only cache read errors are recoverable */ |
602 | s->recoverable = false; | |
603 | } | |
604 | ||
605 | bio_put(bio); | |
606 | closure_put(cl); | |
607 | } | |
608 | ||
cafe5635 KO |
609 | static void bio_complete(struct search *s) |
610 | { | |
611 | if (s->orig_bio) { | |
aae4933d GZ |
612 | generic_end_io_acct(bio_data_dir(s->orig_bio), |
613 | &s->d->disk->part0, s->start_time); | |
cafe5635 | 614 | |
220bb38c | 615 | trace_bcache_request_end(s->d, s->orig_bio); |
4246a0b6 CH |
616 | s->orig_bio->bi_error = s->iop.error; |
617 | bio_endio(s->orig_bio); | |
cafe5635 KO |
618 | s->orig_bio = NULL; |
619 | } | |
620 | } | |
621 | ||
a5ae4300 | 622 | static void do_bio_hook(struct search *s, struct bio *orig_bio) |
cafe5635 KO |
623 | { |
624 | struct bio *bio = &s->bio.bio; | |
cafe5635 | 625 | |
ed9c47be | 626 | bio_init(bio); |
a5ae4300 | 627 | __bio_clone_fast(bio, orig_bio); |
cafe5635 KO |
628 | bio->bi_end_io = request_endio; |
629 | bio->bi_private = &s->cl; | |
ed9c47be | 630 | |
dac56212 | 631 | bio_cnt_set(bio, 3); |
cafe5635 KO |
632 | } |
633 | ||
634 | static void search_free(struct closure *cl) | |
635 | { | |
636 | struct search *s = container_of(cl, struct search, cl); | |
637 | bio_complete(s); | |
638 | ||
220bb38c KO |
639 | if (s->iop.bio) |
640 | bio_put(s->iop.bio); | |
cafe5635 | 641 | |
cafe5635 KO |
642 | closure_debug_destroy(cl); |
643 | mempool_free(s, s->d->c->search); | |
644 | } | |
645 | ||
a5ae4300 KO |
646 | static inline struct search *search_alloc(struct bio *bio, |
647 | struct bcache_device *d) | |
cafe5635 | 648 | { |
0b93207a | 649 | struct search *s; |
0b93207a KO |
650 | |
651 | s = mempool_alloc(d->c->search, GFP_NOIO); | |
cafe5635 | 652 | |
a5ae4300 KO |
653 | closure_init(&s->cl, NULL); |
654 | do_bio_hook(s, bio); | |
cafe5635 | 655 | |
cafe5635 | 656 | s->orig_bio = bio; |
a5ae4300 KO |
657 | s->cache_miss = NULL; |
658 | s->d = d; | |
cafe5635 | 659 | s->recoverable = 1; |
a5ae4300 KO |
660 | s->write = (bio->bi_rw & REQ_WRITE) != 0; |
661 | s->read_dirty_data = 0; | |
cafe5635 | 662 | s->start_time = jiffies; |
a5ae4300 KO |
663 | |
664 | s->iop.c = d->c; | |
665 | s->iop.bio = NULL; | |
666 | s->iop.inode = d->id; | |
667 | s->iop.write_point = hash_long((unsigned long) current, 16); | |
668 | s->iop.write_prio = 0; | |
669 | s->iop.error = 0; | |
670 | s->iop.flags = 0; | |
671 | s->iop.flush_journal = (bio->bi_rw & (REQ_FLUSH|REQ_FUA)) != 0; | |
da415a09 | 672 | s->iop.wq = bcache_wq; |
cafe5635 | 673 | |
cafe5635 KO |
674 | return s; |
675 | } | |
676 | ||
cafe5635 KO |
677 | /* Cached devices */ |
678 | ||
679 | static void cached_dev_bio_complete(struct closure *cl) | |
680 | { | |
681 | struct search *s = container_of(cl, struct search, cl); | |
682 | struct cached_dev *dc = container_of(s->d, struct cached_dev, disk); | |
683 | ||
684 | search_free(cl); | |
685 | cached_dev_put(dc); | |
686 | } | |
687 | ||
688 | /* Process reads */ | |
689 | ||
cdd972b1 | 690 | static void cached_dev_cache_miss_done(struct closure *cl) |
cafe5635 KO |
691 | { |
692 | struct search *s = container_of(cl, struct search, cl); | |
693 | ||
220bb38c KO |
694 | if (s->iop.replace_collision) |
695 | bch_mark_cache_miss_collision(s->iop.c, s->d); | |
cafe5635 | 696 | |
220bb38c | 697 | if (s->iop.bio) { |
cafe5635 KO |
698 | int i; |
699 | struct bio_vec *bv; | |
700 | ||
220bb38c | 701 | bio_for_each_segment_all(bv, s->iop.bio, i) |
cafe5635 KO |
702 | __free_page(bv->bv_page); |
703 | } | |
704 | ||
705 | cached_dev_bio_complete(cl); | |
706 | } | |
707 | ||
cdd972b1 | 708 | static void cached_dev_read_error(struct closure *cl) |
cafe5635 KO |
709 | { |
710 | struct search *s = container_of(cl, struct search, cl); | |
cdd972b1 | 711 | struct bio *bio = &s->bio.bio; |
cafe5635 KO |
712 | |
713 | if (s->recoverable) { | |
c37511b8 KO |
714 | /* Retry from the backing device: */ |
715 | trace_bcache_read_retry(s->orig_bio); | |
cafe5635 | 716 | |
220bb38c | 717 | s->iop.error = 0; |
a5ae4300 | 718 | do_bio_hook(s, s->orig_bio); |
cafe5635 KO |
719 | |
720 | /* XXX: invalidate cache */ | |
721 | ||
cdd972b1 | 722 | closure_bio_submit(bio, cl, s->d); |
cafe5635 KO |
723 | } |
724 | ||
cdd972b1 | 725 | continue_at(cl, cached_dev_cache_miss_done, NULL); |
cafe5635 KO |
726 | } |
727 | ||
cdd972b1 | 728 | static void cached_dev_read_done(struct closure *cl) |
cafe5635 KO |
729 | { |
730 | struct search *s = container_of(cl, struct search, cl); | |
731 | struct cached_dev *dc = container_of(s->d, struct cached_dev, disk); | |
732 | ||
733 | /* | |
cdd972b1 KO |
734 | * We had a cache miss; cache_bio now contains data ready to be inserted |
735 | * into the cache. | |
cafe5635 KO |
736 | * |
737 | * First, we copy the data we just read from cache_bio's bounce buffers | |
738 | * to the buffers the original bio pointed to: | |
739 | */ | |
740 | ||
220bb38c KO |
741 | if (s->iop.bio) { |
742 | bio_reset(s->iop.bio); | |
4f024f37 | 743 | s->iop.bio->bi_iter.bi_sector = s->cache_miss->bi_iter.bi_sector; |
220bb38c | 744 | s->iop.bio->bi_bdev = s->cache_miss->bi_bdev; |
4f024f37 | 745 | s->iop.bio->bi_iter.bi_size = s->insert_bio_sectors << 9; |
220bb38c | 746 | bch_bio_map(s->iop.bio, NULL); |
cafe5635 | 747 | |
220bb38c | 748 | bio_copy_data(s->cache_miss, s->iop.bio); |
cafe5635 KO |
749 | |
750 | bio_put(s->cache_miss); | |
751 | s->cache_miss = NULL; | |
752 | } | |
753 | ||
ed9c47be | 754 | if (verify(dc, &s->bio.bio) && s->recoverable && !s->read_dirty_data) |
220bb38c | 755 | bch_data_verify(dc, s->orig_bio); |
cafe5635 KO |
756 | |
757 | bio_complete(s); | |
758 | ||
220bb38c KO |
759 | if (s->iop.bio && |
760 | !test_bit(CACHE_SET_STOPPING, &s->iop.c->flags)) { | |
761 | BUG_ON(!s->iop.replace); | |
762 | closure_call(&s->iop.cl, bch_data_insert, NULL, cl); | |
cafe5635 KO |
763 | } |
764 | ||
cdd972b1 | 765 | continue_at(cl, cached_dev_cache_miss_done, NULL); |
cafe5635 KO |
766 | } |
767 | ||
cdd972b1 | 768 | static void cached_dev_read_done_bh(struct closure *cl) |
cafe5635 KO |
769 | { |
770 | struct search *s = container_of(cl, struct search, cl); | |
771 | struct cached_dev *dc = container_of(s->d, struct cached_dev, disk); | |
772 | ||
220bb38c KO |
773 | bch_mark_cache_accounting(s->iop.c, s->d, |
774 | !s->cache_miss, s->iop.bypass); | |
775 | trace_bcache_read(s->orig_bio, !s->cache_miss, s->iop.bypass); | |
cafe5635 | 776 | |
220bb38c | 777 | if (s->iop.error) |
cdd972b1 | 778 | continue_at_nobarrier(cl, cached_dev_read_error, bcache_wq); |
220bb38c | 779 | else if (s->iop.bio || verify(dc, &s->bio.bio)) |
cdd972b1 | 780 | continue_at_nobarrier(cl, cached_dev_read_done, bcache_wq); |
cafe5635 | 781 | else |
cdd972b1 | 782 | continue_at_nobarrier(cl, cached_dev_bio_complete, NULL); |
cafe5635 KO |
783 | } |
784 | ||
785 | static int cached_dev_cache_miss(struct btree *b, struct search *s, | |
786 | struct bio *bio, unsigned sectors) | |
787 | { | |
2c1953e2 | 788 | int ret = MAP_CONTINUE; |
e7c590eb | 789 | unsigned reada = 0; |
cafe5635 | 790 | struct cached_dev *dc = container_of(s->d, struct cached_dev, disk); |
cdd972b1 | 791 | struct bio *miss, *cache_bio; |
cafe5635 | 792 | |
220bb38c | 793 | if (s->cache_miss || s->iop.bypass) { |
20d0189b | 794 | miss = bio_next_split(bio, sectors, GFP_NOIO, s->d->bio_split); |
2c1953e2 | 795 | ret = miss == bio ? MAP_DONE : MAP_CONTINUE; |
e7c590eb KO |
796 | goto out_submit; |
797 | } | |
cafe5635 | 798 | |
e7c590eb KO |
799 | if (!(bio->bi_rw & REQ_RAHEAD) && |
800 | !(bio->bi_rw & REQ_META) && | |
220bb38c | 801 | s->iop.c->gc_stats.in_use < CUTOFF_CACHE_READA) |
e7c590eb KO |
802 | reada = min_t(sector_t, dc->readahead >> 9, |
803 | bdev_sectors(bio->bi_bdev) - bio_end_sector(bio)); | |
cafe5635 | 804 | |
220bb38c | 805 | s->insert_bio_sectors = min(sectors, bio_sectors(bio) + reada); |
cafe5635 | 806 | |
220bb38c | 807 | s->iop.replace_key = KEY(s->iop.inode, |
4f024f37 | 808 | bio->bi_iter.bi_sector + s->insert_bio_sectors, |
220bb38c | 809 | s->insert_bio_sectors); |
e7c590eb | 810 | |
220bb38c | 811 | ret = bch_btree_insert_check_key(b, &s->op, &s->iop.replace_key); |
e7c590eb KO |
812 | if (ret) |
813 | return ret; | |
814 | ||
220bb38c | 815 | s->iop.replace = true; |
1b207d80 | 816 | |
20d0189b | 817 | miss = bio_next_split(bio, sectors, GFP_NOIO, s->d->bio_split); |
2c1953e2 KO |
818 | |
819 | /* btree_search_recurse()'s btree iterator is no good anymore */ | |
820 | ret = miss == bio ? MAP_DONE : -EINTR; | |
cafe5635 | 821 | |
cdd972b1 | 822 | cache_bio = bio_alloc_bioset(GFP_NOWAIT, |
220bb38c | 823 | DIV_ROUND_UP(s->insert_bio_sectors, PAGE_SECTORS), |
cafe5635 | 824 | dc->disk.bio_split); |
cdd972b1 | 825 | if (!cache_bio) |
cafe5635 KO |
826 | goto out_submit; |
827 | ||
4f024f37 KO |
828 | cache_bio->bi_iter.bi_sector = miss->bi_iter.bi_sector; |
829 | cache_bio->bi_bdev = miss->bi_bdev; | |
830 | cache_bio->bi_iter.bi_size = s->insert_bio_sectors << 9; | |
cafe5635 | 831 | |
cdd972b1 KO |
832 | cache_bio->bi_end_io = request_endio; |
833 | cache_bio->bi_private = &s->cl; | |
cafe5635 | 834 | |
cdd972b1 KO |
835 | bch_bio_map(cache_bio, NULL); |
836 | if (bio_alloc_pages(cache_bio, __GFP_NOWARN|GFP_NOIO)) | |
cafe5635 KO |
837 | goto out_put; |
838 | ||
220bb38c KO |
839 | if (reada) |
840 | bch_mark_cache_readahead(s->iop.c, s->d); | |
841 | ||
cdd972b1 | 842 | s->cache_miss = miss; |
220bb38c | 843 | s->iop.bio = cache_bio; |
cdd972b1 KO |
844 | bio_get(cache_bio); |
845 | closure_bio_submit(cache_bio, &s->cl, s->d); | |
cafe5635 KO |
846 | |
847 | return ret; | |
848 | out_put: | |
cdd972b1 | 849 | bio_put(cache_bio); |
cafe5635 | 850 | out_submit: |
e7c590eb KO |
851 | miss->bi_end_io = request_endio; |
852 | miss->bi_private = &s->cl; | |
cafe5635 KO |
853 | closure_bio_submit(miss, &s->cl, s->d); |
854 | return ret; | |
855 | } | |
856 | ||
cdd972b1 | 857 | static void cached_dev_read(struct cached_dev *dc, struct search *s) |
cafe5635 KO |
858 | { |
859 | struct closure *cl = &s->cl; | |
860 | ||
220bb38c | 861 | closure_call(&s->iop.cl, cache_lookup, NULL, cl); |
cdd972b1 | 862 | continue_at(cl, cached_dev_read_done_bh, NULL); |
cafe5635 KO |
863 | } |
864 | ||
865 | /* Process writes */ | |
866 | ||
867 | static void cached_dev_write_complete(struct closure *cl) | |
868 | { | |
869 | struct search *s = container_of(cl, struct search, cl); | |
870 | struct cached_dev *dc = container_of(s->d, struct cached_dev, disk); | |
871 | ||
872 | up_read_non_owner(&dc->writeback_lock); | |
873 | cached_dev_bio_complete(cl); | |
874 | } | |
875 | ||
cdd972b1 | 876 | static void cached_dev_write(struct cached_dev *dc, struct search *s) |
cafe5635 KO |
877 | { |
878 | struct closure *cl = &s->cl; | |
879 | struct bio *bio = &s->bio.bio; | |
4f024f37 | 880 | struct bkey start = KEY(dc->disk.id, bio->bi_iter.bi_sector, 0); |
84f0db03 | 881 | struct bkey end = KEY(dc->disk.id, bio_end_sector(bio), 0); |
cafe5635 | 882 | |
220bb38c | 883 | bch_keybuf_check_overlapping(&s->iop.c->moving_gc_keys, &start, &end); |
cafe5635 | 884 | |
cafe5635 | 885 | down_read_non_owner(&dc->writeback_lock); |
cafe5635 | 886 | if (bch_keybuf_check_overlapping(&dc->writeback_keys, &start, &end)) { |
84f0db03 KO |
887 | /* |
888 | * We overlap with some dirty data undergoing background | |
889 | * writeback, force this write to writeback | |
890 | */ | |
220bb38c KO |
891 | s->iop.bypass = false; |
892 | s->iop.writeback = true; | |
cafe5635 KO |
893 | } |
894 | ||
84f0db03 KO |
895 | /* |
896 | * Discards aren't _required_ to do anything, so skipping if | |
897 | * check_overlapping returned true is ok | |
898 | * | |
899 | * But check_overlapping drops dirty keys for which io hasn't started, | |
900 | * so we still want to call it. | |
901 | */ | |
cafe5635 | 902 | if (bio->bi_rw & REQ_DISCARD) |
220bb38c | 903 | s->iop.bypass = true; |
cafe5635 | 904 | |
72c27061 KO |
905 | if (should_writeback(dc, s->orig_bio, |
906 | cache_mode(dc, bio), | |
220bb38c KO |
907 | s->iop.bypass)) { |
908 | s->iop.bypass = false; | |
909 | s->iop.writeback = true; | |
72c27061 KO |
910 | } |
911 | ||
220bb38c KO |
912 | if (s->iop.bypass) { |
913 | s->iop.bio = s->orig_bio; | |
914 | bio_get(s->iop.bio); | |
cafe5635 | 915 | |
84f0db03 KO |
916 | if (!(bio->bi_rw & REQ_DISCARD) || |
917 | blk_queue_discard(bdev_get_queue(dc->bdev))) | |
918 | closure_bio_submit(bio, cl, s->d); | |
220bb38c | 919 | } else if (s->iop.writeback) { |
279afbad | 920 | bch_writeback_add(dc); |
220bb38c | 921 | s->iop.bio = bio; |
e49c7c37 | 922 | |
c0f04d88 | 923 | if (bio->bi_rw & REQ_FLUSH) { |
e49c7c37 | 924 | /* Also need to send a flush to the backing device */ |
d4eddd42 | 925 | struct bio *flush = bio_alloc_bioset(GFP_NOIO, 0, |
c0f04d88 | 926 | dc->disk.bio_split); |
e49c7c37 | 927 | |
c0f04d88 KO |
928 | flush->bi_rw = WRITE_FLUSH; |
929 | flush->bi_bdev = bio->bi_bdev; | |
930 | flush->bi_end_io = request_endio; | |
931 | flush->bi_private = cl; | |
932 | ||
933 | closure_bio_submit(flush, cl, s->d); | |
e49c7c37 | 934 | } |
84f0db03 | 935 | } else { |
59d276fe | 936 | s->iop.bio = bio_clone_fast(bio, GFP_NOIO, dc->disk.bio_split); |
84f0db03 KO |
937 | |
938 | closure_bio_submit(bio, cl, s->d); | |
cafe5635 | 939 | } |
84f0db03 | 940 | |
220bb38c | 941 | closure_call(&s->iop.cl, bch_data_insert, NULL, cl); |
cafe5635 | 942 | continue_at(cl, cached_dev_write_complete, NULL); |
cafe5635 KO |
943 | } |
944 | ||
a34a8bfd | 945 | static void cached_dev_nodata(struct closure *cl) |
cafe5635 | 946 | { |
a34a8bfd | 947 | struct search *s = container_of(cl, struct search, cl); |
cafe5635 KO |
948 | struct bio *bio = &s->bio.bio; |
949 | ||
220bb38c KO |
950 | if (s->iop.flush_journal) |
951 | bch_journal_meta(s->iop.c, cl); | |
cafe5635 | 952 | |
84f0db03 | 953 | /* If it's a flush, we send the flush to the backing device too */ |
cafe5635 KO |
954 | closure_bio_submit(bio, cl, s->d); |
955 | ||
956 | continue_at(cl, cached_dev_bio_complete, NULL); | |
957 | } | |
958 | ||
959 | /* Cached devices - read & write stuff */ | |
960 | ||
cafe5635 KO |
961 | static void cached_dev_make_request(struct request_queue *q, struct bio *bio) |
962 | { | |
963 | struct search *s; | |
964 | struct bcache_device *d = bio->bi_bdev->bd_disk->private_data; | |
965 | struct cached_dev *dc = container_of(d, struct cached_dev, disk); | |
aae4933d | 966 | int rw = bio_data_dir(bio); |
cafe5635 | 967 | |
aae4933d | 968 | generic_start_io_acct(rw, bio_sectors(bio), &d->disk->part0); |
cafe5635 KO |
969 | |
970 | bio->bi_bdev = dc->bdev; | |
4f024f37 | 971 | bio->bi_iter.bi_sector += dc->sb.data_offset; |
cafe5635 KO |
972 | |
973 | if (cached_dev_get(dc)) { | |
974 | s = search_alloc(bio, d); | |
220bb38c | 975 | trace_bcache_request_start(s->d, bio); |
cafe5635 | 976 | |
4f024f37 | 977 | if (!bio->bi_iter.bi_size) { |
a34a8bfd KO |
978 | /* |
979 | * can't call bch_journal_meta from under | |
980 | * generic_make_request | |
981 | */ | |
982 | continue_at_nobarrier(&s->cl, | |
983 | cached_dev_nodata, | |
984 | bcache_wq); | |
985 | } else { | |
220bb38c | 986 | s->iop.bypass = check_should_bypass(dc, bio); |
84f0db03 KO |
987 | |
988 | if (rw) | |
cdd972b1 | 989 | cached_dev_write(dc, s); |
84f0db03 | 990 | else |
cdd972b1 | 991 | cached_dev_read(dc, s); |
84f0db03 | 992 | } |
cafe5635 KO |
993 | } else { |
994 | if ((bio->bi_rw & REQ_DISCARD) && | |
995 | !blk_queue_discard(bdev_get_queue(dc->bdev))) | |
4246a0b6 | 996 | bio_endio(bio); |
cafe5635 KO |
997 | else |
998 | bch_generic_make_request(bio, &d->bio_split_hook); | |
999 | } | |
1000 | } | |
1001 | ||
1002 | static int cached_dev_ioctl(struct bcache_device *d, fmode_t mode, | |
1003 | unsigned int cmd, unsigned long arg) | |
1004 | { | |
1005 | struct cached_dev *dc = container_of(d, struct cached_dev, disk); | |
1006 | return __blkdev_driver_ioctl(dc->bdev, mode, cmd, arg); | |
1007 | } | |
1008 | ||
1009 | static int cached_dev_congested(void *data, int bits) | |
1010 | { | |
1011 | struct bcache_device *d = data; | |
1012 | struct cached_dev *dc = container_of(d, struct cached_dev, disk); | |
1013 | struct request_queue *q = bdev_get_queue(dc->bdev); | |
1014 | int ret = 0; | |
1015 | ||
1016 | if (bdi_congested(&q->backing_dev_info, bits)) | |
1017 | return 1; | |
1018 | ||
1019 | if (cached_dev_get(dc)) { | |
1020 | unsigned i; | |
1021 | struct cache *ca; | |
1022 | ||
1023 | for_each_cache(ca, d->c, i) { | |
1024 | q = bdev_get_queue(ca->bdev); | |
1025 | ret |= bdi_congested(&q->backing_dev_info, bits); | |
1026 | } | |
1027 | ||
1028 | cached_dev_put(dc); | |
1029 | } | |
1030 | ||
1031 | return ret; | |
1032 | } | |
1033 | ||
1034 | void bch_cached_dev_request_init(struct cached_dev *dc) | |
1035 | { | |
1036 | struct gendisk *g = dc->disk.disk; | |
1037 | ||
1038 | g->queue->make_request_fn = cached_dev_make_request; | |
1039 | g->queue->backing_dev_info.congested_fn = cached_dev_congested; | |
1040 | dc->disk.cache_miss = cached_dev_cache_miss; | |
1041 | dc->disk.ioctl = cached_dev_ioctl; | |
1042 | } | |
1043 | ||
1044 | /* Flash backed devices */ | |
1045 | ||
1046 | static int flash_dev_cache_miss(struct btree *b, struct search *s, | |
1047 | struct bio *bio, unsigned sectors) | |
1048 | { | |
1b4eaf3d | 1049 | unsigned bytes = min(sectors, bio_sectors(bio)) << 9; |
cafe5635 | 1050 | |
1b4eaf3d KO |
1051 | swap(bio->bi_iter.bi_size, bytes); |
1052 | zero_fill_bio(bio); | |
1053 | swap(bio->bi_iter.bi_size, bytes); | |
cafe5635 | 1054 | |
1b4eaf3d | 1055 | bio_advance(bio, bytes); |
8e51e414 | 1056 | |
4f024f37 | 1057 | if (!bio->bi_iter.bi_size) |
2c1953e2 | 1058 | return MAP_DONE; |
cafe5635 | 1059 | |
2c1953e2 | 1060 | return MAP_CONTINUE; |
cafe5635 KO |
1061 | } |
1062 | ||
a34a8bfd KO |
1063 | static void flash_dev_nodata(struct closure *cl) |
1064 | { | |
1065 | struct search *s = container_of(cl, struct search, cl); | |
1066 | ||
220bb38c KO |
1067 | if (s->iop.flush_journal) |
1068 | bch_journal_meta(s->iop.c, cl); | |
a34a8bfd KO |
1069 | |
1070 | continue_at(cl, search_free, NULL); | |
1071 | } | |
1072 | ||
cafe5635 KO |
1073 | static void flash_dev_make_request(struct request_queue *q, struct bio *bio) |
1074 | { | |
1075 | struct search *s; | |
1076 | struct closure *cl; | |
1077 | struct bcache_device *d = bio->bi_bdev->bd_disk->private_data; | |
aae4933d | 1078 | int rw = bio_data_dir(bio); |
cafe5635 | 1079 | |
aae4933d | 1080 | generic_start_io_acct(rw, bio_sectors(bio), &d->disk->part0); |
cafe5635 KO |
1081 | |
1082 | s = search_alloc(bio, d); | |
1083 | cl = &s->cl; | |
1084 | bio = &s->bio.bio; | |
1085 | ||
220bb38c | 1086 | trace_bcache_request_start(s->d, bio); |
cafe5635 | 1087 | |
4f024f37 | 1088 | if (!bio->bi_iter.bi_size) { |
a34a8bfd KO |
1089 | /* |
1090 | * can't call bch_journal_meta from under | |
1091 | * generic_make_request | |
1092 | */ | |
1093 | continue_at_nobarrier(&s->cl, | |
1094 | flash_dev_nodata, | |
1095 | bcache_wq); | |
77b5a084 | 1096 | return; |
84f0db03 | 1097 | } else if (rw) { |
220bb38c | 1098 | bch_keybuf_check_overlapping(&s->iop.c->moving_gc_keys, |
4f024f37 | 1099 | &KEY(d->id, bio->bi_iter.bi_sector, 0), |
8e51e414 | 1100 | &KEY(d->id, bio_end_sector(bio), 0)); |
cafe5635 | 1101 | |
220bb38c KO |
1102 | s->iop.bypass = (bio->bi_rw & REQ_DISCARD) != 0; |
1103 | s->iop.writeback = true; | |
1104 | s->iop.bio = bio; | |
cafe5635 | 1105 | |
220bb38c | 1106 | closure_call(&s->iop.cl, bch_data_insert, NULL, cl); |
cafe5635 | 1107 | } else { |
220bb38c | 1108 | closure_call(&s->iop.cl, cache_lookup, NULL, cl); |
cafe5635 KO |
1109 | } |
1110 | ||
1111 | continue_at(cl, search_free, NULL); | |
1112 | } | |
1113 | ||
1114 | static int flash_dev_ioctl(struct bcache_device *d, fmode_t mode, | |
1115 | unsigned int cmd, unsigned long arg) | |
1116 | { | |
1117 | return -ENOTTY; | |
1118 | } | |
1119 | ||
1120 | static int flash_dev_congested(void *data, int bits) | |
1121 | { | |
1122 | struct bcache_device *d = data; | |
1123 | struct request_queue *q; | |
1124 | struct cache *ca; | |
1125 | unsigned i; | |
1126 | int ret = 0; | |
1127 | ||
1128 | for_each_cache(ca, d->c, i) { | |
1129 | q = bdev_get_queue(ca->bdev); | |
1130 | ret |= bdi_congested(&q->backing_dev_info, bits); | |
1131 | } | |
1132 | ||
1133 | return ret; | |
1134 | } | |
1135 | ||
1136 | void bch_flash_dev_request_init(struct bcache_device *d) | |
1137 | { | |
1138 | struct gendisk *g = d->disk; | |
1139 | ||
1140 | g->queue->make_request_fn = flash_dev_make_request; | |
1141 | g->queue->backing_dev_info.congested_fn = flash_dev_congested; | |
1142 | d->cache_miss = flash_dev_cache_miss; | |
1143 | d->ioctl = flash_dev_ioctl; | |
1144 | } | |
1145 | ||
1146 | void bch_request_exit(void) | |
1147 | { | |
cafe5635 KO |
1148 | if (bch_search_cache) |
1149 | kmem_cache_destroy(bch_search_cache); | |
1150 | } | |
1151 | ||
1152 | int __init bch_request_init(void) | |
1153 | { | |
1154 | bch_search_cache = KMEM_CACHE(search, 0); | |
1155 | if (!bch_search_cache) | |
1156 | return -ENOMEM; | |
1157 | ||
cafe5635 KO |
1158 | return 0; |
1159 | } |