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) { |
4e4cbee9 | 84 | op->status = BLK_STS_RESOURCE; |
220bb38c | 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 | |
4e4cbee9 | 181 | if (bio->bi_status) { |
cafe5635 | 182 | /* TODO: We could try to recover from this. */ |
220bb38c | 183 | if (op->writeback) |
4e4cbee9 | 184 | op->status = bio->bi_status; |
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 | ||
4e4cbee9 | 191 | bch_bbio_endio(op->c, bio, bio->bi_status, "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 | |
be628be0 | 199 | if (atomic_sub_return(bio_sectors(bio), &op->c->sectors_to_gc) < 0) |
220bb38c | 200 | wake_up_gc(op->c); |
cafe5635 | 201 | |
e3b4825b NS |
202 | if (op->bypass) |
203 | return bch_data_invalidate(cl); | |
204 | ||
54d12f2b | 205 | /* |
28a8f0d3 | 206 | * Journal writes are marked REQ_PREFLUSH; if the original write was a |
54d12f2b KO |
207 | * flush, it'll wait on the journal write. |
208 | */ | |
1eff9d32 | 209 | bio->bi_opf &= ~(REQ_PREFLUSH|REQ_FUA); |
54d12f2b | 210 | |
cafe5635 KO |
211 | do { |
212 | unsigned i; | |
213 | struct bkey *k; | |
220bb38c | 214 | struct bio_set *split = op->c->bio_split; |
cafe5635 KO |
215 | |
216 | /* 1 for the device pointer and 1 for the chksum */ | |
220bb38c | 217 | if (bch_keylist_realloc(&op->insert_keys, |
085d2a3d | 218 | 3 + (op->csum ? 1 : 0), |
77b5a084 | 219 | op->c)) { |
da415a09 | 220 | continue_at(cl, bch_data_insert_keys, op->wq); |
77b5a084 JA |
221 | return; |
222 | } | |
cafe5635 | 223 | |
220bb38c | 224 | k = op->insert_keys.top; |
cafe5635 | 225 | bkey_init(k); |
220bb38c | 226 | SET_KEY_INODE(k, op->inode); |
4f024f37 | 227 | SET_KEY_OFFSET(k, bio->bi_iter.bi_sector); |
cafe5635 | 228 | |
2599b53b KO |
229 | if (!bch_alloc_sectors(op->c, k, bio_sectors(bio), |
230 | op->write_point, op->write_prio, | |
231 | op->writeback)) | |
cafe5635 KO |
232 | goto err; |
233 | ||
20d0189b | 234 | n = bio_next_split(bio, KEY_SIZE(k), GFP_NOIO, split); |
cafe5635 | 235 | |
a34a8bfd | 236 | n->bi_end_io = bch_data_insert_endio; |
cafe5635 KO |
237 | n->bi_private = cl; |
238 | ||
220bb38c | 239 | if (op->writeback) { |
cafe5635 KO |
240 | SET_KEY_DIRTY(k, true); |
241 | ||
242 | for (i = 0; i < KEY_PTRS(k); i++) | |
220bb38c | 243 | SET_GC_MARK(PTR_BUCKET(op->c, k, i), |
cafe5635 KO |
244 | GC_MARK_DIRTY); |
245 | } | |
246 | ||
220bb38c | 247 | SET_KEY_CSUM(k, op->csum); |
cafe5635 KO |
248 | if (KEY_CSUM(k)) |
249 | bio_csum(n, k); | |
250 | ||
c37511b8 | 251 | trace_bcache_cache_insert(k); |
220bb38c | 252 | bch_keylist_push(&op->insert_keys); |
cafe5635 | 253 | |
ad0d9e76 | 254 | bio_set_op_attrs(n, REQ_OP_WRITE, 0); |
220bb38c | 255 | bch_submit_bbio(n, op->c, k, 0); |
cafe5635 KO |
256 | } while (n != bio); |
257 | ||
220bb38c | 258 | op->insert_data_done = true; |
da415a09 | 259 | continue_at(cl, bch_data_insert_keys, op->wq); |
77b5a084 | 260 | return; |
cafe5635 KO |
261 | err: |
262 | /* bch_alloc_sectors() blocks if s->writeback = true */ | |
220bb38c | 263 | BUG_ON(op->writeback); |
cafe5635 KO |
264 | |
265 | /* | |
266 | * But if it's not a writeback write we'd rather just bail out if | |
267 | * there aren't any buckets ready to write to - it might take awhile and | |
268 | * we might be starving btree writes for gc or something. | |
269 | */ | |
270 | ||
220bb38c | 271 | if (!op->replace) { |
cafe5635 KO |
272 | /* |
273 | * Writethrough write: We can't complete the write until we've | |
274 | * updated the index. But we don't want to delay the write while | |
275 | * we wait for buckets to be freed up, so just invalidate the | |
276 | * rest of the write. | |
277 | */ | |
220bb38c | 278 | op->bypass = true; |
a34a8bfd | 279 | return bch_data_invalidate(cl); |
cafe5635 KO |
280 | } else { |
281 | /* | |
282 | * From a cache miss, we can just insert the keys for the data | |
283 | * we have written or bail out if we didn't do anything. | |
284 | */ | |
220bb38c | 285 | op->insert_data_done = true; |
cafe5635 KO |
286 | bio_put(bio); |
287 | ||
220bb38c | 288 | if (!bch_keylist_empty(&op->insert_keys)) |
da415a09 | 289 | continue_at(cl, bch_data_insert_keys, op->wq); |
cafe5635 KO |
290 | else |
291 | closure_return(cl); | |
292 | } | |
293 | } | |
294 | ||
295 | /** | |
a34a8bfd | 296 | * bch_data_insert - stick some data in the cache |
cafe5635 KO |
297 | * |
298 | * This is the starting point for any data to end up in a cache device; it could | |
299 | * be from a normal write, or a writeback write, or a write to a flash only | |
300 | * volume - it's also used by the moving garbage collector to compact data in | |
301 | * mostly empty buckets. | |
302 | * | |
303 | * It first writes the data to the cache, creating a list of keys to be inserted | |
304 | * (if the data had to be fragmented there will be multiple keys); after the | |
305 | * data is written it calls bch_journal, and after the keys have been added to | |
306 | * the next journal write they're inserted into the btree. | |
307 | * | |
c18536a7 | 308 | * It inserts the data in s->cache_bio; bi_sector is used for the key offset, |
cafe5635 KO |
309 | * and op->inode is used for the key inode. |
310 | * | |
c18536a7 KO |
311 | * If s->bypass is true, instead of inserting the data it invalidates the |
312 | * region of the cache represented by s->cache_bio and op->inode. | |
cafe5635 | 313 | */ |
a34a8bfd | 314 | void bch_data_insert(struct closure *cl) |
cafe5635 | 315 | { |
220bb38c | 316 | struct data_insert_op *op = container_of(cl, struct data_insert_op, cl); |
cafe5635 | 317 | |
60ae81ee SP |
318 | trace_bcache_write(op->c, op->inode, op->bio, |
319 | op->writeback, op->bypass); | |
220bb38c KO |
320 | |
321 | bch_keylist_init(&op->insert_keys); | |
322 | bio_get(op->bio); | |
a34a8bfd | 323 | bch_data_insert_start(cl); |
cafe5635 KO |
324 | } |
325 | ||
220bb38c KO |
326 | /* Congested? */ |
327 | ||
328 | unsigned bch_get_congested(struct cache_set *c) | |
329 | { | |
330 | int i; | |
331 | long rand; | |
332 | ||
333 | if (!c->congested_read_threshold_us && | |
334 | !c->congested_write_threshold_us) | |
335 | return 0; | |
336 | ||
337 | i = (local_clock_us() - c->congested_last_us) / 1024; | |
338 | if (i < 0) | |
339 | return 0; | |
340 | ||
341 | i += atomic_read(&c->congested); | |
342 | if (i >= 0) | |
343 | return 0; | |
344 | ||
345 | i += CONGESTED_MAX; | |
346 | ||
347 | if (i > 0) | |
348 | i = fract_exp_two(i, 6); | |
349 | ||
350 | rand = get_random_int(); | |
351 | i -= bitmap_weight(&rand, BITS_PER_LONG); | |
352 | ||
353 | return i > 0 ? i : 1; | |
354 | } | |
355 | ||
356 | static void add_sequential(struct task_struct *t) | |
357 | { | |
358 | ewma_add(t->sequential_io_avg, | |
359 | t->sequential_io, 8, 0); | |
360 | ||
361 | t->sequential_io = 0; | |
362 | } | |
363 | ||
364 | static struct hlist_head *iohash(struct cached_dev *dc, uint64_t k) | |
365 | { | |
366 | return &dc->io_hash[hash_64(k, RECENT_IO_BITS)]; | |
367 | } | |
368 | ||
369 | static bool check_should_bypass(struct cached_dev *dc, struct bio *bio) | |
370 | { | |
371 | struct cache_set *c = dc->disk.c; | |
372 | unsigned mode = cache_mode(dc, bio); | |
373 | unsigned sectors, congested = bch_get_congested(c); | |
374 | struct task_struct *task = current; | |
8aee1220 | 375 | struct io *i; |
220bb38c | 376 | |
c4d951dd | 377 | if (test_bit(BCACHE_DEV_DETACHING, &dc->disk.flags) || |
220bb38c | 378 | c->gc_stats.in_use > CUTOFF_CACHE_ADD || |
ad0d9e76 | 379 | (bio_op(bio) == REQ_OP_DISCARD)) |
220bb38c KO |
380 | goto skip; |
381 | ||
382 | if (mode == CACHE_MODE_NONE || | |
383 | (mode == CACHE_MODE_WRITEAROUND && | |
c8d93247 | 384 | op_is_write(bio_op(bio)))) |
220bb38c KO |
385 | goto skip; |
386 | ||
4f024f37 | 387 | if (bio->bi_iter.bi_sector & (c->sb.block_size - 1) || |
220bb38c KO |
388 | bio_sectors(bio) & (c->sb.block_size - 1)) { |
389 | pr_debug("skipping unaligned io"); | |
390 | goto skip; | |
391 | } | |
392 | ||
5ceaaad7 KO |
393 | if (bypass_torture_test(dc)) { |
394 | if ((get_random_int() & 3) == 3) | |
395 | goto skip; | |
396 | else | |
397 | goto rescale; | |
398 | } | |
399 | ||
220bb38c KO |
400 | if (!congested && !dc->sequential_cutoff) |
401 | goto rescale; | |
402 | ||
403 | if (!congested && | |
404 | mode == CACHE_MODE_WRITEBACK && | |
83b5df67 CH |
405 | op_is_write(bio->bi_opf) && |
406 | op_is_sync(bio->bi_opf)) | |
220bb38c KO |
407 | goto rescale; |
408 | ||
8aee1220 | 409 | spin_lock(&dc->io_lock); |
220bb38c | 410 | |
4f024f37 KO |
411 | hlist_for_each_entry(i, iohash(dc, bio->bi_iter.bi_sector), hash) |
412 | if (i->last == bio->bi_iter.bi_sector && | |
8aee1220 KO |
413 | time_before(jiffies, i->jiffies)) |
414 | goto found; | |
220bb38c | 415 | |
8aee1220 | 416 | i = list_first_entry(&dc->io_lru, struct io, lru); |
220bb38c | 417 | |
8aee1220 KO |
418 | add_sequential(task); |
419 | i->sequential = 0; | |
220bb38c | 420 | found: |
4f024f37 KO |
421 | if (i->sequential + bio->bi_iter.bi_size > i->sequential) |
422 | i->sequential += bio->bi_iter.bi_size; | |
220bb38c | 423 | |
8aee1220 KO |
424 | i->last = bio_end_sector(bio); |
425 | i->jiffies = jiffies + msecs_to_jiffies(5000); | |
426 | task->sequential_io = i->sequential; | |
220bb38c | 427 | |
8aee1220 KO |
428 | hlist_del(&i->hash); |
429 | hlist_add_head(&i->hash, iohash(dc, i->last)); | |
430 | list_move_tail(&i->lru, &dc->io_lru); | |
220bb38c | 431 | |
8aee1220 | 432 | spin_unlock(&dc->io_lock); |
220bb38c KO |
433 | |
434 | sectors = max(task->sequential_io, | |
435 | task->sequential_io_avg) >> 9; | |
436 | ||
437 | if (dc->sequential_cutoff && | |
438 | sectors >= dc->sequential_cutoff >> 9) { | |
439 | trace_bcache_bypass_sequential(bio); | |
440 | goto skip; | |
441 | } | |
442 | ||
443 | if (congested && sectors >= congested) { | |
444 | trace_bcache_bypass_congested(bio); | |
445 | goto skip; | |
446 | } | |
447 | ||
448 | rescale: | |
449 | bch_rescale_priorities(c, bio_sectors(bio)); | |
450 | return false; | |
451 | skip: | |
452 | bch_mark_sectors_bypassed(c, dc, bio_sectors(bio)); | |
453 | return true; | |
454 | } | |
455 | ||
2c1953e2 | 456 | /* Cache lookup */ |
cafe5635 | 457 | |
220bb38c KO |
458 | struct search { |
459 | /* Stack frame for bio_complete */ | |
460 | struct closure cl; | |
461 | ||
220bb38c KO |
462 | struct bbio bio; |
463 | struct bio *orig_bio; | |
464 | struct bio *cache_miss; | |
a5ae4300 | 465 | struct bcache_device *d; |
220bb38c KO |
466 | |
467 | unsigned insert_bio_sectors; | |
220bb38c | 468 | unsigned recoverable:1; |
220bb38c | 469 | unsigned write:1; |
5ceaaad7 | 470 | unsigned read_dirty_data:1; |
220bb38c KO |
471 | |
472 | unsigned long start_time; | |
473 | ||
474 | struct btree_op op; | |
475 | struct data_insert_op iop; | |
476 | }; | |
477 | ||
4246a0b6 | 478 | static void bch_cache_read_endio(struct bio *bio) |
cafe5635 KO |
479 | { |
480 | struct bbio *b = container_of(bio, struct bbio, bio); | |
481 | struct closure *cl = bio->bi_private; | |
482 | struct search *s = container_of(cl, struct search, cl); | |
483 | ||
484 | /* | |
485 | * If the bucket was reused while our bio was in flight, we might have | |
486 | * read the wrong data. Set s->error but not error so it doesn't get | |
487 | * counted against the cache device, but we'll still reread the data | |
488 | * from the backing device. | |
489 | */ | |
490 | ||
4e4cbee9 CH |
491 | if (bio->bi_status) |
492 | s->iop.status = bio->bi_status; | |
d56d000a KO |
493 | else if (!KEY_DIRTY(&b->key) && |
494 | ptr_stale(s->iop.c, &b->key, 0)) { | |
220bb38c | 495 | atomic_long_inc(&s->iop.c->cache_read_races); |
4e4cbee9 | 496 | s->iop.status = BLK_STS_IOERR; |
cafe5635 KO |
497 | } |
498 | ||
4e4cbee9 | 499 | bch_bbio_endio(s->iop.c, bio, bio->bi_status, "reading from cache"); |
cafe5635 KO |
500 | } |
501 | ||
2c1953e2 KO |
502 | /* |
503 | * Read from a single key, handling the initial cache miss if the key starts in | |
504 | * the middle of the bio | |
505 | */ | |
cc231966 | 506 | static int cache_lookup_fn(struct btree_op *op, struct btree *b, struct bkey *k) |
2c1953e2 KO |
507 | { |
508 | struct search *s = container_of(op, struct search, op); | |
cc231966 KO |
509 | struct bio *n, *bio = &s->bio.bio; |
510 | struct bkey *bio_key; | |
2c1953e2 | 511 | unsigned ptr; |
2c1953e2 | 512 | |
4f024f37 | 513 | if (bkey_cmp(k, &KEY(s->iop.inode, bio->bi_iter.bi_sector, 0)) <= 0) |
cc231966 KO |
514 | return MAP_CONTINUE; |
515 | ||
220bb38c | 516 | if (KEY_INODE(k) != s->iop.inode || |
4f024f37 | 517 | KEY_START(k) > bio->bi_iter.bi_sector) { |
cc231966 | 518 | unsigned bio_sectors = bio_sectors(bio); |
220bb38c | 519 | unsigned sectors = KEY_INODE(k) == s->iop.inode |
cc231966 | 520 | ? min_t(uint64_t, INT_MAX, |
4f024f37 | 521 | KEY_START(k) - bio->bi_iter.bi_sector) |
cc231966 KO |
522 | : INT_MAX; |
523 | ||
524 | int ret = s->d->cache_miss(b, s, bio, sectors); | |
525 | if (ret != MAP_CONTINUE) | |
526 | return ret; | |
527 | ||
528 | /* if this was a complete miss we shouldn't get here */ | |
529 | BUG_ON(bio_sectors <= sectors); | |
530 | } | |
531 | ||
532 | if (!KEY_SIZE(k)) | |
533 | return MAP_CONTINUE; | |
2c1953e2 KO |
534 | |
535 | /* XXX: figure out best pointer - for multiple cache devices */ | |
536 | ptr = 0; | |
537 | ||
538 | PTR_BUCKET(b->c, k, ptr)->prio = INITIAL_PRIO; | |
539 | ||
5ceaaad7 KO |
540 | if (KEY_DIRTY(k)) |
541 | s->read_dirty_data = true; | |
542 | ||
20d0189b KO |
543 | n = bio_next_split(bio, min_t(uint64_t, INT_MAX, |
544 | KEY_OFFSET(k) - bio->bi_iter.bi_sector), | |
545 | GFP_NOIO, s->d->bio_split); | |
2c1953e2 | 546 | |
cc231966 KO |
547 | bio_key = &container_of(n, struct bbio, bio)->key; |
548 | bch_bkey_copy_single_ptr(bio_key, k, ptr); | |
2c1953e2 | 549 | |
4f024f37 | 550 | bch_cut_front(&KEY(s->iop.inode, n->bi_iter.bi_sector, 0), bio_key); |
220bb38c | 551 | bch_cut_back(&KEY(s->iop.inode, bio_end_sector(n), 0), bio_key); |
2c1953e2 | 552 | |
cc231966 KO |
553 | n->bi_end_io = bch_cache_read_endio; |
554 | n->bi_private = &s->cl; | |
2c1953e2 | 555 | |
cc231966 KO |
556 | /* |
557 | * The bucket we're reading from might be reused while our bio | |
558 | * is in flight, and we could then end up reading the wrong | |
559 | * data. | |
560 | * | |
561 | * We guard against this by checking (in cache_read_endio()) if | |
562 | * the pointer is stale again; if so, we treat it as an error | |
563 | * and reread from the backing device (but we don't pass that | |
564 | * error up anywhere). | |
565 | */ | |
2c1953e2 | 566 | |
cc231966 KO |
567 | __bch_submit_bbio(n, b->c); |
568 | return n == bio ? MAP_DONE : MAP_CONTINUE; | |
2c1953e2 KO |
569 | } |
570 | ||
571 | static void cache_lookup(struct closure *cl) | |
572 | { | |
220bb38c | 573 | struct search *s = container_of(cl, struct search, iop.cl); |
2c1953e2 | 574 | struct bio *bio = &s->bio.bio; |
a5ae4300 | 575 | int ret; |
2c1953e2 | 576 | |
a5ae4300 | 577 | bch_btree_op_init(&s->op, -1); |
2c1953e2 | 578 | |
a5ae4300 KO |
579 | ret = bch_btree_map_keys(&s->op, s->iop.c, |
580 | &KEY(s->iop.inode, bio->bi_iter.bi_sector, 0), | |
581 | cache_lookup_fn, MAP_END_KEY); | |
77b5a084 | 582 | if (ret == -EAGAIN) { |
2c1953e2 | 583 | continue_at(cl, cache_lookup, bcache_wq); |
77b5a084 JA |
584 | return; |
585 | } | |
2c1953e2 KO |
586 | |
587 | closure_return(cl); | |
588 | } | |
589 | ||
590 | /* Common code for the make_request functions */ | |
591 | ||
4246a0b6 | 592 | static void request_endio(struct bio *bio) |
2c1953e2 KO |
593 | { |
594 | struct closure *cl = bio->bi_private; | |
595 | ||
4e4cbee9 | 596 | if (bio->bi_status) { |
2c1953e2 | 597 | struct search *s = container_of(cl, struct search, cl); |
4e4cbee9 | 598 | s->iop.status = bio->bi_status; |
2c1953e2 KO |
599 | /* Only cache read errors are recoverable */ |
600 | s->recoverable = false; | |
601 | } | |
602 | ||
603 | bio_put(bio); | |
604 | closure_put(cl); | |
605 | } | |
606 | ||
cafe5635 KO |
607 | static void bio_complete(struct search *s) |
608 | { | |
609 | if (s->orig_bio) { | |
aae4933d GZ |
610 | generic_end_io_acct(bio_data_dir(s->orig_bio), |
611 | &s->d->disk->part0, s->start_time); | |
cafe5635 | 612 | |
220bb38c | 613 | trace_bcache_request_end(s->d, s->orig_bio); |
4e4cbee9 | 614 | s->orig_bio->bi_status = s->iop.status; |
4246a0b6 | 615 | bio_endio(s->orig_bio); |
cafe5635 KO |
616 | s->orig_bio = NULL; |
617 | } | |
618 | } | |
619 | ||
a5ae4300 | 620 | static void do_bio_hook(struct search *s, struct bio *orig_bio) |
cafe5635 KO |
621 | { |
622 | struct bio *bio = &s->bio.bio; | |
cafe5635 | 623 | |
3a83f467 | 624 | bio_init(bio, NULL, 0); |
a5ae4300 | 625 | __bio_clone_fast(bio, orig_bio); |
cafe5635 KO |
626 | bio->bi_end_io = request_endio; |
627 | bio->bi_private = &s->cl; | |
ed9c47be | 628 | |
dac56212 | 629 | bio_cnt_set(bio, 3); |
cafe5635 KO |
630 | } |
631 | ||
632 | static void search_free(struct closure *cl) | |
633 | { | |
634 | struct search *s = container_of(cl, struct search, cl); | |
635 | bio_complete(s); | |
636 | ||
220bb38c KO |
637 | if (s->iop.bio) |
638 | bio_put(s->iop.bio); | |
cafe5635 | 639 | |
cafe5635 KO |
640 | closure_debug_destroy(cl); |
641 | mempool_free(s, s->d->c->search); | |
642 | } | |
643 | ||
a5ae4300 KO |
644 | static inline struct search *search_alloc(struct bio *bio, |
645 | struct bcache_device *d) | |
cafe5635 | 646 | { |
0b93207a | 647 | struct search *s; |
0b93207a KO |
648 | |
649 | s = mempool_alloc(d->c->search, GFP_NOIO); | |
cafe5635 | 650 | |
a5ae4300 KO |
651 | closure_init(&s->cl, NULL); |
652 | do_bio_hook(s, bio); | |
cafe5635 | 653 | |
cafe5635 | 654 | s->orig_bio = bio; |
a5ae4300 KO |
655 | s->cache_miss = NULL; |
656 | s->d = d; | |
cafe5635 | 657 | s->recoverable = 1; |
c8d93247 | 658 | s->write = op_is_write(bio_op(bio)); |
a5ae4300 | 659 | s->read_dirty_data = 0; |
cafe5635 | 660 | s->start_time = jiffies; |
a5ae4300 KO |
661 | |
662 | s->iop.c = d->c; | |
663 | s->iop.bio = NULL; | |
664 | s->iop.inode = d->id; | |
665 | s->iop.write_point = hash_long((unsigned long) current, 16); | |
666 | s->iop.write_prio = 0; | |
4e4cbee9 | 667 | s->iop.status = 0; |
a5ae4300 | 668 | s->iop.flags = 0; |
f73f44eb | 669 | s->iop.flush_journal = op_is_flush(bio->bi_opf); |
da415a09 | 670 | s->iop.wq = bcache_wq; |
cafe5635 | 671 | |
cafe5635 KO |
672 | return s; |
673 | } | |
674 | ||
cafe5635 KO |
675 | /* Cached devices */ |
676 | ||
677 | static void cached_dev_bio_complete(struct closure *cl) | |
678 | { | |
679 | struct search *s = container_of(cl, struct search, cl); | |
680 | struct cached_dev *dc = container_of(s->d, struct cached_dev, disk); | |
681 | ||
682 | search_free(cl); | |
683 | cached_dev_put(dc); | |
684 | } | |
685 | ||
686 | /* Process reads */ | |
687 | ||
cdd972b1 | 688 | static void cached_dev_cache_miss_done(struct closure *cl) |
cafe5635 KO |
689 | { |
690 | struct search *s = container_of(cl, struct search, cl); | |
691 | ||
220bb38c KO |
692 | if (s->iop.replace_collision) |
693 | bch_mark_cache_miss_collision(s->iop.c, s->d); | |
cafe5635 | 694 | |
491221f8 GJ |
695 | if (s->iop.bio) |
696 | bio_free_pages(s->iop.bio); | |
cafe5635 KO |
697 | |
698 | cached_dev_bio_complete(cl); | |
699 | } | |
700 | ||
cdd972b1 | 701 | static void cached_dev_read_error(struct closure *cl) |
cafe5635 KO |
702 | { |
703 | struct search *s = container_of(cl, struct search, cl); | |
cdd972b1 | 704 | struct bio *bio = &s->bio.bio; |
cafe5635 KO |
705 | |
706 | if (s->recoverable) { | |
c37511b8 KO |
707 | /* Retry from the backing device: */ |
708 | trace_bcache_read_retry(s->orig_bio); | |
cafe5635 | 709 | |
4e4cbee9 | 710 | s->iop.status = 0; |
a5ae4300 | 711 | do_bio_hook(s, s->orig_bio); |
cafe5635 KO |
712 | |
713 | /* XXX: invalidate cache */ | |
714 | ||
749b61da | 715 | closure_bio_submit(bio, cl); |
cafe5635 KO |
716 | } |
717 | ||
cdd972b1 | 718 | continue_at(cl, cached_dev_cache_miss_done, NULL); |
cafe5635 KO |
719 | } |
720 | ||
cdd972b1 | 721 | static void cached_dev_read_done(struct closure *cl) |
cafe5635 KO |
722 | { |
723 | struct search *s = container_of(cl, struct search, cl); | |
724 | struct cached_dev *dc = container_of(s->d, struct cached_dev, disk); | |
725 | ||
726 | /* | |
cdd972b1 KO |
727 | * We had a cache miss; cache_bio now contains data ready to be inserted |
728 | * into the cache. | |
cafe5635 KO |
729 | * |
730 | * First, we copy the data we just read from cache_bio's bounce buffers | |
731 | * to the buffers the original bio pointed to: | |
732 | */ | |
733 | ||
220bb38c KO |
734 | if (s->iop.bio) { |
735 | bio_reset(s->iop.bio); | |
4f024f37 | 736 | s->iop.bio->bi_iter.bi_sector = s->cache_miss->bi_iter.bi_sector; |
220bb38c | 737 | s->iop.bio->bi_bdev = s->cache_miss->bi_bdev; |
4f024f37 | 738 | s->iop.bio->bi_iter.bi_size = s->insert_bio_sectors << 9; |
220bb38c | 739 | bch_bio_map(s->iop.bio, NULL); |
cafe5635 | 740 | |
220bb38c | 741 | bio_copy_data(s->cache_miss, s->iop.bio); |
cafe5635 KO |
742 | |
743 | bio_put(s->cache_miss); | |
744 | s->cache_miss = NULL; | |
745 | } | |
746 | ||
ed9c47be | 747 | if (verify(dc, &s->bio.bio) && s->recoverable && !s->read_dirty_data) |
220bb38c | 748 | bch_data_verify(dc, s->orig_bio); |
cafe5635 KO |
749 | |
750 | bio_complete(s); | |
751 | ||
220bb38c KO |
752 | if (s->iop.bio && |
753 | !test_bit(CACHE_SET_STOPPING, &s->iop.c->flags)) { | |
754 | BUG_ON(!s->iop.replace); | |
755 | closure_call(&s->iop.cl, bch_data_insert, NULL, cl); | |
cafe5635 KO |
756 | } |
757 | ||
cdd972b1 | 758 | continue_at(cl, cached_dev_cache_miss_done, NULL); |
cafe5635 KO |
759 | } |
760 | ||
cdd972b1 | 761 | static void cached_dev_read_done_bh(struct closure *cl) |
cafe5635 KO |
762 | { |
763 | struct search *s = container_of(cl, struct search, cl); | |
764 | struct cached_dev *dc = container_of(s->d, struct cached_dev, disk); | |
765 | ||
220bb38c KO |
766 | bch_mark_cache_accounting(s->iop.c, s->d, |
767 | !s->cache_miss, s->iop.bypass); | |
768 | trace_bcache_read(s->orig_bio, !s->cache_miss, s->iop.bypass); | |
cafe5635 | 769 | |
4e4cbee9 | 770 | if (s->iop.status) |
cdd972b1 | 771 | continue_at_nobarrier(cl, cached_dev_read_error, bcache_wq); |
220bb38c | 772 | else if (s->iop.bio || verify(dc, &s->bio.bio)) |
cdd972b1 | 773 | continue_at_nobarrier(cl, cached_dev_read_done, bcache_wq); |
cafe5635 | 774 | else |
cdd972b1 | 775 | continue_at_nobarrier(cl, cached_dev_bio_complete, NULL); |
cafe5635 KO |
776 | } |
777 | ||
778 | static int cached_dev_cache_miss(struct btree *b, struct search *s, | |
779 | struct bio *bio, unsigned sectors) | |
780 | { | |
2c1953e2 | 781 | int ret = MAP_CONTINUE; |
e7c590eb | 782 | unsigned reada = 0; |
cafe5635 | 783 | struct cached_dev *dc = container_of(s->d, struct cached_dev, disk); |
cdd972b1 | 784 | struct bio *miss, *cache_bio; |
cafe5635 | 785 | |
220bb38c | 786 | if (s->cache_miss || s->iop.bypass) { |
20d0189b | 787 | miss = bio_next_split(bio, sectors, GFP_NOIO, s->d->bio_split); |
2c1953e2 | 788 | ret = miss == bio ? MAP_DONE : MAP_CONTINUE; |
e7c590eb KO |
789 | goto out_submit; |
790 | } | |
cafe5635 | 791 | |
1eff9d32 JA |
792 | if (!(bio->bi_opf & REQ_RAHEAD) && |
793 | !(bio->bi_opf & REQ_META) && | |
220bb38c | 794 | s->iop.c->gc_stats.in_use < CUTOFF_CACHE_READA) |
e7c590eb KO |
795 | reada = min_t(sector_t, dc->readahead >> 9, |
796 | bdev_sectors(bio->bi_bdev) - bio_end_sector(bio)); | |
cafe5635 | 797 | |
220bb38c | 798 | s->insert_bio_sectors = min(sectors, bio_sectors(bio) + reada); |
cafe5635 | 799 | |
220bb38c | 800 | s->iop.replace_key = KEY(s->iop.inode, |
4f024f37 | 801 | bio->bi_iter.bi_sector + s->insert_bio_sectors, |
220bb38c | 802 | s->insert_bio_sectors); |
e7c590eb | 803 | |
220bb38c | 804 | ret = bch_btree_insert_check_key(b, &s->op, &s->iop.replace_key); |
e7c590eb KO |
805 | if (ret) |
806 | return ret; | |
807 | ||
220bb38c | 808 | s->iop.replace = true; |
1b207d80 | 809 | |
20d0189b | 810 | miss = bio_next_split(bio, sectors, GFP_NOIO, s->d->bio_split); |
2c1953e2 KO |
811 | |
812 | /* btree_search_recurse()'s btree iterator is no good anymore */ | |
813 | ret = miss == bio ? MAP_DONE : -EINTR; | |
cafe5635 | 814 | |
cdd972b1 | 815 | cache_bio = bio_alloc_bioset(GFP_NOWAIT, |
220bb38c | 816 | DIV_ROUND_UP(s->insert_bio_sectors, PAGE_SECTORS), |
cafe5635 | 817 | dc->disk.bio_split); |
cdd972b1 | 818 | if (!cache_bio) |
cafe5635 KO |
819 | goto out_submit; |
820 | ||
4f024f37 KO |
821 | cache_bio->bi_iter.bi_sector = miss->bi_iter.bi_sector; |
822 | cache_bio->bi_bdev = miss->bi_bdev; | |
823 | cache_bio->bi_iter.bi_size = s->insert_bio_sectors << 9; | |
cafe5635 | 824 | |
cdd972b1 KO |
825 | cache_bio->bi_end_io = request_endio; |
826 | cache_bio->bi_private = &s->cl; | |
cafe5635 | 827 | |
cdd972b1 KO |
828 | bch_bio_map(cache_bio, NULL); |
829 | if (bio_alloc_pages(cache_bio, __GFP_NOWARN|GFP_NOIO)) | |
cafe5635 KO |
830 | goto out_put; |
831 | ||
220bb38c KO |
832 | if (reada) |
833 | bch_mark_cache_readahead(s->iop.c, s->d); | |
834 | ||
cdd972b1 | 835 | s->cache_miss = miss; |
220bb38c | 836 | s->iop.bio = cache_bio; |
cdd972b1 | 837 | bio_get(cache_bio); |
749b61da | 838 | closure_bio_submit(cache_bio, &s->cl); |
cafe5635 KO |
839 | |
840 | return ret; | |
841 | out_put: | |
cdd972b1 | 842 | bio_put(cache_bio); |
cafe5635 | 843 | out_submit: |
e7c590eb KO |
844 | miss->bi_end_io = request_endio; |
845 | miss->bi_private = &s->cl; | |
749b61da | 846 | closure_bio_submit(miss, &s->cl); |
cafe5635 KO |
847 | return ret; |
848 | } | |
849 | ||
cdd972b1 | 850 | static void cached_dev_read(struct cached_dev *dc, struct search *s) |
cafe5635 KO |
851 | { |
852 | struct closure *cl = &s->cl; | |
853 | ||
220bb38c | 854 | closure_call(&s->iop.cl, cache_lookup, NULL, cl); |
cdd972b1 | 855 | continue_at(cl, cached_dev_read_done_bh, NULL); |
cafe5635 KO |
856 | } |
857 | ||
858 | /* Process writes */ | |
859 | ||
860 | static void cached_dev_write_complete(struct closure *cl) | |
861 | { | |
862 | struct search *s = container_of(cl, struct search, cl); | |
863 | struct cached_dev *dc = container_of(s->d, struct cached_dev, disk); | |
864 | ||
865 | up_read_non_owner(&dc->writeback_lock); | |
866 | cached_dev_bio_complete(cl); | |
867 | } | |
868 | ||
cdd972b1 | 869 | static void cached_dev_write(struct cached_dev *dc, struct search *s) |
cafe5635 KO |
870 | { |
871 | struct closure *cl = &s->cl; | |
872 | struct bio *bio = &s->bio.bio; | |
4f024f37 | 873 | struct bkey start = KEY(dc->disk.id, bio->bi_iter.bi_sector, 0); |
84f0db03 | 874 | struct bkey end = KEY(dc->disk.id, bio_end_sector(bio), 0); |
cafe5635 | 875 | |
220bb38c | 876 | bch_keybuf_check_overlapping(&s->iop.c->moving_gc_keys, &start, &end); |
cafe5635 | 877 | |
cafe5635 | 878 | down_read_non_owner(&dc->writeback_lock); |
cafe5635 | 879 | if (bch_keybuf_check_overlapping(&dc->writeback_keys, &start, &end)) { |
84f0db03 KO |
880 | /* |
881 | * We overlap with some dirty data undergoing background | |
882 | * writeback, force this write to writeback | |
883 | */ | |
220bb38c KO |
884 | s->iop.bypass = false; |
885 | s->iop.writeback = true; | |
cafe5635 KO |
886 | } |
887 | ||
84f0db03 KO |
888 | /* |
889 | * Discards aren't _required_ to do anything, so skipping if | |
890 | * check_overlapping returned true is ok | |
891 | * | |
892 | * But check_overlapping drops dirty keys for which io hasn't started, | |
893 | * so we still want to call it. | |
894 | */ | |
ad0d9e76 | 895 | if (bio_op(bio) == REQ_OP_DISCARD) |
220bb38c | 896 | s->iop.bypass = true; |
cafe5635 | 897 | |
72c27061 KO |
898 | if (should_writeback(dc, s->orig_bio, |
899 | cache_mode(dc, bio), | |
220bb38c KO |
900 | s->iop.bypass)) { |
901 | s->iop.bypass = false; | |
902 | s->iop.writeback = true; | |
72c27061 KO |
903 | } |
904 | ||
220bb38c KO |
905 | if (s->iop.bypass) { |
906 | s->iop.bio = s->orig_bio; | |
907 | bio_get(s->iop.bio); | |
cafe5635 | 908 | |
ad0d9e76 | 909 | if ((bio_op(bio) != REQ_OP_DISCARD) || |
84f0db03 | 910 | blk_queue_discard(bdev_get_queue(dc->bdev))) |
749b61da | 911 | closure_bio_submit(bio, cl); |
220bb38c | 912 | } else if (s->iop.writeback) { |
279afbad | 913 | bch_writeback_add(dc); |
220bb38c | 914 | s->iop.bio = bio; |
e49c7c37 | 915 | |
1eff9d32 | 916 | if (bio->bi_opf & REQ_PREFLUSH) { |
e49c7c37 | 917 | /* Also need to send a flush to the backing device */ |
d4eddd42 | 918 | struct bio *flush = bio_alloc_bioset(GFP_NOIO, 0, |
c0f04d88 | 919 | dc->disk.bio_split); |
e49c7c37 | 920 | |
c0f04d88 KO |
921 | flush->bi_bdev = bio->bi_bdev; |
922 | flush->bi_end_io = request_endio; | |
923 | flush->bi_private = cl; | |
70fd7614 | 924 | flush->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; |
c0f04d88 | 925 | |
749b61da | 926 | closure_bio_submit(flush, cl); |
e49c7c37 | 927 | } |
84f0db03 | 928 | } else { |
59d276fe | 929 | s->iop.bio = bio_clone_fast(bio, GFP_NOIO, dc->disk.bio_split); |
84f0db03 | 930 | |
749b61da | 931 | closure_bio_submit(bio, cl); |
cafe5635 | 932 | } |
84f0db03 | 933 | |
220bb38c | 934 | closure_call(&s->iop.cl, bch_data_insert, NULL, cl); |
cafe5635 | 935 | continue_at(cl, cached_dev_write_complete, NULL); |
cafe5635 KO |
936 | } |
937 | ||
a34a8bfd | 938 | static void cached_dev_nodata(struct closure *cl) |
cafe5635 | 939 | { |
a34a8bfd | 940 | struct search *s = container_of(cl, struct search, cl); |
cafe5635 KO |
941 | struct bio *bio = &s->bio.bio; |
942 | ||
220bb38c KO |
943 | if (s->iop.flush_journal) |
944 | bch_journal_meta(s->iop.c, cl); | |
cafe5635 | 945 | |
84f0db03 | 946 | /* If it's a flush, we send the flush to the backing device too */ |
749b61da | 947 | closure_bio_submit(bio, cl); |
cafe5635 KO |
948 | |
949 | continue_at(cl, cached_dev_bio_complete, NULL); | |
950 | } | |
951 | ||
952 | /* Cached devices - read & write stuff */ | |
953 | ||
dece1635 JA |
954 | static blk_qc_t cached_dev_make_request(struct request_queue *q, |
955 | struct bio *bio) | |
cafe5635 KO |
956 | { |
957 | struct search *s; | |
958 | struct bcache_device *d = bio->bi_bdev->bd_disk->private_data; | |
959 | struct cached_dev *dc = container_of(d, struct cached_dev, disk); | |
aae4933d | 960 | int rw = bio_data_dir(bio); |
cafe5635 | 961 | |
aae4933d | 962 | generic_start_io_acct(rw, bio_sectors(bio), &d->disk->part0); |
cafe5635 KO |
963 | |
964 | bio->bi_bdev = dc->bdev; | |
4f024f37 | 965 | bio->bi_iter.bi_sector += dc->sb.data_offset; |
cafe5635 KO |
966 | |
967 | if (cached_dev_get(dc)) { | |
968 | s = search_alloc(bio, d); | |
220bb38c | 969 | trace_bcache_request_start(s->d, bio); |
cafe5635 | 970 | |
4f024f37 | 971 | if (!bio->bi_iter.bi_size) { |
a34a8bfd KO |
972 | /* |
973 | * can't call bch_journal_meta from under | |
974 | * generic_make_request | |
975 | */ | |
976 | continue_at_nobarrier(&s->cl, | |
977 | cached_dev_nodata, | |
978 | bcache_wq); | |
979 | } else { | |
220bb38c | 980 | s->iop.bypass = check_should_bypass(dc, bio); |
84f0db03 KO |
981 | |
982 | if (rw) | |
cdd972b1 | 983 | cached_dev_write(dc, s); |
84f0db03 | 984 | else |
cdd972b1 | 985 | cached_dev_read(dc, s); |
84f0db03 | 986 | } |
cafe5635 | 987 | } else { |
ad0d9e76 | 988 | if ((bio_op(bio) == REQ_OP_DISCARD) && |
cafe5635 | 989 | !blk_queue_discard(bdev_get_queue(dc->bdev))) |
4246a0b6 | 990 | bio_endio(bio); |
cafe5635 | 991 | else |
749b61da | 992 | generic_make_request(bio); |
cafe5635 | 993 | } |
dece1635 JA |
994 | |
995 | return BLK_QC_T_NONE; | |
cafe5635 KO |
996 | } |
997 | ||
998 | static int cached_dev_ioctl(struct bcache_device *d, fmode_t mode, | |
999 | unsigned int cmd, unsigned long arg) | |
1000 | { | |
1001 | struct cached_dev *dc = container_of(d, struct cached_dev, disk); | |
1002 | return __blkdev_driver_ioctl(dc->bdev, mode, cmd, arg); | |
1003 | } | |
1004 | ||
1005 | static int cached_dev_congested(void *data, int bits) | |
1006 | { | |
1007 | struct bcache_device *d = data; | |
1008 | struct cached_dev *dc = container_of(d, struct cached_dev, disk); | |
1009 | struct request_queue *q = bdev_get_queue(dc->bdev); | |
1010 | int ret = 0; | |
1011 | ||
dc3b17cc | 1012 | if (bdi_congested(q->backing_dev_info, bits)) |
cafe5635 KO |
1013 | return 1; |
1014 | ||
1015 | if (cached_dev_get(dc)) { | |
1016 | unsigned i; | |
1017 | struct cache *ca; | |
1018 | ||
1019 | for_each_cache(ca, d->c, i) { | |
1020 | q = bdev_get_queue(ca->bdev); | |
dc3b17cc | 1021 | ret |= bdi_congested(q->backing_dev_info, bits); |
cafe5635 KO |
1022 | } |
1023 | ||
1024 | cached_dev_put(dc); | |
1025 | } | |
1026 | ||
1027 | return ret; | |
1028 | } | |
1029 | ||
1030 | void bch_cached_dev_request_init(struct cached_dev *dc) | |
1031 | { | |
1032 | struct gendisk *g = dc->disk.disk; | |
1033 | ||
1034 | g->queue->make_request_fn = cached_dev_make_request; | |
dc3b17cc | 1035 | g->queue->backing_dev_info->congested_fn = cached_dev_congested; |
cafe5635 KO |
1036 | dc->disk.cache_miss = cached_dev_cache_miss; |
1037 | dc->disk.ioctl = cached_dev_ioctl; | |
1038 | } | |
1039 | ||
1040 | /* Flash backed devices */ | |
1041 | ||
1042 | static int flash_dev_cache_miss(struct btree *b, struct search *s, | |
1043 | struct bio *bio, unsigned sectors) | |
1044 | { | |
1b4eaf3d | 1045 | unsigned bytes = min(sectors, bio_sectors(bio)) << 9; |
cafe5635 | 1046 | |
1b4eaf3d KO |
1047 | swap(bio->bi_iter.bi_size, bytes); |
1048 | zero_fill_bio(bio); | |
1049 | swap(bio->bi_iter.bi_size, bytes); | |
cafe5635 | 1050 | |
1b4eaf3d | 1051 | bio_advance(bio, bytes); |
8e51e414 | 1052 | |
4f024f37 | 1053 | if (!bio->bi_iter.bi_size) |
2c1953e2 | 1054 | return MAP_DONE; |
cafe5635 | 1055 | |
2c1953e2 | 1056 | return MAP_CONTINUE; |
cafe5635 KO |
1057 | } |
1058 | ||
a34a8bfd KO |
1059 | static void flash_dev_nodata(struct closure *cl) |
1060 | { | |
1061 | struct search *s = container_of(cl, struct search, cl); | |
1062 | ||
220bb38c KO |
1063 | if (s->iop.flush_journal) |
1064 | bch_journal_meta(s->iop.c, cl); | |
a34a8bfd KO |
1065 | |
1066 | continue_at(cl, search_free, NULL); | |
1067 | } | |
1068 | ||
dece1635 JA |
1069 | static blk_qc_t flash_dev_make_request(struct request_queue *q, |
1070 | struct bio *bio) | |
cafe5635 KO |
1071 | { |
1072 | struct search *s; | |
1073 | struct closure *cl; | |
1074 | struct bcache_device *d = bio->bi_bdev->bd_disk->private_data; | |
aae4933d | 1075 | int rw = bio_data_dir(bio); |
cafe5635 | 1076 | |
aae4933d | 1077 | generic_start_io_acct(rw, bio_sectors(bio), &d->disk->part0); |
cafe5635 KO |
1078 | |
1079 | s = search_alloc(bio, d); | |
1080 | cl = &s->cl; | |
1081 | bio = &s->bio.bio; | |
1082 | ||
220bb38c | 1083 | trace_bcache_request_start(s->d, bio); |
cafe5635 | 1084 | |
4f024f37 | 1085 | if (!bio->bi_iter.bi_size) { |
a34a8bfd KO |
1086 | /* |
1087 | * can't call bch_journal_meta from under | |
1088 | * generic_make_request | |
1089 | */ | |
1090 | continue_at_nobarrier(&s->cl, | |
1091 | flash_dev_nodata, | |
1092 | bcache_wq); | |
dece1635 | 1093 | return BLK_QC_T_NONE; |
84f0db03 | 1094 | } else if (rw) { |
220bb38c | 1095 | bch_keybuf_check_overlapping(&s->iop.c->moving_gc_keys, |
4f024f37 | 1096 | &KEY(d->id, bio->bi_iter.bi_sector, 0), |
8e51e414 | 1097 | &KEY(d->id, bio_end_sector(bio), 0)); |
cafe5635 | 1098 | |
ad0d9e76 | 1099 | s->iop.bypass = (bio_op(bio) == REQ_OP_DISCARD) != 0; |
220bb38c KO |
1100 | s->iop.writeback = true; |
1101 | s->iop.bio = bio; | |
cafe5635 | 1102 | |
220bb38c | 1103 | closure_call(&s->iop.cl, bch_data_insert, NULL, cl); |
cafe5635 | 1104 | } else { |
220bb38c | 1105 | closure_call(&s->iop.cl, cache_lookup, NULL, cl); |
cafe5635 KO |
1106 | } |
1107 | ||
1108 | continue_at(cl, search_free, NULL); | |
dece1635 | 1109 | return BLK_QC_T_NONE; |
cafe5635 KO |
1110 | } |
1111 | ||
1112 | static int flash_dev_ioctl(struct bcache_device *d, fmode_t mode, | |
1113 | unsigned int cmd, unsigned long arg) | |
1114 | { | |
1115 | return -ENOTTY; | |
1116 | } | |
1117 | ||
1118 | static int flash_dev_congested(void *data, int bits) | |
1119 | { | |
1120 | struct bcache_device *d = data; | |
1121 | struct request_queue *q; | |
1122 | struct cache *ca; | |
1123 | unsigned i; | |
1124 | int ret = 0; | |
1125 | ||
1126 | for_each_cache(ca, d->c, i) { | |
1127 | q = bdev_get_queue(ca->bdev); | |
dc3b17cc | 1128 | ret |= bdi_congested(q->backing_dev_info, bits); |
cafe5635 KO |
1129 | } |
1130 | ||
1131 | return ret; | |
1132 | } | |
1133 | ||
1134 | void bch_flash_dev_request_init(struct bcache_device *d) | |
1135 | { | |
1136 | struct gendisk *g = d->disk; | |
1137 | ||
1138 | g->queue->make_request_fn = flash_dev_make_request; | |
dc3b17cc | 1139 | g->queue->backing_dev_info->congested_fn = flash_dev_congested; |
cafe5635 KO |
1140 | d->cache_miss = flash_dev_cache_miss; |
1141 | d->ioctl = flash_dev_ioctl; | |
1142 | } | |
1143 | ||
1144 | void bch_request_exit(void) | |
1145 | { | |
cafe5635 KO |
1146 | if (bch_search_cache) |
1147 | kmem_cache_destroy(bch_search_cache); | |
1148 | } | |
1149 | ||
1150 | int __init bch_request_init(void) | |
1151 | { | |
1152 | bch_search_cache = KMEM_CACHE(search, 0); | |
1153 | if (!bch_search_cache) | |
1154 | return -ENOMEM; | |
1155 | ||
cafe5635 KO |
1156 | return 0; |
1157 | } |