Commit | Line | Data |
---|---|---|
cafe5635 KO |
1 | /* |
2 | * Copyright (C) 2010 Kent Overstreet <kent.overstreet@gmail.com> | |
3 | * | |
4 | * Uses a block device as cache for other block devices; optimized for SSDs. | |
5 | * All allocation is done in buckets, which should match the erase block size | |
6 | * of the device. | |
7 | * | |
8 | * Buckets containing cached data are kept on a heap sorted by priority; | |
9 | * bucket priority is increased on cache hit, and periodically all the buckets | |
10 | * on the heap have their priority scaled down. This currently is just used as | |
11 | * an LRU but in the future should allow for more intelligent heuristics. | |
12 | * | |
13 | * Buckets have an 8 bit counter; freeing is accomplished by incrementing the | |
14 | * counter. Garbage collection is used to remove stale pointers. | |
15 | * | |
16 | * Indexing is done via a btree; nodes are not necessarily fully sorted, rather | |
17 | * as keys are inserted we only sort the pages that have not yet been written. | |
18 | * When garbage collection is run, we resort the entire node. | |
19 | * | |
20 | * All configuration is done via sysfs; see Documentation/bcache.txt. | |
21 | */ | |
22 | ||
23 | #include "bcache.h" | |
24 | #include "btree.h" | |
25 | #include "debug.h" | |
65d45231 | 26 | #include "extents.h" |
cafe5635 KO |
27 | |
28 | #include <linux/slab.h> | |
29 | #include <linux/bitops.h> | |
72a44517 | 30 | #include <linux/freezer.h> |
cafe5635 | 31 | #include <linux/hash.h> |
72a44517 | 32 | #include <linux/kthread.h> |
cd953ed0 | 33 | #include <linux/prefetch.h> |
cafe5635 KO |
34 | #include <linux/random.h> |
35 | #include <linux/rcupdate.h> | |
36 | #include <trace/events/bcache.h> | |
37 | ||
38 | /* | |
39 | * Todo: | |
40 | * register_bcache: Return errors out to userspace correctly | |
41 | * | |
42 | * Writeback: don't undirty key until after a cache flush | |
43 | * | |
44 | * Create an iterator for key pointers | |
45 | * | |
46 | * On btree write error, mark bucket such that it won't be freed from the cache | |
47 | * | |
48 | * Journalling: | |
49 | * Check for bad keys in replay | |
50 | * Propagate barriers | |
51 | * Refcount journal entries in journal_replay | |
52 | * | |
53 | * Garbage collection: | |
54 | * Finish incremental gc | |
55 | * Gc should free old UUIDs, data for invalid UUIDs | |
56 | * | |
57 | * Provide a way to list backing device UUIDs we have data cached for, and | |
58 | * probably how long it's been since we've seen them, and a way to invalidate | |
59 | * dirty data for devices that will never be attached again | |
60 | * | |
61 | * Keep 1 min/5 min/15 min statistics of how busy a block device has been, so | |
62 | * that based on that and how much dirty data we have we can keep writeback | |
63 | * from being starved | |
64 | * | |
65 | * Add a tracepoint or somesuch to watch for writeback starvation | |
66 | * | |
67 | * When btree depth > 1 and splitting an interior node, we have to make sure | |
68 | * alloc_bucket() cannot fail. This should be true but is not completely | |
69 | * obvious. | |
70 | * | |
cafe5635 KO |
71 | * Plugging? |
72 | * | |
73 | * If data write is less than hard sector size of ssd, round up offset in open | |
74 | * bucket to the next whole sector | |
75 | * | |
cafe5635 KO |
76 | * Superblock needs to be fleshed out for multiple cache devices |
77 | * | |
78 | * Add a sysfs tunable for the number of writeback IOs in flight | |
79 | * | |
80 | * Add a sysfs tunable for the number of open data buckets | |
81 | * | |
82 | * IO tracking: Can we track when one process is doing io on behalf of another? | |
83 | * IO tracking: Don't use just an average, weigh more recent stuff higher | |
84 | * | |
85 | * Test module load/unload | |
86 | */ | |
87 | ||
cafe5635 KO |
88 | #define MAX_NEED_GC 64 |
89 | #define MAX_SAVE_PRIO 72 | |
90 | ||
91 | #define PTR_DIRTY_BIT (((uint64_t) 1 << 36)) | |
92 | ||
93 | #define PTR_HASH(c, k) \ | |
94 | (((k)->ptr[0] >> c->bucket_bits) | PTR_GEN(k, 0)) | |
95 | ||
df8e8970 KO |
96 | #define insert_lock(s, b) ((b)->level <= (s)->lock) |
97 | ||
98 | /* | |
99 | * These macros are for recursing down the btree - they handle the details of | |
100 | * locking and looking up nodes in the cache for you. They're best treated as | |
101 | * mere syntax when reading code that uses them. | |
102 | * | |
103 | * op->lock determines whether we take a read or a write lock at a given depth. | |
104 | * If you've got a read lock and find that you need a write lock (i.e. you're | |
105 | * going to have to split), set op->lock and return -EINTR; btree_root() will | |
106 | * call you again and you'll have the correct lock. | |
107 | */ | |
108 | ||
109 | /** | |
110 | * btree - recurse down the btree on a specified key | |
111 | * @fn: function to call, which will be passed the child node | |
112 | * @key: key to recurse on | |
113 | * @b: parent btree node | |
114 | * @op: pointer to struct btree_op | |
115 | */ | |
116 | #define btree(fn, key, b, op, ...) \ | |
117 | ({ \ | |
118 | int _r, l = (b)->level - 1; \ | |
119 | bool _w = l <= (op)->lock; \ | |
2452cc89 SP |
120 | struct btree *_child = bch_btree_node_get((b)->c, op, key, l, \ |
121 | _w, b); \ | |
df8e8970 | 122 | if (!IS_ERR(_child)) { \ |
df8e8970 KO |
123 | _r = bch_btree_ ## fn(_child, op, ##__VA_ARGS__); \ |
124 | rw_unlock(_w, _child); \ | |
125 | } else \ | |
126 | _r = PTR_ERR(_child); \ | |
127 | _r; \ | |
128 | }) | |
129 | ||
130 | /** | |
131 | * btree_root - call a function on the root of the btree | |
132 | * @fn: function to call, which will be passed the child node | |
133 | * @c: cache set | |
134 | * @op: pointer to struct btree_op | |
135 | */ | |
136 | #define btree_root(fn, c, op, ...) \ | |
137 | ({ \ | |
138 | int _r = -EINTR; \ | |
139 | do { \ | |
140 | struct btree *_b = (c)->root; \ | |
141 | bool _w = insert_lock(op, _b); \ | |
142 | rw_lock(_w, _b, _b->level); \ | |
143 | if (_b == (c)->root && \ | |
144 | _w == insert_lock(op, _b)) { \ | |
df8e8970 KO |
145 | _r = bch_btree_ ## fn(_b, op, ##__VA_ARGS__); \ |
146 | } \ | |
147 | rw_unlock(_w, _b); \ | |
0a63b66d | 148 | bch_cannibalize_unlock(c); \ |
78365411 KO |
149 | if (_r == -EINTR) \ |
150 | schedule(); \ | |
df8e8970 KO |
151 | } while (_r == -EINTR); \ |
152 | \ | |
0a63b66d | 153 | finish_wait(&(c)->btree_cache_wait, &(op)->wait); \ |
df8e8970 KO |
154 | _r; \ |
155 | }) | |
156 | ||
a85e968e KO |
157 | static inline struct bset *write_block(struct btree *b) |
158 | { | |
159 | return ((void *) btree_bset_first(b)) + b->written * block_bytes(b->c); | |
160 | } | |
161 | ||
2a285686 KO |
162 | static void bch_btree_init_next(struct btree *b) |
163 | { | |
164 | /* If not a leaf node, always sort */ | |
165 | if (b->level && b->keys.nsets) | |
166 | bch_btree_sort(&b->keys, &b->c->sort); | |
167 | else | |
168 | bch_btree_sort_lazy(&b->keys, &b->c->sort); | |
169 | ||
170 | if (b->written < btree_blocks(b)) | |
171 | bch_bset_init_next(&b->keys, write_block(b), | |
172 | bset_magic(&b->c->sb)); | |
173 | ||
174 | } | |
175 | ||
cafe5635 KO |
176 | /* Btree key manipulation */ |
177 | ||
3a3b6a4e | 178 | void bkey_put(struct cache_set *c, struct bkey *k) |
e7c590eb KO |
179 | { |
180 | unsigned i; | |
181 | ||
182 | for (i = 0; i < KEY_PTRS(k); i++) | |
183 | if (ptr_available(c, k, i)) | |
184 | atomic_dec_bug(&PTR_BUCKET(c, k, i)->pin); | |
185 | } | |
186 | ||
cafe5635 KO |
187 | /* Btree IO */ |
188 | ||
189 | static uint64_t btree_csum_set(struct btree *b, struct bset *i) | |
190 | { | |
191 | uint64_t crc = b->key.ptr[0]; | |
fafff81c | 192 | void *data = (void *) i + 8, *end = bset_bkey_last(i); |
cafe5635 | 193 | |
169ef1cf | 194 | crc = bch_crc64_update(crc, data, end - data); |
c19ed23a | 195 | return crc ^ 0xffffffffffffffffULL; |
cafe5635 KO |
196 | } |
197 | ||
78b77bf8 | 198 | void bch_btree_node_read_done(struct btree *b) |
cafe5635 | 199 | { |
cafe5635 | 200 | const char *err = "bad btree header"; |
ee811287 | 201 | struct bset *i = btree_bset_first(b); |
57943511 | 202 | struct btree_iter *iter; |
cafe5635 | 203 | |
bcf090e0 | 204 | iter = mempool_alloc(b->c->fill_iter, GFP_NOIO); |
57943511 | 205 | iter->size = b->c->sb.bucket_size / b->c->sb.block_size; |
cafe5635 KO |
206 | iter->used = 0; |
207 | ||
280481d0 | 208 | #ifdef CONFIG_BCACHE_DEBUG |
c052dd9a | 209 | iter->b = &b->keys; |
280481d0 KO |
210 | #endif |
211 | ||
57943511 | 212 | if (!i->seq) |
cafe5635 KO |
213 | goto err; |
214 | ||
215 | for (; | |
a85e968e | 216 | b->written < btree_blocks(b) && i->seq == b->keys.set[0].data->seq; |
cafe5635 KO |
217 | i = write_block(b)) { |
218 | err = "unsupported bset version"; | |
219 | if (i->version > BCACHE_BSET_VERSION) | |
220 | goto err; | |
221 | ||
222 | err = "bad btree header"; | |
ee811287 KO |
223 | if (b->written + set_blocks(i, block_bytes(b->c)) > |
224 | btree_blocks(b)) | |
cafe5635 KO |
225 | goto err; |
226 | ||
227 | err = "bad magic"; | |
81ab4190 | 228 | if (i->magic != bset_magic(&b->c->sb)) |
cafe5635 KO |
229 | goto err; |
230 | ||
231 | err = "bad checksum"; | |
232 | switch (i->version) { | |
233 | case 0: | |
234 | if (i->csum != csum_set(i)) | |
235 | goto err; | |
236 | break; | |
237 | case BCACHE_BSET_VERSION: | |
238 | if (i->csum != btree_csum_set(b, i)) | |
239 | goto err; | |
240 | break; | |
241 | } | |
242 | ||
243 | err = "empty set"; | |
a85e968e | 244 | if (i != b->keys.set[0].data && !i->keys) |
cafe5635 KO |
245 | goto err; |
246 | ||
fafff81c | 247 | bch_btree_iter_push(iter, i->start, bset_bkey_last(i)); |
cafe5635 | 248 | |
ee811287 | 249 | b->written += set_blocks(i, block_bytes(b->c)); |
cafe5635 KO |
250 | } |
251 | ||
252 | err = "corrupted btree"; | |
253 | for (i = write_block(b); | |
a85e968e | 254 | bset_sector_offset(&b->keys, i) < KEY_SIZE(&b->key); |
cafe5635 | 255 | i = ((void *) i) + block_bytes(b->c)) |
a85e968e | 256 | if (i->seq == b->keys.set[0].data->seq) |
cafe5635 KO |
257 | goto err; |
258 | ||
a85e968e | 259 | bch_btree_sort_and_fix_extents(&b->keys, iter, &b->c->sort); |
cafe5635 | 260 | |
a85e968e | 261 | i = b->keys.set[0].data; |
cafe5635 | 262 | err = "short btree key"; |
a85e968e KO |
263 | if (b->keys.set[0].size && |
264 | bkey_cmp(&b->key, &b->keys.set[0].end) < 0) | |
cafe5635 KO |
265 | goto err; |
266 | ||
267 | if (b->written < btree_blocks(b)) | |
a85e968e KO |
268 | bch_bset_init_next(&b->keys, write_block(b), |
269 | bset_magic(&b->c->sb)); | |
cafe5635 | 270 | out: |
57943511 KO |
271 | mempool_free(iter, b->c->fill_iter); |
272 | return; | |
cafe5635 KO |
273 | err: |
274 | set_btree_node_io_error(b); | |
88b9f8c4 | 275 | bch_cache_set_error(b->c, "%s at bucket %zu, block %u, %u keys", |
cafe5635 | 276 | err, PTR_BUCKET_NR(b->c, &b->key, 0), |
88b9f8c4 | 277 | bset_block_offset(b, i), i->keys); |
cafe5635 KO |
278 | goto out; |
279 | } | |
280 | ||
57943511 | 281 | static void btree_node_read_endio(struct bio *bio, int error) |
cafe5635 | 282 | { |
57943511 KO |
283 | struct closure *cl = bio->bi_private; |
284 | closure_put(cl); | |
285 | } | |
cafe5635 | 286 | |
78b77bf8 | 287 | static void bch_btree_node_read(struct btree *b) |
57943511 KO |
288 | { |
289 | uint64_t start_time = local_clock(); | |
290 | struct closure cl; | |
291 | struct bio *bio; | |
cafe5635 | 292 | |
c37511b8 KO |
293 | trace_bcache_btree_read(b); |
294 | ||
57943511 | 295 | closure_init_stack(&cl); |
cafe5635 | 296 | |
57943511 KO |
297 | bio = bch_bbio_alloc(b->c); |
298 | bio->bi_rw = REQ_META|READ_SYNC; | |
4f024f37 | 299 | bio->bi_iter.bi_size = KEY_SIZE(&b->key) << 9; |
57943511 KO |
300 | bio->bi_end_io = btree_node_read_endio; |
301 | bio->bi_private = &cl; | |
cafe5635 | 302 | |
a85e968e | 303 | bch_bio_map(bio, b->keys.set[0].data); |
cafe5635 | 304 | |
57943511 KO |
305 | bch_submit_bbio(bio, b->c, &b->key, 0); |
306 | closure_sync(&cl); | |
cafe5635 | 307 | |
57943511 KO |
308 | if (!test_bit(BIO_UPTODATE, &bio->bi_flags)) |
309 | set_btree_node_io_error(b); | |
310 | ||
311 | bch_bbio_free(bio, b->c); | |
312 | ||
313 | if (btree_node_io_error(b)) | |
314 | goto err; | |
315 | ||
316 | bch_btree_node_read_done(b); | |
57943511 | 317 | bch_time_stats_update(&b->c->btree_read_time, start_time); |
57943511 KO |
318 | |
319 | return; | |
320 | err: | |
61cbd250 | 321 | bch_cache_set_error(b->c, "io error reading bucket %zu", |
57943511 | 322 | PTR_BUCKET_NR(b->c, &b->key, 0)); |
cafe5635 KO |
323 | } |
324 | ||
325 | static void btree_complete_write(struct btree *b, struct btree_write *w) | |
326 | { | |
327 | if (w->prio_blocked && | |
328 | !atomic_sub_return(w->prio_blocked, &b->c->prio_blocked)) | |
119ba0f8 | 329 | wake_up_allocators(b->c); |
cafe5635 KO |
330 | |
331 | if (w->journal) { | |
332 | atomic_dec_bug(w->journal); | |
333 | __closure_wake_up(&b->c->journal.wait); | |
334 | } | |
335 | ||
cafe5635 KO |
336 | w->prio_blocked = 0; |
337 | w->journal = NULL; | |
cafe5635 KO |
338 | } |
339 | ||
cb7a583e KO |
340 | static void btree_node_write_unlock(struct closure *cl) |
341 | { | |
342 | struct btree *b = container_of(cl, struct btree, io); | |
343 | ||
344 | up(&b->io_mutex); | |
345 | } | |
346 | ||
57943511 | 347 | static void __btree_node_write_done(struct closure *cl) |
cafe5635 | 348 | { |
cb7a583e | 349 | struct btree *b = container_of(cl, struct btree, io); |
cafe5635 KO |
350 | struct btree_write *w = btree_prev_write(b); |
351 | ||
352 | bch_bbio_free(b->bio, b->c); | |
353 | b->bio = NULL; | |
354 | btree_complete_write(b, w); | |
355 | ||
356 | if (btree_node_dirty(b)) | |
56b30770 | 357 | schedule_delayed_work(&b->work, 30 * HZ); |
cafe5635 | 358 | |
cb7a583e | 359 | closure_return_with_destructor(cl, btree_node_write_unlock); |
cafe5635 KO |
360 | } |
361 | ||
57943511 | 362 | static void btree_node_write_done(struct closure *cl) |
cafe5635 | 363 | { |
cb7a583e | 364 | struct btree *b = container_of(cl, struct btree, io); |
cafe5635 KO |
365 | struct bio_vec *bv; |
366 | int n; | |
367 | ||
7988613b | 368 | bio_for_each_segment_all(bv, b->bio, n) |
cafe5635 KO |
369 | __free_page(bv->bv_page); |
370 | ||
57943511 | 371 | __btree_node_write_done(cl); |
cafe5635 KO |
372 | } |
373 | ||
57943511 KO |
374 | static void btree_node_write_endio(struct bio *bio, int error) |
375 | { | |
376 | struct closure *cl = bio->bi_private; | |
cb7a583e | 377 | struct btree *b = container_of(cl, struct btree, io); |
57943511 KO |
378 | |
379 | if (error) | |
380 | set_btree_node_io_error(b); | |
381 | ||
382 | bch_bbio_count_io_errors(b->c, bio, error, "writing btree"); | |
383 | closure_put(cl); | |
384 | } | |
385 | ||
386 | static void do_btree_node_write(struct btree *b) | |
cafe5635 | 387 | { |
cb7a583e | 388 | struct closure *cl = &b->io; |
ee811287 | 389 | struct bset *i = btree_bset_last(b); |
cafe5635 KO |
390 | BKEY_PADDED(key) k; |
391 | ||
392 | i->version = BCACHE_BSET_VERSION; | |
393 | i->csum = btree_csum_set(b, i); | |
394 | ||
57943511 KO |
395 | BUG_ON(b->bio); |
396 | b->bio = bch_bbio_alloc(b->c); | |
397 | ||
398 | b->bio->bi_end_io = btree_node_write_endio; | |
faadf0c9 | 399 | b->bio->bi_private = cl; |
e49c7c37 | 400 | b->bio->bi_rw = REQ_META|WRITE_SYNC|REQ_FUA; |
ee811287 | 401 | b->bio->bi_iter.bi_size = roundup(set_bytes(i), block_bytes(b->c)); |
169ef1cf | 402 | bch_bio_map(b->bio, i); |
cafe5635 | 403 | |
e49c7c37 KO |
404 | /* |
405 | * If we're appending to a leaf node, we don't technically need FUA - | |
406 | * this write just needs to be persisted before the next journal write, | |
407 | * which will be marked FLUSH|FUA. | |
408 | * | |
409 | * Similarly if we're writing a new btree root - the pointer is going to | |
410 | * be in the next journal entry. | |
411 | * | |
412 | * But if we're writing a new btree node (that isn't a root) or | |
413 | * appending to a non leaf btree node, we need either FUA or a flush | |
414 | * when we write the parent with the new pointer. FUA is cheaper than a | |
415 | * flush, and writes appending to leaf nodes aren't blocking anything so | |
416 | * just make all btree node writes FUA to keep things sane. | |
417 | */ | |
418 | ||
cafe5635 | 419 | bkey_copy(&k.key, &b->key); |
ee811287 | 420 | SET_PTR_OFFSET(&k.key, 0, PTR_OFFSET(&k.key, 0) + |
a85e968e | 421 | bset_sector_offset(&b->keys, i)); |
cafe5635 | 422 | |
501d52a9 | 423 | if (!bio_alloc_pages(b->bio, __GFP_NOWARN|GFP_NOWAIT)) { |
cafe5635 KO |
424 | int j; |
425 | struct bio_vec *bv; | |
426 | void *base = (void *) ((unsigned long) i & ~(PAGE_SIZE - 1)); | |
427 | ||
7988613b | 428 | bio_for_each_segment_all(bv, b->bio, j) |
cafe5635 KO |
429 | memcpy(page_address(bv->bv_page), |
430 | base + j * PAGE_SIZE, PAGE_SIZE); | |
431 | ||
cafe5635 KO |
432 | bch_submit_bbio(b->bio, b->c, &k.key, 0); |
433 | ||
57943511 | 434 | continue_at(cl, btree_node_write_done, NULL); |
cafe5635 KO |
435 | } else { |
436 | b->bio->bi_vcnt = 0; | |
169ef1cf | 437 | bch_bio_map(b->bio, i); |
cafe5635 | 438 | |
cafe5635 KO |
439 | bch_submit_bbio(b->bio, b->c, &k.key, 0); |
440 | ||
441 | closure_sync(cl); | |
cb7a583e | 442 | continue_at_nobarrier(cl, __btree_node_write_done, NULL); |
cafe5635 KO |
443 | } |
444 | } | |
445 | ||
2a285686 | 446 | void __bch_btree_node_write(struct btree *b, struct closure *parent) |
cafe5635 | 447 | { |
ee811287 | 448 | struct bset *i = btree_bset_last(b); |
cafe5635 | 449 | |
2a285686 KO |
450 | lockdep_assert_held(&b->write_lock); |
451 | ||
c37511b8 KO |
452 | trace_bcache_btree_write(b); |
453 | ||
cafe5635 | 454 | BUG_ON(current->bio_list); |
57943511 KO |
455 | BUG_ON(b->written >= btree_blocks(b)); |
456 | BUG_ON(b->written && !i->keys); | |
ee811287 | 457 | BUG_ON(btree_bset_first(b)->seq != i->seq); |
dc9d98d6 | 458 | bch_check_keys(&b->keys, "writing"); |
cafe5635 | 459 | |
cafe5635 KO |
460 | cancel_delayed_work(&b->work); |
461 | ||
57943511 | 462 | /* If caller isn't waiting for write, parent refcount is cache set */ |
cb7a583e KO |
463 | down(&b->io_mutex); |
464 | closure_init(&b->io, parent ?: &b->c->cl); | |
57943511 | 465 | |
cafe5635 KO |
466 | clear_bit(BTREE_NODE_dirty, &b->flags); |
467 | change_bit(BTREE_NODE_write_idx, &b->flags); | |
468 | ||
57943511 | 469 | do_btree_node_write(b); |
cafe5635 | 470 | |
ee811287 | 471 | atomic_long_add(set_blocks(i, block_bytes(b->c)) * b->c->sb.block_size, |
cafe5635 KO |
472 | &PTR_CACHE(b->c, &b->key, 0)->btree_sectors_written); |
473 | ||
a85e968e | 474 | b->written += set_blocks(i, block_bytes(b->c)); |
2a285686 | 475 | } |
a85e968e | 476 | |
2a285686 KO |
477 | void bch_btree_node_write(struct btree *b, struct closure *parent) |
478 | { | |
479 | unsigned nsets = b->keys.nsets; | |
480 | ||
481 | lockdep_assert_held(&b->lock); | |
482 | ||
483 | __bch_btree_node_write(b, parent); | |
cafe5635 | 484 | |
78b77bf8 KO |
485 | /* |
486 | * do verify if there was more than one set initially (i.e. we did a | |
487 | * sort) and we sorted down to a single set: | |
488 | */ | |
2a285686 | 489 | if (nsets && !b->keys.nsets) |
78b77bf8 KO |
490 | bch_btree_verify(b); |
491 | ||
2a285686 | 492 | bch_btree_init_next(b); |
cafe5635 KO |
493 | } |
494 | ||
f269af5a KO |
495 | static void bch_btree_node_write_sync(struct btree *b) |
496 | { | |
497 | struct closure cl; | |
498 | ||
499 | closure_init_stack(&cl); | |
2a285686 KO |
500 | |
501 | mutex_lock(&b->write_lock); | |
f269af5a | 502 | bch_btree_node_write(b, &cl); |
2a285686 KO |
503 | mutex_unlock(&b->write_lock); |
504 | ||
f269af5a KO |
505 | closure_sync(&cl); |
506 | } | |
507 | ||
57943511 | 508 | static void btree_node_write_work(struct work_struct *w) |
cafe5635 KO |
509 | { |
510 | struct btree *b = container_of(to_delayed_work(w), struct btree, work); | |
511 | ||
2a285686 | 512 | mutex_lock(&b->write_lock); |
cafe5635 | 513 | if (btree_node_dirty(b)) |
2a285686 KO |
514 | __bch_btree_node_write(b, NULL); |
515 | mutex_unlock(&b->write_lock); | |
cafe5635 KO |
516 | } |
517 | ||
c18536a7 | 518 | static void bch_btree_leaf_dirty(struct btree *b, atomic_t *journal_ref) |
cafe5635 | 519 | { |
ee811287 | 520 | struct bset *i = btree_bset_last(b); |
cafe5635 KO |
521 | struct btree_write *w = btree_current_write(b); |
522 | ||
2a285686 KO |
523 | lockdep_assert_held(&b->write_lock); |
524 | ||
57943511 KO |
525 | BUG_ON(!b->written); |
526 | BUG_ON(!i->keys); | |
cafe5635 | 527 | |
57943511 | 528 | if (!btree_node_dirty(b)) |
56b30770 | 529 | schedule_delayed_work(&b->work, 30 * HZ); |
cafe5635 | 530 | |
57943511 | 531 | set_btree_node_dirty(b); |
cafe5635 | 532 | |
c18536a7 | 533 | if (journal_ref) { |
cafe5635 | 534 | if (w->journal && |
c18536a7 | 535 | journal_pin_cmp(b->c, w->journal, journal_ref)) { |
cafe5635 KO |
536 | atomic_dec_bug(w->journal); |
537 | w->journal = NULL; | |
538 | } | |
539 | ||
540 | if (!w->journal) { | |
c18536a7 | 541 | w->journal = journal_ref; |
cafe5635 KO |
542 | atomic_inc(w->journal); |
543 | } | |
544 | } | |
545 | ||
cafe5635 | 546 | /* Force write if set is too big */ |
57943511 KO |
547 | if (set_bytes(i) > PAGE_SIZE - 48 && |
548 | !current->bio_list) | |
549 | bch_btree_node_write(b, NULL); | |
cafe5635 KO |
550 | } |
551 | ||
552 | /* | |
553 | * Btree in memory cache - allocation/freeing | |
554 | * mca -> memory cache | |
555 | */ | |
556 | ||
cafe5635 KO |
557 | #define mca_reserve(c) (((c->root && c->root->level) \ |
558 | ? c->root->level : 1) * 8 + 16) | |
559 | #define mca_can_free(c) \ | |
0a63b66d | 560 | max_t(int, 0, c->btree_cache_used - mca_reserve(c)) |
cafe5635 KO |
561 | |
562 | static void mca_data_free(struct btree *b) | |
563 | { | |
cb7a583e | 564 | BUG_ON(b->io_mutex.count != 1); |
cafe5635 | 565 | |
a85e968e | 566 | bch_btree_keys_free(&b->keys); |
cafe5635 | 567 | |
0a63b66d | 568 | b->c->btree_cache_used--; |
ee811287 | 569 | list_move(&b->list, &b->c->btree_cache_freed); |
cafe5635 KO |
570 | } |
571 | ||
572 | static void mca_bucket_free(struct btree *b) | |
573 | { | |
574 | BUG_ON(btree_node_dirty(b)); | |
575 | ||
576 | b->key.ptr[0] = 0; | |
577 | hlist_del_init_rcu(&b->hash); | |
578 | list_move(&b->list, &b->c->btree_cache_freeable); | |
579 | } | |
580 | ||
581 | static unsigned btree_order(struct bkey *k) | |
582 | { | |
583 | return ilog2(KEY_SIZE(k) / PAGE_SECTORS ?: 1); | |
584 | } | |
585 | ||
586 | static void mca_data_alloc(struct btree *b, struct bkey *k, gfp_t gfp) | |
587 | { | |
a85e968e | 588 | if (!bch_btree_keys_alloc(&b->keys, |
ee811287 KO |
589 | max_t(unsigned, |
590 | ilog2(b->c->btree_pages), | |
591 | btree_order(k)), | |
592 | gfp)) { | |
0a63b66d | 593 | b->c->btree_cache_used++; |
ee811287 KO |
594 | list_move(&b->list, &b->c->btree_cache); |
595 | } else { | |
596 | list_move(&b->list, &b->c->btree_cache_freed); | |
597 | } | |
cafe5635 KO |
598 | } |
599 | ||
600 | static struct btree *mca_bucket_alloc(struct cache_set *c, | |
601 | struct bkey *k, gfp_t gfp) | |
602 | { | |
603 | struct btree *b = kzalloc(sizeof(struct btree), gfp); | |
604 | if (!b) | |
605 | return NULL; | |
606 | ||
607 | init_rwsem(&b->lock); | |
608 | lockdep_set_novalidate_class(&b->lock); | |
2a285686 KO |
609 | mutex_init(&b->write_lock); |
610 | lockdep_set_novalidate_class(&b->write_lock); | |
cafe5635 | 611 | INIT_LIST_HEAD(&b->list); |
57943511 | 612 | INIT_DELAYED_WORK(&b->work, btree_node_write_work); |
cafe5635 | 613 | b->c = c; |
cb7a583e | 614 | sema_init(&b->io_mutex, 1); |
cafe5635 KO |
615 | |
616 | mca_data_alloc(b, k, gfp); | |
617 | return b; | |
618 | } | |
619 | ||
e8e1d468 | 620 | static int mca_reap(struct btree *b, unsigned min_order, bool flush) |
cafe5635 | 621 | { |
e8e1d468 KO |
622 | struct closure cl; |
623 | ||
624 | closure_init_stack(&cl); | |
cafe5635 KO |
625 | lockdep_assert_held(&b->c->bucket_lock); |
626 | ||
627 | if (!down_write_trylock(&b->lock)) | |
628 | return -ENOMEM; | |
629 | ||
a85e968e | 630 | BUG_ON(btree_node_dirty(b) && !b->keys.set[0].data); |
e8e1d468 | 631 | |
a85e968e | 632 | if (b->keys.page_order < min_order) |
cb7a583e KO |
633 | goto out_unlock; |
634 | ||
635 | if (!flush) { | |
636 | if (btree_node_dirty(b)) | |
637 | goto out_unlock; | |
638 | ||
639 | if (down_trylock(&b->io_mutex)) | |
640 | goto out_unlock; | |
641 | up(&b->io_mutex); | |
cafe5635 KO |
642 | } |
643 | ||
2a285686 | 644 | mutex_lock(&b->write_lock); |
f269af5a | 645 | if (btree_node_dirty(b)) |
2a285686 KO |
646 | __bch_btree_node_write(b, &cl); |
647 | mutex_unlock(&b->write_lock); | |
648 | ||
649 | closure_sync(&cl); | |
cafe5635 | 650 | |
e8e1d468 | 651 | /* wait for any in flight btree write */ |
cb7a583e KO |
652 | down(&b->io_mutex); |
653 | up(&b->io_mutex); | |
e8e1d468 | 654 | |
cafe5635 | 655 | return 0; |
cb7a583e KO |
656 | out_unlock: |
657 | rw_unlock(true, b); | |
658 | return -ENOMEM; | |
cafe5635 KO |
659 | } |
660 | ||
7dc19d5a DC |
661 | static unsigned long bch_mca_scan(struct shrinker *shrink, |
662 | struct shrink_control *sc) | |
cafe5635 KO |
663 | { |
664 | struct cache_set *c = container_of(shrink, struct cache_set, shrink); | |
665 | struct btree *b, *t; | |
666 | unsigned long i, nr = sc->nr_to_scan; | |
7dc19d5a | 667 | unsigned long freed = 0; |
cafe5635 KO |
668 | |
669 | if (c->shrinker_disabled) | |
7dc19d5a | 670 | return SHRINK_STOP; |
cafe5635 | 671 | |
0a63b66d | 672 | if (c->btree_cache_alloc_lock) |
7dc19d5a | 673 | return SHRINK_STOP; |
cafe5635 KO |
674 | |
675 | /* Return -1 if we can't do anything right now */ | |
a698e08c | 676 | if (sc->gfp_mask & __GFP_IO) |
cafe5635 KO |
677 | mutex_lock(&c->bucket_lock); |
678 | else if (!mutex_trylock(&c->bucket_lock)) | |
679 | return -1; | |
680 | ||
36c9ea98 KO |
681 | /* |
682 | * It's _really_ critical that we don't free too many btree nodes - we | |
683 | * have to always leave ourselves a reserve. The reserve is how we | |
684 | * guarantee that allocating memory for a new btree node can always | |
685 | * succeed, so that inserting keys into the btree can always succeed and | |
686 | * IO can always make forward progress: | |
687 | */ | |
cafe5635 KO |
688 | nr /= c->btree_pages; |
689 | nr = min_t(unsigned long, nr, mca_can_free(c)); | |
690 | ||
691 | i = 0; | |
692 | list_for_each_entry_safe(b, t, &c->btree_cache_freeable, list) { | |
7dc19d5a | 693 | if (freed >= nr) |
cafe5635 KO |
694 | break; |
695 | ||
696 | if (++i > 3 && | |
e8e1d468 | 697 | !mca_reap(b, 0, false)) { |
cafe5635 KO |
698 | mca_data_free(b); |
699 | rw_unlock(true, b); | |
7dc19d5a | 700 | freed++; |
cafe5635 KO |
701 | } |
702 | } | |
703 | ||
0a63b66d | 704 | for (i = 0; (nr--) && i < c->btree_cache_used; i++) { |
b0f32a56 KO |
705 | if (list_empty(&c->btree_cache)) |
706 | goto out; | |
707 | ||
cafe5635 KO |
708 | b = list_first_entry(&c->btree_cache, struct btree, list); |
709 | list_rotate_left(&c->btree_cache); | |
710 | ||
711 | if (!b->accessed && | |
e8e1d468 | 712 | !mca_reap(b, 0, false)) { |
cafe5635 KO |
713 | mca_bucket_free(b); |
714 | mca_data_free(b); | |
715 | rw_unlock(true, b); | |
7dc19d5a | 716 | freed++; |
cafe5635 KO |
717 | } else |
718 | b->accessed = 0; | |
719 | } | |
720 | out: | |
cafe5635 | 721 | mutex_unlock(&c->bucket_lock); |
7dc19d5a DC |
722 | return freed; |
723 | } | |
724 | ||
725 | static unsigned long bch_mca_count(struct shrinker *shrink, | |
726 | struct shrink_control *sc) | |
727 | { | |
728 | struct cache_set *c = container_of(shrink, struct cache_set, shrink); | |
729 | ||
730 | if (c->shrinker_disabled) | |
731 | return 0; | |
732 | ||
0a63b66d | 733 | if (c->btree_cache_alloc_lock) |
7dc19d5a DC |
734 | return 0; |
735 | ||
736 | return mca_can_free(c) * c->btree_pages; | |
cafe5635 KO |
737 | } |
738 | ||
739 | void bch_btree_cache_free(struct cache_set *c) | |
740 | { | |
741 | struct btree *b; | |
742 | struct closure cl; | |
743 | closure_init_stack(&cl); | |
744 | ||
745 | if (c->shrink.list.next) | |
746 | unregister_shrinker(&c->shrink); | |
747 | ||
748 | mutex_lock(&c->bucket_lock); | |
749 | ||
750 | #ifdef CONFIG_BCACHE_DEBUG | |
751 | if (c->verify_data) | |
752 | list_move(&c->verify_data->list, &c->btree_cache); | |
78b77bf8 KO |
753 | |
754 | free_pages((unsigned long) c->verify_ondisk, ilog2(bucket_pages(c))); | |
cafe5635 KO |
755 | #endif |
756 | ||
757 | list_splice(&c->btree_cache_freeable, | |
758 | &c->btree_cache); | |
759 | ||
760 | while (!list_empty(&c->btree_cache)) { | |
761 | b = list_first_entry(&c->btree_cache, struct btree, list); | |
762 | ||
763 | if (btree_node_dirty(b)) | |
764 | btree_complete_write(b, btree_current_write(b)); | |
765 | clear_bit(BTREE_NODE_dirty, &b->flags); | |
766 | ||
767 | mca_data_free(b); | |
768 | } | |
769 | ||
770 | while (!list_empty(&c->btree_cache_freed)) { | |
771 | b = list_first_entry(&c->btree_cache_freed, | |
772 | struct btree, list); | |
773 | list_del(&b->list); | |
774 | cancel_delayed_work_sync(&b->work); | |
775 | kfree(b); | |
776 | } | |
777 | ||
778 | mutex_unlock(&c->bucket_lock); | |
779 | } | |
780 | ||
781 | int bch_btree_cache_alloc(struct cache_set *c) | |
782 | { | |
783 | unsigned i; | |
784 | ||
cafe5635 | 785 | for (i = 0; i < mca_reserve(c); i++) |
72a44517 KO |
786 | if (!mca_bucket_alloc(c, &ZERO_KEY, GFP_KERNEL)) |
787 | return -ENOMEM; | |
cafe5635 KO |
788 | |
789 | list_splice_init(&c->btree_cache, | |
790 | &c->btree_cache_freeable); | |
791 | ||
792 | #ifdef CONFIG_BCACHE_DEBUG | |
793 | mutex_init(&c->verify_lock); | |
794 | ||
78b77bf8 KO |
795 | c->verify_ondisk = (void *) |
796 | __get_free_pages(GFP_KERNEL, ilog2(bucket_pages(c))); | |
797 | ||
cafe5635 KO |
798 | c->verify_data = mca_bucket_alloc(c, &ZERO_KEY, GFP_KERNEL); |
799 | ||
800 | if (c->verify_data && | |
a85e968e | 801 | c->verify_data->keys.set->data) |
cafe5635 KO |
802 | list_del_init(&c->verify_data->list); |
803 | else | |
804 | c->verify_data = NULL; | |
805 | #endif | |
806 | ||
7dc19d5a DC |
807 | c->shrink.count_objects = bch_mca_count; |
808 | c->shrink.scan_objects = bch_mca_scan; | |
cafe5635 KO |
809 | c->shrink.seeks = 4; |
810 | c->shrink.batch = c->btree_pages * 2; | |
811 | register_shrinker(&c->shrink); | |
812 | ||
813 | return 0; | |
814 | } | |
815 | ||
816 | /* Btree in memory cache - hash table */ | |
817 | ||
818 | static struct hlist_head *mca_hash(struct cache_set *c, struct bkey *k) | |
819 | { | |
820 | return &c->bucket_hash[hash_32(PTR_HASH(c, k), BUCKET_HASH_BITS)]; | |
821 | } | |
822 | ||
823 | static struct btree *mca_find(struct cache_set *c, struct bkey *k) | |
824 | { | |
825 | struct btree *b; | |
826 | ||
827 | rcu_read_lock(); | |
828 | hlist_for_each_entry_rcu(b, mca_hash(c, k), hash) | |
829 | if (PTR_HASH(c, &b->key) == PTR_HASH(c, k)) | |
830 | goto out; | |
831 | b = NULL; | |
832 | out: | |
833 | rcu_read_unlock(); | |
834 | return b; | |
835 | } | |
836 | ||
0a63b66d KO |
837 | static int mca_cannibalize_lock(struct cache_set *c, struct btree_op *op) |
838 | { | |
839 | struct task_struct *old; | |
840 | ||
841 | old = cmpxchg(&c->btree_cache_alloc_lock, NULL, current); | |
842 | if (old && old != current) { | |
843 | if (op) | |
844 | prepare_to_wait(&c->btree_cache_wait, &op->wait, | |
845 | TASK_UNINTERRUPTIBLE); | |
846 | return -EINTR; | |
847 | } | |
848 | ||
849 | return 0; | |
850 | } | |
851 | ||
852 | static struct btree *mca_cannibalize(struct cache_set *c, struct btree_op *op, | |
853 | struct bkey *k) | |
cafe5635 | 854 | { |
e8e1d468 | 855 | struct btree *b; |
cafe5635 | 856 | |
c37511b8 KO |
857 | trace_bcache_btree_cache_cannibalize(c); |
858 | ||
0a63b66d KO |
859 | if (mca_cannibalize_lock(c, op)) |
860 | return ERR_PTR(-EINTR); | |
cafe5635 | 861 | |
e8e1d468 KO |
862 | list_for_each_entry_reverse(b, &c->btree_cache, list) |
863 | if (!mca_reap(b, btree_order(k), false)) | |
864 | return b; | |
cafe5635 | 865 | |
e8e1d468 KO |
866 | list_for_each_entry_reverse(b, &c->btree_cache, list) |
867 | if (!mca_reap(b, btree_order(k), true)) | |
868 | return b; | |
cafe5635 | 869 | |
0a63b66d | 870 | WARN(1, "btree cache cannibalize failed\n"); |
e8e1d468 | 871 | return ERR_PTR(-ENOMEM); |
cafe5635 KO |
872 | } |
873 | ||
874 | /* | |
875 | * We can only have one thread cannibalizing other cached btree nodes at a time, | |
876 | * or we'll deadlock. We use an open coded mutex to ensure that, which a | |
877 | * cannibalize_bucket() will take. This means every time we unlock the root of | |
878 | * the btree, we need to release this lock if we have it held. | |
879 | */ | |
df8e8970 | 880 | static void bch_cannibalize_unlock(struct cache_set *c) |
cafe5635 | 881 | { |
0a63b66d KO |
882 | if (c->btree_cache_alloc_lock == current) { |
883 | c->btree_cache_alloc_lock = NULL; | |
884 | wake_up(&c->btree_cache_wait); | |
cafe5635 KO |
885 | } |
886 | } | |
887 | ||
0a63b66d KO |
888 | static struct btree *mca_alloc(struct cache_set *c, struct btree_op *op, |
889 | struct bkey *k, int level) | |
cafe5635 KO |
890 | { |
891 | struct btree *b; | |
892 | ||
e8e1d468 KO |
893 | BUG_ON(current->bio_list); |
894 | ||
cafe5635 KO |
895 | lockdep_assert_held(&c->bucket_lock); |
896 | ||
897 | if (mca_find(c, k)) | |
898 | return NULL; | |
899 | ||
900 | /* btree_free() doesn't free memory; it sticks the node on the end of | |
901 | * the list. Check if there's any freed nodes there: | |
902 | */ | |
903 | list_for_each_entry(b, &c->btree_cache_freeable, list) | |
e8e1d468 | 904 | if (!mca_reap(b, btree_order(k), false)) |
cafe5635 KO |
905 | goto out; |
906 | ||
907 | /* We never free struct btree itself, just the memory that holds the on | |
908 | * disk node. Check the freed list before allocating a new one: | |
909 | */ | |
910 | list_for_each_entry(b, &c->btree_cache_freed, list) | |
e8e1d468 | 911 | if (!mca_reap(b, 0, false)) { |
cafe5635 | 912 | mca_data_alloc(b, k, __GFP_NOWARN|GFP_NOIO); |
a85e968e | 913 | if (!b->keys.set[0].data) |
cafe5635 KO |
914 | goto err; |
915 | else | |
916 | goto out; | |
917 | } | |
918 | ||
919 | b = mca_bucket_alloc(c, k, __GFP_NOWARN|GFP_NOIO); | |
920 | if (!b) | |
921 | goto err; | |
922 | ||
923 | BUG_ON(!down_write_trylock(&b->lock)); | |
a85e968e | 924 | if (!b->keys.set->data) |
cafe5635 KO |
925 | goto err; |
926 | out: | |
cb7a583e | 927 | BUG_ON(b->io_mutex.count != 1); |
cafe5635 KO |
928 | |
929 | bkey_copy(&b->key, k); | |
930 | list_move(&b->list, &c->btree_cache); | |
931 | hlist_del_init_rcu(&b->hash); | |
932 | hlist_add_head_rcu(&b->hash, mca_hash(c, k)); | |
933 | ||
934 | lock_set_subclass(&b->lock.dep_map, level + 1, _THIS_IP_); | |
d6fd3b11 | 935 | b->parent = (void *) ~0UL; |
a85e968e KO |
936 | b->flags = 0; |
937 | b->written = 0; | |
938 | b->level = level; | |
cafe5635 | 939 | |
65d45231 | 940 | if (!b->level) |
a85e968e KO |
941 | bch_btree_keys_init(&b->keys, &bch_extent_keys_ops, |
942 | &b->c->expensive_debug_checks); | |
65d45231 | 943 | else |
a85e968e KO |
944 | bch_btree_keys_init(&b->keys, &bch_btree_keys_ops, |
945 | &b->c->expensive_debug_checks); | |
cafe5635 KO |
946 | |
947 | return b; | |
948 | err: | |
949 | if (b) | |
950 | rw_unlock(true, b); | |
951 | ||
0a63b66d | 952 | b = mca_cannibalize(c, op, k); |
cafe5635 KO |
953 | if (!IS_ERR(b)) |
954 | goto out; | |
955 | ||
956 | return b; | |
957 | } | |
958 | ||
959 | /** | |
960 | * bch_btree_node_get - find a btree node in the cache and lock it, reading it | |
961 | * in from disk if necessary. | |
962 | * | |
b54d6934 | 963 | * If IO is necessary and running under generic_make_request, returns -EAGAIN. |
cafe5635 KO |
964 | * |
965 | * The btree node will have either a read or a write lock held, depending on | |
966 | * level and op->lock. | |
967 | */ | |
0a63b66d | 968 | struct btree *bch_btree_node_get(struct cache_set *c, struct btree_op *op, |
2452cc89 SP |
969 | struct bkey *k, int level, bool write, |
970 | struct btree *parent) | |
cafe5635 KO |
971 | { |
972 | int i = 0; | |
cafe5635 KO |
973 | struct btree *b; |
974 | ||
975 | BUG_ON(level < 0); | |
976 | retry: | |
977 | b = mca_find(c, k); | |
978 | ||
979 | if (!b) { | |
57943511 KO |
980 | if (current->bio_list) |
981 | return ERR_PTR(-EAGAIN); | |
982 | ||
cafe5635 | 983 | mutex_lock(&c->bucket_lock); |
0a63b66d | 984 | b = mca_alloc(c, op, k, level); |
cafe5635 KO |
985 | mutex_unlock(&c->bucket_lock); |
986 | ||
987 | if (!b) | |
988 | goto retry; | |
989 | if (IS_ERR(b)) | |
990 | return b; | |
991 | ||
57943511 | 992 | bch_btree_node_read(b); |
cafe5635 KO |
993 | |
994 | if (!write) | |
995 | downgrade_write(&b->lock); | |
996 | } else { | |
997 | rw_lock(write, b, level); | |
998 | if (PTR_HASH(c, &b->key) != PTR_HASH(c, k)) { | |
999 | rw_unlock(write, b); | |
1000 | goto retry; | |
1001 | } | |
1002 | BUG_ON(b->level != level); | |
1003 | } | |
1004 | ||
2452cc89 | 1005 | b->parent = parent; |
cafe5635 KO |
1006 | b->accessed = 1; |
1007 | ||
a85e968e KO |
1008 | for (; i <= b->keys.nsets && b->keys.set[i].size; i++) { |
1009 | prefetch(b->keys.set[i].tree); | |
1010 | prefetch(b->keys.set[i].data); | |
cafe5635 KO |
1011 | } |
1012 | ||
a85e968e KO |
1013 | for (; i <= b->keys.nsets; i++) |
1014 | prefetch(b->keys.set[i].data); | |
cafe5635 | 1015 | |
57943511 | 1016 | if (btree_node_io_error(b)) { |
cafe5635 | 1017 | rw_unlock(write, b); |
57943511 KO |
1018 | return ERR_PTR(-EIO); |
1019 | } | |
1020 | ||
1021 | BUG_ON(!b->written); | |
cafe5635 KO |
1022 | |
1023 | return b; | |
1024 | } | |
1025 | ||
2452cc89 | 1026 | static void btree_node_prefetch(struct btree *parent, struct bkey *k) |
cafe5635 KO |
1027 | { |
1028 | struct btree *b; | |
1029 | ||
2452cc89 SP |
1030 | mutex_lock(&parent->c->bucket_lock); |
1031 | b = mca_alloc(parent->c, NULL, k, parent->level - 1); | |
1032 | mutex_unlock(&parent->c->bucket_lock); | |
cafe5635 KO |
1033 | |
1034 | if (!IS_ERR_OR_NULL(b)) { | |
2452cc89 | 1035 | b->parent = parent; |
57943511 | 1036 | bch_btree_node_read(b); |
cafe5635 KO |
1037 | rw_unlock(true, b); |
1038 | } | |
1039 | } | |
1040 | ||
1041 | /* Btree alloc */ | |
1042 | ||
e8e1d468 | 1043 | static void btree_node_free(struct btree *b) |
cafe5635 | 1044 | { |
c37511b8 KO |
1045 | trace_bcache_btree_node_free(b); |
1046 | ||
cafe5635 | 1047 | BUG_ON(b == b->c->root); |
cafe5635 | 1048 | |
2a285686 KO |
1049 | mutex_lock(&b->write_lock); |
1050 | ||
cafe5635 KO |
1051 | if (btree_node_dirty(b)) |
1052 | btree_complete_write(b, btree_current_write(b)); | |
1053 | clear_bit(BTREE_NODE_dirty, &b->flags); | |
1054 | ||
2a285686 KO |
1055 | mutex_unlock(&b->write_lock); |
1056 | ||
cafe5635 KO |
1057 | cancel_delayed_work(&b->work); |
1058 | ||
1059 | mutex_lock(&b->c->bucket_lock); | |
cafe5635 KO |
1060 | bch_bucket_free(b->c, &b->key); |
1061 | mca_bucket_free(b); | |
1062 | mutex_unlock(&b->c->bucket_lock); | |
1063 | } | |
1064 | ||
c5aa4a31 | 1065 | struct btree *__bch_btree_node_alloc(struct cache_set *c, struct btree_op *op, |
2452cc89 SP |
1066 | int level, bool wait, |
1067 | struct btree *parent) | |
cafe5635 KO |
1068 | { |
1069 | BKEY_PADDED(key) k; | |
1070 | struct btree *b = ERR_PTR(-EAGAIN); | |
1071 | ||
1072 | mutex_lock(&c->bucket_lock); | |
1073 | retry: | |
c5aa4a31 | 1074 | if (__bch_bucket_alloc_set(c, RESERVE_BTREE, &k.key, 1, wait)) |
cafe5635 KO |
1075 | goto err; |
1076 | ||
3a3b6a4e | 1077 | bkey_put(c, &k.key); |
cafe5635 KO |
1078 | SET_KEY_SIZE(&k.key, c->btree_pages * PAGE_SECTORS); |
1079 | ||
0a63b66d | 1080 | b = mca_alloc(c, op, &k.key, level); |
cafe5635 KO |
1081 | if (IS_ERR(b)) |
1082 | goto err_free; | |
1083 | ||
1084 | if (!b) { | |
b1a67b0f KO |
1085 | cache_bug(c, |
1086 | "Tried to allocate bucket that was in btree cache"); | |
cafe5635 KO |
1087 | goto retry; |
1088 | } | |
1089 | ||
cafe5635 | 1090 | b->accessed = 1; |
2452cc89 | 1091 | b->parent = parent; |
a85e968e | 1092 | bch_bset_init_next(&b->keys, b->keys.set->data, bset_magic(&b->c->sb)); |
cafe5635 KO |
1093 | |
1094 | mutex_unlock(&c->bucket_lock); | |
c37511b8 KO |
1095 | |
1096 | trace_bcache_btree_node_alloc(b); | |
cafe5635 KO |
1097 | return b; |
1098 | err_free: | |
1099 | bch_bucket_free(c, &k.key); | |
cafe5635 KO |
1100 | err: |
1101 | mutex_unlock(&c->bucket_lock); | |
c37511b8 | 1102 | |
913dc33f | 1103 | trace_bcache_btree_node_alloc_fail(c); |
cafe5635 KO |
1104 | return b; |
1105 | } | |
1106 | ||
c5aa4a31 | 1107 | static struct btree *bch_btree_node_alloc(struct cache_set *c, |
2452cc89 SP |
1108 | struct btree_op *op, int level, |
1109 | struct btree *parent) | |
c5aa4a31 | 1110 | { |
2452cc89 | 1111 | return __bch_btree_node_alloc(c, op, level, op != NULL, parent); |
c5aa4a31 SP |
1112 | } |
1113 | ||
0a63b66d KO |
1114 | static struct btree *btree_node_alloc_replacement(struct btree *b, |
1115 | struct btree_op *op) | |
cafe5635 | 1116 | { |
2452cc89 | 1117 | struct btree *n = bch_btree_node_alloc(b->c, op, b->level, b->parent); |
67539e85 | 1118 | if (!IS_ERR_OR_NULL(n)) { |
2a285686 | 1119 | mutex_lock(&n->write_lock); |
89ebb4a2 | 1120 | bch_btree_sort_into(&b->keys, &n->keys, &b->c->sort); |
67539e85 | 1121 | bkey_copy_key(&n->key, &b->key); |
2a285686 | 1122 | mutex_unlock(&n->write_lock); |
67539e85 | 1123 | } |
cafe5635 KO |
1124 | |
1125 | return n; | |
1126 | } | |
1127 | ||
8835c123 KO |
1128 | static void make_btree_freeing_key(struct btree *b, struct bkey *k) |
1129 | { | |
1130 | unsigned i; | |
1131 | ||
05335cff KO |
1132 | mutex_lock(&b->c->bucket_lock); |
1133 | ||
1134 | atomic_inc(&b->c->prio_blocked); | |
1135 | ||
8835c123 KO |
1136 | bkey_copy(k, &b->key); |
1137 | bkey_copy_key(k, &ZERO_KEY); | |
1138 | ||
05335cff KO |
1139 | for (i = 0; i < KEY_PTRS(k); i++) |
1140 | SET_PTR_GEN(k, i, | |
1141 | bch_inc_gen(PTR_CACHE(b->c, &b->key, i), | |
1142 | PTR_BUCKET(b->c, &b->key, i))); | |
8835c123 | 1143 | |
05335cff | 1144 | mutex_unlock(&b->c->bucket_lock); |
8835c123 KO |
1145 | } |
1146 | ||
78365411 KO |
1147 | static int btree_check_reserve(struct btree *b, struct btree_op *op) |
1148 | { | |
1149 | struct cache_set *c = b->c; | |
1150 | struct cache *ca; | |
0a63b66d | 1151 | unsigned i, reserve = (c->root->level - b->level) * 2 + 1; |
78365411 KO |
1152 | |
1153 | mutex_lock(&c->bucket_lock); | |
1154 | ||
1155 | for_each_cache(ca, c, i) | |
1156 | if (fifo_used(&ca->free[RESERVE_BTREE]) < reserve) { | |
1157 | if (op) | |
0a63b66d | 1158 | prepare_to_wait(&c->btree_cache_wait, &op->wait, |
78365411 | 1159 | TASK_UNINTERRUPTIBLE); |
0a63b66d KO |
1160 | mutex_unlock(&c->bucket_lock); |
1161 | return -EINTR; | |
78365411 KO |
1162 | } |
1163 | ||
1164 | mutex_unlock(&c->bucket_lock); | |
0a63b66d KO |
1165 | |
1166 | return mca_cannibalize_lock(b->c, op); | |
78365411 KO |
1167 | } |
1168 | ||
cafe5635 KO |
1169 | /* Garbage collection */ |
1170 | ||
487dded8 KO |
1171 | static uint8_t __bch_btree_mark_key(struct cache_set *c, int level, |
1172 | struct bkey *k) | |
cafe5635 KO |
1173 | { |
1174 | uint8_t stale = 0; | |
1175 | unsigned i; | |
1176 | struct bucket *g; | |
1177 | ||
1178 | /* | |
1179 | * ptr_invalid() can't return true for the keys that mark btree nodes as | |
1180 | * freed, but since ptr_bad() returns true we'll never actually use them | |
1181 | * for anything and thus we don't want mark their pointers here | |
1182 | */ | |
1183 | if (!bkey_cmp(k, &ZERO_KEY)) | |
1184 | return stale; | |
1185 | ||
1186 | for (i = 0; i < KEY_PTRS(k); i++) { | |
1187 | if (!ptr_available(c, k, i)) | |
1188 | continue; | |
1189 | ||
1190 | g = PTR_BUCKET(c, k, i); | |
1191 | ||
3a2fd9d5 KO |
1192 | if (gen_after(g->last_gc, PTR_GEN(k, i))) |
1193 | g->last_gc = PTR_GEN(k, i); | |
cafe5635 KO |
1194 | |
1195 | if (ptr_stale(c, k, i)) { | |
1196 | stale = max(stale, ptr_stale(c, k, i)); | |
1197 | continue; | |
1198 | } | |
1199 | ||
1200 | cache_bug_on(GC_MARK(g) && | |
1201 | (GC_MARK(g) == GC_MARK_METADATA) != (level != 0), | |
1202 | c, "inconsistent ptrs: mark = %llu, level = %i", | |
1203 | GC_MARK(g), level); | |
1204 | ||
1205 | if (level) | |
1206 | SET_GC_MARK(g, GC_MARK_METADATA); | |
1207 | else if (KEY_DIRTY(k)) | |
1208 | SET_GC_MARK(g, GC_MARK_DIRTY); | |
4fe6a816 KO |
1209 | else if (!GC_MARK(g)) |
1210 | SET_GC_MARK(g, GC_MARK_RECLAIMABLE); | |
cafe5635 KO |
1211 | |
1212 | /* guard against overflow */ | |
1213 | SET_GC_SECTORS_USED(g, min_t(unsigned, | |
1214 | GC_SECTORS_USED(g) + KEY_SIZE(k), | |
94717447 | 1215 | MAX_GC_SECTORS_USED)); |
cafe5635 KO |
1216 | |
1217 | BUG_ON(!GC_SECTORS_USED(g)); | |
1218 | } | |
1219 | ||
1220 | return stale; | |
1221 | } | |
1222 | ||
1223 | #define btree_mark_key(b, k) __bch_btree_mark_key(b->c, b->level, k) | |
1224 | ||
487dded8 KO |
1225 | void bch_initial_mark_key(struct cache_set *c, int level, struct bkey *k) |
1226 | { | |
1227 | unsigned i; | |
1228 | ||
1229 | for (i = 0; i < KEY_PTRS(k); i++) | |
1230 | if (ptr_available(c, k, i) && | |
1231 | !ptr_stale(c, k, i)) { | |
1232 | struct bucket *b = PTR_BUCKET(c, k, i); | |
1233 | ||
1234 | b->gen = PTR_GEN(k, i); | |
1235 | ||
1236 | if (level && bkey_cmp(k, &ZERO_KEY)) | |
1237 | b->prio = BTREE_PRIO; | |
1238 | else if (!level && b->prio == BTREE_PRIO) | |
1239 | b->prio = INITIAL_PRIO; | |
1240 | } | |
1241 | ||
1242 | __bch_btree_mark_key(c, level, k); | |
1243 | } | |
1244 | ||
a1f0358b | 1245 | static bool btree_gc_mark_node(struct btree *b, struct gc_stat *gc) |
cafe5635 KO |
1246 | { |
1247 | uint8_t stale = 0; | |
a1f0358b | 1248 | unsigned keys = 0, good_keys = 0; |
cafe5635 KO |
1249 | struct bkey *k; |
1250 | struct btree_iter iter; | |
1251 | struct bset_tree *t; | |
1252 | ||
1253 | gc->nodes++; | |
1254 | ||
c052dd9a | 1255 | for_each_key_filter(&b->keys, k, &iter, bch_ptr_invalid) { |
cafe5635 | 1256 | stale = max(stale, btree_mark_key(b, k)); |
a1f0358b | 1257 | keys++; |
cafe5635 | 1258 | |
a85e968e | 1259 | if (bch_ptr_bad(&b->keys, k)) |
cafe5635 KO |
1260 | continue; |
1261 | ||
cafe5635 KO |
1262 | gc->key_bytes += bkey_u64s(k); |
1263 | gc->nkeys++; | |
a1f0358b | 1264 | good_keys++; |
cafe5635 KO |
1265 | |
1266 | gc->data += KEY_SIZE(k); | |
cafe5635 KO |
1267 | } |
1268 | ||
a85e968e | 1269 | for (t = b->keys.set; t <= &b->keys.set[b->keys.nsets]; t++) |
cafe5635 | 1270 | btree_bug_on(t->size && |
a85e968e | 1271 | bset_written(&b->keys, t) && |
cafe5635 KO |
1272 | bkey_cmp(&b->key, &t->end) < 0, |
1273 | b, "found short btree key in gc"); | |
1274 | ||
a1f0358b KO |
1275 | if (b->c->gc_always_rewrite) |
1276 | return true; | |
cafe5635 | 1277 | |
a1f0358b KO |
1278 | if (stale > 10) |
1279 | return true; | |
cafe5635 | 1280 | |
a1f0358b KO |
1281 | if ((keys - good_keys) * 2 > keys) |
1282 | return true; | |
cafe5635 | 1283 | |
a1f0358b | 1284 | return false; |
cafe5635 KO |
1285 | } |
1286 | ||
a1f0358b | 1287 | #define GC_MERGE_NODES 4U |
cafe5635 KO |
1288 | |
1289 | struct gc_merge_info { | |
1290 | struct btree *b; | |
cafe5635 KO |
1291 | unsigned keys; |
1292 | }; | |
1293 | ||
a1f0358b KO |
1294 | static int bch_btree_insert_node(struct btree *, struct btree_op *, |
1295 | struct keylist *, atomic_t *, struct bkey *); | |
1296 | ||
1297 | static int btree_gc_coalesce(struct btree *b, struct btree_op *op, | |
0a63b66d | 1298 | struct gc_stat *gc, struct gc_merge_info *r) |
cafe5635 | 1299 | { |
a1f0358b KO |
1300 | unsigned i, nodes = 0, keys = 0, blocks; |
1301 | struct btree *new_nodes[GC_MERGE_NODES]; | |
0a63b66d | 1302 | struct keylist keylist; |
b54d6934 | 1303 | struct closure cl; |
a1f0358b | 1304 | struct bkey *k; |
b54d6934 | 1305 | |
0a63b66d KO |
1306 | bch_keylist_init(&keylist); |
1307 | ||
1308 | if (btree_check_reserve(b, NULL)) | |
1309 | return 0; | |
1310 | ||
a1f0358b | 1311 | memset(new_nodes, 0, sizeof(new_nodes)); |
b54d6934 | 1312 | closure_init_stack(&cl); |
cafe5635 | 1313 | |
a1f0358b | 1314 | while (nodes < GC_MERGE_NODES && !IS_ERR_OR_NULL(r[nodes].b)) |
cafe5635 KO |
1315 | keys += r[nodes++].keys; |
1316 | ||
1317 | blocks = btree_default_blocks(b->c) * 2 / 3; | |
1318 | ||
1319 | if (nodes < 2 || | |
a85e968e | 1320 | __set_blocks(b->keys.set[0].data, keys, |
ee811287 | 1321 | block_bytes(b->c)) > blocks * (nodes - 1)) |
a1f0358b | 1322 | return 0; |
cafe5635 | 1323 | |
a1f0358b | 1324 | for (i = 0; i < nodes; i++) { |
0a63b66d | 1325 | new_nodes[i] = btree_node_alloc_replacement(r[i].b, NULL); |
a1f0358b KO |
1326 | if (IS_ERR_OR_NULL(new_nodes[i])) |
1327 | goto out_nocoalesce; | |
cafe5635 KO |
1328 | } |
1329 | ||
0a63b66d KO |
1330 | /* |
1331 | * We have to check the reserve here, after we've allocated our new | |
1332 | * nodes, to make sure the insert below will succeed - we also check | |
1333 | * before as an optimization to potentially avoid a bunch of expensive | |
1334 | * allocs/sorts | |
1335 | */ | |
1336 | if (btree_check_reserve(b, NULL)) | |
1337 | goto out_nocoalesce; | |
1338 | ||
2a285686 KO |
1339 | for (i = 0; i < nodes; i++) |
1340 | mutex_lock(&new_nodes[i]->write_lock); | |
1341 | ||
cafe5635 | 1342 | for (i = nodes - 1; i > 0; --i) { |
ee811287 KO |
1343 | struct bset *n1 = btree_bset_first(new_nodes[i]); |
1344 | struct bset *n2 = btree_bset_first(new_nodes[i - 1]); | |
cafe5635 KO |
1345 | struct bkey *k, *last = NULL; |
1346 | ||
1347 | keys = 0; | |
1348 | ||
a1f0358b KO |
1349 | if (i > 1) { |
1350 | for (k = n2->start; | |
fafff81c | 1351 | k < bset_bkey_last(n2); |
a1f0358b KO |
1352 | k = bkey_next(k)) { |
1353 | if (__set_blocks(n1, n1->keys + keys + | |
ee811287 KO |
1354 | bkey_u64s(k), |
1355 | block_bytes(b->c)) > blocks) | |
a1f0358b KO |
1356 | break; |
1357 | ||
1358 | last = k; | |
1359 | keys += bkey_u64s(k); | |
1360 | } | |
1361 | } else { | |
cafe5635 KO |
1362 | /* |
1363 | * Last node we're not getting rid of - we're getting | |
1364 | * rid of the node at r[0]. Have to try and fit all of | |
1365 | * the remaining keys into this node; we can't ensure | |
1366 | * they will always fit due to rounding and variable | |
1367 | * length keys (shouldn't be possible in practice, | |
1368 | * though) | |
1369 | */ | |
a1f0358b | 1370 | if (__set_blocks(n1, n1->keys + n2->keys, |
ee811287 KO |
1371 | block_bytes(b->c)) > |
1372 | btree_blocks(new_nodes[i])) | |
a1f0358b | 1373 | goto out_nocoalesce; |
cafe5635 KO |
1374 | |
1375 | keys = n2->keys; | |
a1f0358b | 1376 | /* Take the key of the node we're getting rid of */ |
cafe5635 | 1377 | last = &r->b->key; |
a1f0358b | 1378 | } |
cafe5635 | 1379 | |
ee811287 KO |
1380 | BUG_ON(__set_blocks(n1, n1->keys + keys, block_bytes(b->c)) > |
1381 | btree_blocks(new_nodes[i])); | |
cafe5635 | 1382 | |
a1f0358b KO |
1383 | if (last) |
1384 | bkey_copy_key(&new_nodes[i]->key, last); | |
cafe5635 | 1385 | |
fafff81c | 1386 | memcpy(bset_bkey_last(n1), |
cafe5635 | 1387 | n2->start, |
fafff81c | 1388 | (void *) bset_bkey_idx(n2, keys) - (void *) n2->start); |
cafe5635 KO |
1389 | |
1390 | n1->keys += keys; | |
a1f0358b | 1391 | r[i].keys = n1->keys; |
cafe5635 KO |
1392 | |
1393 | memmove(n2->start, | |
fafff81c KO |
1394 | bset_bkey_idx(n2, keys), |
1395 | (void *) bset_bkey_last(n2) - | |
1396 | (void *) bset_bkey_idx(n2, keys)); | |
cafe5635 KO |
1397 | |
1398 | n2->keys -= keys; | |
1399 | ||
0a63b66d | 1400 | if (__bch_keylist_realloc(&keylist, |
085d2a3d | 1401 | bkey_u64s(&new_nodes[i]->key))) |
a1f0358b KO |
1402 | goto out_nocoalesce; |
1403 | ||
1404 | bch_btree_node_write(new_nodes[i], &cl); | |
0a63b66d | 1405 | bch_keylist_add(&keylist, &new_nodes[i]->key); |
cafe5635 KO |
1406 | } |
1407 | ||
2a285686 KO |
1408 | for (i = 0; i < nodes; i++) |
1409 | mutex_unlock(&new_nodes[i]->write_lock); | |
1410 | ||
05335cff KO |
1411 | closure_sync(&cl); |
1412 | ||
1413 | /* We emptied out this node */ | |
1414 | BUG_ON(btree_bset_first(new_nodes[0])->keys); | |
1415 | btree_node_free(new_nodes[0]); | |
1416 | rw_unlock(true, new_nodes[0]); | |
400ffaa2 | 1417 | new_nodes[0] = NULL; |
05335cff | 1418 | |
a1f0358b | 1419 | for (i = 0; i < nodes; i++) { |
0a63b66d | 1420 | if (__bch_keylist_realloc(&keylist, bkey_u64s(&r[i].b->key))) |
a1f0358b | 1421 | goto out_nocoalesce; |
cafe5635 | 1422 | |
0a63b66d KO |
1423 | make_btree_freeing_key(r[i].b, keylist.top); |
1424 | bch_keylist_push(&keylist); | |
a1f0358b | 1425 | } |
cafe5635 | 1426 | |
0a63b66d KO |
1427 | bch_btree_insert_node(b, op, &keylist, NULL, NULL); |
1428 | BUG_ON(!bch_keylist_empty(&keylist)); | |
a1f0358b KO |
1429 | |
1430 | for (i = 0; i < nodes; i++) { | |
1431 | btree_node_free(r[i].b); | |
1432 | rw_unlock(true, r[i].b); | |
1433 | ||
1434 | r[i].b = new_nodes[i]; | |
1435 | } | |
1436 | ||
a1f0358b KO |
1437 | memmove(r, r + 1, sizeof(r[0]) * (nodes - 1)); |
1438 | r[nodes - 1].b = ERR_PTR(-EINTR); | |
1439 | ||
1440 | trace_bcache_btree_gc_coalesce(nodes); | |
cafe5635 | 1441 | gc->nodes--; |
cafe5635 | 1442 | |
0a63b66d KO |
1443 | bch_keylist_free(&keylist); |
1444 | ||
a1f0358b KO |
1445 | /* Invalidated our iterator */ |
1446 | return -EINTR; | |
1447 | ||
1448 | out_nocoalesce: | |
1449 | closure_sync(&cl); | |
0a63b66d | 1450 | bch_keylist_free(&keylist); |
a1f0358b | 1451 | |
0a63b66d | 1452 | while ((k = bch_keylist_pop(&keylist))) |
a1f0358b KO |
1453 | if (!bkey_cmp(k, &ZERO_KEY)) |
1454 | atomic_dec(&b->c->prio_blocked); | |
1455 | ||
1456 | for (i = 0; i < nodes; i++) | |
1457 | if (!IS_ERR_OR_NULL(new_nodes[i])) { | |
1458 | btree_node_free(new_nodes[i]); | |
1459 | rw_unlock(true, new_nodes[i]); | |
1460 | } | |
1461 | return 0; | |
cafe5635 KO |
1462 | } |
1463 | ||
0a63b66d KO |
1464 | static int btree_gc_rewrite_node(struct btree *b, struct btree_op *op, |
1465 | struct btree *replace) | |
1466 | { | |
1467 | struct keylist keys; | |
1468 | struct btree *n; | |
1469 | ||
1470 | if (btree_check_reserve(b, NULL)) | |
1471 | return 0; | |
1472 | ||
1473 | n = btree_node_alloc_replacement(replace, NULL); | |
1474 | ||
1475 | /* recheck reserve after allocating replacement node */ | |
1476 | if (btree_check_reserve(b, NULL)) { | |
1477 | btree_node_free(n); | |
1478 | rw_unlock(true, n); | |
1479 | return 0; | |
1480 | } | |
1481 | ||
1482 | bch_btree_node_write_sync(n); | |
1483 | ||
1484 | bch_keylist_init(&keys); | |
1485 | bch_keylist_add(&keys, &n->key); | |
1486 | ||
1487 | make_btree_freeing_key(replace, keys.top); | |
1488 | bch_keylist_push(&keys); | |
1489 | ||
1490 | bch_btree_insert_node(b, op, &keys, NULL, NULL); | |
1491 | BUG_ON(!bch_keylist_empty(&keys)); | |
1492 | ||
1493 | btree_node_free(replace); | |
1494 | rw_unlock(true, n); | |
1495 | ||
1496 | /* Invalidated our iterator */ | |
1497 | return -EINTR; | |
1498 | } | |
1499 | ||
a1f0358b | 1500 | static unsigned btree_gc_count_keys(struct btree *b) |
cafe5635 | 1501 | { |
a1f0358b KO |
1502 | struct bkey *k; |
1503 | struct btree_iter iter; | |
1504 | unsigned ret = 0; | |
cafe5635 | 1505 | |
c052dd9a | 1506 | for_each_key_filter(&b->keys, k, &iter, bch_ptr_bad) |
a1f0358b KO |
1507 | ret += bkey_u64s(k); |
1508 | ||
1509 | return ret; | |
1510 | } | |
cafe5635 | 1511 | |
a1f0358b KO |
1512 | static int btree_gc_recurse(struct btree *b, struct btree_op *op, |
1513 | struct closure *writes, struct gc_stat *gc) | |
1514 | { | |
a1f0358b KO |
1515 | int ret = 0; |
1516 | bool should_rewrite; | |
a1f0358b | 1517 | struct bkey *k; |
a1f0358b | 1518 | struct btree_iter iter; |
cafe5635 | 1519 | struct gc_merge_info r[GC_MERGE_NODES]; |
2a285686 | 1520 | struct gc_merge_info *i, *last = r + ARRAY_SIZE(r) - 1; |
cafe5635 | 1521 | |
c052dd9a | 1522 | bch_btree_iter_init(&b->keys, &iter, &b->c->gc_done); |
cafe5635 | 1523 | |
2a285686 KO |
1524 | for (i = r; i < r + ARRAY_SIZE(r); i++) |
1525 | i->b = ERR_PTR(-EINTR); | |
cafe5635 | 1526 | |
a1f0358b | 1527 | while (1) { |
a85e968e | 1528 | k = bch_btree_iter_next_filter(&iter, &b->keys, bch_ptr_bad); |
a1f0358b | 1529 | if (k) { |
0a63b66d | 1530 | r->b = bch_btree_node_get(b->c, op, k, b->level - 1, |
2452cc89 | 1531 | true, b); |
a1f0358b KO |
1532 | if (IS_ERR(r->b)) { |
1533 | ret = PTR_ERR(r->b); | |
1534 | break; | |
1535 | } | |
1536 | ||
1537 | r->keys = btree_gc_count_keys(r->b); | |
1538 | ||
0a63b66d | 1539 | ret = btree_gc_coalesce(b, op, gc, r); |
a1f0358b KO |
1540 | if (ret) |
1541 | break; | |
cafe5635 KO |
1542 | } |
1543 | ||
a1f0358b KO |
1544 | if (!last->b) |
1545 | break; | |
cafe5635 | 1546 | |
a1f0358b KO |
1547 | if (!IS_ERR(last->b)) { |
1548 | should_rewrite = btree_gc_mark_node(last->b, gc); | |
0a63b66d KO |
1549 | if (should_rewrite) { |
1550 | ret = btree_gc_rewrite_node(b, op, last->b); | |
1551 | if (ret) | |
a1f0358b | 1552 | break; |
a1f0358b KO |
1553 | } |
1554 | ||
1555 | if (last->b->level) { | |
1556 | ret = btree_gc_recurse(last->b, op, writes, gc); | |
1557 | if (ret) | |
1558 | break; | |
1559 | } | |
cafe5635 | 1560 | |
a1f0358b KO |
1561 | bkey_copy_key(&b->c->gc_done, &last->b->key); |
1562 | ||
1563 | /* | |
1564 | * Must flush leaf nodes before gc ends, since replace | |
1565 | * operations aren't journalled | |
1566 | */ | |
2a285686 | 1567 | mutex_lock(&last->b->write_lock); |
a1f0358b KO |
1568 | if (btree_node_dirty(last->b)) |
1569 | bch_btree_node_write(last->b, writes); | |
2a285686 | 1570 | mutex_unlock(&last->b->write_lock); |
a1f0358b KO |
1571 | rw_unlock(true, last->b); |
1572 | } | |
1573 | ||
1574 | memmove(r + 1, r, sizeof(r[0]) * (GC_MERGE_NODES - 1)); | |
1575 | r->b = NULL; | |
cafe5635 | 1576 | |
cafe5635 KO |
1577 | if (need_resched()) { |
1578 | ret = -EAGAIN; | |
1579 | break; | |
1580 | } | |
cafe5635 KO |
1581 | } |
1582 | ||
2a285686 KO |
1583 | for (i = r; i < r + ARRAY_SIZE(r); i++) |
1584 | if (!IS_ERR_OR_NULL(i->b)) { | |
1585 | mutex_lock(&i->b->write_lock); | |
1586 | if (btree_node_dirty(i->b)) | |
1587 | bch_btree_node_write(i->b, writes); | |
1588 | mutex_unlock(&i->b->write_lock); | |
1589 | rw_unlock(true, i->b); | |
a1f0358b | 1590 | } |
cafe5635 | 1591 | |
cafe5635 KO |
1592 | return ret; |
1593 | } | |
1594 | ||
1595 | static int bch_btree_gc_root(struct btree *b, struct btree_op *op, | |
1596 | struct closure *writes, struct gc_stat *gc) | |
1597 | { | |
1598 | struct btree *n = NULL; | |
a1f0358b KO |
1599 | int ret = 0; |
1600 | bool should_rewrite; | |
cafe5635 | 1601 | |
a1f0358b KO |
1602 | should_rewrite = btree_gc_mark_node(b, gc); |
1603 | if (should_rewrite) { | |
0a63b66d | 1604 | n = btree_node_alloc_replacement(b, NULL); |
cafe5635 | 1605 | |
a1f0358b KO |
1606 | if (!IS_ERR_OR_NULL(n)) { |
1607 | bch_btree_node_write_sync(n); | |
2a285686 | 1608 | |
a1f0358b KO |
1609 | bch_btree_set_root(n); |
1610 | btree_node_free(b); | |
1611 | rw_unlock(true, n); | |
cafe5635 | 1612 | |
a1f0358b KO |
1613 | return -EINTR; |
1614 | } | |
1615 | } | |
cafe5635 | 1616 | |
487dded8 KO |
1617 | __bch_btree_mark_key(b->c, b->level + 1, &b->key); |
1618 | ||
a1f0358b KO |
1619 | if (b->level) { |
1620 | ret = btree_gc_recurse(b, op, writes, gc); | |
1621 | if (ret) | |
1622 | return ret; | |
cafe5635 KO |
1623 | } |
1624 | ||
a1f0358b KO |
1625 | bkey_copy_key(&b->c->gc_done, &b->key); |
1626 | ||
cafe5635 KO |
1627 | return ret; |
1628 | } | |
1629 | ||
1630 | static void btree_gc_start(struct cache_set *c) | |
1631 | { | |
1632 | struct cache *ca; | |
1633 | struct bucket *b; | |
cafe5635 KO |
1634 | unsigned i; |
1635 | ||
1636 | if (!c->gc_mark_valid) | |
1637 | return; | |
1638 | ||
1639 | mutex_lock(&c->bucket_lock); | |
1640 | ||
1641 | c->gc_mark_valid = 0; | |
1642 | c->gc_done = ZERO_KEY; | |
1643 | ||
1644 | for_each_cache(ca, c, i) | |
1645 | for_each_bucket(b, ca) { | |
3a2fd9d5 | 1646 | b->last_gc = b->gen; |
29ebf465 | 1647 | if (!atomic_read(&b->pin)) { |
4fe6a816 | 1648 | SET_GC_MARK(b, 0); |
29ebf465 KO |
1649 | SET_GC_SECTORS_USED(b, 0); |
1650 | } | |
cafe5635 KO |
1651 | } |
1652 | ||
cafe5635 KO |
1653 | mutex_unlock(&c->bucket_lock); |
1654 | } | |
1655 | ||
2531d9ee | 1656 | static size_t bch_btree_gc_finish(struct cache_set *c) |
cafe5635 KO |
1657 | { |
1658 | size_t available = 0; | |
1659 | struct bucket *b; | |
1660 | struct cache *ca; | |
cafe5635 KO |
1661 | unsigned i; |
1662 | ||
1663 | mutex_lock(&c->bucket_lock); | |
1664 | ||
1665 | set_gc_sectors(c); | |
1666 | c->gc_mark_valid = 1; | |
1667 | c->need_gc = 0; | |
1668 | ||
cafe5635 KO |
1669 | for (i = 0; i < KEY_PTRS(&c->uuid_bucket); i++) |
1670 | SET_GC_MARK(PTR_BUCKET(c, &c->uuid_bucket, i), | |
1671 | GC_MARK_METADATA); | |
1672 | ||
bf0a628a NS |
1673 | /* don't reclaim buckets to which writeback keys point */ |
1674 | rcu_read_lock(); | |
1675 | for (i = 0; i < c->nr_uuids; i++) { | |
1676 | struct bcache_device *d = c->devices[i]; | |
1677 | struct cached_dev *dc; | |
1678 | struct keybuf_key *w, *n; | |
1679 | unsigned j; | |
1680 | ||
1681 | if (!d || UUID_FLASH_ONLY(&c->uuids[i])) | |
1682 | continue; | |
1683 | dc = container_of(d, struct cached_dev, disk); | |
1684 | ||
1685 | spin_lock(&dc->writeback_keys.lock); | |
1686 | rbtree_postorder_for_each_entry_safe(w, n, | |
1687 | &dc->writeback_keys.keys, node) | |
1688 | for (j = 0; j < KEY_PTRS(&w->key); j++) | |
1689 | SET_GC_MARK(PTR_BUCKET(c, &w->key, j), | |
1690 | GC_MARK_DIRTY); | |
1691 | spin_unlock(&dc->writeback_keys.lock); | |
1692 | } | |
1693 | rcu_read_unlock(); | |
1694 | ||
cafe5635 KO |
1695 | for_each_cache(ca, c, i) { |
1696 | uint64_t *i; | |
1697 | ||
1698 | ca->invalidate_needs_gc = 0; | |
1699 | ||
1700 | for (i = ca->sb.d; i < ca->sb.d + ca->sb.keys; i++) | |
1701 | SET_GC_MARK(ca->buckets + *i, GC_MARK_METADATA); | |
1702 | ||
1703 | for (i = ca->prio_buckets; | |
1704 | i < ca->prio_buckets + prio_buckets(ca) * 2; i++) | |
1705 | SET_GC_MARK(ca->buckets + *i, GC_MARK_METADATA); | |
1706 | ||
1707 | for_each_bucket(b, ca) { | |
cafe5635 KO |
1708 | c->need_gc = max(c->need_gc, bucket_gc_gen(b)); |
1709 | ||
4fe6a816 KO |
1710 | if (atomic_read(&b->pin)) |
1711 | continue; | |
1712 | ||
1713 | BUG_ON(!GC_MARK(b) && GC_SECTORS_USED(b)); | |
1714 | ||
1715 | if (!GC_MARK(b) || GC_MARK(b) == GC_MARK_RECLAIMABLE) | |
cafe5635 | 1716 | available++; |
cafe5635 KO |
1717 | } |
1718 | } | |
1719 | ||
cafe5635 KO |
1720 | mutex_unlock(&c->bucket_lock); |
1721 | return available; | |
1722 | } | |
1723 | ||
72a44517 | 1724 | static void bch_btree_gc(struct cache_set *c) |
cafe5635 | 1725 | { |
cafe5635 KO |
1726 | int ret; |
1727 | unsigned long available; | |
1728 | struct gc_stat stats; | |
1729 | struct closure writes; | |
1730 | struct btree_op op; | |
cafe5635 | 1731 | uint64_t start_time = local_clock(); |
57943511 | 1732 | |
c37511b8 | 1733 | trace_bcache_gc_start(c); |
cafe5635 KO |
1734 | |
1735 | memset(&stats, 0, sizeof(struct gc_stat)); | |
1736 | closure_init_stack(&writes); | |
b54d6934 | 1737 | bch_btree_op_init(&op, SHRT_MAX); |
cafe5635 KO |
1738 | |
1739 | btree_gc_start(c); | |
1740 | ||
a1f0358b KO |
1741 | do { |
1742 | ret = btree_root(gc_root, c, &op, &writes, &stats); | |
1743 | closure_sync(&writes); | |
cafe5635 | 1744 | |
a1f0358b KO |
1745 | if (ret && ret != -EAGAIN) |
1746 | pr_warn("gc failed!"); | |
1747 | } while (ret); | |
cafe5635 KO |
1748 | |
1749 | available = bch_btree_gc_finish(c); | |
57943511 KO |
1750 | wake_up_allocators(c); |
1751 | ||
169ef1cf | 1752 | bch_time_stats_update(&c->btree_gc_time, start_time); |
cafe5635 KO |
1753 | |
1754 | stats.key_bytes *= sizeof(uint64_t); | |
cafe5635 KO |
1755 | stats.data <<= 9; |
1756 | stats.in_use = (c->nbuckets - available) * 100 / c->nbuckets; | |
1757 | memcpy(&c->gc_stats, &stats, sizeof(struct gc_stat)); | |
cafe5635 | 1758 | |
c37511b8 | 1759 | trace_bcache_gc_end(c); |
cafe5635 | 1760 | |
72a44517 KO |
1761 | bch_moving_gc(c); |
1762 | } | |
1763 | ||
1764 | static int bch_gc_thread(void *arg) | |
1765 | { | |
1766 | struct cache_set *c = arg; | |
a1f0358b KO |
1767 | struct cache *ca; |
1768 | unsigned i; | |
72a44517 KO |
1769 | |
1770 | while (1) { | |
a1f0358b | 1771 | again: |
72a44517 KO |
1772 | bch_btree_gc(c); |
1773 | ||
1774 | set_current_state(TASK_INTERRUPTIBLE); | |
1775 | if (kthread_should_stop()) | |
1776 | break; | |
1777 | ||
a1f0358b KO |
1778 | mutex_lock(&c->bucket_lock); |
1779 | ||
1780 | for_each_cache(ca, c, i) | |
1781 | if (ca->invalidate_needs_gc) { | |
1782 | mutex_unlock(&c->bucket_lock); | |
1783 | set_current_state(TASK_RUNNING); | |
1784 | goto again; | |
1785 | } | |
1786 | ||
1787 | mutex_unlock(&c->bucket_lock); | |
1788 | ||
72a44517 KO |
1789 | try_to_freeze(); |
1790 | schedule(); | |
1791 | } | |
1792 | ||
1793 | return 0; | |
cafe5635 KO |
1794 | } |
1795 | ||
72a44517 | 1796 | int bch_gc_thread_start(struct cache_set *c) |
cafe5635 | 1797 | { |
72a44517 KO |
1798 | c->gc_thread = kthread_create(bch_gc_thread, c, "bcache_gc"); |
1799 | if (IS_ERR(c->gc_thread)) | |
1800 | return PTR_ERR(c->gc_thread); | |
1801 | ||
1802 | set_task_state(c->gc_thread, TASK_INTERRUPTIBLE); | |
1803 | return 0; | |
cafe5635 KO |
1804 | } |
1805 | ||
1806 | /* Initial partial gc */ | |
1807 | ||
487dded8 | 1808 | static int bch_btree_check_recurse(struct btree *b, struct btree_op *op) |
cafe5635 | 1809 | { |
50310164 | 1810 | int ret = 0; |
50310164 | 1811 | struct bkey *k, *p = NULL; |
cafe5635 KO |
1812 | struct btree_iter iter; |
1813 | ||
487dded8 KO |
1814 | for_each_key_filter(&b->keys, k, &iter, bch_ptr_invalid) |
1815 | bch_initial_mark_key(b->c, b->level, k); | |
cafe5635 | 1816 | |
487dded8 | 1817 | bch_initial_mark_key(b->c, b->level + 1, &b->key); |
cafe5635 KO |
1818 | |
1819 | if (b->level) { | |
c052dd9a | 1820 | bch_btree_iter_init(&b->keys, &iter, NULL); |
cafe5635 | 1821 | |
50310164 | 1822 | do { |
a85e968e KO |
1823 | k = bch_btree_iter_next_filter(&iter, &b->keys, |
1824 | bch_ptr_bad); | |
50310164 | 1825 | if (k) |
2452cc89 | 1826 | btree_node_prefetch(b, k); |
cafe5635 | 1827 | |
50310164 | 1828 | if (p) |
487dded8 | 1829 | ret = btree(check_recurse, p, b, op); |
cafe5635 | 1830 | |
50310164 KO |
1831 | p = k; |
1832 | } while (p && !ret); | |
cafe5635 KO |
1833 | } |
1834 | ||
487dded8 | 1835 | return ret; |
cafe5635 KO |
1836 | } |
1837 | ||
c18536a7 | 1838 | int bch_btree_check(struct cache_set *c) |
cafe5635 | 1839 | { |
c18536a7 | 1840 | struct btree_op op; |
cafe5635 | 1841 | |
b54d6934 | 1842 | bch_btree_op_init(&op, SHRT_MAX); |
cafe5635 | 1843 | |
487dded8 | 1844 | return btree_root(check_recurse, c, &op); |
cafe5635 KO |
1845 | } |
1846 | ||
2531d9ee KO |
1847 | void bch_initial_gc_finish(struct cache_set *c) |
1848 | { | |
1849 | struct cache *ca; | |
1850 | struct bucket *b; | |
1851 | unsigned i; | |
1852 | ||
1853 | bch_btree_gc_finish(c); | |
1854 | ||
1855 | mutex_lock(&c->bucket_lock); | |
1856 | ||
1857 | /* | |
1858 | * We need to put some unused buckets directly on the prio freelist in | |
1859 | * order to get the allocator thread started - it needs freed buckets in | |
1860 | * order to rewrite the prios and gens, and it needs to rewrite prios | |
1861 | * and gens in order to free buckets. | |
1862 | * | |
1863 | * This is only safe for buckets that have no live data in them, which | |
1864 | * there should always be some of. | |
1865 | */ | |
1866 | for_each_cache(ca, c, i) { | |
1867 | for_each_bucket(b, ca) { | |
1868 | if (fifo_full(&ca->free[RESERVE_PRIO])) | |
1869 | break; | |
1870 | ||
1871 | if (bch_can_invalidate_bucket(ca, b) && | |
1872 | !GC_MARK(b)) { | |
1873 | __bch_invalidate_one_bucket(ca, b); | |
1874 | fifo_push(&ca->free[RESERVE_PRIO], | |
1875 | b - ca->buckets); | |
1876 | } | |
1877 | } | |
1878 | } | |
1879 | ||
1880 | mutex_unlock(&c->bucket_lock); | |
1881 | } | |
1882 | ||
cafe5635 KO |
1883 | /* Btree insertion */ |
1884 | ||
829a60b9 KO |
1885 | static bool btree_insert_key(struct btree *b, struct bkey *k, |
1886 | struct bkey *replace_key) | |
cafe5635 | 1887 | { |
829a60b9 | 1888 | unsigned status; |
cafe5635 KO |
1889 | |
1890 | BUG_ON(bkey_cmp(k, &b->key) > 0); | |
1fa8455d | 1891 | |
829a60b9 KO |
1892 | status = bch_btree_insert_key(&b->keys, k, replace_key); |
1893 | if (status != BTREE_INSERT_STATUS_NO_INSERT) { | |
1894 | bch_check_keys(&b->keys, "%u for %s", status, | |
1895 | replace_key ? "replace" : "insert"); | |
cafe5635 | 1896 | |
829a60b9 KO |
1897 | trace_bcache_btree_insert_key(b, k, replace_key != NULL, |
1898 | status); | |
1899 | return true; | |
1900 | } else | |
1901 | return false; | |
cafe5635 KO |
1902 | } |
1903 | ||
59158fde KO |
1904 | static size_t insert_u64s_remaining(struct btree *b) |
1905 | { | |
3572324a | 1906 | long ret = bch_btree_keys_u64s_remaining(&b->keys); |
59158fde KO |
1907 | |
1908 | /* | |
1909 | * Might land in the middle of an existing extent and have to split it | |
1910 | */ | |
1911 | if (b->keys.ops->is_extents) | |
1912 | ret -= KEY_MAX_U64S; | |
1913 | ||
1914 | return max(ret, 0L); | |
1915 | } | |
1916 | ||
26c949f8 | 1917 | static bool bch_btree_insert_keys(struct btree *b, struct btree_op *op, |
1b207d80 KO |
1918 | struct keylist *insert_keys, |
1919 | struct bkey *replace_key) | |
cafe5635 KO |
1920 | { |
1921 | bool ret = false; | |
dc9d98d6 | 1922 | int oldsize = bch_count_data(&b->keys); |
cafe5635 | 1923 | |
26c949f8 | 1924 | while (!bch_keylist_empty(insert_keys)) { |
c2f95ae2 | 1925 | struct bkey *k = insert_keys->keys; |
26c949f8 | 1926 | |
59158fde | 1927 | if (bkey_u64s(k) > insert_u64s_remaining(b)) |
403b6cde KO |
1928 | break; |
1929 | ||
1930 | if (bkey_cmp(k, &b->key) <= 0) { | |
3a3b6a4e KO |
1931 | if (!b->level) |
1932 | bkey_put(b->c, k); | |
26c949f8 | 1933 | |
829a60b9 | 1934 | ret |= btree_insert_key(b, k, replace_key); |
26c949f8 KO |
1935 | bch_keylist_pop_front(insert_keys); |
1936 | } else if (bkey_cmp(&START_KEY(k), &b->key) < 0) { | |
26c949f8 | 1937 | BKEY_PADDED(key) temp; |
c2f95ae2 | 1938 | bkey_copy(&temp.key, insert_keys->keys); |
26c949f8 KO |
1939 | |
1940 | bch_cut_back(&b->key, &temp.key); | |
c2f95ae2 | 1941 | bch_cut_front(&b->key, insert_keys->keys); |
26c949f8 | 1942 | |
829a60b9 | 1943 | ret |= btree_insert_key(b, &temp.key, replace_key); |
26c949f8 KO |
1944 | break; |
1945 | } else { | |
1946 | break; | |
1947 | } | |
cafe5635 KO |
1948 | } |
1949 | ||
829a60b9 KO |
1950 | if (!ret) |
1951 | op->insert_collision = true; | |
1952 | ||
403b6cde KO |
1953 | BUG_ON(!bch_keylist_empty(insert_keys) && b->level); |
1954 | ||
dc9d98d6 | 1955 | BUG_ON(bch_count_data(&b->keys) < oldsize); |
cafe5635 KO |
1956 | return ret; |
1957 | } | |
1958 | ||
26c949f8 KO |
1959 | static int btree_split(struct btree *b, struct btree_op *op, |
1960 | struct keylist *insert_keys, | |
1b207d80 | 1961 | struct bkey *replace_key) |
cafe5635 | 1962 | { |
d6fd3b11 | 1963 | bool split; |
cafe5635 KO |
1964 | struct btree *n1, *n2 = NULL, *n3 = NULL; |
1965 | uint64_t start_time = local_clock(); | |
b54d6934 | 1966 | struct closure cl; |
17e21a9f | 1967 | struct keylist parent_keys; |
b54d6934 KO |
1968 | |
1969 | closure_init_stack(&cl); | |
17e21a9f | 1970 | bch_keylist_init(&parent_keys); |
cafe5635 | 1971 | |
0a63b66d KO |
1972 | if (btree_check_reserve(b, op)) { |
1973 | if (!b->level) | |
1974 | return -EINTR; | |
1975 | else | |
1976 | WARN(1, "insufficient reserve for split\n"); | |
1977 | } | |
78365411 | 1978 | |
0a63b66d | 1979 | n1 = btree_node_alloc_replacement(b, op); |
cafe5635 KO |
1980 | if (IS_ERR(n1)) |
1981 | goto err; | |
1982 | ||
ee811287 KO |
1983 | split = set_blocks(btree_bset_first(n1), |
1984 | block_bytes(n1->c)) > (btree_blocks(b) * 4) / 5; | |
cafe5635 | 1985 | |
cafe5635 KO |
1986 | if (split) { |
1987 | unsigned keys = 0; | |
1988 | ||
ee811287 | 1989 | trace_bcache_btree_node_split(b, btree_bset_first(n1)->keys); |
c37511b8 | 1990 | |
2452cc89 | 1991 | n2 = bch_btree_node_alloc(b->c, op, b->level, b->parent); |
cafe5635 KO |
1992 | if (IS_ERR(n2)) |
1993 | goto err_free1; | |
1994 | ||
d6fd3b11 | 1995 | if (!b->parent) { |
2452cc89 | 1996 | n3 = bch_btree_node_alloc(b->c, op, b->level + 1, NULL); |
cafe5635 KO |
1997 | if (IS_ERR(n3)) |
1998 | goto err_free2; | |
1999 | } | |
2000 | ||
2a285686 KO |
2001 | mutex_lock(&n1->write_lock); |
2002 | mutex_lock(&n2->write_lock); | |
2003 | ||
1b207d80 | 2004 | bch_btree_insert_keys(n1, op, insert_keys, replace_key); |
cafe5635 | 2005 | |
d6fd3b11 KO |
2006 | /* |
2007 | * Has to be a linear search because we don't have an auxiliary | |
cafe5635 KO |
2008 | * search tree yet |
2009 | */ | |
2010 | ||
ee811287 KO |
2011 | while (keys < (btree_bset_first(n1)->keys * 3) / 5) |
2012 | keys += bkey_u64s(bset_bkey_idx(btree_bset_first(n1), | |
fafff81c | 2013 | keys)); |
cafe5635 | 2014 | |
fafff81c | 2015 | bkey_copy_key(&n1->key, |
ee811287 KO |
2016 | bset_bkey_idx(btree_bset_first(n1), keys)); |
2017 | keys += bkey_u64s(bset_bkey_idx(btree_bset_first(n1), keys)); | |
cafe5635 | 2018 | |
ee811287 KO |
2019 | btree_bset_first(n2)->keys = btree_bset_first(n1)->keys - keys; |
2020 | btree_bset_first(n1)->keys = keys; | |
cafe5635 | 2021 | |
ee811287 KO |
2022 | memcpy(btree_bset_first(n2)->start, |
2023 | bset_bkey_last(btree_bset_first(n1)), | |
2024 | btree_bset_first(n2)->keys * sizeof(uint64_t)); | |
cafe5635 KO |
2025 | |
2026 | bkey_copy_key(&n2->key, &b->key); | |
2027 | ||
17e21a9f | 2028 | bch_keylist_add(&parent_keys, &n2->key); |
b54d6934 | 2029 | bch_btree_node_write(n2, &cl); |
2a285686 | 2030 | mutex_unlock(&n2->write_lock); |
cafe5635 | 2031 | rw_unlock(true, n2); |
c37511b8 | 2032 | } else { |
ee811287 | 2033 | trace_bcache_btree_node_compact(b, btree_bset_first(n1)->keys); |
c37511b8 | 2034 | |
2a285686 | 2035 | mutex_lock(&n1->write_lock); |
1b207d80 | 2036 | bch_btree_insert_keys(n1, op, insert_keys, replace_key); |
c37511b8 | 2037 | } |
cafe5635 | 2038 | |
17e21a9f | 2039 | bch_keylist_add(&parent_keys, &n1->key); |
b54d6934 | 2040 | bch_btree_node_write(n1, &cl); |
2a285686 | 2041 | mutex_unlock(&n1->write_lock); |
cafe5635 KO |
2042 | |
2043 | if (n3) { | |
d6fd3b11 | 2044 | /* Depth increases, make a new root */ |
2a285686 | 2045 | mutex_lock(&n3->write_lock); |
cafe5635 | 2046 | bkey_copy_key(&n3->key, &MAX_KEY); |
17e21a9f | 2047 | bch_btree_insert_keys(n3, op, &parent_keys, NULL); |
b54d6934 | 2048 | bch_btree_node_write(n3, &cl); |
2a285686 | 2049 | mutex_unlock(&n3->write_lock); |
cafe5635 | 2050 | |
b54d6934 | 2051 | closure_sync(&cl); |
cafe5635 KO |
2052 | bch_btree_set_root(n3); |
2053 | rw_unlock(true, n3); | |
d6fd3b11 KO |
2054 | } else if (!b->parent) { |
2055 | /* Root filled up but didn't need to be split */ | |
b54d6934 | 2056 | closure_sync(&cl); |
cafe5635 KO |
2057 | bch_btree_set_root(n1); |
2058 | } else { | |
17e21a9f | 2059 | /* Split a non root node */ |
b54d6934 | 2060 | closure_sync(&cl); |
17e21a9f KO |
2061 | make_btree_freeing_key(b, parent_keys.top); |
2062 | bch_keylist_push(&parent_keys); | |
2063 | ||
17e21a9f KO |
2064 | bch_btree_insert_node(b->parent, op, &parent_keys, NULL, NULL); |
2065 | BUG_ON(!bch_keylist_empty(&parent_keys)); | |
cafe5635 KO |
2066 | } |
2067 | ||
05335cff | 2068 | btree_node_free(b); |
cafe5635 | 2069 | rw_unlock(true, n1); |
cafe5635 | 2070 | |
169ef1cf | 2071 | bch_time_stats_update(&b->c->btree_split_time, start_time); |
cafe5635 KO |
2072 | |
2073 | return 0; | |
2074 | err_free2: | |
5f5837d2 | 2075 | bkey_put(b->c, &n2->key); |
e8e1d468 | 2076 | btree_node_free(n2); |
cafe5635 KO |
2077 | rw_unlock(true, n2); |
2078 | err_free1: | |
5f5837d2 | 2079 | bkey_put(b->c, &n1->key); |
e8e1d468 | 2080 | btree_node_free(n1); |
cafe5635 KO |
2081 | rw_unlock(true, n1); |
2082 | err: | |
0a63b66d | 2083 | WARN(1, "bcache: btree split failed (level %u)", b->level); |
5f5837d2 | 2084 | |
cafe5635 KO |
2085 | if (n3 == ERR_PTR(-EAGAIN) || |
2086 | n2 == ERR_PTR(-EAGAIN) || | |
2087 | n1 == ERR_PTR(-EAGAIN)) | |
2088 | return -EAGAIN; | |
2089 | ||
cafe5635 KO |
2090 | return -ENOMEM; |
2091 | } | |
2092 | ||
26c949f8 | 2093 | static int bch_btree_insert_node(struct btree *b, struct btree_op *op, |
c18536a7 | 2094 | struct keylist *insert_keys, |
1b207d80 KO |
2095 | atomic_t *journal_ref, |
2096 | struct bkey *replace_key) | |
cafe5635 | 2097 | { |
2a285686 KO |
2098 | struct closure cl; |
2099 | ||
17e21a9f KO |
2100 | BUG_ON(b->level && replace_key); |
2101 | ||
2a285686 KO |
2102 | closure_init_stack(&cl); |
2103 | ||
2104 | mutex_lock(&b->write_lock); | |
2105 | ||
2106 | if (write_block(b) != btree_bset_last(b) && | |
2107 | b->keys.last_set_unwritten) | |
2108 | bch_btree_init_next(b); /* just wrote a set */ | |
2109 | ||
59158fde | 2110 | if (bch_keylist_nkeys(insert_keys) > insert_u64s_remaining(b)) { |
2a285686 KO |
2111 | mutex_unlock(&b->write_lock); |
2112 | goto split; | |
2113 | } | |
3b3e9e50 | 2114 | |
2a285686 | 2115 | BUG_ON(write_block(b) != btree_bset_last(b)); |
cafe5635 | 2116 | |
2a285686 KO |
2117 | if (bch_btree_insert_keys(b, op, insert_keys, replace_key)) { |
2118 | if (!b->level) | |
2119 | bch_btree_leaf_dirty(b, journal_ref); | |
2120 | else | |
2121 | bch_btree_node_write(b, &cl); | |
2122 | } | |
17e21a9f | 2123 | |
2a285686 KO |
2124 | mutex_unlock(&b->write_lock); |
2125 | ||
2126 | /* wait for btree node write if necessary, after unlock */ | |
2127 | closure_sync(&cl); | |
2128 | ||
2129 | return 0; | |
2130 | split: | |
2131 | if (current->bio_list) { | |
2132 | op->lock = b->c->root->level + 1; | |
2133 | return -EAGAIN; | |
2134 | } else if (op->lock <= b->c->root->level) { | |
2135 | op->lock = b->c->root->level + 1; | |
2136 | return -EINTR; | |
2137 | } else { | |
2138 | /* Invalidated all iterators */ | |
2139 | int ret = btree_split(b, op, insert_keys, replace_key); | |
2140 | ||
2141 | if (bch_keylist_empty(insert_keys)) | |
2142 | return 0; | |
2143 | else if (!ret) | |
2144 | return -EINTR; | |
2145 | return ret; | |
17e21a9f | 2146 | } |
26c949f8 | 2147 | } |
cafe5635 | 2148 | |
e7c590eb KO |
2149 | int bch_btree_insert_check_key(struct btree *b, struct btree_op *op, |
2150 | struct bkey *check_key) | |
2151 | { | |
2152 | int ret = -EINTR; | |
2153 | uint64_t btree_ptr = b->key.ptr[0]; | |
2154 | unsigned long seq = b->seq; | |
2155 | struct keylist insert; | |
2156 | bool upgrade = op->lock == -1; | |
2157 | ||
2158 | bch_keylist_init(&insert); | |
2159 | ||
2160 | if (upgrade) { | |
2161 | rw_unlock(false, b); | |
2162 | rw_lock(true, b, b->level); | |
2163 | ||
2164 | if (b->key.ptr[0] != btree_ptr || | |
2165 | b->seq != seq + 1) | |
2166 | goto out; | |
2167 | } | |
2168 | ||
2169 | SET_KEY_PTRS(check_key, 1); | |
2170 | get_random_bytes(&check_key->ptr[0], sizeof(uint64_t)); | |
2171 | ||
2172 | SET_PTR_DEV(check_key, 0, PTR_CHECK_DEV); | |
2173 | ||
2174 | bch_keylist_add(&insert, check_key); | |
2175 | ||
1b207d80 | 2176 | ret = bch_btree_insert_node(b, op, &insert, NULL, NULL); |
e7c590eb KO |
2177 | |
2178 | BUG_ON(!ret && !bch_keylist_empty(&insert)); | |
2179 | out: | |
2180 | if (upgrade) | |
2181 | downgrade_write(&b->lock); | |
2182 | return ret; | |
2183 | } | |
2184 | ||
cc7b8819 KO |
2185 | struct btree_insert_op { |
2186 | struct btree_op op; | |
2187 | struct keylist *keys; | |
2188 | atomic_t *journal_ref; | |
2189 | struct bkey *replace_key; | |
2190 | }; | |
cafe5635 | 2191 | |
08239ca2 | 2192 | static int btree_insert_fn(struct btree_op *b_op, struct btree *b) |
cc7b8819 KO |
2193 | { |
2194 | struct btree_insert_op *op = container_of(b_op, | |
2195 | struct btree_insert_op, op); | |
cafe5635 | 2196 | |
cc7b8819 KO |
2197 | int ret = bch_btree_insert_node(b, &op->op, op->keys, |
2198 | op->journal_ref, op->replace_key); | |
2199 | if (ret && !bch_keylist_empty(op->keys)) | |
2200 | return ret; | |
2201 | else | |
2202 | return MAP_DONE; | |
cafe5635 KO |
2203 | } |
2204 | ||
cc7b8819 KO |
2205 | int bch_btree_insert(struct cache_set *c, struct keylist *keys, |
2206 | atomic_t *journal_ref, struct bkey *replace_key) | |
cafe5635 | 2207 | { |
cc7b8819 | 2208 | struct btree_insert_op op; |
cafe5635 | 2209 | int ret = 0; |
cafe5635 | 2210 | |
cc7b8819 | 2211 | BUG_ON(current->bio_list); |
4f3d4014 | 2212 | BUG_ON(bch_keylist_empty(keys)); |
cafe5635 | 2213 | |
cc7b8819 KO |
2214 | bch_btree_op_init(&op.op, 0); |
2215 | op.keys = keys; | |
2216 | op.journal_ref = journal_ref; | |
2217 | op.replace_key = replace_key; | |
cafe5635 | 2218 | |
cc7b8819 KO |
2219 | while (!ret && !bch_keylist_empty(keys)) { |
2220 | op.op.lock = 0; | |
2221 | ret = bch_btree_map_leaf_nodes(&op.op, c, | |
2222 | &START_KEY(keys->keys), | |
2223 | btree_insert_fn); | |
2224 | } | |
cafe5635 | 2225 | |
cc7b8819 KO |
2226 | if (ret) { |
2227 | struct bkey *k; | |
cafe5635 | 2228 | |
cc7b8819 | 2229 | pr_err("error %i", ret); |
cafe5635 | 2230 | |
cc7b8819 | 2231 | while ((k = bch_keylist_pop(keys))) |
3a3b6a4e | 2232 | bkey_put(c, k); |
cc7b8819 KO |
2233 | } else if (op.op.insert_collision) |
2234 | ret = -ESRCH; | |
6054c6d4 | 2235 | |
cafe5635 KO |
2236 | return ret; |
2237 | } | |
2238 | ||
2239 | void bch_btree_set_root(struct btree *b) | |
2240 | { | |
2241 | unsigned i; | |
e49c7c37 KO |
2242 | struct closure cl; |
2243 | ||
2244 | closure_init_stack(&cl); | |
cafe5635 | 2245 | |
c37511b8 KO |
2246 | trace_bcache_btree_set_root(b); |
2247 | ||
cafe5635 KO |
2248 | BUG_ON(!b->written); |
2249 | ||
2250 | for (i = 0; i < KEY_PTRS(&b->key); i++) | |
2251 | BUG_ON(PTR_BUCKET(b->c, &b->key, i)->prio != BTREE_PRIO); | |
2252 | ||
2253 | mutex_lock(&b->c->bucket_lock); | |
2254 | list_del_init(&b->list); | |
2255 | mutex_unlock(&b->c->bucket_lock); | |
2256 | ||
2257 | b->c->root = b; | |
cafe5635 | 2258 | |
e49c7c37 KO |
2259 | bch_journal_meta(b->c, &cl); |
2260 | closure_sync(&cl); | |
cafe5635 KO |
2261 | } |
2262 | ||
48dad8ba KO |
2263 | /* Map across nodes or keys */ |
2264 | ||
2265 | static int bch_btree_map_nodes_recurse(struct btree *b, struct btree_op *op, | |
2266 | struct bkey *from, | |
2267 | btree_map_nodes_fn *fn, int flags) | |
2268 | { | |
2269 | int ret = MAP_CONTINUE; | |
2270 | ||
2271 | if (b->level) { | |
2272 | struct bkey *k; | |
2273 | struct btree_iter iter; | |
2274 | ||
c052dd9a | 2275 | bch_btree_iter_init(&b->keys, &iter, from); |
48dad8ba | 2276 | |
a85e968e | 2277 | while ((k = bch_btree_iter_next_filter(&iter, &b->keys, |
48dad8ba KO |
2278 | bch_ptr_bad))) { |
2279 | ret = btree(map_nodes_recurse, k, b, | |
2280 | op, from, fn, flags); | |
2281 | from = NULL; | |
2282 | ||
2283 | if (ret != MAP_CONTINUE) | |
2284 | return ret; | |
2285 | } | |
2286 | } | |
2287 | ||
2288 | if (!b->level || flags == MAP_ALL_NODES) | |
2289 | ret = fn(op, b); | |
2290 | ||
2291 | return ret; | |
2292 | } | |
2293 | ||
2294 | int __bch_btree_map_nodes(struct btree_op *op, struct cache_set *c, | |
2295 | struct bkey *from, btree_map_nodes_fn *fn, int flags) | |
2296 | { | |
b54d6934 | 2297 | return btree_root(map_nodes_recurse, c, op, from, fn, flags); |
48dad8ba KO |
2298 | } |
2299 | ||
2300 | static int bch_btree_map_keys_recurse(struct btree *b, struct btree_op *op, | |
2301 | struct bkey *from, btree_map_keys_fn *fn, | |
2302 | int flags) | |
2303 | { | |
2304 | int ret = MAP_CONTINUE; | |
2305 | struct bkey *k; | |
2306 | struct btree_iter iter; | |
2307 | ||
c052dd9a | 2308 | bch_btree_iter_init(&b->keys, &iter, from); |
48dad8ba | 2309 | |
a85e968e | 2310 | while ((k = bch_btree_iter_next_filter(&iter, &b->keys, bch_ptr_bad))) { |
48dad8ba KO |
2311 | ret = !b->level |
2312 | ? fn(op, b, k) | |
2313 | : btree(map_keys_recurse, k, b, op, from, fn, flags); | |
2314 | from = NULL; | |
2315 | ||
2316 | if (ret != MAP_CONTINUE) | |
2317 | return ret; | |
2318 | } | |
2319 | ||
2320 | if (!b->level && (flags & MAP_END_KEY)) | |
2321 | ret = fn(op, b, &KEY(KEY_INODE(&b->key), | |
2322 | KEY_OFFSET(&b->key), 0)); | |
2323 | ||
2324 | return ret; | |
2325 | } | |
2326 | ||
2327 | int bch_btree_map_keys(struct btree_op *op, struct cache_set *c, | |
2328 | struct bkey *from, btree_map_keys_fn *fn, int flags) | |
2329 | { | |
b54d6934 | 2330 | return btree_root(map_keys_recurse, c, op, from, fn, flags); |
48dad8ba KO |
2331 | } |
2332 | ||
cafe5635 KO |
2333 | /* Keybuf code */ |
2334 | ||
2335 | static inline int keybuf_cmp(struct keybuf_key *l, struct keybuf_key *r) | |
2336 | { | |
2337 | /* Overlapping keys compare equal */ | |
2338 | if (bkey_cmp(&l->key, &START_KEY(&r->key)) <= 0) | |
2339 | return -1; | |
2340 | if (bkey_cmp(&START_KEY(&l->key), &r->key) >= 0) | |
2341 | return 1; | |
2342 | return 0; | |
2343 | } | |
2344 | ||
2345 | static inline int keybuf_nonoverlapping_cmp(struct keybuf_key *l, | |
2346 | struct keybuf_key *r) | |
2347 | { | |
2348 | return clamp_t(int64_t, bkey_cmp(&l->key, &r->key), -1, 1); | |
2349 | } | |
2350 | ||
48dad8ba KO |
2351 | struct refill { |
2352 | struct btree_op op; | |
48a915a8 | 2353 | unsigned nr_found; |
48dad8ba KO |
2354 | struct keybuf *buf; |
2355 | struct bkey *end; | |
2356 | keybuf_pred_fn *pred; | |
2357 | }; | |
cafe5635 | 2358 | |
48dad8ba KO |
2359 | static int refill_keybuf_fn(struct btree_op *op, struct btree *b, |
2360 | struct bkey *k) | |
2361 | { | |
2362 | struct refill *refill = container_of(op, struct refill, op); | |
2363 | struct keybuf *buf = refill->buf; | |
2364 | int ret = MAP_CONTINUE; | |
cafe5635 | 2365 | |
48dad8ba KO |
2366 | if (bkey_cmp(k, refill->end) >= 0) { |
2367 | ret = MAP_DONE; | |
2368 | goto out; | |
2369 | } | |
cafe5635 | 2370 | |
48dad8ba KO |
2371 | if (!KEY_SIZE(k)) /* end key */ |
2372 | goto out; | |
cafe5635 | 2373 | |
48dad8ba KO |
2374 | if (refill->pred(buf, k)) { |
2375 | struct keybuf_key *w; | |
cafe5635 | 2376 | |
48dad8ba | 2377 | spin_lock(&buf->lock); |
cafe5635 | 2378 | |
48dad8ba KO |
2379 | w = array_alloc(&buf->freelist); |
2380 | if (!w) { | |
2381 | spin_unlock(&buf->lock); | |
2382 | return MAP_DONE; | |
2383 | } | |
cafe5635 | 2384 | |
48dad8ba KO |
2385 | w->private = NULL; |
2386 | bkey_copy(&w->key, k); | |
cafe5635 | 2387 | |
48dad8ba KO |
2388 | if (RB_INSERT(&buf->keys, w, node, keybuf_cmp)) |
2389 | array_free(&buf->freelist, w); | |
48a915a8 KO |
2390 | else |
2391 | refill->nr_found++; | |
cafe5635 | 2392 | |
48dad8ba KO |
2393 | if (array_freelist_empty(&buf->freelist)) |
2394 | ret = MAP_DONE; | |
cafe5635 | 2395 | |
48dad8ba | 2396 | spin_unlock(&buf->lock); |
cafe5635 | 2397 | } |
48dad8ba KO |
2398 | out: |
2399 | buf->last_scanned = *k; | |
2400 | return ret; | |
cafe5635 KO |
2401 | } |
2402 | ||
2403 | void bch_refill_keybuf(struct cache_set *c, struct keybuf *buf, | |
72c27061 | 2404 | struct bkey *end, keybuf_pred_fn *pred) |
cafe5635 KO |
2405 | { |
2406 | struct bkey start = buf->last_scanned; | |
48dad8ba | 2407 | struct refill refill; |
cafe5635 KO |
2408 | |
2409 | cond_resched(); | |
2410 | ||
b54d6934 | 2411 | bch_btree_op_init(&refill.op, -1); |
48a915a8 KO |
2412 | refill.nr_found = 0; |
2413 | refill.buf = buf; | |
2414 | refill.end = end; | |
2415 | refill.pred = pred; | |
48dad8ba KO |
2416 | |
2417 | bch_btree_map_keys(&refill.op, c, &buf->last_scanned, | |
2418 | refill_keybuf_fn, MAP_END_KEY); | |
cafe5635 | 2419 | |
48a915a8 KO |
2420 | trace_bcache_keyscan(refill.nr_found, |
2421 | KEY_INODE(&start), KEY_OFFSET(&start), | |
2422 | KEY_INODE(&buf->last_scanned), | |
2423 | KEY_OFFSET(&buf->last_scanned)); | |
cafe5635 KO |
2424 | |
2425 | spin_lock(&buf->lock); | |
2426 | ||
2427 | if (!RB_EMPTY_ROOT(&buf->keys)) { | |
2428 | struct keybuf_key *w; | |
2429 | w = RB_FIRST(&buf->keys, struct keybuf_key, node); | |
2430 | buf->start = START_KEY(&w->key); | |
2431 | ||
2432 | w = RB_LAST(&buf->keys, struct keybuf_key, node); | |
2433 | buf->end = w->key; | |
2434 | } else { | |
2435 | buf->start = MAX_KEY; | |
2436 | buf->end = MAX_KEY; | |
2437 | } | |
2438 | ||
2439 | spin_unlock(&buf->lock); | |
2440 | } | |
2441 | ||
2442 | static void __bch_keybuf_del(struct keybuf *buf, struct keybuf_key *w) | |
2443 | { | |
2444 | rb_erase(&w->node, &buf->keys); | |
2445 | array_free(&buf->freelist, w); | |
2446 | } | |
2447 | ||
2448 | void bch_keybuf_del(struct keybuf *buf, struct keybuf_key *w) | |
2449 | { | |
2450 | spin_lock(&buf->lock); | |
2451 | __bch_keybuf_del(buf, w); | |
2452 | spin_unlock(&buf->lock); | |
2453 | } | |
2454 | ||
2455 | bool bch_keybuf_check_overlapping(struct keybuf *buf, struct bkey *start, | |
2456 | struct bkey *end) | |
2457 | { | |
2458 | bool ret = false; | |
2459 | struct keybuf_key *p, *w, s; | |
2460 | s.key = *start; | |
2461 | ||
2462 | if (bkey_cmp(end, &buf->start) <= 0 || | |
2463 | bkey_cmp(start, &buf->end) >= 0) | |
2464 | return false; | |
2465 | ||
2466 | spin_lock(&buf->lock); | |
2467 | w = RB_GREATER(&buf->keys, s, node, keybuf_nonoverlapping_cmp); | |
2468 | ||
2469 | while (w && bkey_cmp(&START_KEY(&w->key), end) < 0) { | |
2470 | p = w; | |
2471 | w = RB_NEXT(w, node); | |
2472 | ||
2473 | if (p->private) | |
2474 | ret = true; | |
2475 | else | |
2476 | __bch_keybuf_del(buf, p); | |
2477 | } | |
2478 | ||
2479 | spin_unlock(&buf->lock); | |
2480 | return ret; | |
2481 | } | |
2482 | ||
2483 | struct keybuf_key *bch_keybuf_next(struct keybuf *buf) | |
2484 | { | |
2485 | struct keybuf_key *w; | |
2486 | spin_lock(&buf->lock); | |
2487 | ||
2488 | w = RB_FIRST(&buf->keys, struct keybuf_key, node); | |
2489 | ||
2490 | while (w && w->private) | |
2491 | w = RB_NEXT(w, node); | |
2492 | ||
2493 | if (w) | |
2494 | w->private = ERR_PTR(-EINTR); | |
2495 | ||
2496 | spin_unlock(&buf->lock); | |
2497 | return w; | |
2498 | } | |
2499 | ||
2500 | struct keybuf_key *bch_keybuf_next_rescan(struct cache_set *c, | |
48dad8ba KO |
2501 | struct keybuf *buf, |
2502 | struct bkey *end, | |
2503 | keybuf_pred_fn *pred) | |
cafe5635 KO |
2504 | { |
2505 | struct keybuf_key *ret; | |
2506 | ||
2507 | while (1) { | |
2508 | ret = bch_keybuf_next(buf); | |
2509 | if (ret) | |
2510 | break; | |
2511 | ||
2512 | if (bkey_cmp(&buf->last_scanned, end) >= 0) { | |
2513 | pr_debug("scan finished"); | |
2514 | break; | |
2515 | } | |
2516 | ||
72c27061 | 2517 | bch_refill_keybuf(c, buf, end, pred); |
cafe5635 KO |
2518 | } |
2519 | ||
2520 | return ret; | |
2521 | } | |
2522 | ||
72c27061 | 2523 | void bch_keybuf_init(struct keybuf *buf) |
cafe5635 | 2524 | { |
cafe5635 KO |
2525 | buf->last_scanned = MAX_KEY; |
2526 | buf->keys = RB_ROOT; | |
2527 | ||
2528 | spin_lock_init(&buf->lock); | |
2529 | array_allocator_init(&buf->freelist); | |
2530 | } |