1 /*
2 * Ceph - scalable distributed file system
3 *
4 * Copyright (C) 2015 Intel Corporation All Rights Reserved
5 *
6 * This is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License version 2.1, as published by the Free Software
9 * Foundation. See file COPYING.
10 *
11 */
12
13 #ifdef __KERNEL__
14 # include <linux/string.h>
15 # include <linux/slab.h>
16 # include <linux/bug.h>
17 # include <linux/kernel.h>
18 # include <linux/crush/crush.h>
19 # include <linux/crush/hash.h>
20 # include <linux/crush/mapper.h>
21 #else
22 # include "crush_compat.h"
23 # include "crush.h"
24 # include "hash.h"
25 # include "mapper.h"
26 #endif
27 #include "crush_ln_table.h"
28
29 #define dprintk(args...) /* printf(args) */
30
31 /*
32 * Implement the core CRUSH mapping algorithm.
33 */
34
35 /**
36 * crush_find_rule - find a crush_rule id for a given ruleset, type, and size.
37 * @map: the crush_map
38 * @ruleset: the storage ruleset id (user defined)
39 * @type: storage ruleset type (user defined)
40 * @size: output set size
41 */
42 int crush_find_rule(const struct crush_map *map, int ruleset, int type, int size)
43 {
44 __u32 i;
45
46 for (i = 0; i < map->max_rules; i++) {
47 if (map->rules[i] &&
48 map->rules[i]->mask.ruleset == ruleset &&
49 map->rules[i]->mask.type == type &&
50 map->rules[i]->mask.min_size <= size &&
51 map->rules[i]->mask.max_size >= size)
52 return i;
53 }
54 return -1;
55 }
56
57 /*
58 * bucket choose methods
59 *
60 * For each bucket algorithm, we have a "choose" method that, given a
61 * crush input @x and replica position (usually, position in output set) @r,
62 * will produce an item in the bucket.
63 */
64
65 /*
66 * Choose based on a random permutation of the bucket.
67 *
68 * We used to use some prime number arithmetic to do this, but it
69 * wasn't very random, and had some other bad behaviors. Instead, we
70 * calculate an actual random permutation of the bucket members.
71 * Since this is expensive, we optimize for the r=0 case, which
72 * captures the vast majority of calls.
73 */
74 static int bucket_perm_choose(const struct crush_bucket *bucket,
75 struct crush_work_bucket *work,
76 int x, int r)
77 {
78 unsigned int pr = r % bucket->size;
79 unsigned int i, s;
80
81 /* start a new permutation if @x has changed */
82 if (work->perm_x != (__u32)x || work->perm_n == 0) {
83 dprintk("bucket %d new x=%d\n", bucket->id, x);
84 work->perm_x = x;
85
86 /* optimize common r=0 case */
87 if (pr == 0) {
88 s = crush_hash32_3(bucket->hash, x, bucket->id, 0) %
89 bucket->size;
90 work->perm[0] = s;
91 work->perm_n = 0xffff; /* magic value, see below */
92 goto out;
93 }
94
95 for (i = 0; i < bucket->size; i++)
96 work->perm[i] = i;
97 work->perm_n = 0;
98 } else if (work->perm_n == 0xffff) {
99 /* clean up after the r=0 case above */
100 for (i = 1; i < bucket->size; i++)
101 work->perm[i] = i;
102 work->perm[work->perm[0]] = 0;
103 work->perm_n = 1;
104 }
105
106 /* calculate permutation up to pr */
107 for (i = 0; i < work->perm_n; i++)
108 dprintk(" perm_choose have %d: %d\n", i, work->perm[i]);
109 while (work->perm_n <= pr) {
110 unsigned int p = work->perm_n;
111 /* no point in swapping the final entry */
112 if (p < bucket->size - 1) {
113 i = crush_hash32_3(bucket->hash, x, bucket->id, p) %
114 (bucket->size - p);
115 if (i) {
116 unsigned int t = work->perm[p + i];
117 work->perm[p + i] = work->perm[p];
118 work->perm[p] = t;
119 }
120 dprintk(" perm_choose swap %d with %d\n", p, p+i);
121 }
122 work->perm_n++;
123 }
124 for (i = 0; i < bucket->size; i++)
125 dprintk(" perm_choose %d: %d\n", i, work->perm[i]);
126
127 s = work->perm[pr];
128 out:
129 dprintk(" perm_choose %d sz=%d x=%d r=%d (%d) s=%d\n", bucket->id,
130 bucket->size, x, r, pr, s);
131 return bucket->items[s];
132 }
133
134 /* uniform */
135 static int bucket_uniform_choose(const struct crush_bucket_uniform *bucket,
136 struct crush_work_bucket *work, int x, int r)
137 {
138 return bucket_perm_choose(&bucket->h, work, x, r);
139 }
140
141 /* list */
142 static int bucket_list_choose(const struct crush_bucket_list *bucket,
143 int x, int r)
144 {
145 int i;
146
147 for (i = bucket->h.size-1; i >= 0; i--) {
148 __u64 w = crush_hash32_4(bucket->h.hash, x, bucket->h.items[i],
149 r, bucket->h.id);
150 w &= 0xffff;
151 dprintk("list_choose i=%d x=%d r=%d item %d weight %x "
152 "sw %x rand %llx",
153 i, x, r, bucket->h.items[i], bucket->item_weights[i],
154 bucket->sum_weights[i], w);
155 w *= bucket->sum_weights[i];
156 w = w >> 16;
157 /*dprintk(" scaled %llx\n", w);*/
158 if (w < bucket->item_weights[i]) {
159 return bucket->h.items[i];
160 }
161 }
162
163 dprintk("bad list sums for bucket %d\n", bucket->h.id);
164 return bucket->h.items[0];
165 }
166
167
168 /* (binary) tree */
169 static int height(int n)
170 {
171 int h = 0;
172 while ((n & 1) == 0) {
173 h++;
174 n = n >> 1;
175 }
176 return h;
177 }
178
179 static int left(int x)
180 {
181 int h = height(x);
182 return x - (1 << (h-1));
183 }
184
185 static int right(int x)
186 {
187 int h = height(x);
188 return x + (1 << (h-1));
189 }
190
191 static int terminal(int x)
192 {
193 return x & 1;
194 }
195
196 static int bucket_tree_choose(const struct crush_bucket_tree *bucket,
197 int x, int r)
198 {
199 int n;
200 __u32 w;
201 __u64 t;
202
203 /* start at root */
204 n = bucket->num_nodes >> 1;
205
206 while (!terminal(n)) {
207 int l;
208 /* pick point in [0, w) */
209 w = bucket->node_weights[n];
210 t = (__u64)crush_hash32_4(bucket->h.hash, x, n, r,
211 bucket->h.id) * (__u64)w;
212 t = t >> 32;
213
214 /* descend to the left or right? */
215 l = left(n);
216 if (t < bucket->node_weights[l])
217 n = l;
218 else
219 n = right(n);
220 }
221
222 return bucket->h.items[n >> 1];
223 }
224
225
226 /* straw */
227
228 static int bucket_straw_choose(const struct crush_bucket_straw *bucket,
229 int x, int r)
230 {
231 __u32 i;
232 int high = 0;
233 __u64 high_draw = 0;
234 __u64 draw;
235
236 for (i = 0; i < bucket->h.size; i++) {
237 draw = crush_hash32_3(bucket->h.hash, x, bucket->h.items[i], r);
238 draw &= 0xffff;
239 draw *= bucket->straws[i];
240 if (i == 0 || draw > high_draw) {
241 high = i;
242 high_draw = draw;
243 }
244 }
245 return bucket->h.items[high];
246 }
247
248 /* compute 2^44*log2(input+1) */
249 static __u64 crush_ln(unsigned int xin)
250 {
251 unsigned int x = xin;
252 int iexpon, index1, index2;
253 __u64 RH, LH, LL, xl64, result;
254
255 x++;
256
257 /* normalize input */
258 iexpon = 15;
259
260 /*
261 * figure out number of bits we need to shift and
262 * do it in one step instead of iteratively
263 */
264 if (!(x & 0x18000)) {
265 int bits = __builtin_clz(x & 0x1FFFF) - 16;
266 x <<= bits;
267 iexpon = 15 - bits;
268 }
269
270 index1 = (x >> 8) << 1;
271 /* RH ~ 2^56/index1 */
272 RH = __RH_LH_tbl[index1 - 256];
273 /* LH ~ 2^48 * log2(index1/256) */
274 LH = __RH_LH_tbl[index1 + 1 - 256];
275
276 /* RH*x ~ 2^48 * (2^15 + xf), xf<2^8 */
277 xl64 = (__s64)x * RH;
278 xl64 >>= 48;
279
280 result = iexpon;
281 result <<= (12 + 32);
282
283 index2 = xl64 & 0xff;
284 /* LL ~ 2^48*log2(1.0+index2/2^15) */
285 LL = __LL_tbl[index2];
286
287 LH = LH + LL;
288
289 LH >>= (48 - 12 - 32);
290 result += LH;
291
292 return result;
293 }
294
295
296 /*
297 * straw2
298 *
299 * for reference, see:
300 *
301 * https://en.wikipedia.org/wiki/Exponential_distribution#Distribution_of_the_minimum_of_exponential_random_variables
302 *
303 */
304
305 static __u32 *get_choose_arg_weights(const struct crush_bucket_straw2 *bucket,
306 const struct crush_choose_arg *arg,
307 int position)
308 {
309 if (!arg || !arg->weight_set)
310 return bucket->item_weights;
311
312 if (position >= arg->weight_set_size)
313 position = arg->weight_set_size - 1;
314 return arg->weight_set[position].weights;
315 }
316
317 static __s32 *get_choose_arg_ids(const struct crush_bucket_straw2 *bucket,
318 const struct crush_choose_arg *arg)
319 {
320 if (!arg || !arg->ids)
321 return bucket->h.items;
322
323 return arg->ids;
324 }
325
326 static int bucket_straw2_choose(const struct crush_bucket_straw2 *bucket,
327 int x, int r,
328 const struct crush_choose_arg *arg,
329 int position)
330 {
331 unsigned int i, high = 0;
332 unsigned int u;
333 __s64 ln, draw, high_draw = 0;
334 __u32 *weights = get_choose_arg_weights(bucket, arg, position);
335 __s32 *ids = get_choose_arg_ids(bucket, arg);
336
337 for (i = 0; i < bucket->h.size; i++) {
338 dprintk("weight 0x%x item %d\n", weights[i], ids[i]);
339 if (weights[i]) {
340 u = crush_hash32_3(bucket->h.hash, x, ids[i], r);
341 u &= 0xffff;
342
343 /*
344 * for some reason slightly less than 0x10000 produces
345 * a slightly more accurate distribution... probably a
346 * rounding effect.
347 *
348 * the natural log lookup table maps [0,0xffff]
349 * (corresponding to real numbers [1/0x10000, 1] to
350 * [0, 0xffffffffffff] (corresponding to real numbers
351 * [-11.090355,0]).
352 */
353 ln = crush_ln(u) - 0x1000000000000ll;
354
355 /*
356 * divide by 16.16 fixed-point weight. note
357 * that the ln value is negative, so a larger
358 * weight means a larger (less negative) value
359 * for draw.
360 */
361 draw = div64_s64(ln, weights[i]);
362 } else {
363 draw = S64_MIN;
364 }
365
366 if (i == 0 || draw > high_draw) {
367 high = i;
368 high_draw = draw;
369 }
370 }
371
372 return bucket->h.items[high];
373 }
374
375
376 static int crush_bucket_choose(const struct crush_bucket *in,
377 struct crush_work_bucket *work,
378 int x, int r,
379 const struct crush_choose_arg *arg,
380 int position)
381 {
382 dprintk(" crush_bucket_choose %d x=%d r=%d\n", in->id, x, r);
383 BUG_ON(in->size == 0);
384 switch (in->alg) {
385 case CRUSH_BUCKET_UNIFORM:
386 return bucket_uniform_choose(
387 (const struct crush_bucket_uniform *)in,
388 work, x, r);
389 case CRUSH_BUCKET_LIST:
390 return bucket_list_choose((const struct crush_bucket_list *)in,
391 x, r);
392 case CRUSH_BUCKET_TREE:
393 return bucket_tree_choose((const struct crush_bucket_tree *)in,
394 x, r);
395 case CRUSH_BUCKET_STRAW:
396 return bucket_straw_choose(
397 (const struct crush_bucket_straw *)in,
398 x, r);
399 case CRUSH_BUCKET_STRAW2:
400 return bucket_straw2_choose(
401 (const struct crush_bucket_straw2 *)in,
402 x, r, arg, position);
403 default:
404 dprintk("unknown bucket %d alg %d\n", in->id, in->alg);
405 return in->items[0];
406 }
407 }
408
409 /*
410 * true if device is marked "out" (failed, fully offloaded)
411 * of the cluster
412 */
413 static int is_out(const struct crush_map *map,
414 const __u32 *weight, int weight_max,
415 int item, int x)
416 {
417 if (item >= weight_max)
418 return 1;
419 if (weight[item] >= 0x10000)
420 return 0;
421 if (weight[item] == 0)
422 return 1;
423 if ((crush_hash32_2(CRUSH_HASH_RJENKINS1, x, item) & 0xffff)
424 < weight[item])
425 return 0;
426 return 1;
427 }
428
429 /**
430 * crush_choose_firstn - choose numrep distinct items of given type
431 * @map: the crush_map
432 * @work: working space initialized by crush_init_workspace()
433 * @bucket: the bucket we are choose an item from
434 * @weight: weight vector (for map leaves)
435 * @weight_max: size of weight vector
436 * @x: crush input value
437 * @numrep: the number of items to choose
438 * @type: the type of item to choose
439 * @out: pointer to output vector
440 * @outpos: our position in that vector
441 * @out_size: size of the out vector
442 * @tries: number of attempts to make
443 * @recurse_tries: number of attempts to have recursive chooseleaf make
444 * @local_retries: localized retries
445 * @local_fallback_retries: localized fallback retries
446 * @recurse_to_leaf: true if we want one device under each item of given type (chooseleaf instead of choose)
447 * @stable: stable mode starts rep=0 in the recursive call for all replicas
448 * @vary_r: pass r to recursive calls
449 * @out2: second output vector for leaf items (if @recurse_to_leaf)
450 * @parent_r: r value passed from the parent
451 * @choose_args: weights and ids for each known bucket
452 */
453 static int crush_choose_firstn(const struct crush_map *map,
454 struct crush_work *work,
455 const struct crush_bucket *bucket,
456 const __u32 *weight, int weight_max,
457 int x, int numrep, int type,
458 int *out, int outpos,
459 int out_size,
460 unsigned int tries,
461 unsigned int recurse_tries,
462 unsigned int local_retries,
463 unsigned int local_fallback_retries,
464 int recurse_to_leaf,
465 unsigned int vary_r,
466 unsigned int stable,
467 int *out2,
468 int parent_r,
469 const struct crush_choose_arg *choose_args)
470 {
471 int rep;
472 unsigned int ftotal, flocal;
473 int retry_descent, retry_bucket, skip_rep;
474 const struct crush_bucket *in = bucket;
475 int r;
476 int i;
477 int item = 0;
478 int itemtype;
479 int collide, reject;
480 int count = out_size;
481
482 dprintk("CHOOSE%s bucket %d x %d outpos %d numrep %d tries %d recurse_tries %d local_retries %d local_fallback_retries %d parent_r %d stable %d\n",
483 recurse_to_leaf ? "_LEAF" : "",
484 bucket->id, x, outpos, numrep,
485 tries, recurse_tries, local_retries, local_fallback_retries,
486 parent_r, stable);
487
488 for (rep = stable ? 0 : outpos; rep < numrep && count > 0 ; rep++) {
489 /* keep trying until we get a non-out, non-colliding item */
490 ftotal = 0;
491 skip_rep = 0;
492 do {
493 retry_descent = 0;
494 in = bucket; /* initial bucket */
495
496 /* choose through intervening buckets */
497 flocal = 0;
498 do {
499 collide = 0;
500 retry_bucket = 0;
501 r = rep + parent_r;
502 /* r' = r + f_total */
503 r += ftotal;
504
505 /* bucket choose */
506 if (in->size == 0) {
507 reject = 1;
508 goto reject;
509 }
510 if (local_fallback_retries > 0 &&
511 flocal >= (in->size>>1) &&
512 flocal > local_fallback_retries)
513 item = bucket_perm_choose(
514 in, work->work[-1-in->id],
515 x, r);
516 else
517 item = crush_bucket_choose(
518 in, work->work[-1-in->id],
519 x, r,
520 (choose_args ?
521 &choose_args[-1-in->id] : NULL),
522 outpos);
523 if (item >= map->max_devices) {
524 dprintk(" bad item %d\n", item);
525 skip_rep = 1;
526 break;
527 }
528
529 /* desired type? */
530 if (item < 0)
531 itemtype = map->buckets[-1-item]->type;
532 else
533 itemtype = 0;
534 dprintk(" item %d type %d\n", item, itemtype);
535
536 /* keep going? */
537 if (itemtype != type) {
538 if (item >= 0 ||
539 (-1-item) >= map->max_buckets) {
540 dprintk(" bad item type %d\n", type);
541 skip_rep = 1;
542 break;
543 }
544 in = map->buckets[-1-item];
545 retry_bucket = 1;
546 continue;
547 }
548
549 /* collision? */
550 for (i = 0; i < outpos; i++) {
551 if (out[i] == item) {
552 collide = 1;
553 break;
554 }
555 }
556
557 reject = 0;
558 if (!collide && recurse_to_leaf) {
559 if (item < 0) {
560 int sub_r;
561 if (vary_r)
562 sub_r = r >> (vary_r-1);
563 else
564 sub_r = 0;
565 if (crush_choose_firstn(
566 map,
567 work,
568 map->buckets[-1-item],
569 weight, weight_max,
570 x, stable ? 1 : outpos+1, 0,
571 out2, outpos, count,
572 recurse_tries, 0,
573 local_retries,
574 local_fallback_retries,
575 0,
576 vary_r,
577 stable,
578 NULL,
579 sub_r,
580 choose_args) <= outpos)
581 /* didn't get leaf */
582 reject = 1;
583 } else {
584 /* we already have a leaf! */
585 out2[outpos] = item;
586 }
587 }
588
589 if (!reject && !collide) {
590 /* out? */
591 if (itemtype == 0)
592 reject = is_out(map, weight,
593 weight_max,
594 item, x);
595 }
596
597 reject:
598 if (reject || collide) {
599 ftotal++;
600 flocal++;
601
602 if (collide && flocal <= local_retries)
603 /* retry locally a few times */
604 retry_bucket = 1;
605 else if (local_fallback_retries > 0 &&
606 flocal <= in->size + local_fallback_retries)
607 /* exhaustive bucket search */
608 retry_bucket = 1;
609 else if (ftotal < tries)
610 /* then retry descent */
611 retry_descent = 1;
612 else
613 /* else give up */
614 skip_rep = 1;
615 dprintk(" reject %d collide %d "
616 "ftotal %u flocal %u\n",
617 reject, collide, ftotal,
618 flocal);
619 }
620 } while (retry_bucket);
621 } while (retry_descent);
622
623 if (skip_rep) {
624 dprintk("skip rep\n");
625 continue;
626 }
627
628 dprintk("CHOOSE got %d\n", item);
629 out[outpos] = item;
630 outpos++;
631 count--;
632 #ifndef __KERNEL__
633 if (map->choose_tries && ftotal <= map->choose_total_tries)
634 map->choose_tries[ftotal]++;
635 #endif
636 }
637
638 dprintk("CHOOSE returns %d\n", outpos);
639 return outpos;
640 }
641
642
643 /*
644 * crush_choose_indep: alternative breadth-first positionally stable mapping
645 */
646 static void crush_choose_indep(const struct crush_map *map,
647 struct crush_work *work,
648 const struct crush_bucket *bucket,
649 const __u32 *weight, int weight_max,
650 int x, int left, int numrep, int type,
651 int *out, int outpos,
652 unsigned int tries,
653 unsigned int recurse_tries,
654 int recurse_to_leaf,
655 int *out2,
656 int parent_r,
657 const struct crush_choose_arg *choose_args)
658 {
659 const struct crush_bucket *in = bucket;
660 int endpos = outpos + left;
661 int rep;
662 unsigned int ftotal;
663 int r;
664 int i;
665 int item = 0;
666 int itemtype;
667 int collide;
668
669 dprintk("CHOOSE%s INDEP bucket %d x %d outpos %d numrep %d\n", recurse_to_leaf ? "_LEAF" : "",
670 bucket->id, x, outpos, numrep);
671
672 /* initially my result is undefined */
673 for (rep = outpos; rep < endpos; rep++) {
674 out[rep] = CRUSH_ITEM_UNDEF;
675 if (out2)
676 out2[rep] = CRUSH_ITEM_UNDEF;
677 }
678
679 for (ftotal = 0; left > 0 && ftotal < tries; ftotal++) {
680 #ifdef DEBUG_INDEP
681 if (out2 && ftotal) {
682 dprintk("%u %d a: ", ftotal, left);
683 for (rep = outpos; rep < endpos; rep++) {
684 dprintk(" %d", out[rep]);
685 }
686 dprintk("\n");
687 dprintk("%u %d b: ", ftotal, left);
688 for (rep = outpos; rep < endpos; rep++) {
689 dprintk(" %d", out2[rep]);
690 }
691 dprintk("\n");
692 }
693 #endif
694 for (rep = outpos; rep < endpos; rep++) {
695 if (out[rep] != CRUSH_ITEM_UNDEF)
696 continue;
697
698 in = bucket; /* initial bucket */
699
700 /* choose through intervening buckets */
701 for (;;) {
702 /* note: we base the choice on the position
703 * even in the nested call. that means that
704 * if the first layer chooses the same bucket
705 * in a different position, we will tend to
706 * choose a different item in that bucket.
707 * this will involve more devices in data
708 * movement and tend to distribute the load.
709 */
710 r = rep + parent_r;
711
712 /* be careful */
713 if (in->alg == CRUSH_BUCKET_UNIFORM &&
714 in->size % numrep == 0)
715 /* r'=r+(n+1)*f_total */
716 r += (numrep+1) * ftotal;
717 else
718 /* r' = r + n*f_total */
719 r += numrep * ftotal;
720
721 /* bucket choose */
722 if (in->size == 0) {
723 dprintk(" empty bucket\n");
724 break;
725 }
726
727 item = crush_bucket_choose(
728 in, work->work[-1-in->id],
729 x, r,
730 (choose_args ?
731 &choose_args[-1-in->id] : NULL),
732 outpos);
733 if (item >= map->max_devices) {
734 dprintk(" bad item %d\n", item);
735 out[rep] = CRUSH_ITEM_NONE;
736 if (out2)
737 out2[rep] = CRUSH_ITEM_NONE;
738 left--;
739 break;
740 }
741
742 /* desired type? */
743 if (item < 0)
744 itemtype = map->buckets[-1-item]->type;
745 else
746 itemtype = 0;
747 dprintk(" item %d type %d\n", item, itemtype);
748
749 /* keep going? */
750 if (itemtype != type) {
751 if (item >= 0 ||
752 (-1-item) >= map->max_buckets) {
753 dprintk(" bad item type %d\n", type);
754 out[rep] = CRUSH_ITEM_NONE;
755 if (out2)
756 out2[rep] =
757 CRUSH_ITEM_NONE;
758 left--;
759 break;
760 }
761 in = map->buckets[-1-item];
762 continue;
763 }
764
765 /* collision? */
766 collide = 0;
767 for (i = outpos; i < endpos; i++) {
768 if (out[i] == item) {
769 collide = 1;
770 break;
771 }
772 }
773 if (collide)
774 break;
775
776 if (recurse_to_leaf) {
777 if (item < 0) {
778 crush_choose_indep(
779 map,
780 work,
781 map->buckets[-1-item],
782 weight, weight_max,
783 x, 1, numrep, 0,
784 out2, rep,
785 recurse_tries, 0,
786 0, NULL, r,
787 choose_args);
788 if (out2[rep] == CRUSH_ITEM_NONE) {
789 /* placed nothing; no leaf */
790 break;
791 }
792 } else {
793 /* we already have a leaf! */
794 out2[rep] = item;
795 }
796 }
797
798 /* out? */
799 if (itemtype == 0 &&
800 is_out(map, weight, weight_max, item, x))
801 break;
802
803 /* yay! */
804 out[rep] = item;
805 left--;
806 break;
807 }
808 }
809 }
810 for (rep = outpos; rep < endpos; rep++) {
811 if (out[rep] == CRUSH_ITEM_UNDEF) {
812 out[rep] = CRUSH_ITEM_NONE;
813 }
814 if (out2 && out2[rep] == CRUSH_ITEM_UNDEF) {
815 out2[rep] = CRUSH_ITEM_NONE;
816 }
817 }
818 #ifndef __KERNEL__
819 if (map->choose_tries && ftotal <= map->choose_total_tries)
820 map->choose_tries[ftotal]++;
821 #endif
822 #ifdef DEBUG_INDEP
823 if (out2) {
824 dprintk("%u %d a: ", ftotal, left);
825 for (rep = outpos; rep < endpos; rep++) {
826 dprintk(" %d", out[rep]);
827 }
828 dprintk("\n");
829 dprintk("%u %d b: ", ftotal, left);
830 for (rep = outpos; rep < endpos; rep++) {
831 dprintk(" %d", out2[rep]);
832 }
833 dprintk("\n");
834 }
835 #endif
836 }
837
838
839 /*
840 * This takes a chunk of memory and sets it up to be a shiny new
841 * working area for a CRUSH placement computation. It must be called
842 * on any newly allocated memory before passing it in to
843 * crush_do_rule. It may be used repeatedly after that, so long as the
844 * map has not changed. If the map /has/ changed, you must make sure
845 * the working size is no smaller than what was allocated and re-run
846 * crush_init_workspace.
847 *
848 * If you do retain the working space between calls to crush, make it
849 * thread-local.
850 */
851 void crush_init_workspace(const struct crush_map *map, void *v)
852 {
853 struct crush_work *w = v;
854 __s32 b;
855
856 /*
857 * We work by moving through the available space and setting
858 * values and pointers as we go.
859 *
860 * It's a bit like Forth's use of the 'allot' word since we
861 * set the pointer first and then reserve the space for it to
862 * point to by incrementing the point.
863 */
864 v += sizeof(struct crush_work);
865 w->work = v;
866 v += map->max_buckets * sizeof(struct crush_work_bucket *);
867 for (b = 0; b < map->max_buckets; ++b) {
868 if (!map->buckets[b])
869 continue;
870
871 w->work[b] = v;
872 switch (map->buckets[b]->alg) {
873 default:
874 v += sizeof(struct crush_work_bucket);
875 break;
876 }
877 w->work[b]->perm_x = 0;
878 w->work[b]->perm_n = 0;
879 w->work[b]->perm = v;
880 v += map->buckets[b]->size * sizeof(__u32);
881 }
882 BUG_ON(v - (void *)w != map->working_size);
883 }
884
885 /**
886 * crush_do_rule - calculate a mapping with the given input and rule
887 * @map: the crush_map
888 * @ruleno: the rule id
889 * @x: hash input
890 * @result: pointer to result vector
891 * @result_max: maximum result size
892 * @weight: weight vector (for map leaves)
893 * @weight_max: size of weight vector
894 * @cwin: pointer to at least crush_work_size() bytes of memory
895 * @choose_args: weights and ids for each known bucket
896 */
897 int crush_do_rule(const struct crush_map *map,
898 int ruleno, int x, int *result, int result_max,
899 const __u32 *weight, int weight_max,
900 void *cwin, const struct crush_choose_arg *choose_args)
901 {
902 int result_len;
903 struct crush_work *cw = cwin;
904 int *a = cwin + map->working_size;
905 int *b = a + result_max;
906 int *c = b + result_max;
907 int *w = a;
908 int *o = b;
909 int recurse_to_leaf;
910 int wsize = 0;
911 int osize;
912 const struct crush_rule *rule;
913 __u32 step;
914 int i, j;
915 int numrep;
916 int out_size;
917 /*
918 * the original choose_total_tries value was off by one (it
919 * counted "retries" and not "tries"). add one.
920 */
921 int choose_tries = map->choose_total_tries + 1;
922 int choose_leaf_tries = 0;
923 /*
924 * the local tries values were counted as "retries", though,
925 * and need no adjustment
926 */
927 int choose_local_retries = map->choose_local_tries;
928 int choose_local_fallback_retries = map->choose_local_fallback_tries;
929
930 int vary_r = map->chooseleaf_vary_r;
931 int stable = map->chooseleaf_stable;
932
933 if ((__u32)ruleno >= map->max_rules) {
934 dprintk(" bad ruleno %d\n", ruleno);
935 return 0;
936 }
937
938 rule = map->rules[ruleno];
939 result_len = 0;
940
941 for (step = 0; step < rule->len; step++) {
942 int firstn = 0;
943 const struct crush_rule_step *curstep = &rule->steps[step];
944
945 switch (curstep->op) {
946 case CRUSH_RULE_TAKE:
947 if ((curstep->arg1 >= 0 &&
948 curstep->arg1 < map->max_devices) ||
949 (-1-curstep->arg1 >= 0 &&
950 -1-curstep->arg1 < map->max_buckets &&
951 map->buckets[-1-curstep->arg1])) {
952 w[0] = curstep->arg1;
953 wsize = 1;
954 } else {
955 dprintk(" bad take value %d\n", curstep->arg1);
956 }
957 break;
958
959 case CRUSH_RULE_SET_CHOOSE_TRIES:
960 if (curstep->arg1 > 0)
961 choose_tries = curstep->arg1;
962 break;
963
964 case CRUSH_RULE_SET_CHOOSELEAF_TRIES:
965 if (curstep->arg1 > 0)
966 choose_leaf_tries = curstep->arg1;
967 break;
968
969 case CRUSH_RULE_SET_CHOOSE_LOCAL_TRIES:
970 if (curstep->arg1 >= 0)
971 choose_local_retries = curstep->arg1;
972 break;
973
974 case CRUSH_RULE_SET_CHOOSE_LOCAL_FALLBACK_TRIES:
975 if (curstep->arg1 >= 0)
976 choose_local_fallback_retries = curstep->arg1;
977 break;
978
979 case CRUSH_RULE_SET_CHOOSELEAF_VARY_R:
980 if (curstep->arg1 >= 0)
981 vary_r = curstep->arg1;
982 break;
983
984 case CRUSH_RULE_SET_CHOOSELEAF_STABLE:
985 if (curstep->arg1 >= 0)
986 stable = curstep->arg1;
987 break;
988
989 case CRUSH_RULE_CHOOSELEAF_FIRSTN:
990 case CRUSH_RULE_CHOOSE_FIRSTN:
991 firstn = 1;
992 fallthrough;
993 case CRUSH_RULE_CHOOSELEAF_INDEP:
994 case CRUSH_RULE_CHOOSE_INDEP:
995 if (wsize == 0)
996 break;
997
998 recurse_to_leaf =
999 curstep->op ==
1000 CRUSH_RULE_CHOOSELEAF_FIRSTN ||
1001 curstep->op ==
1002 CRUSH_RULE_CHOOSELEAF_INDEP;
1003
1004 /* reset output */
1005 osize = 0;
1006
1007 for (i = 0; i < wsize; i++) {
1008 int bno;
1009 numrep = curstep->arg1;
1010 if (numrep <= 0) {
1011 numrep += result_max;
1012 if (numrep <= 0)
1013 continue;
1014 }
1015 j = 0;
1016 /* make sure bucket id is valid */
1017 bno = -1 - w[i];
1018 if (bno < 0 || bno >= map->max_buckets) {
1019 /* w[i] is probably CRUSH_ITEM_NONE */
1020 dprintk(" bad w[i] %d\n", w[i]);
1021 continue;
1022 }
1023 if (firstn) {
1024 int recurse_tries;
1025 if (choose_leaf_tries)
1026 recurse_tries =
1027 choose_leaf_tries;
1028 else if (map->chooseleaf_descend_once)
1029 recurse_tries = 1;
1030 else
1031 recurse_tries = choose_tries;
1032 osize += crush_choose_firstn(
1033 map,
1034 cw,
1035 map->buckets[bno],
1036 weight, weight_max,
1037 x, numrep,
1038 curstep->arg2,
1039 o+osize, j,
1040 result_max-osize,
1041 choose_tries,
1042 recurse_tries,
1043 choose_local_retries,
1044 choose_local_fallback_retries,
1045 recurse_to_leaf,
1046 vary_r,
1047 stable,
1048 c+osize,
1049 0,
1050 choose_args);
1051 } else {
1052 out_size = ((numrep < (result_max-osize)) ?
1053 numrep : (result_max-osize));
1054 crush_choose_indep(
1055 map,
1056 cw,
1057 map->buckets[bno],
1058 weight, weight_max,
1059 x, out_size, numrep,
1060 curstep->arg2,
1061 o+osize, j,
1062 choose_tries,
1063 choose_leaf_tries ?
1064 choose_leaf_tries : 1,
1065 recurse_to_leaf,
1066 c+osize,
1067 0,
1068 choose_args);
1069 osize += out_size;
1070 }
1071 }
1072
1073 if (recurse_to_leaf)
1074 /* copy final _leaf_ values to output set */
1075 memcpy(o, c, osize*sizeof(*o));
1076
1077 /* swap o and w arrays */
1078 swap(o, w);
1079 wsize = osize;
1080 break;
1081
1082
1083 case CRUSH_RULE_EMIT:
1084 for (i = 0; i < wsize && result_len < result_max; i++) {
1085 result[result_len] = w[i];
1086 result_len++;
1087 }
1088 wsize = 0;
1089 break;
1090
1091 default:
1092 dprintk(" unknown op %d at step %d\n",
1093 curstep->op, step);
1094 break;
1095 }
1096 }
1097
1098 return result_len;
1099 }
1100
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.