]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- |
2 | // vim: ts=8 sw=2 smarttab | |
3 | ||
4 | #include "osd/osd_types.h" | |
5 | #include "common/debug.h" | |
6 | #include "common/Formatter.h" | |
7 | #include "common/errno.h" | |
8 | #include "include/stringify.h" | |
9 | ||
10 | #include "CrushWrapper.h" | |
11 | #include "CrushTreeDumper.h" | |
12 | ||
13 | #define dout_subsys ceph_subsys_crush | |
14 | ||
15 | bool CrushWrapper::has_v2_rules() const | |
16 | { | |
17 | for (unsigned i=0; i<crush->max_rules; i++) { | |
18 | if (is_v2_rule(i)) { | |
19 | return true; | |
20 | } | |
21 | } | |
22 | return false; | |
23 | } | |
24 | ||
25 | bool CrushWrapper::is_v2_rule(unsigned ruleid) const | |
26 | { | |
27 | // check rule for use of indep or new SET_* rule steps | |
28 | if (ruleid >= crush->max_rules) | |
29 | return false; | |
30 | crush_rule *r = crush->rules[ruleid]; | |
31 | if (!r) | |
32 | return false; | |
33 | for (unsigned j=0; j<r->len; j++) { | |
34 | if (r->steps[j].op == CRUSH_RULE_CHOOSE_INDEP || | |
35 | r->steps[j].op == CRUSH_RULE_CHOOSELEAF_INDEP || | |
36 | r->steps[j].op == CRUSH_RULE_SET_CHOOSE_TRIES || | |
37 | r->steps[j].op == CRUSH_RULE_SET_CHOOSELEAF_TRIES) { | |
38 | return true; | |
39 | } | |
40 | } | |
41 | return false; | |
42 | } | |
43 | ||
44 | bool CrushWrapper::has_v3_rules() const | |
45 | { | |
46 | for (unsigned i=0; i<crush->max_rules; i++) { | |
47 | if (is_v3_rule(i)) { | |
48 | return true; | |
49 | } | |
50 | } | |
51 | return false; | |
52 | } | |
53 | ||
54 | bool CrushWrapper::is_v3_rule(unsigned ruleid) const | |
55 | { | |
56 | // check rule for use of SET_CHOOSELEAF_VARY_R step | |
57 | if (ruleid >= crush->max_rules) | |
58 | return false; | |
59 | crush_rule *r = crush->rules[ruleid]; | |
60 | if (!r) | |
61 | return false; | |
62 | for (unsigned j=0; j<r->len; j++) { | |
63 | if (r->steps[j].op == CRUSH_RULE_SET_CHOOSELEAF_VARY_R) { | |
64 | return true; | |
65 | } | |
66 | } | |
67 | return false; | |
68 | } | |
69 | ||
70 | bool CrushWrapper::has_v4_buckets() const | |
71 | { | |
72 | for (int i=0; i<crush->max_buckets; ++i) { | |
73 | crush_bucket *b = crush->buckets[i]; | |
74 | if (!b) | |
75 | continue; | |
76 | if (b->alg == CRUSH_BUCKET_STRAW2) | |
77 | return true; | |
78 | } | |
79 | return false; | |
80 | } | |
81 | ||
82 | bool CrushWrapper::has_v5_rules() const | |
83 | { | |
84 | for (unsigned i=0; i<crush->max_rules; i++) { | |
85 | if (is_v5_rule(i)) { | |
86 | return true; | |
87 | } | |
88 | } | |
89 | return false; | |
90 | } | |
91 | ||
92 | bool CrushWrapper::is_v5_rule(unsigned ruleid) const | |
93 | { | |
94 | // check rule for use of SET_CHOOSELEAF_STABLE step | |
95 | if (ruleid >= crush->max_rules) | |
96 | return false; | |
97 | crush_rule *r = crush->rules[ruleid]; | |
98 | if (!r) | |
99 | return false; | |
100 | for (unsigned j=0; j<r->len; j++) { | |
101 | if (r->steps[j].op == CRUSH_RULE_SET_CHOOSELEAF_STABLE) { | |
102 | return true; | |
103 | } | |
104 | } | |
105 | return false; | |
106 | } | |
107 | ||
108 | bool CrushWrapper::has_chooseargs() const | |
109 | { | |
110 | return !choose_args.empty(); | |
111 | } | |
112 | ||
113 | bool CrushWrapper::has_incompat_chooseargs() const | |
114 | { | |
115 | // FIXME: if the chooseargs all have 1 position *and* do not remap IDs then | |
116 | // we can fabricate a compatible crush map for legacy clients by swapping the | |
117 | // choose_args weights in for the real weights. until then, | |
118 | return has_chooseargs(); | |
119 | } | |
120 | ||
121 | int CrushWrapper::split_id_class(int i, int *idout, int *classout) const | |
122 | { | |
123 | if (!item_exists(i)) | |
124 | return -EINVAL; | |
125 | string name = get_item_name(i); | |
126 | size_t pos = name.find("~"); | |
127 | if (pos == string::npos) { | |
128 | *idout = i; | |
129 | *classout = -1; | |
130 | return 0; | |
131 | } | |
132 | string name_no_class = name.substr(0, pos); | |
133 | if (!name_exists(name_no_class)) | |
134 | return -ENOENT; | |
135 | string class_name = name.substr(pos + 1); | |
136 | if (!class_exists(class_name)) | |
137 | return -ENOENT; | |
138 | *idout = get_item_id(name_no_class); | |
139 | *classout = get_class_id(class_name); | |
140 | return 0; | |
141 | } | |
142 | ||
143 | int CrushWrapper::can_rename_item(const string& srcname, | |
144 | const string& dstname, | |
145 | ostream *ss) const | |
146 | { | |
147 | if (name_exists(srcname)) { | |
148 | if (name_exists(dstname)) { | |
149 | *ss << "dstname = '" << dstname << "' already exists"; | |
150 | return -EEXIST; | |
151 | } | |
152 | if (is_valid_crush_name(dstname)) { | |
153 | return 0; | |
154 | } else { | |
155 | *ss << "dstname = '" << dstname << "' does not match [-_.0-9a-zA-Z]+"; | |
156 | return -EINVAL; | |
157 | } | |
158 | } else { | |
159 | if (name_exists(dstname)) { | |
160 | *ss << "srcname = '" << srcname << "' does not exist " | |
161 | << "and dstname = '" << dstname << "' already exists"; | |
162 | return -EALREADY; | |
163 | } else { | |
164 | *ss << "srcname = '" << srcname << "' does not exist"; | |
165 | return -ENOENT; | |
166 | } | |
167 | } | |
168 | } | |
169 | ||
170 | int CrushWrapper::rename_item(const string& srcname, | |
171 | const string& dstname, | |
172 | ostream *ss) | |
173 | { | |
174 | int ret = can_rename_item(srcname, dstname, ss); | |
175 | if (ret < 0) | |
176 | return ret; | |
177 | int oldid = get_item_id(srcname); | |
178 | return set_item_name(oldid, dstname); | |
179 | } | |
180 | ||
181 | int CrushWrapper::can_rename_bucket(const string& srcname, | |
182 | const string& dstname, | |
183 | ostream *ss) const | |
184 | { | |
185 | int ret = can_rename_item(srcname, dstname, ss); | |
186 | if (ret) | |
187 | return ret; | |
188 | int srcid = get_item_id(srcname); | |
189 | if (srcid >= 0) { | |
190 | *ss << "srcname = '" << srcname << "' is not a bucket " | |
191 | << "because its id = " << srcid << " is >= 0"; | |
192 | return -ENOTDIR; | |
193 | } | |
194 | return 0; | |
195 | } | |
196 | ||
197 | int CrushWrapper::rename_bucket(const string& srcname, | |
198 | const string& dstname, | |
199 | ostream *ss) | |
200 | { | |
201 | int ret = can_rename_bucket(srcname, dstname, ss); | |
202 | if (ret < 0) | |
203 | return ret; | |
204 | int oldid = get_item_id(srcname); | |
205 | return set_item_name(oldid, dstname); | |
206 | } | |
207 | ||
208 | void CrushWrapper::find_takes(set<int>& roots) const | |
209 | { | |
210 | for (unsigned i=0; i<crush->max_rules; i++) { | |
211 | crush_rule *r = crush->rules[i]; | |
212 | if (!r) | |
213 | continue; | |
214 | for (unsigned j=0; j<r->len; j++) { | |
215 | if (r->steps[j].op == CRUSH_RULE_TAKE) | |
216 | roots.insert(r->steps[j].arg1); | |
217 | } | |
218 | } | |
219 | } | |
220 | ||
221 | void CrushWrapper::find_roots(set<int>& roots) const | |
222 | { | |
223 | for (int i = 0; i < crush->max_buckets; i++) { | |
224 | if (!crush->buckets[i]) | |
225 | continue; | |
226 | crush_bucket *b = crush->buckets[i]; | |
227 | if (!_search_item_exists(b->id)) | |
228 | roots.insert(b->id); | |
229 | } | |
230 | } | |
231 | ||
232 | bool CrushWrapper::subtree_contains(int root, int item) const | |
233 | { | |
234 | if (root == item) | |
235 | return true; | |
236 | ||
237 | if (root >= 0) | |
238 | return false; // root is a leaf | |
239 | ||
240 | const crush_bucket *b = get_bucket(root); | |
241 | if (IS_ERR(b)) | |
242 | return false; | |
243 | ||
244 | for (unsigned j=0; j<b->size; j++) { | |
245 | if (subtree_contains(b->items[j], item)) | |
246 | return true; | |
247 | } | |
248 | return false; | |
249 | } | |
250 | ||
251 | bool CrushWrapper::_maybe_remove_last_instance(CephContext *cct, int item, bool unlink_only) | |
252 | { | |
253 | // last instance? | |
254 | if (_search_item_exists(item)) { | |
255 | return false; | |
256 | } | |
257 | if (item < 0 && _bucket_is_in_use(item)) { | |
258 | return false; | |
259 | } | |
260 | ||
261 | if (item < 0 && !unlink_only) { | |
262 | crush_bucket *t = get_bucket(item); | |
263 | ldout(cct, 5) << "_maybe_remove_last_instance removing bucket " << item << dendl; | |
264 | crush_remove_bucket(crush, t); | |
265 | if (class_bucket.count(item) != 0) | |
266 | class_bucket.erase(item); | |
267 | } | |
268 | if ((item >= 0 || !unlink_only) && name_map.count(item)) { | |
269 | ldout(cct, 5) << "_maybe_remove_last_instance removing name for item " << item << dendl; | |
270 | name_map.erase(item); | |
271 | have_rmaps = false; | |
272 | } | |
273 | return true; | |
274 | } | |
275 | ||
276 | int CrushWrapper::remove_root(int item, bool unused) | |
277 | { | |
278 | if (unused && _bucket_is_in_use(item)) | |
279 | return 0; | |
280 | ||
281 | crush_bucket *b = get_bucket(item); | |
282 | if (IS_ERR(b)) | |
283 | return -ENOENT; | |
284 | ||
285 | for (unsigned n = 0; n < b->size; n++) { | |
286 | if (b->items[n] >= 0) | |
287 | continue; | |
288 | int r = remove_root(b->items[n], unused); | |
289 | if (r < 0) | |
290 | return r; | |
291 | } | |
292 | ||
293 | crush_remove_bucket(crush, b); | |
294 | if (name_map.count(item) != 0) { | |
295 | name_map.erase(item); | |
296 | have_rmaps = false; | |
297 | } | |
298 | if (class_bucket.count(item) != 0) | |
299 | class_bucket.erase(item); | |
300 | return 0; | |
301 | } | |
302 | ||
303 | int CrushWrapper::remove_item(CephContext *cct, int item, bool unlink_only) | |
304 | { | |
305 | if (choose_args.size() > 0) { | |
306 | ldout(cct, 1) << "remove_item not implemented when choose_args is not empty" << dendl; | |
307 | return -EDOM; | |
308 | } | |
309 | ||
310 | ldout(cct, 5) << "remove_item " << item << (unlink_only ? " unlink_only":"") << dendl; | |
311 | ||
312 | int ret = -ENOENT; | |
313 | ||
314 | if (item < 0 && !unlink_only) { | |
315 | crush_bucket *t = get_bucket(item); | |
316 | if (IS_ERR(t)) { | |
317 | ldout(cct, 1) << "remove_item bucket " << item << " does not exist" << dendl; | |
318 | return -ENOENT; | |
319 | } | |
320 | ||
321 | if (t->size) { | |
322 | ldout(cct, 1) << "remove_item bucket " << item << " has " << t->size | |
323 | << " items, not empty" << dendl; | |
324 | return -ENOTEMPTY; | |
325 | } | |
326 | if (_bucket_is_in_use(item)) { | |
327 | return -EBUSY; | |
328 | } | |
329 | } | |
330 | ||
331 | for (int i = 0; i < crush->max_buckets; i++) { | |
332 | if (!crush->buckets[i]) | |
333 | continue; | |
334 | crush_bucket *b = crush->buckets[i]; | |
335 | ||
336 | for (unsigned i=0; i<b->size; ++i) { | |
337 | int id = b->items[i]; | |
338 | if (id == item) { | |
339 | ldout(cct, 5) << "remove_item removing item " << item | |
340 | << " from bucket " << b->id << dendl; | |
341 | crush_bucket_remove_item(crush, b, item); | |
342 | adjust_item_weight(cct, b->id, b->weight); | |
343 | ret = 0; | |
344 | } | |
345 | } | |
346 | } | |
347 | ||
348 | if (_maybe_remove_last_instance(cct, item, unlink_only)) | |
349 | ret = 0; | |
350 | ||
351 | return ret; | |
352 | } | |
353 | ||
354 | bool CrushWrapper::_search_item_exists(int item) const | |
355 | { | |
356 | for (int i = 0; i < crush->max_buckets; i++) { | |
357 | if (!crush->buckets[i]) | |
358 | continue; | |
359 | crush_bucket *b = crush->buckets[i]; | |
360 | for (unsigned j=0; j<b->size; ++j) { | |
361 | if (b->items[j] == item) | |
362 | return true; | |
363 | } | |
364 | } | |
365 | return false; | |
366 | } | |
367 | ||
368 | bool CrushWrapper::_bucket_is_in_use(int item) | |
369 | { | |
370 | for (auto &i : class_bucket) | |
371 | for (auto &j : i.second) | |
372 | if (j.second == item) | |
373 | return true; | |
374 | for (unsigned i = 0; i < crush->max_rules; ++i) { | |
375 | crush_rule *r = crush->rules[i]; | |
376 | if (!r) | |
377 | continue; | |
378 | for (unsigned j = 0; j < r->len; ++j) { | |
379 | if (r->steps[j].op == CRUSH_RULE_TAKE) { | |
380 | int step_item = r->steps[j].arg1; | |
381 | int original_item; | |
382 | int c; | |
383 | int res = split_id_class(step_item, &original_item, &c); | |
384 | if (res < 0) | |
385 | return false; | |
386 | if (step_item == item || original_item == item) | |
387 | return true; | |
388 | } | |
389 | } | |
390 | } | |
391 | return false; | |
392 | } | |
393 | ||
394 | int CrushWrapper::_remove_item_under(CephContext *cct, int item, int ancestor, bool unlink_only) | |
395 | { | |
396 | ldout(cct, 5) << "_remove_item_under " << item << " under " << ancestor | |
397 | << (unlink_only ? " unlink_only":"") << dendl; | |
398 | ||
399 | if (ancestor >= 0) { | |
400 | return -EINVAL; | |
401 | } | |
402 | ||
403 | if (!bucket_exists(ancestor)) | |
404 | return -EINVAL; | |
405 | ||
406 | int ret = -ENOENT; | |
407 | ||
408 | crush_bucket *b = get_bucket(ancestor); | |
409 | for (unsigned i=0; i<b->size; ++i) { | |
410 | int id = b->items[i]; | |
411 | if (id == item) { | |
412 | ldout(cct, 5) << "_remove_item_under removing item " << item << " from bucket " << b->id << dendl; | |
413 | crush_bucket_remove_item(crush, b, item); | |
414 | adjust_item_weight(cct, b->id, b->weight); | |
415 | ret = 0; | |
416 | } else if (id < 0) { | |
417 | int r = remove_item_under(cct, item, id, unlink_only); | |
418 | if (r == 0) | |
419 | ret = 0; | |
420 | } | |
421 | } | |
422 | return ret; | |
423 | } | |
424 | ||
425 | int CrushWrapper::remove_item_under(CephContext *cct, int item, int ancestor, bool unlink_only) | |
426 | { | |
427 | ldout(cct, 5) << "remove_item_under " << item << " under " << ancestor | |
428 | << (unlink_only ? " unlink_only":"") << dendl; | |
429 | ||
430 | if (!unlink_only && _bucket_is_in_use(item)) { | |
431 | return -EBUSY; | |
432 | } | |
433 | ||
434 | int ret = _remove_item_under(cct, item, ancestor, unlink_only); | |
435 | if (ret < 0) | |
436 | return ret; | |
437 | ||
438 | if (item < 0 && !unlink_only) { | |
439 | crush_bucket *t = get_bucket(item); | |
440 | if (IS_ERR(t)) { | |
441 | ldout(cct, 1) << "remove_item_under bucket " << item | |
442 | << " does not exist" << dendl; | |
443 | return -ENOENT; | |
444 | } | |
445 | ||
446 | if (t->size) { | |
447 | ldout(cct, 1) << "remove_item_under bucket " << item << " has " << t->size | |
448 | << " items, not empty" << dendl; | |
449 | return -ENOTEMPTY; | |
450 | } | |
451 | } | |
452 | ||
453 | if (_maybe_remove_last_instance(cct, item, unlink_only)) | |
454 | ret = 0; | |
455 | ||
456 | return ret; | |
457 | } | |
458 | ||
459 | int CrushWrapper::get_common_ancestor_distance(CephContext *cct, int id, | |
460 | const std::multimap<string,string>& loc) | |
461 | { | |
462 | ldout(cct, 5) << __func__ << " " << id << " " << loc << dendl; | |
463 | if (!item_exists(id)) | |
464 | return -ENOENT; | |
465 | map<string,string> id_loc = get_full_location(id); | |
466 | ldout(cct, 20) << " id is at " << id_loc << dendl; | |
467 | ||
468 | for (map<int,string>::const_iterator p = type_map.begin(); | |
469 | p != type_map.end(); | |
470 | ++p) { | |
471 | map<string,string>::iterator ip = id_loc.find(p->second); | |
472 | if (ip == id_loc.end()) | |
473 | continue; | |
474 | for (std::multimap<string,string>::const_iterator q = loc.find(p->second); | |
475 | q != loc.end(); | |
476 | ++q) { | |
477 | if (q->first != p->second) | |
478 | break; | |
479 | if (q->second == ip->second) | |
480 | return p->first; | |
481 | } | |
482 | } | |
483 | return -ERANGE; | |
484 | } | |
485 | ||
486 | int CrushWrapper::parse_loc_map(const std::vector<string>& args, | |
487 | std::map<string,string> *ploc) | |
488 | { | |
489 | ploc->clear(); | |
490 | for (unsigned i = 0; i < args.size(); ++i) { | |
491 | const char *s = args[i].c_str(); | |
492 | const char *pos = strchr(s, '='); | |
493 | if (!pos) | |
494 | return -EINVAL; | |
495 | string key(s, 0, pos-s); | |
496 | string value(pos+1); | |
497 | if (value.length()) | |
498 | (*ploc)[key] = value; | |
499 | else | |
500 | return -EINVAL; | |
501 | } | |
502 | return 0; | |
503 | } | |
504 | ||
505 | int CrushWrapper::parse_loc_multimap(const std::vector<string>& args, | |
506 | std::multimap<string,string> *ploc) | |
507 | { | |
508 | ploc->clear(); | |
509 | for (unsigned i = 0; i < args.size(); ++i) { | |
510 | const char *s = args[i].c_str(); | |
511 | const char *pos = strchr(s, '='); | |
512 | if (!pos) | |
513 | return -EINVAL; | |
514 | string key(s, 0, pos-s); | |
515 | string value(pos+1); | |
516 | if (value.length()) | |
517 | ploc->insert(make_pair(key, value)); | |
518 | else | |
519 | return -EINVAL; | |
520 | } | |
521 | return 0; | |
522 | } | |
523 | ||
524 | bool CrushWrapper::check_item_loc(CephContext *cct, int item, const map<string,string>& loc, | |
525 | int *weight) | |
526 | { | |
527 | ldout(cct, 5) << "check_item_loc item " << item << " loc " << loc << dendl; | |
528 | ||
529 | for (map<int,string>::const_iterator p = type_map.begin(); p != type_map.end(); ++p) { | |
530 | // ignore device | |
531 | if (p->first == 0) | |
532 | continue; | |
533 | ||
534 | // ignore types that aren't specified in loc | |
535 | map<string,string>::const_iterator q = loc.find(p->second); | |
536 | if (q == loc.end()) { | |
537 | ldout(cct, 2) << "warning: did not specify location for '" << p->second << "' level (levels are " | |
538 | << type_map << ")" << dendl; | |
539 | continue; | |
540 | } | |
541 | ||
542 | if (!name_exists(q->second)) { | |
543 | ldout(cct, 5) << "check_item_loc bucket " << q->second << " dne" << dendl; | |
544 | return false; | |
545 | } | |
546 | ||
547 | int id = get_item_id(q->second); | |
548 | if (id >= 0) { | |
549 | ldout(cct, 5) << "check_item_loc requested " << q->second << " for type " << p->second | |
550 | << " is a device, not bucket" << dendl; | |
551 | return false; | |
552 | } | |
553 | ||
554 | assert(bucket_exists(id)); | |
555 | crush_bucket *b = get_bucket(id); | |
556 | ||
557 | // see if item exists in this bucket | |
558 | for (unsigned j=0; j<b->size; j++) { | |
559 | if (b->items[j] == item) { | |
560 | ldout(cct, 2) << "check_item_loc " << item << " exists in bucket " << b->id << dendl; | |
561 | if (weight) | |
562 | *weight = crush_get_bucket_item_weight(b, j); | |
563 | return true; | |
564 | } | |
565 | } | |
566 | return false; | |
567 | } | |
568 | ||
569 | ldout(cct, 1) << "check_item_loc item " << item << " loc " << loc << dendl; | |
570 | return false; | |
571 | } | |
572 | ||
573 | map<string, string> CrushWrapper::get_full_location(int id) | |
574 | { | |
575 | vector<pair<string, string> > full_location_ordered; | |
576 | map<string,string> full_location; | |
577 | ||
578 | get_full_location_ordered(id, full_location_ordered); | |
579 | ||
580 | std::copy(full_location_ordered.begin(), | |
581 | full_location_ordered.end(), | |
582 | std::inserter(full_location, full_location.begin())); | |
583 | ||
584 | return full_location; | |
585 | } | |
586 | ||
587 | int CrushWrapper::get_full_location_ordered(int id, vector<pair<string, string> >& path) | |
588 | { | |
589 | if (!item_exists(id)) | |
590 | return -ENOENT; | |
591 | int cur = id; | |
592 | int ret; | |
593 | while (true) { | |
594 | pair<string, string> parent_coord = get_immediate_parent(cur, &ret); | |
595 | if (ret != 0) | |
596 | break; | |
597 | path.push_back(parent_coord); | |
598 | cur = get_item_id(parent_coord.second); | |
599 | } | |
600 | return 0; | |
601 | } | |
602 | ||
603 | ||
604 | map<int, string> CrushWrapper::get_parent_hierarchy(int id) | |
605 | { | |
606 | map<int,string> parent_hierarchy; | |
607 | pair<string, string> parent_coord = get_immediate_parent(id); | |
608 | int parent_id; | |
609 | ||
610 | // get the integer type for id and create a counter from there | |
611 | int type_counter = get_bucket_type(id); | |
612 | ||
613 | // if we get a negative type then we can assume that we have an OSD | |
614 | // change behavior in get_item_type FIXME | |
615 | if (type_counter < 0) | |
616 | type_counter = 0; | |
617 | ||
618 | // read the type map and get the name of the type with the largest ID | |
619 | int high_type = 0; | |
620 | for (map<int, string>::iterator it = type_map.begin(); it != type_map.end(); ++it){ | |
621 | if ( (*it).first > high_type ) | |
622 | high_type = (*it).first; | |
623 | } | |
624 | ||
625 | parent_id = get_item_id(parent_coord.second); | |
626 | ||
627 | while (type_counter < high_type) { | |
628 | type_counter++; | |
629 | parent_hierarchy[ type_counter ] = parent_coord.first; | |
630 | ||
631 | if (type_counter < high_type){ | |
632 | // get the coordinate information for the next parent | |
633 | parent_coord = get_immediate_parent(parent_id); | |
634 | parent_id = get_item_id(parent_coord.second); | |
635 | } | |
636 | } | |
637 | ||
638 | return parent_hierarchy; | |
639 | } | |
640 | ||
641 | int CrushWrapper::get_children(int id, list<int> *children) | |
642 | { | |
643 | // leaf? | |
644 | if (id >= 0) { | |
645 | return 0; | |
646 | } | |
647 | ||
648 | crush_bucket *b = get_bucket(id); | |
649 | if (IS_ERR(b)) { | |
650 | return -ENOENT; | |
651 | } | |
652 | ||
653 | for (unsigned n=0; n<b->size; n++) { | |
654 | children->push_back(b->items[n]); | |
655 | } | |
656 | return b->size; | |
657 | } | |
658 | ||
659 | ||
660 | int CrushWrapper::insert_item(CephContext *cct, int item, float weight, string name, | |
661 | const map<string,string>& loc) // typename -> bucketname | |
662 | { | |
663 | if (choose_args.size() > 0) { | |
664 | ldout(cct, 1) << "insert_item not implemented when choose_args is not empty" << dendl; | |
665 | return -EDOM; | |
666 | } | |
667 | ||
668 | ldout(cct, 5) << "insert_item item " << item << " weight " << weight | |
669 | << " name " << name << " loc " << loc << dendl; | |
670 | ||
671 | if (!is_valid_crush_name(name)) | |
672 | return -EINVAL; | |
673 | ||
674 | if (!is_valid_crush_loc(cct, loc)) | |
675 | return -EINVAL; | |
676 | ||
677 | if (name_exists(name)) { | |
678 | if (get_item_id(name) != item) { | |
679 | ldout(cct, 10) << "device name '" << name << "' already exists as id " | |
680 | << get_item_id(name) << dendl; | |
681 | return -EEXIST; | |
682 | } | |
683 | } else { | |
684 | set_item_name(item, name); | |
685 | } | |
686 | ||
687 | int cur = item; | |
688 | ||
689 | // create locations if locations don't exist and add child in location with 0 weight | |
690 | // the more detail in the insert_item method declaration in CrushWrapper.h | |
691 | for (map<int,string>::iterator p = type_map.begin(); p != type_map.end(); ++p) { | |
692 | // ignore device type | |
693 | if (p->first == 0) | |
694 | continue; | |
695 | ||
696 | // skip types that are unspecified | |
697 | map<string,string>::const_iterator q = loc.find(p->second); | |
698 | if (q == loc.end()) { | |
699 | ldout(cct, 2) << "warning: did not specify location for '" << p->second << "' level (levels are " | |
700 | << type_map << ")" << dendl; | |
701 | continue; | |
702 | } | |
703 | ||
704 | if (!name_exists(q->second)) { | |
705 | ldout(cct, 5) << "insert_item creating bucket " << q->second << dendl; | |
706 | int empty = 0, newid; | |
707 | int r = add_bucket(0, 0, | |
708 | CRUSH_HASH_DEFAULT, p->first, 1, &cur, &empty, &newid); | |
709 | if (r < 0) { | |
710 | ldout(cct, 1) << "add_bucket failure error: " << cpp_strerror(r) << dendl; | |
711 | return r; | |
712 | } | |
713 | set_item_name(newid, q->second); | |
714 | ||
715 | cur = newid; | |
716 | continue; | |
717 | } | |
718 | ||
719 | // add to an existing bucket | |
720 | int id = get_item_id(q->second); | |
721 | if (!bucket_exists(id)) { | |
722 | ldout(cct, 1) << "insert_item doesn't have bucket " << id << dendl; | |
723 | return -EINVAL; | |
724 | } | |
725 | ||
726 | // check that we aren't creating a cycle. | |
727 | if (subtree_contains(id, cur)) { | |
728 | ldout(cct, 1) << "insert_item item " << cur << " already exists beneath " << id << dendl; | |
729 | return -EINVAL; | |
730 | } | |
731 | ||
732 | // we have done sanity check above | |
733 | crush_bucket *b = get_bucket(id); | |
734 | ||
735 | if (p->first != b->type) { | |
736 | ldout(cct, 1) << "insert_item existing bucket has type " | |
737 | << "'" << type_map[b->type] << "' != " | |
738 | << "'" << type_map[p->first] << "'" << dendl; | |
739 | return -EINVAL; | |
740 | } | |
741 | ||
742 | // are we forming a loop? | |
743 | if (subtree_contains(cur, b->id)) { | |
744 | ldout(cct, 1) << "insert_item " << cur << " already contains " << b->id | |
745 | << "; cannot form loop" << dendl; | |
746 | return -ELOOP; | |
747 | } | |
748 | ||
749 | ldout(cct, 5) << "insert_item adding " << cur << " weight " << weight | |
750 | << " to bucket " << id << dendl; | |
751 | int r = crush_bucket_add_item(crush, b, cur, 0); | |
752 | assert (!r); | |
753 | break; | |
754 | } | |
755 | ||
756 | // adjust the item's weight in location | |
757 | if(adjust_item_weightf_in_loc(cct, item, weight, loc) > 0) { | |
758 | if (item >= crush->max_devices) { | |
759 | crush->max_devices = item + 1; | |
760 | ldout(cct, 5) << "insert_item max_devices now " << crush->max_devices << dendl; | |
761 | } | |
762 | return 0; | |
763 | } | |
764 | ||
765 | ldout(cct, 1) << "error: didn't find anywhere to add item " << item << " in " << loc << dendl; | |
766 | return -EINVAL; | |
767 | } | |
768 | ||
769 | int CrushWrapper::move_bucket(CephContext *cct, int id, const map<string,string>& loc) | |
770 | { | |
771 | if (choose_args.size() > 0) { | |
772 | ldout(cct, 1) << "move_bucket not implemented when choose_args is not empty" << dendl; | |
773 | return -EDOM; | |
774 | } | |
775 | ||
776 | // sorry this only works for buckets | |
777 | if (id >= 0) | |
778 | return -EINVAL; | |
779 | ||
780 | if (!item_exists(id)) | |
781 | return -ENOENT; | |
782 | ||
783 | // get the name of the bucket we are trying to move for later | |
784 | string id_name = get_item_name(id); | |
785 | ||
786 | // detach the bucket | |
787 | int bucket_weight = detach_bucket(cct, id); | |
788 | ||
789 | // insert the bucket back into the hierarchy | |
790 | return insert_item(cct, id, bucket_weight / (float)0x10000, id_name, loc); | |
791 | } | |
792 | ||
793 | int CrushWrapper::link_bucket(CephContext *cct, int id, const map<string,string>& loc) | |
794 | { | |
795 | if (choose_args.size() > 0) { | |
796 | ldout(cct, 1) << "link_bucket not implemented when choose_args is not empty" << dendl; | |
797 | return -EDOM; | |
798 | } | |
799 | ||
800 | // sorry this only works for buckets | |
801 | if (id >= 0) | |
802 | return -EINVAL; | |
803 | ||
804 | if (!item_exists(id)) | |
805 | return -ENOENT; | |
806 | ||
807 | // get the name of the bucket we are trying to move for later | |
808 | string id_name = get_item_name(id); | |
809 | ||
810 | crush_bucket *b = get_bucket(id); | |
811 | unsigned bucket_weight = b->weight; | |
812 | ||
813 | return insert_item(cct, id, bucket_weight / (float)0x10000, id_name, loc); | |
814 | } | |
815 | ||
816 | int CrushWrapper::create_or_move_item(CephContext *cct, int item, float weight, string name, | |
817 | const map<string,string>& loc) // typename -> bucketname | |
818 | { | |
819 | if (choose_args.size() > 0) { | |
820 | ldout(cct, 1) << "create_or_move_item not implemented when choose_args is not empty" << dendl; | |
821 | return -EDOM; | |
822 | } | |
823 | ||
824 | int ret = 0; | |
825 | int old_iweight; | |
826 | ||
827 | if (!is_valid_crush_name(name)) | |
828 | return -EINVAL; | |
829 | ||
830 | if (check_item_loc(cct, item, loc, &old_iweight)) { | |
831 | ldout(cct, 5) << "create_or_move_item " << item << " already at " << loc << dendl; | |
832 | } else { | |
833 | if (_search_item_exists(item)) { | |
834 | weight = get_item_weightf(item); | |
835 | ldout(cct, 10) << "create_or_move_item " << item << " exists with weight " << weight << dendl; | |
836 | remove_item(cct, item, true); | |
837 | } | |
838 | ldout(cct, 5) << "create_or_move_item adding " << item << " weight " << weight | |
839 | << " at " << loc << dendl; | |
840 | ret = insert_item(cct, item, weight, name, loc); | |
841 | if (ret == 0) | |
842 | ret = 1; // changed | |
843 | } | |
844 | return ret; | |
845 | } | |
846 | ||
847 | int CrushWrapper::update_item(CephContext *cct, int item, float weight, string name, | |
848 | const map<string,string>& loc) // typename -> bucketname | |
849 | { | |
850 | if (choose_args.size() > 0) { | |
851 | ldout(cct, 1) << "update_item not implemented when choose_args is not empty" << dendl; | |
852 | return -EDOM; | |
853 | } | |
854 | ||
855 | ldout(cct, 5) << "update_item item " << item << " weight " << weight | |
856 | << " name " << name << " loc " << loc << dendl; | |
857 | int ret = 0; | |
858 | ||
859 | if (!is_valid_crush_name(name)) | |
860 | return -EINVAL; | |
861 | ||
862 | if (!is_valid_crush_loc(cct, loc)) | |
863 | return -EINVAL; | |
864 | ||
865 | // compare quantized (fixed-point integer) weights! | |
866 | int iweight = (int)(weight * (float)0x10000); | |
867 | int old_iweight; | |
868 | if (check_item_loc(cct, item, loc, &old_iweight)) { | |
869 | ldout(cct, 5) << "update_item " << item << " already at " << loc << dendl; | |
870 | if (old_iweight != iweight) { | |
871 | ldout(cct, 5) << "update_item " << item << " adjusting weight " | |
872 | << ((float)old_iweight/(float)0x10000) << " -> " << weight << dendl; | |
873 | adjust_item_weight_in_loc(cct, item, iweight, loc); | |
874 | ret = 1; | |
875 | } | |
876 | if (get_item_name(item) != name) { | |
877 | ldout(cct, 5) << "update_item setting " << item << " name to " << name << dendl; | |
878 | set_item_name(item, name); | |
879 | ret = 1; | |
880 | } | |
881 | } else { | |
882 | if (item_exists(item)) { | |
883 | remove_item(cct, item, true); | |
884 | } | |
885 | ldout(cct, 5) << "update_item adding " << item << " weight " << weight | |
886 | << " at " << loc << dendl; | |
887 | ret = insert_item(cct, item, weight, name, loc); | |
888 | if (ret == 0) | |
889 | ret = 1; // changed | |
890 | } | |
891 | return ret; | |
892 | } | |
893 | ||
894 | int CrushWrapper::get_item_weight(int id) const | |
895 | { | |
896 | for (int bidx = 0; bidx < crush->max_buckets; bidx++) { | |
897 | crush_bucket *b = crush->buckets[bidx]; | |
898 | if (b == NULL) | |
899 | continue; | |
900 | if (b->id == id) | |
901 | return b->weight; | |
902 | for (unsigned i = 0; i < b->size; i++) | |
903 | if (b->items[i] == id) | |
904 | return crush_get_bucket_item_weight(b, i); | |
905 | } | |
906 | return -ENOENT; | |
907 | } | |
908 | ||
909 | int CrushWrapper::get_item_weight_in_loc(int id, const map<string,string> &loc) | |
910 | { | |
911 | for (map<string,string>::const_iterator l = loc.begin(); l != loc.end(); ++l) { | |
912 | ||
913 | int bid = get_item_id(l->second); | |
914 | if (!bucket_exists(bid)) | |
915 | continue; | |
916 | crush_bucket *b = get_bucket(bid); | |
917 | for (unsigned int i = 0; i < b->size; i++) { | |
918 | if (b->items[i] == id) { | |
919 | return crush_get_bucket_item_weight(b, i); | |
920 | } | |
921 | } | |
922 | } | |
923 | return -ENOENT; | |
924 | } | |
925 | ||
926 | int CrushWrapper::adjust_item_weight(CephContext *cct, int id, int weight) | |
927 | { | |
928 | ldout(cct, 5) << "adjust_item_weight " << id << " weight " << weight << dendl; | |
929 | int changed = 0; | |
930 | for (int bidx = 0; bidx < crush->max_buckets; bidx++) { | |
931 | crush_bucket *b = crush->buckets[bidx]; | |
932 | if (b == 0) | |
933 | continue; | |
934 | for (unsigned i = 0; i < b->size; i++) { | |
935 | if (b->items[i] == id) { | |
936 | int diff = crush_bucket_adjust_item_weight(crush, b, id, weight); | |
937 | ldout(cct, 5) << "adjust_item_weight " << id << " diff " << diff << " in bucket " << bidx << dendl; | |
938 | adjust_item_weight(cct, -1 - bidx, b->weight); | |
939 | changed++; | |
940 | } | |
941 | } | |
942 | } | |
943 | if (!changed) | |
944 | return -ENOENT; | |
945 | return changed; | |
946 | } | |
947 | ||
948 | int CrushWrapper::adjust_item_weight_in_loc(CephContext *cct, int id, int weight, const map<string,string>& loc) | |
949 | { | |
950 | ldout(cct, 5) << "adjust_item_weight_in_loc " << id << " weight " << weight << " in " << loc << dendl; | |
951 | int changed = 0; | |
952 | ||
953 | for (map<string,string>::const_iterator l = loc.begin(); l != loc.end(); ++l) { | |
954 | int bid = get_item_id(l->second); | |
955 | if (!bucket_exists(bid)) | |
956 | continue; | |
957 | crush_bucket *b = get_bucket(bid); | |
958 | for (unsigned int i = 0; i < b->size; i++) { | |
959 | if (b->items[i] == id) { | |
960 | int diff = crush_bucket_adjust_item_weight(crush, b, id, weight); | |
961 | ldout(cct, 5) << "adjust_item_weight_in_loc " << id << " diff " << diff << " in bucket " << bid << dendl; | |
962 | adjust_item_weight(cct, bid, b->weight); | |
963 | changed++; | |
964 | } | |
965 | } | |
966 | } | |
967 | if (!changed) | |
968 | return -ENOENT; | |
969 | return changed; | |
970 | } | |
971 | ||
972 | int CrushWrapper::adjust_subtree_weight(CephContext *cct, int id, int weight) | |
973 | { | |
974 | ldout(cct, 5) << __func__ << " " << id << " weight " << weight << dendl; | |
975 | crush_bucket *b = get_bucket(id); | |
976 | if (IS_ERR(b)) | |
977 | return PTR_ERR(b); | |
978 | int changed = 0; | |
979 | list<crush_bucket*> q; | |
980 | q.push_back(b); | |
981 | while (!q.empty()) { | |
982 | b = q.front(); | |
983 | q.pop_front(); | |
984 | int local_changed = 0; | |
985 | for (unsigned i=0; i<b->size; ++i) { | |
986 | int n = b->items[i]; | |
987 | if (n >= 0) { | |
988 | crush_bucket_adjust_item_weight(crush, b, n, weight); | |
989 | ++changed; | |
990 | ++local_changed; | |
991 | } else { | |
992 | crush_bucket *sub = get_bucket(n); | |
993 | if (IS_ERR(sub)) | |
994 | continue; | |
995 | q.push_back(sub); | |
996 | } | |
997 | } | |
998 | if (local_changed) { | |
999 | adjust_item_weight(cct, b->id, b->weight); | |
1000 | } | |
1001 | } | |
1002 | return changed; | |
1003 | } | |
1004 | ||
1005 | bool CrushWrapper::check_item_present(int id) const | |
1006 | { | |
1007 | bool found = false; | |
1008 | ||
1009 | for (int bidx = 0; bidx < crush->max_buckets; bidx++) { | |
1010 | crush_bucket *b = crush->buckets[bidx]; | |
1011 | if (b == 0) | |
1012 | continue; | |
1013 | for (unsigned i = 0; i < b->size; i++) | |
1014 | if (b->items[i] == id) | |
1015 | found = true; | |
1016 | } | |
1017 | return found; | |
1018 | } | |
1019 | ||
1020 | ||
1021 | pair<string,string> CrushWrapper::get_immediate_parent(int id, int *_ret) | |
1022 | { | |
1023 | ||
1024 | for (int bidx = 0; bidx < crush->max_buckets; bidx++) { | |
1025 | crush_bucket *b = crush->buckets[bidx]; | |
1026 | if (b == 0) | |
1027 | continue; | |
1028 | for (unsigned i = 0; i < b->size; i++) | |
1029 | if (b->items[i] == id) { | |
1030 | string parent_id = name_map[b->id]; | |
1031 | string parent_bucket_type = type_map[b->type]; | |
1032 | if (_ret) | |
1033 | *_ret = 0; | |
1034 | return make_pair(parent_bucket_type, parent_id); | |
1035 | } | |
1036 | } | |
1037 | ||
1038 | if (_ret) | |
1039 | *_ret = -ENOENT; | |
1040 | ||
1041 | return pair<string, string>(); | |
1042 | } | |
1043 | ||
1044 | int CrushWrapper::get_immediate_parent_id(int id, int *parent) const | |
1045 | { | |
1046 | for (int bidx = 0; bidx < crush->max_buckets; bidx++) { | |
1047 | crush_bucket *b = crush->buckets[bidx]; | |
1048 | if (b == 0) | |
1049 | continue; | |
1050 | for (unsigned i = 0; i < b->size; i++) { | |
1051 | if (b->items[i] == id) { | |
1052 | *parent = b->id; | |
1053 | return 0; | |
1054 | } | |
1055 | } | |
1056 | } | |
1057 | return -ENOENT; | |
1058 | } | |
1059 | ||
1060 | bool CrushWrapper::class_is_in_use(int class_id) | |
1061 | { | |
1062 | for (auto &i : class_bucket) | |
1063 | for (auto &j : i.second) | |
1064 | if (j.first == class_id) | |
1065 | return true; | |
1066 | ||
1067 | for (auto &i : class_map) | |
1068 | if (i.second == class_id) | |
1069 | return true; | |
1070 | ||
1071 | return false; | |
1072 | } | |
1073 | ||
1074 | int CrushWrapper::populate_classes() | |
1075 | { | |
1076 | set<int> roots; | |
1077 | find_roots(roots); | |
1078 | for (auto &r : roots) { | |
1079 | if (r >= 0) | |
1080 | continue; | |
1081 | if (id_has_class(r)) | |
1082 | continue; | |
1083 | for (auto &c : class_name) { | |
1084 | int clone; | |
1085 | int res = device_class_clone(r, c.first, &clone); | |
1086 | if (res < 0) | |
1087 | return res; | |
1088 | } | |
1089 | } | |
1090 | return 0; | |
1091 | } | |
1092 | ||
1093 | int CrushWrapper::cleanup_classes() | |
1094 | { | |
1095 | return trim_roots_with_class(true); | |
1096 | } | |
1097 | ||
1098 | int CrushWrapper::trim_roots_with_class(bool unused) | |
1099 | { | |
1100 | set<int> roots; | |
1101 | find_roots(roots); | |
1102 | for (auto &r : roots) { | |
1103 | if (r >= 0) | |
1104 | continue; | |
1105 | if (!id_has_class(r)) | |
1106 | continue; | |
1107 | int res = remove_root(r, unused); | |
1108 | if (res) | |
1109 | return res; | |
1110 | } | |
1111 | // there is no need to reweight because we only remove from the | |
1112 | // root and down | |
1113 | return 0; | |
1114 | } | |
1115 | ||
1116 | void CrushWrapper::reweight(CephContext *cct) | |
1117 | { | |
1118 | set<int> roots; | |
1119 | find_roots(roots); | |
1120 | for (set<int>::iterator p = roots.begin(); p != roots.end(); ++p) { | |
1121 | if (*p >= 0) | |
1122 | continue; | |
1123 | crush_bucket *b = get_bucket(*p); | |
1124 | ldout(cct, 5) << "reweight bucket " << *p << dendl; | |
1125 | int r = crush_reweight_bucket(crush, b); | |
1126 | assert(r == 0); | |
1127 | } | |
1128 | } | |
1129 | ||
1130 | int CrushWrapper::add_simple_ruleset_at(string name, string root_name, | |
1131 | string failure_domain_name, | |
1132 | string mode, int rule_type, | |
1133 | int rno, ostream *err) | |
1134 | { | |
1135 | if (rule_exists(name)) { | |
1136 | if (err) | |
1137 | *err << "rule " << name << " exists"; | |
1138 | return -EEXIST; | |
1139 | } | |
1140 | if (rno >= 0) { | |
1141 | if (rule_exists(rno)) { | |
1142 | if (err) | |
1143 | *err << "rule with ruleno " << rno << " exists"; | |
1144 | return -EEXIST; | |
1145 | } | |
1146 | if (ruleset_exists(rno)) { | |
1147 | if (err) | |
1148 | *err << "ruleset " << rno << " exists"; | |
1149 | return -EEXIST; | |
1150 | } | |
1151 | } else { | |
1152 | for (rno = 0; rno < get_max_rules(); rno++) { | |
1153 | if (!rule_exists(rno) && !ruleset_exists(rno)) | |
1154 | break; | |
1155 | } | |
1156 | } | |
1157 | if (!name_exists(root_name)) { | |
1158 | if (err) | |
1159 | *err << "root item " << root_name << " does not exist"; | |
1160 | return -ENOENT; | |
1161 | } | |
1162 | int root = get_item_id(root_name); | |
1163 | int type = 0; | |
1164 | if (failure_domain_name.length()) { | |
1165 | type = get_type_id(failure_domain_name); | |
1166 | if (type < 0) { | |
1167 | if (err) | |
1168 | *err << "unknown type " << failure_domain_name; | |
1169 | return -EINVAL; | |
1170 | } | |
1171 | } | |
1172 | if (mode != "firstn" && mode != "indep") { | |
1173 | if (err) | |
1174 | *err << "unknown mode " << mode; | |
1175 | return -EINVAL; | |
1176 | } | |
1177 | ||
1178 | int steps = 3; | |
1179 | if (mode == "indep") | |
1180 | steps = 5; | |
1181 | int min_rep = mode == "firstn" ? 1 : 3; | |
1182 | int max_rep = mode == "firstn" ? 10 : 20; | |
1183 | //set the ruleset the same as rule_id(rno) | |
1184 | crush_rule *rule = crush_make_rule(steps, rno, rule_type, min_rep, max_rep); | |
1185 | assert(rule); | |
1186 | int step = 0; | |
1187 | if (mode == "indep") { | |
1188 | crush_rule_set_step(rule, step++, CRUSH_RULE_SET_CHOOSELEAF_TRIES, 5, 0); | |
1189 | crush_rule_set_step(rule, step++, CRUSH_RULE_SET_CHOOSE_TRIES, 100, 0); | |
1190 | } | |
1191 | crush_rule_set_step(rule, step++, CRUSH_RULE_TAKE, root, 0); | |
1192 | if (type) | |
1193 | crush_rule_set_step(rule, step++, | |
1194 | mode == "firstn" ? CRUSH_RULE_CHOOSELEAF_FIRSTN : | |
1195 | CRUSH_RULE_CHOOSELEAF_INDEP, | |
1196 | CRUSH_CHOOSE_N, | |
1197 | type); | |
1198 | else | |
1199 | crush_rule_set_step(rule, step++, | |
1200 | mode == "firstn" ? CRUSH_RULE_CHOOSE_FIRSTN : | |
1201 | CRUSH_RULE_CHOOSE_INDEP, | |
1202 | CRUSH_CHOOSE_N, | |
1203 | 0); | |
1204 | crush_rule_set_step(rule, step++, CRUSH_RULE_EMIT, 0, 0); | |
1205 | ||
1206 | int ret = crush_add_rule(crush, rule, rno); | |
1207 | if(ret < 0) { | |
1208 | *err << "failed to add rule " << rno << " because " << cpp_strerror(ret); | |
1209 | return ret; | |
1210 | } | |
1211 | set_rule_name(rno, name); | |
1212 | have_rmaps = false; | |
1213 | return rno; | |
1214 | } | |
1215 | ||
1216 | int CrushWrapper::add_simple_ruleset(string name, string root_name, | |
1217 | string failure_domain_name, | |
1218 | string mode, int rule_type, | |
1219 | ostream *err) | |
1220 | { | |
1221 | return add_simple_ruleset_at(name, root_name, failure_domain_name, mode, | |
1222 | rule_type, -1, err); | |
1223 | } | |
1224 | ||
1225 | int CrushWrapper::get_rule_weight_osd_map(unsigned ruleno, map<int,float> *pmap) | |
1226 | { | |
1227 | if (ruleno >= crush->max_rules) | |
1228 | return -ENOENT; | |
1229 | if (crush->rules[ruleno] == NULL) | |
1230 | return -ENOENT; | |
1231 | crush_rule *rule = crush->rules[ruleno]; | |
1232 | ||
1233 | // build a weight map for each TAKE in the rule, and then merge them | |
1234 | for (unsigned i=0; i<rule->len; ++i) { | |
1235 | map<int,float> m; | |
1236 | float sum = 0; | |
1237 | if (rule->steps[i].op == CRUSH_RULE_TAKE) { | |
1238 | int n = rule->steps[i].arg1; | |
1239 | if (n >= 0) { | |
1240 | m[n] = 1.0; | |
1241 | sum = 1.0; | |
1242 | } else { | |
1243 | list<int> q; | |
1244 | q.push_back(n); | |
1245 | //breadth first iterate the OSD tree | |
1246 | while (!q.empty()) { | |
1247 | int bno = q.front(); | |
1248 | q.pop_front(); | |
1249 | crush_bucket *b = crush->buckets[-1-bno]; | |
1250 | assert(b); | |
1251 | for (unsigned j=0; j<b->size; ++j) { | |
1252 | int item_id = b->items[j]; | |
1253 | if (item_id >= 0) { //it's an OSD | |
1254 | float w = crush_get_bucket_item_weight(b, j); | |
1255 | m[item_id] = w; | |
1256 | sum += w; | |
1257 | } else { //not an OSD, expand the child later | |
1258 | q.push_back(item_id); | |
1259 | } | |
1260 | } | |
1261 | } | |
1262 | } | |
1263 | } | |
1264 | for (map<int,float>::iterator p = m.begin(); p != m.end(); ++p) { | |
1265 | map<int,float>::iterator q = pmap->find(p->first); | |
1266 | if (q == pmap->end()) { | |
1267 | (*pmap)[p->first] = p->second / sum; | |
1268 | } else { | |
1269 | q->second += p->second / sum; | |
1270 | } | |
1271 | } | |
1272 | } | |
1273 | ||
1274 | return 0; | |
1275 | } | |
1276 | ||
1277 | int CrushWrapper::remove_rule(int ruleno) | |
1278 | { | |
1279 | if (ruleno >= (int)crush->max_rules) | |
1280 | return -ENOENT; | |
1281 | if (crush->rules[ruleno] == NULL) | |
1282 | return -ENOENT; | |
1283 | crush_destroy_rule(crush->rules[ruleno]); | |
1284 | crush->rules[ruleno] = NULL; | |
1285 | rule_name_map.erase(ruleno); | |
1286 | have_rmaps = false; | |
1287 | return 0; | |
1288 | } | |
1289 | ||
1290 | int CrushWrapper::update_device_class(CephContext *cct, int id, const string& class_name, const string& name) | |
1291 | { | |
1292 | int class_id = get_class_id(class_name); | |
1293 | if (class_id < 0) { | |
1294 | ldout(cct, 0) << "update_device_class class " << class_name << " does not exist " << dendl; | |
1295 | return -ENOENT; | |
1296 | } | |
1297 | if (id < 0) { | |
1298 | ldout(cct, 0) << "update_device_class " << name << " id " << id << " is negative " << dendl; | |
1299 | return -EINVAL; | |
1300 | } | |
1301 | assert(item_exists(id)); | |
1302 | ||
1303 | if (class_map.count(id) != 0 && class_map[id] == class_id) { | |
1304 | ldout(cct, 5) << "update_device_class " << name << " already set to class " << class_name << dendl; | |
1305 | return 0; | |
1306 | } | |
1307 | ||
1308 | set_item_class(id, class_id); | |
1309 | ||
1310 | int r = rebuild_roots_with_classes(); | |
1311 | if (r < 0) | |
1312 | return r; | |
1313 | return 1; | |
1314 | } | |
1315 | ||
1316 | int CrushWrapper::device_class_clone(int original_id, int device_class, int *clone) | |
1317 | { | |
1318 | const char *item_name = get_item_name(original_id); | |
1319 | if (item_name == NULL) | |
1320 | return -ECHILD; | |
1321 | const char *class_name = get_class_name(device_class); | |
1322 | if (class_name == NULL) | |
1323 | return -EBADF; | |
1324 | string copy_name = item_name + string("~") + class_name; | |
1325 | if (name_exists(copy_name)) { | |
1326 | *clone = get_item_id(copy_name); | |
1327 | return 0; | |
1328 | } | |
1329 | crush_bucket *original = get_bucket(original_id); | |
1330 | if (original == NULL) | |
1331 | return -ENOENT; | |
1332 | crush_bucket *copy = crush_make_bucket(crush, | |
1333 | original->alg, | |
1334 | original->hash, | |
1335 | original->type, | |
1336 | 0, NULL, NULL); | |
1337 | if(copy == NULL) | |
1338 | return -ENOMEM; | |
1339 | for (unsigned i = 0; i < original->size; i++) { | |
1340 | int item = original->items[i]; | |
1341 | int weight = crush_get_bucket_item_weight(original, i); | |
1342 | if (item >= 0) { | |
1343 | if (class_map.count(item) != 0 && class_map[item] == device_class) { | |
1344 | int res = crush_bucket_add_item(crush, copy, item, weight); | |
1345 | if (res) | |
1346 | return res; | |
1347 | } | |
1348 | } else { | |
1349 | int child_copy_id; | |
1350 | int res = device_class_clone(item, device_class, &child_copy_id); | |
1351 | if (res < 0) | |
1352 | return res; | |
1353 | crush_bucket *child_copy = get_bucket(child_copy_id); | |
1354 | if (IS_ERR(child_copy)) | |
1355 | return -ENOENT; | |
1356 | res = crush_bucket_add_item(crush, copy, child_copy_id, child_copy->weight); | |
1357 | if (res) | |
1358 | return res; | |
1359 | } | |
1360 | } | |
1361 | int res = crush_add_bucket(crush, 0, copy, clone); | |
1362 | if (res) | |
1363 | return res; | |
1364 | res = set_item_class(*clone, device_class); | |
1365 | if (res < 0) | |
1366 | return res; | |
1367 | // we do not use set_item_name because the name is intentionally invalid | |
1368 | name_map[*clone] = copy_name; | |
1369 | if (have_rmaps) | |
1370 | name_rmap[copy_name] = *clone; | |
1371 | class_bucket[original_id][device_class] = *clone; | |
1372 | return 0; | |
1373 | } | |
1374 | ||
1375 | int CrushWrapper::rebuild_roots_with_classes() | |
1376 | { | |
1377 | int r = trim_roots_with_class(false); | |
1378 | if (r < 0) | |
1379 | return r; | |
1380 | r = populate_classes(); | |
1381 | if (r < 0) | |
1382 | return r; | |
1383 | return trim_roots_with_class(true); | |
1384 | } | |
1385 | ||
1386 | void CrushWrapper::encode(bufferlist& bl, uint64_t features) const | |
1387 | { | |
1388 | assert(crush); | |
1389 | ||
1390 | __u32 magic = CRUSH_MAGIC; | |
1391 | ::encode(magic, bl); | |
1392 | ||
1393 | ::encode(crush->max_buckets, bl); | |
1394 | ::encode(crush->max_rules, bl); | |
1395 | ::encode(crush->max_devices, bl); | |
1396 | ||
1397 | // buckets | |
1398 | for (int i=0; i<crush->max_buckets; i++) { | |
1399 | __u32 alg = 0; | |
1400 | if (crush->buckets[i]) alg = crush->buckets[i]->alg; | |
1401 | ::encode(alg, bl); | |
1402 | if (!alg) | |
1403 | continue; | |
1404 | ||
1405 | ::encode(crush->buckets[i]->id, bl); | |
1406 | ::encode(crush->buckets[i]->type, bl); | |
1407 | ::encode(crush->buckets[i]->alg, bl); | |
1408 | ::encode(crush->buckets[i]->hash, bl); | |
1409 | ::encode(crush->buckets[i]->weight, bl); | |
1410 | ::encode(crush->buckets[i]->size, bl); | |
1411 | for (unsigned j=0; j<crush->buckets[i]->size; j++) | |
1412 | ::encode(crush->buckets[i]->items[j], bl); | |
1413 | ||
1414 | switch (crush->buckets[i]->alg) { | |
1415 | case CRUSH_BUCKET_UNIFORM: | |
1416 | ::encode((reinterpret_cast<crush_bucket_uniform*>(crush->buckets[i]))->item_weight, bl); | |
1417 | break; | |
1418 | ||
1419 | case CRUSH_BUCKET_LIST: | |
1420 | for (unsigned j=0; j<crush->buckets[i]->size; j++) { | |
1421 | ::encode((reinterpret_cast<crush_bucket_list*>(crush->buckets[i]))->item_weights[j], bl); | |
1422 | ::encode((reinterpret_cast<crush_bucket_list*>(crush->buckets[i]))->sum_weights[j], bl); | |
1423 | } | |
1424 | break; | |
1425 | ||
1426 | case CRUSH_BUCKET_TREE: | |
1427 | ::encode((reinterpret_cast<crush_bucket_tree*>(crush->buckets[i]))->num_nodes, bl); | |
1428 | for (unsigned j=0; j<(reinterpret_cast<crush_bucket_tree*>(crush->buckets[i]))->num_nodes; j++) | |
1429 | ::encode((reinterpret_cast<crush_bucket_tree*>(crush->buckets[i]))->node_weights[j], bl); | |
1430 | break; | |
1431 | ||
1432 | case CRUSH_BUCKET_STRAW: | |
1433 | for (unsigned j=0; j<crush->buckets[i]->size; j++) { | |
1434 | ::encode((reinterpret_cast<crush_bucket_straw*>(crush->buckets[i]))->item_weights[j], bl); | |
1435 | ::encode((reinterpret_cast<crush_bucket_straw*>(crush->buckets[i]))->straws[j], bl); | |
1436 | } | |
1437 | break; | |
1438 | ||
1439 | case CRUSH_BUCKET_STRAW2: | |
1440 | for (unsigned j=0; j<crush->buckets[i]->size; j++) { | |
1441 | ::encode((reinterpret_cast<crush_bucket_straw2*>(crush->buckets[i]))->item_weights[j], bl); | |
1442 | } | |
1443 | break; | |
1444 | ||
1445 | default: | |
1446 | ceph_abort(); | |
1447 | break; | |
1448 | } | |
1449 | } | |
1450 | ||
1451 | // rules | |
1452 | for (unsigned i=0; i<crush->max_rules; i++) { | |
1453 | __u32 yes = crush->rules[i] ? 1:0; | |
1454 | ::encode(yes, bl); | |
1455 | if (!yes) | |
1456 | continue; | |
1457 | ||
1458 | ::encode(crush->rules[i]->len, bl); | |
1459 | ::encode(crush->rules[i]->mask, bl); | |
1460 | for (unsigned j=0; j<crush->rules[i]->len; j++) | |
1461 | ::encode(crush->rules[i]->steps[j], bl); | |
1462 | } | |
1463 | ||
1464 | // name info | |
1465 | ::encode(type_map, bl); | |
1466 | ::encode(name_map, bl); | |
1467 | ::encode(rule_name_map, bl); | |
1468 | ||
1469 | // tunables | |
1470 | ::encode(crush->choose_local_tries, bl); | |
1471 | ::encode(crush->choose_local_fallback_tries, bl); | |
1472 | ::encode(crush->choose_total_tries, bl); | |
1473 | ::encode(crush->chooseleaf_descend_once, bl); | |
1474 | ::encode(crush->chooseleaf_vary_r, bl); | |
1475 | ::encode(crush->straw_calc_version, bl); | |
1476 | ::encode(crush->allowed_bucket_algs, bl); | |
1477 | if (features & CEPH_FEATURE_CRUSH_TUNABLES5) { | |
1478 | ::encode(crush->chooseleaf_stable, bl); | |
1479 | } | |
1480 | ||
1481 | if (HAVE_FEATURE(features, SERVER_LUMINOUS)) { | |
1482 | // device classes | |
1483 | ::encode(class_map, bl); | |
1484 | ::encode(class_name, bl); | |
1485 | ::encode(class_bucket, bl); | |
1486 | ||
1487 | ::encode(choose_args.size(), bl); | |
1488 | for (auto c : choose_args) { | |
1489 | ::encode(c.first, bl); | |
1490 | crush_choose_arg_map arg_map = c.second; | |
1491 | __u32 size = 0; | |
1492 | for (__u32 i = 0; i < arg_map.size; i++) { | |
1493 | crush_choose_arg *arg = &arg_map.args[i]; | |
1494 | if (arg->weight_set_size == 0 && | |
1495 | arg->ids_size == 0) | |
1496 | continue; | |
1497 | size++; | |
1498 | } | |
1499 | ::encode(size, bl); | |
1500 | for (__u32 i = 0; i < arg_map.size; i++) { | |
1501 | crush_choose_arg *arg = &arg_map.args[i]; | |
1502 | if (arg->weight_set_size == 0 && | |
1503 | arg->ids_size == 0) | |
1504 | continue; | |
1505 | ::encode(i, bl); | |
1506 | ::encode(arg->weight_set_size, bl); | |
1507 | for (__u32 j = 0; j < arg->weight_set_size; j++) { | |
1508 | crush_weight_set *weight_set = &arg->weight_set[j]; | |
1509 | ::encode(weight_set->size, bl); | |
1510 | for (__u32 k = 0; k < weight_set->size; k++) | |
1511 | ::encode(weight_set->weights[k], bl); | |
1512 | } | |
1513 | ::encode(arg->ids_size, bl); | |
1514 | for (__u32 j = 0; j < arg->ids_size; j++) | |
1515 | ::encode(arg->ids[j], bl); | |
1516 | } | |
1517 | } | |
1518 | } | |
1519 | } | |
1520 | ||
1521 | static void decode_32_or_64_string_map(map<int32_t,string>& m, bufferlist::iterator& blp) | |
1522 | { | |
1523 | m.clear(); | |
1524 | __u32 n; | |
1525 | ::decode(n, blp); | |
1526 | while (n--) { | |
1527 | __s32 key; | |
1528 | ::decode(key, blp); | |
1529 | ||
1530 | __u32 strlen; | |
1531 | ::decode(strlen, blp); | |
1532 | if (strlen == 0) { | |
1533 | // der, key was actually 64-bits! | |
1534 | ::decode(strlen, blp); | |
1535 | } | |
1536 | ::decode_nohead(strlen, m[key], blp); | |
1537 | } | |
1538 | } | |
1539 | ||
1540 | void CrushWrapper::decode(bufferlist::iterator& blp) | |
1541 | { | |
1542 | create(); | |
1543 | ||
1544 | __u32 magic; | |
1545 | ::decode(magic, blp); | |
1546 | if (magic != CRUSH_MAGIC) | |
1547 | throw buffer::malformed_input("bad magic number"); | |
1548 | ||
1549 | ::decode(crush->max_buckets, blp); | |
1550 | ::decode(crush->max_rules, blp); | |
1551 | ::decode(crush->max_devices, blp); | |
1552 | ||
1553 | // legacy tunables, unless we decode something newer | |
1554 | set_tunables_legacy(); | |
1555 | ||
1556 | try { | |
1557 | // buckets | |
1558 | crush->buckets = (crush_bucket**)calloc(1, crush->max_buckets * sizeof(crush_bucket*)); | |
1559 | for (int i=0; i<crush->max_buckets; i++) { | |
1560 | decode_crush_bucket(&crush->buckets[i], blp); | |
1561 | } | |
1562 | ||
1563 | // rules | |
1564 | crush->rules = (crush_rule**)calloc(1, crush->max_rules * sizeof(crush_rule*)); | |
1565 | for (unsigned i = 0; i < crush->max_rules; ++i) { | |
1566 | __u32 yes; | |
1567 | ::decode(yes, blp); | |
1568 | if (!yes) { | |
1569 | crush->rules[i] = NULL; | |
1570 | continue; | |
1571 | } | |
1572 | ||
1573 | __u32 len; | |
1574 | ::decode(len, blp); | |
1575 | crush->rules[i] = reinterpret_cast<crush_rule*>(calloc(1, crush_rule_size(len))); | |
1576 | crush->rules[i]->len = len; | |
1577 | ::decode(crush->rules[i]->mask, blp); | |
1578 | for (unsigned j=0; j<crush->rules[i]->len; j++) | |
1579 | ::decode(crush->rules[i]->steps[j], blp); | |
1580 | } | |
1581 | ||
1582 | // name info | |
1583 | // NOTE: we had a bug where we were incoding int instead of int32, which means the | |
1584 | // 'key' field for these maps may be either 32 or 64 bits, depending. tolerate | |
1585 | // both by assuming the string is always non-empty. | |
1586 | decode_32_or_64_string_map(type_map, blp); | |
1587 | decode_32_or_64_string_map(name_map, blp); | |
1588 | decode_32_or_64_string_map(rule_name_map, blp); | |
1589 | ||
1590 | // tunables | |
1591 | if (!blp.end()) { | |
1592 | ::decode(crush->choose_local_tries, blp); | |
1593 | ::decode(crush->choose_local_fallback_tries, blp); | |
1594 | ::decode(crush->choose_total_tries, blp); | |
1595 | } | |
1596 | if (!blp.end()) { | |
1597 | ::decode(crush->chooseleaf_descend_once, blp); | |
1598 | } | |
1599 | if (!blp.end()) { | |
1600 | ::decode(crush->chooseleaf_vary_r, blp); | |
1601 | } | |
1602 | if (!blp.end()) { | |
1603 | ::decode(crush->straw_calc_version, blp); | |
1604 | } | |
1605 | if (!blp.end()) { | |
1606 | ::decode(crush->allowed_bucket_algs, blp); | |
1607 | } | |
1608 | if (!blp.end()) { | |
1609 | ::decode(crush->chooseleaf_stable, blp); | |
1610 | } | |
1611 | if (!blp.end()) { | |
1612 | ::decode(class_map, blp); | |
1613 | ::decode(class_name, blp); | |
1614 | for (auto &c : class_name) | |
1615 | class_rname[c.second] = c.first; | |
1616 | ::decode(class_bucket, blp); | |
1617 | cleanup_classes(); | |
1618 | } | |
1619 | if (!blp.end()) { | |
1620 | size_t choose_args_size; | |
1621 | ::decode(choose_args_size, blp); | |
1622 | for (size_t i = 0; i < choose_args_size; i++) { | |
1623 | uint64_t choose_args_index; | |
1624 | ::decode(choose_args_index, blp); | |
1625 | crush_choose_arg_map arg_map; | |
1626 | arg_map.size = crush->max_buckets; | |
1627 | arg_map.args = (crush_choose_arg*)calloc(arg_map.size, sizeof(crush_choose_arg)); | |
1628 | __u32 size; | |
1629 | ::decode(size, blp); | |
1630 | for (__u32 j = 0; j < size; j++) { | |
1631 | __u32 bucket_index; | |
1632 | ::decode(bucket_index, blp); | |
1633 | assert(bucket_index < arg_map.size); | |
1634 | crush_choose_arg *arg = &arg_map.args[bucket_index]; | |
1635 | ::decode(arg->weight_set_size, blp); | |
1636 | arg->weight_set = (crush_weight_set*)calloc(arg->weight_set_size, sizeof(crush_weight_set)); | |
1637 | for (__u32 k = 0; k < arg->weight_set_size; k++) { | |
1638 | crush_weight_set *weight_set = &arg->weight_set[k]; | |
1639 | ::decode(weight_set->size, blp); | |
1640 | weight_set->weights = (__u32*)calloc(weight_set->size, sizeof(__u32)); | |
1641 | for (__u32 l = 0; l < weight_set->size; l++) | |
1642 | ::decode(weight_set->weights[l], blp); | |
1643 | } | |
1644 | ::decode(arg->ids_size, blp); | |
1645 | arg->ids = (int*)calloc(arg->ids_size, sizeof(int)); | |
1646 | for (__u32 k = 0; k < arg->ids_size; k++) | |
1647 | ::decode(arg->ids[k], blp); | |
1648 | } | |
1649 | choose_args[choose_args_index] = arg_map; | |
1650 | } | |
1651 | } | |
1652 | finalize(); | |
1653 | } | |
1654 | catch (...) { | |
1655 | crush_destroy(crush); | |
1656 | throw; | |
1657 | } | |
1658 | } | |
1659 | ||
1660 | void CrushWrapper::decode_crush_bucket(crush_bucket** bptr, bufferlist::iterator &blp) | |
1661 | { | |
1662 | __u32 alg; | |
1663 | ::decode(alg, blp); | |
1664 | if (!alg) { | |
1665 | *bptr = NULL; | |
1666 | return; | |
1667 | } | |
1668 | ||
1669 | int size = 0; | |
1670 | switch (alg) { | |
1671 | case CRUSH_BUCKET_UNIFORM: | |
1672 | size = sizeof(crush_bucket_uniform); | |
1673 | break; | |
1674 | case CRUSH_BUCKET_LIST: | |
1675 | size = sizeof(crush_bucket_list); | |
1676 | break; | |
1677 | case CRUSH_BUCKET_TREE: | |
1678 | size = sizeof(crush_bucket_tree); | |
1679 | break; | |
1680 | case CRUSH_BUCKET_STRAW: | |
1681 | size = sizeof(crush_bucket_straw); | |
1682 | break; | |
1683 | case CRUSH_BUCKET_STRAW2: | |
1684 | size = sizeof(crush_bucket_straw2); | |
1685 | break; | |
1686 | default: | |
1687 | { | |
1688 | char str[128]; | |
1689 | snprintf(str, sizeof(str), "unsupported bucket algorithm: %d", alg); | |
1690 | throw buffer::malformed_input(str); | |
1691 | } | |
1692 | } | |
1693 | crush_bucket *bucket = reinterpret_cast<crush_bucket*>(calloc(1, size)); | |
1694 | *bptr = bucket; | |
1695 | ||
1696 | ::decode(bucket->id, blp); | |
1697 | ::decode(bucket->type, blp); | |
1698 | ::decode(bucket->alg, blp); | |
1699 | ::decode(bucket->hash, blp); | |
1700 | ::decode(bucket->weight, blp); | |
1701 | ::decode(bucket->size, blp); | |
1702 | ||
1703 | bucket->items = (__s32*)calloc(1, bucket->size * sizeof(__s32)); | |
1704 | for (unsigned j = 0; j < bucket->size; ++j) { | |
1705 | ::decode(bucket->items[j], blp); | |
1706 | } | |
1707 | ||
1708 | switch (bucket->alg) { | |
1709 | case CRUSH_BUCKET_UNIFORM: | |
1710 | ::decode((reinterpret_cast<crush_bucket_uniform*>(bucket))->item_weight, blp); | |
1711 | break; | |
1712 | ||
1713 | case CRUSH_BUCKET_LIST: { | |
1714 | crush_bucket_list* cbl = reinterpret_cast<crush_bucket_list*>(bucket); | |
1715 | cbl->item_weights = (__u32*)calloc(1, bucket->size * sizeof(__u32)); | |
1716 | cbl->sum_weights = (__u32*)calloc(1, bucket->size * sizeof(__u32)); | |
1717 | ||
1718 | for (unsigned j = 0; j < bucket->size; ++j) { | |
1719 | ::decode(cbl->item_weights[j], blp); | |
1720 | ::decode(cbl->sum_weights[j], blp); | |
1721 | } | |
1722 | break; | |
1723 | } | |
1724 | ||
1725 | case CRUSH_BUCKET_TREE: { | |
1726 | crush_bucket_tree* cbt = reinterpret_cast<crush_bucket_tree*>(bucket); | |
1727 | ::decode(cbt->num_nodes, blp); | |
1728 | cbt->node_weights = (__u32*)calloc(1, cbt->num_nodes * sizeof(__u32)); | |
1729 | for (unsigned j=0; j<cbt->num_nodes; j++) { | |
1730 | ::decode(cbt->node_weights[j], blp); | |
1731 | } | |
1732 | break; | |
1733 | } | |
1734 | ||
1735 | case CRUSH_BUCKET_STRAW: { | |
1736 | crush_bucket_straw* cbs = reinterpret_cast<crush_bucket_straw*>(bucket); | |
1737 | cbs->straws = (__u32*)calloc(1, bucket->size * sizeof(__u32)); | |
1738 | cbs->item_weights = (__u32*)calloc(1, bucket->size * sizeof(__u32)); | |
1739 | for (unsigned j = 0; j < bucket->size; ++j) { | |
1740 | ::decode(cbs->item_weights[j], blp); | |
1741 | ::decode(cbs->straws[j], blp); | |
1742 | } | |
1743 | break; | |
1744 | } | |
1745 | ||
1746 | case CRUSH_BUCKET_STRAW2: { | |
1747 | crush_bucket_straw2* cbs = reinterpret_cast<crush_bucket_straw2*>(bucket); | |
1748 | cbs->item_weights = (__u32*)calloc(1, bucket->size * sizeof(__u32)); | |
1749 | for (unsigned j = 0; j < bucket->size; ++j) { | |
1750 | ::decode(cbs->item_weights[j], blp); | |
1751 | } | |
1752 | break; | |
1753 | } | |
1754 | ||
1755 | default: | |
1756 | // We should have handled this case in the first switch statement | |
1757 | ceph_abort(); | |
1758 | break; | |
1759 | } | |
1760 | } | |
1761 | ||
1762 | ||
1763 | void CrushWrapper::dump(Formatter *f) const | |
1764 | { | |
1765 | f->open_array_section("devices"); | |
1766 | for (int i=0; i<get_max_devices(); i++) { | |
1767 | f->open_object_section("device"); | |
1768 | f->dump_int("id", i); | |
1769 | const char *n = get_item_name(i); | |
1770 | if (n) { | |
1771 | f->dump_string("name", n); | |
1772 | } else { | |
1773 | char name[20]; | |
1774 | sprintf(name, "device%d", i); | |
1775 | f->dump_string("name", name); | |
1776 | } | |
1777 | const char *device_class = get_item_class(i); | |
1778 | if (device_class != NULL) | |
1779 | f->dump_string("class", device_class); | |
1780 | f->close_section(); | |
1781 | } | |
1782 | f->close_section(); | |
1783 | ||
1784 | f->open_array_section("types"); | |
1785 | int n = get_num_type_names(); | |
1786 | for (int i=0; n; i++) { | |
1787 | const char *name = get_type_name(i); | |
1788 | if (!name) { | |
1789 | if (i == 0) { | |
1790 | f->open_object_section("type"); | |
1791 | f->dump_int("type_id", 0); | |
1792 | f->dump_string("name", "device"); | |
1793 | f->close_section(); | |
1794 | } | |
1795 | continue; | |
1796 | } | |
1797 | n--; | |
1798 | f->open_object_section("type"); | |
1799 | f->dump_int("type_id", i); | |
1800 | f->dump_string("name", name); | |
1801 | f->close_section(); | |
1802 | } | |
1803 | f->close_section(); | |
1804 | ||
1805 | f->open_array_section("buckets"); | |
1806 | for (int bucket = -1; bucket > -1-get_max_buckets(); --bucket) { | |
1807 | if (!bucket_exists(bucket)) | |
1808 | continue; | |
1809 | f->open_object_section("bucket"); | |
1810 | f->dump_int("id", bucket); | |
1811 | if (get_item_name(bucket)) | |
1812 | f->dump_string("name", get_item_name(bucket)); | |
1813 | f->dump_int("type_id", get_bucket_type(bucket)); | |
1814 | if (get_type_name(get_bucket_type(bucket))) | |
1815 | f->dump_string("type_name", get_type_name(get_bucket_type(bucket))); | |
1816 | f->dump_int("weight", get_bucket_weight(bucket)); | |
1817 | f->dump_string("alg", crush_bucket_alg_name(get_bucket_alg(bucket))); | |
1818 | f->dump_string("hash", crush_hash_name(get_bucket_hash(bucket))); | |
1819 | f->open_array_section("items"); | |
1820 | for (int j=0; j<get_bucket_size(bucket); j++) { | |
1821 | f->open_object_section("item"); | |
1822 | f->dump_int("id", get_bucket_item(bucket, j)); | |
1823 | f->dump_int("weight", get_bucket_item_weight(bucket, j)); | |
1824 | f->dump_int("pos", j); | |
1825 | f->close_section(); | |
1826 | } | |
1827 | f->close_section(); | |
1828 | f->close_section(); | |
1829 | } | |
1830 | f->close_section(); | |
1831 | ||
1832 | f->open_array_section("rules"); | |
1833 | dump_rules(f); | |
1834 | f->close_section(); | |
1835 | ||
1836 | f->open_object_section("tunables"); | |
1837 | dump_tunables(f); | |
1838 | f->close_section(); | |
1839 | ||
1840 | dump_choose_args(f); | |
1841 | } | |
1842 | ||
1843 | namespace { | |
1844 | // depth first walker | |
1845 | class TreeDumper { | |
1846 | typedef CrushTreeDumper::Item Item; | |
1847 | const CrushWrapper *crush; | |
1848 | public: | |
1849 | explicit TreeDumper(const CrushWrapper *crush) | |
1850 | : crush(crush) {} | |
1851 | ||
1852 | void dump(Formatter *f) { | |
1853 | set<int> roots; | |
1854 | crush->find_roots(roots); | |
1855 | for (set<int>::iterator root = roots.begin(); root != roots.end(); ++root) { | |
1856 | dump_item(Item(*root, 0, crush->get_bucket_weightf(*root)), f); | |
1857 | } | |
1858 | } | |
1859 | ||
1860 | private: | |
1861 | void dump_item(const Item& qi, Formatter* f) { | |
1862 | if (qi.is_bucket()) { | |
1863 | f->open_object_section("bucket"); | |
1864 | CrushTreeDumper::dump_item_fields(crush, qi, f); | |
1865 | dump_bucket_children(qi, f); | |
1866 | f->close_section(); | |
1867 | } else { | |
1868 | f->open_object_section("device"); | |
1869 | CrushTreeDumper::dump_item_fields(crush, qi, f); | |
1870 | f->close_section(); | |
1871 | } | |
1872 | } | |
1873 | ||
1874 | void dump_bucket_children(const Item& parent, Formatter* f) { | |
1875 | f->open_array_section("items"); | |
1876 | const int max_pos = crush->get_bucket_size(parent.id); | |
1877 | for (int pos = 0; pos < max_pos; pos++) { | |
1878 | int id = crush->get_bucket_item(parent.id, pos); | |
1879 | float weight = crush->get_bucket_item_weightf(parent.id, pos); | |
1880 | dump_item(Item(id, parent.depth + 1, weight), f); | |
1881 | } | |
1882 | f->close_section(); | |
1883 | } | |
1884 | }; | |
1885 | } | |
1886 | ||
1887 | void CrushWrapper::dump_tree(Formatter *f) const | |
1888 | { | |
1889 | assert(f); | |
1890 | TreeDumper(this).dump(f); | |
1891 | } | |
1892 | ||
1893 | void CrushWrapper::dump_tunables(Formatter *f) const | |
1894 | { | |
1895 | f->dump_int("choose_local_tries", get_choose_local_tries()); | |
1896 | f->dump_int("choose_local_fallback_tries", get_choose_local_fallback_tries()); | |
1897 | f->dump_int("choose_total_tries", get_choose_total_tries()); | |
1898 | f->dump_int("chooseleaf_descend_once", get_chooseleaf_descend_once()); | |
1899 | f->dump_int("chooseleaf_vary_r", get_chooseleaf_vary_r()); | |
1900 | f->dump_int("chooseleaf_stable", get_chooseleaf_stable()); | |
1901 | f->dump_int("straw_calc_version", get_straw_calc_version()); | |
1902 | f->dump_int("allowed_bucket_algs", get_allowed_bucket_algs()); | |
1903 | ||
1904 | // be helpful about it | |
1905 | if (has_jewel_tunables()) | |
1906 | f->dump_string("profile", "jewel"); | |
1907 | else if (has_hammer_tunables()) | |
1908 | f->dump_string("profile", "hammer"); | |
1909 | else if (has_firefly_tunables()) | |
1910 | f->dump_string("profile", "firefly"); | |
1911 | else if (has_bobtail_tunables()) | |
1912 | f->dump_string("profile", "bobtail"); | |
1913 | else if (has_argonaut_tunables()) | |
1914 | f->dump_string("profile", "argonaut"); | |
1915 | else | |
1916 | f->dump_string("profile", "unknown"); | |
1917 | f->dump_int("optimal_tunables", (int)has_optimal_tunables()); | |
1918 | f->dump_int("legacy_tunables", (int)has_legacy_tunables()); | |
1919 | ||
1920 | // be helpful about minimum version required | |
1921 | f->dump_string("minimum_required_version", get_min_required_version()); | |
1922 | ||
1923 | f->dump_int("require_feature_tunables", (int)has_nondefault_tunables()); | |
1924 | f->dump_int("require_feature_tunables2", (int)has_nondefault_tunables2()); | |
1925 | f->dump_int("has_v2_rules", (int)has_v2_rules()); | |
1926 | f->dump_int("require_feature_tunables3", (int)has_nondefault_tunables3()); | |
1927 | f->dump_int("has_v3_rules", (int)has_v3_rules()); | |
1928 | f->dump_int("has_v4_buckets", (int)has_v4_buckets()); | |
1929 | f->dump_int("require_feature_tunables5", (int)has_nondefault_tunables5()); | |
1930 | f->dump_int("has_v5_rules", (int)has_v5_rules()); | |
1931 | } | |
1932 | ||
1933 | void CrushWrapper::dump_choose_args(Formatter *f) const | |
1934 | { | |
1935 | f->open_object_section("choose_args"); | |
1936 | for (auto c : choose_args) { | |
1937 | crush_choose_arg_map arg_map = c.second; | |
1938 | f->open_array_section(stringify(c.first).c_str()); | |
1939 | for (__u32 i = 0; i < arg_map.size; i++) { | |
1940 | crush_choose_arg *arg = &arg_map.args[i]; | |
1941 | if (arg->weight_set_size == 0 && | |
1942 | arg->ids_size == 0) | |
1943 | continue; | |
1944 | f->open_object_section("choose_args"); | |
1945 | int bucket_index = i; | |
1946 | f->dump_int("bucket_id", -1-bucket_index); | |
1947 | if (arg->weight_set_size > 0) { | |
1948 | f->open_array_section("weight_set"); | |
1949 | for (__u32 j = 0; j < arg->weight_set_size; j++) { | |
1950 | f->open_array_section("weights"); | |
1951 | __u32 *weights = arg->weight_set[j].weights; | |
1952 | __u32 size = arg->weight_set[j].size; | |
1953 | for (__u32 k = 0; k < size; k++) { | |
1954 | f->dump_float("weight", (float)weights[k]/(float)0x10000); | |
1955 | } | |
1956 | f->close_section(); | |
1957 | } | |
1958 | f->close_section(); | |
1959 | } | |
1960 | if (arg->ids_size > 0) { | |
1961 | f->open_array_section("ids"); | |
1962 | for (__u32 j = 0; j < arg->ids_size; j++) | |
1963 | f->dump_int("id", arg->ids[j]); | |
1964 | f->close_section(); | |
1965 | } | |
1966 | f->close_section(); | |
1967 | } | |
1968 | f->close_section(); | |
1969 | } | |
1970 | f->close_section(); | |
1971 | } | |
1972 | ||
1973 | void CrushWrapper::dump_rules(Formatter *f) const | |
1974 | { | |
1975 | for (int i=0; i<get_max_rules(); i++) { | |
1976 | if (!rule_exists(i)) | |
1977 | continue; | |
1978 | dump_rule(i, f); | |
1979 | } | |
1980 | } | |
1981 | ||
1982 | void CrushWrapper::dump_rule(int ruleset, Formatter *f) const | |
1983 | { | |
1984 | f->open_object_section("rule"); | |
1985 | f->dump_int("rule_id", ruleset); | |
1986 | if (get_rule_name(ruleset)) | |
1987 | f->dump_string("rule_name", get_rule_name(ruleset)); | |
1988 | f->dump_int("ruleset", get_rule_mask_ruleset(ruleset)); | |
1989 | f->dump_int("type", get_rule_mask_type(ruleset)); | |
1990 | f->dump_int("min_size", get_rule_mask_min_size(ruleset)); | |
1991 | f->dump_int("max_size", get_rule_mask_max_size(ruleset)); | |
1992 | f->open_array_section("steps"); | |
1993 | for (int j=0; j<get_rule_len(ruleset); j++) { | |
1994 | f->open_object_section("step"); | |
1995 | switch (get_rule_op(ruleset, j)) { | |
1996 | case CRUSH_RULE_NOOP: | |
1997 | f->dump_string("op", "noop"); | |
1998 | break; | |
1999 | case CRUSH_RULE_TAKE: | |
2000 | f->dump_string("op", "take"); | |
2001 | { | |
2002 | int item = get_rule_arg1(ruleset, j); | |
2003 | f->dump_int("item", item); | |
2004 | ||
2005 | const char *name = get_item_name(item); | |
2006 | f->dump_string("item_name", name ? name : ""); | |
2007 | } | |
2008 | break; | |
2009 | case CRUSH_RULE_EMIT: | |
2010 | f->dump_string("op", "emit"); | |
2011 | break; | |
2012 | case CRUSH_RULE_CHOOSE_FIRSTN: | |
2013 | f->dump_string("op", "choose_firstn"); | |
2014 | f->dump_int("num", get_rule_arg1(ruleset, j)); | |
2015 | f->dump_string("type", get_type_name(get_rule_arg2(ruleset, j))); | |
2016 | break; | |
2017 | case CRUSH_RULE_CHOOSE_INDEP: | |
2018 | f->dump_string("op", "choose_indep"); | |
2019 | f->dump_int("num", get_rule_arg1(ruleset, j)); | |
2020 | f->dump_string("type", get_type_name(get_rule_arg2(ruleset, j))); | |
2021 | break; | |
2022 | case CRUSH_RULE_CHOOSELEAF_FIRSTN: | |
2023 | f->dump_string("op", "chooseleaf_firstn"); | |
2024 | f->dump_int("num", get_rule_arg1(ruleset, j)); | |
2025 | f->dump_string("type", get_type_name(get_rule_arg2(ruleset, j))); | |
2026 | break; | |
2027 | case CRUSH_RULE_CHOOSELEAF_INDEP: | |
2028 | f->dump_string("op", "chooseleaf_indep"); | |
2029 | f->dump_int("num", get_rule_arg1(ruleset, j)); | |
2030 | f->dump_string("type", get_type_name(get_rule_arg2(ruleset, j))); | |
2031 | break; | |
2032 | case CRUSH_RULE_SET_CHOOSE_TRIES: | |
2033 | f->dump_string("op", "set_choose_tries"); | |
2034 | f->dump_int("num", get_rule_arg1(ruleset, j)); | |
2035 | break; | |
2036 | case CRUSH_RULE_SET_CHOOSELEAF_TRIES: | |
2037 | f->dump_string("op", "set_chooseleaf_tries"); | |
2038 | f->dump_int("num", get_rule_arg1(ruleset, j)); | |
2039 | break; | |
2040 | default: | |
2041 | f->dump_int("opcode", get_rule_op(ruleset, j)); | |
2042 | f->dump_int("arg1", get_rule_arg1(ruleset, j)); | |
2043 | f->dump_int("arg2", get_rule_arg2(ruleset, j)); | |
2044 | } | |
2045 | f->close_section(); | |
2046 | } | |
2047 | f->close_section(); | |
2048 | f->close_section(); | |
2049 | } | |
2050 | ||
2051 | void CrushWrapper::list_rules(Formatter *f) const | |
2052 | { | |
2053 | for (int rule = 0; rule < get_max_rules(); rule++) { | |
2054 | if (!rule_exists(rule)) | |
2055 | continue; | |
2056 | f->dump_string("name", get_rule_name(rule)); | |
2057 | } | |
2058 | } | |
2059 | ||
2060 | class CrushTreePlainDumper : public CrushTreeDumper::Dumper<ostream> { | |
2061 | public: | |
2062 | typedef CrushTreeDumper::Dumper<ostream> Parent; | |
2063 | ||
2064 | explicit CrushTreePlainDumper(const CrushWrapper *crush) | |
2065 | : Parent(crush) {} | |
2066 | ||
2067 | void dump(ostream *out) { | |
2068 | *out << "ID\tWEIGHT\tTYPE NAME\n"; | |
2069 | Parent::dump(out); | |
2070 | } | |
2071 | ||
2072 | protected: | |
2073 | void dump_item(const CrushTreeDumper::Item &qi, ostream *out) override { | |
2074 | *out << qi.id << "\t" | |
2075 | << weightf_t(qi.weight) << "\t"; | |
2076 | ||
2077 | for (int k=0; k < qi.depth; k++) | |
2078 | *out << "\t"; | |
2079 | ||
2080 | if (qi.is_bucket()) | |
2081 | { | |
2082 | *out << crush->get_type_name(crush->get_bucket_type(qi.id)) << " " | |
2083 | << crush->get_item_name(qi.id); | |
2084 | } | |
2085 | else | |
2086 | { | |
2087 | *out << "osd." << qi.id; | |
2088 | } | |
2089 | *out << "\n"; | |
2090 | } | |
2091 | }; | |
2092 | ||
2093 | ||
2094 | class CrushTreeFormattingDumper : public CrushTreeDumper::FormattingDumper { | |
2095 | public: | |
2096 | typedef CrushTreeDumper::FormattingDumper Parent; | |
2097 | ||
2098 | explicit CrushTreeFormattingDumper(const CrushWrapper *crush) | |
2099 | : Parent(crush) {} | |
2100 | ||
2101 | void dump(Formatter *f) { | |
2102 | f->open_array_section("nodes"); | |
2103 | Parent::dump(f); | |
2104 | f->close_section(); | |
2105 | f->open_array_section("stray"); | |
2106 | f->close_section(); | |
2107 | } | |
2108 | }; | |
2109 | ||
2110 | ||
2111 | void CrushWrapper::dump_tree(ostream *out, Formatter *f) const | |
2112 | { | |
2113 | if (out) | |
2114 | CrushTreePlainDumper(this).dump(out); | |
2115 | if (f) | |
2116 | CrushTreeFormattingDumper(this).dump(f); | |
2117 | } | |
2118 | ||
2119 | void CrushWrapper::generate_test_instances(list<CrushWrapper*>& o) | |
2120 | { | |
2121 | o.push_back(new CrushWrapper); | |
2122 | // fixme | |
2123 | } | |
2124 | ||
2125 | int CrushWrapper::_get_osd_pool_default_crush_replicated_ruleset(CephContext *cct, | |
2126 | bool quiet) | |
2127 | { | |
2128 | int crush_ruleset = cct->_conf->osd_pool_default_crush_rule; | |
2129 | if (crush_ruleset == -1) { | |
2130 | crush_ruleset = cct->_conf->osd_pool_default_crush_replicated_ruleset; | |
2131 | } else if (!quiet) { | |
2132 | ldout(cct, 0) << "osd_pool_default_crush_rule is deprecated " | |
2133 | << "use osd_pool_default_crush_replicated_ruleset instead" | |
2134 | << dendl; | |
2135 | ldout(cct, 0) << "osd_pool_default_crush_rule = " | |
2136 | << cct->_conf-> osd_pool_default_crush_rule << " overrides " | |
2137 | << "osd_pool_default_crush_replicated_ruleset = " | |
2138 | << cct->_conf->osd_pool_default_crush_replicated_ruleset | |
2139 | << dendl; | |
2140 | } | |
2141 | ||
2142 | return crush_ruleset; | |
2143 | } | |
2144 | ||
2145 | /** | |
2146 | * Determine the default CRUSH ruleset ID to be used with | |
2147 | * newly created replicated pools. | |
2148 | * | |
2149 | * @returns a ruleset ID (>=0) or -1 if no suitable ruleset found | |
2150 | */ | |
2151 | int CrushWrapper::get_osd_pool_default_crush_replicated_ruleset(CephContext *cct) | |
2152 | { | |
2153 | int crush_ruleset = _get_osd_pool_default_crush_replicated_ruleset(cct, | |
2154 | false); | |
2155 | if (crush_ruleset == CEPH_DEFAULT_CRUSH_REPLICATED_RULESET) { | |
2156 | crush_ruleset = find_first_ruleset(pg_pool_t::TYPE_REPLICATED); | |
2157 | } else if (!ruleset_exists(crush_ruleset)) { | |
2158 | crush_ruleset = -1; // match find_first_ruleset() retval | |
2159 | } | |
2160 | ||
2161 | return crush_ruleset; | |
2162 | } | |
2163 | ||
2164 | bool CrushWrapper::is_valid_crush_name(const string& s) | |
2165 | { | |
2166 | if (s.empty()) | |
2167 | return false; | |
2168 | for (string::const_iterator p = s.begin(); p != s.end(); ++p) { | |
2169 | if (!(*p == '-') && | |
2170 | !(*p == '_') && | |
2171 | !(*p == '.') && | |
2172 | !(*p >= '0' && *p <= '9') && | |
2173 | !(*p >= 'A' && *p <= 'Z') && | |
2174 | !(*p >= 'a' && *p <= 'z')) | |
2175 | return false; | |
2176 | } | |
2177 | return true; | |
2178 | } | |
2179 | ||
2180 | bool CrushWrapper::is_valid_crush_loc(CephContext *cct, | |
2181 | const map<string,string>& loc) | |
2182 | { | |
2183 | for (map<string,string>::const_iterator l = loc.begin(); l != loc.end(); ++l) { | |
2184 | if (!is_valid_crush_name(l->first) || | |
2185 | !is_valid_crush_name(l->second)) { | |
2186 | ldout(cct, 1) << "loc[" | |
2187 | << l->first << "] = '" | |
2188 | << l->second << "' not a valid crush name ([A-Za-z0-9_-.]+)" | |
2189 | << dendl; | |
2190 | return false; | |
2191 | } | |
2192 | } | |
2193 | return true; | |
2194 | } | |
2195 | ||
2196 | int CrushWrapper::_choose_type_stack( | |
2197 | CephContext *cct, | |
2198 | const vector<pair<int,int>>& stack, | |
2199 | const set<int>& overfull, | |
2200 | const vector<int>& underfull, | |
2201 | const vector<int>& orig, | |
2202 | vector<int>::const_iterator& i, | |
2203 | set<int>& used, | |
2204 | vector<int> *pw) const | |
2205 | { | |
2206 | vector<int> w = *pw; | |
2207 | vector<int> o; | |
2208 | ||
2209 | ldout(cct, 10) << __func__ << " stack " << stack | |
2210 | << " orig " << orig | |
2211 | << " at " << *i | |
2212 | << " pw " << *pw | |
2213 | << dendl; | |
2214 | ||
2215 | vector<int> cumulative_fanout(stack.size()); | |
2216 | int f = 1; | |
2217 | for (int j = (int)stack.size() - 1; j >= 0; --j) { | |
2218 | cumulative_fanout[j] = f; | |
2219 | f *= stack[j].second; | |
2220 | } | |
2221 | ldout(cct, 10) << __func__ << " cumulative_fanout " << cumulative_fanout | |
2222 | << dendl; | |
2223 | ||
2224 | for (unsigned j = 0; j < stack.size(); ++j) { | |
2225 | int type = stack[j].first; | |
2226 | int fanout = stack[j].second; | |
2227 | int cum_fanout = cumulative_fanout[j]; | |
2228 | ldout(cct, 10) << " level " << j << ": type " << type << " fanout " << fanout | |
2229 | << " cumulative " << cum_fanout | |
2230 | << " w " << w << dendl; | |
2231 | vector<int> o; | |
2232 | auto tmpi = i; | |
2233 | for (auto from : w) { | |
2234 | ldout(cct, 10) << " from " << from << dendl; | |
2235 | ||
2236 | for (int pos = 0; pos < fanout; ++pos) { | |
2237 | if (type > 0) { | |
2238 | // non-leaf | |
2239 | int item = *tmpi; | |
2240 | do { | |
2241 | int r = get_immediate_parent_id(item, &item); | |
2242 | if (r < 0) { | |
2243 | ldout(cct, 10) << __func__ << " parent of " << item << " got " | |
2244 | << cpp_strerror(r) << dendl; | |
2245 | return -EINVAL; | |
2246 | } | |
2247 | } while (get_bucket_type(item) != type); | |
2248 | o.push_back(item); | |
2249 | ldout(cct, 10) << __func__ << " from " << *tmpi << " got " << item | |
2250 | << " of type " << type << dendl; | |
2251 | int n = cum_fanout; | |
2252 | while (n-- && tmpi != orig.end()) | |
2253 | ++tmpi; | |
2254 | } else { | |
2255 | // leaf | |
2256 | bool replaced = false; | |
2257 | if (overfull.count(*i)) { | |
2258 | for (auto item : underfull) { | |
2259 | ldout(cct, 10) << __func__ << " pos " << pos | |
2260 | << " was " << *i << " considering " << item | |
2261 | << dendl; | |
2262 | if (used.count(item)) { | |
2263 | ldout(cct, 20) << __func__ << " in used " << used << dendl; | |
2264 | continue; | |
2265 | } | |
2266 | if (!subtree_contains(from, item)) { | |
2267 | ldout(cct, 20) << __func__ << " not in subtree " << from << dendl; | |
2268 | continue; | |
2269 | } | |
2270 | if (std::find(orig.begin(), orig.end(), item) != orig.end()) { | |
2271 | ldout(cct, 20) << __func__ << " in orig " << orig << dendl; | |
2272 | continue; | |
2273 | } | |
2274 | o.push_back(item); | |
2275 | used.insert(item); | |
2276 | ldout(cct, 10) << __func__ << " pos " << pos << " replace " | |
2277 | << *i << " -> " << item << dendl; | |
2278 | replaced = true; | |
2279 | ++i; | |
2280 | break; | |
2281 | } | |
2282 | } | |
2283 | if (!replaced) { | |
2284 | ldout(cct, 10) << __func__ << " pos " << pos << " keep " << *i | |
2285 | << dendl; | |
2286 | o.push_back(*i); | |
2287 | ++i; | |
2288 | } | |
2289 | if (i == orig.end()) { | |
2290 | ldout(cct, 10) << __func__ << " end of orig, break 1" << dendl; | |
2291 | break; | |
2292 | } | |
2293 | } | |
2294 | } | |
2295 | if (i == orig.end()) { | |
2296 | ldout(cct, 10) << __func__ << " end of orig, break 2" << dendl; | |
2297 | break; | |
2298 | } | |
2299 | } | |
2300 | ldout(cct, 10) << __func__ << " w <- " << o << " was " << w << dendl; | |
2301 | w.swap(o); | |
2302 | } | |
2303 | *pw = w; | |
2304 | return 0; | |
2305 | } | |
2306 | ||
2307 | int CrushWrapper::try_remap_rule( | |
2308 | CephContext *cct, | |
2309 | int ruleno, | |
2310 | int maxout, | |
2311 | const set<int>& overfull, | |
2312 | const vector<int>& underfull, | |
2313 | const vector<int>& orig, | |
2314 | vector<int> *out) const | |
2315 | { | |
2316 | const crush_map *map = crush; | |
2317 | const crush_rule *rule = get_rule(ruleno); | |
2318 | assert(rule); | |
2319 | ||
2320 | ldout(cct, 10) << __func__ << " ruleno " << ruleno | |
2321 | << " numrep " << maxout << " overfull " << overfull | |
2322 | << " underfull " << underfull << " orig " << orig | |
2323 | << dendl; | |
2324 | vector<int> w; // working set | |
2325 | out->clear(); | |
2326 | ||
2327 | auto i = orig.begin(); | |
2328 | set<int> used; | |
2329 | ||
2330 | vector<pair<int,int>> type_stack; // (type, fan-out) | |
2331 | ||
2332 | for (unsigned step = 0; step < rule->len; ++step) { | |
2333 | const crush_rule_step *curstep = &rule->steps[step]; | |
2334 | ldout(cct, 10) << __func__ << " step " << step << " w " << w << dendl; | |
2335 | switch (curstep->op) { | |
2336 | case CRUSH_RULE_TAKE: | |
2337 | if ((curstep->arg1 >= 0 && curstep->arg1 < map->max_devices) || | |
2338 | (-1-curstep->arg1 >= 0 && -1-curstep->arg1 < map->max_buckets && | |
2339 | map->buckets[-1-curstep->arg1])) { | |
2340 | w.clear(); | |
2341 | w.push_back(curstep->arg1); | |
2342 | ldout(cct, 10) << __func__ << " take " << w << dendl; | |
2343 | } else { | |
2344 | ldout(cct, 1) << " bad take value " << curstep->arg1 << dendl; | |
2345 | } | |
2346 | break; | |
2347 | ||
2348 | case CRUSH_RULE_CHOOSELEAF_FIRSTN: | |
2349 | case CRUSH_RULE_CHOOSELEAF_INDEP: | |
2350 | { | |
2351 | int numrep = curstep->arg1; | |
2352 | int type = curstep->arg2; | |
2353 | if (numrep <= 0) | |
2354 | numrep += maxout; | |
2355 | type_stack.push_back(make_pair(type, numrep)); | |
2356 | type_stack.push_back(make_pair(0, 1)); | |
2357 | int r = _choose_type_stack(cct, type_stack, overfull, underfull, orig, | |
2358 | i, used, &w); | |
2359 | if (r < 0) | |
2360 | return r; | |
2361 | type_stack.clear(); | |
2362 | } | |
2363 | break; | |
2364 | ||
2365 | case CRUSH_RULE_CHOOSE_FIRSTN: | |
2366 | case CRUSH_RULE_CHOOSE_INDEP: | |
2367 | { | |
2368 | int numrep = curstep->arg1; | |
2369 | int type = curstep->arg2; | |
2370 | if (numrep <= 0) | |
2371 | numrep += maxout; | |
2372 | type_stack.push_back(make_pair(type, numrep)); | |
2373 | } | |
2374 | break; | |
2375 | ||
2376 | case CRUSH_RULE_EMIT: | |
2377 | ldout(cct, 10) << " emit " << w << dendl; | |
2378 | if (!type_stack.empty()) { | |
2379 | int r = _choose_type_stack(cct, type_stack, overfull, underfull, orig, | |
2380 | i, used, &w); | |
2381 | if (r < 0) | |
2382 | return r; | |
2383 | type_stack.clear(); | |
2384 | } | |
2385 | for (auto item : w) { | |
2386 | out->push_back(item); | |
2387 | } | |
2388 | w.clear(); | |
2389 | break; | |
2390 | ||
2391 | default: | |
2392 | // ignore | |
2393 | break; | |
2394 | } | |
2395 | } | |
2396 | ||
2397 | return 0; | |
2398 | } |