blob: aade4a5c1c07f6ab0ca0f1ee66dfacbded110aed [file] [log] [blame]
Sage Weilf24e9982009-10-06 11:31:10 -07001
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07002#include <linux/ceph/ceph_debug.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09003
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07004#include <linux/module.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09005#include <linux/slab.h>
Sage Weilf24e9982009-10-06 11:31:10 -07006#include <asm/div64.h>
7
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07008#include <linux/ceph/libceph.h>
9#include <linux/ceph/osdmap.h>
10#include <linux/ceph/decode.h>
11#include <linux/crush/hash.h>
12#include <linux/crush/mapper.h>
Sage Weilf24e9982009-10-06 11:31:10 -070013
14char *ceph_osdmap_state_str(char *str, int len, int state)
15{
Sage Weilf24e9982009-10-06 11:31:10 -070016 if (!len)
Cong Ding1ec39112013-01-25 17:48:59 -060017 return str;
Sage Weilf24e9982009-10-06 11:31:10 -070018
Cong Ding1ec39112013-01-25 17:48:59 -060019 if ((state & CEPH_OSD_EXISTS) && (state & CEPH_OSD_UP))
20 snprintf(str, len, "exists, up");
21 else if (state & CEPH_OSD_EXISTS)
22 snprintf(str, len, "exists");
23 else if (state & CEPH_OSD_UP)
24 snprintf(str, len, "up");
25 else
Sage Weilf24e9982009-10-06 11:31:10 -070026 snprintf(str, len, "doesn't exist");
Cong Ding1ec39112013-01-25 17:48:59 -060027
Sage Weilf24e9982009-10-06 11:31:10 -070028 return str;
29}
30
31/* maps */
32
Eric Dumazet95c96172012-04-15 05:58:06 +000033static int calc_bits_of(unsigned int t)
Sage Weilf24e9982009-10-06 11:31:10 -070034{
35 int b = 0;
36 while (t) {
37 t = t >> 1;
38 b++;
39 }
40 return b;
41}
42
43/*
44 * the foo_mask is the smallest value 2^n-1 that is >= foo.
45 */
46static void calc_pg_masks(struct ceph_pg_pool_info *pi)
47{
Sage Weil4f6a7e52013-02-23 10:41:09 -080048 pi->pg_num_mask = (1 << calc_bits_of(pi->pg_num-1)) - 1;
49 pi->pgp_num_mask = (1 << calc_bits_of(pi->pgp_num-1)) - 1;
Sage Weilf24e9982009-10-06 11:31:10 -070050}
51
52/*
53 * decode crush map
54 */
55static int crush_decode_uniform_bucket(void **p, void *end,
56 struct crush_bucket_uniform *b)
57{
58 dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
59 ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -070060 b->item_weight = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -070061 return 0;
62bad:
63 return -EINVAL;
64}
65
66static int crush_decode_list_bucket(void **p, void *end,
67 struct crush_bucket_list *b)
68{
69 int j;
70 dout("crush_decode_list_bucket %p to %p\n", *p, end);
71 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
72 if (b->item_weights == NULL)
73 return -ENOMEM;
74 b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
75 if (b->sum_weights == NULL)
76 return -ENOMEM;
77 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
78 for (j = 0; j < b->h.size; j++) {
Sage Weilc89136e2009-10-14 09:59:09 -070079 b->item_weights[j] = ceph_decode_32(p);
80 b->sum_weights[j] = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -070081 }
82 return 0;
83bad:
84 return -EINVAL;
85}
86
87static int crush_decode_tree_bucket(void **p, void *end,
88 struct crush_bucket_tree *b)
89{
90 int j;
91 dout("crush_decode_tree_bucket %p to %p\n", *p, end);
92 ceph_decode_32_safe(p, end, b->num_nodes, bad);
93 b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
94 if (b->node_weights == NULL)
95 return -ENOMEM;
96 ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
97 for (j = 0; j < b->num_nodes; j++)
Sage Weilc89136e2009-10-14 09:59:09 -070098 b->node_weights[j] = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -070099 return 0;
100bad:
101 return -EINVAL;
102}
103
104static int crush_decode_straw_bucket(void **p, void *end,
105 struct crush_bucket_straw *b)
106{
107 int j;
108 dout("crush_decode_straw_bucket %p to %p\n", *p, end);
109 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
110 if (b->item_weights == NULL)
111 return -ENOMEM;
112 b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
113 if (b->straws == NULL)
114 return -ENOMEM;
115 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
116 for (j = 0; j < b->h.size; j++) {
Sage Weilc89136e2009-10-14 09:59:09 -0700117 b->item_weights[j] = ceph_decode_32(p);
118 b->straws[j] = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -0700119 }
120 return 0;
121bad:
122 return -EINVAL;
123}
124
Sage Weil546f04e2012-07-30 18:15:23 -0700125static int skip_name_map(void **p, void *end)
126{
127 int len;
128 ceph_decode_32_safe(p, end, len ,bad);
129 while (len--) {
130 int strlen;
131 *p += sizeof(u32);
132 ceph_decode_32_safe(p, end, strlen, bad);
133 *p += strlen;
134}
135 return 0;
136bad:
137 return -EINVAL;
138}
139
Sage Weilf24e9982009-10-06 11:31:10 -0700140static struct crush_map *crush_decode(void *pbyval, void *end)
141{
142 struct crush_map *c;
143 int err = -EINVAL;
144 int i, j;
145 void **p = &pbyval;
146 void *start = pbyval;
147 u32 magic;
Sage Weil546f04e2012-07-30 18:15:23 -0700148 u32 num_name_maps;
Sage Weilf24e9982009-10-06 11:31:10 -0700149
150 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
151
152 c = kzalloc(sizeof(*c), GFP_NOFS);
153 if (c == NULL)
154 return ERR_PTR(-ENOMEM);
155
Sage Weil546f04e2012-07-30 18:15:23 -0700156 /* set tunables to default values */
157 c->choose_local_tries = 2;
158 c->choose_local_fallback_tries = 5;
159 c->choose_total_tries = 19;
Jim Schutt1604f482012-11-30 09:15:25 -0700160 c->chooseleaf_descend_once = 0;
Sage Weil546f04e2012-07-30 18:15:23 -0700161
Sage Weilf24e9982009-10-06 11:31:10 -0700162 ceph_decode_need(p, end, 4*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -0700163 magic = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -0700164 if (magic != CRUSH_MAGIC) {
165 pr_err("crush_decode magic %x != current %x\n",
Eric Dumazet95c96172012-04-15 05:58:06 +0000166 (unsigned int)magic, (unsigned int)CRUSH_MAGIC);
Sage Weilf24e9982009-10-06 11:31:10 -0700167 goto bad;
168 }
Sage Weilc89136e2009-10-14 09:59:09 -0700169 c->max_buckets = ceph_decode_32(p);
170 c->max_rules = ceph_decode_32(p);
171 c->max_devices = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -0700172
Sage Weilf24e9982009-10-06 11:31:10 -0700173 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
174 if (c->buckets == NULL)
175 goto badmem;
176 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
177 if (c->rules == NULL)
178 goto badmem;
179
180 /* buckets */
181 for (i = 0; i < c->max_buckets; i++) {
182 int size = 0;
183 u32 alg;
184 struct crush_bucket *b;
185
186 ceph_decode_32_safe(p, end, alg, bad);
187 if (alg == 0) {
188 c->buckets[i] = NULL;
189 continue;
190 }
191 dout("crush_decode bucket %d off %x %p to %p\n",
192 i, (int)(*p-start), *p, end);
193
194 switch (alg) {
195 case CRUSH_BUCKET_UNIFORM:
196 size = sizeof(struct crush_bucket_uniform);
197 break;
198 case CRUSH_BUCKET_LIST:
199 size = sizeof(struct crush_bucket_list);
200 break;
201 case CRUSH_BUCKET_TREE:
202 size = sizeof(struct crush_bucket_tree);
203 break;
204 case CRUSH_BUCKET_STRAW:
205 size = sizeof(struct crush_bucket_straw);
206 break;
207 default:
Sage Weil30dc6382009-12-21 14:49:37 -0800208 err = -EINVAL;
Sage Weilf24e9982009-10-06 11:31:10 -0700209 goto bad;
210 }
211 BUG_ON(size == 0);
212 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
213 if (b == NULL)
214 goto badmem;
215
216 ceph_decode_need(p, end, 4*sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -0700217 b->id = ceph_decode_32(p);
218 b->type = ceph_decode_16(p);
Sage Weilfb690392009-11-07 20:18:22 -0800219 b->alg = ceph_decode_8(p);
220 b->hash = ceph_decode_8(p);
Sage Weilc89136e2009-10-14 09:59:09 -0700221 b->weight = ceph_decode_32(p);
222 b->size = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -0700223
224 dout("crush_decode bucket size %d off %x %p to %p\n",
225 b->size, (int)(*p-start), *p, end);
226
227 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
228 if (b->items == NULL)
229 goto badmem;
230 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
231 if (b->perm == NULL)
232 goto badmem;
233 b->perm_n = 0;
234
235 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
236 for (j = 0; j < b->size; j++)
Sage Weilc89136e2009-10-14 09:59:09 -0700237 b->items[j] = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -0700238
239 switch (b->alg) {
240 case CRUSH_BUCKET_UNIFORM:
241 err = crush_decode_uniform_bucket(p, end,
242 (struct crush_bucket_uniform *)b);
243 if (err < 0)
244 goto bad;
245 break;
246 case CRUSH_BUCKET_LIST:
247 err = crush_decode_list_bucket(p, end,
248 (struct crush_bucket_list *)b);
249 if (err < 0)
250 goto bad;
251 break;
252 case CRUSH_BUCKET_TREE:
253 err = crush_decode_tree_bucket(p, end,
254 (struct crush_bucket_tree *)b);
255 if (err < 0)
256 goto bad;
257 break;
258 case CRUSH_BUCKET_STRAW:
259 err = crush_decode_straw_bucket(p, end,
260 (struct crush_bucket_straw *)b);
261 if (err < 0)
262 goto bad;
263 break;
264 }
265 }
266
267 /* rules */
268 dout("rule vec is %p\n", c->rules);
269 for (i = 0; i < c->max_rules; i++) {
270 u32 yes;
271 struct crush_rule *r;
272
273 ceph_decode_32_safe(p, end, yes, bad);
274 if (!yes) {
275 dout("crush_decode NO rule %d off %x %p to %p\n",
276 i, (int)(*p-start), *p, end);
277 c->rules[i] = NULL;
278 continue;
279 }
280
281 dout("crush_decode rule %d off %x %p to %p\n",
282 i, (int)(*p-start), *p, end);
283
284 /* len */
285 ceph_decode_32_safe(p, end, yes, bad);
286#if BITS_PER_LONG == 32
Sage Weil30dc6382009-12-21 14:49:37 -0800287 err = -EINVAL;
Xi Wang64486692012-02-16 11:55:48 -0500288 if (yes > (ULONG_MAX - sizeof(*r))
289 / sizeof(struct crush_rule_step))
Sage Weilf24e9982009-10-06 11:31:10 -0700290 goto bad;
291#endif
292 r = c->rules[i] = kmalloc(sizeof(*r) +
293 yes*sizeof(struct crush_rule_step),
294 GFP_NOFS);
295 if (r == NULL)
296 goto badmem;
297 dout(" rule %d is at %p\n", i, r);
298 r->len = yes;
299 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
300 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
301 for (j = 0; j < r->len; j++) {
Sage Weilc89136e2009-10-14 09:59:09 -0700302 r->steps[j].op = ceph_decode_32(p);
303 r->steps[j].arg1 = ceph_decode_32(p);
304 r->steps[j].arg2 = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -0700305 }
306 }
307
308 /* ignore trailing name maps. */
Sage Weil546f04e2012-07-30 18:15:23 -0700309 for (num_name_maps = 0; num_name_maps < 3; num_name_maps++) {
310 err = skip_name_map(p, end);
311 if (err < 0)
312 goto done;
313 }
Sage Weilf24e9982009-10-06 11:31:10 -0700314
Sage Weil546f04e2012-07-30 18:15:23 -0700315 /* tunables */
316 ceph_decode_need(p, end, 3*sizeof(u32), done);
317 c->choose_local_tries = ceph_decode_32(p);
318 c->choose_local_fallback_tries = ceph_decode_32(p);
319 c->choose_total_tries = ceph_decode_32(p);
320 dout("crush decode tunable choose_local_tries = %d",
321 c->choose_local_tries);
322 dout("crush decode tunable choose_local_fallback_tries = %d",
323 c->choose_local_fallback_tries);
324 dout("crush decode tunable choose_total_tries = %d",
325 c->choose_total_tries);
326
Jim Schutt1604f482012-11-30 09:15:25 -0700327 ceph_decode_need(p, end, sizeof(u32), done);
328 c->chooseleaf_descend_once = ceph_decode_32(p);
329 dout("crush decode tunable chooseleaf_descend_once = %d",
330 c->chooseleaf_descend_once);
331
Sage Weil546f04e2012-07-30 18:15:23 -0700332done:
Sage Weilf24e9982009-10-06 11:31:10 -0700333 dout("crush_decode success\n");
334 return c;
335
336badmem:
337 err = -ENOMEM;
338bad:
339 dout("crush_decode fail %d\n", err);
340 crush_destroy(c);
341 return ERR_PTR(err);
342}
343
Sage Weilf24e9982009-10-06 11:31:10 -0700344/*
Sage Weil9794b142010-02-16 15:53:32 -0800345 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
346 * to a set of osds)
Sage Weilf24e9982009-10-06 11:31:10 -0700347 */
Sage Weil5b191d92013-02-23 10:38:16 -0800348static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
Sage Weil51042122009-11-04 11:39:12 -0800349{
Sage Weil5b191d92013-02-23 10:38:16 -0800350 if (l.pool < r.pool)
Sage Weil51042122009-11-04 11:39:12 -0800351 return -1;
Sage Weil5b191d92013-02-23 10:38:16 -0800352 if (l.pool > r.pool)
353 return 1;
354 if (l.seed < r.seed)
355 return -1;
356 if (l.seed > r.seed)
Sage Weil51042122009-11-04 11:39:12 -0800357 return 1;
358 return 0;
359}
360
Sage Weil991abb62009-10-08 22:22:37 -0700361static int __insert_pg_mapping(struct ceph_pg_mapping *new,
362 struct rb_root *root)
Sage Weilf24e9982009-10-06 11:31:10 -0700363{
364 struct rb_node **p = &root->rb_node;
365 struct rb_node *parent = NULL;
366 struct ceph_pg_mapping *pg = NULL;
Sage Weil51042122009-11-04 11:39:12 -0800367 int c;
Sage Weilf24e9982009-10-06 11:31:10 -0700368
Sage Weil8adc8b32011-09-28 10:11:04 -0700369 dout("__insert_pg_mapping %llx %p\n", *(u64 *)&new->pgid, new);
Sage Weilf24e9982009-10-06 11:31:10 -0700370 while (*p) {
371 parent = *p;
372 pg = rb_entry(parent, struct ceph_pg_mapping, node);
Sage Weil51042122009-11-04 11:39:12 -0800373 c = pgid_cmp(new->pgid, pg->pgid);
374 if (c < 0)
Sage Weilf24e9982009-10-06 11:31:10 -0700375 p = &(*p)->rb_left;
Sage Weil51042122009-11-04 11:39:12 -0800376 else if (c > 0)
Sage Weilf24e9982009-10-06 11:31:10 -0700377 p = &(*p)->rb_right;
378 else
Sage Weil991abb62009-10-08 22:22:37 -0700379 return -EEXIST;
Sage Weilf24e9982009-10-06 11:31:10 -0700380 }
381
382 rb_link_node(&new->node, parent, p);
383 rb_insert_color(&new->node, root);
Sage Weil991abb62009-10-08 22:22:37 -0700384 return 0;
Sage Weilf24e9982009-10-06 11:31:10 -0700385}
386
Sage Weil9794b142010-02-16 15:53:32 -0800387static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
Sage Weil5b191d92013-02-23 10:38:16 -0800388 struct ceph_pg pgid)
Sage Weil9794b142010-02-16 15:53:32 -0800389{
390 struct rb_node *n = root->rb_node;
391 struct ceph_pg_mapping *pg;
392 int c;
393
394 while (n) {
395 pg = rb_entry(n, struct ceph_pg_mapping, node);
396 c = pgid_cmp(pgid, pg->pgid);
Sage Weil8adc8b32011-09-28 10:11:04 -0700397 if (c < 0) {
Sage Weil9794b142010-02-16 15:53:32 -0800398 n = n->rb_left;
Sage Weil8adc8b32011-09-28 10:11:04 -0700399 } else if (c > 0) {
Sage Weil9794b142010-02-16 15:53:32 -0800400 n = n->rb_right;
Sage Weil8adc8b32011-09-28 10:11:04 -0700401 } else {
Sage Weil5b191d92013-02-23 10:38:16 -0800402 dout("__lookup_pg_mapping %lld.%x got %p\n",
403 pgid.pool, pgid.seed, pg);
Sage Weil9794b142010-02-16 15:53:32 -0800404 return pg;
Sage Weil8adc8b32011-09-28 10:11:04 -0700405 }
Sage Weil9794b142010-02-16 15:53:32 -0800406 }
407 return NULL;
408}
409
Sage Weil5b191d92013-02-23 10:38:16 -0800410static int __remove_pg_mapping(struct rb_root *root, struct ceph_pg pgid)
Sage Weil8adc8b32011-09-28 10:11:04 -0700411{
412 struct ceph_pg_mapping *pg = __lookup_pg_mapping(root, pgid);
413
414 if (pg) {
Sage Weil5b191d92013-02-23 10:38:16 -0800415 dout("__remove_pg_mapping %lld.%x %p\n", pgid.pool, pgid.seed,
416 pg);
Sage Weil8adc8b32011-09-28 10:11:04 -0700417 rb_erase(&pg->node, root);
418 kfree(pg);
419 return 0;
420 }
Sage Weil5b191d92013-02-23 10:38:16 -0800421 dout("__remove_pg_mapping %lld.%x dne\n", pgid.pool, pgid.seed);
Sage Weil8adc8b32011-09-28 10:11:04 -0700422 return -ENOENT;
423}
424
Sage Weilf24e9982009-10-06 11:31:10 -0700425/*
Sage Weil4fc51be2010-02-16 15:55:03 -0800426 * rbtree of pg pool info
427 */
428static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
429{
430 struct rb_node **p = &root->rb_node;
431 struct rb_node *parent = NULL;
432 struct ceph_pg_pool_info *pi = NULL;
433
434 while (*p) {
435 parent = *p;
436 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
437 if (new->id < pi->id)
438 p = &(*p)->rb_left;
439 else if (new->id > pi->id)
440 p = &(*p)->rb_right;
441 else
442 return -EEXIST;
443 }
444
445 rb_link_node(&new->node, parent, p);
446 rb_insert_color(&new->node, root);
447 return 0;
448}
449
Sage Weil4f6a7e52013-02-23 10:41:09 -0800450static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, u64 id)
Sage Weil4fc51be2010-02-16 15:55:03 -0800451{
452 struct ceph_pg_pool_info *pi;
453 struct rb_node *n = root->rb_node;
454
455 while (n) {
456 pi = rb_entry(n, struct ceph_pg_pool_info, node);
457 if (id < pi->id)
458 n = n->rb_left;
459 else if (id > pi->id)
460 n = n->rb_right;
461 else
462 return pi;
463 }
464 return NULL;
465}
466
Ilya Dryomovce7f6a22014-01-27 17:40:19 +0200467struct ceph_pg_pool_info *ceph_pg_pool_by_id(struct ceph_osdmap *map, u64 id)
468{
469 return __lookup_pg_pool(&map->pg_pools, id);
470}
471
Alex Elder72afc712012-10-30 19:40:33 -0500472const char *ceph_pg_pool_name_by_id(struct ceph_osdmap *map, u64 id)
473{
474 struct ceph_pg_pool_info *pi;
475
476 if (id == CEPH_NOPOOL)
477 return NULL;
478
479 if (WARN_ON_ONCE(id > (u64) INT_MAX))
480 return NULL;
481
482 pi = __lookup_pg_pool(&map->pg_pools, (int) id);
483
484 return pi ? pi->name : NULL;
485}
486EXPORT_SYMBOL(ceph_pg_pool_name_by_id);
487
Yehuda Sadeh7669a2c2010-05-17 12:31:35 -0700488int ceph_pg_poolid_by_name(struct ceph_osdmap *map, const char *name)
489{
490 struct rb_node *rbp;
491
492 for (rbp = rb_first(&map->pg_pools); rbp; rbp = rb_next(rbp)) {
493 struct ceph_pg_pool_info *pi =
494 rb_entry(rbp, struct ceph_pg_pool_info, node);
495 if (pi->name && strcmp(pi->name, name) == 0)
496 return pi->id;
497 }
498 return -ENOENT;
499}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700500EXPORT_SYMBOL(ceph_pg_poolid_by_name);
Yehuda Sadeh7669a2c2010-05-17 12:31:35 -0700501
Sage Weil2844a762010-04-09 15:46:42 -0700502static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi)
503{
504 rb_erase(&pi->node, root);
505 kfree(pi->name);
506 kfree(pi);
507}
508
Sage Weil73a7e692010-08-02 11:00:55 -0700509static int __decode_pool(void **p, void *end, struct ceph_pg_pool_info *pi)
Sage Weilefd75762010-03-17 10:05:28 -0700510{
Sage Weil4f6a7e52013-02-23 10:41:09 -0800511 u8 ev, cv;
512 unsigned len, num;
513 void *pool_end;
Sage Weil73a7e692010-08-02 11:00:55 -0700514
Sage Weil4f6a7e52013-02-23 10:41:09 -0800515 ceph_decode_need(p, end, 2 + 4, bad);
516 ev = ceph_decode_8(p); /* encoding version */
517 cv = ceph_decode_8(p); /* compat version */
518 if (ev < 5) {
519 pr_warning("got v %d < 5 cv %d of ceph_pg_pool\n", ev, cv);
520 return -EINVAL;
521 }
Ilya Dryomov17a13e42014-01-27 17:40:19 +0200522 if (cv > 9) {
523 pr_warning("got v %d cv %d > 9 of ceph_pg_pool\n", ev, cv);
Sage Weil4f6a7e52013-02-23 10:41:09 -0800524 return -EINVAL;
525 }
526 len = ceph_decode_32(p);
527 ceph_decode_need(p, end, len, bad);
528 pool_end = *p + len;
Sage Weil73a7e692010-08-02 11:00:55 -0700529
Sage Weil4f6a7e52013-02-23 10:41:09 -0800530 pi->type = ceph_decode_8(p);
531 pi->size = ceph_decode_8(p);
532 pi->crush_ruleset = ceph_decode_8(p);
533 pi->object_hash = ceph_decode_8(p);
534
535 pi->pg_num = ceph_decode_32(p);
536 pi->pgp_num = ceph_decode_32(p);
537
538 *p += 4 + 4; /* skip lpg* */
539 *p += 4; /* skip last_change */
540 *p += 8 + 4; /* skip snap_seq, snap_epoch */
541
542 /* skip snaps */
543 num = ceph_decode_32(p);
544 while (num--) {
545 *p += 8; /* snapid key */
546 *p += 1 + 1; /* versions */
547 len = ceph_decode_32(p);
548 *p += len;
Sage Weil73a7e692010-08-02 11:00:55 -0700549 }
550
Ilya Dryomov17a13e42014-01-27 17:40:19 +0200551 /* skip removed_snaps */
Sage Weil4f6a7e52013-02-23 10:41:09 -0800552 num = ceph_decode_32(p);
553 *p += num * (8 + 8);
554
555 *p += 8; /* skip auid */
556 pi->flags = ceph_decode_64(p);
Ilya Dryomov17a13e42014-01-27 17:40:19 +0200557 *p += 4; /* skip crash_replay_interval */
558
559 if (ev >= 7)
560 *p += 1; /* skip min_size */
561
562 if (ev >= 8)
563 *p += 8 + 8; /* skip quota_max_* */
564
565 if (ev >= 9) {
566 /* skip tiers */
567 num = ceph_decode_32(p);
568 *p += num * 8;
569
570 *p += 8; /* skip tier_of */
571 *p += 1; /* skip cache_mode */
572
573 pi->read_tier = ceph_decode_64(p);
574 pi->write_tier = ceph_decode_64(p);
575 } else {
576 pi->read_tier = -1;
577 pi->write_tier = -1;
578 }
Sage Weil4f6a7e52013-02-23 10:41:09 -0800579
580 /* ignore the rest */
581
582 *p = pool_end;
583 calc_pg_masks(pi);
Sage Weil73a7e692010-08-02 11:00:55 -0700584 return 0;
585
586bad:
587 return -EINVAL;
Sage Weilefd75762010-03-17 10:05:28 -0700588}
589
Sage Weil2844a762010-04-09 15:46:42 -0700590static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map)
591{
592 struct ceph_pg_pool_info *pi;
Sage Weil4f6a7e52013-02-23 10:41:09 -0800593 u32 num, len;
594 u64 pool;
Sage Weil2844a762010-04-09 15:46:42 -0700595
596 ceph_decode_32_safe(p, end, num, bad);
597 dout(" %d pool names\n", num);
598 while (num--) {
Sage Weil4f6a7e52013-02-23 10:41:09 -0800599 ceph_decode_64_safe(p, end, pool, bad);
Sage Weil2844a762010-04-09 15:46:42 -0700600 ceph_decode_32_safe(p, end, len, bad);
Sage Weil4f6a7e52013-02-23 10:41:09 -0800601 dout(" pool %llu len %d\n", pool, len);
Xi Wangad3b9042012-06-06 19:35:55 -0500602 ceph_decode_need(p, end, len, bad);
Sage Weil2844a762010-04-09 15:46:42 -0700603 pi = __lookup_pg_pool(&map->pg_pools, pool);
604 if (pi) {
Xi Wangad3b9042012-06-06 19:35:55 -0500605 char *name = kstrndup(*p, len, GFP_NOFS);
606
607 if (!name)
608 return -ENOMEM;
Sage Weil2844a762010-04-09 15:46:42 -0700609 kfree(pi->name);
Xi Wangad3b9042012-06-06 19:35:55 -0500610 pi->name = name;
611 dout(" name is %s\n", pi->name);
Sage Weil2844a762010-04-09 15:46:42 -0700612 }
613 *p += len;
614 }
615 return 0;
616
617bad:
618 return -EINVAL;
619}
620
621/*
622 * osd map
623 */
624void ceph_osdmap_destroy(struct ceph_osdmap *map)
625{
626 dout("osdmap_destroy %p\n", map);
627 if (map->crush)
628 crush_destroy(map->crush);
629 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
630 struct ceph_pg_mapping *pg =
631 rb_entry(rb_first(&map->pg_temp),
632 struct ceph_pg_mapping, node);
633 rb_erase(&pg->node, &map->pg_temp);
634 kfree(pg);
635 }
636 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
637 struct ceph_pg_pool_info *pi =
638 rb_entry(rb_first(&map->pg_pools),
639 struct ceph_pg_pool_info, node);
640 __remove_pg_pool(&map->pg_pools, pi);
641 }
642 kfree(map->osd_state);
643 kfree(map->osd_weight);
644 kfree(map->osd_addr);
645 kfree(map);
646}
647
648/*
649 * adjust max osd value. reallocate arrays.
650 */
651static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
652{
653 u8 *state;
654 struct ceph_entity_addr *addr;
655 u32 *weight;
656
657 state = kcalloc(max, sizeof(*state), GFP_NOFS);
658 addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
659 weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
660 if (state == NULL || addr == NULL || weight == NULL) {
661 kfree(state);
662 kfree(addr);
663 kfree(weight);
664 return -ENOMEM;
665 }
666
667 /* copy old? */
668 if (map->osd_state) {
669 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
670 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
671 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
672 kfree(map->osd_state);
673 kfree(map->osd_addr);
674 kfree(map->osd_weight);
675 }
676
677 map->osd_state = state;
678 map->osd_weight = weight;
679 map->osd_addr = addr;
680 map->max_osd = max;
681 return 0;
682}
683
Sage Weil4fc51be2010-02-16 15:55:03 -0800684/*
Sage Weilf24e9982009-10-06 11:31:10 -0700685 * decode a full map.
686 */
687struct ceph_osdmap *osdmap_decode(void **p, void *end)
688{
689 struct ceph_osdmap *map;
690 u16 version;
691 u32 len, max, i;
692 int err = -EINVAL;
693 void *start = *p;
Sage Weil4fc51be2010-02-16 15:55:03 -0800694 struct ceph_pg_pool_info *pi;
Sage Weilf24e9982009-10-06 11:31:10 -0700695
696 dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
697
698 map = kzalloc(sizeof(*map), GFP_NOFS);
699 if (map == NULL)
700 return ERR_PTR(-ENOMEM);
701 map->pg_temp = RB_ROOT;
702
703 ceph_decode_16_safe(p, end, version, bad);
Sage Weil4f6a7e52013-02-23 10:41:09 -0800704 if (version > 6) {
705 pr_warning("got unknown v %d > 6 of osdmap\n", version);
706 goto bad;
707 }
708 if (version < 6) {
709 pr_warning("got old v %d < 6 of osdmap\n", version);
Sage Weil02f90c62010-02-04 16:18:10 -0800710 goto bad;
711 }
Sage Weilf24e9982009-10-06 11:31:10 -0700712
713 ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
714 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
Sage Weilc89136e2009-10-14 09:59:09 -0700715 map->epoch = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -0700716 ceph_decode_copy(p, &map->created, sizeof(map->created));
717 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
718
Sage Weilf24e9982009-10-06 11:31:10 -0700719 ceph_decode_32_safe(p, end, max, bad);
720 while (max--) {
Sage Weil4f6a7e52013-02-23 10:41:09 -0800721 ceph_decode_need(p, end, 8 + 2, bad);
Sage Weil0ed72852012-10-29 11:01:42 -0700722 err = -ENOMEM;
Sage Weil2844a762010-04-09 15:46:42 -0700723 pi = kzalloc(sizeof(*pi), GFP_NOFS);
Sage Weil4fc51be2010-02-16 15:55:03 -0800724 if (!pi)
Sage Weilf24e9982009-10-06 11:31:10 -0700725 goto bad;
Sage Weil4f6a7e52013-02-23 10:41:09 -0800726 pi->id = ceph_decode_64(p);
Sage Weil73a7e692010-08-02 11:00:55 -0700727 err = __decode_pool(p, end, pi);
Jesper Juhlb0aee352010-12-24 23:01:12 +0100728 if (err < 0) {
729 kfree(pi);
Sage Weil73a7e692010-08-02 11:00:55 -0700730 goto bad;
Jesper Juhlb0aee352010-12-24 23:01:12 +0100731 }
Sage Weil4fc51be2010-02-16 15:55:03 -0800732 __insert_pg_pool(&map->pg_pools, pi);
Sage Weilf24e9982009-10-06 11:31:10 -0700733 }
Sage Weil2844a762010-04-09 15:46:42 -0700734
Sage Weil4f6a7e52013-02-23 10:41:09 -0800735 err = __decode_pool_names(p, end, map);
736 if (err < 0) {
737 dout("fail to decode pool names");
738 goto bad;
Sage Weil0ed72852012-10-29 11:01:42 -0700739 }
Sage Weil2844a762010-04-09 15:46:42 -0700740
Sage Weil4fc51be2010-02-16 15:55:03 -0800741 ceph_decode_32_safe(p, end, map->pool_max, bad);
Sage Weilf24e9982009-10-06 11:31:10 -0700742
743 ceph_decode_32_safe(p, end, map->flags, bad);
744
Sage Weilc89136e2009-10-14 09:59:09 -0700745 max = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -0700746
747 /* (re)alloc osd arrays */
748 err = osdmap_set_max_osd(map, max);
749 if (err < 0)
750 goto bad;
751 dout("osdmap_decode max_osd = %d\n", map->max_osd);
752
753 /* osds */
754 err = -EINVAL;
755 ceph_decode_need(p, end, 3*sizeof(u32) +
756 map->max_osd*(1 + sizeof(*map->osd_weight) +
757 sizeof(*map->osd_addr)), bad);
758 *p += 4; /* skip length field (should match max) */
759 ceph_decode_copy(p, map->osd_state, map->max_osd);
760
761 *p += 4; /* skip length field (should match max) */
762 for (i = 0; i < map->max_osd; i++)
Sage Weilc89136e2009-10-14 09:59:09 -0700763 map->osd_weight[i] = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -0700764
765 *p += 4; /* skip length field (should match max) */
766 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
Sage Weil63f2d212009-11-03 15:17:56 -0800767 for (i = 0; i < map->max_osd; i++)
768 ceph_decode_addr(&map->osd_addr[i]);
Sage Weilf24e9982009-10-06 11:31:10 -0700769
770 /* pg_temp */
771 ceph_decode_32_safe(p, end, len, bad);
772 for (i = 0; i < len; i++) {
773 int n, j;
Sage Weil5b191d92013-02-23 10:38:16 -0800774 struct ceph_pg pgid;
Sage Weilf24e9982009-10-06 11:31:10 -0700775 struct ceph_pg_mapping *pg;
776
Alex Elderef4859d2013-04-01 18:58:26 -0500777 err = ceph_decode_pgid(p, end, &pgid);
Sage Weild6c0dd62013-03-06 14:57:03 -0800778 if (err)
779 goto bad;
780 ceph_decode_need(p, end, sizeof(u32), bad);
Sage Weilc89136e2009-10-14 09:59:09 -0700781 n = ceph_decode_32(p);
Xi Wange91a9b62012-06-06 19:35:55 -0500782 err = -EINVAL;
783 if (n > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
784 goto bad;
Sage Weilf24e9982009-10-06 11:31:10 -0700785 ceph_decode_need(p, end, n * sizeof(u32), bad);
Sage Weil30dc6382009-12-21 14:49:37 -0800786 err = -ENOMEM;
Sage Weilf24e9982009-10-06 11:31:10 -0700787 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
Sage Weil30dc6382009-12-21 14:49:37 -0800788 if (!pg)
Sage Weilf24e9982009-10-06 11:31:10 -0700789 goto bad;
Sage Weilf24e9982009-10-06 11:31:10 -0700790 pg->pgid = pgid;
791 pg->len = n;
792 for (j = 0; j < n; j++)
Sage Weilc89136e2009-10-14 09:59:09 -0700793 pg->osds[j] = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -0700794
Sage Weil991abb62009-10-08 22:22:37 -0700795 err = __insert_pg_mapping(pg, &map->pg_temp);
796 if (err)
797 goto bad;
Sage Weil5b191d92013-02-23 10:38:16 -0800798 dout(" added pg_temp %lld.%x len %d\n", pgid.pool, pgid.seed,
799 len);
Sage Weilf24e9982009-10-06 11:31:10 -0700800 }
801
802 /* crush */
803 ceph_decode_32_safe(p, end, len, bad);
804 dout("osdmap_decode crush len %d from off 0x%x\n", len,
805 (int)(*p - start));
806 ceph_decode_need(p, end, len, bad);
807 map->crush = crush_decode(*p, end);
808 *p += len;
809 if (IS_ERR(map->crush)) {
810 err = PTR_ERR(map->crush);
811 map->crush = NULL;
812 goto bad;
813 }
814
815 /* ignore the rest of the map */
816 *p = end;
817
818 dout("osdmap_decode done %p %p\n", *p, end);
819 return map;
820
821bad:
Sage Weil0ed72852012-10-29 11:01:42 -0700822 dout("osdmap_decode fail err %d\n", err);
Sage Weilf24e9982009-10-06 11:31:10 -0700823 ceph_osdmap_destroy(map);
824 return ERR_PTR(err);
825}
826
827/*
828 * decode and apply an incremental map update.
829 */
830struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
831 struct ceph_osdmap *map,
832 struct ceph_messenger *msgr)
833{
Sage Weilf24e9982009-10-06 11:31:10 -0700834 struct crush_map *newcrush = NULL;
835 struct ceph_fsid fsid;
836 u32 epoch = 0;
837 struct ceph_timespec modified;
Sage Weil4f6a7e52013-02-23 10:41:09 -0800838 s32 len;
839 u64 pool;
840 __s64 new_pool_max;
841 __s32 new_flags, max;
Sage Weilf24e9982009-10-06 11:31:10 -0700842 void *start = *p;
843 int err = -EINVAL;
844 u16 version;
Sage Weilf24e9982009-10-06 11:31:10 -0700845
846 ceph_decode_16_safe(p, end, version, bad);
Sage Weild6c0dd62013-03-06 14:57:03 -0800847 if (version != 6) {
848 pr_warning("got unknown v %d != 6 of inc osdmap\n", version);
Sage Weil02f90c62010-02-04 16:18:10 -0800849 goto bad;
850 }
Sage Weilf24e9982009-10-06 11:31:10 -0700851
852 ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
853 bad);
854 ceph_decode_copy(p, &fsid, sizeof(fsid));
Sage Weilc89136e2009-10-14 09:59:09 -0700855 epoch = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -0700856 BUG_ON(epoch != map->epoch+1);
857 ceph_decode_copy(p, &modified, sizeof(modified));
Sage Weil4f6a7e52013-02-23 10:41:09 -0800858 new_pool_max = ceph_decode_64(p);
Sage Weilc89136e2009-10-14 09:59:09 -0700859 new_flags = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -0700860
861 /* full map? */
862 ceph_decode_32_safe(p, end, len, bad);
863 if (len > 0) {
864 dout("apply_incremental full map len %d, %p to %p\n",
865 len, *p, end);
Sage Weil30dc6382009-12-21 14:49:37 -0800866 return osdmap_decode(p, min(*p+len, end));
Sage Weilf24e9982009-10-06 11:31:10 -0700867 }
868
869 /* new crush? */
870 ceph_decode_32_safe(p, end, len, bad);
871 if (len > 0) {
872 dout("apply_incremental new crush map len %d, %p to %p\n",
873 len, *p, end);
874 newcrush = crush_decode(*p, min(*p+len, end));
875 if (IS_ERR(newcrush))
Julia Lawall7e34bc52010-05-22 12:01:14 +0200876 return ERR_CAST(newcrush);
Sage Weilcebc5be2010-06-17 10:22:48 -0700877 *p += len;
Sage Weilf24e9982009-10-06 11:31:10 -0700878 }
879
880 /* new flags? */
881 if (new_flags >= 0)
882 map->flags = new_flags;
Sage Weil4fc51be2010-02-16 15:55:03 -0800883 if (new_pool_max >= 0)
884 map->pool_max = new_pool_max;
Sage Weilf24e9982009-10-06 11:31:10 -0700885
886 ceph_decode_need(p, end, 5*sizeof(u32), bad);
887
888 /* new max? */
Sage Weilc89136e2009-10-14 09:59:09 -0700889 max = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -0700890 if (max >= 0) {
891 err = osdmap_set_max_osd(map, max);
892 if (err < 0)
893 goto bad;
894 }
895
896 map->epoch++;
Sage Weil31456662011-05-12 15:18:43 -0700897 map->modified = modified;
Sage Weilf24e9982009-10-06 11:31:10 -0700898 if (newcrush) {
899 if (map->crush)
900 crush_destroy(map->crush);
901 map->crush = newcrush;
902 newcrush = NULL;
903 }
904
905 /* new_pool */
906 ceph_decode_32_safe(p, end, len, bad);
907 while (len--) {
Sage Weil4fc51be2010-02-16 15:55:03 -0800908 struct ceph_pg_pool_info *pi;
Sage Weil361be862010-01-25 16:03:02 -0800909
Sage Weil4f6a7e52013-02-23 10:41:09 -0800910 ceph_decode_64_safe(p, end, pool, bad);
Sage Weil4fc51be2010-02-16 15:55:03 -0800911 pi = __lookup_pg_pool(&map->pg_pools, pool);
912 if (!pi) {
Sage Weil2844a762010-04-09 15:46:42 -0700913 pi = kzalloc(sizeof(*pi), GFP_NOFS);
Sage Weil4fc51be2010-02-16 15:55:03 -0800914 if (!pi) {
915 err = -ENOMEM;
916 goto bad;
917 }
918 pi->id = pool;
919 __insert_pg_pool(&map->pg_pools, pi);
920 }
Sage Weil73a7e692010-08-02 11:00:55 -0700921 err = __decode_pool(p, end, pi);
922 if (err < 0)
923 goto bad;
Sage Weilf24e9982009-10-06 11:31:10 -0700924 }
Sage Weil0ed72852012-10-29 11:01:42 -0700925 if (version >= 5) {
926 err = __decode_pool_names(p, end, map);
927 if (err < 0)
928 goto bad;
929 }
Sage Weilf24e9982009-10-06 11:31:10 -0700930
Sage Weil4fc51be2010-02-16 15:55:03 -0800931 /* old_pool */
Sage Weilf24e9982009-10-06 11:31:10 -0700932 ceph_decode_32_safe(p, end, len, bad);
Sage Weil4fc51be2010-02-16 15:55:03 -0800933 while (len--) {
934 struct ceph_pg_pool_info *pi;
935
Sage Weil4f6a7e52013-02-23 10:41:09 -0800936 ceph_decode_64_safe(p, end, pool, bad);
Sage Weil4fc51be2010-02-16 15:55:03 -0800937 pi = __lookup_pg_pool(&map->pg_pools, pool);
Sage Weil2844a762010-04-09 15:46:42 -0700938 if (pi)
939 __remove_pg_pool(&map->pg_pools, pi);
Sage Weil4fc51be2010-02-16 15:55:03 -0800940 }
Sage Weilf24e9982009-10-06 11:31:10 -0700941
942 /* new_up */
943 err = -EINVAL;
944 ceph_decode_32_safe(p, end, len, bad);
945 while (len--) {
946 u32 osd;
947 struct ceph_entity_addr addr;
948 ceph_decode_32_safe(p, end, osd, bad);
949 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
Sage Weil63f2d212009-11-03 15:17:56 -0800950 ceph_decode_addr(&addr);
Sage Weilf24e9982009-10-06 11:31:10 -0700951 pr_info("osd%d up\n", osd);
952 BUG_ON(osd >= map->max_osd);
953 map->osd_state[osd] |= CEPH_OSD_UP;
954 map->osd_addr[osd] = addr;
955 }
956
Sage Weil7662d8f2011-05-03 12:52:05 -0700957 /* new_state */
Sage Weilf24e9982009-10-06 11:31:10 -0700958 ceph_decode_32_safe(p, end, len, bad);
959 while (len--) {
960 u32 osd;
Sage Weil7662d8f2011-05-03 12:52:05 -0700961 u8 xorstate;
Sage Weilf24e9982009-10-06 11:31:10 -0700962 ceph_decode_32_safe(p, end, osd, bad);
Sage Weil7662d8f2011-05-03 12:52:05 -0700963 xorstate = **(u8 **)p;
Sage Weilf24e9982009-10-06 11:31:10 -0700964 (*p)++; /* clean flag */
Sage Weil7662d8f2011-05-03 12:52:05 -0700965 if (xorstate == 0)
966 xorstate = CEPH_OSD_UP;
967 if (xorstate & CEPH_OSD_UP)
968 pr_info("osd%d down\n", osd);
Sage Weilf24e9982009-10-06 11:31:10 -0700969 if (osd < map->max_osd)
Sage Weil7662d8f2011-05-03 12:52:05 -0700970 map->osd_state[osd] ^= xorstate;
Sage Weilf24e9982009-10-06 11:31:10 -0700971 }
972
973 /* new_weight */
974 ceph_decode_32_safe(p, end, len, bad);
975 while (len--) {
976 u32 osd, off;
977 ceph_decode_need(p, end, sizeof(u32)*2, bad);
Sage Weilc89136e2009-10-14 09:59:09 -0700978 osd = ceph_decode_32(p);
979 off = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -0700980 pr_info("osd%d weight 0x%x %s\n", osd, off,
981 off == CEPH_OSD_IN ? "(in)" :
982 (off == CEPH_OSD_OUT ? "(out)" : ""));
983 if (osd < map->max_osd)
984 map->osd_weight[osd] = off;
985 }
986
987 /* new_pg_temp */
Sage Weilf24e9982009-10-06 11:31:10 -0700988 ceph_decode_32_safe(p, end, len, bad);
989 while (len--) {
990 struct ceph_pg_mapping *pg;
991 int j;
Sage Weil5b191d92013-02-23 10:38:16 -0800992 struct ceph_pg pgid;
Sage Weilf24e9982009-10-06 11:31:10 -0700993 u32 pglen;
Sage Weilf24e9982009-10-06 11:31:10 -0700994
Alex Elderef4859d2013-04-01 18:58:26 -0500995 err = ceph_decode_pgid(p, end, &pgid);
Sage Weild6c0dd62013-03-06 14:57:03 -0800996 if (err)
997 goto bad;
998 ceph_decode_need(p, end, sizeof(u32), bad);
999 pglen = ceph_decode_32(p);
Sage Weilf24e9982009-10-06 11:31:10 -07001000 if (pglen) {
Sage Weilf24e9982009-10-06 11:31:10 -07001001 ceph_decode_need(p, end, pglen*sizeof(u32), bad);
Sage Weil6bd9adb2012-05-21 09:45:23 -07001002
1003 /* removing existing (if any) */
1004 (void) __remove_pg_mapping(&map->pg_temp, pgid);
1005
1006 /* insert */
Sage Weil0ed72852012-10-29 11:01:42 -07001007 err = -EINVAL;
1008 if (pglen > (UINT_MAX - sizeof(*pg)) / sizeof(u32))
Xi Wanga5506042012-06-06 19:35:55 -05001009 goto bad;
Sage Weil0ed72852012-10-29 11:01:42 -07001010 err = -ENOMEM;
Sage Weilf24e9982009-10-06 11:31:10 -07001011 pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
Sage Weil0ed72852012-10-29 11:01:42 -07001012 if (!pg)
Sage Weilf24e9982009-10-06 11:31:10 -07001013 goto bad;
Sage Weilf24e9982009-10-06 11:31:10 -07001014 pg->pgid = pgid;
1015 pg->len = pglen;
Sage Weil7067f792009-12-21 16:02:37 -08001016 for (j = 0; j < pglen; j++)
Sage Weilc89136e2009-10-14 09:59:09 -07001017 pg->osds[j] = ceph_decode_32(p);
Sage Weil991abb62009-10-08 22:22:37 -07001018 err = __insert_pg_mapping(pg, &map->pg_temp);
Sage Weilbc4fdca2010-07-20 16:19:56 -07001019 if (err) {
1020 kfree(pg);
Sage Weil991abb62009-10-08 22:22:37 -07001021 goto bad;
Sage Weilbc4fdca2010-07-20 16:19:56 -07001022 }
Sage Weil5b191d92013-02-23 10:38:16 -08001023 dout(" added pg_temp %lld.%x len %d\n", pgid.pool,
1024 pgid.seed, pglen);
Sage Weil8adc8b32011-09-28 10:11:04 -07001025 } else {
1026 /* remove */
1027 __remove_pg_mapping(&map->pg_temp, pgid);
Sage Weilf24e9982009-10-06 11:31:10 -07001028 }
1029 }
Sage Weilf24e9982009-10-06 11:31:10 -07001030
1031 /* ignore the rest */
1032 *p = end;
1033 return map;
1034
1035bad:
1036 pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
1037 epoch, (int)(*p - start), *p, start, end);
Sage Weil9ec7cab2009-12-14 15:13:47 -08001038 print_hex_dump(KERN_DEBUG, "osdmap: ",
1039 DUMP_PREFIX_OFFSET, 16, 1,
1040 start, end - start, true);
Sage Weilf24e9982009-10-06 11:31:10 -07001041 if (newcrush)
1042 crush_destroy(newcrush);
1043 return ERR_PTR(err);
1044}
1045
1046
1047
1048
1049/*
1050 * calculate file layout from given offset, length.
1051 * fill in correct oid, logical length, and object extent
1052 * offset, length.
1053 *
1054 * for now, we write only a single su, until we can
1055 * pass a stride back to the caller.
1056 */
Sage Weild63b77f2012-09-24 20:59:48 -07001057int ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
Alex Eldere8afad62012-11-14 09:38:19 -06001058 u64 off, u64 len,
Sage Weil645a1022009-10-28 15:15:05 -07001059 u64 *ono,
Sage Weilf24e9982009-10-06 11:31:10 -07001060 u64 *oxoff, u64 *oxlen)
1061{
1062 u32 osize = le32_to_cpu(layout->fl_object_size);
1063 u32 su = le32_to_cpu(layout->fl_stripe_unit);
1064 u32 sc = le32_to_cpu(layout->fl_stripe_count);
1065 u32 bl, stripeno, stripepos, objsetno;
1066 u32 su_per_object;
Noah Watkinsff1d1f72009-10-30 12:57:30 -07001067 u64 t, su_offset;
Sage Weilf24e9982009-10-06 11:31:10 -07001068
Alex Eldere8afad62012-11-14 09:38:19 -06001069 dout("mapping %llu~%llu osize %u fl_su %u\n", off, len,
Sage Weilf24e9982009-10-06 11:31:10 -07001070 osize, su);
Sage Weild63b77f2012-09-24 20:59:48 -07001071 if (su == 0 || sc == 0)
1072 goto invalid;
Noah Watkins35e054a2009-10-28 14:04:48 -07001073 su_per_object = osize / su;
Sage Weild63b77f2012-09-24 20:59:48 -07001074 if (su_per_object == 0)
1075 goto invalid;
Sage Weilf24e9982009-10-06 11:31:10 -07001076 dout("osize %u / su %u = su_per_object %u\n", osize, su,
1077 su_per_object);
1078
Sage Weild63b77f2012-09-24 20:59:48 -07001079 if ((su & ~PAGE_MASK) != 0)
1080 goto invalid;
1081
Sage Weilf24e9982009-10-06 11:31:10 -07001082 /* bl = *off / su; */
1083 t = off;
1084 do_div(t, su);
1085 bl = t;
1086 dout("off %llu / su %u = bl %u\n", off, su, bl);
1087
1088 stripeno = bl / sc;
1089 stripepos = bl % sc;
1090 objsetno = stripeno / su_per_object;
1091
Sage Weil645a1022009-10-28 15:15:05 -07001092 *ono = objsetno * sc + stripepos;
Eric Dumazet95c96172012-04-15 05:58:06 +00001093 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned int)*ono);
Sage Weil645a1022009-10-28 15:15:05 -07001094
1095 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
Sage Weilf24e9982009-10-06 11:31:10 -07001096 t = off;
Noah Watkinsff1d1f72009-10-30 12:57:30 -07001097 su_offset = do_div(t, su);
1098 *oxoff = su_offset + (stripeno % su_per_object) * su;
Sage Weil645a1022009-10-28 15:15:05 -07001099
Noah Watkinsff1d1f72009-10-30 12:57:30 -07001100 /*
1101 * Calculate the length of the extent being written to the selected
Alex Eldere8afad62012-11-14 09:38:19 -06001102 * object. This is the minimum of the full length requested (len) or
Noah Watkinsff1d1f72009-10-30 12:57:30 -07001103 * the remainder of the current stripe being written to.
1104 */
Alex Eldere8afad62012-11-14 09:38:19 -06001105 *oxlen = min_t(u64, len, su - su_offset);
Sage Weilf24e9982009-10-06 11:31:10 -07001106
1107 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
Sage Weild63b77f2012-09-24 20:59:48 -07001108 return 0;
1109
1110invalid:
1111 dout(" invalid layout\n");
1112 *ono = 0;
1113 *oxoff = 0;
1114 *oxlen = 0;
1115 return -EINVAL;
Sage Weilf24e9982009-10-06 11:31:10 -07001116}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001117EXPORT_SYMBOL(ceph_calc_file_object_mapping);
Sage Weilf24e9982009-10-06 11:31:10 -07001118
1119/*
Ilya Dryomov7c13cb62014-01-27 17:40:19 +02001120 * Calculate mapping of a (oloc, oid) pair to a PG. Should only be
1121 * called with target's (oloc, oid), since tiering isn't taken into
1122 * account.
Sage Weilf24e9982009-10-06 11:31:10 -07001123 */
Ilya Dryomov7c13cb62014-01-27 17:40:19 +02001124int ceph_oloc_oid_to_pg(struct ceph_osdmap *osdmap,
1125 struct ceph_object_locator *oloc,
1126 struct ceph_object_id *oid,
1127 struct ceph_pg *pg_out)
Sage Weilf24e9982009-10-06 11:31:10 -07001128{
Ilya Dryomov7c13cb62014-01-27 17:40:19 +02001129 struct ceph_pg_pool_info *pi;
Sage Weilf24e9982009-10-06 11:31:10 -07001130
Ilya Dryomov7c13cb62014-01-27 17:40:19 +02001131 pi = __lookup_pg_pool(&osdmap->pg_pools, oloc->pool);
1132 if (!pi)
Sage Weil4fc51be2010-02-16 15:55:03 -08001133 return -EIO;
Sage Weilf24e9982009-10-06 11:31:10 -07001134
Ilya Dryomov7c13cb62014-01-27 17:40:19 +02001135 pg_out->pool = oloc->pool;
1136 pg_out->seed = ceph_str_hash(pi->object_hash, oid->name,
1137 oid->name_len);
1138
1139 dout("%s '%.*s' pgid %llu.%x\n", __func__, oid->name_len, oid->name,
1140 pg_out->pool, pg_out->seed);
Sage Weilf24e9982009-10-06 11:31:10 -07001141 return 0;
1142}
Ilya Dryomov7c13cb62014-01-27 17:40:19 +02001143EXPORT_SYMBOL(ceph_oloc_oid_to_pg);
Sage Weilf24e9982009-10-06 11:31:10 -07001144
Ilya Dryomove8ef19c2013-12-24 21:19:24 +02001145static int crush_do_rule_ary(const struct crush_map *map, int ruleno, int x,
1146 int *result, int result_max,
1147 const __u32 *weight, int weight_max)
1148{
1149 int scratch[result_max * 3];
1150
1151 return crush_do_rule(map, ruleno, x, result, result_max,
1152 weight, weight_max, scratch);
1153}
1154
Sage Weilf24e9982009-10-06 11:31:10 -07001155/*
1156 * Calculate raw osd vector for the given pgid. Return pointer to osd
1157 * array, or NULL on failure.
1158 */
Sage Weil5b191d92013-02-23 10:38:16 -08001159static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
Sage Weilf24e9982009-10-06 11:31:10 -07001160 int *osds, int *num)
1161{
Sage Weilf24e9982009-10-06 11:31:10 -07001162 struct ceph_pg_mapping *pg;
1163 struct ceph_pg_pool_info *pool;
1164 int ruleno;
Sage Weil83ca14f2013-02-26 10:39:09 -08001165 int r;
1166 u32 pps;
Sage Weilf24e9982009-10-06 11:31:10 -07001167
Sage Weil83ca14f2013-02-26 10:39:09 -08001168 pool = __lookup_pg_pool(&osdmap->pg_pools, pgid.pool);
Sage Weil782e1822011-09-28 10:08:27 -07001169 if (!pool)
1170 return NULL;
1171
Sage Weilf24e9982009-10-06 11:31:10 -07001172 /* pg_temp? */
Sage Weil83ca14f2013-02-26 10:39:09 -08001173 pgid.seed = ceph_stable_mod(pgid.seed, pool->pg_num,
Sage Weil9542cf02013-08-28 17:17:29 -07001174 pool->pg_num_mask);
Sage Weil9794b142010-02-16 15:53:32 -08001175 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
1176 if (pg) {
1177 *num = pg->len;
1178 return pg->osds;
Sage Weilf24e9982009-10-06 11:31:10 -07001179 }
1180
1181 /* crush */
Sage Weil4f6a7e52013-02-23 10:41:09 -08001182 ruleno = crush_find_rule(osdmap->crush, pool->crush_ruleset,
1183 pool->type, pool->size);
Sage Weilf24e9982009-10-06 11:31:10 -07001184 if (ruleno < 0) {
Sage Weil83ca14f2013-02-26 10:39:09 -08001185 pr_err("no crush rule pool %lld ruleset %d type %d size %d\n",
1186 pgid.pool, pool->crush_ruleset, pool->type,
Sage Weil4f6a7e52013-02-23 10:41:09 -08001187 pool->size);
Sage Weilf24e9982009-10-06 11:31:10 -07001188 return NULL;
1189 }
1190
Sage Weil83ca14f2013-02-26 10:39:09 -08001191 if (pool->flags & CEPH_POOL_FLAG_HASHPSPOOL) {
1192 /* hash pool id and seed sothat pool PGs do not overlap */
1193 pps = crush_hash32_2(CRUSH_HASH_RJENKINS1,
1194 ceph_stable_mod(pgid.seed, pool->pgp_num,
1195 pool->pgp_num_mask),
1196 pgid.pool);
1197 } else {
1198 /*
1199 * legacy ehavior: add ps and pool together. this is
1200 * not a great approach because the PGs from each pool
1201 * will overlap on top of each other: 0.5 == 1.4 ==
1202 * 2.3 == ...
1203 */
1204 pps = ceph_stable_mod(pgid.seed, pool->pgp_num,
1205 pool->pgp_num_mask) +
1206 (unsigned)pgid.pool;
1207 }
Ilya Dryomove8ef19c2013-12-24 21:19:24 +02001208 r = crush_do_rule_ary(osdmap->crush, ruleno, pps,
1209 osds, min_t(int, pool->size, *num),
1210 osdmap->osd_weight, osdmap->max_osd);
Sage Weil8b393262012-05-07 15:37:23 -07001211 if (r < 0) {
Sage Weil83ca14f2013-02-26 10:39:09 -08001212 pr_err("error %d from crush rule: pool %lld ruleset %d type %d"
1213 " size %d\n", r, pgid.pool, pool->crush_ruleset,
Sage Weil4f6a7e52013-02-23 10:41:09 -08001214 pool->type, pool->size);
Sage Weil8b393262012-05-07 15:37:23 -07001215 return NULL;
1216 }
1217 *num = r;
Sage Weilf24e9982009-10-06 11:31:10 -07001218 return osds;
1219}
1220
1221/*
Sage Weild85b7052010-05-10 10:24:48 -07001222 * Return acting set for given pgid.
1223 */
Sage Weil5b191d92013-02-23 10:38:16 -08001224int ceph_calc_pg_acting(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
Sage Weild85b7052010-05-10 10:24:48 -07001225 int *acting)
1226{
1227 int rawosds[CEPH_PG_MAX_SIZE], *osds;
1228 int i, o, num = CEPH_PG_MAX_SIZE;
1229
1230 osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1231 if (!osds)
1232 return -1;
1233
1234 /* primary is first up osd */
1235 o = 0;
1236 for (i = 0; i < num; i++)
1237 if (ceph_osd_is_up(osdmap, osds[i]))
1238 acting[o++] = osds[i];
1239 return o;
1240}
1241
1242/*
Sage Weilf24e9982009-10-06 11:31:10 -07001243 * Return primary osd for given pgid, or -1 if none.
1244 */
Sage Weil5b191d92013-02-23 10:38:16 -08001245int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
Sage Weilf24e9982009-10-06 11:31:10 -07001246{
Sage Weild85b7052010-05-10 10:24:48 -07001247 int rawosds[CEPH_PG_MAX_SIZE], *osds;
1248 int i, num = CEPH_PG_MAX_SIZE;
Sage Weilf24e9982009-10-06 11:31:10 -07001249
1250 osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1251 if (!osds)
1252 return -1;
1253
1254 /* primary is first up osd */
1255 for (i = 0; i < num; i++)
Sage Weild85b7052010-05-10 10:24:48 -07001256 if (ceph_osd_is_up(osdmap, osds[i]))
Sage Weilf24e9982009-10-06 11:31:10 -07001257 return osds[i];
Sage Weilf24e9982009-10-06 11:31:10 -07001258 return -1;
1259}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001260EXPORT_SYMBOL(ceph_calc_pg_primary);