diff options
Diffstat (limited to 'fs/ceph/osdmap.c')
-rw-r--r-- | fs/ceph/osdmap.c | 1062 |
1 files changed, 1062 insertions, 0 deletions
diff --git a/fs/ceph/osdmap.c b/fs/ceph/osdmap.c new file mode 100644 index 000000000000..2e2c15eed82a --- /dev/null +++ b/fs/ceph/osdmap.c | |||
@@ -0,0 +1,1062 @@ | |||
1 | |||
2 | #include "ceph_debug.h" | ||
3 | |||
4 | #include <linux/slab.h> | ||
5 | #include <asm/div64.h> | ||
6 | |||
7 | #include "super.h" | ||
8 | #include "osdmap.h" | ||
9 | #include "crush/hash.h" | ||
10 | #include "crush/mapper.h" | ||
11 | #include "decode.h" | ||
12 | |||
13 | char *ceph_osdmap_state_str(char *str, int len, int state) | ||
14 | { | ||
15 | int flag = 0; | ||
16 | |||
17 | if (!len) | ||
18 | goto done; | ||
19 | |||
20 | *str = '\0'; | ||
21 | if (state) { | ||
22 | if (state & CEPH_OSD_EXISTS) { | ||
23 | snprintf(str, len, "exists"); | ||
24 | flag = 1; | ||
25 | } | ||
26 | if (state & CEPH_OSD_UP) { | ||
27 | snprintf(str, len, "%s%s%s", str, (flag ? ", " : ""), | ||
28 | "up"); | ||
29 | flag = 1; | ||
30 | } | ||
31 | } else { | ||
32 | snprintf(str, len, "doesn't exist"); | ||
33 | } | ||
34 | done: | ||
35 | return str; | ||
36 | } | ||
37 | |||
38 | /* maps */ | ||
39 | |||
40 | static int calc_bits_of(unsigned t) | ||
41 | { | ||
42 | int b = 0; | ||
43 | while (t) { | ||
44 | t = t >> 1; | ||
45 | b++; | ||
46 | } | ||
47 | return b; | ||
48 | } | ||
49 | |||
50 | /* | ||
51 | * the foo_mask is the smallest value 2^n-1 that is >= foo. | ||
52 | */ | ||
53 | static void calc_pg_masks(struct ceph_pg_pool_info *pi) | ||
54 | { | ||
55 | pi->pg_num_mask = (1 << calc_bits_of(le32_to_cpu(pi->v.pg_num)-1)) - 1; | ||
56 | pi->pgp_num_mask = | ||
57 | (1 << calc_bits_of(le32_to_cpu(pi->v.pgp_num)-1)) - 1; | ||
58 | pi->lpg_num_mask = | ||
59 | (1 << calc_bits_of(le32_to_cpu(pi->v.lpg_num)-1)) - 1; | ||
60 | pi->lpgp_num_mask = | ||
61 | (1 << calc_bits_of(le32_to_cpu(pi->v.lpgp_num)-1)) - 1; | ||
62 | } | ||
63 | |||
64 | /* | ||
65 | * decode crush map | ||
66 | */ | ||
67 | static int crush_decode_uniform_bucket(void **p, void *end, | ||
68 | struct crush_bucket_uniform *b) | ||
69 | { | ||
70 | dout("crush_decode_uniform_bucket %p to %p\n", *p, end); | ||
71 | ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad); | ||
72 | b->item_weight = ceph_decode_32(p); | ||
73 | return 0; | ||
74 | bad: | ||
75 | return -EINVAL; | ||
76 | } | ||
77 | |||
78 | static int crush_decode_list_bucket(void **p, void *end, | ||
79 | struct crush_bucket_list *b) | ||
80 | { | ||
81 | int j; | ||
82 | dout("crush_decode_list_bucket %p to %p\n", *p, end); | ||
83 | b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); | ||
84 | if (b->item_weights == NULL) | ||
85 | return -ENOMEM; | ||
86 | b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); | ||
87 | if (b->sum_weights == NULL) | ||
88 | return -ENOMEM; | ||
89 | ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad); | ||
90 | for (j = 0; j < b->h.size; j++) { | ||
91 | b->item_weights[j] = ceph_decode_32(p); | ||
92 | b->sum_weights[j] = ceph_decode_32(p); | ||
93 | } | ||
94 | return 0; | ||
95 | bad: | ||
96 | return -EINVAL; | ||
97 | } | ||
98 | |||
99 | static int crush_decode_tree_bucket(void **p, void *end, | ||
100 | struct crush_bucket_tree *b) | ||
101 | { | ||
102 | int j; | ||
103 | dout("crush_decode_tree_bucket %p to %p\n", *p, end); | ||
104 | ceph_decode_32_safe(p, end, b->num_nodes, bad); | ||
105 | b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS); | ||
106 | if (b->node_weights == NULL) | ||
107 | return -ENOMEM; | ||
108 | ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad); | ||
109 | for (j = 0; j < b->num_nodes; j++) | ||
110 | b->node_weights[j] = ceph_decode_32(p); | ||
111 | return 0; | ||
112 | bad: | ||
113 | return -EINVAL; | ||
114 | } | ||
115 | |||
116 | static int crush_decode_straw_bucket(void **p, void *end, | ||
117 | struct crush_bucket_straw *b) | ||
118 | { | ||
119 | int j; | ||
120 | dout("crush_decode_straw_bucket %p to %p\n", *p, end); | ||
121 | b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); | ||
122 | if (b->item_weights == NULL) | ||
123 | return -ENOMEM; | ||
124 | b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS); | ||
125 | if (b->straws == NULL) | ||
126 | return -ENOMEM; | ||
127 | ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad); | ||
128 | for (j = 0; j < b->h.size; j++) { | ||
129 | b->item_weights[j] = ceph_decode_32(p); | ||
130 | b->straws[j] = ceph_decode_32(p); | ||
131 | } | ||
132 | return 0; | ||
133 | bad: | ||
134 | return -EINVAL; | ||
135 | } | ||
136 | |||
137 | static struct crush_map *crush_decode(void *pbyval, void *end) | ||
138 | { | ||
139 | struct crush_map *c; | ||
140 | int err = -EINVAL; | ||
141 | int i, j; | ||
142 | void **p = &pbyval; | ||
143 | void *start = pbyval; | ||
144 | u32 magic; | ||
145 | |||
146 | dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p)); | ||
147 | |||
148 | c = kzalloc(sizeof(*c), GFP_NOFS); | ||
149 | if (c == NULL) | ||
150 | return ERR_PTR(-ENOMEM); | ||
151 | |||
152 | ceph_decode_need(p, end, 4*sizeof(u32), bad); | ||
153 | magic = ceph_decode_32(p); | ||
154 | if (magic != CRUSH_MAGIC) { | ||
155 | pr_err("crush_decode magic %x != current %x\n", | ||
156 | (unsigned)magic, (unsigned)CRUSH_MAGIC); | ||
157 | goto bad; | ||
158 | } | ||
159 | c->max_buckets = ceph_decode_32(p); | ||
160 | c->max_rules = ceph_decode_32(p); | ||
161 | c->max_devices = ceph_decode_32(p); | ||
162 | |||
163 | c->device_parents = kcalloc(c->max_devices, sizeof(u32), GFP_NOFS); | ||
164 | if (c->device_parents == NULL) | ||
165 | goto badmem; | ||
166 | c->bucket_parents = kcalloc(c->max_buckets, sizeof(u32), GFP_NOFS); | ||
167 | if (c->bucket_parents == NULL) | ||
168 | goto badmem; | ||
169 | |||
170 | c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS); | ||
171 | if (c->buckets == NULL) | ||
172 | goto badmem; | ||
173 | c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS); | ||
174 | if (c->rules == NULL) | ||
175 | goto badmem; | ||
176 | |||
177 | /* buckets */ | ||
178 | for (i = 0; i < c->max_buckets; i++) { | ||
179 | int size = 0; | ||
180 | u32 alg; | ||
181 | struct crush_bucket *b; | ||
182 | |||
183 | ceph_decode_32_safe(p, end, alg, bad); | ||
184 | if (alg == 0) { | ||
185 | c->buckets[i] = NULL; | ||
186 | continue; | ||
187 | } | ||
188 | dout("crush_decode bucket %d off %x %p to %p\n", | ||
189 | i, (int)(*p-start), *p, end); | ||
190 | |||
191 | switch (alg) { | ||
192 | case CRUSH_BUCKET_UNIFORM: | ||
193 | size = sizeof(struct crush_bucket_uniform); | ||
194 | break; | ||
195 | case CRUSH_BUCKET_LIST: | ||
196 | size = sizeof(struct crush_bucket_list); | ||
197 | break; | ||
198 | case CRUSH_BUCKET_TREE: | ||
199 | size = sizeof(struct crush_bucket_tree); | ||
200 | break; | ||
201 | case CRUSH_BUCKET_STRAW: | ||
202 | size = sizeof(struct crush_bucket_straw); | ||
203 | break; | ||
204 | default: | ||
205 | err = -EINVAL; | ||
206 | goto bad; | ||
207 | } | ||
208 | BUG_ON(size == 0); | ||
209 | b = c->buckets[i] = kzalloc(size, GFP_NOFS); | ||
210 | if (b == NULL) | ||
211 | goto badmem; | ||
212 | |||
213 | ceph_decode_need(p, end, 4*sizeof(u32), bad); | ||
214 | b->id = ceph_decode_32(p); | ||
215 | b->type = ceph_decode_16(p); | ||
216 | b->alg = ceph_decode_8(p); | ||
217 | b->hash = ceph_decode_8(p); | ||
218 | b->weight = ceph_decode_32(p); | ||
219 | b->size = ceph_decode_32(p); | ||
220 | |||
221 | dout("crush_decode bucket size %d off %x %p to %p\n", | ||
222 | b->size, (int)(*p-start), *p, end); | ||
223 | |||
224 | b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS); | ||
225 | if (b->items == NULL) | ||
226 | goto badmem; | ||
227 | b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS); | ||
228 | if (b->perm == NULL) | ||
229 | goto badmem; | ||
230 | b->perm_n = 0; | ||
231 | |||
232 | ceph_decode_need(p, end, b->size*sizeof(u32), bad); | ||
233 | for (j = 0; j < b->size; j++) | ||
234 | b->items[j] = ceph_decode_32(p); | ||
235 | |||
236 | switch (b->alg) { | ||
237 | case CRUSH_BUCKET_UNIFORM: | ||
238 | err = crush_decode_uniform_bucket(p, end, | ||
239 | (struct crush_bucket_uniform *)b); | ||
240 | if (err < 0) | ||
241 | goto bad; | ||
242 | break; | ||
243 | case CRUSH_BUCKET_LIST: | ||
244 | err = crush_decode_list_bucket(p, end, | ||
245 | (struct crush_bucket_list *)b); | ||
246 | if (err < 0) | ||
247 | goto bad; | ||
248 | break; | ||
249 | case CRUSH_BUCKET_TREE: | ||
250 | err = crush_decode_tree_bucket(p, end, | ||
251 | (struct crush_bucket_tree *)b); | ||
252 | if (err < 0) | ||
253 | goto bad; | ||
254 | break; | ||
255 | case CRUSH_BUCKET_STRAW: | ||
256 | err = crush_decode_straw_bucket(p, end, | ||
257 | (struct crush_bucket_straw *)b); | ||
258 | if (err < 0) | ||
259 | goto bad; | ||
260 | break; | ||
261 | } | ||
262 | } | ||
263 | |||
264 | /* rules */ | ||
265 | dout("rule vec is %p\n", c->rules); | ||
266 | for (i = 0; i < c->max_rules; i++) { | ||
267 | u32 yes; | ||
268 | struct crush_rule *r; | ||
269 | |||
270 | ceph_decode_32_safe(p, end, yes, bad); | ||
271 | if (!yes) { | ||
272 | dout("crush_decode NO rule %d off %x %p to %p\n", | ||
273 | i, (int)(*p-start), *p, end); | ||
274 | c->rules[i] = NULL; | ||
275 | continue; | ||
276 | } | ||
277 | |||
278 | dout("crush_decode rule %d off %x %p to %p\n", | ||
279 | i, (int)(*p-start), *p, end); | ||
280 | |||
281 | /* len */ | ||
282 | ceph_decode_32_safe(p, end, yes, bad); | ||
283 | #if BITS_PER_LONG == 32 | ||
284 | err = -EINVAL; | ||
285 | if (yes > ULONG_MAX / sizeof(struct crush_rule_step)) | ||
286 | goto bad; | ||
287 | #endif | ||
288 | r = c->rules[i] = kmalloc(sizeof(*r) + | ||
289 | yes*sizeof(struct crush_rule_step), | ||
290 | GFP_NOFS); | ||
291 | if (r == NULL) | ||
292 | goto badmem; | ||
293 | dout(" rule %d is at %p\n", i, r); | ||
294 | r->len = yes; | ||
295 | ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */ | ||
296 | ceph_decode_need(p, end, r->len*3*sizeof(u32), bad); | ||
297 | for (j = 0; j < r->len; j++) { | ||
298 | r->steps[j].op = ceph_decode_32(p); | ||
299 | r->steps[j].arg1 = ceph_decode_32(p); | ||
300 | r->steps[j].arg2 = ceph_decode_32(p); | ||
301 | } | ||
302 | } | ||
303 | |||
304 | /* ignore trailing name maps. */ | ||
305 | |||
306 | dout("crush_decode success\n"); | ||
307 | return c; | ||
308 | |||
309 | badmem: | ||
310 | err = -ENOMEM; | ||
311 | bad: | ||
312 | dout("crush_decode fail %d\n", err); | ||
313 | crush_destroy(c); | ||
314 | return ERR_PTR(err); | ||
315 | } | ||
316 | |||
317 | /* | ||
318 | * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid | ||
319 | * to a set of osds) | ||
320 | */ | ||
321 | static int pgid_cmp(struct ceph_pg l, struct ceph_pg r) | ||
322 | { | ||
323 | u64 a = *(u64 *)&l; | ||
324 | u64 b = *(u64 *)&r; | ||
325 | |||
326 | if (a < b) | ||
327 | return -1; | ||
328 | if (a > b) | ||
329 | return 1; | ||
330 | return 0; | ||
331 | } | ||
332 | |||
333 | static int __insert_pg_mapping(struct ceph_pg_mapping *new, | ||
334 | struct rb_root *root) | ||
335 | { | ||
336 | struct rb_node **p = &root->rb_node; | ||
337 | struct rb_node *parent = NULL; | ||
338 | struct ceph_pg_mapping *pg = NULL; | ||
339 | int c; | ||
340 | |||
341 | while (*p) { | ||
342 | parent = *p; | ||
343 | pg = rb_entry(parent, struct ceph_pg_mapping, node); | ||
344 | c = pgid_cmp(new->pgid, pg->pgid); | ||
345 | if (c < 0) | ||
346 | p = &(*p)->rb_left; | ||
347 | else if (c > 0) | ||
348 | p = &(*p)->rb_right; | ||
349 | else | ||
350 | return -EEXIST; | ||
351 | } | ||
352 | |||
353 | rb_link_node(&new->node, parent, p); | ||
354 | rb_insert_color(&new->node, root); | ||
355 | return 0; | ||
356 | } | ||
357 | |||
358 | static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root, | ||
359 | struct ceph_pg pgid) | ||
360 | { | ||
361 | struct rb_node *n = root->rb_node; | ||
362 | struct ceph_pg_mapping *pg; | ||
363 | int c; | ||
364 | |||
365 | while (n) { | ||
366 | pg = rb_entry(n, struct ceph_pg_mapping, node); | ||
367 | c = pgid_cmp(pgid, pg->pgid); | ||
368 | if (c < 0) | ||
369 | n = n->rb_left; | ||
370 | else if (c > 0) | ||
371 | n = n->rb_right; | ||
372 | else | ||
373 | return pg; | ||
374 | } | ||
375 | return NULL; | ||
376 | } | ||
377 | |||
378 | /* | ||
379 | * rbtree of pg pool info | ||
380 | */ | ||
381 | static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new) | ||
382 | { | ||
383 | struct rb_node **p = &root->rb_node; | ||
384 | struct rb_node *parent = NULL; | ||
385 | struct ceph_pg_pool_info *pi = NULL; | ||
386 | |||
387 | while (*p) { | ||
388 | parent = *p; | ||
389 | pi = rb_entry(parent, struct ceph_pg_pool_info, node); | ||
390 | if (new->id < pi->id) | ||
391 | p = &(*p)->rb_left; | ||
392 | else if (new->id > pi->id) | ||
393 | p = &(*p)->rb_right; | ||
394 | else | ||
395 | return -EEXIST; | ||
396 | } | ||
397 | |||
398 | rb_link_node(&new->node, parent, p); | ||
399 | rb_insert_color(&new->node, root); | ||
400 | return 0; | ||
401 | } | ||
402 | |||
403 | static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, int id) | ||
404 | { | ||
405 | struct ceph_pg_pool_info *pi; | ||
406 | struct rb_node *n = root->rb_node; | ||
407 | |||
408 | while (n) { | ||
409 | pi = rb_entry(n, struct ceph_pg_pool_info, node); | ||
410 | if (id < pi->id) | ||
411 | n = n->rb_left; | ||
412 | else if (id > pi->id) | ||
413 | n = n->rb_right; | ||
414 | else | ||
415 | return pi; | ||
416 | } | ||
417 | return NULL; | ||
418 | } | ||
419 | |||
420 | static void __remove_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *pi) | ||
421 | { | ||
422 | rb_erase(&pi->node, root); | ||
423 | kfree(pi->name); | ||
424 | kfree(pi); | ||
425 | } | ||
426 | |||
427 | void __decode_pool(void **p, struct ceph_pg_pool_info *pi) | ||
428 | { | ||
429 | ceph_decode_copy(p, &pi->v, sizeof(pi->v)); | ||
430 | calc_pg_masks(pi); | ||
431 | *p += le32_to_cpu(pi->v.num_snaps) * sizeof(u64); | ||
432 | *p += le32_to_cpu(pi->v.num_removed_snap_intervals) * sizeof(u64) * 2; | ||
433 | } | ||
434 | |||
435 | static int __decode_pool_names(void **p, void *end, struct ceph_osdmap *map) | ||
436 | { | ||
437 | struct ceph_pg_pool_info *pi; | ||
438 | u32 num, len, pool; | ||
439 | |||
440 | ceph_decode_32_safe(p, end, num, bad); | ||
441 | dout(" %d pool names\n", num); | ||
442 | while (num--) { | ||
443 | ceph_decode_32_safe(p, end, pool, bad); | ||
444 | ceph_decode_32_safe(p, end, len, bad); | ||
445 | dout(" pool %d len %d\n", pool, len); | ||
446 | pi = __lookup_pg_pool(&map->pg_pools, pool); | ||
447 | if (pi) { | ||
448 | kfree(pi->name); | ||
449 | pi->name = kmalloc(len + 1, GFP_NOFS); | ||
450 | if (pi->name) { | ||
451 | memcpy(pi->name, *p, len); | ||
452 | pi->name[len] = '\0'; | ||
453 | dout(" name is %s\n", pi->name); | ||
454 | } | ||
455 | } | ||
456 | *p += len; | ||
457 | } | ||
458 | return 0; | ||
459 | |||
460 | bad: | ||
461 | return -EINVAL; | ||
462 | } | ||
463 | |||
464 | /* | ||
465 | * osd map | ||
466 | */ | ||
467 | void ceph_osdmap_destroy(struct ceph_osdmap *map) | ||
468 | { | ||
469 | dout("osdmap_destroy %p\n", map); | ||
470 | if (map->crush) | ||
471 | crush_destroy(map->crush); | ||
472 | while (!RB_EMPTY_ROOT(&map->pg_temp)) { | ||
473 | struct ceph_pg_mapping *pg = | ||
474 | rb_entry(rb_first(&map->pg_temp), | ||
475 | struct ceph_pg_mapping, node); | ||
476 | rb_erase(&pg->node, &map->pg_temp); | ||
477 | kfree(pg); | ||
478 | } | ||
479 | while (!RB_EMPTY_ROOT(&map->pg_pools)) { | ||
480 | struct ceph_pg_pool_info *pi = | ||
481 | rb_entry(rb_first(&map->pg_pools), | ||
482 | struct ceph_pg_pool_info, node); | ||
483 | __remove_pg_pool(&map->pg_pools, pi); | ||
484 | } | ||
485 | kfree(map->osd_state); | ||
486 | kfree(map->osd_weight); | ||
487 | kfree(map->osd_addr); | ||
488 | kfree(map); | ||
489 | } | ||
490 | |||
491 | /* | ||
492 | * adjust max osd value. reallocate arrays. | ||
493 | */ | ||
494 | static int osdmap_set_max_osd(struct ceph_osdmap *map, int max) | ||
495 | { | ||
496 | u8 *state; | ||
497 | struct ceph_entity_addr *addr; | ||
498 | u32 *weight; | ||
499 | |||
500 | state = kcalloc(max, sizeof(*state), GFP_NOFS); | ||
501 | addr = kcalloc(max, sizeof(*addr), GFP_NOFS); | ||
502 | weight = kcalloc(max, sizeof(*weight), GFP_NOFS); | ||
503 | if (state == NULL || addr == NULL || weight == NULL) { | ||
504 | kfree(state); | ||
505 | kfree(addr); | ||
506 | kfree(weight); | ||
507 | return -ENOMEM; | ||
508 | } | ||
509 | |||
510 | /* copy old? */ | ||
511 | if (map->osd_state) { | ||
512 | memcpy(state, map->osd_state, map->max_osd*sizeof(*state)); | ||
513 | memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr)); | ||
514 | memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight)); | ||
515 | kfree(map->osd_state); | ||
516 | kfree(map->osd_addr); | ||
517 | kfree(map->osd_weight); | ||
518 | } | ||
519 | |||
520 | map->osd_state = state; | ||
521 | map->osd_weight = weight; | ||
522 | map->osd_addr = addr; | ||
523 | map->max_osd = max; | ||
524 | return 0; | ||
525 | } | ||
526 | |||
527 | /* | ||
528 | * decode a full map. | ||
529 | */ | ||
530 | struct ceph_osdmap *osdmap_decode(void **p, void *end) | ||
531 | { | ||
532 | struct ceph_osdmap *map; | ||
533 | u16 version; | ||
534 | u32 len, max, i; | ||
535 | u8 ev; | ||
536 | int err = -EINVAL; | ||
537 | void *start = *p; | ||
538 | struct ceph_pg_pool_info *pi; | ||
539 | |||
540 | dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p)); | ||
541 | |||
542 | map = kzalloc(sizeof(*map), GFP_NOFS); | ||
543 | if (map == NULL) | ||
544 | return ERR_PTR(-ENOMEM); | ||
545 | map->pg_temp = RB_ROOT; | ||
546 | |||
547 | ceph_decode_16_safe(p, end, version, bad); | ||
548 | if (version > CEPH_OSDMAP_VERSION) { | ||
549 | pr_warning("got unknown v %d > %d of osdmap\n", version, | ||
550 | CEPH_OSDMAP_VERSION); | ||
551 | goto bad; | ||
552 | } | ||
553 | |||
554 | ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad); | ||
555 | ceph_decode_copy(p, &map->fsid, sizeof(map->fsid)); | ||
556 | map->epoch = ceph_decode_32(p); | ||
557 | ceph_decode_copy(p, &map->created, sizeof(map->created)); | ||
558 | ceph_decode_copy(p, &map->modified, sizeof(map->modified)); | ||
559 | |||
560 | ceph_decode_32_safe(p, end, max, bad); | ||
561 | while (max--) { | ||
562 | ceph_decode_need(p, end, 4 + 1 + sizeof(pi->v), bad); | ||
563 | pi = kzalloc(sizeof(*pi), GFP_NOFS); | ||
564 | if (!pi) | ||
565 | goto bad; | ||
566 | pi->id = ceph_decode_32(p); | ||
567 | ev = ceph_decode_8(p); /* encoding version */ | ||
568 | if (ev > CEPH_PG_POOL_VERSION) { | ||
569 | pr_warning("got unknown v %d > %d of ceph_pg_pool\n", | ||
570 | ev, CEPH_PG_POOL_VERSION); | ||
571 | goto bad; | ||
572 | } | ||
573 | __decode_pool(p, pi); | ||
574 | __insert_pg_pool(&map->pg_pools, pi); | ||
575 | } | ||
576 | |||
577 | if (version >= 5 && __decode_pool_names(p, end, map) < 0) | ||
578 | goto bad; | ||
579 | |||
580 | ceph_decode_32_safe(p, end, map->pool_max, bad); | ||
581 | |||
582 | ceph_decode_32_safe(p, end, map->flags, bad); | ||
583 | |||
584 | max = ceph_decode_32(p); | ||
585 | |||
586 | /* (re)alloc osd arrays */ | ||
587 | err = osdmap_set_max_osd(map, max); | ||
588 | if (err < 0) | ||
589 | goto bad; | ||
590 | dout("osdmap_decode max_osd = %d\n", map->max_osd); | ||
591 | |||
592 | /* osds */ | ||
593 | err = -EINVAL; | ||
594 | ceph_decode_need(p, end, 3*sizeof(u32) + | ||
595 | map->max_osd*(1 + sizeof(*map->osd_weight) + | ||
596 | sizeof(*map->osd_addr)), bad); | ||
597 | *p += 4; /* skip length field (should match max) */ | ||
598 | ceph_decode_copy(p, map->osd_state, map->max_osd); | ||
599 | |||
600 | *p += 4; /* skip length field (should match max) */ | ||
601 | for (i = 0; i < map->max_osd; i++) | ||
602 | map->osd_weight[i] = ceph_decode_32(p); | ||
603 | |||
604 | *p += 4; /* skip length field (should match max) */ | ||
605 | ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr)); | ||
606 | for (i = 0; i < map->max_osd; i++) | ||
607 | ceph_decode_addr(&map->osd_addr[i]); | ||
608 | |||
609 | /* pg_temp */ | ||
610 | ceph_decode_32_safe(p, end, len, bad); | ||
611 | for (i = 0; i < len; i++) { | ||
612 | int n, j; | ||
613 | struct ceph_pg pgid; | ||
614 | struct ceph_pg_mapping *pg; | ||
615 | |||
616 | ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad); | ||
617 | ceph_decode_copy(p, &pgid, sizeof(pgid)); | ||
618 | n = ceph_decode_32(p); | ||
619 | ceph_decode_need(p, end, n * sizeof(u32), bad); | ||
620 | err = -ENOMEM; | ||
621 | pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS); | ||
622 | if (!pg) | ||
623 | goto bad; | ||
624 | pg->pgid = pgid; | ||
625 | pg->len = n; | ||
626 | for (j = 0; j < n; j++) | ||
627 | pg->osds[j] = ceph_decode_32(p); | ||
628 | |||
629 | err = __insert_pg_mapping(pg, &map->pg_temp); | ||
630 | if (err) | ||
631 | goto bad; | ||
632 | dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len); | ||
633 | } | ||
634 | |||
635 | /* crush */ | ||
636 | ceph_decode_32_safe(p, end, len, bad); | ||
637 | dout("osdmap_decode crush len %d from off 0x%x\n", len, | ||
638 | (int)(*p - start)); | ||
639 | ceph_decode_need(p, end, len, bad); | ||
640 | map->crush = crush_decode(*p, end); | ||
641 | *p += len; | ||
642 | if (IS_ERR(map->crush)) { | ||
643 | err = PTR_ERR(map->crush); | ||
644 | map->crush = NULL; | ||
645 | goto bad; | ||
646 | } | ||
647 | |||
648 | /* ignore the rest of the map */ | ||
649 | *p = end; | ||
650 | |||
651 | dout("osdmap_decode done %p %p\n", *p, end); | ||
652 | return map; | ||
653 | |||
654 | bad: | ||
655 | dout("osdmap_decode fail\n"); | ||
656 | ceph_osdmap_destroy(map); | ||
657 | return ERR_PTR(err); | ||
658 | } | ||
659 | |||
660 | /* | ||
661 | * decode and apply an incremental map update. | ||
662 | */ | ||
663 | struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end, | ||
664 | struct ceph_osdmap *map, | ||
665 | struct ceph_messenger *msgr) | ||
666 | { | ||
667 | struct crush_map *newcrush = NULL; | ||
668 | struct ceph_fsid fsid; | ||
669 | u32 epoch = 0; | ||
670 | struct ceph_timespec modified; | ||
671 | u32 len, pool; | ||
672 | __s32 new_pool_max, new_flags, max; | ||
673 | void *start = *p; | ||
674 | int err = -EINVAL; | ||
675 | u16 version; | ||
676 | struct rb_node *rbp; | ||
677 | |||
678 | ceph_decode_16_safe(p, end, version, bad); | ||
679 | if (version > CEPH_OSDMAP_INC_VERSION) { | ||
680 | pr_warning("got unknown v %d > %d of inc osdmap\n", version, | ||
681 | CEPH_OSDMAP_INC_VERSION); | ||
682 | goto bad; | ||
683 | } | ||
684 | |||
685 | ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32), | ||
686 | bad); | ||
687 | ceph_decode_copy(p, &fsid, sizeof(fsid)); | ||
688 | epoch = ceph_decode_32(p); | ||
689 | BUG_ON(epoch != map->epoch+1); | ||
690 | ceph_decode_copy(p, &modified, sizeof(modified)); | ||
691 | new_pool_max = ceph_decode_32(p); | ||
692 | new_flags = ceph_decode_32(p); | ||
693 | |||
694 | /* full map? */ | ||
695 | ceph_decode_32_safe(p, end, len, bad); | ||
696 | if (len > 0) { | ||
697 | dout("apply_incremental full map len %d, %p to %p\n", | ||
698 | len, *p, end); | ||
699 | return osdmap_decode(p, min(*p+len, end)); | ||
700 | } | ||
701 | |||
702 | /* new crush? */ | ||
703 | ceph_decode_32_safe(p, end, len, bad); | ||
704 | if (len > 0) { | ||
705 | dout("apply_incremental new crush map len %d, %p to %p\n", | ||
706 | len, *p, end); | ||
707 | newcrush = crush_decode(*p, min(*p+len, end)); | ||
708 | if (IS_ERR(newcrush)) | ||
709 | return ERR_PTR(PTR_ERR(newcrush)); | ||
710 | } | ||
711 | |||
712 | /* new flags? */ | ||
713 | if (new_flags >= 0) | ||
714 | map->flags = new_flags; | ||
715 | if (new_pool_max >= 0) | ||
716 | map->pool_max = new_pool_max; | ||
717 | |||
718 | ceph_decode_need(p, end, 5*sizeof(u32), bad); | ||
719 | |||
720 | /* new max? */ | ||
721 | max = ceph_decode_32(p); | ||
722 | if (max >= 0) { | ||
723 | err = osdmap_set_max_osd(map, max); | ||
724 | if (err < 0) | ||
725 | goto bad; | ||
726 | } | ||
727 | |||
728 | map->epoch++; | ||
729 | map->modified = map->modified; | ||
730 | if (newcrush) { | ||
731 | if (map->crush) | ||
732 | crush_destroy(map->crush); | ||
733 | map->crush = newcrush; | ||
734 | newcrush = NULL; | ||
735 | } | ||
736 | |||
737 | /* new_pool */ | ||
738 | ceph_decode_32_safe(p, end, len, bad); | ||
739 | while (len--) { | ||
740 | __u8 ev; | ||
741 | struct ceph_pg_pool_info *pi; | ||
742 | |||
743 | ceph_decode_32_safe(p, end, pool, bad); | ||
744 | ceph_decode_need(p, end, 1 + sizeof(pi->v), bad); | ||
745 | ev = ceph_decode_8(p); /* encoding version */ | ||
746 | if (ev > CEPH_PG_POOL_VERSION) { | ||
747 | pr_warning("got unknown v %d > %d of ceph_pg_pool\n", | ||
748 | ev, CEPH_PG_POOL_VERSION); | ||
749 | goto bad; | ||
750 | } | ||
751 | pi = __lookup_pg_pool(&map->pg_pools, pool); | ||
752 | if (!pi) { | ||
753 | pi = kzalloc(sizeof(*pi), GFP_NOFS); | ||
754 | if (!pi) { | ||
755 | err = -ENOMEM; | ||
756 | goto bad; | ||
757 | } | ||
758 | pi->id = pool; | ||
759 | __insert_pg_pool(&map->pg_pools, pi); | ||
760 | } | ||
761 | __decode_pool(p, pi); | ||
762 | } | ||
763 | if (version >= 5 && __decode_pool_names(p, end, map) < 0) | ||
764 | goto bad; | ||
765 | |||
766 | /* old_pool */ | ||
767 | ceph_decode_32_safe(p, end, len, bad); | ||
768 | while (len--) { | ||
769 | struct ceph_pg_pool_info *pi; | ||
770 | |||
771 | ceph_decode_32_safe(p, end, pool, bad); | ||
772 | pi = __lookup_pg_pool(&map->pg_pools, pool); | ||
773 | if (pi) | ||
774 | __remove_pg_pool(&map->pg_pools, pi); | ||
775 | } | ||
776 | |||
777 | /* new_up */ | ||
778 | err = -EINVAL; | ||
779 | ceph_decode_32_safe(p, end, len, bad); | ||
780 | while (len--) { | ||
781 | u32 osd; | ||
782 | struct ceph_entity_addr addr; | ||
783 | ceph_decode_32_safe(p, end, osd, bad); | ||
784 | ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad); | ||
785 | ceph_decode_addr(&addr); | ||
786 | pr_info("osd%d up\n", osd); | ||
787 | BUG_ON(osd >= map->max_osd); | ||
788 | map->osd_state[osd] |= CEPH_OSD_UP; | ||
789 | map->osd_addr[osd] = addr; | ||
790 | } | ||
791 | |||
792 | /* new_down */ | ||
793 | ceph_decode_32_safe(p, end, len, bad); | ||
794 | while (len--) { | ||
795 | u32 osd; | ||
796 | ceph_decode_32_safe(p, end, osd, bad); | ||
797 | (*p)++; /* clean flag */ | ||
798 | pr_info("osd%d down\n", osd); | ||
799 | if (osd < map->max_osd) | ||
800 | map->osd_state[osd] &= ~CEPH_OSD_UP; | ||
801 | } | ||
802 | |||
803 | /* new_weight */ | ||
804 | ceph_decode_32_safe(p, end, len, bad); | ||
805 | while (len--) { | ||
806 | u32 osd, off; | ||
807 | ceph_decode_need(p, end, sizeof(u32)*2, bad); | ||
808 | osd = ceph_decode_32(p); | ||
809 | off = ceph_decode_32(p); | ||
810 | pr_info("osd%d weight 0x%x %s\n", osd, off, | ||
811 | off == CEPH_OSD_IN ? "(in)" : | ||
812 | (off == CEPH_OSD_OUT ? "(out)" : "")); | ||
813 | if (osd < map->max_osd) | ||
814 | map->osd_weight[osd] = off; | ||
815 | } | ||
816 | |||
817 | /* new_pg_temp */ | ||
818 | rbp = rb_first(&map->pg_temp); | ||
819 | ceph_decode_32_safe(p, end, len, bad); | ||
820 | while (len--) { | ||
821 | struct ceph_pg_mapping *pg; | ||
822 | int j; | ||
823 | struct ceph_pg pgid; | ||
824 | u32 pglen; | ||
825 | ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad); | ||
826 | ceph_decode_copy(p, &pgid, sizeof(pgid)); | ||
827 | pglen = ceph_decode_32(p); | ||
828 | |||
829 | /* remove any? */ | ||
830 | while (rbp && pgid_cmp(rb_entry(rbp, struct ceph_pg_mapping, | ||
831 | node)->pgid, pgid) <= 0) { | ||
832 | struct rb_node *cur = rbp; | ||
833 | rbp = rb_next(rbp); | ||
834 | dout(" removed pg_temp %llx\n", | ||
835 | *(u64 *)&rb_entry(cur, struct ceph_pg_mapping, | ||
836 | node)->pgid); | ||
837 | rb_erase(cur, &map->pg_temp); | ||
838 | } | ||
839 | |||
840 | if (pglen) { | ||
841 | /* insert */ | ||
842 | ceph_decode_need(p, end, pglen*sizeof(u32), bad); | ||
843 | pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS); | ||
844 | if (!pg) { | ||
845 | err = -ENOMEM; | ||
846 | goto bad; | ||
847 | } | ||
848 | pg->pgid = pgid; | ||
849 | pg->len = pglen; | ||
850 | for (j = 0; j < pglen; j++) | ||
851 | pg->osds[j] = ceph_decode_32(p); | ||
852 | err = __insert_pg_mapping(pg, &map->pg_temp); | ||
853 | if (err) | ||
854 | goto bad; | ||
855 | dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, | ||
856 | pglen); | ||
857 | } | ||
858 | } | ||
859 | while (rbp) { | ||
860 | struct rb_node *cur = rbp; | ||
861 | rbp = rb_next(rbp); | ||
862 | dout(" removed pg_temp %llx\n", | ||
863 | *(u64 *)&rb_entry(cur, struct ceph_pg_mapping, | ||
864 | node)->pgid); | ||
865 | rb_erase(cur, &map->pg_temp); | ||
866 | } | ||
867 | |||
868 | /* ignore the rest */ | ||
869 | *p = end; | ||
870 | return map; | ||
871 | |||
872 | bad: | ||
873 | pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n", | ||
874 | epoch, (int)(*p - start), *p, start, end); | ||
875 | print_hex_dump(KERN_DEBUG, "osdmap: ", | ||
876 | DUMP_PREFIX_OFFSET, 16, 1, | ||
877 | start, end - start, true); | ||
878 | if (newcrush) | ||
879 | crush_destroy(newcrush); | ||
880 | return ERR_PTR(err); | ||
881 | } | ||
882 | |||
883 | |||
884 | |||
885 | |||
886 | /* | ||
887 | * calculate file layout from given offset, length. | ||
888 | * fill in correct oid, logical length, and object extent | ||
889 | * offset, length. | ||
890 | * | ||
891 | * for now, we write only a single su, until we can | ||
892 | * pass a stride back to the caller. | ||
893 | */ | ||
894 | void ceph_calc_file_object_mapping(struct ceph_file_layout *layout, | ||
895 | u64 off, u64 *plen, | ||
896 | u64 *ono, | ||
897 | u64 *oxoff, u64 *oxlen) | ||
898 | { | ||
899 | u32 osize = le32_to_cpu(layout->fl_object_size); | ||
900 | u32 su = le32_to_cpu(layout->fl_stripe_unit); | ||
901 | u32 sc = le32_to_cpu(layout->fl_stripe_count); | ||
902 | u32 bl, stripeno, stripepos, objsetno; | ||
903 | u32 su_per_object; | ||
904 | u64 t, su_offset; | ||
905 | |||
906 | dout("mapping %llu~%llu osize %u fl_su %u\n", off, *plen, | ||
907 | osize, su); | ||
908 | su_per_object = osize / su; | ||
909 | dout("osize %u / su %u = su_per_object %u\n", osize, su, | ||
910 | su_per_object); | ||
911 | |||
912 | BUG_ON((su & ~PAGE_MASK) != 0); | ||
913 | /* bl = *off / su; */ | ||
914 | t = off; | ||
915 | do_div(t, su); | ||
916 | bl = t; | ||
917 | dout("off %llu / su %u = bl %u\n", off, su, bl); | ||
918 | |||
919 | stripeno = bl / sc; | ||
920 | stripepos = bl % sc; | ||
921 | objsetno = stripeno / su_per_object; | ||
922 | |||
923 | *ono = objsetno * sc + stripepos; | ||
924 | dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono); | ||
925 | |||
926 | /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */ | ||
927 | t = off; | ||
928 | su_offset = do_div(t, su); | ||
929 | *oxoff = su_offset + (stripeno % su_per_object) * su; | ||
930 | |||
931 | /* | ||
932 | * Calculate the length of the extent being written to the selected | ||
933 | * object. This is the minimum of the full length requested (plen) or | ||
934 | * the remainder of the current stripe being written to. | ||
935 | */ | ||
936 | *oxlen = min_t(u64, *plen, su - su_offset); | ||
937 | *plen = *oxlen; | ||
938 | |||
939 | dout(" obj extent %llu~%llu\n", *oxoff, *oxlen); | ||
940 | } | ||
941 | |||
942 | /* | ||
943 | * calculate an object layout (i.e. pgid) from an oid, | ||
944 | * file_layout, and osdmap | ||
945 | */ | ||
946 | int ceph_calc_object_layout(struct ceph_object_layout *ol, | ||
947 | const char *oid, | ||
948 | struct ceph_file_layout *fl, | ||
949 | struct ceph_osdmap *osdmap) | ||
950 | { | ||
951 | unsigned num, num_mask; | ||
952 | struct ceph_pg pgid; | ||
953 | s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred); | ||
954 | int poolid = le32_to_cpu(fl->fl_pg_pool); | ||
955 | struct ceph_pg_pool_info *pool; | ||
956 | unsigned ps; | ||
957 | |||
958 | BUG_ON(!osdmap); | ||
959 | |||
960 | pool = __lookup_pg_pool(&osdmap->pg_pools, poolid); | ||
961 | if (!pool) | ||
962 | return -EIO; | ||
963 | ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid)); | ||
964 | if (preferred >= 0) { | ||
965 | ps += preferred; | ||
966 | num = le32_to_cpu(pool->v.lpg_num); | ||
967 | num_mask = pool->lpg_num_mask; | ||
968 | } else { | ||
969 | num = le32_to_cpu(pool->v.pg_num); | ||
970 | num_mask = pool->pg_num_mask; | ||
971 | } | ||
972 | |||
973 | pgid.ps = cpu_to_le16(ps); | ||
974 | pgid.preferred = cpu_to_le16(preferred); | ||
975 | pgid.pool = fl->fl_pg_pool; | ||
976 | if (preferred >= 0) | ||
977 | dout("calc_object_layout '%s' pgid %d.%xp%d\n", oid, poolid, ps, | ||
978 | (int)preferred); | ||
979 | else | ||
980 | dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps); | ||
981 | |||
982 | ol->ol_pgid = pgid; | ||
983 | ol->ol_stripe_unit = fl->fl_object_stripe_unit; | ||
984 | return 0; | ||
985 | } | ||
986 | |||
987 | /* | ||
988 | * Calculate raw osd vector for the given pgid. Return pointer to osd | ||
989 | * array, or NULL on failure. | ||
990 | */ | ||
991 | static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid, | ||
992 | int *osds, int *num) | ||
993 | { | ||
994 | struct ceph_pg_mapping *pg; | ||
995 | struct ceph_pg_pool_info *pool; | ||
996 | int ruleno; | ||
997 | unsigned poolid, ps, pps; | ||
998 | int preferred; | ||
999 | |||
1000 | /* pg_temp? */ | ||
1001 | pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid); | ||
1002 | if (pg) { | ||
1003 | *num = pg->len; | ||
1004 | return pg->osds; | ||
1005 | } | ||
1006 | |||
1007 | /* crush */ | ||
1008 | poolid = le32_to_cpu(pgid.pool); | ||
1009 | ps = le16_to_cpu(pgid.ps); | ||
1010 | preferred = (s16)le16_to_cpu(pgid.preferred); | ||
1011 | |||
1012 | /* don't forcefeed bad device ids to crush */ | ||
1013 | if (preferred >= osdmap->max_osd || | ||
1014 | preferred >= osdmap->crush->max_devices) | ||
1015 | preferred = -1; | ||
1016 | |||
1017 | pool = __lookup_pg_pool(&osdmap->pg_pools, poolid); | ||
1018 | if (!pool) | ||
1019 | return NULL; | ||
1020 | ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset, | ||
1021 | pool->v.type, pool->v.size); | ||
1022 | if (ruleno < 0) { | ||
1023 | pr_err("no crush rule pool %d type %d size %d\n", | ||
1024 | poolid, pool->v.type, pool->v.size); | ||
1025 | return NULL; | ||
1026 | } | ||
1027 | |||
1028 | if (preferred >= 0) | ||
1029 | pps = ceph_stable_mod(ps, | ||
1030 | le32_to_cpu(pool->v.lpgp_num), | ||
1031 | pool->lpgp_num_mask); | ||
1032 | else | ||
1033 | pps = ceph_stable_mod(ps, | ||
1034 | le32_to_cpu(pool->v.pgp_num), | ||
1035 | pool->pgp_num_mask); | ||
1036 | pps += poolid; | ||
1037 | *num = crush_do_rule(osdmap->crush, ruleno, pps, osds, | ||
1038 | min_t(int, pool->v.size, *num), | ||
1039 | preferred, osdmap->osd_weight); | ||
1040 | return osds; | ||
1041 | } | ||
1042 | |||
1043 | /* | ||
1044 | * Return primary osd for given pgid, or -1 if none. | ||
1045 | */ | ||
1046 | int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid) | ||
1047 | { | ||
1048 | int rawosds[10], *osds; | ||
1049 | int i, num = ARRAY_SIZE(rawosds); | ||
1050 | |||
1051 | osds = calc_pg_raw(osdmap, pgid, rawosds, &num); | ||
1052 | if (!osds) | ||
1053 | return -1; | ||
1054 | |||
1055 | /* primary is first up osd */ | ||
1056 | for (i = 0; i < num; i++) | ||
1057 | if (ceph_osd_is_up(osdmap, osds[i])) { | ||
1058 | return osds[i]; | ||
1059 | break; | ||
1060 | } | ||
1061 | return -1; | ||
1062 | } | ||