projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
ceph: fix iterate_caps removal race
[linux-2.6-block.git] / fs / ceph / osdmap.c
CommitLineData
f24e9980
SW		 1
2#include <asm/div64.h>
3
4#include "super.h"
5#include "osdmap.h"
6#include "crush/hash.h"
7#include "crush/mapper.h"
8#include "decode.h"
9#include "ceph_debug.h"
10
11char *ceph_osdmap_state_str(char *str, int len, int state)
12{
13 int flag = 0;
14
15 if (!len)
16 goto done;
17
18 *str = '\0';
19 if (state) {
20 if (state & CEPH_OSD_EXISTS) {
21 snprintf(str, len, "exists");
22 flag = 1;
23 }
24 if (state & CEPH_OSD_UP) {
25 snprintf(str, len, "%s%s%s", str, (flag ? ", " : ""),
26 "up");
27 flag = 1;
28 }
29 } else {
30 snprintf(str, len, "doesn't exist");
31 }
32done:
33 return str;
34}
35
36/* maps */
37
38static int calc_bits_of(unsigned t)
39{
40 int b = 0;
41 while (t) {
42 t = t >> 1;
43 b++;
44 }
45 return b;
46}
47
48/*
49 * the foo_mask is the smallest value 2^n-1 that is >= foo.
50 */
51static void calc_pg_masks(struct ceph_pg_pool_info *pi)
52{
53 pi->pg_num_mask = (1 << calc_bits_of(le32_to_cpu(pi->v.pg_num)-1)) - 1;
54 pi->pgp_num_mask =
55 (1 << calc_bits_of(le32_to_cpu(pi->v.pgp_num)-1)) - 1;
56 pi->lpg_num_mask =
57 (1 << calc_bits_of(le32_to_cpu(pi->v.lpg_num)-1)) - 1;
58 pi->lpgp_num_mask =
59 (1 << calc_bits_of(le32_to_cpu(pi->v.lpgp_num)-1)) - 1;
60}
61
62/*
63 * decode crush map
64 */
65static int crush_decode_uniform_bucket(void **p, void *end,
66 struct crush_bucket_uniform *b)
67{
68 dout("crush_decode_uniform_bucket %p to %p\n", *p, end);
69 ceph_decode_need(p, end, (1+b->h.size) * sizeof(u32), bad);
c89136ea 70 b->item_weight = ceph_decode_32(p);
f24e9980
SW		 71 return 0;
72bad:
73 return -EINVAL;
74}
75
76static int crush_decode_list_bucket(void **p, void *end,
77 struct crush_bucket_list *b)
78{
79 int j;
80 dout("crush_decode_list_bucket %p to %p\n", *p, end);
81 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
82 if (b->item_weights == NULL)
83 return -ENOMEM;
84 b->sum_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
85 if (b->sum_weights == NULL)
86 return -ENOMEM;
87 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
88 for (j = 0; j < b->h.size; j++) {
c89136ea
SW		 89 b->item_weights[j] = ceph_decode_32(p);
90 b->sum_weights[j] = ceph_decode_32(p);
f24e9980
SW		 91 }
92 return 0;
93bad:
94 return -EINVAL;
95}
96
97static int crush_decode_tree_bucket(void **p, void *end,
98 struct crush_bucket_tree *b)
99{
100 int j;
101 dout("crush_decode_tree_bucket %p to %p\n", *p, end);
102 ceph_decode_32_safe(p, end, b->num_nodes, bad);
103 b->node_weights = kcalloc(b->num_nodes, sizeof(u32), GFP_NOFS);
104 if (b->node_weights == NULL)
105 return -ENOMEM;
106 ceph_decode_need(p, end, b->num_nodes * sizeof(u32), bad);
107 for (j = 0; j < b->num_nodes; j++)
c89136ea 108 b->node_weights[j] = ceph_decode_32(p);
f24e9980
SW		 109 return 0;
110bad:
111 return -EINVAL;
112}
113
114static int crush_decode_straw_bucket(void **p, void *end,
115 struct crush_bucket_straw *b)
116{
117 int j;
118 dout("crush_decode_straw_bucket %p to %p\n", *p, end);
119 b->item_weights = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
120 if (b->item_weights == NULL)
121 return -ENOMEM;
122 b->straws = kcalloc(b->h.size, sizeof(u32), GFP_NOFS);
123 if (b->straws == NULL)
124 return -ENOMEM;
125 ceph_decode_need(p, end, 2 * b->h.size * sizeof(u32), bad);
126 for (j = 0; j < b->h.size; j++) {
c89136ea
SW		 127 b->item_weights[j] = ceph_decode_32(p);
128 b->straws[j] = ceph_decode_32(p);
f24e9980
SW		 129 }
130 return 0;
131bad:
132 return -EINVAL;
133}
134
135static struct crush_map *crush_decode(void *pbyval, void *end)
136{
137 struct crush_map *c;
138 int err = -EINVAL;
139 int i, j;
140 void **p = &pbyval;
141 void *start = pbyval;
142 u32 magic;
143
144 dout("crush_decode %p to %p len %d\n", *p, end, (int)(end - *p));
145
146 c = kzalloc(sizeof(*c), GFP_NOFS);
147 if (c == NULL)
148 return ERR_PTR(-ENOMEM);
149
150 ceph_decode_need(p, end, 4*sizeof(u32), bad);
c89136ea 151 magic = ceph_decode_32(p);
f24e9980
SW		 152 if (magic != CRUSH_MAGIC) {
153 pr_err("crush_decode magic %x != current %x\n",
154 (unsigned)magic, (unsigned)CRUSH_MAGIC);
155 goto bad;
156 }
c89136ea
SW		 157 c->max_buckets = ceph_decode_32(p);
158 c->max_rules = ceph_decode_32(p);
159 c->max_devices = ceph_decode_32(p);
f24e9980
SW		 160
161 c->device_parents = kcalloc(c->max_devices, sizeof(u32), GFP_NOFS);
162 if (c->device_parents == NULL)
163 goto badmem;
164 c->bucket_parents = kcalloc(c->max_buckets, sizeof(u32), GFP_NOFS);
165 if (c->bucket_parents == NULL)
166 goto badmem;
167
168 c->buckets = kcalloc(c->max_buckets, sizeof(*c->buckets), GFP_NOFS);
169 if (c->buckets == NULL)
170 goto badmem;
171 c->rules = kcalloc(c->max_rules, sizeof(*c->rules), GFP_NOFS);
172 if (c->rules == NULL)
173 goto badmem;
174
175 /* buckets */
176 for (i = 0; i < c->max_buckets; i++) {
177 int size = 0;
178 u32 alg;
179 struct crush_bucket *b;
180
181 ceph_decode_32_safe(p, end, alg, bad);
182 if (alg == 0) {
183 c->buckets[i] = NULL;
184 continue;
185 }
186 dout("crush_decode bucket %d off %x %p to %p\n",
187 i, (int)(*p-start), *p, end);
188
189 switch (alg) {
190 case CRUSH_BUCKET_UNIFORM:
191 size = sizeof(struct crush_bucket_uniform);
192 break;
193 case CRUSH_BUCKET_LIST:
194 size = sizeof(struct crush_bucket_list);
195 break;
196 case CRUSH_BUCKET_TREE:
197 size = sizeof(struct crush_bucket_tree);
198 break;
199 case CRUSH_BUCKET_STRAW:
200 size = sizeof(struct crush_bucket_straw);
201 break;
202 default:
30dc6381 203 err = -EINVAL;
f24e9980
SW		 204 goto bad;
205 }
206 BUG_ON(size == 0);
207 b = c->buckets[i] = kzalloc(size, GFP_NOFS);
208 if (b == NULL)
209 goto badmem;
210
211 ceph_decode_need(p, end, 4*sizeof(u32), bad);
c89136ea
SW		 212 b->id = ceph_decode_32(p);
213 b->type = ceph_decode_16(p);
fb690390
SW		 214 b->alg = ceph_decode_8(p);
215 b->hash = ceph_decode_8(p);
c89136ea
SW		 216 b->weight = ceph_decode_32(p);
217 b->size = ceph_decode_32(p);
f24e9980
SW		 218
219 dout("crush_decode bucket size %d off %x %p to %p\n",
220 b->size, (int)(*p-start), *p, end);
221
222 b->items = kcalloc(b->size, sizeof(__s32), GFP_NOFS);
223 if (b->items == NULL)
224 goto badmem;
225 b->perm = kcalloc(b->size, sizeof(u32), GFP_NOFS);
226 if (b->perm == NULL)
227 goto badmem;
228 b->perm_n = 0;
229
230 ceph_decode_need(p, end, b->size*sizeof(u32), bad);
231 for (j = 0; j < b->size; j++)
c89136ea 232 b->items[j] = ceph_decode_32(p);
f24e9980
SW		 233
234 switch (b->alg) {
235 case CRUSH_BUCKET_UNIFORM:
236 err = crush_decode_uniform_bucket(p, end,
237 (struct crush_bucket_uniform *)b);
238 if (err < 0)
239 goto bad;
240 break;
241 case CRUSH_BUCKET_LIST:
242 err = crush_decode_list_bucket(p, end,
243 (struct crush_bucket_list *)b);
244 if (err < 0)
245 goto bad;
246 break;
247 case CRUSH_BUCKET_TREE:
248 err = crush_decode_tree_bucket(p, end,
249 (struct crush_bucket_tree *)b);
250 if (err < 0)
251 goto bad;
252 break;
253 case CRUSH_BUCKET_STRAW:
254 err = crush_decode_straw_bucket(p, end,
255 (struct crush_bucket_straw *)b);
256 if (err < 0)
257 goto bad;
258 break;
259 }
260 }
261
262 /* rules */
263 dout("rule vec is %p\n", c->rules);
264 for (i = 0; i < c->max_rules; i++) {
265 u32 yes;
266 struct crush_rule *r;
267
268 ceph_decode_32_safe(p, end, yes, bad);
269 if (!yes) {
270 dout("crush_decode NO rule %d off %x %p to %p\n",
271 i, (int)(*p-start), *p, end);
272 c->rules[i] = NULL;
273 continue;
274 }
275
276 dout("crush_decode rule %d off %x %p to %p\n",
277 i, (int)(*p-start), *p, end);
278
279 /* len */
280 ceph_decode_32_safe(p, end, yes, bad);
281#if BITS_PER_LONG == 32
30dc6381 282 err = -EINVAL;
f24e9980
SW		 283 if (yes > ULONG_MAX / sizeof(struct crush_rule_step))
284 goto bad;
285#endif
286 r = c->rules[i] = kmalloc(sizeof(*r) +
287 yes*sizeof(struct crush_rule_step),
288 GFP_NOFS);
289 if (r == NULL)
290 goto badmem;
291 dout(" rule %d is at %p\n", i, r);
292 r->len = yes;
293 ceph_decode_copy_safe(p, end, &r->mask, 4, bad); /* 4 u8's */
294 ceph_decode_need(p, end, r->len*3*sizeof(u32), bad);
295 for (j = 0; j < r->len; j++) {
c89136ea
SW		 296 r->steps[j].op = ceph_decode_32(p);
297 r->steps[j].arg1 = ceph_decode_32(p);
298 r->steps[j].arg2 = ceph_decode_32(p);
f24e9980
SW		 299 }
300 }
301
302 /* ignore trailing name maps. */
303
304 dout("crush_decode success\n");
305 return c;
306
307badmem:
308 err = -ENOMEM;
309bad:
310 dout("crush_decode fail %d\n", err);
311 crush_destroy(c);
312 return ERR_PTR(err);
313}
314
315
316/*
317 * osd map
318 */
319void ceph_osdmap_destroy(struct ceph_osdmap *map)
320{
321 dout("osdmap_destroy %p\n", map);
322 if (map->crush)
323 crush_destroy(map->crush);
324 while (!RB_EMPTY_ROOT(&map->pg_temp))
325 rb_erase(rb_first(&map->pg_temp), &map->pg_temp);
326 kfree(map->osd_state);
327 kfree(map->osd_weight);
328 kfree(map->pg_pool);
329 kfree(map->osd_addr);
330 kfree(map);
331}
332
333/*
334 * adjust max osd value. reallocate arrays.
335 */
336static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
337{
338 u8 *state;
339 struct ceph_entity_addr *addr;
340 u32 *weight;
341
342 state = kcalloc(max, sizeof(*state), GFP_NOFS);
343 addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
344 weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
345 if (state == NULL || addr == NULL || weight == NULL) {
346 kfree(state);
347 kfree(addr);
348 kfree(weight);
349 return -ENOMEM;
350 }
351
352 /* copy old? */
353 if (map->osd_state) {
354 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
355 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
356 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
357 kfree(map->osd_state);
358 kfree(map->osd_addr);
359 kfree(map->osd_weight);
360 }
361
362 map->osd_state = state;
363 map->osd_weight = weight;
364 map->osd_addr = addr;
365 map->max_osd = max;
366 return 0;
367}
368
369/*
370 * Insert a new pg_temp mapping
371 */
51042122
SW		 372static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
373{
374 u64 a = *(u64 *)&l;
375 u64 b = *(u64 *)&r;
376
377 if (a < b)
378 return -1;
379 if (a > b)
380 return 1;
381 return 0;
382}
383
991abb6e
SW		 384static int __insert_pg_mapping(struct ceph_pg_mapping *new,
385 struct rb_root *root)
f24e9980
SW		 386{
387 struct rb_node **p = &root->rb_node;
388 struct rb_node *parent = NULL;
389 struct ceph_pg_mapping *pg = NULL;
51042122 390 int c;
f24e9980
SW		 391
392 while (*p) {
393 parent = *p;
394 pg = rb_entry(parent, struct ceph_pg_mapping, node);
51042122
SW		 395 c = pgid_cmp(new->pgid, pg->pgid);
396 if (c < 0)
f24e9980 397 p = &(*p)->rb_left;
51042122 398 else if (c > 0)
f24e9980
SW		 399 p = &(*p)->rb_right;
400 else
991abb6e 401 return -EEXIST;
f24e9980
SW		 402 }
403
404 rb_link_node(&new->node, parent, p);
405 rb_insert_color(&new->node, root);
991abb6e 406 return 0;
f24e9980
SW		 407}
408
409/*
410 * decode a full map.
411 */
412struct ceph_osdmap *osdmap_decode(void **p, void *end)
413{
414 struct ceph_osdmap *map;
415 u16 version;
416 u32 len, max, i;
361be860 417 u8 ev;
f24e9980
SW		 418 int err = -EINVAL;
419 void *start = *p;
420
421 dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
422
423 map = kzalloc(sizeof(*map), GFP_NOFS);
424 if (map == NULL)
425 return ERR_PTR(-ENOMEM);
426 map->pg_temp = RB_ROOT;
427
428 ceph_decode_16_safe(p, end, version, bad);
02f90c61
SW		 429 if (version > CEPH_OSDMAP_VERSION) {
430 pr_warning("got unknown v %d > %d of osdmap\n", version,
431 CEPH_OSDMAP_VERSION);
432 goto bad;
433 }
f24e9980
SW		 434
435 ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
436 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
c89136ea 437 map->epoch = ceph_decode_32(p);
f24e9980
SW		 438 ceph_decode_copy(p, &map->created, sizeof(map->created));
439 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
440
c89136ea 441 map->num_pools = ceph_decode_32(p);
f24e9980
SW		 442 map->pg_pool = kcalloc(map->num_pools, sizeof(*map->pg_pool),
443 GFP_NOFS);
444 if (!map->pg_pool) {
445 err = -ENOMEM;
446 goto bad;
447 }
448 ceph_decode_32_safe(p, end, max, bad);
449 while (max--) {
361be860 450 ceph_decode_need(p, end, 4+1+sizeof(map->pg_pool->v), bad);
c89136ea 451 i = ceph_decode_32(p);
f24e9980
SW		 452 if (i >= map->num_pools)
453 goto bad;
361be860 454 ev = ceph_decode_8(p); /* encoding version */
02f90c61
SW		 455 if (ev > CEPH_PG_POOL_VERSION) {
456 pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
457 ev, CEPH_PG_POOL_VERSION);
458 goto bad;
459 }
f24e9980
SW		 460 ceph_decode_copy(p, &map->pg_pool[i].v,
461 sizeof(map->pg_pool->v));
462 calc_pg_masks(&map->pg_pool[i]);
463 p += le32_to_cpu(map->pg_pool[i].v.num_snaps) * sizeof(u64);
464 p += le32_to_cpu(map->pg_pool[i].v.num_removed_snap_intervals)
465 * sizeof(u64) * 2;
466 }
467
468 ceph_decode_32_safe(p, end, map->flags, bad);
469
c89136ea 470 max = ceph_decode_32(p);
f24e9980
SW		 471
472 /* (re)alloc osd arrays */
473 err = osdmap_set_max_osd(map, max);
474 if (err < 0)
475 goto bad;
476 dout("osdmap_decode max_osd = %d\n", map->max_osd);
477
478 /* osds */
479 err = -EINVAL;
480 ceph_decode_need(p, end, 3*sizeof(u32) +
481 map->max_osd*(1 + sizeof(*map->osd_weight) +
482 sizeof(*map->osd_addr)), bad);
483 *p += 4; /* skip length field (should match max) */
484 ceph_decode_copy(p, map->osd_state, map->max_osd);
485
486 *p += 4; /* skip length field (should match max) */
487 for (i = 0; i < map->max_osd; i++)
c89136ea 488 map->osd_weight[i] = ceph_decode_32(p);
f24e9980
SW		 489
490 *p += 4; /* skip length field (should match max) */
491 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
63f2d211
SW		 492 for (i = 0; i < map->max_osd; i++)
493 ceph_decode_addr(&map->osd_addr[i]);
f24e9980
SW		 494
495 /* pg_temp */
496 ceph_decode_32_safe(p, end, len, bad);
497 for (i = 0; i < len; i++) {
498 int n, j;
51042122 499 struct ceph_pg pgid;
f24e9980
SW		 500 struct ceph_pg_mapping *pg;
501
502 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
51042122 503 ceph_decode_copy(p, &pgid, sizeof(pgid));
c89136ea 504 n = ceph_decode_32(p);
f24e9980 505 ceph_decode_need(p, end, n * sizeof(u32), bad);
30dc6381 506 err = -ENOMEM;
f24e9980 507 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
30dc6381 508 if (!pg)
f24e9980 509 goto bad;
f24e9980
SW		 510 pg->pgid = pgid;
511 pg->len = n;
512 for (j = 0; j < n; j++)
c89136ea 513 pg->osds[j] = ceph_decode_32(p);
f24e9980 514
991abb6e
SW		 515 err = __insert_pg_mapping(pg, &map->pg_temp);
516 if (err)
517 goto bad;
51042122 518 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
f24e9980
SW		 519 }
520
521 /* crush */
522 ceph_decode_32_safe(p, end, len, bad);
523 dout("osdmap_decode crush len %d from off 0x%x\n", len,
524 (int)(*p - start));
525 ceph_decode_need(p, end, len, bad);
526 map->crush = crush_decode(*p, end);
527 *p += len;
528 if (IS_ERR(map->crush)) {
529 err = PTR_ERR(map->crush);
530 map->crush = NULL;
531 goto bad;
532 }
533
534 /* ignore the rest of the map */
535 *p = end;
536
537 dout("osdmap_decode done %p %p\n", *p, end);
538 return map;
539
540bad:
541 dout("osdmap_decode fail\n");
542 ceph_osdmap_destroy(map);
543 return ERR_PTR(err);
544}
545
546/*
547 * decode and apply an incremental map update.
548 */
549struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
550 struct ceph_osdmap *map,
551 struct ceph_messenger *msgr)
552{
f24e9980
SW		 553 struct crush_map *newcrush = NULL;
554 struct ceph_fsid fsid;
555 u32 epoch = 0;
556 struct ceph_timespec modified;
557 u32 len, pool;
558 __s32 new_flags, max;
559 void *start = *p;
560 int err = -EINVAL;
561 u16 version;
562 struct rb_node *rbp;
563
564 ceph_decode_16_safe(p, end, version, bad);
02f90c61
SW		 565 if (version > CEPH_OSDMAP_INC_VERSION) {
566 pr_warning("got unknown v %d > %d of inc osdmap\n", version,
567 CEPH_OSDMAP_INC_VERSION);
568 goto bad;
569 }
f24e9980
SW		 570
571 ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
572 bad);
573 ceph_decode_copy(p, &fsid, sizeof(fsid));
c89136ea 574 epoch = ceph_decode_32(p);
f24e9980
SW		 575 BUG_ON(epoch != map->epoch+1);
576 ceph_decode_copy(p, &modified, sizeof(modified));
c89136ea 577 new_flags = ceph_decode_32(p);
f24e9980
SW		 578
579 /* full map? */
580 ceph_decode_32_safe(p, end, len, bad);
581 if (len > 0) {
582 dout("apply_incremental full map len %d, %p to %p\n",
583 len, *p, end);
30dc6381 584 return osdmap_decode(p, min(*p+len, end));
f24e9980
SW		 585 }
586
587 /* new crush? */
588 ceph_decode_32_safe(p, end, len, bad);
589 if (len > 0) {
590 dout("apply_incremental new crush map len %d, %p to %p\n",
591 len, *p, end);
592 newcrush = crush_decode(*p, min(*p+len, end));
593 if (IS_ERR(newcrush))
594 return ERR_PTR(PTR_ERR(newcrush));
595 }
596
597 /* new flags? */
598 if (new_flags >= 0)
599 map->flags = new_flags;
600
601 ceph_decode_need(p, end, 5*sizeof(u32), bad);
602
603 /* new max? */
c89136ea 604 max = ceph_decode_32(p);
f24e9980
SW		 605 if (max >= 0) {
606 err = osdmap_set_max_osd(map, max);
607 if (err < 0)
608 goto bad;
609 }
610
611 map->epoch++;
612 map->modified = map->modified;
613 if (newcrush) {
614 if (map->crush)
615 crush_destroy(map->crush);
616 map->crush = newcrush;
617 newcrush = NULL;
618 }
619
620 /* new_pool */
621 ceph_decode_32_safe(p, end, len, bad);
622 while (len--) {
361be860
SW		 623 __u8 ev;
624
f24e9980
SW		 625 ceph_decode_32_safe(p, end, pool, bad);
626 if (pool >= map->num_pools) {
627 void *pg_pool = kcalloc(pool + 1,
628 sizeof(*map->pg_pool),
629 GFP_NOFS);
630 if (!pg_pool) {
631 err = -ENOMEM;
632 goto bad;
633 }
634 memcpy(pg_pool, map->pg_pool,
635 map->num_pools * sizeof(*map->pg_pool));
636 kfree(map->pg_pool);
637 map->pg_pool = pg_pool;
638 map->num_pools = pool+1;
639 }
361be860
SW		 640 ceph_decode_need(p, end, 1 + sizeof(map->pg_pool->v), bad);
641 ev = ceph_decode_8(p); /* encoding version */
02f90c61
SW		 642 if (ev > CEPH_PG_POOL_VERSION) {
643 pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
644 ev, CEPH_PG_POOL_VERSION);
645 goto bad;
646 }
f24e9980
SW		 647 ceph_decode_copy(p, &map->pg_pool[pool].v,
648 sizeof(map->pg_pool->v));
649 calc_pg_masks(&map->pg_pool[pool]);
650 }
651
652 /* old_pool (ignore) */
653 ceph_decode_32_safe(p, end, len, bad);
654 *p += len * sizeof(u32);
655
656 /* new_up */
657 err = -EINVAL;
658 ceph_decode_32_safe(p, end, len, bad);
659 while (len--) {
660 u32 osd;
661 struct ceph_entity_addr addr;
662 ceph_decode_32_safe(p, end, osd, bad);
663 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
63f2d211 664 ceph_decode_addr(&addr);
f24e9980
SW		 665 pr_info("osd%d up\n", osd);
666 BUG_ON(osd >= map->max_osd);
667 map->osd_state[osd] |= CEPH_OSD_UP;
668 map->osd_addr[osd] = addr;
669 }
670
671 /* new_down */
672 ceph_decode_32_safe(p, end, len, bad);
673 while (len--) {
674 u32 osd;
675 ceph_decode_32_safe(p, end, osd, bad);
676 (*p)++; /* clean flag */
1bdb70e5 677 pr_info("osd%d down\n", osd);
f24e9980
SW		 678 if (osd < map->max_osd)
679 map->osd_state[osd] &= ~CEPH_OSD_UP;
680 }
681
682 /* new_weight */
683 ceph_decode_32_safe(p, end, len, bad);
684 while (len--) {
685 u32 osd, off;
686 ceph_decode_need(p, end, sizeof(u32)*2, bad);
c89136ea
SW		 687 osd = ceph_decode_32(p);
688 off = ceph_decode_32(p);
f24e9980
SW		 689 pr_info("osd%d weight 0x%x %s\n", osd, off,
690 off == CEPH_OSD_IN ? "(in)" :
691 (off == CEPH_OSD_OUT ? "(out)" : ""));
692 if (osd < map->max_osd)
693 map->osd_weight[osd] = off;
694 }
695
696 /* new_pg_temp */
697 rbp = rb_first(&map->pg_temp);
698 ceph_decode_32_safe(p, end, len, bad);
699 while (len--) {
700 struct ceph_pg_mapping *pg;
701 int j;
51042122 702 struct ceph_pg pgid;
f24e9980
SW		 703 u32 pglen;
704 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
51042122 705 ceph_decode_copy(p, &pgid, sizeof(pgid));
c89136ea 706 pglen = ceph_decode_32(p);
f24e9980
SW		 707
708 /* remove any? */
51042122
SW		 709 while (rbp && pgid_cmp(rb_entry(rbp, struct ceph_pg_mapping,
710 node)->pgid, pgid) <= 0) {
f24e9980
SW		 711 struct rb_node *cur = rbp;
712 rbp = rb_next(rbp);
713 dout(" removed pg_temp %llx\n",
51042122
SW		 714 *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
715 node)->pgid);
f24e9980
SW		 716 rb_erase(cur, &map->pg_temp);
717 }
718
719 if (pglen) {
720 /* insert */
721 ceph_decode_need(p, end, pglen*sizeof(u32), bad);
722 pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
723 if (!pg) {
724 err = -ENOMEM;
725 goto bad;
726 }
727 pg->pgid = pgid;
728 pg->len = pglen;
7067f797 729 for (j = 0; j < pglen; j++)
c89136ea 730 pg->osds[j] = ceph_decode_32(p);
991abb6e
SW		 731 err = __insert_pg_mapping(pg, &map->pg_temp);
732 if (err)
733 goto bad;
51042122
SW		 734 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
735 pglen);
f24e9980
SW		 736 }
737 }
738 while (rbp) {
739 struct rb_node *cur = rbp;
740 rbp = rb_next(rbp);
741 dout(" removed pg_temp %llx\n",
51042122
SW		 742 *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
743 node)->pgid);
f24e9980
SW		 744 rb_erase(cur, &map->pg_temp);
745 }
746
747 /* ignore the rest */
748 *p = end;
749 return map;
750
751bad:
752 pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
753 epoch, (int)(*p - start), *p, start, end);
9ec7cab1
SW		 754 print_hex_dump(KERN_DEBUG, "osdmap: ",
755 DUMP_PREFIX_OFFSET, 16, 1,
756 start, end - start, true);
f24e9980
SW		 757 if (newcrush)
758 crush_destroy(newcrush);
759 return ERR_PTR(err);
760}
761
762
763
764
765/*
766 * calculate file layout from given offset, length.
767 * fill in correct oid, logical length, and object extent
768 * offset, length.
769 *
770 * for now, we write only a single su, until we can
771 * pass a stride back to the caller.
772 */
773void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
774 u64 off, u64 *plen,
645a1025 775 u64 *ono,
f24e9980
SW		 776 u64 *oxoff, u64 *oxlen)
777{
778 u32 osize = le32_to_cpu(layout->fl_object_size);
779 u32 su = le32_to_cpu(layout->fl_stripe_unit);
780 u32 sc = le32_to_cpu(layout->fl_stripe_count);
781 u32 bl, stripeno, stripepos, objsetno;
782 u32 su_per_object;
ff1d1f71 783 u64 t, su_offset;
f24e9980
SW		 784
785 dout("mapping %llu~%llu osize %u fl_su %u\n", off, *plen,
786 osize, su);
35e054a6 787 su_per_object = osize / su;
f24e9980
SW		 788 dout("osize %u / su %u = su_per_object %u\n", osize, su,
789 su_per_object);
790
791 BUG_ON((su & ~PAGE_MASK) != 0);
792 /* bl = *off / su; */
793 t = off;
794 do_div(t, su);
795 bl = t;
796 dout("off %llu / su %u = bl %u\n", off, su, bl);
797
798 stripeno = bl / sc;
799 stripepos = bl % sc;
800 objsetno = stripeno / su_per_object;
801
645a1025
SW		 802 *ono = objsetno * sc + stripepos;
803 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono);
804
805 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
f24e9980 806 t = off;
ff1d1f71
NW		 807 su_offset = do_div(t, su);
808 *oxoff = su_offset + (stripeno % su_per_object) * su;
809
810 /*
811 * Calculate the length of the extent being written to the selected
812 * object. This is the minimum of the full length requested (plen) or
813 * the remainder of the current stripe being written to.
814 */
815 *oxlen = min_t(u64, *plen, su - su_offset);
f24e9980
SW		 816 *plen = *oxlen;
817
818 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
819}
820
821/*
822 * calculate an object layout (i.e. pgid) from an oid,
823 * file_layout, and osdmap
824 */
825int ceph_calc_object_layout(struct ceph_object_layout *ol,
826 const char *oid,
827 struct ceph_file_layout *fl,
828 struct ceph_osdmap *osdmap)
829{
830 unsigned num, num_mask;
51042122 831 struct ceph_pg pgid;
f24e9980
SW		 832 s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred);
833 int poolid = le32_to_cpu(fl->fl_pg_pool);
834 struct ceph_pg_pool_info *pool;
51042122 835 unsigned ps;
f24e9980 836
30dc6381 837 BUG_ON(!osdmap);
f24e9980
SW		 838 if (poolid >= osdmap->num_pools)
839 return -EIO;
f24e9980 840
51042122 841 pool = &osdmap->pg_pool[poolid];
1654dd0c 842 ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
f24e9980 843 if (preferred >= 0) {
51042122 844 ps += preferred;
f24e9980
SW		 845 num = le32_to_cpu(pool->v.lpg_num);
846 num_mask = pool->lpg_num_mask;
847 } else {
848 num = le32_to_cpu(pool->v.pg_num);
849 num_mask = pool->pg_num_mask;
850 }
851
51042122
SW		 852 pgid.ps = cpu_to_le16(ps);
853 pgid.preferred = cpu_to_le16(preferred);
854 pgid.pool = fl->fl_pg_pool;
f24e9980 855 if (preferred >= 0)
51042122
SW		 856 dout("calc_object_layout '%s' pgid %d.%xp%d\n", oid, poolid, ps,
857 (int)preferred);
f24e9980 858 else
51042122 859 dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
f24e9980 860
51042122 861 ol->ol_pgid = pgid;
f24e9980 862 ol->ol_stripe_unit = fl->fl_object_stripe_unit;
f24e9980
SW		 863 return 0;
864}
865
866/*
867 * Calculate raw osd vector for the given pgid. Return pointer to osd
868 * array, or NULL on failure.
869 */
51042122 870static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
f24e9980
SW		 871 int *osds, int *num)
872{
873 struct rb_node *n = osdmap->pg_temp.rb_node;
874 struct ceph_pg_mapping *pg;
875 struct ceph_pg_pool_info *pool;
876 int ruleno;
51042122
SW		 877 unsigned poolid, ps, pps;
878 int preferred;
879 int c;
f24e9980
SW		 880
881 /* pg_temp? */
882 while (n) {
883 pg = rb_entry(n, struct ceph_pg_mapping, node);
51042122
SW		 884 c = pgid_cmp(pgid, pg->pgid);
885 if (c < 0)
f24e9980 886 n = n->rb_left;
51042122 887 else if (c > 0)
f24e9980
SW		 888 n = n->rb_right;
889 else {
890 *num = pg->len;
891 return pg->osds;
892 }
893 }
894
895 /* crush */
51042122
SW		 896 poolid = le32_to_cpu(pgid.pool);
897 ps = le16_to_cpu(pgid.ps);
898 preferred = (s16)le16_to_cpu(pgid.preferred);
899
767ea5c3
SW		 900 /* don't forcefeed bad device ids to crush */
901 if (preferred >= osdmap->max_osd ||
902 preferred >= osdmap->crush->max_devices)
903 preferred = -1;
904
51042122 905 if (poolid >= osdmap->num_pools)
f24e9980 906 return NULL;
51042122 907 pool = &osdmap->pg_pool[poolid];
f24e9980
SW		 908 ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
909 pool->v.type, pool->v.size);
910 if (ruleno < 0) {
911 pr_err("no crush rule pool %d type %d size %d\n",
51042122 912 poolid, pool->v.type, pool->v.size);
f24e9980
SW		 913 return NULL;
914 }
915
51042122
SW		 916 if (preferred >= 0)
917 pps = ceph_stable_mod(ps,
f24e9980
SW		 918 le32_to_cpu(pool->v.lpgp_num),
919 pool->lpgp_num_mask);
920 else
51042122 921 pps = ceph_stable_mod(ps,
f24e9980
SW		 922 le32_to_cpu(pool->v.pgp_num),
923 pool->pgp_num_mask);
51042122 924 pps += poolid;
f24e9980
SW		 925 *num = crush_do_rule(osdmap->crush, ruleno, pps, osds,
926 min_t(int, pool->v.size, *num),
51042122 927 preferred, osdmap->osd_weight);
f24e9980
SW		 928 return osds;
929}
930
931/*
932 * Return primary osd for given pgid, or -1 if none.
933 */
51042122 934int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
f24e9980
SW		 935{
936 int rawosds[10], *osds;
937 int i, num = ARRAY_SIZE(rawosds);
938
939 osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
940 if (!osds)
941 return -1;
942
943 /* primary is first up osd */
944 for (i = 0; i < num; i++)
945 if (ceph_osd_is_up(osdmap, osds[i])) {
946 return osds[i];
947 break;
948 }
949 return -1;
950}
Linux 6.x block layer and io_uring tree(s)