]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - fs/ceph/osdmap.c
ceph: fix mds sync() race with completing requests
[mirror_ubuntu-zesty-kernel.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);
9794b146
SW
324 while (!RB_EMPTY_ROOT(&map->pg_temp)) {
325 struct ceph_pg_mapping *pg =
326 rb_entry(rb_first(&map->pg_temp),
327 struct ceph_pg_mapping, node);
328 rb_erase(&pg->node, &map->pg_temp);
329 kfree(pg);
330 }
4fc51be8
SW
331 while (!RB_EMPTY_ROOT(&map->pg_pools)) {
332 struct ceph_pg_pool_info *pi =
333 rb_entry(rb_first(&map->pg_pools),
334 struct ceph_pg_pool_info, node);
335 rb_erase(&pi->node, &map->pg_pools);
336 kfree(pi);
337 }
f24e9980
SW
338 kfree(map->osd_state);
339 kfree(map->osd_weight);
f24e9980
SW
340 kfree(map->osd_addr);
341 kfree(map);
342}
343
344/*
345 * adjust max osd value. reallocate arrays.
346 */
347static int osdmap_set_max_osd(struct ceph_osdmap *map, int max)
348{
349 u8 *state;
350 struct ceph_entity_addr *addr;
351 u32 *weight;
352
353 state = kcalloc(max, sizeof(*state), GFP_NOFS);
354 addr = kcalloc(max, sizeof(*addr), GFP_NOFS);
355 weight = kcalloc(max, sizeof(*weight), GFP_NOFS);
356 if (state == NULL || addr == NULL || weight == NULL) {
357 kfree(state);
358 kfree(addr);
359 kfree(weight);
360 return -ENOMEM;
361 }
362
363 /* copy old? */
364 if (map->osd_state) {
365 memcpy(state, map->osd_state, map->max_osd*sizeof(*state));
366 memcpy(addr, map->osd_addr, map->max_osd*sizeof(*addr));
367 memcpy(weight, map->osd_weight, map->max_osd*sizeof(*weight));
368 kfree(map->osd_state);
369 kfree(map->osd_addr);
370 kfree(map->osd_weight);
371 }
372
373 map->osd_state = state;
374 map->osd_weight = weight;
375 map->osd_addr = addr;
376 map->max_osd = max;
377 return 0;
378}
379
380/*
9794b146
SW
381 * rbtree of pg_mapping for handling pg_temp (explicit mapping of pgid
382 * to a set of osds)
f24e9980 383 */
51042122
SW
384static int pgid_cmp(struct ceph_pg l, struct ceph_pg r)
385{
386 u64 a = *(u64 *)&l;
387 u64 b = *(u64 *)&r;
388
389 if (a < b)
390 return -1;
391 if (a > b)
392 return 1;
393 return 0;
394}
395
991abb6e
SW
396static int __insert_pg_mapping(struct ceph_pg_mapping *new,
397 struct rb_root *root)
f24e9980
SW
398{
399 struct rb_node **p = &root->rb_node;
400 struct rb_node *parent = NULL;
401 struct ceph_pg_mapping *pg = NULL;
51042122 402 int c;
f24e9980
SW
403
404 while (*p) {
405 parent = *p;
406 pg = rb_entry(parent, struct ceph_pg_mapping, node);
51042122
SW
407 c = pgid_cmp(new->pgid, pg->pgid);
408 if (c < 0)
f24e9980 409 p = &(*p)->rb_left;
51042122 410 else if (c > 0)
f24e9980
SW
411 p = &(*p)->rb_right;
412 else
991abb6e 413 return -EEXIST;
f24e9980
SW
414 }
415
416 rb_link_node(&new->node, parent, p);
417 rb_insert_color(&new->node, root);
991abb6e 418 return 0;
f24e9980
SW
419}
420
9794b146
SW
421static struct ceph_pg_mapping *__lookup_pg_mapping(struct rb_root *root,
422 struct ceph_pg pgid)
423{
424 struct rb_node *n = root->rb_node;
425 struct ceph_pg_mapping *pg;
426 int c;
427
428 while (n) {
429 pg = rb_entry(n, struct ceph_pg_mapping, node);
430 c = pgid_cmp(pgid, pg->pgid);
431 if (c < 0)
432 n = n->rb_left;
433 else if (c > 0)
434 n = n->rb_right;
435 else
436 return pg;
437 }
438 return NULL;
439}
440
4fc51be8
SW
441/*
442 * rbtree of pg pool info
443 */
444static int __insert_pg_pool(struct rb_root *root, struct ceph_pg_pool_info *new)
445{
446 struct rb_node **p = &root->rb_node;
447 struct rb_node *parent = NULL;
448 struct ceph_pg_pool_info *pi = NULL;
449
450 while (*p) {
451 parent = *p;
452 pi = rb_entry(parent, struct ceph_pg_pool_info, node);
453 if (new->id < pi->id)
454 p = &(*p)->rb_left;
455 else if (new->id > pi->id)
456 p = &(*p)->rb_right;
457 else
458 return -EEXIST;
459 }
460
461 rb_link_node(&new->node, parent, p);
462 rb_insert_color(&new->node, root);
463 return 0;
464}
465
466static struct ceph_pg_pool_info *__lookup_pg_pool(struct rb_root *root, int id)
467{
468 struct ceph_pg_pool_info *pi;
469 struct rb_node *n = root->rb_node;
470
471 while (n) {
472 pi = rb_entry(n, struct ceph_pg_pool_info, node);
473 if (id < pi->id)
474 n = n->rb_left;
475 else if (id > pi->id)
476 n = n->rb_right;
477 else
478 return pi;
479 }
480 return NULL;
481}
482
f24e9980
SW
483/*
484 * decode a full map.
485 */
486struct ceph_osdmap *osdmap_decode(void **p, void *end)
487{
488 struct ceph_osdmap *map;
489 u16 version;
490 u32 len, max, i;
361be860 491 u8 ev;
f24e9980
SW
492 int err = -EINVAL;
493 void *start = *p;
4fc51be8 494 struct ceph_pg_pool_info *pi;
f24e9980
SW
495
496 dout("osdmap_decode %p to %p len %d\n", *p, end, (int)(end - *p));
497
498 map = kzalloc(sizeof(*map), GFP_NOFS);
499 if (map == NULL)
500 return ERR_PTR(-ENOMEM);
501 map->pg_temp = RB_ROOT;
502
503 ceph_decode_16_safe(p, end, version, bad);
02f90c61
SW
504 if (version > CEPH_OSDMAP_VERSION) {
505 pr_warning("got unknown v %d > %d of osdmap\n", version,
506 CEPH_OSDMAP_VERSION);
507 goto bad;
508 }
f24e9980
SW
509
510 ceph_decode_need(p, end, 2*sizeof(u64)+6*sizeof(u32), bad);
511 ceph_decode_copy(p, &map->fsid, sizeof(map->fsid));
c89136ea 512 map->epoch = ceph_decode_32(p);
f24e9980
SW
513 ceph_decode_copy(p, &map->created, sizeof(map->created));
514 ceph_decode_copy(p, &map->modified, sizeof(map->modified));
515
f24e9980
SW
516 ceph_decode_32_safe(p, end, max, bad);
517 while (max--) {
4fc51be8
SW
518 ceph_decode_need(p, end, 4 + 1 + sizeof(pi->v), bad);
519 pi = kmalloc(sizeof(*pi), GFP_NOFS);
520 if (!pi)
f24e9980 521 goto bad;
4fc51be8 522 pi->id = ceph_decode_32(p);
361be860 523 ev = ceph_decode_8(p); /* encoding version */
02f90c61
SW
524 if (ev > CEPH_PG_POOL_VERSION) {
525 pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
526 ev, CEPH_PG_POOL_VERSION);
527 goto bad;
528 }
4fc51be8
SW
529 ceph_decode_copy(p, &pi->v, sizeof(pi->v));
530 __insert_pg_pool(&map->pg_pools, pi);
531 calc_pg_masks(pi);
e53a8fd7
SW
532 *p += le32_to_cpu(pi->v.num_snaps) * sizeof(u64);
533 *p += le32_to_cpu(pi->v.num_removed_snap_intervals)
f24e9980
SW
534 * sizeof(u64) * 2;
535 }
4fc51be8 536 ceph_decode_32_safe(p, end, map->pool_max, bad);
f24e9980
SW
537
538 ceph_decode_32_safe(p, end, map->flags, bad);
539
c89136ea 540 max = ceph_decode_32(p);
f24e9980
SW
541
542 /* (re)alloc osd arrays */
543 err = osdmap_set_max_osd(map, max);
544 if (err < 0)
545 goto bad;
546 dout("osdmap_decode max_osd = %d\n", map->max_osd);
547
548 /* osds */
549 err = -EINVAL;
550 ceph_decode_need(p, end, 3*sizeof(u32) +
551 map->max_osd*(1 + sizeof(*map->osd_weight) +
552 sizeof(*map->osd_addr)), bad);
553 *p += 4; /* skip length field (should match max) */
554 ceph_decode_copy(p, map->osd_state, map->max_osd);
555
556 *p += 4; /* skip length field (should match max) */
557 for (i = 0; i < map->max_osd; i++)
c89136ea 558 map->osd_weight[i] = ceph_decode_32(p);
f24e9980
SW
559
560 *p += 4; /* skip length field (should match max) */
561 ceph_decode_copy(p, map->osd_addr, map->max_osd*sizeof(*map->osd_addr));
63f2d211
SW
562 for (i = 0; i < map->max_osd; i++)
563 ceph_decode_addr(&map->osd_addr[i]);
f24e9980
SW
564
565 /* pg_temp */
566 ceph_decode_32_safe(p, end, len, bad);
567 for (i = 0; i < len; i++) {
568 int n, j;
51042122 569 struct ceph_pg pgid;
f24e9980
SW
570 struct ceph_pg_mapping *pg;
571
572 ceph_decode_need(p, end, sizeof(u32) + sizeof(u64), bad);
51042122 573 ceph_decode_copy(p, &pgid, sizeof(pgid));
c89136ea 574 n = ceph_decode_32(p);
f24e9980 575 ceph_decode_need(p, end, n * sizeof(u32), bad);
30dc6381 576 err = -ENOMEM;
f24e9980 577 pg = kmalloc(sizeof(*pg) + n*sizeof(u32), GFP_NOFS);
30dc6381 578 if (!pg)
f24e9980 579 goto bad;
f24e9980
SW
580 pg->pgid = pgid;
581 pg->len = n;
582 for (j = 0; j < n; j++)
c89136ea 583 pg->osds[j] = ceph_decode_32(p);
f24e9980 584
991abb6e
SW
585 err = __insert_pg_mapping(pg, &map->pg_temp);
586 if (err)
587 goto bad;
51042122 588 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid, len);
f24e9980
SW
589 }
590
591 /* crush */
592 ceph_decode_32_safe(p, end, len, bad);
593 dout("osdmap_decode crush len %d from off 0x%x\n", len,
594 (int)(*p - start));
595 ceph_decode_need(p, end, len, bad);
596 map->crush = crush_decode(*p, end);
597 *p += len;
598 if (IS_ERR(map->crush)) {
599 err = PTR_ERR(map->crush);
600 map->crush = NULL;
601 goto bad;
602 }
603
604 /* ignore the rest of the map */
605 *p = end;
606
607 dout("osdmap_decode done %p %p\n", *p, end);
608 return map;
609
610bad:
611 dout("osdmap_decode fail\n");
612 ceph_osdmap_destroy(map);
613 return ERR_PTR(err);
614}
615
616/*
617 * decode and apply an incremental map update.
618 */
619struct ceph_osdmap *osdmap_apply_incremental(void **p, void *end,
620 struct ceph_osdmap *map,
621 struct ceph_messenger *msgr)
622{
f24e9980
SW
623 struct crush_map *newcrush = NULL;
624 struct ceph_fsid fsid;
625 u32 epoch = 0;
626 struct ceph_timespec modified;
627 u32 len, pool;
4fc51be8 628 __s32 new_pool_max, new_flags, max;
f24e9980
SW
629 void *start = *p;
630 int err = -EINVAL;
631 u16 version;
632 struct rb_node *rbp;
633
634 ceph_decode_16_safe(p, end, version, bad);
02f90c61
SW
635 if (version > CEPH_OSDMAP_INC_VERSION) {
636 pr_warning("got unknown v %d > %d of inc osdmap\n", version,
637 CEPH_OSDMAP_INC_VERSION);
638 goto bad;
639 }
f24e9980
SW
640
641 ceph_decode_need(p, end, sizeof(fsid)+sizeof(modified)+2*sizeof(u32),
642 bad);
643 ceph_decode_copy(p, &fsid, sizeof(fsid));
c89136ea 644 epoch = ceph_decode_32(p);
f24e9980
SW
645 BUG_ON(epoch != map->epoch+1);
646 ceph_decode_copy(p, &modified, sizeof(modified));
4fc51be8 647 new_pool_max = ceph_decode_32(p);
c89136ea 648 new_flags = ceph_decode_32(p);
f24e9980
SW
649
650 /* full map? */
651 ceph_decode_32_safe(p, end, len, bad);
652 if (len > 0) {
653 dout("apply_incremental full map len %d, %p to %p\n",
654 len, *p, end);
30dc6381 655 return osdmap_decode(p, min(*p+len, end));
f24e9980
SW
656 }
657
658 /* new crush? */
659 ceph_decode_32_safe(p, end, len, bad);
660 if (len > 0) {
661 dout("apply_incremental new crush map len %d, %p to %p\n",
662 len, *p, end);
663 newcrush = crush_decode(*p, min(*p+len, end));
664 if (IS_ERR(newcrush))
665 return ERR_PTR(PTR_ERR(newcrush));
666 }
667
668 /* new flags? */
669 if (new_flags >= 0)
670 map->flags = new_flags;
4fc51be8
SW
671 if (new_pool_max >= 0)
672 map->pool_max = new_pool_max;
f24e9980
SW
673
674 ceph_decode_need(p, end, 5*sizeof(u32), bad);
675
676 /* new max? */
c89136ea 677 max = ceph_decode_32(p);
f24e9980
SW
678 if (max >= 0) {
679 err = osdmap_set_max_osd(map, max);
680 if (err < 0)
681 goto bad;
682 }
683
684 map->epoch++;
685 map->modified = map->modified;
686 if (newcrush) {
687 if (map->crush)
688 crush_destroy(map->crush);
689 map->crush = newcrush;
690 newcrush = NULL;
691 }
692
693 /* new_pool */
694 ceph_decode_32_safe(p, end, len, bad);
695 while (len--) {
361be860 696 __u8 ev;
4fc51be8 697 struct ceph_pg_pool_info *pi;
361be860 698
f24e9980 699 ceph_decode_32_safe(p, end, pool, bad);
4fc51be8 700 ceph_decode_need(p, end, 1 + sizeof(pi->v), bad);
361be860 701 ev = ceph_decode_8(p); /* encoding version */
02f90c61
SW
702 if (ev > CEPH_PG_POOL_VERSION) {
703 pr_warning("got unknown v %d > %d of ceph_pg_pool\n",
704 ev, CEPH_PG_POOL_VERSION);
705 goto bad;
706 }
4fc51be8
SW
707 pi = __lookup_pg_pool(&map->pg_pools, pool);
708 if (!pi) {
709 pi = kmalloc(sizeof(*pi), GFP_NOFS);
710 if (!pi) {
711 err = -ENOMEM;
712 goto bad;
713 }
714 pi->id = pool;
715 __insert_pg_pool(&map->pg_pools, pi);
716 }
717 ceph_decode_copy(p, &pi->v, sizeof(pi->v));
718 calc_pg_masks(pi);
f24e9980
SW
719 }
720
4fc51be8 721 /* old_pool */
f24e9980 722 ceph_decode_32_safe(p, end, len, bad);
4fc51be8
SW
723 while (len--) {
724 struct ceph_pg_pool_info *pi;
725
726 ceph_decode_32_safe(p, end, pool, bad);
727 pi = __lookup_pg_pool(&map->pg_pools, pool);
728 if (pi) {
729 rb_erase(&pi->node, &map->pg_pools);
730 kfree(pi);
731 }
732 }
f24e9980
SW
733
734 /* new_up */
735 err = -EINVAL;
736 ceph_decode_32_safe(p, end, len, bad);
737 while (len--) {
738 u32 osd;
739 struct ceph_entity_addr addr;
740 ceph_decode_32_safe(p, end, osd, bad);
741 ceph_decode_copy_safe(p, end, &addr, sizeof(addr), bad);
63f2d211 742 ceph_decode_addr(&addr);
f24e9980
SW
743 pr_info("osd%d up\n", osd);
744 BUG_ON(osd >= map->max_osd);
745 map->osd_state[osd] |= CEPH_OSD_UP;
746 map->osd_addr[osd] = addr;
747 }
748
749 /* new_down */
750 ceph_decode_32_safe(p, end, len, bad);
751 while (len--) {
752 u32 osd;
753 ceph_decode_32_safe(p, end, osd, bad);
754 (*p)++; /* clean flag */
1bdb70e5 755 pr_info("osd%d down\n", osd);
f24e9980
SW
756 if (osd < map->max_osd)
757 map->osd_state[osd] &= ~CEPH_OSD_UP;
758 }
759
760 /* new_weight */
761 ceph_decode_32_safe(p, end, len, bad);
762 while (len--) {
763 u32 osd, off;
764 ceph_decode_need(p, end, sizeof(u32)*2, bad);
c89136ea
SW
765 osd = ceph_decode_32(p);
766 off = ceph_decode_32(p);
f24e9980
SW
767 pr_info("osd%d weight 0x%x %s\n", osd, off,
768 off == CEPH_OSD_IN ? "(in)" :
769 (off == CEPH_OSD_OUT ? "(out)" : ""));
770 if (osd < map->max_osd)
771 map->osd_weight[osd] = off;
772 }
773
774 /* new_pg_temp */
775 rbp = rb_first(&map->pg_temp);
776 ceph_decode_32_safe(p, end, len, bad);
777 while (len--) {
778 struct ceph_pg_mapping *pg;
779 int j;
51042122 780 struct ceph_pg pgid;
f24e9980
SW
781 u32 pglen;
782 ceph_decode_need(p, end, sizeof(u64) + sizeof(u32), bad);
51042122 783 ceph_decode_copy(p, &pgid, sizeof(pgid));
c89136ea 784 pglen = ceph_decode_32(p);
f24e9980
SW
785
786 /* remove any? */
51042122
SW
787 while (rbp && pgid_cmp(rb_entry(rbp, struct ceph_pg_mapping,
788 node)->pgid, pgid) <= 0) {
f24e9980
SW
789 struct rb_node *cur = rbp;
790 rbp = rb_next(rbp);
791 dout(" removed pg_temp %llx\n",
51042122
SW
792 *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
793 node)->pgid);
f24e9980
SW
794 rb_erase(cur, &map->pg_temp);
795 }
796
797 if (pglen) {
798 /* insert */
799 ceph_decode_need(p, end, pglen*sizeof(u32), bad);
800 pg = kmalloc(sizeof(*pg) + sizeof(u32)*pglen, GFP_NOFS);
801 if (!pg) {
802 err = -ENOMEM;
803 goto bad;
804 }
805 pg->pgid = pgid;
806 pg->len = pglen;
7067f797 807 for (j = 0; j < pglen; j++)
c89136ea 808 pg->osds[j] = ceph_decode_32(p);
991abb6e
SW
809 err = __insert_pg_mapping(pg, &map->pg_temp);
810 if (err)
811 goto bad;
51042122
SW
812 dout(" added pg_temp %llx len %d\n", *(u64 *)&pgid,
813 pglen);
f24e9980
SW
814 }
815 }
816 while (rbp) {
817 struct rb_node *cur = rbp;
818 rbp = rb_next(rbp);
819 dout(" removed pg_temp %llx\n",
51042122
SW
820 *(u64 *)&rb_entry(cur, struct ceph_pg_mapping,
821 node)->pgid);
f24e9980
SW
822 rb_erase(cur, &map->pg_temp);
823 }
824
825 /* ignore the rest */
826 *p = end;
827 return map;
828
829bad:
830 pr_err("corrupt inc osdmap epoch %d off %d (%p of %p-%p)\n",
831 epoch, (int)(*p - start), *p, start, end);
9ec7cab1
SW
832 print_hex_dump(KERN_DEBUG, "osdmap: ",
833 DUMP_PREFIX_OFFSET, 16, 1,
834 start, end - start, true);
f24e9980
SW
835 if (newcrush)
836 crush_destroy(newcrush);
837 return ERR_PTR(err);
838}
839
840
841
842
843/*
844 * calculate file layout from given offset, length.
845 * fill in correct oid, logical length, and object extent
846 * offset, length.
847 *
848 * for now, we write only a single su, until we can
849 * pass a stride back to the caller.
850 */
851void ceph_calc_file_object_mapping(struct ceph_file_layout *layout,
852 u64 off, u64 *plen,
645a1025 853 u64 *ono,
f24e9980
SW
854 u64 *oxoff, u64 *oxlen)
855{
856 u32 osize = le32_to_cpu(layout->fl_object_size);
857 u32 su = le32_to_cpu(layout->fl_stripe_unit);
858 u32 sc = le32_to_cpu(layout->fl_stripe_count);
859 u32 bl, stripeno, stripepos, objsetno;
860 u32 su_per_object;
ff1d1f71 861 u64 t, su_offset;
f24e9980
SW
862
863 dout("mapping %llu~%llu osize %u fl_su %u\n", off, *plen,
864 osize, su);
35e054a6 865 su_per_object = osize / su;
f24e9980
SW
866 dout("osize %u / su %u = su_per_object %u\n", osize, su,
867 su_per_object);
868
869 BUG_ON((su & ~PAGE_MASK) != 0);
870 /* bl = *off / su; */
871 t = off;
872 do_div(t, su);
873 bl = t;
874 dout("off %llu / su %u = bl %u\n", off, su, bl);
875
876 stripeno = bl / sc;
877 stripepos = bl % sc;
878 objsetno = stripeno / su_per_object;
879
645a1025
SW
880 *ono = objsetno * sc + stripepos;
881 dout("objset %u * sc %u = ono %u\n", objsetno, sc, (unsigned)*ono);
882
883 /* *oxoff = *off % layout->fl_stripe_unit; # offset in su */
f24e9980 884 t = off;
ff1d1f71
NW
885 su_offset = do_div(t, su);
886 *oxoff = su_offset + (stripeno % su_per_object) * su;
887
888 /*
889 * Calculate the length of the extent being written to the selected
890 * object. This is the minimum of the full length requested (plen) or
891 * the remainder of the current stripe being written to.
892 */
893 *oxlen = min_t(u64, *plen, su - su_offset);
f24e9980
SW
894 *plen = *oxlen;
895
896 dout(" obj extent %llu~%llu\n", *oxoff, *oxlen);
897}
898
899/*
900 * calculate an object layout (i.e. pgid) from an oid,
901 * file_layout, and osdmap
902 */
903int ceph_calc_object_layout(struct ceph_object_layout *ol,
904 const char *oid,
905 struct ceph_file_layout *fl,
906 struct ceph_osdmap *osdmap)
907{
908 unsigned num, num_mask;
51042122 909 struct ceph_pg pgid;
f24e9980
SW
910 s32 preferred = (s32)le32_to_cpu(fl->fl_pg_preferred);
911 int poolid = le32_to_cpu(fl->fl_pg_pool);
912 struct ceph_pg_pool_info *pool;
51042122 913 unsigned ps;
f24e9980 914
30dc6381 915 BUG_ON(!osdmap);
f24e9980 916
4fc51be8
SW
917 pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
918 if (!pool)
919 return -EIO;
1654dd0c 920 ps = ceph_str_hash(pool->v.object_hash, oid, strlen(oid));
f24e9980 921 if (preferred >= 0) {
51042122 922 ps += preferred;
f24e9980
SW
923 num = le32_to_cpu(pool->v.lpg_num);
924 num_mask = pool->lpg_num_mask;
925 } else {
926 num = le32_to_cpu(pool->v.pg_num);
927 num_mask = pool->pg_num_mask;
928 }
929
51042122
SW
930 pgid.ps = cpu_to_le16(ps);
931 pgid.preferred = cpu_to_le16(preferred);
932 pgid.pool = fl->fl_pg_pool;
f24e9980 933 if (preferred >= 0)
51042122
SW
934 dout("calc_object_layout '%s' pgid %d.%xp%d\n", oid, poolid, ps,
935 (int)preferred);
f24e9980 936 else
51042122 937 dout("calc_object_layout '%s' pgid %d.%x\n", oid, poolid, ps);
f24e9980 938
51042122 939 ol->ol_pgid = pgid;
f24e9980 940 ol->ol_stripe_unit = fl->fl_object_stripe_unit;
f24e9980
SW
941 return 0;
942}
943
944/*
945 * Calculate raw osd vector for the given pgid. Return pointer to osd
946 * array, or NULL on failure.
947 */
51042122 948static int *calc_pg_raw(struct ceph_osdmap *osdmap, struct ceph_pg pgid,
f24e9980
SW
949 int *osds, int *num)
950{
f24e9980
SW
951 struct ceph_pg_mapping *pg;
952 struct ceph_pg_pool_info *pool;
953 int ruleno;
51042122
SW
954 unsigned poolid, ps, pps;
955 int preferred;
f24e9980
SW
956
957 /* pg_temp? */
9794b146
SW
958 pg = __lookup_pg_mapping(&osdmap->pg_temp, pgid);
959 if (pg) {
960 *num = pg->len;
961 return pg->osds;
f24e9980
SW
962 }
963
964 /* crush */
51042122
SW
965 poolid = le32_to_cpu(pgid.pool);
966 ps = le16_to_cpu(pgid.ps);
967 preferred = (s16)le16_to_cpu(pgid.preferred);
968
767ea5c3
SW
969 /* don't forcefeed bad device ids to crush */
970 if (preferred >= osdmap->max_osd ||
971 preferred >= osdmap->crush->max_devices)
972 preferred = -1;
973
4fc51be8
SW
974 pool = __lookup_pg_pool(&osdmap->pg_pools, poolid);
975 if (!pool)
f24e9980 976 return NULL;
f24e9980
SW
977 ruleno = crush_find_rule(osdmap->crush, pool->v.crush_ruleset,
978 pool->v.type, pool->v.size);
979 if (ruleno < 0) {
980 pr_err("no crush rule pool %d type %d size %d\n",
51042122 981 poolid, pool->v.type, pool->v.size);
f24e9980
SW
982 return NULL;
983 }
984
51042122
SW
985 if (preferred >= 0)
986 pps = ceph_stable_mod(ps,
f24e9980
SW
987 le32_to_cpu(pool->v.lpgp_num),
988 pool->lpgp_num_mask);
989 else
51042122 990 pps = ceph_stable_mod(ps,
f24e9980
SW
991 le32_to_cpu(pool->v.pgp_num),
992 pool->pgp_num_mask);
51042122 993 pps += poolid;
f24e9980
SW
994 *num = crush_do_rule(osdmap->crush, ruleno, pps, osds,
995 min_t(int, pool->v.size, *num),
51042122 996 preferred, osdmap->osd_weight);
f24e9980
SW
997 return osds;
998}
999
1000/*
1001 * Return primary osd for given pgid, or -1 if none.
1002 */
51042122 1003int ceph_calc_pg_primary(struct ceph_osdmap *osdmap, struct ceph_pg pgid)
f24e9980
SW
1004{
1005 int rawosds[10], *osds;
1006 int i, num = ARRAY_SIZE(rawosds);
1007
1008 osds = calc_pg_raw(osdmap, pgid, rawosds, &num);
1009 if (!osds)
1010 return -1;
1011
1012 /* primary is first up osd */
1013 for (i = 0; i < num; i++)
1014 if (ceph_osd_is_up(osdmap, osds[i])) {
1015 return osds[i];
1016 break;
1017 }
1018 return -1;
1019}