?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
libceph: nuke unused fields and functions
[linux-2.6/btrfs-unstable.git] / net / ceph / mon_client.c
blob3dfafdad92aab0c028484a8c2ce1cb53db488cbf
1 #include <linux/ceph/ceph_debug.h>
2
3 #include <linux/module.h>
4 #include <linux/types.h>
5 #include <linux/slab.h>
6 #include <linux/random.h>
7 #include <linux/sched.h>
8
9 #include <linux/ceph/mon_client.h>
10 #include <linux/ceph/libceph.h>
11 #include <linux/ceph/debugfs.h>
12 #include <linux/ceph/decode.h>
13 #include <linux/ceph/auth.h>
14
15 /*
16 * Interact with Ceph monitor cluster. Handle requests for new map
17 * versions, and periodically resend as needed. Also implement
18 * statfs() and umount().
19 *
20 * A small cluster of Ceph "monitors" are responsible for managing critical
21 * cluster configuration and state information. An odd number (e.g., 3, 5)
22 * of cmon daemons use a modified version of the Paxos part-time parliament
23 * algorithm to manage the MDS map (mds cluster membership), OSD map, and
24 * list of clients who have mounted the file system.
25 *
26 * We maintain an open, active session with a monitor at all times in order to
27 * receive timely MDSMap updates. We periodically send a keepalive byte on the
28 * TCP socket to ensure we detect a failure. If the connection does break, we
29 * randomly hunt for a new monitor. Once the connection is reestablished, we
30 * resend any outstanding requests.
31 */
32
33 static const struct ceph_connection_operations mon_con_ops;
34
35 static int __validate_auth(struct ceph_mon_client *monc);
36
37 /*
38 * Decode a monmap blob (e.g., during mount).
39 */
40 struct ceph_monmap *ceph_monmap_decode(void *p, void *end)
41 {
42 struct ceph_monmap *m = NULL;
43 int i, err = -EINVAL;
44 struct ceph_fsid fsid;
45 u32 epoch, num_mon;
46 u16 version;
47 u32 len;
48
49 ceph_decode_32_safe(&p, end, len, bad);
50 ceph_decode_need(&p, end, len, bad);
51
52 dout("monmap_decode %p %p len %d\n", p, end, (int)(end-p));
53
54 ceph_decode_16_safe(&p, end, version, bad);
55
56 ceph_decode_need(&p, end, sizeof(fsid) + 2*sizeof(u32), bad);
57 ceph_decode_copy(&p, &fsid, sizeof(fsid));
58 epoch = ceph_decode_32(&p);
59
60 num_mon = ceph_decode_32(&p);
61 ceph_decode_need(&p, end, num_mon*sizeof(m->mon_inst[0]), bad);
62
63 if (num_mon >= CEPH_MAX_MON)
64 goto bad;
65 m = kmalloc(sizeof(*m) + sizeof(m->mon_inst[0])*num_mon, GFP_NOFS);
66 if (m == NULL)
67 return ERR_PTR(-ENOMEM);
68 m->fsid = fsid;
69 m->epoch = epoch;
70 m->num_mon = num_mon;
71 ceph_decode_copy(&p, m->mon_inst, num_mon*sizeof(m->mon_inst[0]));
72 for (i = 0; i < num_mon; i++)
73 ceph_decode_addr(&m->mon_inst[i].addr);
74
75 dout("monmap_decode epoch %d, num_mon %d\n", m->epoch,
76 m->num_mon);
77 for (i = 0; i < m->num_mon; i++)
78 dout("monmap_decode mon%d is %s\n", i,
79 ceph_pr_addr(&m->mon_inst[i].addr.in_addr));
80 return m;
81
82 bad:
83 dout("monmap_decode failed with %d\n", err);
84 kfree(m);
85 return ERR_PTR(err);
86 }
87
88 /*
89 * return true if *addr is included in the monmap.
90 */
91 int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr)
92 {
93 int i;
94
95 for (i = 0; i < m->num_mon; i++)
96 if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0)
97 return 1;
98 return 0;
99 }
100
101 /*
102 * Send an auth request.
103 */
104 static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len)
105 {
106 monc->pending_auth = 1;
107 monc->m_auth->front.iov_len = len;
108 monc->m_auth->hdr.front_len = cpu_to_le32(len);
109 ceph_msg_revoke(monc->m_auth);
110 ceph_msg_get(monc->m_auth); /* keep our ref */
111 ceph_con_send(&monc->con, monc->m_auth);
112 }
113
114 /*
115 * Close monitor session, if any.
116 */
117 static void __close_session(struct ceph_mon_client *monc)
118 {
119 dout("__close_session closing mon%d\n", monc->cur_mon);
120 ceph_msg_revoke(monc->m_auth);
121 ceph_msg_revoke_incoming(monc->m_auth_reply);
122 ceph_msg_revoke(monc->m_subscribe);
123 ceph_msg_revoke_incoming(monc->m_subscribe_ack);
124 ceph_con_close(&monc->con);
125
126 monc->pending_auth = 0;
127 ceph_auth_reset(monc->auth);
128 }
129
130 /*
131 * Pick a new monitor at random and set cur_mon. If we are repicking
132 * (i.e. cur_mon is already set), be sure to pick a different one.
133 */
134 static void pick_new_mon(struct ceph_mon_client *monc)
135 {
136 int old_mon = monc->cur_mon;
137
138 BUG_ON(monc->monmap->num_mon < 1);
139
140 if (monc->monmap->num_mon == 1) {
141 monc->cur_mon = 0;
142 } else {
143 int max = monc->monmap->num_mon;
144 int o = -1;
145 int n;
146
147 if (monc->cur_mon >= 0) {
148 if (monc->cur_mon < monc->monmap->num_mon)
149 o = monc->cur_mon;
150 if (o >= 0)
151 max--;
152 }
153
154 n = prandom_u32() % max;
155 if (o >= 0 && n >= o)
156 n++;
157
158 monc->cur_mon = n;
159 }
160
161 dout("%s mon%d -> mon%d out of %d mons\n", __func__, old_mon,
162 monc->cur_mon, monc->monmap->num_mon);
163 }
164
165 /*
166 * Open a session with a new monitor.
167 */
168 static void __open_session(struct ceph_mon_client *monc)
169 {
170 int ret;
171
172 pick_new_mon(monc);
173
174 monc->hunting = true;
175 if (monc->had_a_connection) {
176 monc->hunt_mult *= CEPH_MONC_HUNT_BACKOFF;
177 if (monc->hunt_mult > CEPH_MONC_HUNT_MAX_MULT)
178 monc->hunt_mult = CEPH_MONC_HUNT_MAX_MULT;
179 }
180
181 monc->sub_renew_after = jiffies; /* i.e., expired */
182 monc->sub_renew_sent = 0;
183
184 dout("%s opening mon%d\n", __func__, monc->cur_mon);
185 ceph_con_open(&monc->con, CEPH_ENTITY_TYPE_MON, monc->cur_mon,
186 &monc->monmap->mon_inst[monc->cur_mon].addr);
187
188 /*
189 * send an initial keepalive to ensure our timestamp is valid
190 * by the time we are in an OPENED state
191 */
192 ceph_con_keepalive(&monc->con);
193
194 /* initiate authentication handshake */
195 ret = ceph_auth_build_hello(monc->auth,
196 monc->m_auth->front.iov_base,
197 monc->m_auth->front_alloc_len);
198 BUG_ON(ret <= 0);
199 __send_prepared_auth_request(monc, ret);
200 }
201
202 static void reopen_session(struct ceph_mon_client *monc)
203 {
204 if (!monc->hunting)
205 pr_info("mon%d %s session lost, hunting for new mon\n",
206 monc->cur_mon, ceph_pr_addr(&monc->con.peer_addr.in_addr));
207
208 __close_session(monc);
209 __open_session(monc);
210 }
211
212 /*
213 * Reschedule delayed work timer.
214 */
215 static void __schedule_delayed(struct ceph_mon_client *monc)
216 {
217 unsigned long delay;
218
219 if (monc->hunting)
220 delay = CEPH_MONC_HUNT_INTERVAL * monc->hunt_mult;
221 else
222 delay = CEPH_MONC_PING_INTERVAL;
223
224 dout("__schedule_delayed after %lu\n", delay);
225 mod_delayed_work(system_wq, &monc->delayed_work,
226 round_jiffies_relative(delay));
227 }
228
229 const char *ceph_sub_str[] = {
230 [CEPH_SUB_MDSMAP] = "mdsmap",
231 [CEPH_SUB_MONMAP] = "monmap",
232 [CEPH_SUB_OSDMAP] = "osdmap",
233 };
234
235 /*
236 * Send subscribe request for one or more maps, according to
237 * monc->subs.
238 */
239 static void __send_subscribe(struct ceph_mon_client *monc)
240 {
241 struct ceph_msg *msg = monc->m_subscribe;
242 void *p = msg->front.iov_base;
243 void *const end = p + msg->front_alloc_len;
244 int num = 0;
245 int i;
246
247 dout("%s sent %lu\n", __func__, monc->sub_renew_sent);
248
249 BUG_ON(monc->cur_mon < 0);
250
251 if (!monc->sub_renew_sent)
252 monc->sub_renew_sent = jiffies | 1; /* never 0 */
253
254 msg->hdr.version = cpu_to_le16(2);
255
256 for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
257 if (monc->subs[i].want)
258 num++;
259 }
260 BUG_ON(num < 1); /* monmap sub is always there */
261 ceph_encode_32(&p, num);
262 for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
263 const char *s = ceph_sub_str[i];
264
265 if (!monc->subs[i].want)
266 continue;
267
268 dout("%s %s start %llu flags 0x%x\n", __func__, s,
269 le64_to_cpu(monc->subs[i].item.start),
270 monc->subs[i].item.flags);
271 ceph_encode_string(&p, end, s, strlen(s));
272 memcpy(p, &monc->subs[i].item, sizeof(monc->subs[i].item));
273 p += sizeof(monc->subs[i].item);
274 }
275
276 BUG_ON(p != (end - 35 - (ARRAY_SIZE(monc->subs) - num) * 19));
277 msg->front.iov_len = p - msg->front.iov_base;
278 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
279 ceph_msg_revoke(msg);
280 ceph_con_send(&monc->con, ceph_msg_get(msg));
281 }
282
283 static void handle_subscribe_ack(struct ceph_mon_client *monc,
284 struct ceph_msg *msg)
285 {
286 unsigned int seconds;
287 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
288
289 if (msg->front.iov_len < sizeof(*h))
290 goto bad;
291 seconds = le32_to_cpu(h->duration);
292
293 mutex_lock(&monc->mutex);
294 if (monc->sub_renew_sent) {
295 monc->sub_renew_after = monc->sub_renew_sent +
296 (seconds >> 1) * HZ - 1;
297 dout("%s sent %lu duration %d renew after %lu\n", __func__,
298 monc->sub_renew_sent, seconds, monc->sub_renew_after);
299 monc->sub_renew_sent = 0;
300 } else {
301 dout("%s sent %lu renew after %lu, ignoring\n", __func__,
302 monc->sub_renew_sent, monc->sub_renew_after);
303 }
304 mutex_unlock(&monc->mutex);
305 return;
306 bad:
307 pr_err("got corrupt subscribe-ack msg\n");
308 ceph_msg_dump(msg);
309 }
310
311 /*
312 * Register interest in a map
313 *
314 * @sub: one of CEPH_SUB_*
315 * @epoch: X for "every map since X", or 0 for "just the latest"
316 */
317 static bool __ceph_monc_want_map(struct ceph_mon_client *monc, int sub,
318 u32 epoch, bool continuous)
319 {
320 __le64 start = cpu_to_le64(epoch);
321 u8 flags = !continuous ? CEPH_SUBSCRIBE_ONETIME : 0;
322
323 dout("%s %s epoch %u continuous %d\n", __func__, ceph_sub_str[sub],
324 epoch, continuous);
325
326 if (monc->subs[sub].want &&
327 monc->subs[sub].item.start == start &&
328 monc->subs[sub].item.flags == flags)
329 return false;
330
331 monc->subs[sub].item.start = start;
332 monc->subs[sub].item.flags = flags;
333 monc->subs[sub].want = true;
334
335 return true;
336 }
337
338 bool ceph_monc_want_map(struct ceph_mon_client *monc, int sub, u32 epoch,
339 bool continuous)
340 {
341 bool need_request;
342
343 mutex_lock(&monc->mutex);
344 need_request = __ceph_monc_want_map(monc, sub, epoch, continuous);
345 mutex_unlock(&monc->mutex);
346
347 return need_request;
348 }
349 EXPORT_SYMBOL(ceph_monc_want_map);
350
351 /*
352 * Keep track of which maps we have
353 *
354 * @sub: one of CEPH_SUB_*
355 */
356 static void __ceph_monc_got_map(struct ceph_mon_client *monc, int sub,
357 u32 epoch)
358 {
359 dout("%s %s epoch %u\n", __func__, ceph_sub_str[sub], epoch);
360
361 if (monc->subs[sub].want) {
362 if (monc->subs[sub].item.flags & CEPH_SUBSCRIBE_ONETIME)
363 monc->subs[sub].want = false;
364 else
365 monc->subs[sub].item.start = cpu_to_le64(epoch + 1);
366 }
367
368 monc->subs[sub].have = epoch;
369 }
370
371 void ceph_monc_got_map(struct ceph_mon_client *monc, int sub, u32 epoch)
372 {
373 mutex_lock(&monc->mutex);
374 __ceph_monc_got_map(monc, sub, epoch);
375 mutex_unlock(&monc->mutex);
376 }
377 EXPORT_SYMBOL(ceph_monc_got_map);
378
379 /*
380 * Register interest in the next osdmap
381 */
382 void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
383 {
384 dout("%s have %u\n", __func__, monc->subs[CEPH_SUB_OSDMAP].have);
385 mutex_lock(&monc->mutex);
386 if (__ceph_monc_want_map(monc, CEPH_SUB_OSDMAP,
387 monc->subs[CEPH_SUB_OSDMAP].have + 1, false))
388 __send_subscribe(monc);
389 mutex_unlock(&monc->mutex);
390 }
391 EXPORT_SYMBOL(ceph_monc_request_next_osdmap);
392
393 /*
394 * Wait for an osdmap with a given epoch.
395 *
396 * @epoch: epoch to wait for
397 * @timeout: in jiffies, 0 means "wait forever"
398 */
399 int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
400 unsigned long timeout)
401 {
402 unsigned long started = jiffies;
403 long ret;
404
405 mutex_lock(&monc->mutex);
406 while (monc->subs[CEPH_SUB_OSDMAP].have < epoch) {
407 mutex_unlock(&monc->mutex);
408
409 if (timeout && time_after_eq(jiffies, started + timeout))
410 return -ETIMEDOUT;
411
412 ret = wait_event_interruptible_timeout(monc->client->auth_wq,
413 monc->subs[CEPH_SUB_OSDMAP].have >= epoch,
414 ceph_timeout_jiffies(timeout));
415 if (ret < 0)
416 return ret;
417
418 mutex_lock(&monc->mutex);
419 }
420
421 mutex_unlock(&monc->mutex);
422 return 0;
423 }
424 EXPORT_SYMBOL(ceph_monc_wait_osdmap);
425
426 /*
427 * Open a session with a random monitor. Request monmap and osdmap,
428 * which are waited upon in __ceph_open_session().
429 */
430 int ceph_monc_open_session(struct ceph_mon_client *monc)
431 {
432 mutex_lock(&monc->mutex);
433 __ceph_monc_want_map(monc, CEPH_SUB_MONMAP, 0, true);
434 __ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, 0, false);
435 __open_session(monc);
436 __schedule_delayed(monc);
437 mutex_unlock(&monc->mutex);
438 return 0;
439 }
440 EXPORT_SYMBOL(ceph_monc_open_session);
441
442 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
443 struct ceph_msg *msg)
444 {
445 struct ceph_client *client = monc->client;
446 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
447 void *p, *end;
448
449 mutex_lock(&monc->mutex);
450
451 dout("handle_monmap\n");
452 p = msg->front.iov_base;
453 end = p + msg->front.iov_len;
454
455 monmap = ceph_monmap_decode(p, end);
456 if (IS_ERR(monmap)) {
457 pr_err("problem decoding monmap, %d\n",
458 (int)PTR_ERR(monmap));
459 goto out;
460 }
461
462 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
463 kfree(monmap);
464 goto out;
465 }
466
467 client->monc.monmap = monmap;
468 kfree(old);
469
470 __ceph_monc_got_map(monc, CEPH_SUB_MONMAP, monc->monmap->epoch);
471 client->have_fsid = true;
472
473 out:
474 mutex_unlock(&monc->mutex);
475 wake_up_all(&client->auth_wq);
476 }
477
478 /*
479 * generic requests (currently statfs, mon_get_version)
480 */
481 static struct ceph_mon_generic_request *__lookup_generic_req(
482 struct ceph_mon_client *monc, u64 tid)
483 {
484 struct ceph_mon_generic_request *req;
485 struct rb_node *n = monc->generic_request_tree.rb_node;
486
487 while (n) {
488 req = rb_entry(n, struct ceph_mon_generic_request, node);
489 if (tid < req->tid)
490 n = n->rb_left;
491 else if (tid > req->tid)
492 n = n->rb_right;
493 else
494 return req;
495 }
496 return NULL;
497 }
498
499 static void __insert_generic_request(struct ceph_mon_client *monc,
500 struct ceph_mon_generic_request *new)
501 {
502 struct rb_node **p = &monc->generic_request_tree.rb_node;
503 struct rb_node *parent = NULL;
504 struct ceph_mon_generic_request *req = NULL;
505
506 while (*p) {
507 parent = *p;
508 req = rb_entry(parent, struct ceph_mon_generic_request, node);
509 if (new->tid < req->tid)
510 p = &(*p)->rb_left;
511 else if (new->tid > req->tid)
512 p = &(*p)->rb_right;
513 else
514 BUG();
515 }
516
517 rb_link_node(&new->node, parent, p);
518 rb_insert_color(&new->node, &monc->generic_request_tree);
519 }
520
521 static void release_generic_request(struct kref *kref)
522 {
523 struct ceph_mon_generic_request *req =
524 container_of(kref, struct ceph_mon_generic_request, kref);
525
526 if (req->reply)
527 ceph_msg_put(req->reply);
528 if (req->request)
529 ceph_msg_put(req->request);
530
531 kfree(req);
532 }
533
534 static void put_generic_request(struct ceph_mon_generic_request *req)
535 {
536 kref_put(&req->kref, release_generic_request);
537 }
538
539 static void get_generic_request(struct ceph_mon_generic_request *req)
540 {
541 kref_get(&req->kref);
542 }
543
544 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
545 struct ceph_msg_header *hdr,
546 int *skip)
547 {
548 struct ceph_mon_client *monc = con->private;
549 struct ceph_mon_generic_request *req;
550 u64 tid = le64_to_cpu(hdr->tid);
551 struct ceph_msg *m;
552
553 mutex_lock(&monc->mutex);
554 req = __lookup_generic_req(monc, tid);
555 if (!req) {
556 dout("get_generic_reply %lld dne\n", tid);
557 *skip = 1;
558 m = NULL;
559 } else {
560 dout("get_generic_reply %lld got %p\n", tid, req->reply);
561 *skip = 0;
562 m = ceph_msg_get(req->reply);
563 /*
564 * we don't need to track the connection reading into
565 * this reply because we only have one open connection
566 * at a time, ever.
567 */
568 }
569 mutex_unlock(&monc->mutex);
570 return m;
571 }
572
573 static int __do_generic_request(struct ceph_mon_client *monc, u64 tid,
574 struct ceph_mon_generic_request *req)
575 {
576 int err;
577
578 /* register request */
579 req->tid = tid != 0 ? tid : ++monc->last_tid;
580 req->request->hdr.tid = cpu_to_le64(req->tid);
581 __insert_generic_request(monc, req);
582 ceph_con_send(&monc->con, ceph_msg_get(req->request));
583 mutex_unlock(&monc->mutex);
584
585 err = wait_for_completion_interruptible(&req->completion);
586
587 mutex_lock(&monc->mutex);
588 rb_erase(&req->node, &monc->generic_request_tree);
589
590 if (!err)
591 err = req->result;
592 return err;
593 }
594
595 static int do_generic_request(struct ceph_mon_client *monc,
596 struct ceph_mon_generic_request *req)
597 {
598 int err;
599
600 mutex_lock(&monc->mutex);
601 err = __do_generic_request(monc, 0, req);
602 mutex_unlock(&monc->mutex);
603
604 return err;
605 }
606
607 /*
608 * statfs
609 */
610 static void handle_statfs_reply(struct ceph_mon_client *monc,
611 struct ceph_msg *msg)
612 {
613 struct ceph_mon_generic_request *req;
614 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
615 u64 tid = le64_to_cpu(msg->hdr.tid);
616
617 if (msg->front.iov_len != sizeof(*reply))
618 goto bad;
619 dout("handle_statfs_reply %p tid %llu\n", msg, tid);
620
621 mutex_lock(&monc->mutex);
622 req = __lookup_generic_req(monc, tid);
623 if (req) {
624 *(struct ceph_statfs *)req->buf = reply->st;
625 req->result = 0;
626 get_generic_request(req);
627 }
628 mutex_unlock(&monc->mutex);
629 if (req) {
630 complete_all(&req->completion);
631 put_generic_request(req);
632 }
633 return;
634
635 bad:
636 pr_err("corrupt statfs reply, tid %llu\n", tid);
637 ceph_msg_dump(msg);
638 }
639
640 /*
641 * Do a synchronous statfs().
642 */
643 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
644 {
645 struct ceph_mon_generic_request *req;
646 struct ceph_mon_statfs *h;
647 int err;
648
649 req = kzalloc(sizeof(*req), GFP_NOFS);
650 if (!req)
651 return -ENOMEM;
652
653 kref_init(&req->kref);
654 req->buf = buf;
655 init_completion(&req->completion);
656
657 err = -ENOMEM;
658 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
659 true);
660 if (!req->request)
661 goto out;
662 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
663 true);
664 if (!req->reply)
665 goto out;
666
667 /* fill out request */
668 h = req->request->front.iov_base;
669 h->monhdr.have_version = 0;
670 h->monhdr.session_mon = cpu_to_le16(-1);
671 h->monhdr.session_mon_tid = 0;
672 h->fsid = monc->monmap->fsid;
673
674 err = do_generic_request(monc, req);
675
676 out:
677 put_generic_request(req);
678 return err;
679 }
680 EXPORT_SYMBOL(ceph_monc_do_statfs);
681
682 static void handle_get_version_reply(struct ceph_mon_client *monc,
683 struct ceph_msg *msg)
684 {
685 struct ceph_mon_generic_request *req;
686 u64 tid = le64_to_cpu(msg->hdr.tid);
687 void *p = msg->front.iov_base;
688 void *end = p + msg->front_alloc_len;
689 u64 handle;
690
691 dout("%s %p tid %llu\n", __func__, msg, tid);
692
693 ceph_decode_need(&p, end, 2*sizeof(u64), bad);
694 handle = ceph_decode_64(&p);
695 if (tid != 0 && tid != handle)
696 goto bad;
697
698 mutex_lock(&monc->mutex);
699 req = __lookup_generic_req(monc, handle);
700 if (req) {
701 *(u64 *)req->buf = ceph_decode_64(&p);
702 req->result = 0;
703 get_generic_request(req);
704 }
705 mutex_unlock(&monc->mutex);
706 if (req) {
707 complete_all(&req->completion);
708 put_generic_request(req);
709 }
710
711 return;
712 bad:
713 pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
714 ceph_msg_dump(msg);
715 }
716
717 /*
718 * Send MMonGetVersion and wait for the reply.
719 *
720 * @what: one of "mdsmap", "osdmap" or "monmap"
721 */
722 int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what,
723 u64 *newest)
724 {
725 struct ceph_mon_generic_request *req;
726 void *p, *end;
727 u64 tid;
728 int err;
729
730 req = kzalloc(sizeof(*req), GFP_NOFS);
731 if (!req)
732 return -ENOMEM;
733
734 kref_init(&req->kref);
735 req->buf = newest;
736 init_completion(&req->completion);
737
738 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
739 sizeof(u64) + sizeof(u32) + strlen(what),
740 GFP_NOFS, true);
741 if (!req->request) {
742 err = -ENOMEM;
743 goto out;
744 }
745
746 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024,
747 GFP_NOFS, true);
748 if (!req->reply) {
749 err = -ENOMEM;
750 goto out;
751 }
752
753 p = req->request->front.iov_base;
754 end = p + req->request->front_alloc_len;
755
756 /* fill out request */
757 mutex_lock(&monc->mutex);
758 tid = ++monc->last_tid;
759 ceph_encode_64(&p, tid); /* handle */
760 ceph_encode_string(&p, end, what, strlen(what));
761
762 err = __do_generic_request(monc, tid, req);
763
764 mutex_unlock(&monc->mutex);
765 out:
766 put_generic_request(req);
767 return err;
768 }
769 EXPORT_SYMBOL(ceph_monc_do_get_version);
770
771 /*
772 * Resend pending generic requests.
773 */
774 static void __resend_generic_request(struct ceph_mon_client *monc)
775 {
776 struct ceph_mon_generic_request *req;
777 struct rb_node *p;
778
779 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
780 req = rb_entry(p, struct ceph_mon_generic_request, node);
781 ceph_msg_revoke(req->request);
782 ceph_msg_revoke_incoming(req->reply);
783 ceph_con_send(&monc->con, ceph_msg_get(req->request));
784 }
785 }
786
787 /*
788 * Delayed work. If we haven't mounted yet, retry. Otherwise,
789 * renew/retry subscription as needed (in case it is timing out, or we
790 * got an ENOMEM). And keep the monitor connection alive.
791 */
792 static void delayed_work(struct work_struct *work)
793 {
794 struct ceph_mon_client *monc =
795 container_of(work, struct ceph_mon_client, delayed_work.work);
796
797 dout("monc delayed_work\n");
798 mutex_lock(&monc->mutex);
799 if (monc->hunting) {
800 dout("%s continuing hunt\n", __func__);
801 reopen_session(monc);
802 } else {
803 int is_auth = ceph_auth_is_authenticated(monc->auth);
804 if (ceph_con_keepalive_expired(&monc->con,
805 CEPH_MONC_PING_TIMEOUT)) {
806 dout("monc keepalive timeout\n");
807 is_auth = 0;
808 reopen_session(monc);
809 }
810
811 if (!monc->hunting) {
812 ceph_con_keepalive(&monc->con);
813 __validate_auth(monc);
814 }
815
816 if (is_auth) {
817 unsigned long now = jiffies;
818
819 dout("%s renew subs? now %lu renew after %lu\n",
820 __func__, now, monc->sub_renew_after);
821 if (time_after_eq(now, monc->sub_renew_after))
822 __send_subscribe(monc);
823 }
824 }
825 __schedule_delayed(monc);
826 mutex_unlock(&monc->mutex);
827 }
828
829 /*
830 * On startup, we build a temporary monmap populated with the IPs
831 * provided by mount(2).
832 */
833 static int build_initial_monmap(struct ceph_mon_client *monc)
834 {
835 struct ceph_options *opt = monc->client->options;
836 struct ceph_entity_addr *mon_addr = opt->mon_addr;
837 int num_mon = opt->num_mon;
838 int i;
839
840 /* build initial monmap */
841 monc->monmap = kzalloc(sizeof(*monc->monmap) +
842 num_mon*sizeof(monc->monmap->mon_inst[0]),
843 GFP_KERNEL);
844 if (!monc->monmap)
845 return -ENOMEM;
846 for (i = 0; i < num_mon; i++) {
847 monc->monmap->mon_inst[i].addr = mon_addr[i];
848 monc->monmap->mon_inst[i].addr.nonce = 0;
849 monc->monmap->mon_inst[i].name.type =
850 CEPH_ENTITY_TYPE_MON;
851 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
852 }
853 monc->monmap->num_mon = num_mon;
854 return 0;
855 }
856
857 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
858 {
859 int err = 0;
860
861 dout("init\n");
862 memset(monc, 0, sizeof(*monc));
863 monc->client = cl;
864 monc->monmap = NULL;
865 mutex_init(&monc->mutex);
866
867 err = build_initial_monmap(monc);
868 if (err)
869 goto out;
870
871 /* connection */
872 /* authentication */
873 monc->auth = ceph_auth_init(cl->options->name,
874 cl->options->key);
875 if (IS_ERR(monc->auth)) {
876 err = PTR_ERR(monc->auth);
877 goto out_monmap;
878 }
879 monc->auth->want_keys =
880 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
881 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
882
883 /* msgs */
884 err = -ENOMEM;
885 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
886 sizeof(struct ceph_mon_subscribe_ack),
887 GFP_NOFS, true);
888 if (!monc->m_subscribe_ack)
889 goto out_auth;
890
891 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
892 true);
893 if (!monc->m_subscribe)
894 goto out_subscribe_ack;
895
896 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
897 true);
898 if (!monc->m_auth_reply)
899 goto out_subscribe;
900
901 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
902 monc->pending_auth = 0;
903 if (!monc->m_auth)
904 goto out_auth_reply;
905
906 ceph_con_init(&monc->con, monc, &mon_con_ops,
907 &monc->client->msgr);
908
909 monc->cur_mon = -1;
910 monc->had_a_connection = false;
911 monc->hunt_mult = 1;
912
913 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
914 monc->generic_request_tree = RB_ROOT;
915 monc->last_tid = 0;
916
917 return 0;
918
919 out_auth_reply:
920 ceph_msg_put(monc->m_auth_reply);
921 out_subscribe:
922 ceph_msg_put(monc->m_subscribe);
923 out_subscribe_ack:
924 ceph_msg_put(monc->m_subscribe_ack);
925 out_auth:
926 ceph_auth_destroy(monc->auth);
927 out_monmap:
928 kfree(monc->monmap);
929 out:
930 return err;
931 }
932 EXPORT_SYMBOL(ceph_monc_init);
933
934 void ceph_monc_stop(struct ceph_mon_client *monc)
935 {
936 dout("stop\n");
937 cancel_delayed_work_sync(&monc->delayed_work);
938
939 mutex_lock(&monc->mutex);
940 __close_session(monc);
941 monc->cur_mon = -1;
942 mutex_unlock(&monc->mutex);
943
944 /*
945 * flush msgr queue before we destroy ourselves to ensure that:
946 * - any work that references our embedded con is finished.
947 * - any osd_client or other work that may reference an authorizer
948 * finishes before we shut down the auth subsystem.
949 */
950 ceph_msgr_flush();
951
952 ceph_auth_destroy(monc->auth);
953
954 ceph_msg_put(monc->m_auth);
955 ceph_msg_put(monc->m_auth_reply);
956 ceph_msg_put(monc->m_subscribe);
957 ceph_msg_put(monc->m_subscribe_ack);
958
959 kfree(monc->monmap);
960 }
961 EXPORT_SYMBOL(ceph_monc_stop);
962
963 static void finish_hunting(struct ceph_mon_client *monc)
964 {
965 if (monc->hunting) {
966 dout("%s found mon%d\n", __func__, monc->cur_mon);
967 monc->hunting = false;
968 monc->had_a_connection = true;
969 monc->hunt_mult /= 2; /* reduce by 50% */
970 if (monc->hunt_mult < 1)
971 monc->hunt_mult = 1;
972 }
973 }
974
975 static void handle_auth_reply(struct ceph_mon_client *monc,
976 struct ceph_msg *msg)
977 {
978 int ret;
979 int was_auth = 0;
980
981 mutex_lock(&monc->mutex);
982 was_auth = ceph_auth_is_authenticated(monc->auth);
983 monc->pending_auth = 0;
984 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
985 msg->front.iov_len,
986 monc->m_auth->front.iov_base,
987 monc->m_auth->front_alloc_len);
988 if (ret > 0) {
989 __send_prepared_auth_request(monc, ret);
990 goto out;
991 }
992
993 finish_hunting(monc);
994
995 if (ret < 0) {
996 monc->client->auth_err = ret;
997 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
998 dout("authenticated, starting session\n");
999
1000 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
1001 monc->client->msgr.inst.name.num =
1002 cpu_to_le64(monc->auth->global_id);
1003
1004 __send_subscribe(monc);
1005 __resend_generic_request(monc);
1006
1007 pr_info("mon%d %s session established\n", monc->cur_mon,
1008 ceph_pr_addr(&monc->con.peer_addr.in_addr));
1009 }
1010
1011 out:
1012 mutex_unlock(&monc->mutex);
1013 if (monc->client->auth_err < 0)
1014 wake_up_all(&monc->client->auth_wq);
1015 }
1016
1017 static int __validate_auth(struct ceph_mon_client *monc)
1018 {
1019 int ret;
1020
1021 if (monc->pending_auth)
1022 return 0;
1023
1024 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
1025 monc->m_auth->front_alloc_len);
1026 if (ret <= 0)
1027 return ret; /* either an error, or no need to authenticate */
1028 __send_prepared_auth_request(monc, ret);
1029 return 0;
1030 }
1031
1032 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
1033 {
1034 int ret;
1035
1036 mutex_lock(&monc->mutex);
1037 ret = __validate_auth(monc);
1038 mutex_unlock(&monc->mutex);
1039 return ret;
1040 }
1041 EXPORT_SYMBOL(ceph_monc_validate_auth);
1042
1043 /*
1044 * handle incoming message
1045 */
1046 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1047 {
1048 struct ceph_mon_client *monc = con->private;
1049 int type = le16_to_cpu(msg->hdr.type);
1050
1051 if (!monc)
1052 return;
1053
1054 switch (type) {
1055 case CEPH_MSG_AUTH_REPLY:
1056 handle_auth_reply(monc, msg);
1057 break;
1058
1059 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1060 handle_subscribe_ack(monc, msg);
1061 break;
1062
1063 case CEPH_MSG_STATFS_REPLY:
1064 handle_statfs_reply(monc, msg);
1065 break;
1066
1067 case CEPH_MSG_MON_GET_VERSION_REPLY:
1068 handle_get_version_reply(monc, msg);
1069 break;
1070
1071 case CEPH_MSG_MON_MAP:
1072 ceph_monc_handle_map(monc, msg);
1073 break;
1074
1075 case CEPH_MSG_OSD_MAP:
1076 ceph_osdc_handle_map(&monc->client->osdc, msg);
1077 break;
1078
1079 default:
1080 /* can the chained handler handle it? */
1081 if (monc->client->extra_mon_dispatch &&
1082 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1083 break;
1084
1085 pr_err("received unknown message type %d %s\n", type,
1086 ceph_msg_type_name(type));
1087 }
1088 ceph_msg_put(msg);
1089 }
1090
1091 /*
1092 * Allocate memory for incoming message
1093 */
1094 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1095 struct ceph_msg_header *hdr,
1096 int *skip)
1097 {
1098 struct ceph_mon_client *monc = con->private;
1099 int type = le16_to_cpu(hdr->type);
1100 int front_len = le32_to_cpu(hdr->front_len);
1101 struct ceph_msg *m = NULL;
1102
1103 *skip = 0;
1104
1105 switch (type) {
1106 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1107 m = ceph_msg_get(monc->m_subscribe_ack);
1108 break;
1109 case CEPH_MSG_STATFS_REPLY:
1110 return get_generic_reply(con, hdr, skip);
1111 case CEPH_MSG_AUTH_REPLY:
1112 m = ceph_msg_get(monc->m_auth_reply);
1113 break;
1114 case CEPH_MSG_MON_GET_VERSION_REPLY:
1115 if (le64_to_cpu(hdr->tid) != 0)
1116 return get_generic_reply(con, hdr, skip);
1117
1118 /*
1119 * Older OSDs don't set reply tid even if the orignal
1120 * request had a non-zero tid. Workaround this weirdness
1121 * by falling through to the allocate case.
1122 */
1123 case CEPH_MSG_MON_MAP:
1124 case CEPH_MSG_MDS_MAP:
1125 case CEPH_MSG_OSD_MAP:
1126 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1127 if (!m)
1128 return NULL; /* ENOMEM--return skip == 0 */
1129 break;
1130 }
1131
1132 if (!m) {
1133 pr_info("alloc_msg unknown type %d\n", type);
1134 *skip = 1;
1135 } else if (front_len > m->front_alloc_len) {
1136 pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1137 front_len, m->front_alloc_len,
1138 (unsigned int)con->peer_name.type,
1139 le64_to_cpu(con->peer_name.num));
1140 ceph_msg_put(m);
1141 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1142 }
1143
1144 return m;
1145 }
1146
1147 /*
1148 * If the monitor connection resets, pick a new monitor and resubmit
1149 * any pending requests.
1150 */
1151 static void mon_fault(struct ceph_connection *con)
1152 {
1153 struct ceph_mon_client *monc = con->private;
1154
1155 mutex_lock(&monc->mutex);
1156 dout("%s mon%d\n", __func__, monc->cur_mon);
1157 if (monc->cur_mon >= 0) {
1158 if (!monc->hunting) {
1159 dout("%s hunting for new mon\n", __func__);
1160 reopen_session(monc);
1161 __schedule_delayed(monc);
1162 } else {
1163 dout("%s already hunting\n", __func__);
1164 }
1165 }
1166 mutex_unlock(&monc->mutex);
1167 }
1168
1169 /*
1170 * We can ignore refcounting on the connection struct, as all references
1171 * will come from the messenger workqueue, which is drained prior to
1172 * mon_client destruction.
1173 */
1174 static struct ceph_connection *con_get(struct ceph_connection *con)
1175 {
1176 return con;
1177 }
1178
1179 static void con_put(struct ceph_connection *con)
1180 {
1181 }
1182
1183 static const struct ceph_connection_operations mon_con_ops = {
1184 .get = con_get,
1185 .put = con_put,
1186 .dispatch = dispatch,
1187 .fault = mon_fault,
1188 .alloc_msg = mon_alloc_msg,
1189 };
(deprecated) Btrfs devel tree