projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
libceph: switch to calc_target(), part 1
[deliverable/linux.git] / net / ceph / mon_client.c
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 DEFINE_RB_FUNCS(generic_request, struct ceph_mon_generic_request, tid, node)
482
483 static void release_generic_request(struct kref *kref)
484 {
485 struct ceph_mon_generic_request *req =
486 container_of(kref, struct ceph_mon_generic_request, kref);
487
488 if (req->reply)
489 ceph_msg_put(req->reply);
490 if (req->request)
491 ceph_msg_put(req->request);
492
493 kfree(req);
494 }
495
496 static void put_generic_request(struct ceph_mon_generic_request *req)
497 {
498 kref_put(&req->kref, release_generic_request);
499 }
500
501 static void get_generic_request(struct ceph_mon_generic_request *req)
502 {
503 kref_get(&req->kref);
504 }
505
506 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
507 struct ceph_msg_header *hdr,
508 int *skip)
509 {
510 struct ceph_mon_client *monc = con->private;
511 struct ceph_mon_generic_request *req;
512 u64 tid = le64_to_cpu(hdr->tid);
513 struct ceph_msg *m;
514
515 mutex_lock(&monc->mutex);
516 req = lookup_generic_request(&monc->generic_request_tree, tid);
517 if (!req) {
518 dout("get_generic_reply %lld dne\n", tid);
519 *skip = 1;
520 m = NULL;
521 } else {
522 dout("get_generic_reply %lld got %p\n", tid, req->reply);
523 *skip = 0;
524 m = ceph_msg_get(req->reply);
525 /*
526 * we don't need to track the connection reading into
527 * this reply because we only have one open connection
528 * at a time, ever.
529 */
530 }
531 mutex_unlock(&monc->mutex);
532 return m;
533 }
534
535 static int __do_generic_request(struct ceph_mon_client *monc, u64 tid,
536 struct ceph_mon_generic_request *req)
537 {
538 int err;
539
540 /* register request */
541 req->tid = tid != 0 ? tid : ++monc->last_tid;
542 req->request->hdr.tid = cpu_to_le64(req->tid);
543 insert_generic_request(&monc->generic_request_tree, req);
544 ceph_con_send(&monc->con, ceph_msg_get(req->request));
545 mutex_unlock(&monc->mutex);
546
547 err = wait_for_completion_interruptible(&req->completion);
548
549 mutex_lock(&monc->mutex);
550 erase_generic_request(&monc->generic_request_tree, req);
551
552 if (!err)
553 err = req->result;
554 return err;
555 }
556
557 static int do_generic_request(struct ceph_mon_client *monc,
558 struct ceph_mon_generic_request *req)
559 {
560 int err;
561
562 mutex_lock(&monc->mutex);
563 err = __do_generic_request(monc, 0, req);
564 mutex_unlock(&monc->mutex);
565
566 return err;
567 }
568
569 /*
570 * statfs
571 */
572 static void handle_statfs_reply(struct ceph_mon_client *monc,
573 struct ceph_msg *msg)
574 {
575 struct ceph_mon_generic_request *req;
576 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
577 u64 tid = le64_to_cpu(msg->hdr.tid);
578
579 if (msg->front.iov_len != sizeof(*reply))
580 goto bad;
581 dout("handle_statfs_reply %p tid %llu\n", msg, tid);
582
583 mutex_lock(&monc->mutex);
584 req = lookup_generic_request(&monc->generic_request_tree, tid);
585 if (req) {
586 *(struct ceph_statfs *)req->buf = reply->st;
587 req->result = 0;
588 get_generic_request(req);
589 }
590 mutex_unlock(&monc->mutex);
591 if (req) {
592 complete_all(&req->completion);
593 put_generic_request(req);
594 }
595 return;
596
597 bad:
598 pr_err("corrupt statfs reply, tid %llu\n", tid);
599 ceph_msg_dump(msg);
600 }
601
602 /*
603 * Do a synchronous statfs().
604 */
605 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
606 {
607 struct ceph_mon_generic_request *req;
608 struct ceph_mon_statfs *h;
609 int err;
610
611 req = kzalloc(sizeof(*req), GFP_NOFS);
612 if (!req)
613 return -ENOMEM;
614
615 kref_init(&req->kref);
616 RB_CLEAR_NODE(&req->node);
617 req->buf = buf;
618 init_completion(&req->completion);
619
620 err = -ENOMEM;
621 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
622 true);
623 if (!req->request)
624 goto out;
625 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
626 true);
627 if (!req->reply)
628 goto out;
629
630 /* fill out request */
631 h = req->request->front.iov_base;
632 h->monhdr.have_version = 0;
633 h->monhdr.session_mon = cpu_to_le16(-1);
634 h->monhdr.session_mon_tid = 0;
635 h->fsid = monc->monmap->fsid;
636
637 err = do_generic_request(monc, req);
638
639 out:
640 put_generic_request(req);
641 return err;
642 }
643 EXPORT_SYMBOL(ceph_monc_do_statfs);
644
645 static void handle_get_version_reply(struct ceph_mon_client *monc,
646 struct ceph_msg *msg)
647 {
648 struct ceph_mon_generic_request *req;
649 u64 tid = le64_to_cpu(msg->hdr.tid);
650 void *p = msg->front.iov_base;
651 void *end = p + msg->front_alloc_len;
652 u64 handle;
653
654 dout("%s %p tid %llu\n", __func__, msg, tid);
655
656 ceph_decode_need(&p, end, 2*sizeof(u64), bad);
657 handle = ceph_decode_64(&p);
658 if (tid != 0 && tid != handle)
659 goto bad;
660
661 mutex_lock(&monc->mutex);
662 req = lookup_generic_request(&monc->generic_request_tree, handle);
663 if (req) {
664 *(u64 *)req->buf = ceph_decode_64(&p);
665 req->result = 0;
666 get_generic_request(req);
667 }
668 mutex_unlock(&monc->mutex);
669 if (req) {
670 complete_all(&req->completion);
671 put_generic_request(req);
672 }
673
674 return;
675 bad:
676 pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
677 ceph_msg_dump(msg);
678 }
679
680 /*
681 * Send MMonGetVersion and wait for the reply.
682 *
683 * @what: one of "mdsmap", "osdmap" or "monmap"
684 */
685 int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what,
686 u64 *newest)
687 {
688 struct ceph_mon_generic_request *req;
689 void *p, *end;
690 u64 tid;
691 int err;
692
693 req = kzalloc(sizeof(*req), GFP_NOFS);
694 if (!req)
695 return -ENOMEM;
696
697 kref_init(&req->kref);
698 RB_CLEAR_NODE(&req->node);
699 req->buf = newest;
700 init_completion(&req->completion);
701
702 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
703 sizeof(u64) + sizeof(u32) + strlen(what),
704 GFP_NOFS, true);
705 if (!req->request) {
706 err = -ENOMEM;
707 goto out;
708 }
709
710 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024,
711 GFP_NOFS, true);
712 if (!req->reply) {
713 err = -ENOMEM;
714 goto out;
715 }
716
717 p = req->request->front.iov_base;
718 end = p + req->request->front_alloc_len;
719
720 /* fill out request */
721 mutex_lock(&monc->mutex);
722 tid = ++monc->last_tid;
723 ceph_encode_64(&p, tid); /* handle */
724 ceph_encode_string(&p, end, what, strlen(what));
725
726 err = __do_generic_request(monc, tid, req);
727
728 mutex_unlock(&monc->mutex);
729 out:
730 put_generic_request(req);
731 return err;
732 }
733 EXPORT_SYMBOL(ceph_monc_do_get_version);
734
735 /*
736 * Resend pending generic requests.
737 */
738 static void __resend_generic_request(struct ceph_mon_client *monc)
739 {
740 struct ceph_mon_generic_request *req;
741 struct rb_node *p;
742
743 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
744 req = rb_entry(p, struct ceph_mon_generic_request, node);
745 ceph_msg_revoke(req->request);
746 ceph_msg_revoke_incoming(req->reply);
747 ceph_con_send(&monc->con, ceph_msg_get(req->request));
748 }
749 }
750
751 /*
752 * Delayed work. If we haven't mounted yet, retry. Otherwise,
753 * renew/retry subscription as needed (in case it is timing out, or we
754 * got an ENOMEM). And keep the monitor connection alive.
755 */
756 static void delayed_work(struct work_struct *work)
757 {
758 struct ceph_mon_client *monc =
759 container_of(work, struct ceph_mon_client, delayed_work.work);
760
761 dout("monc delayed_work\n");
762 mutex_lock(&monc->mutex);
763 if (monc->hunting) {
764 dout("%s continuing hunt\n", __func__);
765 reopen_session(monc);
766 } else {
767 int is_auth = ceph_auth_is_authenticated(monc->auth);
768 if (ceph_con_keepalive_expired(&monc->con,
769 CEPH_MONC_PING_TIMEOUT)) {
770 dout("monc keepalive timeout\n");
771 is_auth = 0;
772 reopen_session(monc);
773 }
774
775 if (!monc->hunting) {
776 ceph_con_keepalive(&monc->con);
777 __validate_auth(monc);
778 }
779
780 if (is_auth) {
781 unsigned long now = jiffies;
782
783 dout("%s renew subs? now %lu renew after %lu\n",
784 __func__, now, monc->sub_renew_after);
785 if (time_after_eq(now, monc->sub_renew_after))
786 __send_subscribe(monc);
787 }
788 }
789 __schedule_delayed(monc);
790 mutex_unlock(&monc->mutex);
791 }
792
793 /*
794 * On startup, we build a temporary monmap populated with the IPs
795 * provided by mount(2).
796 */
797 static int build_initial_monmap(struct ceph_mon_client *monc)
798 {
799 struct ceph_options *opt = monc->client->options;
800 struct ceph_entity_addr *mon_addr = opt->mon_addr;
801 int num_mon = opt->num_mon;
802 int i;
803
804 /* build initial monmap */
805 monc->monmap = kzalloc(sizeof(*monc->monmap) +
806 num_mon*sizeof(monc->monmap->mon_inst[0]),
807 GFP_KERNEL);
808 if (!monc->monmap)
809 return -ENOMEM;
810 for (i = 0; i < num_mon; i++) {
811 monc->monmap->mon_inst[i].addr = mon_addr[i];
812 monc->monmap->mon_inst[i].addr.nonce = 0;
813 monc->monmap->mon_inst[i].name.type =
814 CEPH_ENTITY_TYPE_MON;
815 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
816 }
817 monc->monmap->num_mon = num_mon;
818 return 0;
819 }
820
821 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
822 {
823 int err = 0;
824
825 dout("init\n");
826 memset(monc, 0, sizeof(*monc));
827 monc->client = cl;
828 monc->monmap = NULL;
829 mutex_init(&monc->mutex);
830
831 err = build_initial_monmap(monc);
832 if (err)
833 goto out;
834
835 /* connection */
836 /* authentication */
837 monc->auth = ceph_auth_init(cl->options->name,
838 cl->options->key);
839 if (IS_ERR(monc->auth)) {
840 err = PTR_ERR(monc->auth);
841 goto out_monmap;
842 }
843 monc->auth->want_keys =
844 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
845 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
846
847 /* msgs */
848 err = -ENOMEM;
849 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
850 sizeof(struct ceph_mon_subscribe_ack),
851 GFP_NOFS, true);
852 if (!monc->m_subscribe_ack)
853 goto out_auth;
854
855 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
856 true);
857 if (!monc->m_subscribe)
858 goto out_subscribe_ack;
859
860 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
861 true);
862 if (!monc->m_auth_reply)
863 goto out_subscribe;
864
865 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
866 monc->pending_auth = 0;
867 if (!monc->m_auth)
868 goto out_auth_reply;
869
870 ceph_con_init(&monc->con, monc, &mon_con_ops,
871 &monc->client->msgr);
872
873 monc->cur_mon = -1;
874 monc->had_a_connection = false;
875 monc->hunt_mult = 1;
876
877 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
878 monc->generic_request_tree = RB_ROOT;
879 monc->last_tid = 0;
880
881 return 0;
882
883 out_auth_reply:
884 ceph_msg_put(monc->m_auth_reply);
885 out_subscribe:
886 ceph_msg_put(monc->m_subscribe);
887 out_subscribe_ack:
888 ceph_msg_put(monc->m_subscribe_ack);
889 out_auth:
890 ceph_auth_destroy(monc->auth);
891 out_monmap:
892 kfree(monc->monmap);
893 out:
894 return err;
895 }
896 EXPORT_SYMBOL(ceph_monc_init);
897
898 void ceph_monc_stop(struct ceph_mon_client *monc)
899 {
900 dout("stop\n");
901 cancel_delayed_work_sync(&monc->delayed_work);
902
903 mutex_lock(&monc->mutex);
904 __close_session(monc);
905 monc->cur_mon = -1;
906 mutex_unlock(&monc->mutex);
907
908 /*
909 * flush msgr queue before we destroy ourselves to ensure that:
910 * - any work that references our embedded con is finished.
911 * - any osd_client or other work that may reference an authorizer
912 * finishes before we shut down the auth subsystem.
913 */
914 ceph_msgr_flush();
915
916 ceph_auth_destroy(monc->auth);
917
918 ceph_msg_put(monc->m_auth);
919 ceph_msg_put(monc->m_auth_reply);
920 ceph_msg_put(monc->m_subscribe);
921 ceph_msg_put(monc->m_subscribe_ack);
922
923 kfree(monc->monmap);
924 }
925 EXPORT_SYMBOL(ceph_monc_stop);
926
927 static void finish_hunting(struct ceph_mon_client *monc)
928 {
929 if (monc->hunting) {
930 dout("%s found mon%d\n", __func__, monc->cur_mon);
931 monc->hunting = false;
932 monc->had_a_connection = true;
933 monc->hunt_mult /= 2; /* reduce by 50% */
934 if (monc->hunt_mult < 1)
935 monc->hunt_mult = 1;
936 }
937 }
938
939 static void handle_auth_reply(struct ceph_mon_client *monc,
940 struct ceph_msg *msg)
941 {
942 int ret;
943 int was_auth = 0;
944
945 mutex_lock(&monc->mutex);
946 was_auth = ceph_auth_is_authenticated(monc->auth);
947 monc->pending_auth = 0;
948 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
949 msg->front.iov_len,
950 monc->m_auth->front.iov_base,
951 monc->m_auth->front_alloc_len);
952 if (ret > 0) {
953 __send_prepared_auth_request(monc, ret);
954 goto out;
955 }
956
957 finish_hunting(monc);
958
959 if (ret < 0) {
960 monc->client->auth_err = ret;
961 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
962 dout("authenticated, starting session\n");
963
964 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
965 monc->client->msgr.inst.name.num =
966 cpu_to_le64(monc->auth->global_id);
967
968 __send_subscribe(monc);
969 __resend_generic_request(monc);
970
971 pr_info("mon%d %s session established\n", monc->cur_mon,
972 ceph_pr_addr(&monc->con.peer_addr.in_addr));
973 }
974
975 out:
976 mutex_unlock(&monc->mutex);
977 if (monc->client->auth_err < 0)
978 wake_up_all(&monc->client->auth_wq);
979 }
980
981 static int __validate_auth(struct ceph_mon_client *monc)
982 {
983 int ret;
984
985 if (monc->pending_auth)
986 return 0;
987
988 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
989 monc->m_auth->front_alloc_len);
990 if (ret <= 0)
991 return ret; /* either an error, or no need to authenticate */
992 __send_prepared_auth_request(monc, ret);
993 return 0;
994 }
995
996 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
997 {
998 int ret;
999
1000 mutex_lock(&monc->mutex);
1001 ret = __validate_auth(monc);
1002 mutex_unlock(&monc->mutex);
1003 return ret;
1004 }
1005 EXPORT_SYMBOL(ceph_monc_validate_auth);
1006
1007 /*
1008 * handle incoming message
1009 */
1010 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1011 {
1012 struct ceph_mon_client *monc = con->private;
1013 int type = le16_to_cpu(msg->hdr.type);
1014
1015 if (!monc)
1016 return;
1017
1018 switch (type) {
1019 case CEPH_MSG_AUTH_REPLY:
1020 handle_auth_reply(monc, msg);
1021 break;
1022
1023 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1024 handle_subscribe_ack(monc, msg);
1025 break;
1026
1027 case CEPH_MSG_STATFS_REPLY:
1028 handle_statfs_reply(monc, msg);
1029 break;
1030
1031 case CEPH_MSG_MON_GET_VERSION_REPLY:
1032 handle_get_version_reply(monc, msg);
1033 break;
1034
1035 case CEPH_MSG_MON_MAP:
1036 ceph_monc_handle_map(monc, msg);
1037 break;
1038
1039 case CEPH_MSG_OSD_MAP:
1040 ceph_osdc_handle_map(&monc->client->osdc, msg);
1041 break;
1042
1043 default:
1044 /* can the chained handler handle it? */
1045 if (monc->client->extra_mon_dispatch &&
1046 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1047 break;
1048
1049 pr_err("received unknown message type %d %s\n", type,
1050 ceph_msg_type_name(type));
1051 }
1052 ceph_msg_put(msg);
1053 }
1054
1055 /*
1056 * Allocate memory for incoming message
1057 */
1058 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1059 struct ceph_msg_header *hdr,
1060 int *skip)
1061 {
1062 struct ceph_mon_client *monc = con->private;
1063 int type = le16_to_cpu(hdr->type);
1064 int front_len = le32_to_cpu(hdr->front_len);
1065 struct ceph_msg *m = NULL;
1066
1067 *skip = 0;
1068
1069 switch (type) {
1070 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1071 m = ceph_msg_get(monc->m_subscribe_ack);
1072 break;
1073 case CEPH_MSG_STATFS_REPLY:
1074 return get_generic_reply(con, hdr, skip);
1075 case CEPH_MSG_AUTH_REPLY:
1076 m = ceph_msg_get(monc->m_auth_reply);
1077 break;
1078 case CEPH_MSG_MON_GET_VERSION_REPLY:
1079 if (le64_to_cpu(hdr->tid) != 0)
1080 return get_generic_reply(con, hdr, skip);
1081
1082 /*
1083 * Older OSDs don't set reply tid even if the orignal
1084 * request had a non-zero tid. Workaround this weirdness
1085 * by falling through to the allocate case.
1086 */
1087 case CEPH_MSG_MON_MAP:
1088 case CEPH_MSG_MDS_MAP:
1089 case CEPH_MSG_OSD_MAP:
1090 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1091 if (!m)
1092 return NULL; /* ENOMEM--return skip == 0 */
1093 break;
1094 }
1095
1096 if (!m) {
1097 pr_info("alloc_msg unknown type %d\n", type);
1098 *skip = 1;
1099 } else if (front_len > m->front_alloc_len) {
1100 pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1101 front_len, m->front_alloc_len,
1102 (unsigned int)con->peer_name.type,
1103 le64_to_cpu(con->peer_name.num));
1104 ceph_msg_put(m);
1105 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1106 }
1107
1108 return m;
1109 }
1110
1111 /*
1112 * If the monitor connection resets, pick a new monitor and resubmit
1113 * any pending requests.
1114 */
1115 static void mon_fault(struct ceph_connection *con)
1116 {
1117 struct ceph_mon_client *monc = con->private;
1118
1119 mutex_lock(&monc->mutex);
1120 dout("%s mon%d\n", __func__, monc->cur_mon);
1121 if (monc->cur_mon >= 0) {
1122 if (!monc->hunting) {
1123 dout("%s hunting for new mon\n", __func__);
1124 reopen_session(monc);
1125 __schedule_delayed(monc);
1126 } else {
1127 dout("%s already hunting\n", __func__);
1128 }
1129 }
1130 mutex_unlock(&monc->mutex);
1131 }
1132
1133 /*
1134 * We can ignore refcounting on the connection struct, as all references
1135 * will come from the messenger workqueue, which is drained prior to
1136 * mon_client destruction.
1137 */
1138 static struct ceph_connection *con_get(struct ceph_connection *con)
1139 {
1140 return con;
1141 }
1142
1143 static void con_put(struct ceph_connection *con)
1144 {
1145 }
1146
1147 static const struct ceph_connection_operations mon_con_ops = {
1148 .get = con_get,
1149 .put = con_put,
1150 .dispatch = dispatch,
1151 .fault = mon_fault,
1152 .alloc_msg = mon_alloc_msg,
1153 };
Linux kernel - Deliverables
This page took 0.087061 seconds and 5 git commands to generate.