projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
libceph: decouple hunting and subs management
[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 monc->cur_mon = -1;
126 monc->pending_auth = 0;
127 ceph_auth_reset(monc->auth);
128 }
129
130 /*
131 * Open a session with a (new) monitor.
132 */
133 static int __open_session(struct ceph_mon_client *monc)
134 {
135 char r;
136 int ret;
137
138 if (monc->cur_mon < 0) {
139 get_random_bytes(&r, 1);
140 monc->cur_mon = r % monc->monmap->num_mon;
141 dout("open_session num=%d r=%d -> mon%d\n",
142 monc->monmap->num_mon, r, monc->cur_mon);
143 monc->sub_sent = 0;
144 monc->sub_renew_after = jiffies; /* i.e., expired */
145 monc->want_next_osdmap = !!monc->want_next_osdmap;
146
147 dout("open_session mon%d opening\n", monc->cur_mon);
148 ceph_con_open(&monc->con,
149 CEPH_ENTITY_TYPE_MON, monc->cur_mon,
150 &monc->monmap->mon_inst[monc->cur_mon].addr);
151
152 /* send an initial keepalive to ensure our timestamp is
153 * valid by the time we are in an OPENED state */
154 ceph_con_keepalive(&monc->con);
155
156 /* initiatiate authentication handshake */
157 ret = ceph_auth_build_hello(monc->auth,
158 monc->m_auth->front.iov_base,
159 monc->m_auth->front_alloc_len);
160 __send_prepared_auth_request(monc, ret);
161 } else {
162 dout("open_session mon%d already open\n", monc->cur_mon);
163 }
164 return 0;
165 }
166
167 static bool __sub_expired(struct ceph_mon_client *monc)
168 {
169 return time_after_eq(jiffies, monc->sub_renew_after);
170 }
171
172 /*
173 * Reschedule delayed work timer.
174 */
175 static void __schedule_delayed(struct ceph_mon_client *monc)
176 {
177 struct ceph_options *opt = monc->client->options;
178 unsigned long delay;
179
180 if (monc->cur_mon < 0 || __sub_expired(monc)) {
181 delay = 10 * HZ;
182 } else {
183 delay = 20 * HZ;
184 if (opt->monc_ping_timeout > 0)
185 delay = min(delay, opt->monc_ping_timeout / 3);
186 }
187 dout("__schedule_delayed after %lu\n", delay);
188 schedule_delayed_work(&monc->delayed_work,
189 round_jiffies_relative(delay));
190 }
191
192 /*
193 * Send subscribe request for mdsmap and/or osdmap.
194 */
195 static void __send_subscribe(struct ceph_mon_client *monc)
196 {
197 dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n",
198 (unsigned int)monc->sub_sent, __sub_expired(monc),
199 monc->want_next_osdmap);
200 if ((__sub_expired(monc) && !monc->sub_sent) ||
201 monc->want_next_osdmap == 1) {
202 struct ceph_msg *msg = monc->m_subscribe;
203 struct ceph_mon_subscribe_item *i;
204 void *p, *end;
205 int num;
206
207 p = msg->front.iov_base;
208 end = p + msg->front_alloc_len;
209
210 num = 1 + !!monc->want_next_osdmap + !!monc->want_mdsmap;
211 ceph_encode_32(&p, num);
212
213 if (monc->want_next_osdmap) {
214 dout("__send_subscribe to 'osdmap' %u\n",
215 (unsigned int)monc->have_osdmap);
216 ceph_encode_string(&p, end, "osdmap", 6);
217 i = p;
218 i->have = cpu_to_le64(monc->have_osdmap);
219 i->onetime = 1;
220 p += sizeof(*i);
221 monc->want_next_osdmap = 2; /* requested */
222 }
223 if (monc->want_mdsmap) {
224 dout("__send_subscribe to 'mdsmap' %u+\n",
225 (unsigned int)monc->have_mdsmap);
226 ceph_encode_string(&p, end, "mdsmap", 6);
227 i = p;
228 i->have = cpu_to_le64(monc->have_mdsmap);
229 i->onetime = 0;
230 p += sizeof(*i);
231 }
232 ceph_encode_string(&p, end, "monmap", 6);
233 i = p;
234 i->have = 0;
235 i->onetime = 0;
236 p += sizeof(*i);
237
238 msg->front.iov_len = p - msg->front.iov_base;
239 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
240 ceph_msg_revoke(msg);
241 ceph_con_send(&monc->con, ceph_msg_get(msg));
242
243 monc->sub_sent = jiffies | 1; /* never 0 */
244 }
245 }
246
247 static void handle_subscribe_ack(struct ceph_mon_client *monc,
248 struct ceph_msg *msg)
249 {
250 unsigned int seconds;
251 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
252
253 if (msg->front.iov_len < sizeof(*h))
254 goto bad;
255 seconds = le32_to_cpu(h->duration);
256
257 mutex_lock(&monc->mutex);
258 dout("handle_subscribe_ack after %d seconds\n", seconds);
259 monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1;
260 monc->sub_sent = 0;
261 mutex_unlock(&monc->mutex);
262 return;
263 bad:
264 pr_err("got corrupt subscribe-ack msg\n");
265 ceph_msg_dump(msg);
266 }
267
268 /*
269 * Keep track of which maps we have
270 */
271 int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got)
272 {
273 mutex_lock(&monc->mutex);
274 monc->have_mdsmap = got;
275 mutex_unlock(&monc->mutex);
276 return 0;
277 }
278 EXPORT_SYMBOL(ceph_monc_got_mdsmap);
279
280 int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got)
281 {
282 mutex_lock(&monc->mutex);
283 monc->have_osdmap = got;
284 monc->want_next_osdmap = 0;
285 mutex_unlock(&monc->mutex);
286 return 0;
287 }
288
289 /*
290 * Register interest in the next osdmap
291 */
292 void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
293 {
294 dout("request_next_osdmap have %u\n", monc->have_osdmap);
295 mutex_lock(&monc->mutex);
296 if (!monc->want_next_osdmap)
297 monc->want_next_osdmap = 1;
298 if (monc->want_next_osdmap < 2)
299 __send_subscribe(monc);
300 mutex_unlock(&monc->mutex);
301 }
302 EXPORT_SYMBOL(ceph_monc_request_next_osdmap);
303
304 /*
305 * Wait for an osdmap with a given epoch.
306 *
307 * @epoch: epoch to wait for
308 * @timeout: in jiffies, 0 means "wait forever"
309 */
310 int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
311 unsigned long timeout)
312 {
313 unsigned long started = jiffies;
314 long ret;
315
316 mutex_lock(&monc->mutex);
317 while (monc->have_osdmap < epoch) {
318 mutex_unlock(&monc->mutex);
319
320 if (timeout && time_after_eq(jiffies, started + timeout))
321 return -ETIMEDOUT;
322
323 ret = wait_event_interruptible_timeout(monc->client->auth_wq,
324 monc->have_osdmap >= epoch,
325 ceph_timeout_jiffies(timeout));
326 if (ret < 0)
327 return ret;
328
329 mutex_lock(&monc->mutex);
330 }
331
332 mutex_unlock(&monc->mutex);
333 return 0;
334 }
335 EXPORT_SYMBOL(ceph_monc_wait_osdmap);
336
337 /*
338 *
339 */
340 int ceph_monc_open_session(struct ceph_mon_client *monc)
341 {
342 mutex_lock(&monc->mutex);
343 __open_session(monc);
344 __schedule_delayed(monc);
345 mutex_unlock(&monc->mutex);
346 return 0;
347 }
348 EXPORT_SYMBOL(ceph_monc_open_session);
349
350 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
351 struct ceph_msg *msg)
352 {
353 struct ceph_client *client = monc->client;
354 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
355 void *p, *end;
356
357 mutex_lock(&monc->mutex);
358
359 dout("handle_monmap\n");
360 p = msg->front.iov_base;
361 end = p + msg->front.iov_len;
362
363 monmap = ceph_monmap_decode(p, end);
364 if (IS_ERR(monmap)) {
365 pr_err("problem decoding monmap, %d\n",
366 (int)PTR_ERR(monmap));
367 goto out;
368 }
369
370 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
371 kfree(monmap);
372 goto out;
373 }
374
375 client->monc.monmap = monmap;
376 kfree(old);
377
378 client->have_fsid = true;
379
380 out:
381 mutex_unlock(&monc->mutex);
382 wake_up_all(&client->auth_wq);
383 }
384
385 /*
386 * generic requests (currently statfs, mon_get_version)
387 */
388 static struct ceph_mon_generic_request *__lookup_generic_req(
389 struct ceph_mon_client *monc, u64 tid)
390 {
391 struct ceph_mon_generic_request *req;
392 struct rb_node *n = monc->generic_request_tree.rb_node;
393
394 while (n) {
395 req = rb_entry(n, struct ceph_mon_generic_request, node);
396 if (tid < req->tid)
397 n = n->rb_left;
398 else if (tid > req->tid)
399 n = n->rb_right;
400 else
401 return req;
402 }
403 return NULL;
404 }
405
406 static void __insert_generic_request(struct ceph_mon_client *monc,
407 struct ceph_mon_generic_request *new)
408 {
409 struct rb_node **p = &monc->generic_request_tree.rb_node;
410 struct rb_node *parent = NULL;
411 struct ceph_mon_generic_request *req = NULL;
412
413 while (*p) {
414 parent = *p;
415 req = rb_entry(parent, struct ceph_mon_generic_request, node);
416 if (new->tid < req->tid)
417 p = &(*p)->rb_left;
418 else if (new->tid > req->tid)
419 p = &(*p)->rb_right;
420 else
421 BUG();
422 }
423
424 rb_link_node(&new->node, parent, p);
425 rb_insert_color(&new->node, &monc->generic_request_tree);
426 }
427
428 static void release_generic_request(struct kref *kref)
429 {
430 struct ceph_mon_generic_request *req =
431 container_of(kref, struct ceph_mon_generic_request, kref);
432
433 if (req->reply)
434 ceph_msg_put(req->reply);
435 if (req->request)
436 ceph_msg_put(req->request);
437
438 kfree(req);
439 }
440
441 static void put_generic_request(struct ceph_mon_generic_request *req)
442 {
443 kref_put(&req->kref, release_generic_request);
444 }
445
446 static void get_generic_request(struct ceph_mon_generic_request *req)
447 {
448 kref_get(&req->kref);
449 }
450
451 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
452 struct ceph_msg_header *hdr,
453 int *skip)
454 {
455 struct ceph_mon_client *monc = con->private;
456 struct ceph_mon_generic_request *req;
457 u64 tid = le64_to_cpu(hdr->tid);
458 struct ceph_msg *m;
459
460 mutex_lock(&monc->mutex);
461 req = __lookup_generic_req(monc, tid);
462 if (!req) {
463 dout("get_generic_reply %lld dne\n", tid);
464 *skip = 1;
465 m = NULL;
466 } else {
467 dout("get_generic_reply %lld got %p\n", tid, req->reply);
468 *skip = 0;
469 m = ceph_msg_get(req->reply);
470 /*
471 * we don't need to track the connection reading into
472 * this reply because we only have one open connection
473 * at a time, ever.
474 */
475 }
476 mutex_unlock(&monc->mutex);
477 return m;
478 }
479
480 static int __do_generic_request(struct ceph_mon_client *monc, u64 tid,
481 struct ceph_mon_generic_request *req)
482 {
483 int err;
484
485 /* register request */
486 req->tid = tid != 0 ? tid : ++monc->last_tid;
487 req->request->hdr.tid = cpu_to_le64(req->tid);
488 __insert_generic_request(monc, req);
489 monc->num_generic_requests++;
490 ceph_con_send(&monc->con, ceph_msg_get(req->request));
491 mutex_unlock(&monc->mutex);
492
493 err = wait_for_completion_interruptible(&req->completion);
494
495 mutex_lock(&monc->mutex);
496 rb_erase(&req->node, &monc->generic_request_tree);
497 monc->num_generic_requests--;
498
499 if (!err)
500 err = req->result;
501 return err;
502 }
503
504 static int do_generic_request(struct ceph_mon_client *monc,
505 struct ceph_mon_generic_request *req)
506 {
507 int err;
508
509 mutex_lock(&monc->mutex);
510 err = __do_generic_request(monc, 0, req);
511 mutex_unlock(&monc->mutex);
512
513 return err;
514 }
515
516 /*
517 * statfs
518 */
519 static void handle_statfs_reply(struct ceph_mon_client *monc,
520 struct ceph_msg *msg)
521 {
522 struct ceph_mon_generic_request *req;
523 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
524 u64 tid = le64_to_cpu(msg->hdr.tid);
525
526 if (msg->front.iov_len != sizeof(*reply))
527 goto bad;
528 dout("handle_statfs_reply %p tid %llu\n", msg, tid);
529
530 mutex_lock(&monc->mutex);
531 req = __lookup_generic_req(monc, tid);
532 if (req) {
533 *(struct ceph_statfs *)req->buf = reply->st;
534 req->result = 0;
535 get_generic_request(req);
536 }
537 mutex_unlock(&monc->mutex);
538 if (req) {
539 complete_all(&req->completion);
540 put_generic_request(req);
541 }
542 return;
543
544 bad:
545 pr_err("corrupt statfs reply, tid %llu\n", tid);
546 ceph_msg_dump(msg);
547 }
548
549 /*
550 * Do a synchronous statfs().
551 */
552 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
553 {
554 struct ceph_mon_generic_request *req;
555 struct ceph_mon_statfs *h;
556 int err;
557
558 req = kzalloc(sizeof(*req), GFP_NOFS);
559 if (!req)
560 return -ENOMEM;
561
562 kref_init(&req->kref);
563 req->buf = buf;
564 init_completion(&req->completion);
565
566 err = -ENOMEM;
567 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
568 true);
569 if (!req->request)
570 goto out;
571 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
572 true);
573 if (!req->reply)
574 goto out;
575
576 /* fill out request */
577 h = req->request->front.iov_base;
578 h->monhdr.have_version = 0;
579 h->monhdr.session_mon = cpu_to_le16(-1);
580 h->monhdr.session_mon_tid = 0;
581 h->fsid = monc->monmap->fsid;
582
583 err = do_generic_request(monc, req);
584
585 out:
586 put_generic_request(req);
587 return err;
588 }
589 EXPORT_SYMBOL(ceph_monc_do_statfs);
590
591 static void handle_get_version_reply(struct ceph_mon_client *monc,
592 struct ceph_msg *msg)
593 {
594 struct ceph_mon_generic_request *req;
595 u64 tid = le64_to_cpu(msg->hdr.tid);
596 void *p = msg->front.iov_base;
597 void *end = p + msg->front_alloc_len;
598 u64 handle;
599
600 dout("%s %p tid %llu\n", __func__, msg, tid);
601
602 ceph_decode_need(&p, end, 2*sizeof(u64), bad);
603 handle = ceph_decode_64(&p);
604 if (tid != 0 && tid != handle)
605 goto bad;
606
607 mutex_lock(&monc->mutex);
608 req = __lookup_generic_req(monc, handle);
609 if (req) {
610 *(u64 *)req->buf = ceph_decode_64(&p);
611 req->result = 0;
612 get_generic_request(req);
613 }
614 mutex_unlock(&monc->mutex);
615 if (req) {
616 complete_all(&req->completion);
617 put_generic_request(req);
618 }
619
620 return;
621 bad:
622 pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
623 ceph_msg_dump(msg);
624 }
625
626 /*
627 * Send MMonGetVersion and wait for the reply.
628 *
629 * @what: one of "mdsmap", "osdmap" or "monmap"
630 */
631 int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what,
632 u64 *newest)
633 {
634 struct ceph_mon_generic_request *req;
635 void *p, *end;
636 u64 tid;
637 int err;
638
639 req = kzalloc(sizeof(*req), GFP_NOFS);
640 if (!req)
641 return -ENOMEM;
642
643 kref_init(&req->kref);
644 req->buf = newest;
645 init_completion(&req->completion);
646
647 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
648 sizeof(u64) + sizeof(u32) + strlen(what),
649 GFP_NOFS, true);
650 if (!req->request) {
651 err = -ENOMEM;
652 goto out;
653 }
654
655 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024,
656 GFP_NOFS, true);
657 if (!req->reply) {
658 err = -ENOMEM;
659 goto out;
660 }
661
662 p = req->request->front.iov_base;
663 end = p + req->request->front_alloc_len;
664
665 /* fill out request */
666 mutex_lock(&monc->mutex);
667 tid = ++monc->last_tid;
668 ceph_encode_64(&p, tid); /* handle */
669 ceph_encode_string(&p, end, what, strlen(what));
670
671 err = __do_generic_request(monc, tid, req);
672
673 mutex_unlock(&monc->mutex);
674 out:
675 put_generic_request(req);
676 return err;
677 }
678 EXPORT_SYMBOL(ceph_monc_do_get_version);
679
680 /*
681 * Resend pending generic requests.
682 */
683 static void __resend_generic_request(struct ceph_mon_client *monc)
684 {
685 struct ceph_mon_generic_request *req;
686 struct rb_node *p;
687
688 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
689 req = rb_entry(p, struct ceph_mon_generic_request, node);
690 ceph_msg_revoke(req->request);
691 ceph_msg_revoke_incoming(req->reply);
692 ceph_con_send(&monc->con, ceph_msg_get(req->request));
693 }
694 }
695
696 /*
697 * Delayed work. If we haven't mounted yet, retry. Otherwise,
698 * renew/retry subscription as needed (in case it is timing out, or we
699 * got an ENOMEM). And keep the monitor connection alive.
700 */
701 static void delayed_work(struct work_struct *work)
702 {
703 struct ceph_mon_client *monc =
704 container_of(work, struct ceph_mon_client, delayed_work.work);
705
706 dout("monc delayed_work\n");
707 mutex_lock(&monc->mutex);
708 if (monc->hunting) {
709 __close_session(monc);
710 __open_session(monc); /* continue hunting */
711 } else {
712 struct ceph_options *opt = monc->client->options;
713 int is_auth = ceph_auth_is_authenticated(monc->auth);
714 if (ceph_con_keepalive_expired(&monc->con,
715 opt->monc_ping_timeout)) {
716 dout("monc keepalive timeout\n");
717 is_auth = 0;
718 __close_session(monc);
719 monc->hunting = true;
720 __open_session(monc);
721 }
722
723 if (!monc->hunting) {
724 ceph_con_keepalive(&monc->con);
725 __validate_auth(monc);
726 }
727
728 if (is_auth)
729 __send_subscribe(monc);
730 }
731 __schedule_delayed(monc);
732 mutex_unlock(&monc->mutex);
733 }
734
735 /*
736 * On startup, we build a temporary monmap populated with the IPs
737 * provided by mount(2).
738 */
739 static int build_initial_monmap(struct ceph_mon_client *monc)
740 {
741 struct ceph_options *opt = monc->client->options;
742 struct ceph_entity_addr *mon_addr = opt->mon_addr;
743 int num_mon = opt->num_mon;
744 int i;
745
746 /* build initial monmap */
747 monc->monmap = kzalloc(sizeof(*monc->monmap) +
748 num_mon*sizeof(monc->monmap->mon_inst[0]),
749 GFP_KERNEL);
750 if (!monc->monmap)
751 return -ENOMEM;
752 for (i = 0; i < num_mon; i++) {
753 monc->monmap->mon_inst[i].addr = mon_addr[i];
754 monc->monmap->mon_inst[i].addr.nonce = 0;
755 monc->monmap->mon_inst[i].name.type =
756 CEPH_ENTITY_TYPE_MON;
757 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
758 }
759 monc->monmap->num_mon = num_mon;
760 return 0;
761 }
762
763 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
764 {
765 int err = 0;
766
767 dout("init\n");
768 memset(monc, 0, sizeof(*monc));
769 monc->client = cl;
770 monc->monmap = NULL;
771 mutex_init(&monc->mutex);
772
773 err = build_initial_monmap(monc);
774 if (err)
775 goto out;
776
777 /* connection */
778 /* authentication */
779 monc->auth = ceph_auth_init(cl->options->name,
780 cl->options->key);
781 if (IS_ERR(monc->auth)) {
782 err = PTR_ERR(monc->auth);
783 goto out_monmap;
784 }
785 monc->auth->want_keys =
786 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
787 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
788
789 /* msgs */
790 err = -ENOMEM;
791 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
792 sizeof(struct ceph_mon_subscribe_ack),
793 GFP_NOFS, true);
794 if (!monc->m_subscribe_ack)
795 goto out_auth;
796
797 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
798 true);
799 if (!monc->m_subscribe)
800 goto out_subscribe_ack;
801
802 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
803 true);
804 if (!monc->m_auth_reply)
805 goto out_subscribe;
806
807 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
808 monc->pending_auth = 0;
809 if (!monc->m_auth)
810 goto out_auth_reply;
811
812 ceph_con_init(&monc->con, monc, &mon_con_ops,
813 &monc->client->msgr);
814
815 monc->cur_mon = -1;
816 monc->hunting = true;
817 monc->sub_renew_after = jiffies;
818 monc->sub_sent = 0;
819
820 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
821 monc->generic_request_tree = RB_ROOT;
822 monc->num_generic_requests = 0;
823 monc->last_tid = 0;
824
825 monc->have_mdsmap = 0;
826 monc->have_osdmap = 0;
827 monc->want_next_osdmap = 1;
828 return 0;
829
830 out_auth_reply:
831 ceph_msg_put(monc->m_auth_reply);
832 out_subscribe:
833 ceph_msg_put(monc->m_subscribe);
834 out_subscribe_ack:
835 ceph_msg_put(monc->m_subscribe_ack);
836 out_auth:
837 ceph_auth_destroy(monc->auth);
838 out_monmap:
839 kfree(monc->monmap);
840 out:
841 return err;
842 }
843 EXPORT_SYMBOL(ceph_monc_init);
844
845 void ceph_monc_stop(struct ceph_mon_client *monc)
846 {
847 dout("stop\n");
848 cancel_delayed_work_sync(&monc->delayed_work);
849
850 mutex_lock(&monc->mutex);
851 __close_session(monc);
852
853 mutex_unlock(&monc->mutex);
854
855 /*
856 * flush msgr queue before we destroy ourselves to ensure that:
857 * - any work that references our embedded con is finished.
858 * - any osd_client or other work that may reference an authorizer
859 * finishes before we shut down the auth subsystem.
860 */
861 ceph_msgr_flush();
862
863 ceph_auth_destroy(monc->auth);
864
865 ceph_msg_put(monc->m_auth);
866 ceph_msg_put(monc->m_auth_reply);
867 ceph_msg_put(monc->m_subscribe);
868 ceph_msg_put(monc->m_subscribe_ack);
869
870 kfree(monc->monmap);
871 }
872 EXPORT_SYMBOL(ceph_monc_stop);
873
874 static void finish_hunting(struct ceph_mon_client *monc)
875 {
876 if (monc->hunting) {
877 dout("%s found mon%d\n", __func__, monc->cur_mon);
878 monc->hunting = false;
879 }
880 }
881
882 static void handle_auth_reply(struct ceph_mon_client *monc,
883 struct ceph_msg *msg)
884 {
885 int ret;
886 int was_auth = 0;
887
888 mutex_lock(&monc->mutex);
889 was_auth = ceph_auth_is_authenticated(monc->auth);
890 monc->pending_auth = 0;
891 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
892 msg->front.iov_len,
893 monc->m_auth->front.iov_base,
894 monc->m_auth->front_alloc_len);
895 if (ret > 0) {
896 __send_prepared_auth_request(monc, ret);
897 goto out;
898 }
899
900 finish_hunting(monc);
901
902 if (ret < 0) {
903 monc->client->auth_err = ret;
904 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
905 dout("authenticated, starting session\n");
906
907 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
908 monc->client->msgr.inst.name.num =
909 cpu_to_le64(monc->auth->global_id);
910
911 __send_subscribe(monc);
912 __resend_generic_request(monc);
913
914 pr_info("mon%d %s session established\n", monc->cur_mon,
915 ceph_pr_addr(&monc->con.peer_addr.in_addr));
916 }
917
918 out:
919 mutex_unlock(&monc->mutex);
920 if (monc->client->auth_err < 0)
921 wake_up_all(&monc->client->auth_wq);
922 }
923
924 static int __validate_auth(struct ceph_mon_client *monc)
925 {
926 int ret;
927
928 if (monc->pending_auth)
929 return 0;
930
931 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
932 monc->m_auth->front_alloc_len);
933 if (ret <= 0)
934 return ret; /* either an error, or no need to authenticate */
935 __send_prepared_auth_request(monc, ret);
936 return 0;
937 }
938
939 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
940 {
941 int ret;
942
943 mutex_lock(&monc->mutex);
944 ret = __validate_auth(monc);
945 mutex_unlock(&monc->mutex);
946 return ret;
947 }
948 EXPORT_SYMBOL(ceph_monc_validate_auth);
949
950 /*
951 * handle incoming message
952 */
953 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
954 {
955 struct ceph_mon_client *monc = con->private;
956 int type = le16_to_cpu(msg->hdr.type);
957
958 if (!monc)
959 return;
960
961 switch (type) {
962 case CEPH_MSG_AUTH_REPLY:
963 handle_auth_reply(monc, msg);
964 break;
965
966 case CEPH_MSG_MON_SUBSCRIBE_ACK:
967 handle_subscribe_ack(monc, msg);
968 break;
969
970 case CEPH_MSG_STATFS_REPLY:
971 handle_statfs_reply(monc, msg);
972 break;
973
974 case CEPH_MSG_MON_GET_VERSION_REPLY:
975 handle_get_version_reply(monc, msg);
976 break;
977
978 case CEPH_MSG_MON_MAP:
979 ceph_monc_handle_map(monc, msg);
980 break;
981
982 case CEPH_MSG_OSD_MAP:
983 ceph_osdc_handle_map(&monc->client->osdc, msg);
984 break;
985
986 default:
987 /* can the chained handler handle it? */
988 if (monc->client->extra_mon_dispatch &&
989 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
990 break;
991
992 pr_err("received unknown message type %d %s\n", type,
993 ceph_msg_type_name(type));
994 }
995 ceph_msg_put(msg);
996 }
997
998 /*
999 * Allocate memory for incoming message
1000 */
1001 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1002 struct ceph_msg_header *hdr,
1003 int *skip)
1004 {
1005 struct ceph_mon_client *monc = con->private;
1006 int type = le16_to_cpu(hdr->type);
1007 int front_len = le32_to_cpu(hdr->front_len);
1008 struct ceph_msg *m = NULL;
1009
1010 *skip = 0;
1011
1012 switch (type) {
1013 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1014 m = ceph_msg_get(monc->m_subscribe_ack);
1015 break;
1016 case CEPH_MSG_STATFS_REPLY:
1017 return get_generic_reply(con, hdr, skip);
1018 case CEPH_MSG_AUTH_REPLY:
1019 m = ceph_msg_get(monc->m_auth_reply);
1020 break;
1021 case CEPH_MSG_MON_GET_VERSION_REPLY:
1022 if (le64_to_cpu(hdr->tid) != 0)
1023 return get_generic_reply(con, hdr, skip);
1024
1025 /*
1026 * Older OSDs don't set reply tid even if the orignal
1027 * request had a non-zero tid. Workaround this weirdness
1028 * by falling through to the allocate case.
1029 */
1030 case CEPH_MSG_MON_MAP:
1031 case CEPH_MSG_MDS_MAP:
1032 case CEPH_MSG_OSD_MAP:
1033 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1034 if (!m)
1035 return NULL; /* ENOMEM--return skip == 0 */
1036 break;
1037 }
1038
1039 if (!m) {
1040 pr_info("alloc_msg unknown type %d\n", type);
1041 *skip = 1;
1042 } else if (front_len > m->front_alloc_len) {
1043 pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1044 front_len, m->front_alloc_len,
1045 (unsigned int)con->peer_name.type,
1046 le64_to_cpu(con->peer_name.num));
1047 ceph_msg_put(m);
1048 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1049 }
1050
1051 return m;
1052 }
1053
1054 /*
1055 * If the monitor connection resets, pick a new monitor and resubmit
1056 * any pending requests.
1057 */
1058 static void mon_fault(struct ceph_connection *con)
1059 {
1060 struct ceph_mon_client *monc = con->private;
1061
1062 if (!monc)
1063 return;
1064
1065 dout("mon_fault\n");
1066 mutex_lock(&monc->mutex);
1067 if (!con->private)
1068 goto out;
1069
1070 if (!monc->hunting)
1071 pr_info("mon%d %s session lost, "
1072 "hunting for new mon\n", monc->cur_mon,
1073 ceph_pr_addr(&monc->con.peer_addr.in_addr));
1074
1075 __close_session(monc);
1076 if (!monc->hunting) {
1077 /* start hunting */
1078 monc->hunting = true;
1079 __open_session(monc);
1080 } else {
1081 /* already hunting, let's wait a bit */
1082 __schedule_delayed(monc);
1083 }
1084 out:
1085 mutex_unlock(&monc->mutex);
1086 }
1087
1088 /*
1089 * We can ignore refcounting on the connection struct, as all references
1090 * will come from the messenger workqueue, which is drained prior to
1091 * mon_client destruction.
1092 */
1093 static struct ceph_connection *con_get(struct ceph_connection *con)
1094 {
1095 return con;
1096 }
1097
1098 static void con_put(struct ceph_connection *con)
1099 {
1100 }
1101
1102 static const struct ceph_connection_operations mon_con_ops = {
1103 .get = con_get,
1104 .put = con_put,
1105 .dispatch = dispatch,
1106 .fault = mon_fault,
1107 .alloc_msg = mon_alloc_msg,
1108 };
Linux kernel - Deliverables
This page took 0.80832 seconds and 6 git commands to generate.