xref: /openbmc/linux/net/ceph/mon_client.c (revision a8fe58ce)
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 	if (monc->hunting) {
259 		pr_info("mon%d %s session established\n",
260 			monc->cur_mon,
261 			ceph_pr_addr(&monc->con.peer_addr.in_addr));
262 		monc->hunting = false;
263 	}
264 	dout("handle_subscribe_ack after %d seconds\n", seconds);
265 	monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1;
266 	monc->sub_sent = 0;
267 	mutex_unlock(&monc->mutex);
268 	return;
269 bad:
270 	pr_err("got corrupt subscribe-ack msg\n");
271 	ceph_msg_dump(msg);
272 }
273 
274 /*
275  * Keep track of which maps we have
276  */
277 int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got)
278 {
279 	mutex_lock(&monc->mutex);
280 	monc->have_mdsmap = got;
281 	mutex_unlock(&monc->mutex);
282 	return 0;
283 }
284 EXPORT_SYMBOL(ceph_monc_got_mdsmap);
285 
286 int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got)
287 {
288 	mutex_lock(&monc->mutex);
289 	monc->have_osdmap = got;
290 	monc->want_next_osdmap = 0;
291 	mutex_unlock(&monc->mutex);
292 	return 0;
293 }
294 
295 /*
296  * Register interest in the next osdmap
297  */
298 void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
299 {
300 	dout("request_next_osdmap have %u\n", monc->have_osdmap);
301 	mutex_lock(&monc->mutex);
302 	if (!monc->want_next_osdmap)
303 		monc->want_next_osdmap = 1;
304 	if (monc->want_next_osdmap < 2)
305 		__send_subscribe(monc);
306 	mutex_unlock(&monc->mutex);
307 }
308 EXPORT_SYMBOL(ceph_monc_request_next_osdmap);
309 
310 /*
311  * Wait for an osdmap with a given epoch.
312  *
313  * @epoch: epoch to wait for
314  * @timeout: in jiffies, 0 means "wait forever"
315  */
316 int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
317 			  unsigned long timeout)
318 {
319 	unsigned long started = jiffies;
320 	long ret;
321 
322 	mutex_lock(&monc->mutex);
323 	while (monc->have_osdmap < epoch) {
324 		mutex_unlock(&monc->mutex);
325 
326 		if (timeout && time_after_eq(jiffies, started + timeout))
327 			return -ETIMEDOUT;
328 
329 		ret = wait_event_interruptible_timeout(monc->client->auth_wq,
330 						monc->have_osdmap >= epoch,
331 						ceph_timeout_jiffies(timeout));
332 		if (ret < 0)
333 			return ret;
334 
335 		mutex_lock(&monc->mutex);
336 	}
337 
338 	mutex_unlock(&monc->mutex);
339 	return 0;
340 }
341 EXPORT_SYMBOL(ceph_monc_wait_osdmap);
342 
343 /*
344  *
345  */
346 int ceph_monc_open_session(struct ceph_mon_client *monc)
347 {
348 	mutex_lock(&monc->mutex);
349 	__open_session(monc);
350 	__schedule_delayed(monc);
351 	mutex_unlock(&monc->mutex);
352 	return 0;
353 }
354 EXPORT_SYMBOL(ceph_monc_open_session);
355 
356 /*
357  * We require the fsid and global_id in order to initialize our
358  * debugfs dir.
359  */
360 static bool have_debugfs_info(struct ceph_mon_client *monc)
361 {
362 	dout("have_debugfs_info fsid %d globalid %lld\n",
363 	     (int)monc->client->have_fsid, monc->auth->global_id);
364 	return monc->client->have_fsid && monc->auth->global_id > 0;
365 }
366 
367 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
368 				 struct ceph_msg *msg)
369 {
370 	struct ceph_client *client = monc->client;
371 	struct ceph_monmap *monmap = NULL, *old = monc->monmap;
372 	void *p, *end;
373 	int had_debugfs_info, init_debugfs = 0;
374 
375 	mutex_lock(&monc->mutex);
376 
377 	had_debugfs_info = have_debugfs_info(monc);
378 
379 	dout("handle_monmap\n");
380 	p = msg->front.iov_base;
381 	end = p + msg->front.iov_len;
382 
383 	monmap = ceph_monmap_decode(p, end);
384 	if (IS_ERR(monmap)) {
385 		pr_err("problem decoding monmap, %d\n",
386 		       (int)PTR_ERR(monmap));
387 		goto out;
388 	}
389 
390 	if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
391 		kfree(monmap);
392 		goto out;
393 	}
394 
395 	client->monc.monmap = monmap;
396 	kfree(old);
397 
398 	if (!client->have_fsid) {
399 		client->have_fsid = true;
400 		if (!had_debugfs_info && have_debugfs_info(monc)) {
401 			pr_info("client%lld fsid %pU\n",
402 				ceph_client_id(monc->client),
403 				&monc->client->fsid);
404 			init_debugfs = 1;
405 		}
406 		mutex_unlock(&monc->mutex);
407 
408 		if (init_debugfs) {
409 			/*
410 			 * do debugfs initialization without mutex to avoid
411 			 * creating a locking dependency
412 			 */
413 			ceph_debugfs_client_init(monc->client);
414 		}
415 
416 		goto out_unlocked;
417 	}
418 out:
419 	mutex_unlock(&monc->mutex);
420 out_unlocked:
421 	wake_up_all(&client->auth_wq);
422 }
423 
424 /*
425  * generic requests (currently statfs, mon_get_version)
426  */
427 static struct ceph_mon_generic_request *__lookup_generic_req(
428 	struct ceph_mon_client *monc, u64 tid)
429 {
430 	struct ceph_mon_generic_request *req;
431 	struct rb_node *n = monc->generic_request_tree.rb_node;
432 
433 	while (n) {
434 		req = rb_entry(n, struct ceph_mon_generic_request, node);
435 		if (tid < req->tid)
436 			n = n->rb_left;
437 		else if (tid > req->tid)
438 			n = n->rb_right;
439 		else
440 			return req;
441 	}
442 	return NULL;
443 }
444 
445 static void __insert_generic_request(struct ceph_mon_client *monc,
446 			    struct ceph_mon_generic_request *new)
447 {
448 	struct rb_node **p = &monc->generic_request_tree.rb_node;
449 	struct rb_node *parent = NULL;
450 	struct ceph_mon_generic_request *req = NULL;
451 
452 	while (*p) {
453 		parent = *p;
454 		req = rb_entry(parent, struct ceph_mon_generic_request, node);
455 		if (new->tid < req->tid)
456 			p = &(*p)->rb_left;
457 		else if (new->tid > req->tid)
458 			p = &(*p)->rb_right;
459 		else
460 			BUG();
461 	}
462 
463 	rb_link_node(&new->node, parent, p);
464 	rb_insert_color(&new->node, &monc->generic_request_tree);
465 }
466 
467 static void release_generic_request(struct kref *kref)
468 {
469 	struct ceph_mon_generic_request *req =
470 		container_of(kref, struct ceph_mon_generic_request, kref);
471 
472 	if (req->reply)
473 		ceph_msg_put(req->reply);
474 	if (req->request)
475 		ceph_msg_put(req->request);
476 
477 	kfree(req);
478 }
479 
480 static void put_generic_request(struct ceph_mon_generic_request *req)
481 {
482 	kref_put(&req->kref, release_generic_request);
483 }
484 
485 static void get_generic_request(struct ceph_mon_generic_request *req)
486 {
487 	kref_get(&req->kref);
488 }
489 
490 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
491 					 struct ceph_msg_header *hdr,
492 					 int *skip)
493 {
494 	struct ceph_mon_client *monc = con->private;
495 	struct ceph_mon_generic_request *req;
496 	u64 tid = le64_to_cpu(hdr->tid);
497 	struct ceph_msg *m;
498 
499 	mutex_lock(&monc->mutex);
500 	req = __lookup_generic_req(monc, tid);
501 	if (!req) {
502 		dout("get_generic_reply %lld dne\n", tid);
503 		*skip = 1;
504 		m = NULL;
505 	} else {
506 		dout("get_generic_reply %lld got %p\n", tid, req->reply);
507 		*skip = 0;
508 		m = ceph_msg_get(req->reply);
509 		/*
510 		 * we don't need to track the connection reading into
511 		 * this reply because we only have one open connection
512 		 * at a time, ever.
513 		 */
514 	}
515 	mutex_unlock(&monc->mutex);
516 	return m;
517 }
518 
519 static int __do_generic_request(struct ceph_mon_client *monc, u64 tid,
520 				struct ceph_mon_generic_request *req)
521 {
522 	int err;
523 
524 	/* register request */
525 	req->tid = tid != 0 ? tid : ++monc->last_tid;
526 	req->request->hdr.tid = cpu_to_le64(req->tid);
527 	__insert_generic_request(monc, req);
528 	monc->num_generic_requests++;
529 	ceph_con_send(&monc->con, ceph_msg_get(req->request));
530 	mutex_unlock(&monc->mutex);
531 
532 	err = wait_for_completion_interruptible(&req->completion);
533 
534 	mutex_lock(&monc->mutex);
535 	rb_erase(&req->node, &monc->generic_request_tree);
536 	monc->num_generic_requests--;
537 
538 	if (!err)
539 		err = req->result;
540 	return err;
541 }
542 
543 static int do_generic_request(struct ceph_mon_client *monc,
544 			      struct ceph_mon_generic_request *req)
545 {
546 	int err;
547 
548 	mutex_lock(&monc->mutex);
549 	err = __do_generic_request(monc, 0, req);
550 	mutex_unlock(&monc->mutex);
551 
552 	return err;
553 }
554 
555 /*
556  * statfs
557  */
558 static void handle_statfs_reply(struct ceph_mon_client *monc,
559 				struct ceph_msg *msg)
560 {
561 	struct ceph_mon_generic_request *req;
562 	struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
563 	u64 tid = le64_to_cpu(msg->hdr.tid);
564 
565 	if (msg->front.iov_len != sizeof(*reply))
566 		goto bad;
567 	dout("handle_statfs_reply %p tid %llu\n", msg, tid);
568 
569 	mutex_lock(&monc->mutex);
570 	req = __lookup_generic_req(monc, tid);
571 	if (req) {
572 		*(struct ceph_statfs *)req->buf = reply->st;
573 		req->result = 0;
574 		get_generic_request(req);
575 	}
576 	mutex_unlock(&monc->mutex);
577 	if (req) {
578 		complete_all(&req->completion);
579 		put_generic_request(req);
580 	}
581 	return;
582 
583 bad:
584 	pr_err("corrupt statfs reply, tid %llu\n", tid);
585 	ceph_msg_dump(msg);
586 }
587 
588 /*
589  * Do a synchronous statfs().
590  */
591 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
592 {
593 	struct ceph_mon_generic_request *req;
594 	struct ceph_mon_statfs *h;
595 	int err;
596 
597 	req = kzalloc(sizeof(*req), GFP_NOFS);
598 	if (!req)
599 		return -ENOMEM;
600 
601 	kref_init(&req->kref);
602 	req->buf = buf;
603 	init_completion(&req->completion);
604 
605 	err = -ENOMEM;
606 	req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
607 				    true);
608 	if (!req->request)
609 		goto out;
610 	req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
611 				  true);
612 	if (!req->reply)
613 		goto out;
614 
615 	/* fill out request */
616 	h = req->request->front.iov_base;
617 	h->monhdr.have_version = 0;
618 	h->monhdr.session_mon = cpu_to_le16(-1);
619 	h->monhdr.session_mon_tid = 0;
620 	h->fsid = monc->monmap->fsid;
621 
622 	err = do_generic_request(monc, req);
623 
624 out:
625 	put_generic_request(req);
626 	return err;
627 }
628 EXPORT_SYMBOL(ceph_monc_do_statfs);
629 
630 static void handle_get_version_reply(struct ceph_mon_client *monc,
631 				     struct ceph_msg *msg)
632 {
633 	struct ceph_mon_generic_request *req;
634 	u64 tid = le64_to_cpu(msg->hdr.tid);
635 	void *p = msg->front.iov_base;
636 	void *end = p + msg->front_alloc_len;
637 	u64 handle;
638 
639 	dout("%s %p tid %llu\n", __func__, msg, tid);
640 
641 	ceph_decode_need(&p, end, 2*sizeof(u64), bad);
642 	handle = ceph_decode_64(&p);
643 	if (tid != 0 && tid != handle)
644 		goto bad;
645 
646 	mutex_lock(&monc->mutex);
647 	req = __lookup_generic_req(monc, handle);
648 	if (req) {
649 		*(u64 *)req->buf = ceph_decode_64(&p);
650 		req->result = 0;
651 		get_generic_request(req);
652 	}
653 	mutex_unlock(&monc->mutex);
654 	if (req) {
655 		complete_all(&req->completion);
656 		put_generic_request(req);
657 	}
658 
659 	return;
660 bad:
661 	pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
662 	ceph_msg_dump(msg);
663 }
664 
665 /*
666  * Send MMonGetVersion and wait for the reply.
667  *
668  * @what: one of "mdsmap", "osdmap" or "monmap"
669  */
670 int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what,
671 			     u64 *newest)
672 {
673 	struct ceph_mon_generic_request *req;
674 	void *p, *end;
675 	u64 tid;
676 	int err;
677 
678 	req = kzalloc(sizeof(*req), GFP_NOFS);
679 	if (!req)
680 		return -ENOMEM;
681 
682 	kref_init(&req->kref);
683 	req->buf = newest;
684 	init_completion(&req->completion);
685 
686 	req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
687 				    sizeof(u64) + sizeof(u32) + strlen(what),
688 				    GFP_NOFS, true);
689 	if (!req->request) {
690 		err = -ENOMEM;
691 		goto out;
692 	}
693 
694 	req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024,
695 				  GFP_NOFS, true);
696 	if (!req->reply) {
697 		err = -ENOMEM;
698 		goto out;
699 	}
700 
701 	p = req->request->front.iov_base;
702 	end = p + req->request->front_alloc_len;
703 
704 	/* fill out request */
705 	mutex_lock(&monc->mutex);
706 	tid = ++monc->last_tid;
707 	ceph_encode_64(&p, tid); /* handle */
708 	ceph_encode_string(&p, end, what, strlen(what));
709 
710 	err = __do_generic_request(monc, tid, req);
711 
712 	mutex_unlock(&monc->mutex);
713 out:
714 	put_generic_request(req);
715 	return err;
716 }
717 EXPORT_SYMBOL(ceph_monc_do_get_version);
718 
719 /*
720  * Resend pending generic requests.
721  */
722 static void __resend_generic_request(struct ceph_mon_client *monc)
723 {
724 	struct ceph_mon_generic_request *req;
725 	struct rb_node *p;
726 
727 	for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
728 		req = rb_entry(p, struct ceph_mon_generic_request, node);
729 		ceph_msg_revoke(req->request);
730 		ceph_msg_revoke_incoming(req->reply);
731 		ceph_con_send(&monc->con, ceph_msg_get(req->request));
732 	}
733 }
734 
735 /*
736  * Delayed work.  If we haven't mounted yet, retry.  Otherwise,
737  * renew/retry subscription as needed (in case it is timing out, or we
738  * got an ENOMEM).  And keep the monitor connection alive.
739  */
740 static void delayed_work(struct work_struct *work)
741 {
742 	struct ceph_mon_client *monc =
743 		container_of(work, struct ceph_mon_client, delayed_work.work);
744 
745 	dout("monc delayed_work\n");
746 	mutex_lock(&monc->mutex);
747 	if (monc->hunting) {
748 		__close_session(monc);
749 		__open_session(monc);  /* continue hunting */
750 	} else {
751 		struct ceph_options *opt = monc->client->options;
752 		int is_auth = ceph_auth_is_authenticated(monc->auth);
753 		if (ceph_con_keepalive_expired(&monc->con,
754 					       opt->monc_ping_timeout)) {
755 			dout("monc keepalive timeout\n");
756 			is_auth = 0;
757 			__close_session(monc);
758 			monc->hunting = true;
759 			__open_session(monc);
760 		}
761 
762 		if (!monc->hunting) {
763 			ceph_con_keepalive(&monc->con);
764 			__validate_auth(monc);
765 		}
766 
767 		if (is_auth)
768 			__send_subscribe(monc);
769 	}
770 	__schedule_delayed(monc);
771 	mutex_unlock(&monc->mutex);
772 }
773 
774 /*
775  * On startup, we build a temporary monmap populated with the IPs
776  * provided by mount(2).
777  */
778 static int build_initial_monmap(struct ceph_mon_client *monc)
779 {
780 	struct ceph_options *opt = monc->client->options;
781 	struct ceph_entity_addr *mon_addr = opt->mon_addr;
782 	int num_mon = opt->num_mon;
783 	int i;
784 
785 	/* build initial monmap */
786 	monc->monmap = kzalloc(sizeof(*monc->monmap) +
787 			       num_mon*sizeof(monc->monmap->mon_inst[0]),
788 			       GFP_KERNEL);
789 	if (!monc->monmap)
790 		return -ENOMEM;
791 	for (i = 0; i < num_mon; i++) {
792 		monc->monmap->mon_inst[i].addr = mon_addr[i];
793 		monc->monmap->mon_inst[i].addr.nonce = 0;
794 		monc->monmap->mon_inst[i].name.type =
795 			CEPH_ENTITY_TYPE_MON;
796 		monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
797 	}
798 	monc->monmap->num_mon = num_mon;
799 	return 0;
800 }
801 
802 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
803 {
804 	int err = 0;
805 
806 	dout("init\n");
807 	memset(monc, 0, sizeof(*monc));
808 	monc->client = cl;
809 	monc->monmap = NULL;
810 	mutex_init(&monc->mutex);
811 
812 	err = build_initial_monmap(monc);
813 	if (err)
814 		goto out;
815 
816 	/* connection */
817 	/* authentication */
818 	monc->auth = ceph_auth_init(cl->options->name,
819 				    cl->options->key);
820 	if (IS_ERR(monc->auth)) {
821 		err = PTR_ERR(monc->auth);
822 		goto out_monmap;
823 	}
824 	monc->auth->want_keys =
825 		CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
826 		CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
827 
828 	/* msgs */
829 	err = -ENOMEM;
830 	monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
831 				     sizeof(struct ceph_mon_subscribe_ack),
832 				     GFP_NOFS, true);
833 	if (!monc->m_subscribe_ack)
834 		goto out_auth;
835 
836 	monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
837 					 true);
838 	if (!monc->m_subscribe)
839 		goto out_subscribe_ack;
840 
841 	monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
842 					  true);
843 	if (!monc->m_auth_reply)
844 		goto out_subscribe;
845 
846 	monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
847 	monc->pending_auth = 0;
848 	if (!monc->m_auth)
849 		goto out_auth_reply;
850 
851 	ceph_con_init(&monc->con, monc, &mon_con_ops,
852 		      &monc->client->msgr);
853 
854 	monc->cur_mon = -1;
855 	monc->hunting = true;
856 	monc->sub_renew_after = jiffies;
857 	monc->sub_sent = 0;
858 
859 	INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
860 	monc->generic_request_tree = RB_ROOT;
861 	monc->num_generic_requests = 0;
862 	monc->last_tid = 0;
863 
864 	monc->have_mdsmap = 0;
865 	monc->have_osdmap = 0;
866 	monc->want_next_osdmap = 1;
867 	return 0;
868 
869 out_auth_reply:
870 	ceph_msg_put(monc->m_auth_reply);
871 out_subscribe:
872 	ceph_msg_put(monc->m_subscribe);
873 out_subscribe_ack:
874 	ceph_msg_put(monc->m_subscribe_ack);
875 out_auth:
876 	ceph_auth_destroy(monc->auth);
877 out_monmap:
878 	kfree(monc->monmap);
879 out:
880 	return err;
881 }
882 EXPORT_SYMBOL(ceph_monc_init);
883 
884 void ceph_monc_stop(struct ceph_mon_client *monc)
885 {
886 	dout("stop\n");
887 	cancel_delayed_work_sync(&monc->delayed_work);
888 
889 	mutex_lock(&monc->mutex);
890 	__close_session(monc);
891 
892 	mutex_unlock(&monc->mutex);
893 
894 	/*
895 	 * flush msgr queue before we destroy ourselves to ensure that:
896 	 *  - any work that references our embedded con is finished.
897 	 *  - any osd_client or other work that may reference an authorizer
898 	 *    finishes before we shut down the auth subsystem.
899 	 */
900 	ceph_msgr_flush();
901 
902 	ceph_auth_destroy(monc->auth);
903 
904 	ceph_msg_put(monc->m_auth);
905 	ceph_msg_put(monc->m_auth_reply);
906 	ceph_msg_put(monc->m_subscribe);
907 	ceph_msg_put(monc->m_subscribe_ack);
908 
909 	kfree(monc->monmap);
910 }
911 EXPORT_SYMBOL(ceph_monc_stop);
912 
913 static void handle_auth_reply(struct ceph_mon_client *monc,
914 			      struct ceph_msg *msg)
915 {
916 	int ret;
917 	int was_auth = 0;
918 	int had_debugfs_info, init_debugfs = 0;
919 
920 	mutex_lock(&monc->mutex);
921 	had_debugfs_info = have_debugfs_info(monc);
922 	was_auth = ceph_auth_is_authenticated(monc->auth);
923 	monc->pending_auth = 0;
924 	ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
925 				     msg->front.iov_len,
926 				     monc->m_auth->front.iov_base,
927 				     monc->m_auth->front_alloc_len);
928 	if (ret < 0) {
929 		monc->client->auth_err = ret;
930 		wake_up_all(&monc->client->auth_wq);
931 	} else if (ret > 0) {
932 		__send_prepared_auth_request(monc, ret);
933 	} else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
934 		dout("authenticated, starting session\n");
935 
936 		monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
937 		monc->client->msgr.inst.name.num =
938 					cpu_to_le64(monc->auth->global_id);
939 
940 		__send_subscribe(monc);
941 		__resend_generic_request(monc);
942 	}
943 
944 	if (!had_debugfs_info && have_debugfs_info(monc)) {
945 		pr_info("client%lld fsid %pU\n",
946 			ceph_client_id(monc->client),
947 			&monc->client->fsid);
948 		init_debugfs = 1;
949 	}
950 	mutex_unlock(&monc->mutex);
951 
952 	if (init_debugfs) {
953 		/*
954 		 * do debugfs initialization without mutex to avoid
955 		 * creating a locking dependency
956 		 */
957 		ceph_debugfs_client_init(monc->client);
958 	}
959 }
960 
961 static int __validate_auth(struct ceph_mon_client *monc)
962 {
963 	int ret;
964 
965 	if (monc->pending_auth)
966 		return 0;
967 
968 	ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
969 			      monc->m_auth->front_alloc_len);
970 	if (ret <= 0)
971 		return ret; /* either an error, or no need to authenticate */
972 	__send_prepared_auth_request(monc, ret);
973 	return 0;
974 }
975 
976 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
977 {
978 	int ret;
979 
980 	mutex_lock(&monc->mutex);
981 	ret = __validate_auth(monc);
982 	mutex_unlock(&monc->mutex);
983 	return ret;
984 }
985 EXPORT_SYMBOL(ceph_monc_validate_auth);
986 
987 /*
988  * handle incoming message
989  */
990 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
991 {
992 	struct ceph_mon_client *monc = con->private;
993 	int type = le16_to_cpu(msg->hdr.type);
994 
995 	if (!monc)
996 		return;
997 
998 	switch (type) {
999 	case CEPH_MSG_AUTH_REPLY:
1000 		handle_auth_reply(monc, msg);
1001 		break;
1002 
1003 	case CEPH_MSG_MON_SUBSCRIBE_ACK:
1004 		handle_subscribe_ack(monc, msg);
1005 		break;
1006 
1007 	case CEPH_MSG_STATFS_REPLY:
1008 		handle_statfs_reply(monc, msg);
1009 		break;
1010 
1011 	case CEPH_MSG_MON_GET_VERSION_REPLY:
1012 		handle_get_version_reply(monc, msg);
1013 		break;
1014 
1015 	case CEPH_MSG_MON_MAP:
1016 		ceph_monc_handle_map(monc, msg);
1017 		break;
1018 
1019 	case CEPH_MSG_OSD_MAP:
1020 		ceph_osdc_handle_map(&monc->client->osdc, msg);
1021 		break;
1022 
1023 	default:
1024 		/* can the chained handler handle it? */
1025 		if (monc->client->extra_mon_dispatch &&
1026 		    monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1027 			break;
1028 
1029 		pr_err("received unknown message type %d %s\n", type,
1030 		       ceph_msg_type_name(type));
1031 	}
1032 	ceph_msg_put(msg);
1033 }
1034 
1035 /*
1036  * Allocate memory for incoming message
1037  */
1038 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1039 				      struct ceph_msg_header *hdr,
1040 				      int *skip)
1041 {
1042 	struct ceph_mon_client *monc = con->private;
1043 	int type = le16_to_cpu(hdr->type);
1044 	int front_len = le32_to_cpu(hdr->front_len);
1045 	struct ceph_msg *m = NULL;
1046 
1047 	*skip = 0;
1048 
1049 	switch (type) {
1050 	case CEPH_MSG_MON_SUBSCRIBE_ACK:
1051 		m = ceph_msg_get(monc->m_subscribe_ack);
1052 		break;
1053 	case CEPH_MSG_STATFS_REPLY:
1054 		return get_generic_reply(con, hdr, skip);
1055 	case CEPH_MSG_AUTH_REPLY:
1056 		m = ceph_msg_get(monc->m_auth_reply);
1057 		break;
1058 	case CEPH_MSG_MON_GET_VERSION_REPLY:
1059 		if (le64_to_cpu(hdr->tid) != 0)
1060 			return get_generic_reply(con, hdr, skip);
1061 
1062 		/*
1063 		 * Older OSDs don't set reply tid even if the orignal
1064 		 * request had a non-zero tid.  Workaround this weirdness
1065 		 * by falling through to the allocate case.
1066 		 */
1067 	case CEPH_MSG_MON_MAP:
1068 	case CEPH_MSG_MDS_MAP:
1069 	case CEPH_MSG_OSD_MAP:
1070 		m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1071 		if (!m)
1072 			return NULL;	/* ENOMEM--return skip == 0 */
1073 		break;
1074 	}
1075 
1076 	if (!m) {
1077 		pr_info("alloc_msg unknown type %d\n", type);
1078 		*skip = 1;
1079 	} else if (front_len > m->front_alloc_len) {
1080 		pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1081 			front_len, m->front_alloc_len,
1082 			(unsigned int)con->peer_name.type,
1083 			le64_to_cpu(con->peer_name.num));
1084 		ceph_msg_put(m);
1085 		m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1086 	}
1087 
1088 	return m;
1089 }
1090 
1091 /*
1092  * If the monitor connection resets, pick a new monitor and resubmit
1093  * any pending requests.
1094  */
1095 static void mon_fault(struct ceph_connection *con)
1096 {
1097 	struct ceph_mon_client *monc = con->private;
1098 
1099 	if (!monc)
1100 		return;
1101 
1102 	dout("mon_fault\n");
1103 	mutex_lock(&monc->mutex);
1104 	if (!con->private)
1105 		goto out;
1106 
1107 	if (!monc->hunting)
1108 		pr_info("mon%d %s session lost, "
1109 			"hunting for new mon\n", monc->cur_mon,
1110 			ceph_pr_addr(&monc->con.peer_addr.in_addr));
1111 
1112 	__close_session(monc);
1113 	if (!monc->hunting) {
1114 		/* start hunting */
1115 		monc->hunting = true;
1116 		__open_session(monc);
1117 	} else {
1118 		/* already hunting, let's wait a bit */
1119 		__schedule_delayed(monc);
1120 	}
1121 out:
1122 	mutex_unlock(&monc->mutex);
1123 }
1124 
1125 /*
1126  * We can ignore refcounting on the connection struct, as all references
1127  * will come from the messenger workqueue, which is drained prior to
1128  * mon_client destruction.
1129  */
1130 static struct ceph_connection *con_get(struct ceph_connection *con)
1131 {
1132 	return con;
1133 }
1134 
1135 static void con_put(struct ceph_connection *con)
1136 {
1137 }
1138 
1139 static const struct ceph_connection_operations mon_con_ops = {
1140 	.get = con_get,
1141 	.put = con_put,
1142 	.dispatch = dispatch,
1143 	.fault = mon_fault,
1144 	.alloc_msg = mon_alloc_msg,
1145 };
1146