xref: /openbmc/linux/net/ceph/mon_client.c (revision f5ad1c74)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3 
4 #include <linux/module.h>
5 #include <linux/types.h>
6 #include <linux/slab.h>
7 #include <linux/random.h>
8 #include <linux/sched.h>
9 
10 #include <linux/ceph/ceph_features.h>
11 #include <linux/ceph/mon_client.h>
12 #include <linux/ceph/libceph.h>
13 #include <linux/ceph/debugfs.h>
14 #include <linux/ceph/decode.h>
15 #include <linux/ceph/auth.h>
16 
17 /*
18  * Interact with Ceph monitor cluster.  Handle requests for new map
19  * versions, and periodically resend as needed.  Also implement
20  * statfs() and umount().
21  *
22  * A small cluster of Ceph "monitors" are responsible for managing critical
23  * cluster configuration and state information.  An odd number (e.g., 3, 5)
24  * of cmon daemons use a modified version of the Paxos part-time parliament
25  * algorithm to manage the MDS map (mds cluster membership), OSD map, and
26  * list of clients who have mounted the file system.
27  *
28  * We maintain an open, active session with a monitor at all times in order to
29  * receive timely MDSMap updates.  We periodically send a keepalive byte on the
30  * TCP socket to ensure we detect a failure.  If the connection does break, we
31  * randomly hunt for a new monitor.  Once the connection is reestablished, we
32  * resend any outstanding requests.
33  */
34 
35 static const struct ceph_connection_operations mon_con_ops;
36 
37 static int __validate_auth(struct ceph_mon_client *monc);
38 
39 /*
40  * Decode a monmap blob (e.g., during mount).
41  */
42 static struct ceph_monmap *ceph_monmap_decode(void *p, void *end)
43 {
44 	struct ceph_monmap *m = NULL;
45 	int i, err = -EINVAL;
46 	struct ceph_fsid fsid;
47 	u32 epoch, num_mon;
48 	u32 len;
49 
50 	ceph_decode_32_safe(&p, end, len, bad);
51 	ceph_decode_need(&p, end, len, bad);
52 
53 	dout("monmap_decode %p %p len %d (%d)\n", p, end, len, (int)(end-p));
54 	p += sizeof(u16);  /* skip version */
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 
62 	if (num_mon > CEPH_MAX_MON)
63 		goto bad;
64 	m = kmalloc(struct_size(m, mon_inst, num_mon), GFP_NOFS);
65 	if (m == NULL)
66 		return ERR_PTR(-ENOMEM);
67 	m->fsid = fsid;
68 	m->epoch = epoch;
69 	m->num_mon = num_mon;
70 	for (i = 0; i < num_mon; ++i) {
71 		struct ceph_entity_inst *inst = &m->mon_inst[i];
72 
73 		/* copy name portion */
74 		ceph_decode_copy_safe(&p, end, &inst->name,
75 					sizeof(inst->name), bad);
76 		err = ceph_decode_entity_addr(&p, end, &inst->addr);
77 		if (err)
78 			goto bad;
79 	}
80 	dout("monmap_decode epoch %d, num_mon %d\n", m->epoch,
81 	     m->num_mon);
82 	for (i = 0; i < m->num_mon; i++)
83 		dout("monmap_decode  mon%d is %s\n", i,
84 		     ceph_pr_addr(&m->mon_inst[i].addr));
85 	return m;
86 bad:
87 	dout("monmap_decode failed with %d\n", err);
88 	kfree(m);
89 	return ERR_PTR(err);
90 }
91 
92 /*
93  * return true if *addr is included in the monmap.
94  */
95 int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr)
96 {
97 	int i;
98 
99 	for (i = 0; i < m->num_mon; i++)
100 		if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0)
101 			return 1;
102 	return 0;
103 }
104 
105 /*
106  * Send an auth request.
107  */
108 static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len)
109 {
110 	monc->pending_auth = 1;
111 	monc->m_auth->front.iov_len = len;
112 	monc->m_auth->hdr.front_len = cpu_to_le32(len);
113 	ceph_msg_revoke(monc->m_auth);
114 	ceph_msg_get(monc->m_auth);  /* keep our ref */
115 	ceph_con_send(&monc->con, monc->m_auth);
116 }
117 
118 /*
119  * Close monitor session, if any.
120  */
121 static void __close_session(struct ceph_mon_client *monc)
122 {
123 	dout("__close_session closing mon%d\n", monc->cur_mon);
124 	ceph_msg_revoke(monc->m_auth);
125 	ceph_msg_revoke_incoming(monc->m_auth_reply);
126 	ceph_msg_revoke(monc->m_subscribe);
127 	ceph_msg_revoke_incoming(monc->m_subscribe_ack);
128 	ceph_con_close(&monc->con);
129 
130 	monc->pending_auth = 0;
131 	ceph_auth_reset(monc->auth);
132 }
133 
134 /*
135  * Pick a new monitor at random and set cur_mon.  If we are repicking
136  * (i.e. cur_mon is already set), be sure to pick a different one.
137  */
138 static void pick_new_mon(struct ceph_mon_client *monc)
139 {
140 	int old_mon = monc->cur_mon;
141 
142 	BUG_ON(monc->monmap->num_mon < 1);
143 
144 	if (monc->monmap->num_mon == 1) {
145 		monc->cur_mon = 0;
146 	} else {
147 		int max = monc->monmap->num_mon;
148 		int o = -1;
149 		int n;
150 
151 		if (monc->cur_mon >= 0) {
152 			if (monc->cur_mon < monc->monmap->num_mon)
153 				o = monc->cur_mon;
154 			if (o >= 0)
155 				max--;
156 		}
157 
158 		n = prandom_u32() % max;
159 		if (o >= 0 && n >= o)
160 			n++;
161 
162 		monc->cur_mon = n;
163 	}
164 
165 	dout("%s mon%d -> mon%d out of %d mons\n", __func__, old_mon,
166 	     monc->cur_mon, monc->monmap->num_mon);
167 }
168 
169 /*
170  * Open a session with a new monitor.
171  */
172 static void __open_session(struct ceph_mon_client *monc)
173 {
174 	int ret;
175 
176 	pick_new_mon(monc);
177 
178 	monc->hunting = true;
179 	if (monc->had_a_connection) {
180 		monc->hunt_mult *= CEPH_MONC_HUNT_BACKOFF;
181 		if (monc->hunt_mult > CEPH_MONC_HUNT_MAX_MULT)
182 			monc->hunt_mult = CEPH_MONC_HUNT_MAX_MULT;
183 	}
184 
185 	monc->sub_renew_after = jiffies; /* i.e., expired */
186 	monc->sub_renew_sent = 0;
187 
188 	dout("%s opening mon%d\n", __func__, monc->cur_mon);
189 	ceph_con_open(&monc->con, CEPH_ENTITY_TYPE_MON, monc->cur_mon,
190 		      &monc->monmap->mon_inst[monc->cur_mon].addr);
191 
192 	/*
193 	 * send an initial keepalive to ensure our timestamp is valid
194 	 * by the time we are in an OPENED state
195 	 */
196 	ceph_con_keepalive(&monc->con);
197 
198 	/* initiate authentication handshake */
199 	ret = ceph_auth_build_hello(monc->auth,
200 				    monc->m_auth->front.iov_base,
201 				    monc->m_auth->front_alloc_len);
202 	BUG_ON(ret <= 0);
203 	__send_prepared_auth_request(monc, ret);
204 }
205 
206 static void reopen_session(struct ceph_mon_client *monc)
207 {
208 	if (!monc->hunting)
209 		pr_info("mon%d %s session lost, hunting for new mon\n",
210 		    monc->cur_mon, ceph_pr_addr(&monc->con.peer_addr));
211 
212 	__close_session(monc);
213 	__open_session(monc);
214 }
215 
216 void ceph_monc_reopen_session(struct ceph_mon_client *monc)
217 {
218 	mutex_lock(&monc->mutex);
219 	reopen_session(monc);
220 	mutex_unlock(&monc->mutex);
221 }
222 
223 static void un_backoff(struct ceph_mon_client *monc)
224 {
225 	monc->hunt_mult /= 2; /* reduce by 50% */
226 	if (monc->hunt_mult < 1)
227 		monc->hunt_mult = 1;
228 	dout("%s hunt_mult now %d\n", __func__, monc->hunt_mult);
229 }
230 
231 /*
232  * Reschedule delayed work timer.
233  */
234 static void __schedule_delayed(struct ceph_mon_client *monc)
235 {
236 	unsigned long delay;
237 
238 	if (monc->hunting)
239 		delay = CEPH_MONC_HUNT_INTERVAL * monc->hunt_mult;
240 	else
241 		delay = CEPH_MONC_PING_INTERVAL;
242 
243 	dout("__schedule_delayed after %lu\n", delay);
244 	mod_delayed_work(system_wq, &monc->delayed_work,
245 			 round_jiffies_relative(delay));
246 }
247 
248 const char *ceph_sub_str[] = {
249 	[CEPH_SUB_MONMAP] = "monmap",
250 	[CEPH_SUB_OSDMAP] = "osdmap",
251 	[CEPH_SUB_FSMAP]  = "fsmap.user",
252 	[CEPH_SUB_MDSMAP] = "mdsmap",
253 };
254 
255 /*
256  * Send subscribe request for one or more maps, according to
257  * monc->subs.
258  */
259 static void __send_subscribe(struct ceph_mon_client *monc)
260 {
261 	struct ceph_msg *msg = monc->m_subscribe;
262 	void *p = msg->front.iov_base;
263 	void *const end = p + msg->front_alloc_len;
264 	int num = 0;
265 	int i;
266 
267 	dout("%s sent %lu\n", __func__, monc->sub_renew_sent);
268 
269 	BUG_ON(monc->cur_mon < 0);
270 
271 	if (!monc->sub_renew_sent)
272 		monc->sub_renew_sent = jiffies | 1; /* never 0 */
273 
274 	msg->hdr.version = cpu_to_le16(2);
275 
276 	for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
277 		if (monc->subs[i].want)
278 			num++;
279 	}
280 	BUG_ON(num < 1); /* monmap sub is always there */
281 	ceph_encode_32(&p, num);
282 	for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
283 		char buf[32];
284 		int len;
285 
286 		if (!monc->subs[i].want)
287 			continue;
288 
289 		len = sprintf(buf, "%s", ceph_sub_str[i]);
290 		if (i == CEPH_SUB_MDSMAP &&
291 		    monc->fs_cluster_id != CEPH_FS_CLUSTER_ID_NONE)
292 			len += sprintf(buf + len, ".%d", monc->fs_cluster_id);
293 
294 		dout("%s %s start %llu flags 0x%x\n", __func__, buf,
295 		     le64_to_cpu(monc->subs[i].item.start),
296 		     monc->subs[i].item.flags);
297 		ceph_encode_string(&p, end, buf, len);
298 		memcpy(p, &monc->subs[i].item, sizeof(monc->subs[i].item));
299 		p += sizeof(monc->subs[i].item);
300 	}
301 
302 	BUG_ON(p > end);
303 	msg->front.iov_len = p - msg->front.iov_base;
304 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
305 	ceph_msg_revoke(msg);
306 	ceph_con_send(&monc->con, ceph_msg_get(msg));
307 }
308 
309 static void handle_subscribe_ack(struct ceph_mon_client *monc,
310 				 struct ceph_msg *msg)
311 {
312 	unsigned int seconds;
313 	struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
314 
315 	if (msg->front.iov_len < sizeof(*h))
316 		goto bad;
317 	seconds = le32_to_cpu(h->duration);
318 
319 	mutex_lock(&monc->mutex);
320 	if (monc->sub_renew_sent) {
321 		/*
322 		 * This is only needed for legacy (infernalis or older)
323 		 * MONs -- see delayed_work().
324 		 */
325 		monc->sub_renew_after = monc->sub_renew_sent +
326 					    (seconds >> 1) * HZ - 1;
327 		dout("%s sent %lu duration %d renew after %lu\n", __func__,
328 		     monc->sub_renew_sent, seconds, monc->sub_renew_after);
329 		monc->sub_renew_sent = 0;
330 	} else {
331 		dout("%s sent %lu renew after %lu, ignoring\n", __func__,
332 		     monc->sub_renew_sent, monc->sub_renew_after);
333 	}
334 	mutex_unlock(&monc->mutex);
335 	return;
336 bad:
337 	pr_err("got corrupt subscribe-ack msg\n");
338 	ceph_msg_dump(msg);
339 }
340 
341 /*
342  * Register interest in a map
343  *
344  * @sub: one of CEPH_SUB_*
345  * @epoch: X for "every map since X", or 0 for "just the latest"
346  */
347 static bool __ceph_monc_want_map(struct ceph_mon_client *monc, int sub,
348 				 u32 epoch, bool continuous)
349 {
350 	__le64 start = cpu_to_le64(epoch);
351 	u8 flags = !continuous ? CEPH_SUBSCRIBE_ONETIME : 0;
352 
353 	dout("%s %s epoch %u continuous %d\n", __func__, ceph_sub_str[sub],
354 	     epoch, continuous);
355 
356 	if (monc->subs[sub].want &&
357 	    monc->subs[sub].item.start == start &&
358 	    monc->subs[sub].item.flags == flags)
359 		return false;
360 
361 	monc->subs[sub].item.start = start;
362 	monc->subs[sub].item.flags = flags;
363 	monc->subs[sub].want = true;
364 
365 	return true;
366 }
367 
368 bool ceph_monc_want_map(struct ceph_mon_client *monc, int sub, u32 epoch,
369 			bool continuous)
370 {
371 	bool need_request;
372 
373 	mutex_lock(&monc->mutex);
374 	need_request = __ceph_monc_want_map(monc, sub, epoch, continuous);
375 	mutex_unlock(&monc->mutex);
376 
377 	return need_request;
378 }
379 EXPORT_SYMBOL(ceph_monc_want_map);
380 
381 /*
382  * Keep track of which maps we have
383  *
384  * @sub: one of CEPH_SUB_*
385  */
386 static void __ceph_monc_got_map(struct ceph_mon_client *monc, int sub,
387 				u32 epoch)
388 {
389 	dout("%s %s epoch %u\n", __func__, ceph_sub_str[sub], epoch);
390 
391 	if (monc->subs[sub].want) {
392 		if (monc->subs[sub].item.flags & CEPH_SUBSCRIBE_ONETIME)
393 			monc->subs[sub].want = false;
394 		else
395 			monc->subs[sub].item.start = cpu_to_le64(epoch + 1);
396 	}
397 
398 	monc->subs[sub].have = epoch;
399 }
400 
401 void ceph_monc_got_map(struct ceph_mon_client *monc, int sub, u32 epoch)
402 {
403 	mutex_lock(&monc->mutex);
404 	__ceph_monc_got_map(monc, sub, epoch);
405 	mutex_unlock(&monc->mutex);
406 }
407 EXPORT_SYMBOL(ceph_monc_got_map);
408 
409 void ceph_monc_renew_subs(struct ceph_mon_client *monc)
410 {
411 	mutex_lock(&monc->mutex);
412 	__send_subscribe(monc);
413 	mutex_unlock(&monc->mutex);
414 }
415 EXPORT_SYMBOL(ceph_monc_renew_subs);
416 
417 /*
418  * Wait for an osdmap with a given epoch.
419  *
420  * @epoch: epoch to wait for
421  * @timeout: in jiffies, 0 means "wait forever"
422  */
423 int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
424 			  unsigned long timeout)
425 {
426 	unsigned long started = jiffies;
427 	long ret;
428 
429 	mutex_lock(&monc->mutex);
430 	while (monc->subs[CEPH_SUB_OSDMAP].have < epoch) {
431 		mutex_unlock(&monc->mutex);
432 
433 		if (timeout && time_after_eq(jiffies, started + timeout))
434 			return -ETIMEDOUT;
435 
436 		ret = wait_event_interruptible_timeout(monc->client->auth_wq,
437 				     monc->subs[CEPH_SUB_OSDMAP].have >= epoch,
438 				     ceph_timeout_jiffies(timeout));
439 		if (ret < 0)
440 			return ret;
441 
442 		mutex_lock(&monc->mutex);
443 	}
444 
445 	mutex_unlock(&monc->mutex);
446 	return 0;
447 }
448 EXPORT_SYMBOL(ceph_monc_wait_osdmap);
449 
450 /*
451  * Open a session with a random monitor.  Request monmap and osdmap,
452  * which are waited upon in __ceph_open_session().
453  */
454 int ceph_monc_open_session(struct ceph_mon_client *monc)
455 {
456 	mutex_lock(&monc->mutex);
457 	__ceph_monc_want_map(monc, CEPH_SUB_MONMAP, 0, true);
458 	__ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, 0, false);
459 	__open_session(monc);
460 	__schedule_delayed(monc);
461 	mutex_unlock(&monc->mutex);
462 	return 0;
463 }
464 EXPORT_SYMBOL(ceph_monc_open_session);
465 
466 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
467 				 struct ceph_msg *msg)
468 {
469 	struct ceph_client *client = monc->client;
470 	struct ceph_monmap *monmap;
471 	void *p, *end;
472 
473 	mutex_lock(&monc->mutex);
474 
475 	dout("handle_monmap\n");
476 	p = msg->front.iov_base;
477 	end = p + msg->front.iov_len;
478 
479 	monmap = ceph_monmap_decode(p, end);
480 	if (IS_ERR(monmap)) {
481 		pr_err("problem decoding monmap, %d\n",
482 		       (int)PTR_ERR(monmap));
483 		ceph_msg_dump(msg);
484 		goto out;
485 	}
486 
487 	if (ceph_check_fsid(client, &monmap->fsid) < 0) {
488 		kfree(monmap);
489 		goto out;
490 	}
491 
492 	kfree(monc->monmap);
493 	monc->monmap = monmap;
494 
495 	__ceph_monc_got_map(monc, CEPH_SUB_MONMAP, monc->monmap->epoch);
496 	client->have_fsid = true;
497 
498 out:
499 	mutex_unlock(&monc->mutex);
500 	wake_up_all(&client->auth_wq);
501 }
502 
503 /*
504  * generic requests (currently statfs, mon_get_version)
505  */
506 DEFINE_RB_FUNCS(generic_request, struct ceph_mon_generic_request, tid, node)
507 
508 static void release_generic_request(struct kref *kref)
509 {
510 	struct ceph_mon_generic_request *req =
511 		container_of(kref, struct ceph_mon_generic_request, kref);
512 
513 	dout("%s greq %p request %p reply %p\n", __func__, req, req->request,
514 	     req->reply);
515 	WARN_ON(!RB_EMPTY_NODE(&req->node));
516 
517 	if (req->reply)
518 		ceph_msg_put(req->reply);
519 	if (req->request)
520 		ceph_msg_put(req->request);
521 
522 	kfree(req);
523 }
524 
525 static void put_generic_request(struct ceph_mon_generic_request *req)
526 {
527 	if (req)
528 		kref_put(&req->kref, release_generic_request);
529 }
530 
531 static void get_generic_request(struct ceph_mon_generic_request *req)
532 {
533 	kref_get(&req->kref);
534 }
535 
536 static struct ceph_mon_generic_request *
537 alloc_generic_request(struct ceph_mon_client *monc, gfp_t gfp)
538 {
539 	struct ceph_mon_generic_request *req;
540 
541 	req = kzalloc(sizeof(*req), gfp);
542 	if (!req)
543 		return NULL;
544 
545 	req->monc = monc;
546 	kref_init(&req->kref);
547 	RB_CLEAR_NODE(&req->node);
548 	init_completion(&req->completion);
549 
550 	dout("%s greq %p\n", __func__, req);
551 	return req;
552 }
553 
554 static void register_generic_request(struct ceph_mon_generic_request *req)
555 {
556 	struct ceph_mon_client *monc = req->monc;
557 
558 	WARN_ON(req->tid);
559 
560 	get_generic_request(req);
561 	req->tid = ++monc->last_tid;
562 	insert_generic_request(&monc->generic_request_tree, req);
563 }
564 
565 static void send_generic_request(struct ceph_mon_client *monc,
566 				 struct ceph_mon_generic_request *req)
567 {
568 	WARN_ON(!req->tid);
569 
570 	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
571 	req->request->hdr.tid = cpu_to_le64(req->tid);
572 	ceph_con_send(&monc->con, ceph_msg_get(req->request));
573 }
574 
575 static void __finish_generic_request(struct ceph_mon_generic_request *req)
576 {
577 	struct ceph_mon_client *monc = req->monc;
578 
579 	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
580 	erase_generic_request(&monc->generic_request_tree, req);
581 
582 	ceph_msg_revoke(req->request);
583 	ceph_msg_revoke_incoming(req->reply);
584 }
585 
586 static void finish_generic_request(struct ceph_mon_generic_request *req)
587 {
588 	__finish_generic_request(req);
589 	put_generic_request(req);
590 }
591 
592 static void complete_generic_request(struct ceph_mon_generic_request *req)
593 {
594 	if (req->complete_cb)
595 		req->complete_cb(req);
596 	else
597 		complete_all(&req->completion);
598 	put_generic_request(req);
599 }
600 
601 static void cancel_generic_request(struct ceph_mon_generic_request *req)
602 {
603 	struct ceph_mon_client *monc = req->monc;
604 	struct ceph_mon_generic_request *lookup_req;
605 
606 	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
607 
608 	mutex_lock(&monc->mutex);
609 	lookup_req = lookup_generic_request(&monc->generic_request_tree,
610 					    req->tid);
611 	if (lookup_req) {
612 		WARN_ON(lookup_req != req);
613 		finish_generic_request(req);
614 	}
615 
616 	mutex_unlock(&monc->mutex);
617 }
618 
619 static int wait_generic_request(struct ceph_mon_generic_request *req)
620 {
621 	int ret;
622 
623 	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
624 	ret = wait_for_completion_interruptible(&req->completion);
625 	if (ret)
626 		cancel_generic_request(req);
627 	else
628 		ret = req->result; /* completed */
629 
630 	return ret;
631 }
632 
633 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
634 					 struct ceph_msg_header *hdr,
635 					 int *skip)
636 {
637 	struct ceph_mon_client *monc = con->private;
638 	struct ceph_mon_generic_request *req;
639 	u64 tid = le64_to_cpu(hdr->tid);
640 	struct ceph_msg *m;
641 
642 	mutex_lock(&monc->mutex);
643 	req = lookup_generic_request(&monc->generic_request_tree, tid);
644 	if (!req) {
645 		dout("get_generic_reply %lld dne\n", tid);
646 		*skip = 1;
647 		m = NULL;
648 	} else {
649 		dout("get_generic_reply %lld got %p\n", tid, req->reply);
650 		*skip = 0;
651 		m = ceph_msg_get(req->reply);
652 		/*
653 		 * we don't need to track the connection reading into
654 		 * this reply because we only have one open connection
655 		 * at a time, ever.
656 		 */
657 	}
658 	mutex_unlock(&monc->mutex);
659 	return m;
660 }
661 
662 /*
663  * statfs
664  */
665 static void handle_statfs_reply(struct ceph_mon_client *monc,
666 				struct ceph_msg *msg)
667 {
668 	struct ceph_mon_generic_request *req;
669 	struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
670 	u64 tid = le64_to_cpu(msg->hdr.tid);
671 
672 	dout("%s msg %p tid %llu\n", __func__, msg, tid);
673 
674 	if (msg->front.iov_len != sizeof(*reply))
675 		goto bad;
676 
677 	mutex_lock(&monc->mutex);
678 	req = lookup_generic_request(&monc->generic_request_tree, tid);
679 	if (!req) {
680 		mutex_unlock(&monc->mutex);
681 		return;
682 	}
683 
684 	req->result = 0;
685 	*req->u.st = reply->st; /* struct */
686 	__finish_generic_request(req);
687 	mutex_unlock(&monc->mutex);
688 
689 	complete_generic_request(req);
690 	return;
691 
692 bad:
693 	pr_err("corrupt statfs reply, tid %llu\n", tid);
694 	ceph_msg_dump(msg);
695 }
696 
697 /*
698  * Do a synchronous statfs().
699  */
700 int ceph_monc_do_statfs(struct ceph_mon_client *monc, u64 data_pool,
701 			struct ceph_statfs *buf)
702 {
703 	struct ceph_mon_generic_request *req;
704 	struct ceph_mon_statfs *h;
705 	int ret = -ENOMEM;
706 
707 	req = alloc_generic_request(monc, GFP_NOFS);
708 	if (!req)
709 		goto out;
710 
711 	req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
712 				    true);
713 	if (!req->request)
714 		goto out;
715 
716 	req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 64, GFP_NOFS, true);
717 	if (!req->reply)
718 		goto out;
719 
720 	req->u.st = buf;
721 	req->request->hdr.version = cpu_to_le16(2);
722 
723 	mutex_lock(&monc->mutex);
724 	register_generic_request(req);
725 	/* fill out request */
726 	h = req->request->front.iov_base;
727 	h->monhdr.have_version = 0;
728 	h->monhdr.session_mon = cpu_to_le16(-1);
729 	h->monhdr.session_mon_tid = 0;
730 	h->fsid = monc->monmap->fsid;
731 	h->contains_data_pool = (data_pool != CEPH_NOPOOL);
732 	h->data_pool = cpu_to_le64(data_pool);
733 	send_generic_request(monc, req);
734 	mutex_unlock(&monc->mutex);
735 
736 	ret = wait_generic_request(req);
737 out:
738 	put_generic_request(req);
739 	return ret;
740 }
741 EXPORT_SYMBOL(ceph_monc_do_statfs);
742 
743 static void handle_get_version_reply(struct ceph_mon_client *monc,
744 				     struct ceph_msg *msg)
745 {
746 	struct ceph_mon_generic_request *req;
747 	u64 tid = le64_to_cpu(msg->hdr.tid);
748 	void *p = msg->front.iov_base;
749 	void *end = p + msg->front_alloc_len;
750 	u64 handle;
751 
752 	dout("%s msg %p tid %llu\n", __func__, msg, tid);
753 
754 	ceph_decode_need(&p, end, 2*sizeof(u64), bad);
755 	handle = ceph_decode_64(&p);
756 	if (tid != 0 && tid != handle)
757 		goto bad;
758 
759 	mutex_lock(&monc->mutex);
760 	req = lookup_generic_request(&monc->generic_request_tree, handle);
761 	if (!req) {
762 		mutex_unlock(&monc->mutex);
763 		return;
764 	}
765 
766 	req->result = 0;
767 	req->u.newest = ceph_decode_64(&p);
768 	__finish_generic_request(req);
769 	mutex_unlock(&monc->mutex);
770 
771 	complete_generic_request(req);
772 	return;
773 
774 bad:
775 	pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
776 	ceph_msg_dump(msg);
777 }
778 
779 static struct ceph_mon_generic_request *
780 __ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
781 			ceph_monc_callback_t cb, u64 private_data)
782 {
783 	struct ceph_mon_generic_request *req;
784 
785 	req = alloc_generic_request(monc, GFP_NOIO);
786 	if (!req)
787 		goto err_put_req;
788 
789 	req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
790 				    sizeof(u64) + sizeof(u32) + strlen(what),
791 				    GFP_NOIO, true);
792 	if (!req->request)
793 		goto err_put_req;
794 
795 	req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 32, GFP_NOIO,
796 				  true);
797 	if (!req->reply)
798 		goto err_put_req;
799 
800 	req->complete_cb = cb;
801 	req->private_data = private_data;
802 
803 	mutex_lock(&monc->mutex);
804 	register_generic_request(req);
805 	{
806 		void *p = req->request->front.iov_base;
807 		void *const end = p + req->request->front_alloc_len;
808 
809 		ceph_encode_64(&p, req->tid); /* handle */
810 		ceph_encode_string(&p, end, what, strlen(what));
811 		WARN_ON(p != end);
812 	}
813 	send_generic_request(monc, req);
814 	mutex_unlock(&monc->mutex);
815 
816 	return req;
817 
818 err_put_req:
819 	put_generic_request(req);
820 	return ERR_PTR(-ENOMEM);
821 }
822 
823 /*
824  * Send MMonGetVersion and wait for the reply.
825  *
826  * @what: one of "mdsmap", "osdmap" or "monmap"
827  */
828 int ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
829 			  u64 *newest)
830 {
831 	struct ceph_mon_generic_request *req;
832 	int ret;
833 
834 	req = __ceph_monc_get_version(monc, what, NULL, 0);
835 	if (IS_ERR(req))
836 		return PTR_ERR(req);
837 
838 	ret = wait_generic_request(req);
839 	if (!ret)
840 		*newest = req->u.newest;
841 
842 	put_generic_request(req);
843 	return ret;
844 }
845 EXPORT_SYMBOL(ceph_monc_get_version);
846 
847 /*
848  * Send MMonGetVersion,
849  *
850  * @what: one of "mdsmap", "osdmap" or "monmap"
851  */
852 int ceph_monc_get_version_async(struct ceph_mon_client *monc, const char *what,
853 				ceph_monc_callback_t cb, u64 private_data)
854 {
855 	struct ceph_mon_generic_request *req;
856 
857 	req = __ceph_monc_get_version(monc, what, cb, private_data);
858 	if (IS_ERR(req))
859 		return PTR_ERR(req);
860 
861 	put_generic_request(req);
862 	return 0;
863 }
864 EXPORT_SYMBOL(ceph_monc_get_version_async);
865 
866 static void handle_command_ack(struct ceph_mon_client *monc,
867 			       struct ceph_msg *msg)
868 {
869 	struct ceph_mon_generic_request *req;
870 	void *p = msg->front.iov_base;
871 	void *const end = p + msg->front_alloc_len;
872 	u64 tid = le64_to_cpu(msg->hdr.tid);
873 
874 	dout("%s msg %p tid %llu\n", __func__, msg, tid);
875 
876 	ceph_decode_need(&p, end, sizeof(struct ceph_mon_request_header) +
877 							    sizeof(u32), bad);
878 	p += sizeof(struct ceph_mon_request_header);
879 
880 	mutex_lock(&monc->mutex);
881 	req = lookup_generic_request(&monc->generic_request_tree, tid);
882 	if (!req) {
883 		mutex_unlock(&monc->mutex);
884 		return;
885 	}
886 
887 	req->result = ceph_decode_32(&p);
888 	__finish_generic_request(req);
889 	mutex_unlock(&monc->mutex);
890 
891 	complete_generic_request(req);
892 	return;
893 
894 bad:
895 	pr_err("corrupt mon_command ack, tid %llu\n", tid);
896 	ceph_msg_dump(msg);
897 }
898 
899 static __printf(2, 0)
900 int do_mon_command_vargs(struct ceph_mon_client *monc, const char *fmt,
901 			 va_list ap)
902 {
903 	struct ceph_mon_generic_request *req;
904 	struct ceph_mon_command *h;
905 	int ret = -ENOMEM;
906 	int len;
907 
908 	req = alloc_generic_request(monc, GFP_NOIO);
909 	if (!req)
910 		goto out;
911 
912 	req->request = ceph_msg_new(CEPH_MSG_MON_COMMAND, 256, GFP_NOIO, true);
913 	if (!req->request)
914 		goto out;
915 
916 	req->reply = ceph_msg_new(CEPH_MSG_MON_COMMAND_ACK, 512, GFP_NOIO,
917 				  true);
918 	if (!req->reply)
919 		goto out;
920 
921 	mutex_lock(&monc->mutex);
922 	register_generic_request(req);
923 	h = req->request->front.iov_base;
924 	h->monhdr.have_version = 0;
925 	h->monhdr.session_mon = cpu_to_le16(-1);
926 	h->monhdr.session_mon_tid = 0;
927 	h->fsid = monc->monmap->fsid;
928 	h->num_strs = cpu_to_le32(1);
929 	len = vsprintf(h->str, fmt, ap);
930 	h->str_len = cpu_to_le32(len);
931 	send_generic_request(monc, req);
932 	mutex_unlock(&monc->mutex);
933 
934 	ret = wait_generic_request(req);
935 out:
936 	put_generic_request(req);
937 	return ret;
938 }
939 
940 static __printf(2, 3)
941 int do_mon_command(struct ceph_mon_client *monc, const char *fmt, ...)
942 {
943 	va_list ap;
944 	int ret;
945 
946 	va_start(ap, fmt);
947 	ret = do_mon_command_vargs(monc, fmt, ap);
948 	va_end(ap);
949 	return ret;
950 }
951 
952 int ceph_monc_blocklist_add(struct ceph_mon_client *monc,
953 			    struct ceph_entity_addr *client_addr)
954 {
955 	int ret;
956 
957 	ret = do_mon_command(monc,
958 			     "{ \"prefix\": \"osd blocklist\", \
959 				\"blocklistop\": \"add\", \
960 				\"addr\": \"%pISpc/%u\" }",
961 			     &client_addr->in_addr,
962 			     le32_to_cpu(client_addr->nonce));
963 	if (ret == -EINVAL) {
964 		/*
965 		 * The monitor returns EINVAL on an unrecognized command.
966 		 * Try the legacy command -- it is exactly the same except
967 		 * for the name.
968 		 */
969 		ret = do_mon_command(monc,
970 				     "{ \"prefix\": \"osd blacklist\", \
971 					\"blacklistop\": \"add\", \
972 					\"addr\": \"%pISpc/%u\" }",
973 				     &client_addr->in_addr,
974 				     le32_to_cpu(client_addr->nonce));
975 	}
976 	if (ret)
977 		return ret;
978 
979 	/*
980 	 * Make sure we have the osdmap that includes the blocklist
981 	 * entry.  This is needed to ensure that the OSDs pick up the
982 	 * new blocklist before processing any future requests from
983 	 * this client.
984 	 */
985 	return ceph_wait_for_latest_osdmap(monc->client, 0);
986 }
987 EXPORT_SYMBOL(ceph_monc_blocklist_add);
988 
989 /*
990  * Resend pending generic requests.
991  */
992 static void __resend_generic_request(struct ceph_mon_client *monc)
993 {
994 	struct ceph_mon_generic_request *req;
995 	struct rb_node *p;
996 
997 	for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
998 		req = rb_entry(p, struct ceph_mon_generic_request, node);
999 		ceph_msg_revoke(req->request);
1000 		ceph_msg_revoke_incoming(req->reply);
1001 		ceph_con_send(&monc->con, ceph_msg_get(req->request));
1002 	}
1003 }
1004 
1005 /*
1006  * Delayed work.  If we haven't mounted yet, retry.  Otherwise,
1007  * renew/retry subscription as needed (in case it is timing out, or we
1008  * got an ENOMEM).  And keep the monitor connection alive.
1009  */
1010 static void delayed_work(struct work_struct *work)
1011 {
1012 	struct ceph_mon_client *monc =
1013 		container_of(work, struct ceph_mon_client, delayed_work.work);
1014 
1015 	dout("monc delayed_work\n");
1016 	mutex_lock(&monc->mutex);
1017 	if (monc->hunting) {
1018 		dout("%s continuing hunt\n", __func__);
1019 		reopen_session(monc);
1020 	} else {
1021 		int is_auth = ceph_auth_is_authenticated(monc->auth);
1022 		if (ceph_con_keepalive_expired(&monc->con,
1023 					       CEPH_MONC_PING_TIMEOUT)) {
1024 			dout("monc keepalive timeout\n");
1025 			is_auth = 0;
1026 			reopen_session(monc);
1027 		}
1028 
1029 		if (!monc->hunting) {
1030 			ceph_con_keepalive(&monc->con);
1031 			__validate_auth(monc);
1032 			un_backoff(monc);
1033 		}
1034 
1035 		if (is_auth &&
1036 		    !(monc->con.peer_features & CEPH_FEATURE_MON_STATEFUL_SUB)) {
1037 			unsigned long now = jiffies;
1038 
1039 			dout("%s renew subs? now %lu renew after %lu\n",
1040 			     __func__, now, monc->sub_renew_after);
1041 			if (time_after_eq(now, monc->sub_renew_after))
1042 				__send_subscribe(monc);
1043 		}
1044 	}
1045 	__schedule_delayed(monc);
1046 	mutex_unlock(&monc->mutex);
1047 }
1048 
1049 /*
1050  * On startup, we build a temporary monmap populated with the IPs
1051  * provided by mount(2).
1052  */
1053 static int build_initial_monmap(struct ceph_mon_client *monc)
1054 {
1055 	struct ceph_options *opt = monc->client->options;
1056 	struct ceph_entity_addr *mon_addr = opt->mon_addr;
1057 	int num_mon = opt->num_mon;
1058 	int i;
1059 
1060 	/* build initial monmap */
1061 	monc->monmap = kzalloc(struct_size(monc->monmap, mon_inst, num_mon),
1062 			       GFP_KERNEL);
1063 	if (!monc->monmap)
1064 		return -ENOMEM;
1065 	for (i = 0; i < num_mon; i++) {
1066 		monc->monmap->mon_inst[i].addr = mon_addr[i];
1067 		monc->monmap->mon_inst[i].addr.nonce = 0;
1068 		monc->monmap->mon_inst[i].name.type =
1069 			CEPH_ENTITY_TYPE_MON;
1070 		monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
1071 	}
1072 	monc->monmap->num_mon = num_mon;
1073 	return 0;
1074 }
1075 
1076 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
1077 {
1078 	int err = 0;
1079 
1080 	dout("init\n");
1081 	memset(monc, 0, sizeof(*monc));
1082 	monc->client = cl;
1083 	monc->monmap = NULL;
1084 	mutex_init(&monc->mutex);
1085 
1086 	err = build_initial_monmap(monc);
1087 	if (err)
1088 		goto out;
1089 
1090 	/* connection */
1091 	/* authentication */
1092 	monc->auth = ceph_auth_init(cl->options->name,
1093 				    cl->options->key);
1094 	if (IS_ERR(monc->auth)) {
1095 		err = PTR_ERR(monc->auth);
1096 		goto out_monmap;
1097 	}
1098 	monc->auth->want_keys =
1099 		CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
1100 		CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
1101 
1102 	/* msgs */
1103 	err = -ENOMEM;
1104 	monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
1105 				     sizeof(struct ceph_mon_subscribe_ack),
1106 				     GFP_KERNEL, true);
1107 	if (!monc->m_subscribe_ack)
1108 		goto out_auth;
1109 
1110 	monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 128,
1111 					 GFP_KERNEL, true);
1112 	if (!monc->m_subscribe)
1113 		goto out_subscribe_ack;
1114 
1115 	monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096,
1116 					  GFP_KERNEL, true);
1117 	if (!monc->m_auth_reply)
1118 		goto out_subscribe;
1119 
1120 	monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_KERNEL, true);
1121 	monc->pending_auth = 0;
1122 	if (!monc->m_auth)
1123 		goto out_auth_reply;
1124 
1125 	ceph_con_init(&monc->con, monc, &mon_con_ops,
1126 		      &monc->client->msgr);
1127 
1128 	monc->cur_mon = -1;
1129 	monc->had_a_connection = false;
1130 	monc->hunt_mult = 1;
1131 
1132 	INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
1133 	monc->generic_request_tree = RB_ROOT;
1134 	monc->last_tid = 0;
1135 
1136 	monc->fs_cluster_id = CEPH_FS_CLUSTER_ID_NONE;
1137 
1138 	return 0;
1139 
1140 out_auth_reply:
1141 	ceph_msg_put(monc->m_auth_reply);
1142 out_subscribe:
1143 	ceph_msg_put(monc->m_subscribe);
1144 out_subscribe_ack:
1145 	ceph_msg_put(monc->m_subscribe_ack);
1146 out_auth:
1147 	ceph_auth_destroy(monc->auth);
1148 out_monmap:
1149 	kfree(monc->monmap);
1150 out:
1151 	return err;
1152 }
1153 EXPORT_SYMBOL(ceph_monc_init);
1154 
1155 void ceph_monc_stop(struct ceph_mon_client *monc)
1156 {
1157 	dout("stop\n");
1158 	cancel_delayed_work_sync(&monc->delayed_work);
1159 
1160 	mutex_lock(&monc->mutex);
1161 	__close_session(monc);
1162 	monc->cur_mon = -1;
1163 	mutex_unlock(&monc->mutex);
1164 
1165 	/*
1166 	 * flush msgr queue before we destroy ourselves to ensure that:
1167 	 *  - any work that references our embedded con is finished.
1168 	 *  - any osd_client or other work that may reference an authorizer
1169 	 *    finishes before we shut down the auth subsystem.
1170 	 */
1171 	ceph_msgr_flush();
1172 
1173 	ceph_auth_destroy(monc->auth);
1174 
1175 	WARN_ON(!RB_EMPTY_ROOT(&monc->generic_request_tree));
1176 
1177 	ceph_msg_put(monc->m_auth);
1178 	ceph_msg_put(monc->m_auth_reply);
1179 	ceph_msg_put(monc->m_subscribe);
1180 	ceph_msg_put(monc->m_subscribe_ack);
1181 
1182 	kfree(monc->monmap);
1183 }
1184 EXPORT_SYMBOL(ceph_monc_stop);
1185 
1186 static void finish_hunting(struct ceph_mon_client *monc)
1187 {
1188 	if (monc->hunting) {
1189 		dout("%s found mon%d\n", __func__, monc->cur_mon);
1190 		monc->hunting = false;
1191 		monc->had_a_connection = true;
1192 		un_backoff(monc);
1193 		__schedule_delayed(monc);
1194 	}
1195 }
1196 
1197 static void handle_auth_reply(struct ceph_mon_client *monc,
1198 			      struct ceph_msg *msg)
1199 {
1200 	int ret;
1201 	int was_auth = 0;
1202 
1203 	mutex_lock(&monc->mutex);
1204 	was_auth = ceph_auth_is_authenticated(monc->auth);
1205 	monc->pending_auth = 0;
1206 	ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
1207 				     msg->front.iov_len,
1208 				     monc->m_auth->front.iov_base,
1209 				     monc->m_auth->front_alloc_len);
1210 	if (ret > 0) {
1211 		__send_prepared_auth_request(monc, ret);
1212 		goto out;
1213 	}
1214 
1215 	finish_hunting(monc);
1216 
1217 	if (ret < 0) {
1218 		monc->client->auth_err = ret;
1219 	} else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
1220 		dout("authenticated, starting session\n");
1221 
1222 		monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
1223 		monc->client->msgr.inst.name.num =
1224 					cpu_to_le64(monc->auth->global_id);
1225 
1226 		__send_subscribe(monc);
1227 		__resend_generic_request(monc);
1228 
1229 		pr_info("mon%d %s session established\n", monc->cur_mon,
1230 			ceph_pr_addr(&monc->con.peer_addr));
1231 	}
1232 
1233 out:
1234 	mutex_unlock(&monc->mutex);
1235 	if (monc->client->auth_err < 0)
1236 		wake_up_all(&monc->client->auth_wq);
1237 }
1238 
1239 static int __validate_auth(struct ceph_mon_client *monc)
1240 {
1241 	int ret;
1242 
1243 	if (monc->pending_auth)
1244 		return 0;
1245 
1246 	ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
1247 			      monc->m_auth->front_alloc_len);
1248 	if (ret <= 0)
1249 		return ret; /* either an error, or no need to authenticate */
1250 	__send_prepared_auth_request(monc, ret);
1251 	return 0;
1252 }
1253 
1254 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
1255 {
1256 	int ret;
1257 
1258 	mutex_lock(&monc->mutex);
1259 	ret = __validate_auth(monc);
1260 	mutex_unlock(&monc->mutex);
1261 	return ret;
1262 }
1263 EXPORT_SYMBOL(ceph_monc_validate_auth);
1264 
1265 /*
1266  * handle incoming message
1267  */
1268 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1269 {
1270 	struct ceph_mon_client *monc = con->private;
1271 	int type = le16_to_cpu(msg->hdr.type);
1272 
1273 	switch (type) {
1274 	case CEPH_MSG_AUTH_REPLY:
1275 		handle_auth_reply(monc, msg);
1276 		break;
1277 
1278 	case CEPH_MSG_MON_SUBSCRIBE_ACK:
1279 		handle_subscribe_ack(monc, msg);
1280 		break;
1281 
1282 	case CEPH_MSG_STATFS_REPLY:
1283 		handle_statfs_reply(monc, msg);
1284 		break;
1285 
1286 	case CEPH_MSG_MON_GET_VERSION_REPLY:
1287 		handle_get_version_reply(monc, msg);
1288 		break;
1289 
1290 	case CEPH_MSG_MON_COMMAND_ACK:
1291 		handle_command_ack(monc, msg);
1292 		break;
1293 
1294 	case CEPH_MSG_MON_MAP:
1295 		ceph_monc_handle_map(monc, msg);
1296 		break;
1297 
1298 	case CEPH_MSG_OSD_MAP:
1299 		ceph_osdc_handle_map(&monc->client->osdc, msg);
1300 		break;
1301 
1302 	default:
1303 		/* can the chained handler handle it? */
1304 		if (monc->client->extra_mon_dispatch &&
1305 		    monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1306 			break;
1307 
1308 		pr_err("received unknown message type %d %s\n", type,
1309 		       ceph_msg_type_name(type));
1310 	}
1311 	ceph_msg_put(msg);
1312 }
1313 
1314 /*
1315  * Allocate memory for incoming message
1316  */
1317 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1318 				      struct ceph_msg_header *hdr,
1319 				      int *skip)
1320 {
1321 	struct ceph_mon_client *monc = con->private;
1322 	int type = le16_to_cpu(hdr->type);
1323 	int front_len = le32_to_cpu(hdr->front_len);
1324 	struct ceph_msg *m = NULL;
1325 
1326 	*skip = 0;
1327 
1328 	switch (type) {
1329 	case CEPH_MSG_MON_SUBSCRIBE_ACK:
1330 		m = ceph_msg_get(monc->m_subscribe_ack);
1331 		break;
1332 	case CEPH_MSG_STATFS_REPLY:
1333 	case CEPH_MSG_MON_COMMAND_ACK:
1334 		return get_generic_reply(con, hdr, skip);
1335 	case CEPH_MSG_AUTH_REPLY:
1336 		m = ceph_msg_get(monc->m_auth_reply);
1337 		break;
1338 	case CEPH_MSG_MON_GET_VERSION_REPLY:
1339 		if (le64_to_cpu(hdr->tid) != 0)
1340 			return get_generic_reply(con, hdr, skip);
1341 
1342 		/*
1343 		 * Older OSDs don't set reply tid even if the orignal
1344 		 * request had a non-zero tid.  Work around this weirdness
1345 		 * by allocating a new message.
1346 		 */
1347 		fallthrough;
1348 	case CEPH_MSG_MON_MAP:
1349 	case CEPH_MSG_MDS_MAP:
1350 	case CEPH_MSG_OSD_MAP:
1351 	case CEPH_MSG_FS_MAP_USER:
1352 		m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1353 		if (!m)
1354 			return NULL;	/* ENOMEM--return skip == 0 */
1355 		break;
1356 	}
1357 
1358 	if (!m) {
1359 		pr_info("alloc_msg unknown type %d\n", type);
1360 		*skip = 1;
1361 	} else if (front_len > m->front_alloc_len) {
1362 		pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1363 			front_len, m->front_alloc_len,
1364 			(unsigned int)con->peer_name.type,
1365 			le64_to_cpu(con->peer_name.num));
1366 		ceph_msg_put(m);
1367 		m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1368 	}
1369 
1370 	return m;
1371 }
1372 
1373 /*
1374  * If the monitor connection resets, pick a new monitor and resubmit
1375  * any pending requests.
1376  */
1377 static void mon_fault(struct ceph_connection *con)
1378 {
1379 	struct ceph_mon_client *monc = con->private;
1380 
1381 	mutex_lock(&monc->mutex);
1382 	dout("%s mon%d\n", __func__, monc->cur_mon);
1383 	if (monc->cur_mon >= 0) {
1384 		if (!monc->hunting) {
1385 			dout("%s hunting for new mon\n", __func__);
1386 			reopen_session(monc);
1387 			__schedule_delayed(monc);
1388 		} else {
1389 			dout("%s already hunting\n", __func__);
1390 		}
1391 	}
1392 	mutex_unlock(&monc->mutex);
1393 }
1394 
1395 /*
1396  * We can ignore refcounting on the connection struct, as all references
1397  * will come from the messenger workqueue, which is drained prior to
1398  * mon_client destruction.
1399  */
1400 static struct ceph_connection *con_get(struct ceph_connection *con)
1401 {
1402 	return con;
1403 }
1404 
1405 static void con_put(struct ceph_connection *con)
1406 {
1407 }
1408 
1409 static const struct ceph_connection_operations mon_con_ops = {
1410 	.get = con_get,
1411 	.put = con_put,
1412 	.dispatch = dispatch,
1413 	.fault = mon_fault,
1414 	.alloc_msg = mon_alloc_msg,
1415 };
1416