xref: /linux/net/ceph/mon_client.c (revision ac2dc6d57425ffa9629941d7c9d7c0e51082cb5a)
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 
decode_mon_info(void ** p,void * end,bool msgr2,struct ceph_entity_addr * addr)39 static int decode_mon_info(void **p, void *end, bool msgr2,
40 			   struct ceph_entity_addr *addr)
41 {
42 	void *mon_info_end;
43 	u32 struct_len;
44 	u8 struct_v;
45 	int ret;
46 
47 	ret = ceph_start_decoding(p, end, 1, "mon_info_t", &struct_v,
48 				  &struct_len);
49 	if (ret)
50 		return ret;
51 
52 	mon_info_end = *p + struct_len;
53 	ceph_decode_skip_string(p, end, e_inval);  /* skip mon name */
54 	ret = ceph_decode_entity_addrvec(p, end, msgr2, addr);
55 	if (ret)
56 		return ret;
57 
58 	*p = mon_info_end;
59 	return 0;
60 
61 e_inval:
62 	return -EINVAL;
63 }
64 
65 /*
66  * Decode a monmap blob (e.g., during mount).
67  *
68  * Assume MonMap v3 (i.e. encoding with MONNAMES and MONENC).
69  */
ceph_monmap_decode(void ** p,void * end,bool msgr2)70 static struct ceph_monmap *ceph_monmap_decode(void **p, void *end, bool msgr2)
71 {
72 	struct ceph_monmap *monmap = NULL;
73 	struct ceph_fsid fsid;
74 	u32 struct_len;
75 	u32 blob_len;
76 	u32 num_mon;
77 	u8 struct_v;
78 	u32 epoch;
79 	int ret;
80 	int i;
81 
82 	ceph_decode_32_safe(p, end, blob_len, e_inval);
83 	ceph_decode_need(p, end, blob_len, e_inval);
84 
85 	ret = ceph_start_decoding(p, end, 6, "monmap", &struct_v, &struct_len);
86 	if (ret)
87 		goto fail;
88 
89 	dout("%s struct_v %d\n", __func__, struct_v);
90 	ceph_decode_copy_safe(p, end, &fsid, sizeof(fsid), e_inval);
91 	ceph_decode_32_safe(p, end, epoch, e_inval);
92 	if (struct_v >= 6) {
93 		u32 feat_struct_len;
94 		u8 feat_struct_v;
95 
96 		*p += sizeof(struct ceph_timespec);  /* skip last_changed */
97 		*p += sizeof(struct ceph_timespec);  /* skip created */
98 
99 		ret = ceph_start_decoding(p, end, 1, "mon_feature_t",
100 					  &feat_struct_v, &feat_struct_len);
101 		if (ret)
102 			goto fail;
103 
104 		*p += feat_struct_len;  /* skip persistent_features */
105 
106 		ret = ceph_start_decoding(p, end, 1, "mon_feature_t",
107 					  &feat_struct_v, &feat_struct_len);
108 		if (ret)
109 			goto fail;
110 
111 		*p += feat_struct_len;  /* skip optional_features */
112 	}
113 	ceph_decode_32_safe(p, end, num_mon, e_inval);
114 
115 	dout("%s fsid %pU epoch %u num_mon %u\n", __func__, &fsid, epoch,
116 	     num_mon);
117 	if (num_mon > CEPH_MAX_MON)
118 		goto e_inval;
119 
120 	monmap = kmalloc_flex(*monmap, mon_inst, num_mon, GFP_NOIO);
121 	if (!monmap) {
122 		ret = -ENOMEM;
123 		goto fail;
124 	}
125 	monmap->fsid = fsid;
126 	monmap->epoch = epoch;
127 	monmap->num_mon = num_mon;
128 
129 	/* legacy_mon_addr map or mon_info map */
130 	for (i = 0; i < num_mon; i++) {
131 		struct ceph_entity_inst *inst = &monmap->mon_inst[i];
132 
133 		ceph_decode_skip_string(p, end, e_inval);  /* skip mon name */
134 		inst->name.type = CEPH_ENTITY_TYPE_MON;
135 		inst->name.num = cpu_to_le64(i);
136 
137 		if (struct_v >= 6)
138 			ret = decode_mon_info(p, end, msgr2, &inst->addr);
139 		else
140 			ret = ceph_decode_entity_addr(p, end, &inst->addr);
141 		if (ret)
142 			goto fail;
143 
144 		dout("%s mon%d addr %s\n", __func__, i,
145 		     ceph_pr_addr(&inst->addr));
146 	}
147 
148 	return monmap;
149 
150 e_inval:
151 	ret = -EINVAL;
152 fail:
153 	kfree(monmap);
154 	return ERR_PTR(ret);
155 }
156 
157 /*
158  * return true if *addr is included in the monmap.
159  */
ceph_monmap_contains(struct ceph_monmap * m,struct ceph_entity_addr * addr)160 int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr)
161 {
162 	int i;
163 
164 	for (i = 0; i < m->num_mon; i++) {
165 		if (ceph_addr_equal_no_type(addr, &m->mon_inst[i].addr))
166 			return 1;
167 	}
168 
169 	return 0;
170 }
171 
172 /*
173  * Send an auth request.
174  */
__send_prepared_auth_request(struct ceph_mon_client * monc,int len)175 static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len)
176 {
177 	BUG_ON(len > monc->m_auth->front_alloc_len);
178 
179 	monc->pending_auth = 1;
180 	monc->m_auth->front.iov_len = len;
181 	monc->m_auth->hdr.front_len = cpu_to_le32(len);
182 	ceph_msg_revoke(monc->m_auth);
183 	ceph_msg_get(monc->m_auth);  /* keep our ref */
184 	ceph_con_send(&monc->con, monc->m_auth);
185 }
186 
187 /*
188  * Close monitor session, if any.
189  */
__close_session(struct ceph_mon_client * monc)190 static void __close_session(struct ceph_mon_client *monc)
191 {
192 	dout("__close_session closing mon%d\n", monc->cur_mon);
193 	ceph_msg_revoke(monc->m_auth);
194 	ceph_msg_revoke_incoming(monc->m_auth_reply);
195 	ceph_msg_revoke(monc->m_subscribe);
196 	ceph_msg_revoke_incoming(monc->m_subscribe_ack);
197 	ceph_con_close(&monc->con);
198 
199 	monc->pending_auth = 0;
200 	ceph_auth_reset(monc->auth);
201 }
202 
203 /*
204  * Pick a new monitor at random and set cur_mon.  If we are repicking
205  * (i.e. cur_mon is already set), be sure to pick a different one.
206  */
pick_new_mon(struct ceph_mon_client * monc)207 static void pick_new_mon(struct ceph_mon_client *monc)
208 {
209 	int old_mon = monc->cur_mon;
210 
211 	BUG_ON(monc->monmap->num_mon < 1);
212 
213 	if (monc->monmap->num_mon == 1) {
214 		monc->cur_mon = 0;
215 	} else {
216 		int max = monc->monmap->num_mon;
217 		int o = -1;
218 		int n;
219 
220 		if (monc->cur_mon >= 0) {
221 			if (monc->cur_mon < monc->monmap->num_mon)
222 				o = monc->cur_mon;
223 			if (o >= 0)
224 				max--;
225 		}
226 
227 		n = get_random_u32_below(max);
228 		if (o >= 0 && n >= o)
229 			n++;
230 
231 		monc->cur_mon = n;
232 	}
233 
234 	dout("%s mon%d -> mon%d out of %d mons\n", __func__, old_mon,
235 	     monc->cur_mon, monc->monmap->num_mon);
236 }
237 
238 /*
239  * Open a session with a new monitor.
240  */
__open_session(struct ceph_mon_client * monc)241 static void __open_session(struct ceph_mon_client *monc)
242 {
243 	int ret;
244 
245 	pick_new_mon(monc);
246 
247 	monc->hunting = true;
248 	if (monc->had_a_connection) {
249 		monc->hunt_mult *= CEPH_MONC_HUNT_BACKOFF;
250 		if (monc->hunt_mult > CEPH_MONC_HUNT_MAX_MULT)
251 			monc->hunt_mult = CEPH_MONC_HUNT_MAX_MULT;
252 	}
253 
254 	monc->sub_renew_after = jiffies; /* i.e., expired */
255 	monc->sub_renew_sent = 0;
256 
257 	dout("%s opening mon%d\n", __func__, monc->cur_mon);
258 	ceph_con_open(&monc->con, CEPH_ENTITY_TYPE_MON, monc->cur_mon,
259 		      &monc->monmap->mon_inst[monc->cur_mon].addr);
260 
261 	/*
262 	 * Queue a keepalive to ensure that in case of an early fault
263 	 * the messenger doesn't put us into STANDBY state and instead
264 	 * retries.  This also ensures that our timestamp is valid by
265 	 * the time we finish hunting and delayed_work() checks it.
266 	 */
267 	ceph_con_keepalive(&monc->con);
268 	if (ceph_msgr2(monc->client)) {
269 		monc->pending_auth = 1;
270 		return;
271 	}
272 
273 	/* initiate authentication handshake */
274 	ret = ceph_auth_build_hello(monc->auth,
275 				    monc->m_auth->front.iov_base,
276 				    monc->m_auth->front_alloc_len);
277 	BUG_ON(ret <= 0);
278 	__send_prepared_auth_request(monc, ret);
279 }
280 
reopen_session(struct ceph_mon_client * monc)281 static void reopen_session(struct ceph_mon_client *monc)
282 {
283 	if (!monc->hunting)
284 		pr_info("mon%d %s session lost, hunting for new mon\n",
285 		    monc->cur_mon, ceph_pr_addr(&monc->con.peer_addr));
286 
287 	__close_session(monc);
288 	__open_session(monc);
289 }
290 
ceph_monc_reopen_session(struct ceph_mon_client * monc)291 void ceph_monc_reopen_session(struct ceph_mon_client *monc)
292 {
293 	mutex_lock(&monc->mutex);
294 	reopen_session(monc);
295 	mutex_unlock(&monc->mutex);
296 }
297 
un_backoff(struct ceph_mon_client * monc)298 static void un_backoff(struct ceph_mon_client *monc)
299 {
300 	monc->hunt_mult /= 2; /* reduce by 50% */
301 	if (monc->hunt_mult < 1)
302 		monc->hunt_mult = 1;
303 	dout("%s hunt_mult now %d\n", __func__, monc->hunt_mult);
304 }
305 
306 /*
307  * Reschedule delayed work timer.
308  */
__schedule_delayed(struct ceph_mon_client * monc)309 static void __schedule_delayed(struct ceph_mon_client *monc)
310 {
311 	unsigned long delay;
312 
313 	if (monc->hunting)
314 		delay = CEPH_MONC_HUNT_INTERVAL * monc->hunt_mult;
315 	else
316 		delay = CEPH_MONC_PING_INTERVAL;
317 
318 	dout("__schedule_delayed after %lu\n", delay);
319 	mod_delayed_work(system_percpu_wq, &monc->delayed_work,
320 			 round_jiffies_relative(delay));
321 }
322 
323 const char *ceph_sub_str[] = {
324 	[CEPH_SUB_MONMAP] = "monmap",
325 	[CEPH_SUB_OSDMAP] = "osdmap",
326 	[CEPH_SUB_FSMAP]  = "fsmap.user",
327 	[CEPH_SUB_MDSMAP] = "mdsmap",
328 };
329 
330 /*
331  * Send subscribe request for one or more maps, according to
332  * monc->subs.
333  */
__send_subscribe(struct ceph_mon_client * monc)334 static void __send_subscribe(struct ceph_mon_client *monc)
335 {
336 	struct ceph_msg *msg = monc->m_subscribe;
337 	void *p = msg->front.iov_base;
338 	void *const end = p + msg->front_alloc_len;
339 	int num = 0;
340 	int i;
341 
342 	dout("%s sent %lu\n", __func__, monc->sub_renew_sent);
343 
344 	BUG_ON(monc->cur_mon < 0);
345 
346 	if (!monc->sub_renew_sent)
347 		monc->sub_renew_sent = jiffies | 1; /* never 0 */
348 
349 	msg->hdr.version = cpu_to_le16(2);
350 
351 	for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
352 		if (monc->subs[i].want)
353 			num++;
354 	}
355 	BUG_ON(num < 1); /* monmap sub is always there */
356 	ceph_encode_32(&p, num);
357 	for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
358 		char buf[32];
359 		int len;
360 
361 		if (!monc->subs[i].want)
362 			continue;
363 
364 		len = sprintf(buf, "%s", ceph_sub_str[i]);
365 		if (i == CEPH_SUB_MDSMAP &&
366 		    monc->fs_cluster_id != CEPH_FS_CLUSTER_ID_NONE)
367 			len += sprintf(buf + len, ".%d", monc->fs_cluster_id);
368 
369 		dout("%s %s start %llu flags 0x%x\n", __func__, buf,
370 		     le64_to_cpu(monc->subs[i].item.start),
371 		     monc->subs[i].item.flags);
372 		ceph_encode_string(&p, end, buf, len);
373 		memcpy(p, &monc->subs[i].item, sizeof(monc->subs[i].item));
374 		p += sizeof(monc->subs[i].item);
375 	}
376 
377 	BUG_ON(p > end);
378 	msg->front.iov_len = p - msg->front.iov_base;
379 	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
380 	ceph_msg_revoke(msg);
381 	ceph_con_send(&monc->con, ceph_msg_get(msg));
382 }
383 
handle_subscribe_ack(struct ceph_mon_client * monc,struct ceph_msg * msg)384 static void handle_subscribe_ack(struct ceph_mon_client *monc,
385 				 struct ceph_msg *msg)
386 {
387 	unsigned int seconds;
388 	struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
389 
390 	if (msg->front.iov_len < sizeof(*h))
391 		goto bad;
392 	seconds = le32_to_cpu(h->duration);
393 
394 	mutex_lock(&monc->mutex);
395 	if (monc->sub_renew_sent) {
396 		/*
397 		 * This is only needed for legacy (infernalis or older)
398 		 * MONs -- see delayed_work().
399 		 */
400 		monc->sub_renew_after = monc->sub_renew_sent +
401 					    (seconds >> 1) * HZ - 1;
402 		dout("%s sent %lu duration %d renew after %lu\n", __func__,
403 		     monc->sub_renew_sent, seconds, monc->sub_renew_after);
404 		monc->sub_renew_sent = 0;
405 	} else {
406 		dout("%s sent %lu renew after %lu, ignoring\n", __func__,
407 		     monc->sub_renew_sent, monc->sub_renew_after);
408 	}
409 	mutex_unlock(&monc->mutex);
410 	return;
411 bad:
412 	pr_err("got corrupt subscribe-ack msg\n");
413 	ceph_msg_dump(msg);
414 }
415 
416 /*
417  * Register interest in a map
418  *
419  * @sub: one of CEPH_SUB_*
420  * @epoch: X for "every map since X", or 0 for "just the latest"
421  */
__ceph_monc_want_map(struct ceph_mon_client * monc,int sub,u32 epoch,bool continuous)422 static bool __ceph_monc_want_map(struct ceph_mon_client *monc, int sub,
423 				 u32 epoch, bool continuous)
424 {
425 	__le64 start = cpu_to_le64(epoch);
426 	u8 flags = !continuous ? CEPH_SUBSCRIBE_ONETIME : 0;
427 
428 	dout("%s %s epoch %u continuous %d\n", __func__, ceph_sub_str[sub],
429 	     epoch, continuous);
430 
431 	if (monc->subs[sub].want &&
432 	    monc->subs[sub].item.start == start &&
433 	    monc->subs[sub].item.flags == flags)
434 		return false;
435 
436 	monc->subs[sub].item.start = start;
437 	monc->subs[sub].item.flags = flags;
438 	monc->subs[sub].want = true;
439 
440 	return true;
441 }
442 
ceph_monc_want_map(struct ceph_mon_client * monc,int sub,u32 epoch,bool continuous)443 bool ceph_monc_want_map(struct ceph_mon_client *monc, int sub, u32 epoch,
444 			bool continuous)
445 {
446 	bool need_request;
447 
448 	mutex_lock(&monc->mutex);
449 	need_request = __ceph_monc_want_map(monc, sub, epoch, continuous);
450 	mutex_unlock(&monc->mutex);
451 
452 	return need_request;
453 }
454 EXPORT_SYMBOL(ceph_monc_want_map);
455 
456 /*
457  * Keep track of which maps we have
458  *
459  * @sub: one of CEPH_SUB_*
460  */
__ceph_monc_got_map(struct ceph_mon_client * monc,int sub,u32 epoch)461 static void __ceph_monc_got_map(struct ceph_mon_client *monc, int sub,
462 				u32 epoch)
463 {
464 	dout("%s %s epoch %u\n", __func__, ceph_sub_str[sub], epoch);
465 
466 	if (monc->subs[sub].want) {
467 		if (monc->subs[sub].item.flags & CEPH_SUBSCRIBE_ONETIME)
468 			monc->subs[sub].want = false;
469 		else
470 			monc->subs[sub].item.start = cpu_to_le64(epoch + 1);
471 	}
472 
473 	monc->subs[sub].have = epoch;
474 }
475 
ceph_monc_got_map(struct ceph_mon_client * monc,int sub,u32 epoch)476 void ceph_monc_got_map(struct ceph_mon_client *monc, int sub, u32 epoch)
477 {
478 	mutex_lock(&monc->mutex);
479 	__ceph_monc_got_map(monc, sub, epoch);
480 	mutex_unlock(&monc->mutex);
481 }
482 EXPORT_SYMBOL(ceph_monc_got_map);
483 
ceph_monc_renew_subs(struct ceph_mon_client * monc)484 void ceph_monc_renew_subs(struct ceph_mon_client *monc)
485 {
486 	mutex_lock(&monc->mutex);
487 	__send_subscribe(monc);
488 	mutex_unlock(&monc->mutex);
489 }
490 EXPORT_SYMBOL(ceph_monc_renew_subs);
491 
492 /*
493  * Wait for an osdmap with a given epoch.
494  *
495  * @epoch: epoch to wait for
496  * @timeout: in jiffies, 0 means "wait forever"
497  */
ceph_monc_wait_osdmap(struct ceph_mon_client * monc,u32 epoch,unsigned long timeout)498 int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
499 			  unsigned long timeout)
500 {
501 	unsigned long started = jiffies;
502 	long ret;
503 
504 	mutex_lock(&monc->mutex);
505 	while (monc->subs[CEPH_SUB_OSDMAP].have < epoch) {
506 		mutex_unlock(&monc->mutex);
507 
508 		if (timeout && time_after_eq(jiffies, started + timeout))
509 			return -ETIMEDOUT;
510 
511 		ret = wait_event_interruptible_timeout(monc->client->auth_wq,
512 				     monc->subs[CEPH_SUB_OSDMAP].have >= epoch,
513 				     ceph_timeout_jiffies(timeout));
514 		if (ret < 0)
515 			return ret;
516 
517 		mutex_lock(&monc->mutex);
518 	}
519 
520 	mutex_unlock(&monc->mutex);
521 	return 0;
522 }
523 EXPORT_SYMBOL(ceph_monc_wait_osdmap);
524 
525 /*
526  * Open a session with a random monitor.  Request monmap and osdmap,
527  * which are waited upon in __ceph_open_session().
528  */
ceph_monc_open_session(struct ceph_mon_client * monc)529 int ceph_monc_open_session(struct ceph_mon_client *monc)
530 {
531 	mutex_lock(&monc->mutex);
532 	__ceph_monc_want_map(monc, CEPH_SUB_MONMAP, 0, true);
533 	__ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, 0, false);
534 	__open_session(monc);
535 	__schedule_delayed(monc);
536 	mutex_unlock(&monc->mutex);
537 	return 0;
538 }
539 EXPORT_SYMBOL(ceph_monc_open_session);
540 
ceph_monc_handle_map(struct ceph_mon_client * monc,struct ceph_msg * msg)541 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
542 				 struct ceph_msg *msg)
543 {
544 	struct ceph_client *client = monc->client;
545 	struct ceph_monmap *monmap;
546 	void *p, *end;
547 
548 	mutex_lock(&monc->mutex);
549 
550 	dout("handle_monmap\n");
551 	p = msg->front.iov_base;
552 	end = p + msg->front.iov_len;
553 
554 	monmap = ceph_monmap_decode(&p, end, ceph_msgr2(client));
555 	if (IS_ERR(monmap)) {
556 		pr_err("problem decoding monmap, %d\n",
557 		       (int)PTR_ERR(monmap));
558 		ceph_msg_dump(msg);
559 		goto out;
560 	}
561 
562 	if (ceph_check_fsid(client, &monmap->fsid) < 0) {
563 		kfree(monmap);
564 		goto out;
565 	}
566 
567 	kfree(monc->monmap);
568 	monc->monmap = monmap;
569 
570 	__ceph_monc_got_map(monc, CEPH_SUB_MONMAP, monc->monmap->epoch);
571 	client->have_fsid = true;
572 
573 out:
574 	mutex_unlock(&monc->mutex);
575 	wake_up_all(&client->auth_wq);
576 }
577 
578 /*
579  * generic requests (currently statfs, mon_get_version)
580  */
DEFINE_RB_FUNCS(generic_request,struct ceph_mon_generic_request,tid,node)581 DEFINE_RB_FUNCS(generic_request, struct ceph_mon_generic_request, tid, node)
582 
583 static void release_generic_request(struct kref *kref)
584 {
585 	struct ceph_mon_generic_request *req =
586 		container_of(kref, struct ceph_mon_generic_request, kref);
587 
588 	dout("%s greq %p request %p reply %p\n", __func__, req, req->request,
589 	     req->reply);
590 	WARN_ON(!RB_EMPTY_NODE(&req->node));
591 
592 	if (req->reply)
593 		ceph_msg_put(req->reply);
594 	if (req->request)
595 		ceph_msg_put(req->request);
596 
597 	kfree(req);
598 }
599 
put_generic_request(struct ceph_mon_generic_request * req)600 static void put_generic_request(struct ceph_mon_generic_request *req)
601 {
602 	if (req)
603 		kref_put(&req->kref, release_generic_request);
604 }
605 
get_generic_request(struct ceph_mon_generic_request * req)606 static void get_generic_request(struct ceph_mon_generic_request *req)
607 {
608 	kref_get(&req->kref);
609 }
610 
611 static struct ceph_mon_generic_request *
alloc_generic_request(struct ceph_mon_client * monc,gfp_t gfp)612 alloc_generic_request(struct ceph_mon_client *monc, gfp_t gfp)
613 {
614 	struct ceph_mon_generic_request *req;
615 
616 	req = kzalloc_obj(*req, gfp);
617 	if (!req)
618 		return NULL;
619 
620 	req->monc = monc;
621 	kref_init(&req->kref);
622 	RB_CLEAR_NODE(&req->node);
623 	init_completion(&req->completion);
624 
625 	dout("%s greq %p\n", __func__, req);
626 	return req;
627 }
628 
register_generic_request(struct ceph_mon_generic_request * req)629 static void register_generic_request(struct ceph_mon_generic_request *req)
630 {
631 	struct ceph_mon_client *monc = req->monc;
632 
633 	WARN_ON(req->tid);
634 
635 	get_generic_request(req);
636 	req->tid = ++monc->last_tid;
637 	insert_generic_request(&monc->generic_request_tree, req);
638 }
639 
send_generic_request(struct ceph_mon_client * monc,struct ceph_mon_generic_request * req)640 static void send_generic_request(struct ceph_mon_client *monc,
641 				 struct ceph_mon_generic_request *req)
642 {
643 	WARN_ON(!req->tid);
644 
645 	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
646 	req->request->hdr.tid = cpu_to_le64(req->tid);
647 	ceph_con_send(&monc->con, ceph_msg_get(req->request));
648 }
649 
__finish_generic_request(struct ceph_mon_generic_request * req)650 static void __finish_generic_request(struct ceph_mon_generic_request *req)
651 {
652 	struct ceph_mon_client *monc = req->monc;
653 
654 	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
655 	erase_generic_request(&monc->generic_request_tree, req);
656 
657 	ceph_msg_revoke(req->request);
658 	ceph_msg_revoke_incoming(req->reply);
659 }
660 
finish_generic_request(struct ceph_mon_generic_request * req)661 static void finish_generic_request(struct ceph_mon_generic_request *req)
662 {
663 	__finish_generic_request(req);
664 	put_generic_request(req);
665 }
666 
complete_generic_request(struct ceph_mon_generic_request * req)667 static void complete_generic_request(struct ceph_mon_generic_request *req)
668 {
669 	if (req->complete_cb)
670 		req->complete_cb(req);
671 	else
672 		complete_all(&req->completion);
673 	put_generic_request(req);
674 }
675 
cancel_generic_request(struct ceph_mon_generic_request * req)676 static void cancel_generic_request(struct ceph_mon_generic_request *req)
677 {
678 	struct ceph_mon_client *monc = req->monc;
679 	struct ceph_mon_generic_request *lookup_req;
680 
681 	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
682 
683 	mutex_lock(&monc->mutex);
684 	lookup_req = lookup_generic_request(&monc->generic_request_tree,
685 					    req->tid);
686 	if (lookup_req) {
687 		WARN_ON(lookup_req != req);
688 		finish_generic_request(req);
689 	}
690 
691 	mutex_unlock(&monc->mutex);
692 }
693 
wait_generic_request(struct ceph_mon_generic_request * req)694 static int wait_generic_request(struct ceph_mon_generic_request *req)
695 {
696 	int ret;
697 
698 	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
699 	ret = wait_for_completion_interruptible(&req->completion);
700 	if (ret)
701 		cancel_generic_request(req);
702 	else
703 		ret = req->result; /* completed */
704 
705 	return ret;
706 }
707 
get_generic_reply(struct ceph_connection * con,struct ceph_msg_header * hdr,int * skip)708 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
709 					 struct ceph_msg_header *hdr,
710 					 int *skip)
711 {
712 	struct ceph_mon_client *monc = con->private;
713 	struct ceph_mon_generic_request *req;
714 	u64 tid = le64_to_cpu(hdr->tid);
715 	struct ceph_msg *m;
716 
717 	mutex_lock(&monc->mutex);
718 	req = lookup_generic_request(&monc->generic_request_tree, tid);
719 	if (!req) {
720 		dout("get_generic_reply %lld dne\n", tid);
721 		*skip = 1;
722 		m = NULL;
723 	} else {
724 		dout("get_generic_reply %lld got %p\n", tid, req->reply);
725 		*skip = 0;
726 		m = ceph_msg_get(req->reply);
727 		/*
728 		 * we don't need to track the connection reading into
729 		 * this reply because we only have one open connection
730 		 * at a time, ever.
731 		 */
732 	}
733 	mutex_unlock(&monc->mutex);
734 	return m;
735 }
736 
737 /*
738  * statfs
739  */
handle_statfs_reply(struct ceph_mon_client * monc,struct ceph_msg * msg)740 static void handle_statfs_reply(struct ceph_mon_client *monc,
741 				struct ceph_msg *msg)
742 {
743 	struct ceph_mon_generic_request *req;
744 	struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
745 	u64 tid = le64_to_cpu(msg->hdr.tid);
746 
747 	dout("%s msg %p tid %llu\n", __func__, msg, tid);
748 
749 	if (msg->front.iov_len != sizeof(*reply))
750 		goto bad;
751 
752 	mutex_lock(&monc->mutex);
753 	req = lookup_generic_request(&monc->generic_request_tree, tid);
754 	if (!req) {
755 		mutex_unlock(&monc->mutex);
756 		return;
757 	}
758 
759 	req->result = 0;
760 	*req->u.st = reply->st; /* struct */
761 	__finish_generic_request(req);
762 	mutex_unlock(&monc->mutex);
763 
764 	complete_generic_request(req);
765 	return;
766 
767 bad:
768 	pr_err("corrupt statfs reply, tid %llu\n", tid);
769 	ceph_msg_dump(msg);
770 }
771 
772 /*
773  * Do a synchronous statfs().
774  */
ceph_monc_do_statfs(struct ceph_mon_client * monc,u64 data_pool,struct ceph_statfs * buf)775 int ceph_monc_do_statfs(struct ceph_mon_client *monc, u64 data_pool,
776 			struct ceph_statfs *buf)
777 {
778 	struct ceph_mon_generic_request *req;
779 	struct ceph_mon_statfs *h;
780 	int ret = -ENOMEM;
781 
782 	req = alloc_generic_request(monc, GFP_NOFS);
783 	if (!req)
784 		goto out;
785 
786 	req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
787 				    true);
788 	if (!req->request)
789 		goto out;
790 
791 	req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 64, GFP_NOFS, true);
792 	if (!req->reply)
793 		goto out;
794 
795 	req->u.st = buf;
796 	req->request->hdr.version = cpu_to_le16(2);
797 
798 	mutex_lock(&monc->mutex);
799 	register_generic_request(req);
800 	/* fill out request */
801 	h = req->request->front.iov_base;
802 	h->monhdr.have_version = 0;
803 	h->monhdr.session_mon = cpu_to_le16(-1);
804 	h->monhdr.session_mon_tid = 0;
805 	h->fsid = monc->monmap->fsid;
806 	h->contains_data_pool = (data_pool != CEPH_NOPOOL);
807 	h->data_pool = cpu_to_le64(data_pool);
808 	send_generic_request(monc, req);
809 	mutex_unlock(&monc->mutex);
810 
811 	ret = wait_generic_request(req);
812 out:
813 	put_generic_request(req);
814 	return ret;
815 }
816 EXPORT_SYMBOL(ceph_monc_do_statfs);
817 
handle_get_version_reply(struct ceph_mon_client * monc,struct ceph_msg * msg)818 static void handle_get_version_reply(struct ceph_mon_client *monc,
819 				     struct ceph_msg *msg)
820 {
821 	struct ceph_mon_generic_request *req;
822 	u64 tid = le64_to_cpu(msg->hdr.tid);
823 	void *p = msg->front.iov_base;
824 	void *end = p + msg->front_alloc_len;
825 	u64 handle;
826 
827 	dout("%s msg %p tid %llu\n", __func__, msg, tid);
828 
829 	ceph_decode_need(&p, end, 2*sizeof(u64), bad);
830 	handle = ceph_decode_64(&p);
831 	if (tid != 0 && tid != handle)
832 		goto bad;
833 
834 	mutex_lock(&monc->mutex);
835 	req = lookup_generic_request(&monc->generic_request_tree, handle);
836 	if (!req) {
837 		mutex_unlock(&monc->mutex);
838 		return;
839 	}
840 
841 	req->result = 0;
842 	req->u.newest = ceph_decode_64(&p);
843 	__finish_generic_request(req);
844 	mutex_unlock(&monc->mutex);
845 
846 	complete_generic_request(req);
847 	return;
848 
849 bad:
850 	pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
851 	ceph_msg_dump(msg);
852 }
853 
854 static struct ceph_mon_generic_request *
__ceph_monc_get_version(struct ceph_mon_client * monc,const char * what,ceph_monc_callback_t cb,u64 private_data)855 __ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
856 			ceph_monc_callback_t cb, u64 private_data)
857 {
858 	struct ceph_mon_generic_request *req;
859 
860 	req = alloc_generic_request(monc, GFP_NOIO);
861 	if (!req)
862 		goto err_put_req;
863 
864 	req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
865 				    sizeof(u64) + sizeof(u32) + strlen(what),
866 				    GFP_NOIO, true);
867 	if (!req->request)
868 		goto err_put_req;
869 
870 	req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 32, GFP_NOIO,
871 				  true);
872 	if (!req->reply)
873 		goto err_put_req;
874 
875 	req->complete_cb = cb;
876 	req->private_data = private_data;
877 
878 	mutex_lock(&monc->mutex);
879 	register_generic_request(req);
880 	{
881 		void *p = req->request->front.iov_base;
882 		void *const end = p + req->request->front_alloc_len;
883 
884 		ceph_encode_64(&p, req->tid); /* handle */
885 		ceph_encode_string(&p, end, what, strlen(what));
886 		WARN_ON(p != end);
887 	}
888 	send_generic_request(monc, req);
889 	mutex_unlock(&monc->mutex);
890 
891 	return req;
892 
893 err_put_req:
894 	put_generic_request(req);
895 	return ERR_PTR(-ENOMEM);
896 }
897 
898 /*
899  * Send MMonGetVersion and wait for the reply.
900  *
901  * @what: one of "mdsmap", "osdmap" or "monmap"
902  */
ceph_monc_get_version(struct ceph_mon_client * monc,const char * what,u64 * newest)903 int ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
904 			  u64 *newest)
905 {
906 	struct ceph_mon_generic_request *req;
907 	int ret;
908 
909 	req = __ceph_monc_get_version(monc, what, NULL, 0);
910 	if (IS_ERR(req))
911 		return PTR_ERR(req);
912 
913 	ret = wait_generic_request(req);
914 	if (!ret)
915 		*newest = req->u.newest;
916 
917 	put_generic_request(req);
918 	return ret;
919 }
920 EXPORT_SYMBOL(ceph_monc_get_version);
921 
922 /*
923  * Send MMonGetVersion,
924  *
925  * @what: one of "mdsmap", "osdmap" or "monmap"
926  */
ceph_monc_get_version_async(struct ceph_mon_client * monc,const char * what,ceph_monc_callback_t cb,u64 private_data)927 int ceph_monc_get_version_async(struct ceph_mon_client *monc, const char *what,
928 				ceph_monc_callback_t cb, u64 private_data)
929 {
930 	struct ceph_mon_generic_request *req;
931 
932 	req = __ceph_monc_get_version(monc, what, cb, private_data);
933 	if (IS_ERR(req))
934 		return PTR_ERR(req);
935 
936 	put_generic_request(req);
937 	return 0;
938 }
939 EXPORT_SYMBOL(ceph_monc_get_version_async);
940 
handle_command_ack(struct ceph_mon_client * monc,struct ceph_msg * msg)941 static void handle_command_ack(struct ceph_mon_client *monc,
942 			       struct ceph_msg *msg)
943 {
944 	struct ceph_mon_generic_request *req;
945 	void *p = msg->front.iov_base;
946 	void *const end = p + msg->front_alloc_len;
947 	u64 tid = le64_to_cpu(msg->hdr.tid);
948 
949 	dout("%s msg %p tid %llu\n", __func__, msg, tid);
950 
951 	ceph_decode_need(&p, end, sizeof(struct ceph_mon_request_header) +
952 							    sizeof(u32), bad);
953 	p += sizeof(struct ceph_mon_request_header);
954 
955 	mutex_lock(&monc->mutex);
956 	req = lookup_generic_request(&monc->generic_request_tree, tid);
957 	if (!req) {
958 		mutex_unlock(&monc->mutex);
959 		return;
960 	}
961 
962 	req->result = ceph_decode_32(&p);
963 	__finish_generic_request(req);
964 	mutex_unlock(&monc->mutex);
965 
966 	complete_generic_request(req);
967 	return;
968 
969 bad:
970 	pr_err("corrupt mon_command ack, tid %llu\n", tid);
971 	ceph_msg_dump(msg);
972 }
973 
974 static __printf(2, 0)
do_mon_command_vargs(struct ceph_mon_client * monc,const char * fmt,va_list ap)975 int do_mon_command_vargs(struct ceph_mon_client *monc, const char *fmt,
976 			 va_list ap)
977 {
978 	struct ceph_mon_generic_request *req;
979 	struct ceph_mon_command *h;
980 	int ret = -ENOMEM;
981 	int len;
982 
983 	req = alloc_generic_request(monc, GFP_NOIO);
984 	if (!req)
985 		goto out;
986 
987 	req->request = ceph_msg_new(CEPH_MSG_MON_COMMAND, 256, GFP_NOIO, true);
988 	if (!req->request)
989 		goto out;
990 
991 	req->reply = ceph_msg_new(CEPH_MSG_MON_COMMAND_ACK, 512, GFP_NOIO,
992 				  true);
993 	if (!req->reply)
994 		goto out;
995 
996 	mutex_lock(&monc->mutex);
997 	register_generic_request(req);
998 	h = req->request->front.iov_base;
999 	h->monhdr.have_version = 0;
1000 	h->monhdr.session_mon = cpu_to_le16(-1);
1001 	h->monhdr.session_mon_tid = 0;
1002 	h->fsid = monc->monmap->fsid;
1003 	h->num_strs = cpu_to_le32(1);
1004 	len = vsprintf(h->str, fmt, ap);
1005 	h->str_len = cpu_to_le32(len);
1006 	send_generic_request(monc, req);
1007 	mutex_unlock(&monc->mutex);
1008 
1009 	ret = wait_generic_request(req);
1010 out:
1011 	put_generic_request(req);
1012 	return ret;
1013 }
1014 
1015 static __printf(2, 3)
do_mon_command(struct ceph_mon_client * monc,const char * fmt,...)1016 int do_mon_command(struct ceph_mon_client *monc, const char *fmt, ...)
1017 {
1018 	va_list ap;
1019 	int ret;
1020 
1021 	va_start(ap, fmt);
1022 	ret = do_mon_command_vargs(monc, fmt, ap);
1023 	va_end(ap);
1024 	return ret;
1025 }
1026 
ceph_monc_blocklist_add(struct ceph_mon_client * monc,struct ceph_entity_addr * client_addr)1027 int ceph_monc_blocklist_add(struct ceph_mon_client *monc,
1028 			    struct ceph_entity_addr *client_addr)
1029 {
1030 	int ret;
1031 
1032 	ret = do_mon_command(monc,
1033 			     "{ \"prefix\": \"osd blocklist\", \
1034 				\"blocklistop\": \"add\", \
1035 				\"addr\": \"%pISpc/%u\" }",
1036 			     &client_addr->in_addr,
1037 			     le32_to_cpu(client_addr->nonce));
1038 	if (ret == -EINVAL) {
1039 		/*
1040 		 * The monitor returns EINVAL on an unrecognized command.
1041 		 * Try the legacy command -- it is exactly the same except
1042 		 * for the name.
1043 		 */
1044 		ret = do_mon_command(monc,
1045 				     "{ \"prefix\": \"osd blacklist\", \
1046 					\"blacklistop\": \"add\", \
1047 					\"addr\": \"%pISpc/%u\" }",
1048 				     &client_addr->in_addr,
1049 				     le32_to_cpu(client_addr->nonce));
1050 	}
1051 	if (ret)
1052 		return ret;
1053 
1054 	/*
1055 	 * Make sure we have the osdmap that includes the blocklist
1056 	 * entry.  This is needed to ensure that the OSDs pick up the
1057 	 * new blocklist before processing any future requests from
1058 	 * this client.
1059 	 */
1060 	return ceph_wait_for_latest_osdmap(monc->client, 0);
1061 }
1062 EXPORT_SYMBOL(ceph_monc_blocklist_add);
1063 
1064 /*
1065  * Resend pending generic requests.
1066  */
__resend_generic_request(struct ceph_mon_client * monc)1067 static void __resend_generic_request(struct ceph_mon_client *monc)
1068 {
1069 	struct ceph_mon_generic_request *req;
1070 	struct rb_node *p;
1071 
1072 	for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
1073 		req = rb_entry(p, struct ceph_mon_generic_request, node);
1074 		ceph_msg_revoke(req->request);
1075 		ceph_msg_revoke_incoming(req->reply);
1076 		ceph_con_send(&monc->con, ceph_msg_get(req->request));
1077 	}
1078 }
1079 
1080 /*
1081  * Delayed work.  If we haven't mounted yet, retry.  Otherwise,
1082  * renew/retry subscription as needed (in case it is timing out, or we
1083  * got an ENOMEM).  And keep the monitor connection alive.
1084  */
delayed_work(struct work_struct * work)1085 static void delayed_work(struct work_struct *work)
1086 {
1087 	struct ceph_mon_client *monc =
1088 		container_of(work, struct ceph_mon_client, delayed_work.work);
1089 
1090 	mutex_lock(&monc->mutex);
1091 	dout("%s mon%d\n", __func__, monc->cur_mon);
1092 	if (monc->cur_mon < 0) {
1093 		goto out;
1094 	}
1095 
1096 	if (monc->hunting) {
1097 		dout("%s continuing hunt\n", __func__);
1098 		reopen_session(monc);
1099 	} else {
1100 		int is_auth = ceph_auth_is_authenticated(monc->auth);
1101 
1102 		dout("%s is_authed %d\n", __func__, is_auth);
1103 		if (ceph_con_keepalive_expired(&monc->con,
1104 					       CEPH_MONC_PING_TIMEOUT)) {
1105 			dout("monc keepalive timeout\n");
1106 			is_auth = 0;
1107 			reopen_session(monc);
1108 		}
1109 
1110 		if (!monc->hunting) {
1111 			ceph_con_keepalive(&monc->con);
1112 			__validate_auth(monc);
1113 			un_backoff(monc);
1114 		}
1115 
1116 		if (is_auth &&
1117 		    !(monc->con.peer_features & CEPH_FEATURE_MON_STATEFUL_SUB)) {
1118 			unsigned long now = jiffies;
1119 
1120 			dout("%s renew subs? now %lu renew after %lu\n",
1121 			     __func__, now, monc->sub_renew_after);
1122 			if (time_after_eq(now, monc->sub_renew_after))
1123 				__send_subscribe(monc);
1124 		}
1125 	}
1126 	__schedule_delayed(monc);
1127 
1128 out:
1129 	mutex_unlock(&monc->mutex);
1130 }
1131 
1132 /*
1133  * On startup, we build a temporary monmap populated with the IPs
1134  * provided by mount(2).
1135  */
build_initial_monmap(struct ceph_mon_client * monc)1136 static int build_initial_monmap(struct ceph_mon_client *monc)
1137 {
1138 	__le32 my_type = ceph_msgr2(monc->client) ?
1139 		CEPH_ENTITY_ADDR_TYPE_MSGR2 : CEPH_ENTITY_ADDR_TYPE_LEGACY;
1140 	struct ceph_options *opt = monc->client->options;
1141 	int num_mon = opt->num_mon;
1142 	int i;
1143 
1144 	/* build initial monmap */
1145 	monc->monmap = kzalloc_flex(*monc->monmap, mon_inst, num_mon);
1146 	if (!monc->monmap)
1147 		return -ENOMEM;
1148 	monc->monmap->num_mon = num_mon;
1149 
1150 	for (i = 0; i < num_mon; i++) {
1151 		struct ceph_entity_inst *inst = &monc->monmap->mon_inst[i];
1152 
1153 		memcpy(&inst->addr.in_addr, &opt->mon_addr[i].in_addr,
1154 		       sizeof(inst->addr.in_addr));
1155 		inst->addr.type = my_type;
1156 		inst->addr.nonce = 0;
1157 		inst->name.type = CEPH_ENTITY_TYPE_MON;
1158 		inst->name.num = cpu_to_le64(i);
1159 	}
1160 	return 0;
1161 }
1162 
ceph_monc_init(struct ceph_mon_client * monc,struct ceph_client * cl)1163 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
1164 {
1165 	int err;
1166 
1167 	dout("init\n");
1168 	memset(monc, 0, sizeof(*monc));
1169 	monc->client = cl;
1170 	mutex_init(&monc->mutex);
1171 
1172 	err = build_initial_monmap(monc);
1173 	if (err)
1174 		goto out;
1175 
1176 	/* connection */
1177 	/* authentication */
1178 	monc->auth = ceph_auth_init(cl->options->name, cl->options->key,
1179 				    cl->options->con_modes);
1180 	if (IS_ERR(monc->auth)) {
1181 		err = PTR_ERR(monc->auth);
1182 		goto out_monmap;
1183 	}
1184 	monc->auth->want_keys =
1185 		CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
1186 		CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
1187 
1188 	/* msgs */
1189 	err = -ENOMEM;
1190 	monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
1191 				     sizeof(struct ceph_mon_subscribe_ack),
1192 				     GFP_KERNEL, true);
1193 	if (!monc->m_subscribe_ack)
1194 		goto out_auth;
1195 
1196 	monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 128,
1197 					 GFP_KERNEL, true);
1198 	if (!monc->m_subscribe)
1199 		goto out_subscribe_ack;
1200 
1201 	monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096,
1202 					  GFP_KERNEL, true);
1203 	if (!monc->m_auth_reply)
1204 		goto out_subscribe;
1205 
1206 	monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_KERNEL, true);
1207 	monc->pending_auth = 0;
1208 	if (!monc->m_auth)
1209 		goto out_auth_reply;
1210 
1211 	ceph_con_init(&monc->con, monc, &mon_con_ops,
1212 		      &monc->client->msgr);
1213 
1214 	monc->cur_mon = -1;
1215 	monc->had_a_connection = false;
1216 	monc->hunt_mult = 1;
1217 
1218 	INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
1219 	monc->generic_request_tree = RB_ROOT;
1220 	monc->last_tid = 0;
1221 
1222 	monc->fs_cluster_id = CEPH_FS_CLUSTER_ID_NONE;
1223 
1224 	return 0;
1225 
1226 out_auth_reply:
1227 	ceph_msg_put(monc->m_auth_reply);
1228 out_subscribe:
1229 	ceph_msg_put(monc->m_subscribe);
1230 out_subscribe_ack:
1231 	ceph_msg_put(monc->m_subscribe_ack);
1232 out_auth:
1233 	ceph_auth_destroy(monc->auth);
1234 out_monmap:
1235 	kfree(monc->monmap);
1236 out:
1237 	return err;
1238 }
1239 EXPORT_SYMBOL(ceph_monc_init);
1240 
ceph_monc_stop(struct ceph_mon_client * monc)1241 void ceph_monc_stop(struct ceph_mon_client *monc)
1242 {
1243 	dout("stop\n");
1244 
1245 	mutex_lock(&monc->mutex);
1246 	__close_session(monc);
1247 	monc->hunting = false;
1248 	monc->cur_mon = -1;
1249 	mutex_unlock(&monc->mutex);
1250 
1251 	cancel_delayed_work_sync(&monc->delayed_work);
1252 
1253 	/*
1254 	 * flush msgr queue before we destroy ourselves to ensure that:
1255 	 *  - any work that references our embedded con is finished.
1256 	 *  - any osd_client or other work that may reference an authorizer
1257 	 *    finishes before we shut down the auth subsystem.
1258 	 */
1259 	ceph_msgr_flush();
1260 
1261 	ceph_auth_destroy(monc->auth);
1262 
1263 	WARN_ON(!RB_EMPTY_ROOT(&monc->generic_request_tree));
1264 
1265 	ceph_msg_put(monc->m_auth);
1266 	ceph_msg_put(monc->m_auth_reply);
1267 	ceph_msg_put(monc->m_subscribe);
1268 	ceph_msg_put(monc->m_subscribe_ack);
1269 
1270 	kfree(monc->monmap);
1271 }
1272 EXPORT_SYMBOL(ceph_monc_stop);
1273 
finish_hunting(struct ceph_mon_client * monc)1274 static void finish_hunting(struct ceph_mon_client *monc)
1275 {
1276 	if (monc->hunting) {
1277 		dout("%s found mon%d\n", __func__, monc->cur_mon);
1278 		monc->hunting = false;
1279 		monc->had_a_connection = true;
1280 		un_backoff(monc);
1281 		__schedule_delayed(monc);
1282 	}
1283 }
1284 
finish_auth(struct ceph_mon_client * monc,int auth_err,bool was_authed)1285 static void finish_auth(struct ceph_mon_client *monc, int auth_err,
1286 			bool was_authed)
1287 {
1288 	dout("%s auth_err %d was_authed %d\n", __func__, auth_err, was_authed);
1289 	WARN_ON(auth_err > 0);
1290 
1291 	monc->pending_auth = 0;
1292 	if (auth_err) {
1293 		monc->client->auth_err = auth_err;
1294 		wake_up_all(&monc->client->auth_wq);
1295 		return;
1296 	}
1297 
1298 	if (!was_authed && ceph_auth_is_authenticated(monc->auth)) {
1299 		dout("%s authenticated, starting session global_id %llu\n",
1300 		     __func__, monc->auth->global_id);
1301 
1302 		monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
1303 		monc->client->msgr.inst.name.num =
1304 					cpu_to_le64(monc->auth->global_id);
1305 
1306 		__send_subscribe(monc);
1307 		__resend_generic_request(monc);
1308 
1309 		pr_info("mon%d %s session established\n", monc->cur_mon,
1310 			ceph_pr_addr(&monc->con.peer_addr));
1311 	}
1312 }
1313 
handle_auth_reply(struct ceph_mon_client * monc,struct ceph_msg * msg)1314 static void handle_auth_reply(struct ceph_mon_client *monc,
1315 			      struct ceph_msg *msg)
1316 {
1317 	bool was_authed;
1318 	int ret;
1319 
1320 	mutex_lock(&monc->mutex);
1321 	was_authed = ceph_auth_is_authenticated(monc->auth);
1322 	ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
1323 				     msg->front.iov_len,
1324 				     monc->m_auth->front.iov_base,
1325 				     monc->m_auth->front_alloc_len);
1326 	if (ret > 0) {
1327 		__send_prepared_auth_request(monc, ret);
1328 	} else {
1329 		finish_auth(monc, ret, was_authed);
1330 		finish_hunting(monc);
1331 	}
1332 	mutex_unlock(&monc->mutex);
1333 }
1334 
__validate_auth(struct ceph_mon_client * monc)1335 static int __validate_auth(struct ceph_mon_client *monc)
1336 {
1337 	int ret;
1338 
1339 	if (monc->pending_auth)
1340 		return 0;
1341 
1342 	ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
1343 			      monc->m_auth->front_alloc_len);
1344 	if (ret <= 0)
1345 		return ret; /* either an error, or no need to authenticate */
1346 	__send_prepared_auth_request(monc, ret);
1347 	return 0;
1348 }
1349 
ceph_monc_validate_auth(struct ceph_mon_client * monc)1350 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
1351 {
1352 	int ret;
1353 
1354 	mutex_lock(&monc->mutex);
1355 	ret = __validate_auth(monc);
1356 	mutex_unlock(&monc->mutex);
1357 	return ret;
1358 }
1359 EXPORT_SYMBOL(ceph_monc_validate_auth);
1360 
mon_get_auth_request(struct ceph_connection * con,void * buf,int * buf_len,void ** authorizer,int * authorizer_len)1361 static int mon_get_auth_request(struct ceph_connection *con,
1362 				void *buf, int *buf_len,
1363 				void **authorizer, int *authorizer_len)
1364 {
1365 	struct ceph_mon_client *monc = con->private;
1366 	int ret;
1367 
1368 	mutex_lock(&monc->mutex);
1369 	ret = ceph_auth_get_request(monc->auth, buf, *buf_len);
1370 	mutex_unlock(&monc->mutex);
1371 	if (ret < 0)
1372 		return ret;
1373 
1374 	*buf_len = ret;
1375 	*authorizer = NULL;
1376 	*authorizer_len = 0;
1377 	return 0;
1378 }
1379 
mon_handle_auth_reply_more(struct ceph_connection * con,void * reply,int reply_len,void * buf,int * buf_len,void ** authorizer,int * authorizer_len)1380 static int mon_handle_auth_reply_more(struct ceph_connection *con,
1381 				      void *reply, int reply_len,
1382 				      void *buf, int *buf_len,
1383 				      void **authorizer, int *authorizer_len)
1384 {
1385 	struct ceph_mon_client *monc = con->private;
1386 	int ret;
1387 
1388 	mutex_lock(&monc->mutex);
1389 	ret = ceph_auth_handle_reply_more(monc->auth, reply, reply_len,
1390 					  buf, *buf_len);
1391 	mutex_unlock(&monc->mutex);
1392 	if (ret < 0)
1393 		return ret;
1394 
1395 	*buf_len = ret;
1396 	*authorizer = NULL;
1397 	*authorizer_len = 0;
1398 	return 0;
1399 }
1400 
mon_handle_auth_done(struct ceph_connection * con,u64 global_id,void * reply,int reply_len,u8 * session_key,int * session_key_len,u8 * con_secret,int * con_secret_len)1401 static int mon_handle_auth_done(struct ceph_connection *con,
1402 				u64 global_id, void *reply, int reply_len,
1403 				u8 *session_key, int *session_key_len,
1404 				u8 *con_secret, int *con_secret_len)
1405 {
1406 	struct ceph_mon_client *monc = con->private;
1407 	bool was_authed;
1408 	int ret;
1409 
1410 	mutex_lock(&monc->mutex);
1411 	WARN_ON(!monc->hunting);
1412 	was_authed = ceph_auth_is_authenticated(monc->auth);
1413 	ret = ceph_auth_handle_reply_done(monc->auth, global_id,
1414 					  reply, reply_len,
1415 					  session_key, session_key_len,
1416 					  con_secret, con_secret_len);
1417 	finish_auth(monc, ret, was_authed);
1418 	if (!ret)
1419 		finish_hunting(monc);
1420 	mutex_unlock(&monc->mutex);
1421 	return ret;
1422 }
1423 
mon_handle_auth_bad_method(struct ceph_connection * con,int used_proto,int result,const int * allowed_protos,int proto_cnt,const int * allowed_modes,int mode_cnt)1424 static int mon_handle_auth_bad_method(struct ceph_connection *con,
1425 				      int used_proto, int result,
1426 				      const int *allowed_protos, int proto_cnt,
1427 				      const int *allowed_modes, int mode_cnt)
1428 {
1429 	struct ceph_mon_client *monc = con->private;
1430 	bool was_authed;
1431 
1432 	mutex_lock(&monc->mutex);
1433 	WARN_ON(!monc->hunting);
1434 	was_authed = ceph_auth_is_authenticated(monc->auth);
1435 	ceph_auth_handle_bad_method(monc->auth, used_proto, result,
1436 				    allowed_protos, proto_cnt,
1437 				    allowed_modes, mode_cnt);
1438 	finish_auth(monc, -EACCES, was_authed);
1439 	mutex_unlock(&monc->mutex);
1440 	return 0;
1441 }
1442 
1443 /*
1444  * handle incoming message
1445  */
mon_dispatch(struct ceph_connection * con,struct ceph_msg * msg)1446 static void mon_dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1447 {
1448 	struct ceph_mon_client *monc = con->private;
1449 	int type = le16_to_cpu(msg->hdr.type);
1450 
1451 	switch (type) {
1452 	case CEPH_MSG_AUTH_REPLY:
1453 		handle_auth_reply(monc, msg);
1454 		break;
1455 
1456 	case CEPH_MSG_MON_SUBSCRIBE_ACK:
1457 		handle_subscribe_ack(monc, msg);
1458 		break;
1459 
1460 	case CEPH_MSG_STATFS_REPLY:
1461 		handle_statfs_reply(monc, msg);
1462 		break;
1463 
1464 	case CEPH_MSG_MON_GET_VERSION_REPLY:
1465 		handle_get_version_reply(monc, msg);
1466 		break;
1467 
1468 	case CEPH_MSG_MON_COMMAND_ACK:
1469 		handle_command_ack(monc, msg);
1470 		break;
1471 
1472 	case CEPH_MSG_MON_MAP:
1473 		ceph_monc_handle_map(monc, msg);
1474 		break;
1475 
1476 	case CEPH_MSG_OSD_MAP:
1477 		ceph_osdc_handle_map(&monc->client->osdc, msg);
1478 		break;
1479 
1480 	default:
1481 		/* can the chained handler handle it? */
1482 		if (monc->client->extra_mon_dispatch &&
1483 		    monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1484 			break;
1485 
1486 		pr_err("received unknown message type %d %s\n", type,
1487 		       ceph_msg_type_name(type));
1488 	}
1489 	ceph_msg_put(msg);
1490 }
1491 
1492 /*
1493  * Allocate memory for incoming message
1494  */
mon_alloc_msg(struct ceph_connection * con,struct ceph_msg_header * hdr,int * skip)1495 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1496 				      struct ceph_msg_header *hdr,
1497 				      int *skip)
1498 {
1499 	struct ceph_mon_client *monc = con->private;
1500 	int type = le16_to_cpu(hdr->type);
1501 	int front_len = le32_to_cpu(hdr->front_len);
1502 	struct ceph_msg *m = NULL;
1503 
1504 	*skip = 0;
1505 
1506 	switch (type) {
1507 	case CEPH_MSG_MON_SUBSCRIBE_ACK:
1508 		m = ceph_msg_get(monc->m_subscribe_ack);
1509 		break;
1510 	case CEPH_MSG_STATFS_REPLY:
1511 	case CEPH_MSG_MON_COMMAND_ACK:
1512 		return get_generic_reply(con, hdr, skip);
1513 	case CEPH_MSG_AUTH_REPLY:
1514 		m = ceph_msg_get(monc->m_auth_reply);
1515 		break;
1516 	case CEPH_MSG_MON_GET_VERSION_REPLY:
1517 		if (le64_to_cpu(hdr->tid) != 0)
1518 			return get_generic_reply(con, hdr, skip);
1519 
1520 		/*
1521 		 * Older OSDs don't set reply tid even if the original
1522 		 * request had a non-zero tid.  Work around this weirdness
1523 		 * by allocating a new message.
1524 		 */
1525 		fallthrough;
1526 	case CEPH_MSG_MON_MAP:
1527 	case CEPH_MSG_MDS_MAP:
1528 	case CEPH_MSG_OSD_MAP:
1529 	case CEPH_MSG_FS_MAP_USER:
1530 		m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1531 		if (!m)
1532 			return NULL;	/* ENOMEM--return skip == 0 */
1533 		break;
1534 	}
1535 
1536 	if (!m) {
1537 		pr_info("alloc_msg unknown type %d\n", type);
1538 		*skip = 1;
1539 	} else if (front_len > m->front_alloc_len) {
1540 		pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1541 			front_len, m->front_alloc_len,
1542 			(unsigned int)con->peer_name.type,
1543 			le64_to_cpu(con->peer_name.num));
1544 		ceph_msg_put(m);
1545 		m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1546 	}
1547 
1548 	return m;
1549 }
1550 
1551 /*
1552  * If the monitor connection resets, pick a new monitor and resubmit
1553  * any pending requests.
1554  */
mon_fault(struct ceph_connection * con)1555 static void mon_fault(struct ceph_connection *con)
1556 {
1557 	struct ceph_mon_client *monc = con->private;
1558 
1559 	mutex_lock(&monc->mutex);
1560 	dout("%s mon%d\n", __func__, monc->cur_mon);
1561 	if (monc->cur_mon >= 0) {
1562 		if (!monc->hunting) {
1563 			dout("%s hunting for new mon\n", __func__);
1564 			reopen_session(monc);
1565 			__schedule_delayed(monc);
1566 		} else {
1567 			dout("%s already hunting\n", __func__);
1568 		}
1569 	}
1570 	mutex_unlock(&monc->mutex);
1571 }
1572 
1573 /*
1574  * We can ignore refcounting on the connection struct, as all references
1575  * will come from the messenger workqueue, which is drained prior to
1576  * mon_client destruction.
1577  */
mon_get_con(struct ceph_connection * con)1578 static struct ceph_connection *mon_get_con(struct ceph_connection *con)
1579 {
1580 	return con;
1581 }
1582 
mon_put_con(struct ceph_connection * con)1583 static void mon_put_con(struct ceph_connection *con)
1584 {
1585 }
1586 
1587 static const struct ceph_connection_operations mon_con_ops = {
1588 	.get = mon_get_con,
1589 	.put = mon_put_con,
1590 	.alloc_msg = mon_alloc_msg,
1591 	.dispatch = mon_dispatch,
1592 	.fault = mon_fault,
1593 	.get_auth_request = mon_get_auth_request,
1594 	.handle_auth_reply_more = mon_handle_auth_reply_more,
1595 	.handle_auth_done = mon_handle_auth_done,
1596 	.handle_auth_bad_method = mon_handle_auth_bad_method,
1597 };
1598