xref: /kernel/linux/linux-5.10/net/ceph/mon_client.c (revision 8c2ecf20)
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
35static const struct ceph_connection_operations mon_con_ops;
36
37static int __validate_auth(struct ceph_mon_client *monc);
38
39/*
40 * Decode a monmap blob (e.g., during mount).
41 */
42static 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;
86bad:
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 */
95int 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 (ceph_addr_equal_no_type(addr, &m->mon_inst[i].addr))
101			return 1;
102	}
103
104	return 0;
105}
106
107/*
108 * Send an auth request.
109 */
110static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len)
111{
112	monc->pending_auth = 1;
113	monc->m_auth->front.iov_len = len;
114	monc->m_auth->hdr.front_len = cpu_to_le32(len);
115	ceph_msg_revoke(monc->m_auth);
116	ceph_msg_get(monc->m_auth);  /* keep our ref */
117	ceph_con_send(&monc->con, monc->m_auth);
118}
119
120/*
121 * Close monitor session, if any.
122 */
123static void __close_session(struct ceph_mon_client *monc)
124{
125	dout("__close_session closing mon%d\n", monc->cur_mon);
126	ceph_msg_revoke(monc->m_auth);
127	ceph_msg_revoke_incoming(monc->m_auth_reply);
128	ceph_msg_revoke(monc->m_subscribe);
129	ceph_msg_revoke_incoming(monc->m_subscribe_ack);
130	ceph_con_close(&monc->con);
131
132	monc->pending_auth = 0;
133	ceph_auth_reset(monc->auth);
134}
135
136/*
137 * Pick a new monitor at random and set cur_mon.  If we are repicking
138 * (i.e. cur_mon is already set), be sure to pick a different one.
139 */
140static void pick_new_mon(struct ceph_mon_client *monc)
141{
142	int old_mon = monc->cur_mon;
143
144	BUG_ON(monc->monmap->num_mon < 1);
145
146	if (monc->monmap->num_mon == 1) {
147		monc->cur_mon = 0;
148	} else {
149		int max = monc->monmap->num_mon;
150		int o = -1;
151		int n;
152
153		if (monc->cur_mon >= 0) {
154			if (monc->cur_mon < monc->monmap->num_mon)
155				o = monc->cur_mon;
156			if (o >= 0)
157				max--;
158		}
159
160		n = prandom_u32() % max;
161		if (o >= 0 && n >= o)
162			n++;
163
164		monc->cur_mon = n;
165	}
166
167	dout("%s mon%d -> mon%d out of %d mons\n", __func__, old_mon,
168	     monc->cur_mon, monc->monmap->num_mon);
169}
170
171/*
172 * Open a session with a new monitor.
173 */
174static void __open_session(struct ceph_mon_client *monc)
175{
176	int ret;
177
178	pick_new_mon(monc);
179
180	monc->hunting = true;
181	if (monc->had_a_connection) {
182		monc->hunt_mult *= CEPH_MONC_HUNT_BACKOFF;
183		if (monc->hunt_mult > CEPH_MONC_HUNT_MAX_MULT)
184			monc->hunt_mult = CEPH_MONC_HUNT_MAX_MULT;
185	}
186
187	monc->sub_renew_after = jiffies; /* i.e., expired */
188	monc->sub_renew_sent = 0;
189
190	dout("%s opening mon%d\n", __func__, monc->cur_mon);
191	ceph_con_open(&monc->con, CEPH_ENTITY_TYPE_MON, monc->cur_mon,
192		      &monc->monmap->mon_inst[monc->cur_mon].addr);
193
194	/*
195	 * send an initial keepalive to ensure our timestamp is valid
196	 * by the time we are in an OPENED state
197	 */
198	ceph_con_keepalive(&monc->con);
199
200	/* initiate authentication handshake */
201	ret = ceph_auth_build_hello(monc->auth,
202				    monc->m_auth->front.iov_base,
203				    monc->m_auth->front_alloc_len);
204	BUG_ON(ret <= 0);
205	__send_prepared_auth_request(monc, ret);
206}
207
208static void reopen_session(struct ceph_mon_client *monc)
209{
210	if (!monc->hunting)
211		pr_info("mon%d %s session lost, hunting for new mon\n",
212		    monc->cur_mon, ceph_pr_addr(&monc->con.peer_addr));
213
214	__close_session(monc);
215	__open_session(monc);
216}
217
218void ceph_monc_reopen_session(struct ceph_mon_client *monc)
219{
220	mutex_lock(&monc->mutex);
221	reopen_session(monc);
222	mutex_unlock(&monc->mutex);
223}
224
225static void un_backoff(struct ceph_mon_client *monc)
226{
227	monc->hunt_mult /= 2; /* reduce by 50% */
228	if (monc->hunt_mult < 1)
229		monc->hunt_mult = 1;
230	dout("%s hunt_mult now %d\n", __func__, monc->hunt_mult);
231}
232
233/*
234 * Reschedule delayed work timer.
235 */
236static void __schedule_delayed(struct ceph_mon_client *monc)
237{
238	unsigned long delay;
239
240	if (monc->hunting)
241		delay = CEPH_MONC_HUNT_INTERVAL * monc->hunt_mult;
242	else
243		delay = CEPH_MONC_PING_INTERVAL;
244
245	dout("__schedule_delayed after %lu\n", delay);
246	mod_delayed_work(system_wq, &monc->delayed_work,
247			 round_jiffies_relative(delay));
248}
249
250const char *ceph_sub_str[] = {
251	[CEPH_SUB_MONMAP] = "monmap",
252	[CEPH_SUB_OSDMAP] = "osdmap",
253	[CEPH_SUB_FSMAP]  = "fsmap.user",
254	[CEPH_SUB_MDSMAP] = "mdsmap",
255};
256
257/*
258 * Send subscribe request for one or more maps, according to
259 * monc->subs.
260 */
261static void __send_subscribe(struct ceph_mon_client *monc)
262{
263	struct ceph_msg *msg = monc->m_subscribe;
264	void *p = msg->front.iov_base;
265	void *const end = p + msg->front_alloc_len;
266	int num = 0;
267	int i;
268
269	dout("%s sent %lu\n", __func__, monc->sub_renew_sent);
270
271	BUG_ON(monc->cur_mon < 0);
272
273	if (!monc->sub_renew_sent)
274		monc->sub_renew_sent = jiffies | 1; /* never 0 */
275
276	msg->hdr.version = cpu_to_le16(2);
277
278	for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
279		if (monc->subs[i].want)
280			num++;
281	}
282	BUG_ON(num < 1); /* monmap sub is always there */
283	ceph_encode_32(&p, num);
284	for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
285		char buf[32];
286		int len;
287
288		if (!monc->subs[i].want)
289			continue;
290
291		len = sprintf(buf, "%s", ceph_sub_str[i]);
292		if (i == CEPH_SUB_MDSMAP &&
293		    monc->fs_cluster_id != CEPH_FS_CLUSTER_ID_NONE)
294			len += sprintf(buf + len, ".%d", monc->fs_cluster_id);
295
296		dout("%s %s start %llu flags 0x%x\n", __func__, buf,
297		     le64_to_cpu(monc->subs[i].item.start),
298		     monc->subs[i].item.flags);
299		ceph_encode_string(&p, end, buf, len);
300		memcpy(p, &monc->subs[i].item, sizeof(monc->subs[i].item));
301		p += sizeof(monc->subs[i].item);
302	}
303
304	BUG_ON(p > end);
305	msg->front.iov_len = p - msg->front.iov_base;
306	msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
307	ceph_msg_revoke(msg);
308	ceph_con_send(&monc->con, ceph_msg_get(msg));
309}
310
311static void handle_subscribe_ack(struct ceph_mon_client *monc,
312				 struct ceph_msg *msg)
313{
314	unsigned int seconds;
315	struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
316
317	if (msg->front.iov_len < sizeof(*h))
318		goto bad;
319	seconds = le32_to_cpu(h->duration);
320
321	mutex_lock(&monc->mutex);
322	if (monc->sub_renew_sent) {
323		/*
324		 * This is only needed for legacy (infernalis or older)
325		 * MONs -- see delayed_work().
326		 */
327		monc->sub_renew_after = monc->sub_renew_sent +
328					    (seconds >> 1) * HZ - 1;
329		dout("%s sent %lu duration %d renew after %lu\n", __func__,
330		     monc->sub_renew_sent, seconds, monc->sub_renew_after);
331		monc->sub_renew_sent = 0;
332	} else {
333		dout("%s sent %lu renew after %lu, ignoring\n", __func__,
334		     monc->sub_renew_sent, monc->sub_renew_after);
335	}
336	mutex_unlock(&monc->mutex);
337	return;
338bad:
339	pr_err("got corrupt subscribe-ack msg\n");
340	ceph_msg_dump(msg);
341}
342
343/*
344 * Register interest in a map
345 *
346 * @sub: one of CEPH_SUB_*
347 * @epoch: X for "every map since X", or 0 for "just the latest"
348 */
349static bool __ceph_monc_want_map(struct ceph_mon_client *monc, int sub,
350				 u32 epoch, bool continuous)
351{
352	__le64 start = cpu_to_le64(epoch);
353	u8 flags = !continuous ? CEPH_SUBSCRIBE_ONETIME : 0;
354
355	dout("%s %s epoch %u continuous %d\n", __func__, ceph_sub_str[sub],
356	     epoch, continuous);
357
358	if (monc->subs[sub].want &&
359	    monc->subs[sub].item.start == start &&
360	    monc->subs[sub].item.flags == flags)
361		return false;
362
363	monc->subs[sub].item.start = start;
364	monc->subs[sub].item.flags = flags;
365	monc->subs[sub].want = true;
366
367	return true;
368}
369
370bool ceph_monc_want_map(struct ceph_mon_client *monc, int sub, u32 epoch,
371			bool continuous)
372{
373	bool need_request;
374
375	mutex_lock(&monc->mutex);
376	need_request = __ceph_monc_want_map(monc, sub, epoch, continuous);
377	mutex_unlock(&monc->mutex);
378
379	return need_request;
380}
381EXPORT_SYMBOL(ceph_monc_want_map);
382
383/*
384 * Keep track of which maps we have
385 *
386 * @sub: one of CEPH_SUB_*
387 */
388static void __ceph_monc_got_map(struct ceph_mon_client *monc, int sub,
389				u32 epoch)
390{
391	dout("%s %s epoch %u\n", __func__, ceph_sub_str[sub], epoch);
392
393	if (monc->subs[sub].want) {
394		if (monc->subs[sub].item.flags & CEPH_SUBSCRIBE_ONETIME)
395			monc->subs[sub].want = false;
396		else
397			monc->subs[sub].item.start = cpu_to_le64(epoch + 1);
398	}
399
400	monc->subs[sub].have = epoch;
401}
402
403void ceph_monc_got_map(struct ceph_mon_client *monc, int sub, u32 epoch)
404{
405	mutex_lock(&monc->mutex);
406	__ceph_monc_got_map(monc, sub, epoch);
407	mutex_unlock(&monc->mutex);
408}
409EXPORT_SYMBOL(ceph_monc_got_map);
410
411void ceph_monc_renew_subs(struct ceph_mon_client *monc)
412{
413	mutex_lock(&monc->mutex);
414	__send_subscribe(monc);
415	mutex_unlock(&monc->mutex);
416}
417EXPORT_SYMBOL(ceph_monc_renew_subs);
418
419/*
420 * Wait for an osdmap with a given epoch.
421 *
422 * @epoch: epoch to wait for
423 * @timeout: in jiffies, 0 means "wait forever"
424 */
425int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
426			  unsigned long timeout)
427{
428	unsigned long started = jiffies;
429	long ret;
430
431	mutex_lock(&monc->mutex);
432	while (monc->subs[CEPH_SUB_OSDMAP].have < epoch) {
433		mutex_unlock(&monc->mutex);
434
435		if (timeout && time_after_eq(jiffies, started + timeout))
436			return -ETIMEDOUT;
437
438		ret = wait_event_interruptible_timeout(monc->client->auth_wq,
439				     monc->subs[CEPH_SUB_OSDMAP].have >= epoch,
440				     ceph_timeout_jiffies(timeout));
441		if (ret < 0)
442			return ret;
443
444		mutex_lock(&monc->mutex);
445	}
446
447	mutex_unlock(&monc->mutex);
448	return 0;
449}
450EXPORT_SYMBOL(ceph_monc_wait_osdmap);
451
452/*
453 * Open a session with a random monitor.  Request monmap and osdmap,
454 * which are waited upon in __ceph_open_session().
455 */
456int ceph_monc_open_session(struct ceph_mon_client *monc)
457{
458	mutex_lock(&monc->mutex);
459	__ceph_monc_want_map(monc, CEPH_SUB_MONMAP, 0, true);
460	__ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, 0, false);
461	__open_session(monc);
462	__schedule_delayed(monc);
463	mutex_unlock(&monc->mutex);
464	return 0;
465}
466EXPORT_SYMBOL(ceph_monc_open_session);
467
468static void ceph_monc_handle_map(struct ceph_mon_client *monc,
469				 struct ceph_msg *msg)
470{
471	struct ceph_client *client = monc->client;
472	struct ceph_monmap *monmap;
473	void *p, *end;
474
475	mutex_lock(&monc->mutex);
476
477	dout("handle_monmap\n");
478	p = msg->front.iov_base;
479	end = p + msg->front.iov_len;
480
481	monmap = ceph_monmap_decode(p, end);
482	if (IS_ERR(monmap)) {
483		pr_err("problem decoding monmap, %d\n",
484		       (int)PTR_ERR(monmap));
485		ceph_msg_dump(msg);
486		goto out;
487	}
488
489	if (ceph_check_fsid(client, &monmap->fsid) < 0) {
490		kfree(monmap);
491		goto out;
492	}
493
494	kfree(monc->monmap);
495	monc->monmap = monmap;
496
497	__ceph_monc_got_map(monc, CEPH_SUB_MONMAP, monc->monmap->epoch);
498	client->have_fsid = true;
499
500out:
501	mutex_unlock(&monc->mutex);
502	wake_up_all(&client->auth_wq);
503}
504
505/*
506 * generic requests (currently statfs, mon_get_version)
507 */
508DEFINE_RB_FUNCS(generic_request, struct ceph_mon_generic_request, tid, node)
509
510static void release_generic_request(struct kref *kref)
511{
512	struct ceph_mon_generic_request *req =
513		container_of(kref, struct ceph_mon_generic_request, kref);
514
515	dout("%s greq %p request %p reply %p\n", __func__, req, req->request,
516	     req->reply);
517	WARN_ON(!RB_EMPTY_NODE(&req->node));
518
519	if (req->reply)
520		ceph_msg_put(req->reply);
521	if (req->request)
522		ceph_msg_put(req->request);
523
524	kfree(req);
525}
526
527static void put_generic_request(struct ceph_mon_generic_request *req)
528{
529	if (req)
530		kref_put(&req->kref, release_generic_request);
531}
532
533static void get_generic_request(struct ceph_mon_generic_request *req)
534{
535	kref_get(&req->kref);
536}
537
538static struct ceph_mon_generic_request *
539alloc_generic_request(struct ceph_mon_client *monc, gfp_t gfp)
540{
541	struct ceph_mon_generic_request *req;
542
543	req = kzalloc(sizeof(*req), gfp);
544	if (!req)
545		return NULL;
546
547	req->monc = monc;
548	kref_init(&req->kref);
549	RB_CLEAR_NODE(&req->node);
550	init_completion(&req->completion);
551
552	dout("%s greq %p\n", __func__, req);
553	return req;
554}
555
556static void register_generic_request(struct ceph_mon_generic_request *req)
557{
558	struct ceph_mon_client *monc = req->monc;
559
560	WARN_ON(req->tid);
561
562	get_generic_request(req);
563	req->tid = ++monc->last_tid;
564	insert_generic_request(&monc->generic_request_tree, req);
565}
566
567static void send_generic_request(struct ceph_mon_client *monc,
568				 struct ceph_mon_generic_request *req)
569{
570	WARN_ON(!req->tid);
571
572	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
573	req->request->hdr.tid = cpu_to_le64(req->tid);
574	ceph_con_send(&monc->con, ceph_msg_get(req->request));
575}
576
577static void __finish_generic_request(struct ceph_mon_generic_request *req)
578{
579	struct ceph_mon_client *monc = req->monc;
580
581	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
582	erase_generic_request(&monc->generic_request_tree, req);
583
584	ceph_msg_revoke(req->request);
585	ceph_msg_revoke_incoming(req->reply);
586}
587
588static void finish_generic_request(struct ceph_mon_generic_request *req)
589{
590	__finish_generic_request(req);
591	put_generic_request(req);
592}
593
594static void complete_generic_request(struct ceph_mon_generic_request *req)
595{
596	if (req->complete_cb)
597		req->complete_cb(req);
598	else
599		complete_all(&req->completion);
600	put_generic_request(req);
601}
602
603static void cancel_generic_request(struct ceph_mon_generic_request *req)
604{
605	struct ceph_mon_client *monc = req->monc;
606	struct ceph_mon_generic_request *lookup_req;
607
608	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
609
610	mutex_lock(&monc->mutex);
611	lookup_req = lookup_generic_request(&monc->generic_request_tree,
612					    req->tid);
613	if (lookup_req) {
614		WARN_ON(lookup_req != req);
615		finish_generic_request(req);
616	}
617
618	mutex_unlock(&monc->mutex);
619}
620
621static int wait_generic_request(struct ceph_mon_generic_request *req)
622{
623	int ret;
624
625	dout("%s greq %p tid %llu\n", __func__, req, req->tid);
626	ret = wait_for_completion_interruptible(&req->completion);
627	if (ret)
628		cancel_generic_request(req);
629	else
630		ret = req->result; /* completed */
631
632	return ret;
633}
634
635static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
636					 struct ceph_msg_header *hdr,
637					 int *skip)
638{
639	struct ceph_mon_client *monc = con->private;
640	struct ceph_mon_generic_request *req;
641	u64 tid = le64_to_cpu(hdr->tid);
642	struct ceph_msg *m;
643
644	mutex_lock(&monc->mutex);
645	req = lookup_generic_request(&monc->generic_request_tree, tid);
646	if (!req) {
647		dout("get_generic_reply %lld dne\n", tid);
648		*skip = 1;
649		m = NULL;
650	} else {
651		dout("get_generic_reply %lld got %p\n", tid, req->reply);
652		*skip = 0;
653		m = ceph_msg_get(req->reply);
654		/*
655		 * we don't need to track the connection reading into
656		 * this reply because we only have one open connection
657		 * at a time, ever.
658		 */
659	}
660	mutex_unlock(&monc->mutex);
661	return m;
662}
663
664/*
665 * statfs
666 */
667static void handle_statfs_reply(struct ceph_mon_client *monc,
668				struct ceph_msg *msg)
669{
670	struct ceph_mon_generic_request *req;
671	struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
672	u64 tid = le64_to_cpu(msg->hdr.tid);
673
674	dout("%s msg %p tid %llu\n", __func__, msg, tid);
675
676	if (msg->front.iov_len != sizeof(*reply))
677		goto bad;
678
679	mutex_lock(&monc->mutex);
680	req = lookup_generic_request(&monc->generic_request_tree, tid);
681	if (!req) {
682		mutex_unlock(&monc->mutex);
683		return;
684	}
685
686	req->result = 0;
687	*req->u.st = reply->st; /* struct */
688	__finish_generic_request(req);
689	mutex_unlock(&monc->mutex);
690
691	complete_generic_request(req);
692	return;
693
694bad:
695	pr_err("corrupt statfs reply, tid %llu\n", tid);
696	ceph_msg_dump(msg);
697}
698
699/*
700 * Do a synchronous statfs().
701 */
702int ceph_monc_do_statfs(struct ceph_mon_client *monc, u64 data_pool,
703			struct ceph_statfs *buf)
704{
705	struct ceph_mon_generic_request *req;
706	struct ceph_mon_statfs *h;
707	int ret = -ENOMEM;
708
709	req = alloc_generic_request(monc, GFP_NOFS);
710	if (!req)
711		goto out;
712
713	req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
714				    true);
715	if (!req->request)
716		goto out;
717
718	req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 64, GFP_NOFS, true);
719	if (!req->reply)
720		goto out;
721
722	req->u.st = buf;
723	req->request->hdr.version = cpu_to_le16(2);
724
725	mutex_lock(&monc->mutex);
726	register_generic_request(req);
727	/* fill out request */
728	h = req->request->front.iov_base;
729	h->monhdr.have_version = 0;
730	h->monhdr.session_mon = cpu_to_le16(-1);
731	h->monhdr.session_mon_tid = 0;
732	h->fsid = monc->monmap->fsid;
733	h->contains_data_pool = (data_pool != CEPH_NOPOOL);
734	h->data_pool = cpu_to_le64(data_pool);
735	send_generic_request(monc, req);
736	mutex_unlock(&monc->mutex);
737
738	ret = wait_generic_request(req);
739out:
740	put_generic_request(req);
741	return ret;
742}
743EXPORT_SYMBOL(ceph_monc_do_statfs);
744
745static void handle_get_version_reply(struct ceph_mon_client *monc,
746				     struct ceph_msg *msg)
747{
748	struct ceph_mon_generic_request *req;
749	u64 tid = le64_to_cpu(msg->hdr.tid);
750	void *p = msg->front.iov_base;
751	void *end = p + msg->front_alloc_len;
752	u64 handle;
753
754	dout("%s msg %p tid %llu\n", __func__, msg, tid);
755
756	ceph_decode_need(&p, end, 2*sizeof(u64), bad);
757	handle = ceph_decode_64(&p);
758	if (tid != 0 && tid != handle)
759		goto bad;
760
761	mutex_lock(&monc->mutex);
762	req = lookup_generic_request(&monc->generic_request_tree, handle);
763	if (!req) {
764		mutex_unlock(&monc->mutex);
765		return;
766	}
767
768	req->result = 0;
769	req->u.newest = ceph_decode_64(&p);
770	__finish_generic_request(req);
771	mutex_unlock(&monc->mutex);
772
773	complete_generic_request(req);
774	return;
775
776bad:
777	pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
778	ceph_msg_dump(msg);
779}
780
781static struct ceph_mon_generic_request *
782__ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
783			ceph_monc_callback_t cb, u64 private_data)
784{
785	struct ceph_mon_generic_request *req;
786
787	req = alloc_generic_request(monc, GFP_NOIO);
788	if (!req)
789		goto err_put_req;
790
791	req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
792				    sizeof(u64) + sizeof(u32) + strlen(what),
793				    GFP_NOIO, true);
794	if (!req->request)
795		goto err_put_req;
796
797	req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 32, GFP_NOIO,
798				  true);
799	if (!req->reply)
800		goto err_put_req;
801
802	req->complete_cb = cb;
803	req->private_data = private_data;
804
805	mutex_lock(&monc->mutex);
806	register_generic_request(req);
807	{
808		void *p = req->request->front.iov_base;
809		void *const end = p + req->request->front_alloc_len;
810
811		ceph_encode_64(&p, req->tid); /* handle */
812		ceph_encode_string(&p, end, what, strlen(what));
813		WARN_ON(p != end);
814	}
815	send_generic_request(monc, req);
816	mutex_unlock(&monc->mutex);
817
818	return req;
819
820err_put_req:
821	put_generic_request(req);
822	return ERR_PTR(-ENOMEM);
823}
824
825/*
826 * Send MMonGetVersion and wait for the reply.
827 *
828 * @what: one of "mdsmap", "osdmap" or "monmap"
829 */
830int ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
831			  u64 *newest)
832{
833	struct ceph_mon_generic_request *req;
834	int ret;
835
836	req = __ceph_monc_get_version(monc, what, NULL, 0);
837	if (IS_ERR(req))
838		return PTR_ERR(req);
839
840	ret = wait_generic_request(req);
841	if (!ret)
842		*newest = req->u.newest;
843
844	put_generic_request(req);
845	return ret;
846}
847EXPORT_SYMBOL(ceph_monc_get_version);
848
849/*
850 * Send MMonGetVersion,
851 *
852 * @what: one of "mdsmap", "osdmap" or "monmap"
853 */
854int ceph_monc_get_version_async(struct ceph_mon_client *monc, const char *what,
855				ceph_monc_callback_t cb, u64 private_data)
856{
857	struct ceph_mon_generic_request *req;
858
859	req = __ceph_monc_get_version(monc, what, cb, private_data);
860	if (IS_ERR(req))
861		return PTR_ERR(req);
862
863	put_generic_request(req);
864	return 0;
865}
866EXPORT_SYMBOL(ceph_monc_get_version_async);
867
868static void handle_command_ack(struct ceph_mon_client *monc,
869			       struct ceph_msg *msg)
870{
871	struct ceph_mon_generic_request *req;
872	void *p = msg->front.iov_base;
873	void *const end = p + msg->front_alloc_len;
874	u64 tid = le64_to_cpu(msg->hdr.tid);
875
876	dout("%s msg %p tid %llu\n", __func__, msg, tid);
877
878	ceph_decode_need(&p, end, sizeof(struct ceph_mon_request_header) +
879							    sizeof(u32), bad);
880	p += sizeof(struct ceph_mon_request_header);
881
882	mutex_lock(&monc->mutex);
883	req = lookup_generic_request(&monc->generic_request_tree, tid);
884	if (!req) {
885		mutex_unlock(&monc->mutex);
886		return;
887	}
888
889	req->result = ceph_decode_32(&p);
890	__finish_generic_request(req);
891	mutex_unlock(&monc->mutex);
892
893	complete_generic_request(req);
894	return;
895
896bad:
897	pr_err("corrupt mon_command ack, tid %llu\n", tid);
898	ceph_msg_dump(msg);
899}
900
901static __printf(2, 0)
902int do_mon_command_vargs(struct ceph_mon_client *monc, const char *fmt,
903			 va_list ap)
904{
905	struct ceph_mon_generic_request *req;
906	struct ceph_mon_command *h;
907	int ret = -ENOMEM;
908	int len;
909
910	req = alloc_generic_request(monc, GFP_NOIO);
911	if (!req)
912		goto out;
913
914	req->request = ceph_msg_new(CEPH_MSG_MON_COMMAND, 256, GFP_NOIO, true);
915	if (!req->request)
916		goto out;
917
918	req->reply = ceph_msg_new(CEPH_MSG_MON_COMMAND_ACK, 512, GFP_NOIO,
919				  true);
920	if (!req->reply)
921		goto out;
922
923	mutex_lock(&monc->mutex);
924	register_generic_request(req);
925	h = req->request->front.iov_base;
926	h->monhdr.have_version = 0;
927	h->monhdr.session_mon = cpu_to_le16(-1);
928	h->monhdr.session_mon_tid = 0;
929	h->fsid = monc->monmap->fsid;
930	h->num_strs = cpu_to_le32(1);
931	len = vsprintf(h->str, fmt, ap);
932	h->str_len = cpu_to_le32(len);
933	send_generic_request(monc, req);
934	mutex_unlock(&monc->mutex);
935
936	ret = wait_generic_request(req);
937out:
938	put_generic_request(req);
939	return ret;
940}
941
942static __printf(2, 3)
943int do_mon_command(struct ceph_mon_client *monc, const char *fmt, ...)
944{
945	va_list ap;
946	int ret;
947
948	va_start(ap, fmt);
949	ret = do_mon_command_vargs(monc, fmt, ap);
950	va_end(ap);
951	return ret;
952}
953
954int ceph_monc_blocklist_add(struct ceph_mon_client *monc,
955			    struct ceph_entity_addr *client_addr)
956{
957	int ret;
958
959	ret = do_mon_command(monc,
960			     "{ \"prefix\": \"osd blocklist\", \
961				\"blocklistop\": \"add\", \
962				\"addr\": \"%pISpc/%u\" }",
963			     &client_addr->in_addr,
964			     le32_to_cpu(client_addr->nonce));
965	if (ret == -EINVAL) {
966		/*
967		 * The monitor returns EINVAL on an unrecognized command.
968		 * Try the legacy command -- it is exactly the same except
969		 * for the name.
970		 */
971		ret = do_mon_command(monc,
972				     "{ \"prefix\": \"osd blacklist\", \
973					\"blacklistop\": \"add\", \
974					\"addr\": \"%pISpc/%u\" }",
975				     &client_addr->in_addr,
976				     le32_to_cpu(client_addr->nonce));
977	}
978	if (ret)
979		return ret;
980
981	/*
982	 * Make sure we have the osdmap that includes the blocklist
983	 * entry.  This is needed to ensure that the OSDs pick up the
984	 * new blocklist before processing any future requests from
985	 * this client.
986	 */
987	return ceph_wait_for_latest_osdmap(monc->client, 0);
988}
989EXPORT_SYMBOL(ceph_monc_blocklist_add);
990
991/*
992 * Resend pending generic requests.
993 */
994static void __resend_generic_request(struct ceph_mon_client *monc)
995{
996	struct ceph_mon_generic_request *req;
997	struct rb_node *p;
998
999	for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
1000		req = rb_entry(p, struct ceph_mon_generic_request, node);
1001		ceph_msg_revoke(req->request);
1002		ceph_msg_revoke_incoming(req->reply);
1003		ceph_con_send(&monc->con, ceph_msg_get(req->request));
1004	}
1005}
1006
1007/*
1008 * Delayed work.  If we haven't mounted yet, retry.  Otherwise,
1009 * renew/retry subscription as needed (in case it is timing out, or we
1010 * got an ENOMEM).  And keep the monitor connection alive.
1011 */
1012static void delayed_work(struct work_struct *work)
1013{
1014	struct ceph_mon_client *monc =
1015		container_of(work, struct ceph_mon_client, delayed_work.work);
1016
1017	mutex_lock(&monc->mutex);
1018	dout("%s mon%d\n", __func__, monc->cur_mon);
1019	if (monc->cur_mon < 0) {
1020		goto out;
1021	}
1022
1023	if (monc->hunting) {
1024		dout("%s continuing hunt\n", __func__);
1025		reopen_session(monc);
1026	} else {
1027		int is_auth = ceph_auth_is_authenticated(monc->auth);
1028
1029		dout("%s is_authed %d\n", __func__, is_auth);
1030		if (ceph_con_keepalive_expired(&monc->con,
1031					       CEPH_MONC_PING_TIMEOUT)) {
1032			dout("monc keepalive timeout\n");
1033			is_auth = 0;
1034			reopen_session(monc);
1035		}
1036
1037		if (!monc->hunting) {
1038			ceph_con_keepalive(&monc->con);
1039			__validate_auth(monc);
1040			un_backoff(monc);
1041		}
1042
1043		if (is_auth &&
1044		    !(monc->con.peer_features & CEPH_FEATURE_MON_STATEFUL_SUB)) {
1045			unsigned long now = jiffies;
1046
1047			dout("%s renew subs? now %lu renew after %lu\n",
1048			     __func__, now, monc->sub_renew_after);
1049			if (time_after_eq(now, monc->sub_renew_after))
1050				__send_subscribe(monc);
1051		}
1052	}
1053	__schedule_delayed(monc);
1054
1055out:
1056	mutex_unlock(&monc->mutex);
1057}
1058
1059/*
1060 * On startup, we build a temporary monmap populated with the IPs
1061 * provided by mount(2).
1062 */
1063static int build_initial_monmap(struct ceph_mon_client *monc)
1064{
1065	struct ceph_options *opt = monc->client->options;
1066	struct ceph_entity_addr *mon_addr = opt->mon_addr;
1067	int num_mon = opt->num_mon;
1068	int i;
1069
1070	/* build initial monmap */
1071	monc->monmap = kzalloc(struct_size(monc->monmap, mon_inst, num_mon),
1072			       GFP_KERNEL);
1073	if (!monc->monmap)
1074		return -ENOMEM;
1075	for (i = 0; i < num_mon; i++) {
1076		monc->monmap->mon_inst[i].addr = mon_addr[i];
1077		monc->monmap->mon_inst[i].addr.nonce = 0;
1078		monc->monmap->mon_inst[i].name.type =
1079			CEPH_ENTITY_TYPE_MON;
1080		monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
1081	}
1082	monc->monmap->num_mon = num_mon;
1083	return 0;
1084}
1085
1086int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
1087{
1088	int err = 0;
1089
1090	dout("init\n");
1091	memset(monc, 0, sizeof(*monc));
1092	monc->client = cl;
1093	monc->monmap = NULL;
1094	mutex_init(&monc->mutex);
1095
1096	err = build_initial_monmap(monc);
1097	if (err)
1098		goto out;
1099
1100	/* connection */
1101	/* authentication */
1102	monc->auth = ceph_auth_init(cl->options->name,
1103				    cl->options->key);
1104	if (IS_ERR(monc->auth)) {
1105		err = PTR_ERR(monc->auth);
1106		goto out_monmap;
1107	}
1108	monc->auth->want_keys =
1109		CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
1110		CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
1111
1112	/* msgs */
1113	err = -ENOMEM;
1114	monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
1115				     sizeof(struct ceph_mon_subscribe_ack),
1116				     GFP_KERNEL, true);
1117	if (!monc->m_subscribe_ack)
1118		goto out_auth;
1119
1120	monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 128,
1121					 GFP_KERNEL, true);
1122	if (!monc->m_subscribe)
1123		goto out_subscribe_ack;
1124
1125	monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096,
1126					  GFP_KERNEL, true);
1127	if (!monc->m_auth_reply)
1128		goto out_subscribe;
1129
1130	monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_KERNEL, true);
1131	monc->pending_auth = 0;
1132	if (!monc->m_auth)
1133		goto out_auth_reply;
1134
1135	ceph_con_init(&monc->con, monc, &mon_con_ops,
1136		      &monc->client->msgr);
1137
1138	monc->cur_mon = -1;
1139	monc->had_a_connection = false;
1140	monc->hunt_mult = 1;
1141
1142	INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
1143	monc->generic_request_tree = RB_ROOT;
1144	monc->last_tid = 0;
1145
1146	monc->fs_cluster_id = CEPH_FS_CLUSTER_ID_NONE;
1147
1148	return 0;
1149
1150out_auth_reply:
1151	ceph_msg_put(monc->m_auth_reply);
1152out_subscribe:
1153	ceph_msg_put(monc->m_subscribe);
1154out_subscribe_ack:
1155	ceph_msg_put(monc->m_subscribe_ack);
1156out_auth:
1157	ceph_auth_destroy(monc->auth);
1158out_monmap:
1159	kfree(monc->monmap);
1160out:
1161	return err;
1162}
1163EXPORT_SYMBOL(ceph_monc_init);
1164
1165void ceph_monc_stop(struct ceph_mon_client *monc)
1166{
1167	dout("stop\n");
1168
1169	mutex_lock(&monc->mutex);
1170	__close_session(monc);
1171	monc->hunting = false;
1172	monc->cur_mon = -1;
1173	mutex_unlock(&monc->mutex);
1174
1175	cancel_delayed_work_sync(&monc->delayed_work);
1176
1177	/*
1178	 * flush msgr queue before we destroy ourselves to ensure that:
1179	 *  - any work that references our embedded con is finished.
1180	 *  - any osd_client or other work that may reference an authorizer
1181	 *    finishes before we shut down the auth subsystem.
1182	 */
1183	ceph_msgr_flush();
1184
1185	ceph_auth_destroy(monc->auth);
1186
1187	WARN_ON(!RB_EMPTY_ROOT(&monc->generic_request_tree));
1188
1189	ceph_msg_put(monc->m_auth);
1190	ceph_msg_put(monc->m_auth_reply);
1191	ceph_msg_put(monc->m_subscribe);
1192	ceph_msg_put(monc->m_subscribe_ack);
1193
1194	kfree(monc->monmap);
1195}
1196EXPORT_SYMBOL(ceph_monc_stop);
1197
1198static void finish_hunting(struct ceph_mon_client *monc)
1199{
1200	if (monc->hunting) {
1201		dout("%s found mon%d\n", __func__, monc->cur_mon);
1202		monc->hunting = false;
1203		monc->had_a_connection = true;
1204		un_backoff(monc);
1205		__schedule_delayed(monc);
1206	}
1207}
1208
1209static void handle_auth_reply(struct ceph_mon_client *monc,
1210			      struct ceph_msg *msg)
1211{
1212	int ret;
1213	int was_auth = 0;
1214
1215	mutex_lock(&monc->mutex);
1216	was_auth = ceph_auth_is_authenticated(monc->auth);
1217	monc->pending_auth = 0;
1218	ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
1219				     msg->front.iov_len,
1220				     monc->m_auth->front.iov_base,
1221				     monc->m_auth->front_alloc_len);
1222	if (ret > 0) {
1223		__send_prepared_auth_request(monc, ret);
1224		goto out;
1225	}
1226
1227	finish_hunting(monc);
1228
1229	if (ret < 0) {
1230		monc->client->auth_err = ret;
1231	} else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
1232		dout("authenticated, starting session\n");
1233
1234		monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
1235		monc->client->msgr.inst.name.num =
1236					cpu_to_le64(monc->auth->global_id);
1237
1238		__send_subscribe(monc);
1239		__resend_generic_request(monc);
1240
1241		pr_info("mon%d %s session established\n", monc->cur_mon,
1242			ceph_pr_addr(&monc->con.peer_addr));
1243	}
1244
1245out:
1246	mutex_unlock(&monc->mutex);
1247	if (monc->client->auth_err < 0)
1248		wake_up_all(&monc->client->auth_wq);
1249}
1250
1251static int __validate_auth(struct ceph_mon_client *monc)
1252{
1253	int ret;
1254
1255	if (monc->pending_auth)
1256		return 0;
1257
1258	ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
1259			      monc->m_auth->front_alloc_len);
1260	if (ret <= 0)
1261		return ret; /* either an error, or no need to authenticate */
1262	__send_prepared_auth_request(monc, ret);
1263	return 0;
1264}
1265
1266int ceph_monc_validate_auth(struct ceph_mon_client *monc)
1267{
1268	int ret;
1269
1270	mutex_lock(&monc->mutex);
1271	ret = __validate_auth(monc);
1272	mutex_unlock(&monc->mutex);
1273	return ret;
1274}
1275EXPORT_SYMBOL(ceph_monc_validate_auth);
1276
1277/*
1278 * handle incoming message
1279 */
1280static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1281{
1282	struct ceph_mon_client *monc = con->private;
1283	int type = le16_to_cpu(msg->hdr.type);
1284
1285	switch (type) {
1286	case CEPH_MSG_AUTH_REPLY:
1287		handle_auth_reply(monc, msg);
1288		break;
1289
1290	case CEPH_MSG_MON_SUBSCRIBE_ACK:
1291		handle_subscribe_ack(monc, msg);
1292		break;
1293
1294	case CEPH_MSG_STATFS_REPLY:
1295		handle_statfs_reply(monc, msg);
1296		break;
1297
1298	case CEPH_MSG_MON_GET_VERSION_REPLY:
1299		handle_get_version_reply(monc, msg);
1300		break;
1301
1302	case CEPH_MSG_MON_COMMAND_ACK:
1303		handle_command_ack(monc, msg);
1304		break;
1305
1306	case CEPH_MSG_MON_MAP:
1307		ceph_monc_handle_map(monc, msg);
1308		break;
1309
1310	case CEPH_MSG_OSD_MAP:
1311		ceph_osdc_handle_map(&monc->client->osdc, msg);
1312		break;
1313
1314	default:
1315		/* can the chained handler handle it? */
1316		if (monc->client->extra_mon_dispatch &&
1317		    monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1318			break;
1319
1320		pr_err("received unknown message type %d %s\n", type,
1321		       ceph_msg_type_name(type));
1322	}
1323	ceph_msg_put(msg);
1324}
1325
1326/*
1327 * Allocate memory for incoming message
1328 */
1329static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1330				      struct ceph_msg_header *hdr,
1331				      int *skip)
1332{
1333	struct ceph_mon_client *monc = con->private;
1334	int type = le16_to_cpu(hdr->type);
1335	int front_len = le32_to_cpu(hdr->front_len);
1336	struct ceph_msg *m = NULL;
1337
1338	*skip = 0;
1339
1340	switch (type) {
1341	case CEPH_MSG_MON_SUBSCRIBE_ACK:
1342		m = ceph_msg_get(monc->m_subscribe_ack);
1343		break;
1344	case CEPH_MSG_STATFS_REPLY:
1345	case CEPH_MSG_MON_COMMAND_ACK:
1346		return get_generic_reply(con, hdr, skip);
1347	case CEPH_MSG_AUTH_REPLY:
1348		m = ceph_msg_get(monc->m_auth_reply);
1349		break;
1350	case CEPH_MSG_MON_GET_VERSION_REPLY:
1351		if (le64_to_cpu(hdr->tid) != 0)
1352			return get_generic_reply(con, hdr, skip);
1353
1354		/*
1355		 * Older OSDs don't set reply tid even if the orignal
1356		 * request had a non-zero tid.  Work around this weirdness
1357		 * by allocating a new message.
1358		 */
1359		fallthrough;
1360	case CEPH_MSG_MON_MAP:
1361	case CEPH_MSG_MDS_MAP:
1362	case CEPH_MSG_OSD_MAP:
1363	case CEPH_MSG_FS_MAP_USER:
1364		m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1365		if (!m)
1366			return NULL;	/* ENOMEM--return skip == 0 */
1367		break;
1368	}
1369
1370	if (!m) {
1371		pr_info("alloc_msg unknown type %d\n", type);
1372		*skip = 1;
1373	} else if (front_len > m->front_alloc_len) {
1374		pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1375			front_len, m->front_alloc_len,
1376			(unsigned int)con->peer_name.type,
1377			le64_to_cpu(con->peer_name.num));
1378		ceph_msg_put(m);
1379		m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1380	}
1381
1382	return m;
1383}
1384
1385/*
1386 * If the monitor connection resets, pick a new monitor and resubmit
1387 * any pending requests.
1388 */
1389static void mon_fault(struct ceph_connection *con)
1390{
1391	struct ceph_mon_client *monc = con->private;
1392
1393	mutex_lock(&monc->mutex);
1394	dout("%s mon%d\n", __func__, monc->cur_mon);
1395	if (monc->cur_mon >= 0) {
1396		if (!monc->hunting) {
1397			dout("%s hunting for new mon\n", __func__);
1398			reopen_session(monc);
1399			__schedule_delayed(monc);
1400		} else {
1401			dout("%s already hunting\n", __func__);
1402		}
1403	}
1404	mutex_unlock(&monc->mutex);
1405}
1406
1407/*
1408 * We can ignore refcounting on the connection struct, as all references
1409 * will come from the messenger workqueue, which is drained prior to
1410 * mon_client destruction.
1411 */
1412static struct ceph_connection *con_get(struct ceph_connection *con)
1413{
1414	return con;
1415}
1416
1417static void con_put(struct ceph_connection *con)
1418{
1419}
1420
1421static const struct ceph_connection_operations mon_con_ops = {
1422	.get = con_get,
1423	.put = con_put,
1424	.dispatch = dispatch,
1425	.fault = mon_fault,
1426	.alloc_msg = mon_alloc_msg,
1427};
1428