mon_client.c revision b9a678994b4a64b1106ab2cf7cfe7cbc10bb6f40
1#include <linux/ceph/ceph_debug.h>
2
3#include <linux/module.h>
4#include <linux/types.h>
5#include <linux/slab.h>
6#include <linux/random.h>
7#include <linux/sched.h>
8
9#include <linux/ceph/mon_client.h>
10#include <linux/ceph/libceph.h>
11#include <linux/ceph/debugfs.h>
12#include <linux/ceph/decode.h>
13#include <linux/ceph/auth.h>
14
15/*
16 * Interact with Ceph monitor cluster.  Handle requests for new map
17 * versions, and periodically resend as needed.  Also implement
18 * statfs() and umount().
19 *
20 * A small cluster of Ceph "monitors" are responsible for managing critical
21 * cluster configuration and state information.  An odd number (e.g., 3, 5)
22 * of cmon daemons use a modified version of the Paxos part-time parliament
23 * algorithm to manage the MDS map (mds cluster membership), OSD map, and
24 * list of clients who have mounted the file system.
25 *
26 * We maintain an open, active session with a monitor at all times in order to
27 * receive timely MDSMap updates.  We periodically send a keepalive byte on the
28 * TCP socket to ensure we detect a failure.  If the connection does break, we
29 * randomly hunt for a new monitor.  Once the connection is reestablished, we
30 * resend any outstanding requests.
31 */
32
33static const struct ceph_connection_operations mon_con_ops;
34
35static int __validate_auth(struct ceph_mon_client *monc);
36
37/*
38 * Decode a monmap blob (e.g., during mount).
39 */
40struct ceph_monmap *ceph_monmap_decode(void *p, void *end)
41{
42	struct ceph_monmap *m = NULL;
43	int i, err = -EINVAL;
44	struct ceph_fsid fsid;
45	u32 epoch, num_mon;
46	u16 version;
47	u32 len;
48
49	ceph_decode_32_safe(&p, end, len, bad);
50	ceph_decode_need(&p, end, len, bad);
51
52	dout("monmap_decode %p %p len %d\n", p, end, (int)(end-p));
53
54	ceph_decode_16_safe(&p, end, version, bad);
55
56	ceph_decode_need(&p, end, sizeof(fsid) + 2*sizeof(u32), bad);
57	ceph_decode_copy(&p, &fsid, sizeof(fsid));
58	epoch = ceph_decode_32(&p);
59
60	num_mon = ceph_decode_32(&p);
61	ceph_decode_need(&p, end, num_mon*sizeof(m->mon_inst[0]), bad);
62
63	if (num_mon >= CEPH_MAX_MON)
64		goto bad;
65	m = kmalloc(sizeof(*m) + sizeof(m->mon_inst[0])*num_mon, GFP_NOFS);
66	if (m == NULL)
67		return ERR_PTR(-ENOMEM);
68	m->fsid = fsid;
69	m->epoch = epoch;
70	m->num_mon = num_mon;
71	ceph_decode_copy(&p, m->mon_inst, num_mon*sizeof(m->mon_inst[0]));
72	for (i = 0; i < num_mon; i++)
73		ceph_decode_addr(&m->mon_inst[i].addr);
74
75	dout("monmap_decode epoch %d, num_mon %d\n", m->epoch,
76	     m->num_mon);
77	for (i = 0; i < m->num_mon; i++)
78		dout("monmap_decode  mon%d is %s\n", i,
79		     ceph_pr_addr(&m->mon_inst[i].addr.in_addr));
80	return m;
81
82bad:
83	dout("monmap_decode failed with %d\n", err);
84	kfree(m);
85	return ERR_PTR(err);
86}
87
88/*
89 * return true if *addr is included in the monmap.
90 */
91int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr)
92{
93	int i;
94
95	for (i = 0; i < m->num_mon; i++)
96		if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0)
97			return 1;
98	return 0;
99}
100
101/*
102 * Send an auth request.
103 */
104static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len)
105{
106	monc->pending_auth = 1;
107	monc->m_auth->front.iov_len = len;
108	monc->m_auth->hdr.front_len = cpu_to_le32(len);
109	ceph_msg_revoke(monc->m_auth);
110	ceph_msg_get(monc->m_auth);  /* keep our ref */
111	ceph_con_send(&monc->con, monc->m_auth);
112}
113
114/*
115 * Close monitor session, if any.
116 */
117static void __close_session(struct ceph_mon_client *monc)
118{
119	dout("__close_session closing mon%d\n", monc->cur_mon);
120	ceph_msg_revoke(monc->m_auth);
121	ceph_msg_revoke_incoming(monc->m_auth_reply);
122	ceph_msg_revoke(monc->m_subscribe);
123	ceph_msg_revoke_incoming(monc->m_subscribe_ack);
124	ceph_con_close(&monc->con);
125	monc->cur_mon = -1;
126	monc->pending_auth = 0;
127	ceph_auth_reset(monc->auth);
128}
129
130/*
131 * Open a session with a (new) monitor.
132 */
133static int __open_session(struct ceph_mon_client *monc)
134{
135	char r;
136	int ret;
137
138	if (monc->cur_mon < 0) {
139		get_random_bytes(&r, 1);
140		monc->cur_mon = r % monc->monmap->num_mon;
141		dout("open_session num=%d r=%d -> mon%d\n",
142		     monc->monmap->num_mon, r, monc->cur_mon);
143		monc->sub_sent = 0;
144		monc->sub_renew_after = jiffies;  /* i.e., expired */
145		monc->want_next_osdmap = !!monc->want_next_osdmap;
146
147		dout("open_session mon%d opening\n", monc->cur_mon);
148		ceph_con_open(&monc->con,
149			      CEPH_ENTITY_TYPE_MON, monc->cur_mon,
150			      &monc->monmap->mon_inst[monc->cur_mon].addr);
151
152		/* initiatiate authentication handshake */
153		ret = ceph_auth_build_hello(monc->auth,
154					    monc->m_auth->front.iov_base,
155					    monc->m_auth->front_alloc_len);
156		__send_prepared_auth_request(monc, ret);
157	} else {
158		dout("open_session mon%d already open\n", monc->cur_mon);
159	}
160	return 0;
161}
162
163static bool __sub_expired(struct ceph_mon_client *monc)
164{
165	return time_after_eq(jiffies, monc->sub_renew_after);
166}
167
168/*
169 * Reschedule delayed work timer.
170 */
171static void __schedule_delayed(struct ceph_mon_client *monc)
172{
173	unsigned int delay;
174
175	if (monc->cur_mon < 0 || __sub_expired(monc))
176		delay = 10 * HZ;
177	else
178		delay = 20 * HZ;
179	dout("__schedule_delayed after %u\n", delay);
180	schedule_delayed_work(&monc->delayed_work, delay);
181}
182
183/*
184 * Send subscribe request for mdsmap and/or osdmap.
185 */
186static void __send_subscribe(struct ceph_mon_client *monc)
187{
188	dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n",
189	     (unsigned int)monc->sub_sent, __sub_expired(monc),
190	     monc->want_next_osdmap);
191	if ((__sub_expired(monc) && !monc->sub_sent) ||
192	    monc->want_next_osdmap == 1) {
193		struct ceph_msg *msg = monc->m_subscribe;
194		struct ceph_mon_subscribe_item *i;
195		void *p, *end;
196		int num;
197
198		p = msg->front.iov_base;
199		end = p + msg->front_alloc_len;
200
201		num = 1 + !!monc->want_next_osdmap + !!monc->want_mdsmap;
202		ceph_encode_32(&p, num);
203
204		if (monc->want_next_osdmap) {
205			dout("__send_subscribe to 'osdmap' %u\n",
206			     (unsigned int)monc->have_osdmap);
207			ceph_encode_string(&p, end, "osdmap", 6);
208			i = p;
209			i->have = cpu_to_le64(monc->have_osdmap);
210			i->onetime = 1;
211			p += sizeof(*i);
212			monc->want_next_osdmap = 2;  /* requested */
213		}
214		if (monc->want_mdsmap) {
215			dout("__send_subscribe to 'mdsmap' %u+\n",
216			     (unsigned int)monc->have_mdsmap);
217			ceph_encode_string(&p, end, "mdsmap", 6);
218			i = p;
219			i->have = cpu_to_le64(monc->have_mdsmap);
220			i->onetime = 0;
221			p += sizeof(*i);
222		}
223		ceph_encode_string(&p, end, "monmap", 6);
224		i = p;
225		i->have = 0;
226		i->onetime = 0;
227		p += sizeof(*i);
228
229		msg->front.iov_len = p - msg->front.iov_base;
230		msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
231		ceph_msg_revoke(msg);
232		ceph_con_send(&monc->con, ceph_msg_get(msg));
233
234		monc->sub_sent = jiffies | 1;  /* never 0 */
235	}
236}
237
238static void handle_subscribe_ack(struct ceph_mon_client *monc,
239				 struct ceph_msg *msg)
240{
241	unsigned int seconds;
242	struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
243
244	if (msg->front.iov_len < sizeof(*h))
245		goto bad;
246	seconds = le32_to_cpu(h->duration);
247
248	mutex_lock(&monc->mutex);
249	if (monc->hunting) {
250		pr_info("mon%d %s session established\n",
251			monc->cur_mon,
252			ceph_pr_addr(&monc->con.peer_addr.in_addr));
253		monc->hunting = false;
254	}
255	dout("handle_subscribe_ack after %d seconds\n", seconds);
256	monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1;
257	monc->sub_sent = 0;
258	mutex_unlock(&monc->mutex);
259	return;
260bad:
261	pr_err("got corrupt subscribe-ack msg\n");
262	ceph_msg_dump(msg);
263}
264
265/*
266 * Keep track of which maps we have
267 */
268int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got)
269{
270	mutex_lock(&monc->mutex);
271	monc->have_mdsmap = got;
272	mutex_unlock(&monc->mutex);
273	return 0;
274}
275EXPORT_SYMBOL(ceph_monc_got_mdsmap);
276
277int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got)
278{
279	mutex_lock(&monc->mutex);
280	monc->have_osdmap = got;
281	monc->want_next_osdmap = 0;
282	mutex_unlock(&monc->mutex);
283	return 0;
284}
285
286/*
287 * Register interest in the next osdmap
288 */
289void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
290{
291	dout("request_next_osdmap have %u\n", monc->have_osdmap);
292	mutex_lock(&monc->mutex);
293	if (!monc->want_next_osdmap)
294		monc->want_next_osdmap = 1;
295	if (monc->want_next_osdmap < 2)
296		__send_subscribe(monc);
297	mutex_unlock(&monc->mutex);
298}
299EXPORT_SYMBOL(ceph_monc_request_next_osdmap);
300
301int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
302			  unsigned long timeout)
303{
304	unsigned long started = jiffies;
305	int ret;
306
307	mutex_lock(&monc->mutex);
308	while (monc->have_osdmap < epoch) {
309		mutex_unlock(&monc->mutex);
310
311		if (timeout != 0 && time_after_eq(jiffies, started + timeout))
312			return -ETIMEDOUT;
313
314		ret = wait_event_interruptible_timeout(monc->client->auth_wq,
315					 monc->have_osdmap >= epoch, timeout);
316		if (ret < 0)
317			return ret;
318
319		mutex_lock(&monc->mutex);
320	}
321
322	mutex_unlock(&monc->mutex);
323	return 0;
324}
325EXPORT_SYMBOL(ceph_monc_wait_osdmap);
326
327/*
328 *
329 */
330int ceph_monc_open_session(struct ceph_mon_client *monc)
331{
332	mutex_lock(&monc->mutex);
333	__open_session(monc);
334	__schedule_delayed(monc);
335	mutex_unlock(&monc->mutex);
336	return 0;
337}
338EXPORT_SYMBOL(ceph_monc_open_session);
339
340/*
341 * We require the fsid and global_id in order to initialize our
342 * debugfs dir.
343 */
344static bool have_debugfs_info(struct ceph_mon_client *monc)
345{
346	dout("have_debugfs_info fsid %d globalid %lld\n",
347	     (int)monc->client->have_fsid, monc->auth->global_id);
348	return monc->client->have_fsid && monc->auth->global_id > 0;
349}
350
351/*
352 * The monitor responds with mount ack indicate mount success.  The
353 * included client ticket allows the client to talk to MDSs and OSDs.
354 */
355static void ceph_monc_handle_map(struct ceph_mon_client *monc,
356				 struct ceph_msg *msg)
357{
358	struct ceph_client *client = monc->client;
359	struct ceph_monmap *monmap = NULL, *old = monc->monmap;
360	void *p, *end;
361	int had_debugfs_info, init_debugfs = 0;
362
363	mutex_lock(&monc->mutex);
364
365	had_debugfs_info = have_debugfs_info(monc);
366
367	dout("handle_monmap\n");
368	p = msg->front.iov_base;
369	end = p + msg->front.iov_len;
370
371	monmap = ceph_monmap_decode(p, end);
372	if (IS_ERR(monmap)) {
373		pr_err("problem decoding monmap, %d\n",
374		       (int)PTR_ERR(monmap));
375		goto out;
376	}
377
378	if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
379		kfree(monmap);
380		goto out;
381	}
382
383	client->monc.monmap = monmap;
384	kfree(old);
385
386	if (!client->have_fsid) {
387		client->have_fsid = true;
388		if (!had_debugfs_info && have_debugfs_info(monc)) {
389			pr_info("client%lld fsid %pU\n",
390				ceph_client_id(monc->client),
391				&monc->client->fsid);
392			init_debugfs = 1;
393		}
394		mutex_unlock(&monc->mutex);
395
396		if (init_debugfs) {
397			/*
398			 * do debugfs initialization without mutex to avoid
399			 * creating a locking dependency
400			 */
401			ceph_debugfs_client_init(monc->client);
402		}
403
404		goto out_unlocked;
405	}
406out:
407	mutex_unlock(&monc->mutex);
408out_unlocked:
409	wake_up_all(&client->auth_wq);
410}
411
412/*
413 * generic requests (e.g., statfs, poolop)
414 */
415static struct ceph_mon_generic_request *__lookup_generic_req(
416	struct ceph_mon_client *monc, u64 tid)
417{
418	struct ceph_mon_generic_request *req;
419	struct rb_node *n = monc->generic_request_tree.rb_node;
420
421	while (n) {
422		req = rb_entry(n, struct ceph_mon_generic_request, node);
423		if (tid < req->tid)
424			n = n->rb_left;
425		else if (tid > req->tid)
426			n = n->rb_right;
427		else
428			return req;
429	}
430	return NULL;
431}
432
433static void __insert_generic_request(struct ceph_mon_client *monc,
434			    struct ceph_mon_generic_request *new)
435{
436	struct rb_node **p = &monc->generic_request_tree.rb_node;
437	struct rb_node *parent = NULL;
438	struct ceph_mon_generic_request *req = NULL;
439
440	while (*p) {
441		parent = *p;
442		req = rb_entry(parent, struct ceph_mon_generic_request, node);
443		if (new->tid < req->tid)
444			p = &(*p)->rb_left;
445		else if (new->tid > req->tid)
446			p = &(*p)->rb_right;
447		else
448			BUG();
449	}
450
451	rb_link_node(&new->node, parent, p);
452	rb_insert_color(&new->node, &monc->generic_request_tree);
453}
454
455static void release_generic_request(struct kref *kref)
456{
457	struct ceph_mon_generic_request *req =
458		container_of(kref, struct ceph_mon_generic_request, kref);
459
460	if (req->reply)
461		ceph_msg_put(req->reply);
462	if (req->request)
463		ceph_msg_put(req->request);
464
465	kfree(req);
466}
467
468static void put_generic_request(struct ceph_mon_generic_request *req)
469{
470	kref_put(&req->kref, release_generic_request);
471}
472
473static void get_generic_request(struct ceph_mon_generic_request *req)
474{
475	kref_get(&req->kref);
476}
477
478static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
479					 struct ceph_msg_header *hdr,
480					 int *skip)
481{
482	struct ceph_mon_client *monc = con->private;
483	struct ceph_mon_generic_request *req;
484	u64 tid = le64_to_cpu(hdr->tid);
485	struct ceph_msg *m;
486
487	mutex_lock(&monc->mutex);
488	req = __lookup_generic_req(monc, tid);
489	if (!req) {
490		dout("get_generic_reply %lld dne\n", tid);
491		*skip = 1;
492		m = NULL;
493	} else {
494		dout("get_generic_reply %lld got %p\n", tid, req->reply);
495		*skip = 0;
496		m = ceph_msg_get(req->reply);
497		/*
498		 * we don't need to track the connection reading into
499		 * this reply because we only have one open connection
500		 * at a time, ever.
501		 */
502	}
503	mutex_unlock(&monc->mutex);
504	return m;
505}
506
507static int __do_generic_request(struct ceph_mon_client *monc, u64 tid,
508				struct ceph_mon_generic_request *req)
509{
510	int err;
511
512	/* register request */
513	req->tid = tid != 0 ? tid : ++monc->last_tid;
514	req->request->hdr.tid = cpu_to_le64(req->tid);
515	__insert_generic_request(monc, req);
516	monc->num_generic_requests++;
517	ceph_con_send(&monc->con, ceph_msg_get(req->request));
518	mutex_unlock(&monc->mutex);
519
520	err = wait_for_completion_interruptible(&req->completion);
521
522	mutex_lock(&monc->mutex);
523	rb_erase(&req->node, &monc->generic_request_tree);
524	monc->num_generic_requests--;
525
526	if (!err)
527		err = req->result;
528	return err;
529}
530
531static int do_generic_request(struct ceph_mon_client *monc,
532			      struct ceph_mon_generic_request *req)
533{
534	int err;
535
536	mutex_lock(&monc->mutex);
537	err = __do_generic_request(monc, 0, req);
538	mutex_unlock(&monc->mutex);
539
540	return err;
541}
542
543/*
544 * statfs
545 */
546static void handle_statfs_reply(struct ceph_mon_client *monc,
547				struct ceph_msg *msg)
548{
549	struct ceph_mon_generic_request *req;
550	struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
551	u64 tid = le64_to_cpu(msg->hdr.tid);
552
553	if (msg->front.iov_len != sizeof(*reply))
554		goto bad;
555	dout("handle_statfs_reply %p tid %llu\n", msg, tid);
556
557	mutex_lock(&monc->mutex);
558	req = __lookup_generic_req(monc, tid);
559	if (req) {
560		*(struct ceph_statfs *)req->buf = reply->st;
561		req->result = 0;
562		get_generic_request(req);
563	}
564	mutex_unlock(&monc->mutex);
565	if (req) {
566		complete_all(&req->completion);
567		put_generic_request(req);
568	}
569	return;
570
571bad:
572	pr_err("corrupt generic reply, tid %llu\n", tid);
573	ceph_msg_dump(msg);
574}
575
576/*
577 * Do a synchronous statfs().
578 */
579int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
580{
581	struct ceph_mon_generic_request *req;
582	struct ceph_mon_statfs *h;
583	int err;
584
585	req = kzalloc(sizeof(*req), GFP_NOFS);
586	if (!req)
587		return -ENOMEM;
588
589	kref_init(&req->kref);
590	req->buf = buf;
591	req->buf_len = sizeof(*buf);
592	init_completion(&req->completion);
593
594	err = -ENOMEM;
595	req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
596				    true);
597	if (!req->request)
598		goto out;
599	req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
600				  true);
601	if (!req->reply)
602		goto out;
603
604	/* fill out request */
605	h = req->request->front.iov_base;
606	h->monhdr.have_version = 0;
607	h->monhdr.session_mon = cpu_to_le16(-1);
608	h->monhdr.session_mon_tid = 0;
609	h->fsid = monc->monmap->fsid;
610
611	err = do_generic_request(monc, req);
612
613out:
614	kref_put(&req->kref, release_generic_request);
615	return err;
616}
617EXPORT_SYMBOL(ceph_monc_do_statfs);
618
619static void handle_get_version_reply(struct ceph_mon_client *monc,
620				     struct ceph_msg *msg)
621{
622	struct ceph_mon_generic_request *req;
623	u64 tid = le64_to_cpu(msg->hdr.tid);
624	void *p = msg->front.iov_base;
625	void *end = p + msg->front_alloc_len;
626	u64 handle;
627
628	dout("%s %p tid %llu\n", __func__, msg, tid);
629
630	ceph_decode_need(&p, end, 2*sizeof(u64), bad);
631	handle = ceph_decode_64(&p);
632	if (tid != 0 && tid != handle)
633		goto bad;
634
635	mutex_lock(&monc->mutex);
636	req = __lookup_generic_req(monc, handle);
637	if (req) {
638		*(u64 *)req->buf = ceph_decode_64(&p);
639		req->result = 0;
640		get_generic_request(req);
641	}
642	mutex_unlock(&monc->mutex);
643	if (req) {
644		complete_all(&req->completion);
645		put_generic_request(req);
646	}
647
648	return;
649bad:
650	pr_err("corrupt mon_get_version reply\n");
651	ceph_msg_dump(msg);
652}
653
654/*
655 * Send MMonGetVersion and wait for the reply.
656 *
657 * @what: one of "mdsmap", "osdmap" or "monmap"
658 */
659int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what,
660			     u64 *newest)
661{
662	struct ceph_mon_generic_request *req;
663	void *p, *end;
664	u64 tid;
665	int err;
666
667	req = kzalloc(sizeof(*req), GFP_NOFS);
668	if (!req)
669		return -ENOMEM;
670
671	kref_init(&req->kref);
672	req->buf = newest;
673	req->buf_len = sizeof(*newest);
674	init_completion(&req->completion);
675
676	req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
677				    sizeof(u64) + sizeof(u32) + strlen(what),
678				    GFP_NOFS, true);
679	if (!req->request) {
680		err = -ENOMEM;
681		goto out;
682	}
683
684	req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024,
685				  GFP_NOFS, true);
686	if (!req->reply) {
687		err = -ENOMEM;
688		goto out;
689	}
690
691	p = req->request->front.iov_base;
692	end = p + req->request->front_alloc_len;
693
694	/* fill out request */
695	mutex_lock(&monc->mutex);
696	tid = ++monc->last_tid;
697	ceph_encode_64(&p, tid); /* handle */
698	ceph_encode_string(&p, end, what, strlen(what));
699
700	err = __do_generic_request(monc, tid, req);
701
702	mutex_unlock(&monc->mutex);
703out:
704	kref_put(&req->kref, release_generic_request);
705	return err;
706}
707EXPORT_SYMBOL(ceph_monc_do_get_version);
708
709/*
710 * pool ops
711 */
712static int get_poolop_reply_buf(const char *src, size_t src_len,
713				char *dst, size_t dst_len)
714{
715	u32 buf_len;
716
717	if (src_len != sizeof(u32) + dst_len)
718		return -EINVAL;
719
720	buf_len = le32_to_cpu(*(u32 *)src);
721	if (buf_len != dst_len)
722		return -EINVAL;
723
724	memcpy(dst, src + sizeof(u32), dst_len);
725	return 0;
726}
727
728static void handle_poolop_reply(struct ceph_mon_client *monc,
729				struct ceph_msg *msg)
730{
731	struct ceph_mon_generic_request *req;
732	struct ceph_mon_poolop_reply *reply = msg->front.iov_base;
733	u64 tid = le64_to_cpu(msg->hdr.tid);
734
735	if (msg->front.iov_len < sizeof(*reply))
736		goto bad;
737	dout("handle_poolop_reply %p tid %llu\n", msg, tid);
738
739	mutex_lock(&monc->mutex);
740	req = __lookup_generic_req(monc, tid);
741	if (req) {
742		if (req->buf_len &&
743		    get_poolop_reply_buf(msg->front.iov_base + sizeof(*reply),
744				     msg->front.iov_len - sizeof(*reply),
745				     req->buf, req->buf_len) < 0) {
746			mutex_unlock(&monc->mutex);
747			goto bad;
748		}
749		req->result = le32_to_cpu(reply->reply_code);
750		get_generic_request(req);
751	}
752	mutex_unlock(&monc->mutex);
753	if (req) {
754		complete(&req->completion);
755		put_generic_request(req);
756	}
757	return;
758
759bad:
760	pr_err("corrupt generic reply, tid %llu\n", tid);
761	ceph_msg_dump(msg);
762}
763
764/*
765 * Do a synchronous pool op.
766 */
767static int do_poolop(struct ceph_mon_client *monc, u32 op,
768			u32 pool, u64 snapid,
769			char *buf, int len)
770{
771	struct ceph_mon_generic_request *req;
772	struct ceph_mon_poolop *h;
773	int err;
774
775	req = kzalloc(sizeof(*req), GFP_NOFS);
776	if (!req)
777		return -ENOMEM;
778
779	kref_init(&req->kref);
780	req->buf = buf;
781	req->buf_len = len;
782	init_completion(&req->completion);
783
784	err = -ENOMEM;
785	req->request = ceph_msg_new(CEPH_MSG_POOLOP, sizeof(*h), GFP_NOFS,
786				    true);
787	if (!req->request)
788		goto out;
789	req->reply = ceph_msg_new(CEPH_MSG_POOLOP_REPLY, 1024, GFP_NOFS,
790				  true);
791	if (!req->reply)
792		goto out;
793
794	/* fill out request */
795	req->request->hdr.version = cpu_to_le16(2);
796	h = req->request->front.iov_base;
797	h->monhdr.have_version = 0;
798	h->monhdr.session_mon = cpu_to_le16(-1);
799	h->monhdr.session_mon_tid = 0;
800	h->fsid = monc->monmap->fsid;
801	h->pool = cpu_to_le32(pool);
802	h->op = cpu_to_le32(op);
803	h->auid = 0;
804	h->snapid = cpu_to_le64(snapid);
805	h->name_len = 0;
806
807	err = do_generic_request(monc, req);
808
809out:
810	kref_put(&req->kref, release_generic_request);
811	return err;
812}
813
814int ceph_monc_create_snapid(struct ceph_mon_client *monc,
815			    u32 pool, u64 *snapid)
816{
817	return do_poolop(monc,  POOL_OP_CREATE_UNMANAGED_SNAP,
818				   pool, 0, (char *)snapid, sizeof(*snapid));
819
820}
821EXPORT_SYMBOL(ceph_monc_create_snapid);
822
823int ceph_monc_delete_snapid(struct ceph_mon_client *monc,
824			    u32 pool, u64 snapid)
825{
826	return do_poolop(monc,  POOL_OP_CREATE_UNMANAGED_SNAP,
827				   pool, snapid, NULL, 0);
828
829}
830
831/*
832 * Resend pending generic requests.
833 */
834static void __resend_generic_request(struct ceph_mon_client *monc)
835{
836	struct ceph_mon_generic_request *req;
837	struct rb_node *p;
838
839	for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
840		req = rb_entry(p, struct ceph_mon_generic_request, node);
841		ceph_msg_revoke(req->request);
842		ceph_msg_revoke_incoming(req->reply);
843		ceph_con_send(&monc->con, ceph_msg_get(req->request));
844	}
845}
846
847/*
848 * Delayed work.  If we haven't mounted yet, retry.  Otherwise,
849 * renew/retry subscription as needed (in case it is timing out, or we
850 * got an ENOMEM).  And keep the monitor connection alive.
851 */
852static void delayed_work(struct work_struct *work)
853{
854	struct ceph_mon_client *monc =
855		container_of(work, struct ceph_mon_client, delayed_work.work);
856
857	dout("monc delayed_work\n");
858	mutex_lock(&monc->mutex);
859	if (monc->hunting) {
860		__close_session(monc);
861		__open_session(monc);  /* continue hunting */
862	} else {
863		ceph_con_keepalive(&monc->con);
864
865		__validate_auth(monc);
866
867		if (ceph_auth_is_authenticated(monc->auth))
868			__send_subscribe(monc);
869	}
870	__schedule_delayed(monc);
871	mutex_unlock(&monc->mutex);
872}
873
874/*
875 * On startup, we build a temporary monmap populated with the IPs
876 * provided by mount(2).
877 */
878static int build_initial_monmap(struct ceph_mon_client *monc)
879{
880	struct ceph_options *opt = monc->client->options;
881	struct ceph_entity_addr *mon_addr = opt->mon_addr;
882	int num_mon = opt->num_mon;
883	int i;
884
885	/* build initial monmap */
886	monc->monmap = kzalloc(sizeof(*monc->monmap) +
887			       num_mon*sizeof(monc->monmap->mon_inst[0]),
888			       GFP_KERNEL);
889	if (!monc->monmap)
890		return -ENOMEM;
891	for (i = 0; i < num_mon; i++) {
892		monc->monmap->mon_inst[i].addr = mon_addr[i];
893		monc->monmap->mon_inst[i].addr.nonce = 0;
894		monc->monmap->mon_inst[i].name.type =
895			CEPH_ENTITY_TYPE_MON;
896		monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
897	}
898	monc->monmap->num_mon = num_mon;
899	return 0;
900}
901
902int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
903{
904	int err = 0;
905
906	dout("init\n");
907	memset(monc, 0, sizeof(*monc));
908	monc->client = cl;
909	monc->monmap = NULL;
910	mutex_init(&monc->mutex);
911
912	err = build_initial_monmap(monc);
913	if (err)
914		goto out;
915
916	/* connection */
917	/* authentication */
918	monc->auth = ceph_auth_init(cl->options->name,
919				    cl->options->key);
920	if (IS_ERR(monc->auth)) {
921		err = PTR_ERR(monc->auth);
922		goto out_monmap;
923	}
924	monc->auth->want_keys =
925		CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
926		CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
927
928	/* msgs */
929	err = -ENOMEM;
930	monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
931				     sizeof(struct ceph_mon_subscribe_ack),
932				     GFP_NOFS, true);
933	if (!monc->m_subscribe_ack)
934		goto out_auth;
935
936	monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
937					 true);
938	if (!monc->m_subscribe)
939		goto out_subscribe_ack;
940
941	monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
942					  true);
943	if (!monc->m_auth_reply)
944		goto out_subscribe;
945
946	monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
947	monc->pending_auth = 0;
948	if (!monc->m_auth)
949		goto out_auth_reply;
950
951	ceph_con_init(&monc->con, monc, &mon_con_ops,
952		      &monc->client->msgr);
953
954	monc->cur_mon = -1;
955	monc->hunting = true;
956	monc->sub_renew_after = jiffies;
957	monc->sub_sent = 0;
958
959	INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
960	monc->generic_request_tree = RB_ROOT;
961	monc->num_generic_requests = 0;
962	monc->last_tid = 0;
963
964	monc->have_mdsmap = 0;
965	monc->have_osdmap = 0;
966	monc->want_next_osdmap = 1;
967	return 0;
968
969out_auth_reply:
970	ceph_msg_put(monc->m_auth_reply);
971out_subscribe:
972	ceph_msg_put(monc->m_subscribe);
973out_subscribe_ack:
974	ceph_msg_put(monc->m_subscribe_ack);
975out_auth:
976	ceph_auth_destroy(monc->auth);
977out_monmap:
978	kfree(monc->monmap);
979out:
980	return err;
981}
982EXPORT_SYMBOL(ceph_monc_init);
983
984void ceph_monc_stop(struct ceph_mon_client *monc)
985{
986	dout("stop\n");
987	cancel_delayed_work_sync(&monc->delayed_work);
988
989	mutex_lock(&monc->mutex);
990	__close_session(monc);
991
992	mutex_unlock(&monc->mutex);
993
994	/*
995	 * flush msgr queue before we destroy ourselves to ensure that:
996	 *  - any work that references our embedded con is finished.
997	 *  - any osd_client or other work that may reference an authorizer
998	 *    finishes before we shut down the auth subsystem.
999	 */
1000	ceph_msgr_flush();
1001
1002	ceph_auth_destroy(monc->auth);
1003
1004	ceph_msg_put(monc->m_auth);
1005	ceph_msg_put(monc->m_auth_reply);
1006	ceph_msg_put(monc->m_subscribe);
1007	ceph_msg_put(monc->m_subscribe_ack);
1008
1009	kfree(monc->monmap);
1010}
1011EXPORT_SYMBOL(ceph_monc_stop);
1012
1013static void handle_auth_reply(struct ceph_mon_client *monc,
1014			      struct ceph_msg *msg)
1015{
1016	int ret;
1017	int was_auth = 0;
1018	int had_debugfs_info, init_debugfs = 0;
1019
1020	mutex_lock(&monc->mutex);
1021	had_debugfs_info = have_debugfs_info(monc);
1022	was_auth = ceph_auth_is_authenticated(monc->auth);
1023	monc->pending_auth = 0;
1024	ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
1025				     msg->front.iov_len,
1026				     monc->m_auth->front.iov_base,
1027				     monc->m_auth->front_alloc_len);
1028	if (ret < 0) {
1029		monc->client->auth_err = ret;
1030		wake_up_all(&monc->client->auth_wq);
1031	} else if (ret > 0) {
1032		__send_prepared_auth_request(monc, ret);
1033	} else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
1034		dout("authenticated, starting session\n");
1035
1036		monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
1037		monc->client->msgr.inst.name.num =
1038					cpu_to_le64(monc->auth->global_id);
1039
1040		__send_subscribe(monc);
1041		__resend_generic_request(monc);
1042	}
1043
1044	if (!had_debugfs_info && have_debugfs_info(monc)) {
1045		pr_info("client%lld fsid %pU\n",
1046			ceph_client_id(monc->client),
1047			&monc->client->fsid);
1048		init_debugfs = 1;
1049	}
1050	mutex_unlock(&monc->mutex);
1051
1052	if (init_debugfs) {
1053		/*
1054		 * do debugfs initialization without mutex to avoid
1055		 * creating a locking dependency
1056		 */
1057		ceph_debugfs_client_init(monc->client);
1058	}
1059}
1060
1061static int __validate_auth(struct ceph_mon_client *monc)
1062{
1063	int ret;
1064
1065	if (monc->pending_auth)
1066		return 0;
1067
1068	ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
1069			      monc->m_auth->front_alloc_len);
1070	if (ret <= 0)
1071		return ret; /* either an error, or no need to authenticate */
1072	__send_prepared_auth_request(monc, ret);
1073	return 0;
1074}
1075
1076int ceph_monc_validate_auth(struct ceph_mon_client *monc)
1077{
1078	int ret;
1079
1080	mutex_lock(&monc->mutex);
1081	ret = __validate_auth(monc);
1082	mutex_unlock(&monc->mutex);
1083	return ret;
1084}
1085EXPORT_SYMBOL(ceph_monc_validate_auth);
1086
1087/*
1088 * handle incoming message
1089 */
1090static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1091{
1092	struct ceph_mon_client *monc = con->private;
1093	int type = le16_to_cpu(msg->hdr.type);
1094
1095	if (!monc)
1096		return;
1097
1098	switch (type) {
1099	case CEPH_MSG_AUTH_REPLY:
1100		handle_auth_reply(monc, msg);
1101		break;
1102
1103	case CEPH_MSG_MON_SUBSCRIBE_ACK:
1104		handle_subscribe_ack(monc, msg);
1105		break;
1106
1107	case CEPH_MSG_STATFS_REPLY:
1108		handle_statfs_reply(monc, msg);
1109		break;
1110
1111	case CEPH_MSG_MON_GET_VERSION_REPLY:
1112		handle_get_version_reply(monc, msg);
1113		break;
1114
1115	case CEPH_MSG_POOLOP_REPLY:
1116		handle_poolop_reply(monc, msg);
1117		break;
1118
1119	case CEPH_MSG_MON_MAP:
1120		ceph_monc_handle_map(monc, msg);
1121		break;
1122
1123	case CEPH_MSG_OSD_MAP:
1124		ceph_osdc_handle_map(&monc->client->osdc, msg);
1125		break;
1126
1127	default:
1128		/* can the chained handler handle it? */
1129		if (monc->client->extra_mon_dispatch &&
1130		    monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1131			break;
1132
1133		pr_err("received unknown message type %d %s\n", type,
1134		       ceph_msg_type_name(type));
1135	}
1136	ceph_msg_put(msg);
1137}
1138
1139/*
1140 * Allocate memory for incoming message
1141 */
1142static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1143				      struct ceph_msg_header *hdr,
1144				      int *skip)
1145{
1146	struct ceph_mon_client *monc = con->private;
1147	int type = le16_to_cpu(hdr->type);
1148	int front_len = le32_to_cpu(hdr->front_len);
1149	struct ceph_msg *m = NULL;
1150
1151	*skip = 0;
1152
1153	switch (type) {
1154	case CEPH_MSG_MON_SUBSCRIBE_ACK:
1155		m = ceph_msg_get(monc->m_subscribe_ack);
1156		break;
1157	case CEPH_MSG_POOLOP_REPLY:
1158	case CEPH_MSG_STATFS_REPLY:
1159		return get_generic_reply(con, hdr, skip);
1160	case CEPH_MSG_AUTH_REPLY:
1161		m = ceph_msg_get(monc->m_auth_reply);
1162		break;
1163	case CEPH_MSG_MON_GET_VERSION_REPLY:
1164		if (le64_to_cpu(hdr->tid) != 0)
1165			return get_generic_reply(con, hdr, skip);
1166
1167		/*
1168		 * Older OSDs don't set reply tid even if the orignal
1169		 * request had a non-zero tid.  Workaround this weirdness
1170		 * by falling through to the allocate case.
1171		 */
1172	case CEPH_MSG_MON_MAP:
1173	case CEPH_MSG_MDS_MAP:
1174	case CEPH_MSG_OSD_MAP:
1175		m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1176		if (!m)
1177			return NULL;	/* ENOMEM--return skip == 0 */
1178		break;
1179	}
1180
1181	if (!m) {
1182		pr_info("alloc_msg unknown type %d\n", type);
1183		*skip = 1;
1184	} else if (front_len > m->front_alloc_len) {
1185		pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1186			front_len, m->front_alloc_len,
1187			(unsigned int)con->peer_name.type,
1188			le64_to_cpu(con->peer_name.num));
1189		ceph_msg_put(m);
1190		m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1191	}
1192
1193	return m;
1194}
1195
1196/*
1197 * If the monitor connection resets, pick a new monitor and resubmit
1198 * any pending requests.
1199 */
1200static void mon_fault(struct ceph_connection *con)
1201{
1202	struct ceph_mon_client *monc = con->private;
1203
1204	if (!monc)
1205		return;
1206
1207	dout("mon_fault\n");
1208	mutex_lock(&monc->mutex);
1209	if (!con->private)
1210		goto out;
1211
1212	if (!monc->hunting)
1213		pr_info("mon%d %s session lost, "
1214			"hunting for new mon\n", monc->cur_mon,
1215			ceph_pr_addr(&monc->con.peer_addr.in_addr));
1216
1217	__close_session(monc);
1218	if (!monc->hunting) {
1219		/* start hunting */
1220		monc->hunting = true;
1221		__open_session(monc);
1222	} else {
1223		/* already hunting, let's wait a bit */
1224		__schedule_delayed(monc);
1225	}
1226out:
1227	mutex_unlock(&monc->mutex);
1228}
1229
1230/*
1231 * We can ignore refcounting on the connection struct, as all references
1232 * will come from the messenger workqueue, which is drained prior to
1233 * mon_client destruction.
1234 */
1235static struct ceph_connection *con_get(struct ceph_connection *con)
1236{
1237	return con;
1238}
1239
1240static void con_put(struct ceph_connection *con)
1241{
1242}
1243
1244static const struct ceph_connection_operations mon_con_ops = {
1245	.get = con_get,
1246	.put = con_put,
1247	.dispatch = dispatch,
1248	.fault = mon_fault,
1249	.alloc_msg = mon_alloc_msg,
1250};
1251