1 #include <linux/ceph/ceph_debug.h>
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>
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>
16 * Interact with Ceph monitor cluster. Handle requests for new map
17 * versions, and periodically resend as needed. Also implement
18 * statfs() and umount().
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.
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.
33 static const struct ceph_connection_operations mon_con_ops;
35 static int __validate_auth(struct ceph_mon_client *monc);
38 * Decode a monmap blob (e.g., during mount).
40 struct ceph_monmap *ceph_monmap_decode(void *p, void *end)
42 struct ceph_monmap *m = NULL;
44 struct ceph_fsid fsid;
49 ceph_decode_32_safe(&p, end, len, bad);
50 ceph_decode_need(&p, end, len, bad);
52 dout("monmap_decode %p %p len %d\n", p, end, (int)(end-p));
54 ceph_decode_16_safe(&p, end, version, bad);
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);
60 num_mon = ceph_decode_32(&p);
61 ceph_decode_need(&p, end, num_mon*sizeof(m->mon_inst[0]), bad);
63 if (num_mon >= CEPH_MAX_MON)
65 m = kmalloc(sizeof(*m) + sizeof(m->mon_inst[0])*num_mon, GFP_NOFS);
67 return ERR_PTR(-ENOMEM);
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);
75 dout("monmap_decode epoch %d, num_mon %d\n", m->epoch,
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));
83 dout("monmap_decode failed with %d\n", err);
89 * return true if *addr is included in the monmap.
91 int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr)
95 for (i = 0; i < m->num_mon; i++)
96 if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0)
102 * Send an auth request.
104 static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len)
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);
115 * Close monitor session, if any.
117 static void __close_session(struct ceph_mon_client *monc)
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);
126 monc->pending_auth = 0;
127 ceph_auth_reset(monc->auth);
131 * Pick a new monitor at random and set cur_mon. If we are repicking
132 * (i.e. cur_mon is already set), be sure to pick a different one.
134 static void pick_new_mon(struct ceph_mon_client *monc)
136 int old_mon = monc->cur_mon;
138 BUG_ON(monc->monmap->num_mon < 1);
140 if (monc->monmap->num_mon == 1) {
143 int max = monc->monmap->num_mon;
147 if (monc->cur_mon >= 0) {
148 if (monc->cur_mon < monc->monmap->num_mon)
154 n = prandom_u32() % max;
155 if (o >= 0 && n >= o)
161 dout("%s mon%d -> mon%d out of %d mons\n", __func__, old_mon,
162 monc->cur_mon, monc->monmap->num_mon);
166 * Open a session with a new monitor.
168 static void __open_session(struct ceph_mon_client *monc)
174 monc->hunting = true;
175 if (monc->had_a_connection) {
176 monc->hunt_mult *= CEPH_MONC_HUNT_BACKOFF;
177 if (monc->hunt_mult > CEPH_MONC_HUNT_MAX_MULT)
178 monc->hunt_mult = CEPH_MONC_HUNT_MAX_MULT;
181 monc->sub_renew_after = jiffies; /* i.e., expired */
182 monc->sub_renew_sent = 0;
184 dout("%s opening mon%d\n", __func__, monc->cur_mon);
185 ceph_con_open(&monc->con, CEPH_ENTITY_TYPE_MON, monc->cur_mon,
186 &monc->monmap->mon_inst[monc->cur_mon].addr);
189 * send an initial keepalive to ensure our timestamp is valid
190 * by the time we are in an OPENED state
192 ceph_con_keepalive(&monc->con);
194 /* initiate authentication handshake */
195 ret = ceph_auth_build_hello(monc->auth,
196 monc->m_auth->front.iov_base,
197 monc->m_auth->front_alloc_len);
199 __send_prepared_auth_request(monc, ret);
202 static void reopen_session(struct ceph_mon_client *monc)
205 pr_info("mon%d %s session lost, hunting for new mon\n",
206 monc->cur_mon, ceph_pr_addr(&monc->con.peer_addr.in_addr));
208 __close_session(monc);
209 __open_session(monc);
213 * Reschedule delayed work timer.
215 static void __schedule_delayed(struct ceph_mon_client *monc)
220 delay = CEPH_MONC_HUNT_INTERVAL * monc->hunt_mult;
222 delay = CEPH_MONC_PING_INTERVAL;
224 dout("__schedule_delayed after %lu\n", delay);
225 mod_delayed_work(system_wq, &monc->delayed_work,
226 round_jiffies_relative(delay));
229 const char *ceph_sub_str[] = {
230 [CEPH_SUB_MDSMAP] = "mdsmap",
231 [CEPH_SUB_MONMAP] = "monmap",
232 [CEPH_SUB_OSDMAP] = "osdmap",
236 * Send subscribe request for one or more maps, according to
239 static void __send_subscribe(struct ceph_mon_client *monc)
241 struct ceph_msg *msg = monc->m_subscribe;
242 void *p = msg->front.iov_base;
243 void *const end = p + msg->front_alloc_len;
247 dout("%s sent %lu\n", __func__, monc->sub_renew_sent);
249 BUG_ON(monc->cur_mon < 0);
251 if (!monc->sub_renew_sent)
252 monc->sub_renew_sent = jiffies | 1; /* never 0 */
254 msg->hdr.version = cpu_to_le16(2);
256 for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
257 if (monc->subs[i].want)
260 BUG_ON(num < 1); /* monmap sub is always there */
261 ceph_encode_32(&p, num);
262 for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
263 const char *s = ceph_sub_str[i];
265 if (!monc->subs[i].want)
268 dout("%s %s start %llu flags 0x%x\n", __func__, s,
269 le64_to_cpu(monc->subs[i].item.start),
270 monc->subs[i].item.flags);
271 ceph_encode_string(&p, end, s, strlen(s));
272 memcpy(p, &monc->subs[i].item, sizeof(monc->subs[i].item));
273 p += sizeof(monc->subs[i].item);
276 BUG_ON(p != (end - 35 - (ARRAY_SIZE(monc->subs) - num) * 19));
277 msg->front.iov_len = p - msg->front.iov_base;
278 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
279 ceph_msg_revoke(msg);
280 ceph_con_send(&monc->con, ceph_msg_get(msg));
283 static void handle_subscribe_ack(struct ceph_mon_client *monc,
284 struct ceph_msg *msg)
286 unsigned int seconds;
287 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
289 if (msg->front.iov_len < sizeof(*h))
291 seconds = le32_to_cpu(h->duration);
293 mutex_lock(&monc->mutex);
294 if (monc->sub_renew_sent) {
295 monc->sub_renew_after = monc->sub_renew_sent +
296 (seconds >> 1) * HZ - 1;
297 dout("%s sent %lu duration %d renew after %lu\n", __func__,
298 monc->sub_renew_sent, seconds, monc->sub_renew_after);
299 monc->sub_renew_sent = 0;
301 dout("%s sent %lu renew after %lu, ignoring\n", __func__,
302 monc->sub_renew_sent, monc->sub_renew_after);
304 mutex_unlock(&monc->mutex);
307 pr_err("got corrupt subscribe-ack msg\n");
312 * Register interest in a map
314 * @sub: one of CEPH_SUB_*
315 * @epoch: X for "every map since X", or 0 for "just the latest"
317 static bool __ceph_monc_want_map(struct ceph_mon_client *monc, int sub,
318 u32 epoch, bool continuous)
320 __le64 start = cpu_to_le64(epoch);
321 u8 flags = !continuous ? CEPH_SUBSCRIBE_ONETIME : 0;
323 dout("%s %s epoch %u continuous %d\n", __func__, ceph_sub_str[sub],
326 if (monc->subs[sub].want &&
327 monc->subs[sub].item.start == start &&
328 monc->subs[sub].item.flags == flags)
331 monc->subs[sub].item.start = start;
332 monc->subs[sub].item.flags = flags;
333 monc->subs[sub].want = true;
338 bool ceph_monc_want_map(struct ceph_mon_client *monc, int sub, u32 epoch,
343 mutex_lock(&monc->mutex);
344 need_request = __ceph_monc_want_map(monc, sub, epoch, continuous);
345 mutex_unlock(&monc->mutex);
349 EXPORT_SYMBOL(ceph_monc_want_map);
352 * Keep track of which maps we have
354 * @sub: one of CEPH_SUB_*
356 static void __ceph_monc_got_map(struct ceph_mon_client *monc, int sub,
359 dout("%s %s epoch %u\n", __func__, ceph_sub_str[sub], epoch);
361 if (monc->subs[sub].want) {
362 if (monc->subs[sub].item.flags & CEPH_SUBSCRIBE_ONETIME)
363 monc->subs[sub].want = false;
365 monc->subs[sub].item.start = cpu_to_le64(epoch + 1);
368 monc->subs[sub].have = epoch;
371 void ceph_monc_got_map(struct ceph_mon_client *monc, int sub, u32 epoch)
373 mutex_lock(&monc->mutex);
374 __ceph_monc_got_map(monc, sub, epoch);
375 mutex_unlock(&monc->mutex);
377 EXPORT_SYMBOL(ceph_monc_got_map);
379 void ceph_monc_renew_subs(struct ceph_mon_client *monc)
381 mutex_lock(&monc->mutex);
382 __send_subscribe(monc);
383 mutex_unlock(&monc->mutex);
385 EXPORT_SYMBOL(ceph_monc_renew_subs);
388 * Register interest in the next osdmap
390 void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
392 dout("%s have %u\n", __func__, monc->subs[CEPH_SUB_OSDMAP].have);
393 mutex_lock(&monc->mutex);
394 if (__ceph_monc_want_map(monc, CEPH_SUB_OSDMAP,
395 monc->subs[CEPH_SUB_OSDMAP].have + 1, false))
396 __send_subscribe(monc);
397 mutex_unlock(&monc->mutex);
399 EXPORT_SYMBOL(ceph_monc_request_next_osdmap);
402 * Wait for an osdmap with a given epoch.
404 * @epoch: epoch to wait for
405 * @timeout: in jiffies, 0 means "wait forever"
407 int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
408 unsigned long timeout)
410 unsigned long started = jiffies;
413 mutex_lock(&monc->mutex);
414 while (monc->subs[CEPH_SUB_OSDMAP].have < epoch) {
415 mutex_unlock(&monc->mutex);
417 if (timeout && time_after_eq(jiffies, started + timeout))
420 ret = wait_event_interruptible_timeout(monc->client->auth_wq,
421 monc->subs[CEPH_SUB_OSDMAP].have >= epoch,
422 ceph_timeout_jiffies(timeout));
426 mutex_lock(&monc->mutex);
429 mutex_unlock(&monc->mutex);
432 EXPORT_SYMBOL(ceph_monc_wait_osdmap);
435 * Open a session with a random monitor. Request monmap and osdmap,
436 * which are waited upon in __ceph_open_session().
438 int ceph_monc_open_session(struct ceph_mon_client *monc)
440 mutex_lock(&monc->mutex);
441 __ceph_monc_want_map(monc, CEPH_SUB_MONMAP, 0, true);
442 __ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, 0, false);
443 __open_session(monc);
444 __schedule_delayed(monc);
445 mutex_unlock(&monc->mutex);
448 EXPORT_SYMBOL(ceph_monc_open_session);
450 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
451 struct ceph_msg *msg)
453 struct ceph_client *client = monc->client;
454 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
457 mutex_lock(&monc->mutex);
459 dout("handle_monmap\n");
460 p = msg->front.iov_base;
461 end = p + msg->front.iov_len;
463 monmap = ceph_monmap_decode(p, end);
464 if (IS_ERR(monmap)) {
465 pr_err("problem decoding monmap, %d\n",
466 (int)PTR_ERR(monmap));
470 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
475 client->monc.monmap = monmap;
478 __ceph_monc_got_map(monc, CEPH_SUB_MONMAP, monc->monmap->epoch);
479 client->have_fsid = true;
482 mutex_unlock(&monc->mutex);
483 wake_up_all(&client->auth_wq);
487 * generic requests (currently statfs, mon_get_version)
489 DEFINE_RB_FUNCS(generic_request, struct ceph_mon_generic_request, tid, node)
491 static void release_generic_request(struct kref *kref)
493 struct ceph_mon_generic_request *req =
494 container_of(kref, struct ceph_mon_generic_request, kref);
497 ceph_msg_put(req->reply);
499 ceph_msg_put(req->request);
504 static void put_generic_request(struct ceph_mon_generic_request *req)
506 kref_put(&req->kref, release_generic_request);
509 static void get_generic_request(struct ceph_mon_generic_request *req)
511 kref_get(&req->kref);
514 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
515 struct ceph_msg_header *hdr,
518 struct ceph_mon_client *monc = con->private;
519 struct ceph_mon_generic_request *req;
520 u64 tid = le64_to_cpu(hdr->tid);
523 mutex_lock(&monc->mutex);
524 req = lookup_generic_request(&monc->generic_request_tree, tid);
526 dout("get_generic_reply %lld dne\n", tid);
530 dout("get_generic_reply %lld got %p\n", tid, req->reply);
532 m = ceph_msg_get(req->reply);
534 * we don't need to track the connection reading into
535 * this reply because we only have one open connection
539 mutex_unlock(&monc->mutex);
543 static int __do_generic_request(struct ceph_mon_client *monc, u64 tid,
544 struct ceph_mon_generic_request *req)
548 /* register request */
549 req->tid = tid != 0 ? tid : ++monc->last_tid;
550 req->request->hdr.tid = cpu_to_le64(req->tid);
551 insert_generic_request(&monc->generic_request_tree, req);
552 ceph_con_send(&monc->con, ceph_msg_get(req->request));
553 mutex_unlock(&monc->mutex);
555 err = wait_for_completion_interruptible(&req->completion);
557 mutex_lock(&monc->mutex);
558 erase_generic_request(&monc->generic_request_tree, req);
565 static int do_generic_request(struct ceph_mon_client *monc,
566 struct ceph_mon_generic_request *req)
570 mutex_lock(&monc->mutex);
571 err = __do_generic_request(monc, 0, req);
572 mutex_unlock(&monc->mutex);
580 static void handle_statfs_reply(struct ceph_mon_client *monc,
581 struct ceph_msg *msg)
583 struct ceph_mon_generic_request *req;
584 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
585 u64 tid = le64_to_cpu(msg->hdr.tid);
587 if (msg->front.iov_len != sizeof(*reply))
589 dout("handle_statfs_reply %p tid %llu\n", msg, tid);
591 mutex_lock(&monc->mutex);
592 req = lookup_generic_request(&monc->generic_request_tree, tid);
594 *(struct ceph_statfs *)req->buf = reply->st;
596 get_generic_request(req);
598 mutex_unlock(&monc->mutex);
600 complete_all(&req->completion);
601 put_generic_request(req);
606 pr_err("corrupt statfs reply, tid %llu\n", tid);
611 * Do a synchronous statfs().
613 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
615 struct ceph_mon_generic_request *req;
616 struct ceph_mon_statfs *h;
619 req = kzalloc(sizeof(*req), GFP_NOFS);
623 kref_init(&req->kref);
624 RB_CLEAR_NODE(&req->node);
626 init_completion(&req->completion);
629 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
633 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
638 /* fill out request */
639 h = req->request->front.iov_base;
640 h->monhdr.have_version = 0;
641 h->monhdr.session_mon = cpu_to_le16(-1);
642 h->monhdr.session_mon_tid = 0;
643 h->fsid = monc->monmap->fsid;
645 err = do_generic_request(monc, req);
648 put_generic_request(req);
651 EXPORT_SYMBOL(ceph_monc_do_statfs);
653 static void handle_get_version_reply(struct ceph_mon_client *monc,
654 struct ceph_msg *msg)
656 struct ceph_mon_generic_request *req;
657 u64 tid = le64_to_cpu(msg->hdr.tid);
658 void *p = msg->front.iov_base;
659 void *end = p + msg->front_alloc_len;
662 dout("%s %p tid %llu\n", __func__, msg, tid);
664 ceph_decode_need(&p, end, 2*sizeof(u64), bad);
665 handle = ceph_decode_64(&p);
666 if (tid != 0 && tid != handle)
669 mutex_lock(&monc->mutex);
670 req = lookup_generic_request(&monc->generic_request_tree, handle);
672 *(u64 *)req->buf = ceph_decode_64(&p);
674 get_generic_request(req);
676 mutex_unlock(&monc->mutex);
678 complete_all(&req->completion);
679 put_generic_request(req);
684 pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
689 * Send MMonGetVersion and wait for the reply.
691 * @what: one of "mdsmap", "osdmap" or "monmap"
693 int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what,
696 struct ceph_mon_generic_request *req;
701 req = kzalloc(sizeof(*req), GFP_NOFS);
705 kref_init(&req->kref);
706 RB_CLEAR_NODE(&req->node);
708 init_completion(&req->completion);
710 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
711 sizeof(u64) + sizeof(u32) + strlen(what),
718 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024,
725 p = req->request->front.iov_base;
726 end = p + req->request->front_alloc_len;
728 /* fill out request */
729 mutex_lock(&monc->mutex);
730 tid = ++monc->last_tid;
731 ceph_encode_64(&p, tid); /* handle */
732 ceph_encode_string(&p, end, what, strlen(what));
734 err = __do_generic_request(monc, tid, req);
736 mutex_unlock(&monc->mutex);
738 put_generic_request(req);
741 EXPORT_SYMBOL(ceph_monc_do_get_version);
744 * Resend pending generic requests.
746 static void __resend_generic_request(struct ceph_mon_client *monc)
748 struct ceph_mon_generic_request *req;
751 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
752 req = rb_entry(p, struct ceph_mon_generic_request, node);
753 ceph_msg_revoke(req->request);
754 ceph_msg_revoke_incoming(req->reply);
755 ceph_con_send(&monc->con, ceph_msg_get(req->request));
760 * Delayed work. If we haven't mounted yet, retry. Otherwise,
761 * renew/retry subscription as needed (in case it is timing out, or we
762 * got an ENOMEM). And keep the monitor connection alive.
764 static void delayed_work(struct work_struct *work)
766 struct ceph_mon_client *monc =
767 container_of(work, struct ceph_mon_client, delayed_work.work);
769 dout("monc delayed_work\n");
770 mutex_lock(&monc->mutex);
772 dout("%s continuing hunt\n", __func__);
773 reopen_session(monc);
775 int is_auth = ceph_auth_is_authenticated(monc->auth);
776 if (ceph_con_keepalive_expired(&monc->con,
777 CEPH_MONC_PING_TIMEOUT)) {
778 dout("monc keepalive timeout\n");
780 reopen_session(monc);
783 if (!monc->hunting) {
784 ceph_con_keepalive(&monc->con);
785 __validate_auth(monc);
789 unsigned long now = jiffies;
791 dout("%s renew subs? now %lu renew after %lu\n",
792 __func__, now, monc->sub_renew_after);
793 if (time_after_eq(now, monc->sub_renew_after))
794 __send_subscribe(monc);
797 __schedule_delayed(monc);
798 mutex_unlock(&monc->mutex);
802 * On startup, we build a temporary monmap populated with the IPs
803 * provided by mount(2).
805 static int build_initial_monmap(struct ceph_mon_client *monc)
807 struct ceph_options *opt = monc->client->options;
808 struct ceph_entity_addr *mon_addr = opt->mon_addr;
809 int num_mon = opt->num_mon;
812 /* build initial monmap */
813 monc->monmap = kzalloc(sizeof(*monc->monmap) +
814 num_mon*sizeof(monc->monmap->mon_inst[0]),
818 for (i = 0; i < num_mon; i++) {
819 monc->monmap->mon_inst[i].addr = mon_addr[i];
820 monc->monmap->mon_inst[i].addr.nonce = 0;
821 monc->monmap->mon_inst[i].name.type =
822 CEPH_ENTITY_TYPE_MON;
823 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
825 monc->monmap->num_mon = num_mon;
829 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
834 memset(monc, 0, sizeof(*monc));
837 mutex_init(&monc->mutex);
839 err = build_initial_monmap(monc);
845 monc->auth = ceph_auth_init(cl->options->name,
847 if (IS_ERR(monc->auth)) {
848 err = PTR_ERR(monc->auth);
851 monc->auth->want_keys =
852 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
853 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
857 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
858 sizeof(struct ceph_mon_subscribe_ack),
860 if (!monc->m_subscribe_ack)
863 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
865 if (!monc->m_subscribe)
866 goto out_subscribe_ack;
868 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
870 if (!monc->m_auth_reply)
873 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
874 monc->pending_auth = 0;
878 ceph_con_init(&monc->con, monc, &mon_con_ops,
879 &monc->client->msgr);
882 monc->had_a_connection = false;
885 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
886 monc->generic_request_tree = RB_ROOT;
892 ceph_msg_put(monc->m_auth_reply);
894 ceph_msg_put(monc->m_subscribe);
896 ceph_msg_put(monc->m_subscribe_ack);
898 ceph_auth_destroy(monc->auth);
904 EXPORT_SYMBOL(ceph_monc_init);
906 void ceph_monc_stop(struct ceph_mon_client *monc)
909 cancel_delayed_work_sync(&monc->delayed_work);
911 mutex_lock(&monc->mutex);
912 __close_session(monc);
914 mutex_unlock(&monc->mutex);
917 * flush msgr queue before we destroy ourselves to ensure that:
918 * - any work that references our embedded con is finished.
919 * - any osd_client or other work that may reference an authorizer
920 * finishes before we shut down the auth subsystem.
924 ceph_auth_destroy(monc->auth);
926 ceph_msg_put(monc->m_auth);
927 ceph_msg_put(monc->m_auth_reply);
928 ceph_msg_put(monc->m_subscribe);
929 ceph_msg_put(monc->m_subscribe_ack);
933 EXPORT_SYMBOL(ceph_monc_stop);
935 static void finish_hunting(struct ceph_mon_client *monc)
938 dout("%s found mon%d\n", __func__, monc->cur_mon);
939 monc->hunting = false;
940 monc->had_a_connection = true;
941 monc->hunt_mult /= 2; /* reduce by 50% */
942 if (monc->hunt_mult < 1)
947 static void handle_auth_reply(struct ceph_mon_client *monc,
948 struct ceph_msg *msg)
953 mutex_lock(&monc->mutex);
954 was_auth = ceph_auth_is_authenticated(monc->auth);
955 monc->pending_auth = 0;
956 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
958 monc->m_auth->front.iov_base,
959 monc->m_auth->front_alloc_len);
961 __send_prepared_auth_request(monc, ret);
965 finish_hunting(monc);
968 monc->client->auth_err = ret;
969 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
970 dout("authenticated, starting session\n");
972 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
973 monc->client->msgr.inst.name.num =
974 cpu_to_le64(monc->auth->global_id);
976 __send_subscribe(monc);
977 __resend_generic_request(monc);
979 pr_info("mon%d %s session established\n", monc->cur_mon,
980 ceph_pr_addr(&monc->con.peer_addr.in_addr));
984 mutex_unlock(&monc->mutex);
985 if (monc->client->auth_err < 0)
986 wake_up_all(&monc->client->auth_wq);
989 static int __validate_auth(struct ceph_mon_client *monc)
993 if (monc->pending_auth)
996 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
997 monc->m_auth->front_alloc_len);
999 return ret; /* either an error, or no need to authenticate */
1000 __send_prepared_auth_request(monc, ret);
1004 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
1008 mutex_lock(&monc->mutex);
1009 ret = __validate_auth(monc);
1010 mutex_unlock(&monc->mutex);
1013 EXPORT_SYMBOL(ceph_monc_validate_auth);
1016 * handle incoming message
1018 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1020 struct ceph_mon_client *monc = con->private;
1021 int type = le16_to_cpu(msg->hdr.type);
1027 case CEPH_MSG_AUTH_REPLY:
1028 handle_auth_reply(monc, msg);
1031 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1032 handle_subscribe_ack(monc, msg);
1035 case CEPH_MSG_STATFS_REPLY:
1036 handle_statfs_reply(monc, msg);
1039 case CEPH_MSG_MON_GET_VERSION_REPLY:
1040 handle_get_version_reply(monc, msg);
1043 case CEPH_MSG_MON_MAP:
1044 ceph_monc_handle_map(monc, msg);
1047 case CEPH_MSG_OSD_MAP:
1048 ceph_osdc_handle_map(&monc->client->osdc, msg);
1052 /* can the chained handler handle it? */
1053 if (monc->client->extra_mon_dispatch &&
1054 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1057 pr_err("received unknown message type %d %s\n", type,
1058 ceph_msg_type_name(type));
1064 * Allocate memory for incoming message
1066 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1067 struct ceph_msg_header *hdr,
1070 struct ceph_mon_client *monc = con->private;
1071 int type = le16_to_cpu(hdr->type);
1072 int front_len = le32_to_cpu(hdr->front_len);
1073 struct ceph_msg *m = NULL;
1078 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1079 m = ceph_msg_get(monc->m_subscribe_ack);
1081 case CEPH_MSG_STATFS_REPLY:
1082 return get_generic_reply(con, hdr, skip);
1083 case CEPH_MSG_AUTH_REPLY:
1084 m = ceph_msg_get(monc->m_auth_reply);
1086 case CEPH_MSG_MON_GET_VERSION_REPLY:
1087 if (le64_to_cpu(hdr->tid) != 0)
1088 return get_generic_reply(con, hdr, skip);
1091 * Older OSDs don't set reply tid even if the orignal
1092 * request had a non-zero tid. Workaround this weirdness
1093 * by falling through to the allocate case.
1095 case CEPH_MSG_MON_MAP:
1096 case CEPH_MSG_MDS_MAP:
1097 case CEPH_MSG_OSD_MAP:
1098 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1100 return NULL; /* ENOMEM--return skip == 0 */
1105 pr_info("alloc_msg unknown type %d\n", type);
1107 } else if (front_len > m->front_alloc_len) {
1108 pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1109 front_len, m->front_alloc_len,
1110 (unsigned int)con->peer_name.type,
1111 le64_to_cpu(con->peer_name.num));
1113 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1120 * If the monitor connection resets, pick a new monitor and resubmit
1121 * any pending requests.
1123 static void mon_fault(struct ceph_connection *con)
1125 struct ceph_mon_client *monc = con->private;
1127 mutex_lock(&monc->mutex);
1128 dout("%s mon%d\n", __func__, monc->cur_mon);
1129 if (monc->cur_mon >= 0) {
1130 if (!monc->hunting) {
1131 dout("%s hunting for new mon\n", __func__);
1132 reopen_session(monc);
1133 __schedule_delayed(monc);
1135 dout("%s already hunting\n", __func__);
1138 mutex_unlock(&monc->mutex);
1142 * We can ignore refcounting on the connection struct, as all references
1143 * will come from the messenger workqueue, which is drained prior to
1144 * mon_client destruction.
1146 static struct ceph_connection *con_get(struct ceph_connection *con)
1151 static void con_put(struct ceph_connection *con)
1155 static const struct ceph_connection_operations mon_con_ops = {
1158 .dispatch = dispatch,
1160 .alloc_msg = mon_alloc_msg,