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_con_close(&monc->con);
123 monc->pending_auth = 0;
124 ceph_auth_reset(monc->auth);
128 * Open a session with a (new) monitor.
130 static int __open_session(struct ceph_mon_client *monc)
135 if (monc->cur_mon < 0) {
136 get_random_bytes(&r, 1);
137 monc->cur_mon = r % monc->monmap->num_mon;
138 dout("open_session num=%d r=%d -> mon%d\n",
139 monc->monmap->num_mon, r, monc->cur_mon);
141 monc->sub_renew_after = jiffies; /* i.e., expired */
142 monc->want_next_osdmap = !!monc->want_next_osdmap;
144 dout("open_session mon%d opening\n", monc->cur_mon);
145 ceph_con_open(&monc->con,
146 CEPH_ENTITY_TYPE_MON, monc->cur_mon,
147 &monc->monmap->mon_inst[monc->cur_mon].addr);
149 /* initiatiate authentication handshake */
150 ret = ceph_auth_build_hello(monc->auth,
151 monc->m_auth->front.iov_base,
152 monc->m_auth->front_max);
153 __send_prepared_auth_request(monc, ret);
155 dout("open_session mon%d already open\n", monc->cur_mon);
160 static bool __sub_expired(struct ceph_mon_client *monc)
162 return time_after_eq(jiffies, monc->sub_renew_after);
166 * Reschedule delayed work timer.
168 static void __schedule_delayed(struct ceph_mon_client *monc)
172 if (monc->cur_mon < 0 || __sub_expired(monc))
176 dout("__schedule_delayed after %u\n", delay);
177 schedule_delayed_work(&monc->delayed_work, delay);
181 * Send subscribe request for mdsmap and/or osdmap.
183 static void __send_subscribe(struct ceph_mon_client *monc)
185 dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n",
186 (unsigned int)monc->sub_sent, __sub_expired(monc),
187 monc->want_next_osdmap);
188 if ((__sub_expired(monc) && !monc->sub_sent) ||
189 monc->want_next_osdmap == 1) {
190 struct ceph_msg *msg = monc->m_subscribe;
191 struct ceph_mon_subscribe_item *i;
195 p = msg->front.iov_base;
196 end = p + msg->front_max;
198 num = 1 + !!monc->want_next_osdmap + !!monc->want_mdsmap;
199 ceph_encode_32(&p, num);
201 if (monc->want_next_osdmap) {
202 dout("__send_subscribe to 'osdmap' %u\n",
203 (unsigned int)monc->have_osdmap);
204 ceph_encode_string(&p, end, "osdmap", 6);
206 i->have = cpu_to_le64(monc->have_osdmap);
209 monc->want_next_osdmap = 2; /* requested */
211 if (monc->want_mdsmap) {
212 dout("__send_subscribe to 'mdsmap' %u+\n",
213 (unsigned int)monc->have_mdsmap);
214 ceph_encode_string(&p, end, "mdsmap", 6);
216 i->have = cpu_to_le64(monc->have_mdsmap);
220 ceph_encode_string(&p, end, "monmap", 6);
226 msg->front.iov_len = p - msg->front.iov_base;
227 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
228 ceph_msg_revoke(msg);
229 ceph_con_send(&monc->con, ceph_msg_get(msg));
231 monc->sub_sent = jiffies | 1; /* never 0 */
235 static void handle_subscribe_ack(struct ceph_mon_client *monc,
236 struct ceph_msg *msg)
238 unsigned int seconds;
239 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
241 if (msg->front.iov_len < sizeof(*h))
243 seconds = le32_to_cpu(h->duration);
245 mutex_lock(&monc->mutex);
247 pr_info("mon%d %s session established\n",
249 ceph_pr_addr(&monc->con.peer_addr.in_addr));
250 monc->hunting = false;
252 dout("handle_subscribe_ack after %d seconds\n", seconds);
253 monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1;
255 mutex_unlock(&monc->mutex);
258 pr_err("got corrupt subscribe-ack msg\n");
263 * Keep track of which maps we have
265 int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got)
267 mutex_lock(&monc->mutex);
268 monc->have_mdsmap = got;
269 mutex_unlock(&monc->mutex);
272 EXPORT_SYMBOL(ceph_monc_got_mdsmap);
274 int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got)
276 mutex_lock(&monc->mutex);
277 monc->have_osdmap = got;
278 monc->want_next_osdmap = 0;
279 mutex_unlock(&monc->mutex);
284 * Register interest in the next osdmap
286 void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
288 dout("request_next_osdmap have %u\n", monc->have_osdmap);
289 mutex_lock(&monc->mutex);
290 if (!monc->want_next_osdmap)
291 monc->want_next_osdmap = 1;
292 if (monc->want_next_osdmap < 2)
293 __send_subscribe(monc);
294 mutex_unlock(&monc->mutex);
300 int ceph_monc_open_session(struct ceph_mon_client *monc)
302 mutex_lock(&monc->mutex);
303 __open_session(monc);
304 __schedule_delayed(monc);
305 mutex_unlock(&monc->mutex);
308 EXPORT_SYMBOL(ceph_monc_open_session);
311 * The monitor responds with mount ack indicate mount success. The
312 * included client ticket allows the client to talk to MDSs and OSDs.
314 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
315 struct ceph_msg *msg)
317 struct ceph_client *client = monc->client;
318 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
321 mutex_lock(&monc->mutex);
323 dout("handle_monmap\n");
324 p = msg->front.iov_base;
325 end = p + msg->front.iov_len;
327 monmap = ceph_monmap_decode(p, end);
328 if (IS_ERR(monmap)) {
329 pr_err("problem decoding monmap, %d\n",
330 (int)PTR_ERR(monmap));
334 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
339 client->monc.monmap = monmap;
342 if (!client->have_fsid) {
343 client->have_fsid = true;
344 mutex_unlock(&monc->mutex);
346 * do debugfs initialization without mutex to avoid
347 * creating a locking dependency
349 ceph_debugfs_client_init(client);
353 mutex_unlock(&monc->mutex);
355 wake_up_all(&client->auth_wq);
359 * generic requests (e.g., statfs, poolop)
361 static struct ceph_mon_generic_request *__lookup_generic_req(
362 struct ceph_mon_client *monc, u64 tid)
364 struct ceph_mon_generic_request *req;
365 struct rb_node *n = monc->generic_request_tree.rb_node;
368 req = rb_entry(n, struct ceph_mon_generic_request, node);
371 else if (tid > req->tid)
379 static void __insert_generic_request(struct ceph_mon_client *monc,
380 struct ceph_mon_generic_request *new)
382 struct rb_node **p = &monc->generic_request_tree.rb_node;
383 struct rb_node *parent = NULL;
384 struct ceph_mon_generic_request *req = NULL;
388 req = rb_entry(parent, struct ceph_mon_generic_request, node);
389 if (new->tid < req->tid)
391 else if (new->tid > req->tid)
397 rb_link_node(&new->node, parent, p);
398 rb_insert_color(&new->node, &monc->generic_request_tree);
401 static void release_generic_request(struct kref *kref)
403 struct ceph_mon_generic_request *req =
404 container_of(kref, struct ceph_mon_generic_request, kref);
407 ceph_msg_put(req->reply);
409 ceph_msg_put(req->request);
414 static void put_generic_request(struct ceph_mon_generic_request *req)
416 kref_put(&req->kref, release_generic_request);
419 static void get_generic_request(struct ceph_mon_generic_request *req)
421 kref_get(&req->kref);
424 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
425 struct ceph_msg_header *hdr,
428 struct ceph_mon_client *monc = con->private;
429 struct ceph_mon_generic_request *req;
430 u64 tid = le64_to_cpu(hdr->tid);
433 mutex_lock(&monc->mutex);
434 req = __lookup_generic_req(monc, tid);
436 dout("get_generic_reply %lld dne\n", tid);
440 dout("get_generic_reply %lld got %p\n", tid, req->reply);
442 m = ceph_msg_get(req->reply);
444 * we don't need to track the connection reading into
445 * this reply because we only have one open connection
449 mutex_unlock(&monc->mutex);
453 static int do_generic_request(struct ceph_mon_client *monc,
454 struct ceph_mon_generic_request *req)
458 /* register request */
459 mutex_lock(&monc->mutex);
460 req->tid = ++monc->last_tid;
461 req->request->hdr.tid = cpu_to_le64(req->tid);
462 __insert_generic_request(monc, req);
463 monc->num_generic_requests++;
464 ceph_con_send(&monc->con, ceph_msg_get(req->request));
465 mutex_unlock(&monc->mutex);
467 err = wait_for_completion_interruptible(&req->completion);
469 mutex_lock(&monc->mutex);
470 rb_erase(&req->node, &monc->generic_request_tree);
471 monc->num_generic_requests--;
472 mutex_unlock(&monc->mutex);
482 static void handle_statfs_reply(struct ceph_mon_client *monc,
483 struct ceph_msg *msg)
485 struct ceph_mon_generic_request *req;
486 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
487 u64 tid = le64_to_cpu(msg->hdr.tid);
489 if (msg->front.iov_len != sizeof(*reply))
491 dout("handle_statfs_reply %p tid %llu\n", msg, tid);
493 mutex_lock(&monc->mutex);
494 req = __lookup_generic_req(monc, tid);
496 *(struct ceph_statfs *)req->buf = reply->st;
498 get_generic_request(req);
500 mutex_unlock(&monc->mutex);
502 complete_all(&req->completion);
503 put_generic_request(req);
508 pr_err("corrupt generic reply, tid %llu\n", tid);
513 * Do a synchronous statfs().
515 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
517 struct ceph_mon_generic_request *req;
518 struct ceph_mon_statfs *h;
521 req = kzalloc(sizeof(*req), GFP_NOFS);
525 kref_init(&req->kref);
527 req->buf_len = sizeof(*buf);
528 init_completion(&req->completion);
531 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
535 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
540 /* fill out request */
541 h = req->request->front.iov_base;
542 h->monhdr.have_version = 0;
543 h->monhdr.session_mon = cpu_to_le16(-1);
544 h->monhdr.session_mon_tid = 0;
545 h->fsid = monc->monmap->fsid;
547 err = do_generic_request(monc, req);
550 kref_put(&req->kref, release_generic_request);
553 EXPORT_SYMBOL(ceph_monc_do_statfs);
558 static int get_poolop_reply_buf(const char *src, size_t src_len,
559 char *dst, size_t dst_len)
563 if (src_len != sizeof(u32) + dst_len)
566 buf_len = le32_to_cpu(*(u32 *)src);
567 if (buf_len != dst_len)
570 memcpy(dst, src + sizeof(u32), dst_len);
574 static void handle_poolop_reply(struct ceph_mon_client *monc,
575 struct ceph_msg *msg)
577 struct ceph_mon_generic_request *req;
578 struct ceph_mon_poolop_reply *reply = msg->front.iov_base;
579 u64 tid = le64_to_cpu(msg->hdr.tid);
581 if (msg->front.iov_len < sizeof(*reply))
583 dout("handle_poolop_reply %p tid %llu\n", msg, tid);
585 mutex_lock(&monc->mutex);
586 req = __lookup_generic_req(monc, tid);
589 get_poolop_reply_buf(msg->front.iov_base + sizeof(*reply),
590 msg->front.iov_len - sizeof(*reply),
591 req->buf, req->buf_len) < 0) {
592 mutex_unlock(&monc->mutex);
595 req->result = le32_to_cpu(reply->reply_code);
596 get_generic_request(req);
598 mutex_unlock(&monc->mutex);
600 complete(&req->completion);
601 put_generic_request(req);
606 pr_err("corrupt generic reply, tid %llu\n", tid);
611 * Do a synchronous pool op.
613 int ceph_monc_do_poolop(struct ceph_mon_client *monc, u32 op,
614 u32 pool, u64 snapid,
617 struct ceph_mon_generic_request *req;
618 struct ceph_mon_poolop *h;
621 req = kzalloc(sizeof(*req), GFP_NOFS);
625 kref_init(&req->kref);
628 init_completion(&req->completion);
631 req->request = ceph_msg_new(CEPH_MSG_POOLOP, sizeof(*h), GFP_NOFS,
635 req->reply = ceph_msg_new(CEPH_MSG_POOLOP_REPLY, 1024, GFP_NOFS,
640 /* fill out request */
641 req->request->hdr.version = cpu_to_le16(2);
642 h = req->request->front.iov_base;
643 h->monhdr.have_version = 0;
644 h->monhdr.session_mon = cpu_to_le16(-1);
645 h->monhdr.session_mon_tid = 0;
646 h->fsid = monc->monmap->fsid;
647 h->pool = cpu_to_le32(pool);
648 h->op = cpu_to_le32(op);
650 h->snapid = cpu_to_le64(snapid);
653 err = do_generic_request(monc, req);
656 kref_put(&req->kref, release_generic_request);
660 int ceph_monc_create_snapid(struct ceph_mon_client *monc,
661 u32 pool, u64 *snapid)
663 return ceph_monc_do_poolop(monc, POOL_OP_CREATE_UNMANAGED_SNAP,
664 pool, 0, (char *)snapid, sizeof(*snapid));
667 EXPORT_SYMBOL(ceph_monc_create_snapid);
669 int ceph_monc_delete_snapid(struct ceph_mon_client *monc,
670 u32 pool, u64 snapid)
672 return ceph_monc_do_poolop(monc, POOL_OP_CREATE_UNMANAGED_SNAP,
678 * Resend pending generic requests.
680 static void __resend_generic_request(struct ceph_mon_client *monc)
682 struct ceph_mon_generic_request *req;
685 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
686 req = rb_entry(p, struct ceph_mon_generic_request, node);
687 ceph_msg_revoke(req->request);
688 ceph_con_send(&monc->con, ceph_msg_get(req->request));
693 * Delayed work. If we haven't mounted yet, retry. Otherwise,
694 * renew/retry subscription as needed (in case it is timing out, or we
695 * got an ENOMEM). And keep the monitor connection alive.
697 static void delayed_work(struct work_struct *work)
699 struct ceph_mon_client *monc =
700 container_of(work, struct ceph_mon_client, delayed_work.work);
702 dout("monc delayed_work\n");
703 mutex_lock(&monc->mutex);
705 __close_session(monc);
706 __open_session(monc); /* continue hunting */
708 ceph_con_keepalive(&monc->con);
710 __validate_auth(monc);
712 if (monc->auth->ops->is_authenticated(monc->auth))
713 __send_subscribe(monc);
715 __schedule_delayed(monc);
716 mutex_unlock(&monc->mutex);
720 * On startup, we build a temporary monmap populated with the IPs
721 * provided by mount(2).
723 static int build_initial_monmap(struct ceph_mon_client *monc)
725 struct ceph_options *opt = monc->client->options;
726 struct ceph_entity_addr *mon_addr = opt->mon_addr;
727 int num_mon = opt->num_mon;
730 /* build initial monmap */
731 monc->monmap = kzalloc(sizeof(*monc->monmap) +
732 num_mon*sizeof(monc->monmap->mon_inst[0]),
736 for (i = 0; i < num_mon; i++) {
737 monc->monmap->mon_inst[i].addr = mon_addr[i];
738 monc->monmap->mon_inst[i].addr.nonce = 0;
739 monc->monmap->mon_inst[i].name.type =
740 CEPH_ENTITY_TYPE_MON;
741 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
743 monc->monmap->num_mon = num_mon;
744 monc->have_fsid = false;
748 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
753 memset(monc, 0, sizeof(*monc));
756 mutex_init(&monc->mutex);
758 err = build_initial_monmap(monc);
764 monc->auth = ceph_auth_init(cl->options->name,
766 if (IS_ERR(monc->auth)) {
767 err = PTR_ERR(monc->auth);
770 monc->auth->want_keys =
771 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
772 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
776 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
777 sizeof(struct ceph_mon_subscribe_ack),
779 if (!monc->m_subscribe_ack)
782 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
784 if (!monc->m_subscribe)
785 goto out_subscribe_ack;
787 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
789 if (!monc->m_auth_reply)
792 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
793 monc->pending_auth = 0;
797 ceph_con_init(&monc->con, monc, &mon_con_ops,
798 &monc->client->msgr);
801 monc->hunting = true;
802 monc->sub_renew_after = jiffies;
805 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
806 monc->generic_request_tree = RB_ROOT;
807 monc->num_generic_requests = 0;
810 monc->have_mdsmap = 0;
811 monc->have_osdmap = 0;
812 monc->want_next_osdmap = 1;
816 ceph_msg_put(monc->m_auth_reply);
818 ceph_msg_put(monc->m_subscribe);
820 ceph_msg_put(monc->m_subscribe_ack);
822 ceph_auth_destroy(monc->auth);
828 EXPORT_SYMBOL(ceph_monc_init);
830 void ceph_monc_stop(struct ceph_mon_client *monc)
833 cancel_delayed_work_sync(&monc->delayed_work);
835 mutex_lock(&monc->mutex);
836 __close_session(monc);
838 mutex_unlock(&monc->mutex);
841 * flush msgr queue before we destroy ourselves to ensure that:
842 * - any work that references our embedded con is finished.
843 * - any osd_client or other work that may reference an authorizer
844 * finishes before we shut down the auth subsystem.
848 ceph_auth_destroy(monc->auth);
850 ceph_msg_put(monc->m_auth);
851 ceph_msg_put(monc->m_auth_reply);
852 ceph_msg_put(monc->m_subscribe);
853 ceph_msg_put(monc->m_subscribe_ack);
857 EXPORT_SYMBOL(ceph_monc_stop);
859 static void handle_auth_reply(struct ceph_mon_client *monc,
860 struct ceph_msg *msg)
865 mutex_lock(&monc->mutex);
867 was_auth = monc->auth->ops->is_authenticated(monc->auth);
868 monc->pending_auth = 0;
869 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
871 monc->m_auth->front.iov_base,
872 monc->m_auth->front_max);
874 monc->client->auth_err = ret;
875 wake_up_all(&monc->client->auth_wq);
876 } else if (ret > 0) {
877 __send_prepared_auth_request(monc, ret);
878 } else if (!was_auth && monc->auth->ops->is_authenticated(monc->auth)) {
879 dout("authenticated, starting session\n");
881 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
882 monc->client->msgr.inst.name.num =
883 cpu_to_le64(monc->auth->global_id);
885 __send_subscribe(monc);
886 __resend_generic_request(monc);
888 mutex_unlock(&monc->mutex);
891 static int __validate_auth(struct ceph_mon_client *monc)
895 if (monc->pending_auth)
898 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
899 monc->m_auth->front_max);
901 return ret; /* either an error, or no need to authenticate */
902 __send_prepared_auth_request(monc, ret);
906 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
910 mutex_lock(&monc->mutex);
911 ret = __validate_auth(monc);
912 mutex_unlock(&monc->mutex);
915 EXPORT_SYMBOL(ceph_monc_validate_auth);
918 * handle incoming message
920 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
922 struct ceph_mon_client *monc = con->private;
923 int type = le16_to_cpu(msg->hdr.type);
929 case CEPH_MSG_AUTH_REPLY:
930 handle_auth_reply(monc, msg);
933 case CEPH_MSG_MON_SUBSCRIBE_ACK:
934 handle_subscribe_ack(monc, msg);
937 case CEPH_MSG_STATFS_REPLY:
938 handle_statfs_reply(monc, msg);
941 case CEPH_MSG_POOLOP_REPLY:
942 handle_poolop_reply(monc, msg);
945 case CEPH_MSG_MON_MAP:
946 ceph_monc_handle_map(monc, msg);
949 case CEPH_MSG_OSD_MAP:
950 ceph_osdc_handle_map(&monc->client->osdc, msg);
954 /* can the chained handler handle it? */
955 if (monc->client->extra_mon_dispatch &&
956 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
959 pr_err("received unknown message type %d %s\n", type,
960 ceph_msg_type_name(type));
966 * Allocate memory for incoming message
968 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
969 struct ceph_msg_header *hdr,
972 struct ceph_mon_client *monc = con->private;
973 int type = le16_to_cpu(hdr->type);
974 int front_len = le32_to_cpu(hdr->front_len);
975 struct ceph_msg *m = NULL;
980 case CEPH_MSG_MON_SUBSCRIBE_ACK:
981 m = ceph_msg_get(monc->m_subscribe_ack);
983 case CEPH_MSG_POOLOP_REPLY:
984 case CEPH_MSG_STATFS_REPLY:
985 return get_generic_reply(con, hdr, skip);
986 case CEPH_MSG_AUTH_REPLY:
987 m = ceph_msg_get(monc->m_auth_reply);
989 case CEPH_MSG_MON_MAP:
990 case CEPH_MSG_MDS_MAP:
991 case CEPH_MSG_OSD_MAP:
992 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
994 return NULL; /* ENOMEM--return skip == 0 */
999 pr_info("alloc_msg unknown type %d\n", type);
1006 * If the monitor connection resets, pick a new monitor and resubmit
1007 * any pending requests.
1009 static void mon_fault(struct ceph_connection *con)
1011 struct ceph_mon_client *monc = con->private;
1016 dout("mon_fault\n");
1017 mutex_lock(&monc->mutex);
1022 pr_info("mon%d %s session lost, "
1023 "hunting for new mon\n", monc->cur_mon,
1024 ceph_pr_addr(&monc->con.peer_addr.in_addr));
1026 __close_session(monc);
1027 if (!monc->hunting) {
1029 monc->hunting = true;
1030 __open_session(monc);
1032 /* already hunting, let's wait a bit */
1033 __schedule_delayed(monc);
1036 mutex_unlock(&monc->mutex);
1040 * We can ignore refcounting on the connection struct, as all references
1041 * will come from the messenger workqueue, which is drained prior to
1042 * mon_client destruction.
1044 static struct ceph_connection *con_get(struct ceph_connection *con)
1049 static void con_put(struct ceph_connection *con)
1053 static const struct ceph_connection_operations mon_con_ops = {
1056 .dispatch = dispatch,
1058 .alloc_msg = mon_alloc_msg,