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/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_con_revoke(monc->con, 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_con_revoke(monc->con, 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 monc->con->peer_name.type = CEPH_ENTITY_TYPE_MON;
146 monc->con->peer_name.num = cpu_to_le64(monc->cur_mon);
147 ceph_con_open(monc->con,
148 &monc->monmap->mon_inst[monc->cur_mon].addr);
150 /* initiatiate authentication handshake */
151 ret = ceph_auth_build_hello(monc->auth,
152 monc->m_auth->front.iov_base,
153 monc->m_auth->front_max);
154 __send_prepared_auth_request(monc, ret);
156 dout("open_session mon%d already open\n", monc->cur_mon);
161 static bool __sub_expired(struct ceph_mon_client *monc)
163 return time_after_eq(jiffies, monc->sub_renew_after);
167 * Reschedule delayed work timer.
169 static void __schedule_delayed(struct ceph_mon_client *monc)
173 if (monc->cur_mon < 0 || __sub_expired(monc))
177 dout("__schedule_delayed after %u\n", delay);
178 schedule_delayed_work(&monc->delayed_work, delay);
182 * Send subscribe request for mdsmap and/or osdmap.
184 static void __send_subscribe(struct ceph_mon_client *monc)
186 dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n",
187 (unsigned)monc->sub_sent, __sub_expired(monc),
188 monc->want_next_osdmap);
189 if ((__sub_expired(monc) && !monc->sub_sent) ||
190 monc->want_next_osdmap == 1) {
191 struct ceph_msg *msg = monc->m_subscribe;
192 struct ceph_mon_subscribe_item *i;
196 p = msg->front.iov_base;
197 end = p + msg->front_max;
199 num = 1 + !!monc->want_next_osdmap + !!monc->want_mdsmap;
200 ceph_encode_32(&p, num);
202 if (monc->want_next_osdmap) {
203 dout("__send_subscribe to 'osdmap' %u\n",
204 (unsigned)monc->have_osdmap);
205 ceph_encode_string(&p, end, "osdmap", 6);
207 i->have = cpu_to_le64(monc->have_osdmap);
210 monc->want_next_osdmap = 2; /* requested */
212 if (monc->want_mdsmap) {
213 dout("__send_subscribe to 'mdsmap' %u+\n",
214 (unsigned)monc->have_mdsmap);
215 ceph_encode_string(&p, end, "mdsmap", 6);
217 i->have = cpu_to_le64(monc->have_mdsmap);
221 ceph_encode_string(&p, end, "monmap", 6);
227 msg->front.iov_len = p - msg->front.iov_base;
228 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
229 ceph_con_revoke(monc->con, msg);
230 ceph_con_send(monc->con, ceph_msg_get(msg));
232 monc->sub_sent = jiffies | 1; /* never 0 */
236 static void handle_subscribe_ack(struct ceph_mon_client *monc,
237 struct ceph_msg *msg)
240 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
242 if (msg->front.iov_len < sizeof(*h))
244 seconds = le32_to_cpu(h->duration);
246 mutex_lock(&monc->mutex);
248 pr_info("mon%d %s session established\n",
250 ceph_pr_addr(&monc->con->peer_addr.in_addr));
251 monc->hunting = false;
253 dout("handle_subscribe_ack after %d seconds\n", seconds);
254 monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1;
256 mutex_unlock(&monc->mutex);
259 pr_err("got corrupt subscribe-ack msg\n");
264 * Keep track of which maps we have
266 int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got)
268 mutex_lock(&monc->mutex);
269 monc->have_mdsmap = got;
270 mutex_unlock(&monc->mutex);
273 EXPORT_SYMBOL(ceph_monc_got_mdsmap);
275 int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got)
277 mutex_lock(&monc->mutex);
278 monc->have_osdmap = got;
279 monc->want_next_osdmap = 0;
280 mutex_unlock(&monc->mutex);
285 * Register interest in the next osdmap
287 void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
289 dout("request_next_osdmap have %u\n", monc->have_osdmap);
290 mutex_lock(&monc->mutex);
291 if (!monc->want_next_osdmap)
292 monc->want_next_osdmap = 1;
293 if (monc->want_next_osdmap < 2)
294 __send_subscribe(monc);
295 mutex_unlock(&monc->mutex);
301 int ceph_monc_open_session(struct ceph_mon_client *monc)
303 mutex_lock(&monc->mutex);
304 __open_session(monc);
305 __schedule_delayed(monc);
306 mutex_unlock(&monc->mutex);
309 EXPORT_SYMBOL(ceph_monc_open_session);
312 * The monitor responds with mount ack indicate mount success. The
313 * included client ticket allows the client to talk to MDSs and OSDs.
315 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
316 struct ceph_msg *msg)
318 struct ceph_client *client = monc->client;
319 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
322 mutex_lock(&monc->mutex);
324 dout("handle_monmap\n");
325 p = msg->front.iov_base;
326 end = p + msg->front.iov_len;
328 monmap = ceph_monmap_decode(p, end);
329 if (IS_ERR(monmap)) {
330 pr_err("problem decoding monmap, %d\n",
331 (int)PTR_ERR(monmap));
335 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
340 client->monc.monmap = monmap;
344 mutex_unlock(&monc->mutex);
345 wake_up_all(&client->auth_wq);
349 * generic requests (e.g., statfs, poolop)
351 static struct ceph_mon_generic_request *__lookup_generic_req(
352 struct ceph_mon_client *monc, u64 tid)
354 struct ceph_mon_generic_request *req;
355 struct rb_node *n = monc->generic_request_tree.rb_node;
358 req = rb_entry(n, struct ceph_mon_generic_request, node);
361 else if (tid > req->tid)
369 static void __insert_generic_request(struct ceph_mon_client *monc,
370 struct ceph_mon_generic_request *new)
372 struct rb_node **p = &monc->generic_request_tree.rb_node;
373 struct rb_node *parent = NULL;
374 struct ceph_mon_generic_request *req = NULL;
378 req = rb_entry(parent, struct ceph_mon_generic_request, node);
379 if (new->tid < req->tid)
381 else if (new->tid > req->tid)
387 rb_link_node(&new->node, parent, p);
388 rb_insert_color(&new->node, &monc->generic_request_tree);
391 static void release_generic_request(struct kref *kref)
393 struct ceph_mon_generic_request *req =
394 container_of(kref, struct ceph_mon_generic_request, kref);
397 ceph_msg_put(req->reply);
399 ceph_msg_put(req->request);
404 static void put_generic_request(struct ceph_mon_generic_request *req)
406 kref_put(&req->kref, release_generic_request);
409 static void get_generic_request(struct ceph_mon_generic_request *req)
411 kref_get(&req->kref);
414 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
415 struct ceph_msg_header *hdr,
418 struct ceph_mon_client *monc = con->private;
419 struct ceph_mon_generic_request *req;
420 u64 tid = le64_to_cpu(hdr->tid);
423 mutex_lock(&monc->mutex);
424 req = __lookup_generic_req(monc, tid);
426 dout("get_generic_reply %lld dne\n", tid);
430 dout("get_generic_reply %lld got %p\n", tid, req->reply);
431 m = ceph_msg_get(req->reply);
433 * we don't need to track the connection reading into
434 * this reply because we only have one open connection
438 mutex_unlock(&monc->mutex);
442 static int do_generic_request(struct ceph_mon_client *monc,
443 struct ceph_mon_generic_request *req)
447 /* register request */
448 mutex_lock(&monc->mutex);
449 req->tid = ++monc->last_tid;
450 req->request->hdr.tid = cpu_to_le64(req->tid);
451 __insert_generic_request(monc, req);
452 monc->num_generic_requests++;
453 ceph_con_send(monc->con, ceph_msg_get(req->request));
454 mutex_unlock(&monc->mutex);
456 err = wait_for_completion_interruptible(&req->completion);
458 mutex_lock(&monc->mutex);
459 rb_erase(&req->node, &monc->generic_request_tree);
460 monc->num_generic_requests--;
461 mutex_unlock(&monc->mutex);
471 static void handle_statfs_reply(struct ceph_mon_client *monc,
472 struct ceph_msg *msg)
474 struct ceph_mon_generic_request *req;
475 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
476 u64 tid = le64_to_cpu(msg->hdr.tid);
478 if (msg->front.iov_len != sizeof(*reply))
480 dout("handle_statfs_reply %p tid %llu\n", msg, tid);
482 mutex_lock(&monc->mutex);
483 req = __lookup_generic_req(monc, tid);
485 *(struct ceph_statfs *)req->buf = reply->st;
487 get_generic_request(req);
489 mutex_unlock(&monc->mutex);
491 complete_all(&req->completion);
492 put_generic_request(req);
497 pr_err("corrupt generic reply, tid %llu\n", tid);
502 * Do a synchronous statfs().
504 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
506 struct ceph_mon_generic_request *req;
507 struct ceph_mon_statfs *h;
510 req = kzalloc(sizeof(*req), GFP_NOFS);
514 kref_init(&req->kref);
516 req->buf_len = sizeof(*buf);
517 init_completion(&req->completion);
520 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
524 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
529 /* fill out request */
530 h = req->request->front.iov_base;
531 h->monhdr.have_version = 0;
532 h->monhdr.session_mon = cpu_to_le16(-1);
533 h->monhdr.session_mon_tid = 0;
534 h->fsid = monc->monmap->fsid;
536 err = do_generic_request(monc, req);
539 kref_put(&req->kref, release_generic_request);
542 EXPORT_SYMBOL(ceph_monc_do_statfs);
547 static int get_poolop_reply_buf(const char *src, size_t src_len,
548 char *dst, size_t dst_len)
552 if (src_len != sizeof(u32) + dst_len)
555 buf_len = le32_to_cpu(*(u32 *)src);
556 if (buf_len != dst_len)
559 memcpy(dst, src + sizeof(u32), dst_len);
563 static void handle_poolop_reply(struct ceph_mon_client *monc,
564 struct ceph_msg *msg)
566 struct ceph_mon_generic_request *req;
567 struct ceph_mon_poolop_reply *reply = msg->front.iov_base;
568 u64 tid = le64_to_cpu(msg->hdr.tid);
570 if (msg->front.iov_len < sizeof(*reply))
572 dout("handle_poolop_reply %p tid %llu\n", msg, tid);
574 mutex_lock(&monc->mutex);
575 req = __lookup_generic_req(monc, tid);
578 get_poolop_reply_buf(msg->front.iov_base + sizeof(*reply),
579 msg->front.iov_len - sizeof(*reply),
580 req->buf, req->buf_len) < 0) {
581 mutex_unlock(&monc->mutex);
584 req->result = le32_to_cpu(reply->reply_code);
585 get_generic_request(req);
587 mutex_unlock(&monc->mutex);
589 complete(&req->completion);
590 put_generic_request(req);
595 pr_err("corrupt generic reply, tid %llu\n", tid);
600 * Do a synchronous pool op.
602 int ceph_monc_do_poolop(struct ceph_mon_client *monc, u32 op,
603 u32 pool, u64 snapid,
606 struct ceph_mon_generic_request *req;
607 struct ceph_mon_poolop *h;
610 req = kzalloc(sizeof(*req), GFP_NOFS);
614 kref_init(&req->kref);
617 init_completion(&req->completion);
620 req->request = ceph_msg_new(CEPH_MSG_POOLOP, sizeof(*h), GFP_NOFS,
624 req->reply = ceph_msg_new(CEPH_MSG_POOLOP_REPLY, 1024, GFP_NOFS,
629 /* fill out request */
630 req->request->hdr.version = cpu_to_le16(2);
631 h = req->request->front.iov_base;
632 h->monhdr.have_version = 0;
633 h->monhdr.session_mon = cpu_to_le16(-1);
634 h->monhdr.session_mon_tid = 0;
635 h->fsid = monc->monmap->fsid;
636 h->pool = cpu_to_le32(pool);
637 h->op = cpu_to_le32(op);
639 h->snapid = cpu_to_le64(snapid);
642 err = do_generic_request(monc, req);
645 kref_put(&req->kref, release_generic_request);
649 int ceph_monc_create_snapid(struct ceph_mon_client *monc,
650 u32 pool, u64 *snapid)
652 return ceph_monc_do_poolop(monc, POOL_OP_CREATE_UNMANAGED_SNAP,
653 pool, 0, (char *)snapid, sizeof(*snapid));
656 EXPORT_SYMBOL(ceph_monc_create_snapid);
658 int ceph_monc_delete_snapid(struct ceph_mon_client *monc,
659 u32 pool, u64 snapid)
661 return ceph_monc_do_poolop(monc, POOL_OP_CREATE_UNMANAGED_SNAP,
667 * Resend pending generic requests.
669 static void __resend_generic_request(struct ceph_mon_client *monc)
671 struct ceph_mon_generic_request *req;
674 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
675 req = rb_entry(p, struct ceph_mon_generic_request, node);
676 ceph_con_revoke(monc->con, req->request);
677 ceph_con_send(monc->con, ceph_msg_get(req->request));
682 * Delayed work. If we haven't mounted yet, retry. Otherwise,
683 * renew/retry subscription as needed (in case it is timing out, or we
684 * got an ENOMEM). And keep the monitor connection alive.
686 static void delayed_work(struct work_struct *work)
688 struct ceph_mon_client *monc =
689 container_of(work, struct ceph_mon_client, delayed_work.work);
691 dout("monc delayed_work\n");
692 mutex_lock(&monc->mutex);
694 __close_session(monc);
695 __open_session(monc); /* continue hunting */
697 ceph_con_keepalive(monc->con);
699 __validate_auth(monc);
701 if (monc->auth->ops->is_authenticated(monc->auth))
702 __send_subscribe(monc);
704 __schedule_delayed(monc);
705 mutex_unlock(&monc->mutex);
709 * On startup, we build a temporary monmap populated with the IPs
710 * provided by mount(2).
712 static int build_initial_monmap(struct ceph_mon_client *monc)
714 struct ceph_options *opt = monc->client->options;
715 struct ceph_entity_addr *mon_addr = opt->mon_addr;
716 int num_mon = opt->num_mon;
719 /* build initial monmap */
720 monc->monmap = kzalloc(sizeof(*monc->monmap) +
721 num_mon*sizeof(monc->monmap->mon_inst[0]),
725 for (i = 0; i < num_mon; i++) {
726 monc->monmap->mon_inst[i].addr = mon_addr[i];
727 monc->monmap->mon_inst[i].addr.nonce = 0;
728 monc->monmap->mon_inst[i].name.type =
729 CEPH_ENTITY_TYPE_MON;
730 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
732 monc->monmap->num_mon = num_mon;
733 monc->have_fsid = false;
737 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
742 memset(monc, 0, sizeof(*monc));
745 mutex_init(&monc->mutex);
747 err = build_initial_monmap(monc);
752 monc->con = kmalloc(sizeof(*monc->con), GFP_KERNEL);
755 ceph_con_init(monc->client->msgr, monc->con);
756 monc->con->private = monc;
757 monc->con->ops = &mon_con_ops;
760 monc->auth = ceph_auth_init(cl->options->name,
762 if (IS_ERR(monc->auth)) {
763 err = PTR_ERR(monc->auth);
766 monc->auth->want_keys =
767 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
768 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
772 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
773 sizeof(struct ceph_mon_subscribe_ack),
775 if (!monc->m_subscribe_ack)
778 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
780 if (!monc->m_subscribe)
781 goto out_subscribe_ack;
783 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
785 if (!monc->m_auth_reply)
788 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
789 monc->pending_auth = 0;
794 monc->hunting = true;
795 monc->sub_renew_after = jiffies;
798 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
799 monc->generic_request_tree = RB_ROOT;
800 monc->num_generic_requests = 0;
803 monc->have_mdsmap = 0;
804 monc->have_osdmap = 0;
805 monc->want_next_osdmap = 1;
809 ceph_msg_put(monc->m_auth_reply);
811 ceph_msg_put(monc->m_subscribe);
813 ceph_msg_put(monc->m_subscribe_ack);
815 ceph_auth_destroy(monc->auth);
817 monc->con->ops->put(monc->con);
823 EXPORT_SYMBOL(ceph_monc_init);
825 void ceph_monc_stop(struct ceph_mon_client *monc)
828 cancel_delayed_work_sync(&monc->delayed_work);
830 mutex_lock(&monc->mutex);
831 __close_session(monc);
833 monc->con->private = NULL;
834 monc->con->ops->put(monc->con);
837 mutex_unlock(&monc->mutex);
839 ceph_auth_destroy(monc->auth);
841 ceph_msg_put(monc->m_auth);
842 ceph_msg_put(monc->m_auth_reply);
843 ceph_msg_put(monc->m_subscribe);
844 ceph_msg_put(monc->m_subscribe_ack);
848 EXPORT_SYMBOL(ceph_monc_stop);
850 static void handle_auth_reply(struct ceph_mon_client *monc,
851 struct ceph_msg *msg)
856 mutex_lock(&monc->mutex);
858 was_auth = monc->auth->ops->is_authenticated(monc->auth);
859 monc->pending_auth = 0;
860 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
862 monc->m_auth->front.iov_base,
863 monc->m_auth->front_max);
865 monc->client->auth_err = ret;
866 wake_up_all(&monc->client->auth_wq);
867 } else if (ret > 0) {
868 __send_prepared_auth_request(monc, ret);
869 } else if (!was_auth && monc->auth->ops->is_authenticated(monc->auth)) {
870 dout("authenticated, starting session\n");
872 monc->client->msgr->inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
873 monc->client->msgr->inst.name.num =
874 cpu_to_le64(monc->auth->global_id);
876 __send_subscribe(monc);
877 __resend_generic_request(monc);
879 mutex_unlock(&monc->mutex);
882 static int __validate_auth(struct ceph_mon_client *monc)
886 if (monc->pending_auth)
889 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
890 monc->m_auth->front_max);
892 return ret; /* either an error, or no need to authenticate */
893 __send_prepared_auth_request(monc, ret);
897 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
901 mutex_lock(&monc->mutex);
902 ret = __validate_auth(monc);
903 mutex_unlock(&monc->mutex);
906 EXPORT_SYMBOL(ceph_monc_validate_auth);
909 * handle incoming message
911 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
913 struct ceph_mon_client *monc = con->private;
914 int type = le16_to_cpu(msg->hdr.type);
920 case CEPH_MSG_AUTH_REPLY:
921 handle_auth_reply(monc, msg);
924 case CEPH_MSG_MON_SUBSCRIBE_ACK:
925 handle_subscribe_ack(monc, msg);
928 case CEPH_MSG_STATFS_REPLY:
929 handle_statfs_reply(monc, msg);
932 case CEPH_MSG_POOLOP_REPLY:
933 handle_poolop_reply(monc, msg);
936 case CEPH_MSG_MON_MAP:
937 ceph_monc_handle_map(monc, msg);
940 case CEPH_MSG_OSD_MAP:
941 ceph_osdc_handle_map(&monc->client->osdc, msg);
945 /* can the chained handler handle it? */
946 if (monc->client->extra_mon_dispatch &&
947 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
950 pr_err("received unknown message type %d %s\n", type,
951 ceph_msg_type_name(type));
957 * Allocate memory for incoming message
959 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
960 struct ceph_msg_header *hdr,
963 struct ceph_mon_client *monc = con->private;
964 int type = le16_to_cpu(hdr->type);
965 int front_len = le32_to_cpu(hdr->front_len);
966 struct ceph_msg *m = NULL;
971 case CEPH_MSG_MON_SUBSCRIBE_ACK:
972 m = ceph_msg_get(monc->m_subscribe_ack);
974 case CEPH_MSG_POOLOP_REPLY:
975 case CEPH_MSG_STATFS_REPLY:
976 return get_generic_reply(con, hdr, skip);
977 case CEPH_MSG_AUTH_REPLY:
978 m = ceph_msg_get(monc->m_auth_reply);
980 case CEPH_MSG_MON_MAP:
981 case CEPH_MSG_MDS_MAP:
982 case CEPH_MSG_OSD_MAP:
983 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
988 pr_info("alloc_msg unknown type %d\n", type);
995 * If the monitor connection resets, pick a new monitor and resubmit
996 * any pending requests.
998 static void mon_fault(struct ceph_connection *con)
1000 struct ceph_mon_client *monc = con->private;
1005 dout("mon_fault\n");
1006 mutex_lock(&monc->mutex);
1011 pr_info("mon%d %s session lost, "
1012 "hunting for new mon\n", monc->cur_mon,
1013 ceph_pr_addr(&monc->con->peer_addr.in_addr));
1015 __close_session(monc);
1016 if (!monc->hunting) {
1018 monc->hunting = true;
1019 __open_session(monc);
1021 /* already hunting, let's wait a bit */
1022 __schedule_delayed(monc);
1025 mutex_unlock(&monc->mutex);
1028 static const struct ceph_connection_operations mon_con_ops = {
1029 .get = ceph_con_get,
1030 .put = ceph_con_put,
1031 .dispatch = dispatch,
1033 .alloc_msg = mon_alloc_msg,