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 * Open a session with a (new) monitor.
133 static int __open_session(struct ceph_mon_client *monc)
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);
144 monc->sub_renew_after = jiffies; /* i.e., expired */
145 monc->want_next_osdmap = !!monc->want_next_osdmap;
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);
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);
158 dout("open_session mon%d already open\n", monc->cur_mon);
163 static bool __sub_expired(struct ceph_mon_client *monc)
165 return time_after_eq(jiffies, monc->sub_renew_after);
169 * Reschedule delayed work timer.
171 static void __schedule_delayed(struct ceph_mon_client *monc)
175 if (monc->cur_mon < 0 || __sub_expired(monc))
179 dout("__schedule_delayed after %u\n", delay);
180 schedule_delayed_work(&monc->delayed_work, delay);
184 * Send subscribe request for mdsmap and/or osdmap.
186 static void __send_subscribe(struct ceph_mon_client *monc)
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;
198 p = msg->front.iov_base;
199 end = p + msg->front_alloc_len;
201 num = 1 + !!monc->want_next_osdmap + !!monc->want_mdsmap;
202 ceph_encode_32(&p, num);
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);
209 i->have = cpu_to_le64(monc->have_osdmap);
212 monc->want_next_osdmap = 2; /* requested */
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);
219 i->have = cpu_to_le64(monc->have_mdsmap);
223 ceph_encode_string(&p, end, "monmap", 6);
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));
234 monc->sub_sent = jiffies | 1; /* never 0 */
238 static void handle_subscribe_ack(struct ceph_mon_client *monc,
239 struct ceph_msg *msg)
241 unsigned int seconds;
242 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
244 if (msg->front.iov_len < sizeof(*h))
246 seconds = le32_to_cpu(h->duration);
248 mutex_lock(&monc->mutex);
250 pr_info("mon%d %s session established\n",
252 ceph_pr_addr(&monc->con.peer_addr.in_addr));
253 monc->hunting = false;
255 dout("handle_subscribe_ack after %d seconds\n", seconds);
256 monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1;
258 mutex_unlock(&monc->mutex);
261 pr_err("got corrupt subscribe-ack msg\n");
266 * Keep track of which maps we have
268 int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got)
270 mutex_lock(&monc->mutex);
271 monc->have_mdsmap = got;
272 mutex_unlock(&monc->mutex);
275 EXPORT_SYMBOL(ceph_monc_got_mdsmap);
277 int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got)
279 mutex_lock(&monc->mutex);
280 monc->have_osdmap = got;
281 monc->want_next_osdmap = 0;
282 mutex_unlock(&monc->mutex);
287 * Register interest in the next osdmap
289 void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
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);
299 EXPORT_SYMBOL(ceph_monc_request_next_osdmap);
302 * Wait for an osdmap with a given epoch.
304 * @epoch: epoch to wait for
305 * @timeout: in jiffies, 0 means "wait forever"
307 int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
308 unsigned long timeout)
310 unsigned long started = jiffies;
313 mutex_lock(&monc->mutex);
314 while (monc->have_osdmap < epoch) {
315 mutex_unlock(&monc->mutex);
317 if (timeout && time_after_eq(jiffies, started + timeout))
320 ret = wait_event_interruptible_timeout(monc->client->auth_wq,
321 monc->have_osdmap >= epoch,
322 ceph_timeout_jiffies(timeout));
326 mutex_lock(&monc->mutex);
329 mutex_unlock(&monc->mutex);
332 EXPORT_SYMBOL(ceph_monc_wait_osdmap);
337 int ceph_monc_open_session(struct ceph_mon_client *monc)
339 mutex_lock(&monc->mutex);
340 __open_session(monc);
341 __schedule_delayed(monc);
342 mutex_unlock(&monc->mutex);
345 EXPORT_SYMBOL(ceph_monc_open_session);
348 * We require the fsid and global_id in order to initialize our
351 static bool have_debugfs_info(struct ceph_mon_client *monc)
353 dout("have_debugfs_info fsid %d globalid %lld\n",
354 (int)monc->client->have_fsid, monc->auth->global_id);
355 return monc->client->have_fsid && monc->auth->global_id > 0;
359 * The monitor responds with mount ack indicate mount success. The
360 * included client ticket allows the client to talk to MDSs and OSDs.
362 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
363 struct ceph_msg *msg)
365 struct ceph_client *client = monc->client;
366 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
368 int had_debugfs_info, init_debugfs = 0;
370 mutex_lock(&monc->mutex);
372 had_debugfs_info = have_debugfs_info(monc);
374 dout("handle_monmap\n");
375 p = msg->front.iov_base;
376 end = p + msg->front.iov_len;
378 monmap = ceph_monmap_decode(p, end);
379 if (IS_ERR(monmap)) {
380 pr_err("problem decoding monmap, %d\n",
381 (int)PTR_ERR(monmap));
385 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
390 client->monc.monmap = monmap;
393 if (!client->have_fsid) {
394 client->have_fsid = true;
395 if (!had_debugfs_info && have_debugfs_info(monc)) {
396 pr_info("client%lld fsid %pU\n",
397 ceph_client_id(monc->client),
398 &monc->client->fsid);
401 mutex_unlock(&monc->mutex);
405 * do debugfs initialization without mutex to avoid
406 * creating a locking dependency
408 ceph_debugfs_client_init(monc->client);
414 mutex_unlock(&monc->mutex);
416 wake_up_all(&client->auth_wq);
420 * generic requests (currently statfs, mon_get_version)
422 static struct ceph_mon_generic_request *__lookup_generic_req(
423 struct ceph_mon_client *monc, u64 tid)
425 struct ceph_mon_generic_request *req;
426 struct rb_node *n = monc->generic_request_tree.rb_node;
429 req = rb_entry(n, struct ceph_mon_generic_request, node);
432 else if (tid > req->tid)
440 static void __insert_generic_request(struct ceph_mon_client *monc,
441 struct ceph_mon_generic_request *new)
443 struct rb_node **p = &monc->generic_request_tree.rb_node;
444 struct rb_node *parent = NULL;
445 struct ceph_mon_generic_request *req = NULL;
449 req = rb_entry(parent, struct ceph_mon_generic_request, node);
450 if (new->tid < req->tid)
452 else if (new->tid > req->tid)
458 rb_link_node(&new->node, parent, p);
459 rb_insert_color(&new->node, &monc->generic_request_tree);
462 static void release_generic_request(struct kref *kref)
464 struct ceph_mon_generic_request *req =
465 container_of(kref, struct ceph_mon_generic_request, kref);
468 ceph_msg_put(req->reply);
470 ceph_msg_put(req->request);
475 static void put_generic_request(struct ceph_mon_generic_request *req)
477 kref_put(&req->kref, release_generic_request);
480 static void get_generic_request(struct ceph_mon_generic_request *req)
482 kref_get(&req->kref);
485 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
486 struct ceph_msg_header *hdr,
489 struct ceph_mon_client *monc = con->private;
490 struct ceph_mon_generic_request *req;
491 u64 tid = le64_to_cpu(hdr->tid);
494 mutex_lock(&monc->mutex);
495 req = __lookup_generic_req(monc, tid);
497 dout("get_generic_reply %lld dne\n", tid);
501 dout("get_generic_reply %lld got %p\n", tid, req->reply);
503 m = ceph_msg_get(req->reply);
505 * we don't need to track the connection reading into
506 * this reply because we only have one open connection
510 mutex_unlock(&monc->mutex);
514 static int __do_generic_request(struct ceph_mon_client *monc, u64 tid,
515 struct ceph_mon_generic_request *req)
519 /* register request */
520 req->tid = tid != 0 ? tid : ++monc->last_tid;
521 req->request->hdr.tid = cpu_to_le64(req->tid);
522 __insert_generic_request(monc, req);
523 monc->num_generic_requests++;
524 ceph_con_send(&monc->con, ceph_msg_get(req->request));
525 mutex_unlock(&monc->mutex);
527 err = wait_for_completion_interruptible(&req->completion);
529 mutex_lock(&monc->mutex);
530 rb_erase(&req->node, &monc->generic_request_tree);
531 monc->num_generic_requests--;
538 static int do_generic_request(struct ceph_mon_client *monc,
539 struct ceph_mon_generic_request *req)
543 mutex_lock(&monc->mutex);
544 err = __do_generic_request(monc, 0, req);
545 mutex_unlock(&monc->mutex);
553 static void handle_statfs_reply(struct ceph_mon_client *monc,
554 struct ceph_msg *msg)
556 struct ceph_mon_generic_request *req;
557 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
558 u64 tid = le64_to_cpu(msg->hdr.tid);
560 if (msg->front.iov_len != sizeof(*reply))
562 dout("handle_statfs_reply %p tid %llu\n", msg, tid);
564 mutex_lock(&monc->mutex);
565 req = __lookup_generic_req(monc, tid);
567 *(struct ceph_statfs *)req->buf = reply->st;
569 get_generic_request(req);
571 mutex_unlock(&monc->mutex);
573 complete_all(&req->completion);
574 put_generic_request(req);
579 pr_err("corrupt statfs reply, tid %llu\n", tid);
584 * Do a synchronous statfs().
586 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
588 struct ceph_mon_generic_request *req;
589 struct ceph_mon_statfs *h;
592 req = kzalloc(sizeof(*req), GFP_NOFS);
596 kref_init(&req->kref);
598 init_completion(&req->completion);
601 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
605 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
610 /* fill out request */
611 h = req->request->front.iov_base;
612 h->monhdr.have_version = 0;
613 h->monhdr.session_mon = cpu_to_le16(-1);
614 h->monhdr.session_mon_tid = 0;
615 h->fsid = monc->monmap->fsid;
617 err = do_generic_request(monc, req);
620 put_generic_request(req);
623 EXPORT_SYMBOL(ceph_monc_do_statfs);
625 static void handle_get_version_reply(struct ceph_mon_client *monc,
626 struct ceph_msg *msg)
628 struct ceph_mon_generic_request *req;
629 u64 tid = le64_to_cpu(msg->hdr.tid);
630 void *p = msg->front.iov_base;
631 void *end = p + msg->front_alloc_len;
634 dout("%s %p tid %llu\n", __func__, msg, tid);
636 ceph_decode_need(&p, end, 2*sizeof(u64), bad);
637 handle = ceph_decode_64(&p);
638 if (tid != 0 && tid != handle)
641 mutex_lock(&monc->mutex);
642 req = __lookup_generic_req(monc, handle);
644 *(u64 *)req->buf = ceph_decode_64(&p);
646 get_generic_request(req);
648 mutex_unlock(&monc->mutex);
650 complete_all(&req->completion);
651 put_generic_request(req);
656 pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
661 * Send MMonGetVersion and wait for the reply.
663 * @what: one of "mdsmap", "osdmap" or "monmap"
665 int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what,
668 struct ceph_mon_generic_request *req;
673 req = kzalloc(sizeof(*req), GFP_NOFS);
677 kref_init(&req->kref);
679 init_completion(&req->completion);
681 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
682 sizeof(u64) + sizeof(u32) + strlen(what),
689 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024,
696 p = req->request->front.iov_base;
697 end = p + req->request->front_alloc_len;
699 /* fill out request */
700 mutex_lock(&monc->mutex);
701 tid = ++monc->last_tid;
702 ceph_encode_64(&p, tid); /* handle */
703 ceph_encode_string(&p, end, what, strlen(what));
705 err = __do_generic_request(monc, tid, req);
707 mutex_unlock(&monc->mutex);
709 put_generic_request(req);
712 EXPORT_SYMBOL(ceph_monc_do_get_version);
715 * Resend pending generic requests.
717 static void __resend_generic_request(struct ceph_mon_client *monc)
719 struct ceph_mon_generic_request *req;
722 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
723 req = rb_entry(p, struct ceph_mon_generic_request, node);
724 ceph_msg_revoke(req->request);
725 ceph_msg_revoke_incoming(req->reply);
726 ceph_con_send(&monc->con, ceph_msg_get(req->request));
731 * Delayed work. If we haven't mounted yet, retry. Otherwise,
732 * renew/retry subscription as needed (in case it is timing out, or we
733 * got an ENOMEM). And keep the monitor connection alive.
735 static void delayed_work(struct work_struct *work)
737 struct ceph_mon_client *monc =
738 container_of(work, struct ceph_mon_client, delayed_work.work);
740 dout("monc delayed_work\n");
741 mutex_lock(&monc->mutex);
743 __close_session(monc);
744 __open_session(monc); /* continue hunting */
746 ceph_con_keepalive(&monc->con);
748 __validate_auth(monc);
750 if (ceph_auth_is_authenticated(monc->auth))
751 __send_subscribe(monc);
753 __schedule_delayed(monc);
754 mutex_unlock(&monc->mutex);
758 * On startup, we build a temporary monmap populated with the IPs
759 * provided by mount(2).
761 static int build_initial_monmap(struct ceph_mon_client *monc)
763 struct ceph_options *opt = monc->client->options;
764 struct ceph_entity_addr *mon_addr = opt->mon_addr;
765 int num_mon = opt->num_mon;
768 /* build initial monmap */
769 monc->monmap = kzalloc(sizeof(*monc->monmap) +
770 num_mon*sizeof(monc->monmap->mon_inst[0]),
774 for (i = 0; i < num_mon; i++) {
775 monc->monmap->mon_inst[i].addr = mon_addr[i];
776 monc->monmap->mon_inst[i].addr.nonce = 0;
777 monc->monmap->mon_inst[i].name.type =
778 CEPH_ENTITY_TYPE_MON;
779 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
781 monc->monmap->num_mon = num_mon;
785 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
790 memset(monc, 0, sizeof(*monc));
793 mutex_init(&monc->mutex);
795 err = build_initial_monmap(monc);
801 monc->auth = ceph_auth_init(cl->options->name,
803 if (IS_ERR(monc->auth)) {
804 err = PTR_ERR(monc->auth);
807 monc->auth->want_keys =
808 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
809 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
813 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
814 sizeof(struct ceph_mon_subscribe_ack),
816 if (!monc->m_subscribe_ack)
819 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
821 if (!monc->m_subscribe)
822 goto out_subscribe_ack;
824 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
826 if (!monc->m_auth_reply)
829 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
830 monc->pending_auth = 0;
834 ceph_con_init(&monc->con, monc, &mon_con_ops,
835 &monc->client->msgr);
838 monc->hunting = true;
839 monc->sub_renew_after = jiffies;
842 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
843 monc->generic_request_tree = RB_ROOT;
844 monc->num_generic_requests = 0;
847 monc->have_mdsmap = 0;
848 monc->have_osdmap = 0;
849 monc->want_next_osdmap = 1;
853 ceph_msg_put(monc->m_auth_reply);
855 ceph_msg_put(monc->m_subscribe);
857 ceph_msg_put(monc->m_subscribe_ack);
859 ceph_auth_destroy(monc->auth);
865 EXPORT_SYMBOL(ceph_monc_init);
867 void ceph_monc_stop(struct ceph_mon_client *monc)
870 cancel_delayed_work_sync(&monc->delayed_work);
872 mutex_lock(&monc->mutex);
873 __close_session(monc);
875 mutex_unlock(&monc->mutex);
878 * flush msgr queue before we destroy ourselves to ensure that:
879 * - any work that references our embedded con is finished.
880 * - any osd_client or other work that may reference an authorizer
881 * finishes before we shut down the auth subsystem.
885 ceph_auth_destroy(monc->auth);
887 ceph_msg_put(monc->m_auth);
888 ceph_msg_put(monc->m_auth_reply);
889 ceph_msg_put(monc->m_subscribe);
890 ceph_msg_put(monc->m_subscribe_ack);
894 EXPORT_SYMBOL(ceph_monc_stop);
896 static void handle_auth_reply(struct ceph_mon_client *monc,
897 struct ceph_msg *msg)
901 int had_debugfs_info, init_debugfs = 0;
903 mutex_lock(&monc->mutex);
904 had_debugfs_info = have_debugfs_info(monc);
905 was_auth = ceph_auth_is_authenticated(monc->auth);
906 monc->pending_auth = 0;
907 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
909 monc->m_auth->front.iov_base,
910 monc->m_auth->front_alloc_len);
912 monc->client->auth_err = ret;
913 wake_up_all(&monc->client->auth_wq);
914 } else if (ret > 0) {
915 __send_prepared_auth_request(monc, ret);
916 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
917 dout("authenticated, starting session\n");
919 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
920 monc->client->msgr.inst.name.num =
921 cpu_to_le64(monc->auth->global_id);
923 __send_subscribe(monc);
924 __resend_generic_request(monc);
927 if (!had_debugfs_info && have_debugfs_info(monc)) {
928 pr_info("client%lld fsid %pU\n",
929 ceph_client_id(monc->client),
930 &monc->client->fsid);
933 mutex_unlock(&monc->mutex);
937 * do debugfs initialization without mutex to avoid
938 * creating a locking dependency
940 ceph_debugfs_client_init(monc->client);
944 static int __validate_auth(struct ceph_mon_client *monc)
948 if (monc->pending_auth)
951 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
952 monc->m_auth->front_alloc_len);
954 return ret; /* either an error, or no need to authenticate */
955 __send_prepared_auth_request(monc, ret);
959 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
963 mutex_lock(&monc->mutex);
964 ret = __validate_auth(monc);
965 mutex_unlock(&monc->mutex);
968 EXPORT_SYMBOL(ceph_monc_validate_auth);
971 * handle incoming message
973 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
975 struct ceph_mon_client *monc = con->private;
976 int type = le16_to_cpu(msg->hdr.type);
982 case CEPH_MSG_AUTH_REPLY:
983 handle_auth_reply(monc, msg);
986 case CEPH_MSG_MON_SUBSCRIBE_ACK:
987 handle_subscribe_ack(monc, msg);
990 case CEPH_MSG_STATFS_REPLY:
991 handle_statfs_reply(monc, msg);
994 case CEPH_MSG_MON_GET_VERSION_REPLY:
995 handle_get_version_reply(monc, msg);
998 case CEPH_MSG_MON_MAP:
999 ceph_monc_handle_map(monc, msg);
1002 case CEPH_MSG_OSD_MAP:
1003 ceph_osdc_handle_map(&monc->client->osdc, msg);
1007 /* can the chained handler handle it? */
1008 if (monc->client->extra_mon_dispatch &&
1009 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1012 pr_err("received unknown message type %d %s\n", type,
1013 ceph_msg_type_name(type));
1019 * Allocate memory for incoming message
1021 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1022 struct ceph_msg_header *hdr,
1025 struct ceph_mon_client *monc = con->private;
1026 int type = le16_to_cpu(hdr->type);
1027 int front_len = le32_to_cpu(hdr->front_len);
1028 struct ceph_msg *m = NULL;
1033 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1034 m = ceph_msg_get(monc->m_subscribe_ack);
1036 case CEPH_MSG_STATFS_REPLY:
1037 return get_generic_reply(con, hdr, skip);
1038 case CEPH_MSG_AUTH_REPLY:
1039 m = ceph_msg_get(monc->m_auth_reply);
1041 case CEPH_MSG_MON_GET_VERSION_REPLY:
1042 if (le64_to_cpu(hdr->tid) != 0)
1043 return get_generic_reply(con, hdr, skip);
1046 * Older OSDs don't set reply tid even if the orignal
1047 * request had a non-zero tid. Workaround this weirdness
1048 * by falling through to the allocate case.
1050 case CEPH_MSG_MON_MAP:
1051 case CEPH_MSG_MDS_MAP:
1052 case CEPH_MSG_OSD_MAP:
1053 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1055 return NULL; /* ENOMEM--return skip == 0 */
1060 pr_info("alloc_msg unknown type %d\n", type);
1062 } else if (front_len > m->front_alloc_len) {
1063 pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1064 front_len, m->front_alloc_len,
1065 (unsigned int)con->peer_name.type,
1066 le64_to_cpu(con->peer_name.num));
1068 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1075 * If the monitor connection resets, pick a new monitor and resubmit
1076 * any pending requests.
1078 static void mon_fault(struct ceph_connection *con)
1080 struct ceph_mon_client *monc = con->private;
1085 dout("mon_fault\n");
1086 mutex_lock(&monc->mutex);
1091 pr_info("mon%d %s session lost, "
1092 "hunting for new mon\n", monc->cur_mon,
1093 ceph_pr_addr(&monc->con.peer_addr.in_addr));
1095 __close_session(monc);
1096 if (!monc->hunting) {
1098 monc->hunting = true;
1099 __open_session(monc);
1101 /* already hunting, let's wait a bit */
1102 __schedule_delayed(monc);
1105 mutex_unlock(&monc->mutex);
1109 * We can ignore refcounting on the connection struct, as all references
1110 * will come from the messenger workqueue, which is drained prior to
1111 * mon_client destruction.
1113 static struct ceph_connection *con_get(struct ceph_connection *con)
1118 static void con_put(struct ceph_connection *con)
1122 static const struct ceph_connection_operations mon_con_ops = {
1125 .dispatch = dispatch,
1127 .alloc_msg = mon_alloc_msg,