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 /* send an initial keepalive to ensure our timestamp is
153 * valid by the time we are in an OPENED state */
154 ceph_con_keepalive(&monc->con);
156 /* initiatiate authentication handshake */
157 ret = ceph_auth_build_hello(monc->auth,
158 monc->m_auth->front.iov_base,
159 monc->m_auth->front_alloc_len);
160 __send_prepared_auth_request(monc, ret);
162 dout("open_session mon%d already open\n", monc->cur_mon);
167 static bool __sub_expired(struct ceph_mon_client *monc)
169 return time_after_eq(jiffies, monc->sub_renew_after);
173 * Reschedule delayed work timer.
175 static void __schedule_delayed(struct ceph_mon_client *monc)
177 struct ceph_options *opt = monc->client->options;
180 if (monc->cur_mon < 0 || __sub_expired(monc)) {
184 if (opt->monc_ping_timeout > 0)
185 delay = min(delay, opt->monc_ping_timeout / 3);
187 dout("__schedule_delayed after %lu\n", delay);
188 schedule_delayed_work(&monc->delayed_work,
189 round_jiffies_relative(delay));
193 * Send subscribe request for mdsmap and/or osdmap.
195 static void __send_subscribe(struct ceph_mon_client *monc)
197 dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n",
198 (unsigned int)monc->sub_sent, __sub_expired(monc),
199 monc->want_next_osdmap);
200 if ((__sub_expired(monc) && !monc->sub_sent) ||
201 monc->want_next_osdmap == 1) {
202 struct ceph_msg *msg = monc->m_subscribe;
203 struct ceph_mon_subscribe_item *i;
207 p = msg->front.iov_base;
208 end = p + msg->front_alloc_len;
210 num = 1 + !!monc->want_next_osdmap + !!monc->want_mdsmap;
211 ceph_encode_32(&p, num);
213 if (monc->want_next_osdmap) {
214 dout("__send_subscribe to 'osdmap' %u\n",
215 (unsigned int)monc->have_osdmap);
216 ceph_encode_string(&p, end, "osdmap", 6);
218 i->have = cpu_to_le64(monc->have_osdmap);
221 monc->want_next_osdmap = 2; /* requested */
223 if (monc->want_mdsmap) {
224 dout("__send_subscribe to 'mdsmap' %u+\n",
225 (unsigned int)monc->have_mdsmap);
226 ceph_encode_string(&p, end, "mdsmap", 6);
228 i->have = cpu_to_le64(monc->have_mdsmap);
232 ceph_encode_string(&p, end, "monmap", 6);
238 msg->front.iov_len = p - msg->front.iov_base;
239 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
240 ceph_msg_revoke(msg);
241 ceph_con_send(&monc->con, ceph_msg_get(msg));
243 monc->sub_sent = jiffies | 1; /* never 0 */
247 static void handle_subscribe_ack(struct ceph_mon_client *monc,
248 struct ceph_msg *msg)
250 unsigned int seconds;
251 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
253 if (msg->front.iov_len < sizeof(*h))
255 seconds = le32_to_cpu(h->duration);
257 mutex_lock(&monc->mutex);
259 pr_info("mon%d %s session established\n",
261 ceph_pr_addr(&monc->con.peer_addr.in_addr));
262 monc->hunting = false;
264 dout("handle_subscribe_ack after %d seconds\n", seconds);
265 monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1;
267 mutex_unlock(&monc->mutex);
270 pr_err("got corrupt subscribe-ack msg\n");
275 * Keep track of which maps we have
277 int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got)
279 mutex_lock(&monc->mutex);
280 monc->have_mdsmap = got;
281 mutex_unlock(&monc->mutex);
284 EXPORT_SYMBOL(ceph_monc_got_mdsmap);
286 int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got)
288 mutex_lock(&monc->mutex);
289 monc->have_osdmap = got;
290 monc->want_next_osdmap = 0;
291 mutex_unlock(&monc->mutex);
296 * Register interest in the next osdmap
298 void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
300 dout("request_next_osdmap have %u\n", monc->have_osdmap);
301 mutex_lock(&monc->mutex);
302 if (!monc->want_next_osdmap)
303 monc->want_next_osdmap = 1;
304 if (monc->want_next_osdmap < 2)
305 __send_subscribe(monc);
306 mutex_unlock(&monc->mutex);
308 EXPORT_SYMBOL(ceph_monc_request_next_osdmap);
311 * Wait for an osdmap with a given epoch.
313 * @epoch: epoch to wait for
314 * @timeout: in jiffies, 0 means "wait forever"
316 int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
317 unsigned long timeout)
319 unsigned long started = jiffies;
322 mutex_lock(&monc->mutex);
323 while (monc->have_osdmap < epoch) {
324 mutex_unlock(&monc->mutex);
326 if (timeout && time_after_eq(jiffies, started + timeout))
329 ret = wait_event_interruptible_timeout(monc->client->auth_wq,
330 monc->have_osdmap >= epoch,
331 ceph_timeout_jiffies(timeout));
335 mutex_lock(&monc->mutex);
338 mutex_unlock(&monc->mutex);
341 EXPORT_SYMBOL(ceph_monc_wait_osdmap);
346 int ceph_monc_open_session(struct ceph_mon_client *monc)
348 mutex_lock(&monc->mutex);
349 __open_session(monc);
350 __schedule_delayed(monc);
351 mutex_unlock(&monc->mutex);
354 EXPORT_SYMBOL(ceph_monc_open_session);
357 * We require the fsid and global_id in order to initialize our
360 static bool have_debugfs_info(struct ceph_mon_client *monc)
362 dout("have_debugfs_info fsid %d globalid %lld\n",
363 (int)monc->client->have_fsid, monc->auth->global_id);
364 return monc->client->have_fsid && monc->auth->global_id > 0;
367 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
368 struct ceph_msg *msg)
370 struct ceph_client *client = monc->client;
371 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
373 int had_debugfs_info, init_debugfs = 0;
375 mutex_lock(&monc->mutex);
377 had_debugfs_info = have_debugfs_info(monc);
379 dout("handle_monmap\n");
380 p = msg->front.iov_base;
381 end = p + msg->front.iov_len;
383 monmap = ceph_monmap_decode(p, end);
384 if (IS_ERR(monmap)) {
385 pr_err("problem decoding monmap, %d\n",
386 (int)PTR_ERR(monmap));
390 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
395 client->monc.monmap = monmap;
398 if (!client->have_fsid) {
399 client->have_fsid = true;
400 if (!had_debugfs_info && have_debugfs_info(monc)) {
401 pr_info("client%lld fsid %pU\n",
402 ceph_client_id(monc->client),
403 &monc->client->fsid);
406 mutex_unlock(&monc->mutex);
410 * do debugfs initialization without mutex to avoid
411 * creating a locking dependency
413 ceph_debugfs_client_init(monc->client);
419 mutex_unlock(&monc->mutex);
421 wake_up_all(&client->auth_wq);
425 * generic requests (currently statfs, mon_get_version)
427 static struct ceph_mon_generic_request *__lookup_generic_req(
428 struct ceph_mon_client *monc, u64 tid)
430 struct ceph_mon_generic_request *req;
431 struct rb_node *n = monc->generic_request_tree.rb_node;
434 req = rb_entry(n, struct ceph_mon_generic_request, node);
437 else if (tid > req->tid)
445 static void __insert_generic_request(struct ceph_mon_client *monc,
446 struct ceph_mon_generic_request *new)
448 struct rb_node **p = &monc->generic_request_tree.rb_node;
449 struct rb_node *parent = NULL;
450 struct ceph_mon_generic_request *req = NULL;
454 req = rb_entry(parent, struct ceph_mon_generic_request, node);
455 if (new->tid < req->tid)
457 else if (new->tid > req->tid)
463 rb_link_node(&new->node, parent, p);
464 rb_insert_color(&new->node, &monc->generic_request_tree);
467 static void release_generic_request(struct kref *kref)
469 struct ceph_mon_generic_request *req =
470 container_of(kref, struct ceph_mon_generic_request, kref);
473 ceph_msg_put(req->reply);
475 ceph_msg_put(req->request);
480 static void put_generic_request(struct ceph_mon_generic_request *req)
482 kref_put(&req->kref, release_generic_request);
485 static void get_generic_request(struct ceph_mon_generic_request *req)
487 kref_get(&req->kref);
490 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
491 struct ceph_msg_header *hdr,
494 struct ceph_mon_client *monc = con->private;
495 struct ceph_mon_generic_request *req;
496 u64 tid = le64_to_cpu(hdr->tid);
499 mutex_lock(&monc->mutex);
500 req = __lookup_generic_req(monc, tid);
502 dout("get_generic_reply %lld dne\n", tid);
506 dout("get_generic_reply %lld got %p\n", tid, req->reply);
508 m = ceph_msg_get(req->reply);
510 * we don't need to track the connection reading into
511 * this reply because we only have one open connection
515 mutex_unlock(&monc->mutex);
519 static int __do_generic_request(struct ceph_mon_client *monc, u64 tid,
520 struct ceph_mon_generic_request *req)
524 /* register request */
525 req->tid = tid != 0 ? tid : ++monc->last_tid;
526 req->request->hdr.tid = cpu_to_le64(req->tid);
527 __insert_generic_request(monc, req);
528 monc->num_generic_requests++;
529 ceph_con_send(&monc->con, ceph_msg_get(req->request));
530 mutex_unlock(&monc->mutex);
532 err = wait_for_completion_interruptible(&req->completion);
534 mutex_lock(&monc->mutex);
535 rb_erase(&req->node, &monc->generic_request_tree);
536 monc->num_generic_requests--;
543 static int do_generic_request(struct ceph_mon_client *monc,
544 struct ceph_mon_generic_request *req)
548 mutex_lock(&monc->mutex);
549 err = __do_generic_request(monc, 0, req);
550 mutex_unlock(&monc->mutex);
558 static void handle_statfs_reply(struct ceph_mon_client *monc,
559 struct ceph_msg *msg)
561 struct ceph_mon_generic_request *req;
562 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
563 u64 tid = le64_to_cpu(msg->hdr.tid);
565 if (msg->front.iov_len != sizeof(*reply))
567 dout("handle_statfs_reply %p tid %llu\n", msg, tid);
569 mutex_lock(&monc->mutex);
570 req = __lookup_generic_req(monc, tid);
572 *(struct ceph_statfs *)req->buf = reply->st;
574 get_generic_request(req);
576 mutex_unlock(&monc->mutex);
578 complete_all(&req->completion);
579 put_generic_request(req);
584 pr_err("corrupt statfs reply, tid %llu\n", tid);
589 * Do a synchronous statfs().
591 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
593 struct ceph_mon_generic_request *req;
594 struct ceph_mon_statfs *h;
597 req = kzalloc(sizeof(*req), GFP_NOFS);
601 kref_init(&req->kref);
603 init_completion(&req->completion);
606 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
610 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
615 /* fill out request */
616 h = req->request->front.iov_base;
617 h->monhdr.have_version = 0;
618 h->monhdr.session_mon = cpu_to_le16(-1);
619 h->monhdr.session_mon_tid = 0;
620 h->fsid = monc->monmap->fsid;
622 err = do_generic_request(monc, req);
625 put_generic_request(req);
628 EXPORT_SYMBOL(ceph_monc_do_statfs);
630 static void handle_get_version_reply(struct ceph_mon_client *monc,
631 struct ceph_msg *msg)
633 struct ceph_mon_generic_request *req;
634 u64 tid = le64_to_cpu(msg->hdr.tid);
635 void *p = msg->front.iov_base;
636 void *end = p + msg->front_alloc_len;
639 dout("%s %p tid %llu\n", __func__, msg, tid);
641 ceph_decode_need(&p, end, 2*sizeof(u64), bad);
642 handle = ceph_decode_64(&p);
643 if (tid != 0 && tid != handle)
646 mutex_lock(&monc->mutex);
647 req = __lookup_generic_req(monc, handle);
649 *(u64 *)req->buf = ceph_decode_64(&p);
651 get_generic_request(req);
653 mutex_unlock(&monc->mutex);
655 complete_all(&req->completion);
656 put_generic_request(req);
661 pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
666 * Send MMonGetVersion and wait for the reply.
668 * @what: one of "mdsmap", "osdmap" or "monmap"
670 int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what,
673 struct ceph_mon_generic_request *req;
678 req = kzalloc(sizeof(*req), GFP_NOFS);
682 kref_init(&req->kref);
684 init_completion(&req->completion);
686 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
687 sizeof(u64) + sizeof(u32) + strlen(what),
694 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024,
701 p = req->request->front.iov_base;
702 end = p + req->request->front_alloc_len;
704 /* fill out request */
705 mutex_lock(&monc->mutex);
706 tid = ++monc->last_tid;
707 ceph_encode_64(&p, tid); /* handle */
708 ceph_encode_string(&p, end, what, strlen(what));
710 err = __do_generic_request(monc, tid, req);
712 mutex_unlock(&monc->mutex);
714 put_generic_request(req);
717 EXPORT_SYMBOL(ceph_monc_do_get_version);
720 * Resend pending generic requests.
722 static void __resend_generic_request(struct ceph_mon_client *monc)
724 struct ceph_mon_generic_request *req;
727 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
728 req = rb_entry(p, struct ceph_mon_generic_request, node);
729 ceph_msg_revoke(req->request);
730 ceph_msg_revoke_incoming(req->reply);
731 ceph_con_send(&monc->con, ceph_msg_get(req->request));
736 * Delayed work. If we haven't mounted yet, retry. Otherwise,
737 * renew/retry subscription as needed (in case it is timing out, or we
738 * got an ENOMEM). And keep the monitor connection alive.
740 static void delayed_work(struct work_struct *work)
742 struct ceph_mon_client *monc =
743 container_of(work, struct ceph_mon_client, delayed_work.work);
745 dout("monc delayed_work\n");
746 mutex_lock(&monc->mutex);
748 __close_session(monc);
749 __open_session(monc); /* continue hunting */
751 struct ceph_options *opt = monc->client->options;
752 int is_auth = ceph_auth_is_authenticated(monc->auth);
753 if (ceph_con_keepalive_expired(&monc->con,
754 opt->monc_ping_timeout)) {
755 dout("monc keepalive timeout\n");
757 __close_session(monc);
758 monc->hunting = true;
759 __open_session(monc);
762 if (!monc->hunting) {
763 ceph_con_keepalive(&monc->con);
764 __validate_auth(monc);
768 __send_subscribe(monc);
770 __schedule_delayed(monc);
771 mutex_unlock(&monc->mutex);
775 * On startup, we build a temporary monmap populated with the IPs
776 * provided by mount(2).
778 static int build_initial_monmap(struct ceph_mon_client *monc)
780 struct ceph_options *opt = monc->client->options;
781 struct ceph_entity_addr *mon_addr = opt->mon_addr;
782 int num_mon = opt->num_mon;
785 /* build initial monmap */
786 monc->monmap = kzalloc(sizeof(*monc->monmap) +
787 num_mon*sizeof(monc->monmap->mon_inst[0]),
791 for (i = 0; i < num_mon; i++) {
792 monc->monmap->mon_inst[i].addr = mon_addr[i];
793 monc->monmap->mon_inst[i].addr.nonce = 0;
794 monc->monmap->mon_inst[i].name.type =
795 CEPH_ENTITY_TYPE_MON;
796 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
798 monc->monmap->num_mon = num_mon;
802 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
807 memset(monc, 0, sizeof(*monc));
810 mutex_init(&monc->mutex);
812 err = build_initial_monmap(monc);
818 monc->auth = ceph_auth_init(cl->options->name,
820 if (IS_ERR(monc->auth)) {
821 err = PTR_ERR(monc->auth);
824 monc->auth->want_keys =
825 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
826 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
830 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
831 sizeof(struct ceph_mon_subscribe_ack),
833 if (!monc->m_subscribe_ack)
836 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
838 if (!monc->m_subscribe)
839 goto out_subscribe_ack;
841 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
843 if (!monc->m_auth_reply)
846 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
847 monc->pending_auth = 0;
851 ceph_con_init(&monc->con, monc, &mon_con_ops,
852 &monc->client->msgr);
855 monc->hunting = true;
856 monc->sub_renew_after = jiffies;
859 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
860 monc->generic_request_tree = RB_ROOT;
861 monc->num_generic_requests = 0;
864 monc->have_mdsmap = 0;
865 monc->have_osdmap = 0;
866 monc->want_next_osdmap = 1;
870 ceph_msg_put(monc->m_auth_reply);
872 ceph_msg_put(monc->m_subscribe);
874 ceph_msg_put(monc->m_subscribe_ack);
876 ceph_auth_destroy(monc->auth);
882 EXPORT_SYMBOL(ceph_monc_init);
884 void ceph_monc_stop(struct ceph_mon_client *monc)
887 cancel_delayed_work_sync(&monc->delayed_work);
889 mutex_lock(&monc->mutex);
890 __close_session(monc);
892 mutex_unlock(&monc->mutex);
895 * flush msgr queue before we destroy ourselves to ensure that:
896 * - any work that references our embedded con is finished.
897 * - any osd_client or other work that may reference an authorizer
898 * finishes before we shut down the auth subsystem.
902 ceph_auth_destroy(monc->auth);
904 ceph_msg_put(monc->m_auth);
905 ceph_msg_put(monc->m_auth_reply);
906 ceph_msg_put(monc->m_subscribe);
907 ceph_msg_put(monc->m_subscribe_ack);
911 EXPORT_SYMBOL(ceph_monc_stop);
913 static void handle_auth_reply(struct ceph_mon_client *monc,
914 struct ceph_msg *msg)
918 int had_debugfs_info, init_debugfs = 0;
920 mutex_lock(&monc->mutex);
921 had_debugfs_info = have_debugfs_info(monc);
922 was_auth = ceph_auth_is_authenticated(monc->auth);
923 monc->pending_auth = 0;
924 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
926 monc->m_auth->front.iov_base,
927 monc->m_auth->front_alloc_len);
929 monc->client->auth_err = ret;
930 wake_up_all(&monc->client->auth_wq);
931 } else if (ret > 0) {
932 __send_prepared_auth_request(monc, ret);
933 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
934 dout("authenticated, starting session\n");
936 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
937 monc->client->msgr.inst.name.num =
938 cpu_to_le64(monc->auth->global_id);
940 __send_subscribe(monc);
941 __resend_generic_request(monc);
944 if (!had_debugfs_info && have_debugfs_info(monc)) {
945 pr_info("client%lld fsid %pU\n",
946 ceph_client_id(monc->client),
947 &monc->client->fsid);
950 mutex_unlock(&monc->mutex);
954 * do debugfs initialization without mutex to avoid
955 * creating a locking dependency
957 ceph_debugfs_client_init(monc->client);
961 static int __validate_auth(struct ceph_mon_client *monc)
965 if (monc->pending_auth)
968 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
969 monc->m_auth->front_alloc_len);
971 return ret; /* either an error, or no need to authenticate */
972 __send_prepared_auth_request(monc, ret);
976 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
980 mutex_lock(&monc->mutex);
981 ret = __validate_auth(monc);
982 mutex_unlock(&monc->mutex);
985 EXPORT_SYMBOL(ceph_monc_validate_auth);
988 * handle incoming message
990 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
992 struct ceph_mon_client *monc = con->private;
993 int type = le16_to_cpu(msg->hdr.type);
999 case CEPH_MSG_AUTH_REPLY:
1000 handle_auth_reply(monc, msg);
1003 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1004 handle_subscribe_ack(monc, msg);
1007 case CEPH_MSG_STATFS_REPLY:
1008 handle_statfs_reply(monc, msg);
1011 case CEPH_MSG_MON_GET_VERSION_REPLY:
1012 handle_get_version_reply(monc, msg);
1015 case CEPH_MSG_MON_MAP:
1016 ceph_monc_handle_map(monc, msg);
1019 case CEPH_MSG_OSD_MAP:
1020 ceph_osdc_handle_map(&monc->client->osdc, msg);
1024 /* can the chained handler handle it? */
1025 if (monc->client->extra_mon_dispatch &&
1026 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1029 pr_err("received unknown message type %d %s\n", type,
1030 ceph_msg_type_name(type));
1036 * Allocate memory for incoming message
1038 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1039 struct ceph_msg_header *hdr,
1042 struct ceph_mon_client *monc = con->private;
1043 int type = le16_to_cpu(hdr->type);
1044 int front_len = le32_to_cpu(hdr->front_len);
1045 struct ceph_msg *m = NULL;
1050 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1051 m = ceph_msg_get(monc->m_subscribe_ack);
1053 case CEPH_MSG_STATFS_REPLY:
1054 return get_generic_reply(con, hdr, skip);
1055 case CEPH_MSG_AUTH_REPLY:
1056 m = ceph_msg_get(monc->m_auth_reply);
1058 case CEPH_MSG_MON_GET_VERSION_REPLY:
1059 if (le64_to_cpu(hdr->tid) != 0)
1060 return get_generic_reply(con, hdr, skip);
1063 * Older OSDs don't set reply tid even if the orignal
1064 * request had a non-zero tid. Workaround this weirdness
1065 * by falling through to the allocate case.
1067 case CEPH_MSG_MON_MAP:
1068 case CEPH_MSG_MDS_MAP:
1069 case CEPH_MSG_OSD_MAP:
1070 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1072 return NULL; /* ENOMEM--return skip == 0 */
1077 pr_info("alloc_msg unknown type %d\n", type);
1079 } else if (front_len > m->front_alloc_len) {
1080 pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1081 front_len, m->front_alloc_len,
1082 (unsigned int)con->peer_name.type,
1083 le64_to_cpu(con->peer_name.num));
1085 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1092 * If the monitor connection resets, pick a new monitor and resubmit
1093 * any pending requests.
1095 static void mon_fault(struct ceph_connection *con)
1097 struct ceph_mon_client *monc = con->private;
1102 dout("mon_fault\n");
1103 mutex_lock(&monc->mutex);
1108 pr_info("mon%d %s session lost, "
1109 "hunting for new mon\n", monc->cur_mon,
1110 ceph_pr_addr(&monc->con.peer_addr.in_addr));
1112 __close_session(monc);
1113 if (!monc->hunting) {
1115 monc->hunting = true;
1116 __open_session(monc);
1118 /* already hunting, let's wait a bit */
1119 __schedule_delayed(monc);
1122 mutex_unlock(&monc->mutex);
1126 * We can ignore refcounting on the connection struct, as all references
1127 * will come from the messenger workqueue, which is drained prior to
1128 * mon_client destruction.
1130 static struct ceph_connection *con_get(struct ceph_connection *con)
1135 static void con_put(struct ceph_connection *con)
1139 static const struct ceph_connection_operations mon_con_ops = {
1142 .dispatch = dispatch,
1144 .alloc_msg = mon_alloc_msg,