1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/module.h>
5 #include <linux/types.h>
6 #include <linux/slab.h>
7 #include <linux/random.h>
8 #include <linux/sched.h>
10 #include <linux/ceph/ceph_features.h>
11 #include <linux/ceph/mon_client.h>
12 #include <linux/ceph/libceph.h>
13 #include <linux/ceph/debugfs.h>
14 #include <linux/ceph/decode.h>
15 #include <linux/ceph/auth.h>
18 * Interact with Ceph monitor cluster. Handle requests for new map
19 * versions, and periodically resend as needed. Also implement
20 * statfs() and umount().
22 * A small cluster of Ceph "monitors" are responsible for managing critical
23 * cluster configuration and state information. An odd number (e.g., 3, 5)
24 * of cmon daemons use a modified version of the Paxos part-time parliament
25 * algorithm to manage the MDS map (mds cluster membership), OSD map, and
26 * list of clients who have mounted the file system.
28 * We maintain an open, active session with a monitor at all times in order to
29 * receive timely MDSMap updates. We periodically send a keepalive byte on the
30 * TCP socket to ensure we detect a failure. If the connection does break, we
31 * randomly hunt for a new monitor. Once the connection is reestablished, we
32 * resend any outstanding requests.
35 static const struct ceph_connection_operations mon_con_ops;
37 static int __validate_auth(struct ceph_mon_client *monc);
40 * Decode a monmap blob (e.g., during mount).
42 struct ceph_monmap *ceph_monmap_decode(void *p, void *end)
44 struct ceph_monmap *m = NULL;
46 struct ceph_fsid fsid;
50 ceph_decode_32_safe(&p, end, len, bad);
51 ceph_decode_need(&p, end, len, bad);
53 dout("monmap_decode %p %p len %d\n", p, end, (int)(end-p));
54 p += sizeof(u16); /* skip version */
56 ceph_decode_need(&p, end, sizeof(fsid) + 2*sizeof(u32), bad);
57 ceph_decode_copy(&p, &fsid, sizeof(fsid));
58 epoch = ceph_decode_32(&p);
60 num_mon = ceph_decode_32(&p);
61 ceph_decode_need(&p, end, num_mon*sizeof(m->mon_inst[0]), bad);
63 if (num_mon >= CEPH_MAX_MON)
65 m = kmalloc(sizeof(*m) + sizeof(m->mon_inst[0])*num_mon, GFP_NOFS);
67 return ERR_PTR(-ENOMEM);
71 ceph_decode_copy(&p, m->mon_inst, num_mon*sizeof(m->mon_inst[0]));
72 for (i = 0; i < num_mon; i++)
73 ceph_decode_addr(&m->mon_inst[i].addr);
75 dout("monmap_decode epoch %d, num_mon %d\n", m->epoch,
77 for (i = 0; i < m->num_mon; i++)
78 dout("monmap_decode mon%d is %s\n", i,
79 ceph_pr_addr(&m->mon_inst[i].addr.in_addr));
83 dout("monmap_decode failed with %d\n", err);
89 * return true if *addr is included in the monmap.
91 int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr)
95 for (i = 0; i < m->num_mon; i++)
96 if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0)
102 * Send an auth request.
104 static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len)
106 monc->pending_auth = 1;
107 monc->m_auth->front.iov_len = len;
108 monc->m_auth->hdr.front_len = cpu_to_le32(len);
109 ceph_msg_revoke(monc->m_auth);
110 ceph_msg_get(monc->m_auth); /* keep our ref */
111 ceph_con_send(&monc->con, monc->m_auth);
115 * Close monitor session, if any.
117 static void __close_session(struct ceph_mon_client *monc)
119 dout("__close_session closing mon%d\n", monc->cur_mon);
120 ceph_msg_revoke(monc->m_auth);
121 ceph_msg_revoke_incoming(monc->m_auth_reply);
122 ceph_msg_revoke(monc->m_subscribe);
123 ceph_msg_revoke_incoming(monc->m_subscribe_ack);
124 ceph_con_close(&monc->con);
126 monc->pending_auth = 0;
127 ceph_auth_reset(monc->auth);
131 * Pick a new monitor at random and set cur_mon. If we are repicking
132 * (i.e. cur_mon is already set), be sure to pick a different one.
134 static void pick_new_mon(struct ceph_mon_client *monc)
136 int old_mon = monc->cur_mon;
138 BUG_ON(monc->monmap->num_mon < 1);
140 if (monc->monmap->num_mon == 1) {
143 int max = monc->monmap->num_mon;
147 if (monc->cur_mon >= 0) {
148 if (monc->cur_mon < monc->monmap->num_mon)
154 n = prandom_u32() % max;
155 if (o >= 0 && n >= o)
161 dout("%s mon%d -> mon%d out of %d mons\n", __func__, old_mon,
162 monc->cur_mon, monc->monmap->num_mon);
166 * Open a session with a new monitor.
168 static void __open_session(struct ceph_mon_client *monc)
174 monc->hunting = true;
175 if (monc->had_a_connection) {
176 monc->hunt_mult *= CEPH_MONC_HUNT_BACKOFF;
177 if (monc->hunt_mult > CEPH_MONC_HUNT_MAX_MULT)
178 monc->hunt_mult = CEPH_MONC_HUNT_MAX_MULT;
181 monc->sub_renew_after = jiffies; /* i.e., expired */
182 monc->sub_renew_sent = 0;
184 dout("%s opening mon%d\n", __func__, monc->cur_mon);
185 ceph_con_open(&monc->con, CEPH_ENTITY_TYPE_MON, monc->cur_mon,
186 &monc->monmap->mon_inst[monc->cur_mon].addr);
189 * send an initial keepalive to ensure our timestamp is valid
190 * by the time we are in an OPENED state
192 ceph_con_keepalive(&monc->con);
194 /* initiate authentication handshake */
195 ret = ceph_auth_build_hello(monc->auth,
196 monc->m_auth->front.iov_base,
197 monc->m_auth->front_alloc_len);
199 __send_prepared_auth_request(monc, ret);
202 static void reopen_session(struct ceph_mon_client *monc)
205 pr_info("mon%d %s session lost, hunting for new mon\n",
206 monc->cur_mon, ceph_pr_addr(&monc->con.peer_addr.in_addr));
208 __close_session(monc);
209 __open_session(monc);
213 * Reschedule delayed work timer.
215 static void __schedule_delayed(struct ceph_mon_client *monc)
220 delay = CEPH_MONC_HUNT_INTERVAL * monc->hunt_mult;
222 delay = CEPH_MONC_PING_INTERVAL;
224 dout("__schedule_delayed after %lu\n", delay);
225 mod_delayed_work(system_wq, &monc->delayed_work,
226 round_jiffies_relative(delay));
229 const char *ceph_sub_str[] = {
230 [CEPH_SUB_MONMAP] = "monmap",
231 [CEPH_SUB_OSDMAP] = "osdmap",
232 [CEPH_SUB_FSMAP] = "fsmap.user",
233 [CEPH_SUB_MDSMAP] = "mdsmap",
237 * Send subscribe request for one or more maps, according to
240 static void __send_subscribe(struct ceph_mon_client *monc)
242 struct ceph_msg *msg = monc->m_subscribe;
243 void *p = msg->front.iov_base;
244 void *const end = p + msg->front_alloc_len;
248 dout("%s sent %lu\n", __func__, monc->sub_renew_sent);
250 BUG_ON(monc->cur_mon < 0);
252 if (!monc->sub_renew_sent)
253 monc->sub_renew_sent = jiffies | 1; /* never 0 */
255 msg->hdr.version = cpu_to_le16(2);
257 for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
258 if (monc->subs[i].want)
261 BUG_ON(num < 1); /* monmap sub is always there */
262 ceph_encode_32(&p, num);
263 for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
267 if (!monc->subs[i].want)
270 len = sprintf(buf, "%s", ceph_sub_str[i]);
271 if (i == CEPH_SUB_MDSMAP &&
272 monc->fs_cluster_id != CEPH_FS_CLUSTER_ID_NONE)
273 len += sprintf(buf + len, ".%d", monc->fs_cluster_id);
275 dout("%s %s start %llu flags 0x%x\n", __func__, buf,
276 le64_to_cpu(monc->subs[i].item.start),
277 monc->subs[i].item.flags);
278 ceph_encode_string(&p, end, buf, len);
279 memcpy(p, &monc->subs[i].item, sizeof(monc->subs[i].item));
280 p += sizeof(monc->subs[i].item);
284 msg->front.iov_len = p - msg->front.iov_base;
285 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
286 ceph_msg_revoke(msg);
287 ceph_con_send(&monc->con, ceph_msg_get(msg));
290 static void handle_subscribe_ack(struct ceph_mon_client *monc,
291 struct ceph_msg *msg)
293 unsigned int seconds;
294 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
296 if (msg->front.iov_len < sizeof(*h))
298 seconds = le32_to_cpu(h->duration);
300 mutex_lock(&monc->mutex);
301 if (monc->sub_renew_sent) {
303 * This is only needed for legacy (infernalis or older)
304 * MONs -- see delayed_work().
306 monc->sub_renew_after = monc->sub_renew_sent +
307 (seconds >> 1) * HZ - 1;
308 dout("%s sent %lu duration %d renew after %lu\n", __func__,
309 monc->sub_renew_sent, seconds, monc->sub_renew_after);
310 monc->sub_renew_sent = 0;
312 dout("%s sent %lu renew after %lu, ignoring\n", __func__,
313 monc->sub_renew_sent, monc->sub_renew_after);
315 mutex_unlock(&monc->mutex);
318 pr_err("got corrupt subscribe-ack msg\n");
323 * Register interest in a map
325 * @sub: one of CEPH_SUB_*
326 * @epoch: X for "every map since X", or 0 for "just the latest"
328 static bool __ceph_monc_want_map(struct ceph_mon_client *monc, int sub,
329 u32 epoch, bool continuous)
331 __le64 start = cpu_to_le64(epoch);
332 u8 flags = !continuous ? CEPH_SUBSCRIBE_ONETIME : 0;
334 dout("%s %s epoch %u continuous %d\n", __func__, ceph_sub_str[sub],
337 if (monc->subs[sub].want &&
338 monc->subs[sub].item.start == start &&
339 monc->subs[sub].item.flags == flags)
342 monc->subs[sub].item.start = start;
343 monc->subs[sub].item.flags = flags;
344 monc->subs[sub].want = true;
349 bool ceph_monc_want_map(struct ceph_mon_client *monc, int sub, u32 epoch,
354 mutex_lock(&monc->mutex);
355 need_request = __ceph_monc_want_map(monc, sub, epoch, continuous);
356 mutex_unlock(&monc->mutex);
360 EXPORT_SYMBOL(ceph_monc_want_map);
363 * Keep track of which maps we have
365 * @sub: one of CEPH_SUB_*
367 static void __ceph_monc_got_map(struct ceph_mon_client *monc, int sub,
370 dout("%s %s epoch %u\n", __func__, ceph_sub_str[sub], epoch);
372 if (monc->subs[sub].want) {
373 if (monc->subs[sub].item.flags & CEPH_SUBSCRIBE_ONETIME)
374 monc->subs[sub].want = false;
376 monc->subs[sub].item.start = cpu_to_le64(epoch + 1);
379 monc->subs[sub].have = epoch;
382 void ceph_monc_got_map(struct ceph_mon_client *monc, int sub, u32 epoch)
384 mutex_lock(&monc->mutex);
385 __ceph_monc_got_map(monc, sub, epoch);
386 mutex_unlock(&monc->mutex);
388 EXPORT_SYMBOL(ceph_monc_got_map);
390 void ceph_monc_renew_subs(struct ceph_mon_client *monc)
392 mutex_lock(&monc->mutex);
393 __send_subscribe(monc);
394 mutex_unlock(&monc->mutex);
396 EXPORT_SYMBOL(ceph_monc_renew_subs);
399 * Wait for an osdmap with a given epoch.
401 * @epoch: epoch to wait for
402 * @timeout: in jiffies, 0 means "wait forever"
404 int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
405 unsigned long timeout)
407 unsigned long started = jiffies;
410 mutex_lock(&monc->mutex);
411 while (monc->subs[CEPH_SUB_OSDMAP].have < epoch) {
412 mutex_unlock(&monc->mutex);
414 if (timeout && time_after_eq(jiffies, started + timeout))
417 ret = wait_event_interruptible_timeout(monc->client->auth_wq,
418 monc->subs[CEPH_SUB_OSDMAP].have >= epoch,
419 ceph_timeout_jiffies(timeout));
423 mutex_lock(&monc->mutex);
426 mutex_unlock(&monc->mutex);
429 EXPORT_SYMBOL(ceph_monc_wait_osdmap);
432 * Open a session with a random monitor. Request monmap and osdmap,
433 * which are waited upon in __ceph_open_session().
435 int ceph_monc_open_session(struct ceph_mon_client *monc)
437 mutex_lock(&monc->mutex);
438 __ceph_monc_want_map(monc, CEPH_SUB_MONMAP, 0, true);
439 __ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, 0, false);
440 __open_session(monc);
441 __schedule_delayed(monc);
442 mutex_unlock(&monc->mutex);
445 EXPORT_SYMBOL(ceph_monc_open_session);
447 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
448 struct ceph_msg *msg)
450 struct ceph_client *client = monc->client;
451 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
454 mutex_lock(&monc->mutex);
456 dout("handle_monmap\n");
457 p = msg->front.iov_base;
458 end = p + msg->front.iov_len;
460 monmap = ceph_monmap_decode(p, end);
461 if (IS_ERR(monmap)) {
462 pr_err("problem decoding monmap, %d\n",
463 (int)PTR_ERR(monmap));
467 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
472 client->monc.monmap = monmap;
475 __ceph_monc_got_map(monc, CEPH_SUB_MONMAP, monc->monmap->epoch);
476 client->have_fsid = true;
479 mutex_unlock(&monc->mutex);
480 wake_up_all(&client->auth_wq);
484 * generic requests (currently statfs, mon_get_version)
486 DEFINE_RB_FUNCS(generic_request, struct ceph_mon_generic_request, tid, node)
488 static void release_generic_request(struct kref *kref)
490 struct ceph_mon_generic_request *req =
491 container_of(kref, struct ceph_mon_generic_request, kref);
493 dout("%s greq %p request %p reply %p\n", __func__, req, req->request,
495 WARN_ON(!RB_EMPTY_NODE(&req->node));
498 ceph_msg_put(req->reply);
500 ceph_msg_put(req->request);
505 static void put_generic_request(struct ceph_mon_generic_request *req)
508 kref_put(&req->kref, release_generic_request);
511 static void get_generic_request(struct ceph_mon_generic_request *req)
513 kref_get(&req->kref);
516 static struct ceph_mon_generic_request *
517 alloc_generic_request(struct ceph_mon_client *monc, gfp_t gfp)
519 struct ceph_mon_generic_request *req;
521 req = kzalloc(sizeof(*req), gfp);
526 kref_init(&req->kref);
527 RB_CLEAR_NODE(&req->node);
528 init_completion(&req->completion);
530 dout("%s greq %p\n", __func__, req);
534 static void register_generic_request(struct ceph_mon_generic_request *req)
536 struct ceph_mon_client *monc = req->monc;
540 get_generic_request(req);
541 req->tid = ++monc->last_tid;
542 insert_generic_request(&monc->generic_request_tree, req);
545 static void send_generic_request(struct ceph_mon_client *monc,
546 struct ceph_mon_generic_request *req)
550 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
551 req->request->hdr.tid = cpu_to_le64(req->tid);
552 ceph_con_send(&monc->con, ceph_msg_get(req->request));
555 static void __finish_generic_request(struct ceph_mon_generic_request *req)
557 struct ceph_mon_client *monc = req->monc;
559 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
560 erase_generic_request(&monc->generic_request_tree, req);
562 ceph_msg_revoke(req->request);
563 ceph_msg_revoke_incoming(req->reply);
566 static void finish_generic_request(struct ceph_mon_generic_request *req)
568 __finish_generic_request(req);
569 put_generic_request(req);
572 static void complete_generic_request(struct ceph_mon_generic_request *req)
574 if (req->complete_cb)
575 req->complete_cb(req);
577 complete_all(&req->completion);
578 put_generic_request(req);
581 static void cancel_generic_request(struct ceph_mon_generic_request *req)
583 struct ceph_mon_client *monc = req->monc;
584 struct ceph_mon_generic_request *lookup_req;
586 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
588 mutex_lock(&monc->mutex);
589 lookup_req = lookup_generic_request(&monc->generic_request_tree,
592 WARN_ON(lookup_req != req);
593 finish_generic_request(req);
596 mutex_unlock(&monc->mutex);
599 static int wait_generic_request(struct ceph_mon_generic_request *req)
603 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
604 ret = wait_for_completion_interruptible(&req->completion);
606 cancel_generic_request(req);
608 ret = req->result; /* completed */
613 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
614 struct ceph_msg_header *hdr,
617 struct ceph_mon_client *monc = con->private;
618 struct ceph_mon_generic_request *req;
619 u64 tid = le64_to_cpu(hdr->tid);
622 mutex_lock(&monc->mutex);
623 req = lookup_generic_request(&monc->generic_request_tree, tid);
625 dout("get_generic_reply %lld dne\n", tid);
629 dout("get_generic_reply %lld got %p\n", tid, req->reply);
631 m = ceph_msg_get(req->reply);
633 * we don't need to track the connection reading into
634 * this reply because we only have one open connection
638 mutex_unlock(&monc->mutex);
645 static void handle_statfs_reply(struct ceph_mon_client *monc,
646 struct ceph_msg *msg)
648 struct ceph_mon_generic_request *req;
649 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
650 u64 tid = le64_to_cpu(msg->hdr.tid);
652 dout("%s msg %p tid %llu\n", __func__, msg, tid);
654 if (msg->front.iov_len != sizeof(*reply))
657 mutex_lock(&monc->mutex);
658 req = lookup_generic_request(&monc->generic_request_tree, tid);
660 mutex_unlock(&monc->mutex);
665 *req->u.st = reply->st; /* struct */
666 __finish_generic_request(req);
667 mutex_unlock(&monc->mutex);
669 complete_generic_request(req);
673 pr_err("corrupt statfs reply, tid %llu\n", tid);
678 * Do a synchronous statfs().
680 int ceph_monc_do_statfs(struct ceph_mon_client *monc, u64 data_pool,
681 struct ceph_statfs *buf)
683 struct ceph_mon_generic_request *req;
684 struct ceph_mon_statfs *h;
687 req = alloc_generic_request(monc, GFP_NOFS);
691 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
696 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 64, GFP_NOFS, true);
701 req->request->hdr.version = cpu_to_le16(2);
703 mutex_lock(&monc->mutex);
704 register_generic_request(req);
705 /* fill out request */
706 h = req->request->front.iov_base;
707 h->monhdr.have_version = 0;
708 h->monhdr.session_mon = cpu_to_le16(-1);
709 h->monhdr.session_mon_tid = 0;
710 h->fsid = monc->monmap->fsid;
711 h->contains_data_pool = (data_pool != CEPH_NOPOOL);
712 h->data_pool = cpu_to_le64(data_pool);
713 send_generic_request(monc, req);
714 mutex_unlock(&monc->mutex);
716 ret = wait_generic_request(req);
718 put_generic_request(req);
721 EXPORT_SYMBOL(ceph_monc_do_statfs);
723 static void handle_get_version_reply(struct ceph_mon_client *monc,
724 struct ceph_msg *msg)
726 struct ceph_mon_generic_request *req;
727 u64 tid = le64_to_cpu(msg->hdr.tid);
728 void *p = msg->front.iov_base;
729 void *end = p + msg->front_alloc_len;
732 dout("%s msg %p tid %llu\n", __func__, msg, tid);
734 ceph_decode_need(&p, end, 2*sizeof(u64), bad);
735 handle = ceph_decode_64(&p);
736 if (tid != 0 && tid != handle)
739 mutex_lock(&monc->mutex);
740 req = lookup_generic_request(&monc->generic_request_tree, handle);
742 mutex_unlock(&monc->mutex);
747 req->u.newest = ceph_decode_64(&p);
748 __finish_generic_request(req);
749 mutex_unlock(&monc->mutex);
751 complete_generic_request(req);
755 pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
759 static struct ceph_mon_generic_request *
760 __ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
761 ceph_monc_callback_t cb, u64 private_data)
763 struct ceph_mon_generic_request *req;
765 req = alloc_generic_request(monc, GFP_NOIO);
769 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
770 sizeof(u64) + sizeof(u32) + strlen(what),
775 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 32, GFP_NOIO,
780 req->complete_cb = cb;
781 req->private_data = private_data;
783 mutex_lock(&monc->mutex);
784 register_generic_request(req);
786 void *p = req->request->front.iov_base;
787 void *const end = p + req->request->front_alloc_len;
789 ceph_encode_64(&p, req->tid); /* handle */
790 ceph_encode_string(&p, end, what, strlen(what));
793 send_generic_request(monc, req);
794 mutex_unlock(&monc->mutex);
799 put_generic_request(req);
800 return ERR_PTR(-ENOMEM);
804 * Send MMonGetVersion and wait for the reply.
806 * @what: one of "mdsmap", "osdmap" or "monmap"
808 int ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
811 struct ceph_mon_generic_request *req;
814 req = __ceph_monc_get_version(monc, what, NULL, 0);
818 ret = wait_generic_request(req);
820 *newest = req->u.newest;
822 put_generic_request(req);
825 EXPORT_SYMBOL(ceph_monc_get_version);
828 * Send MMonGetVersion,
830 * @what: one of "mdsmap", "osdmap" or "monmap"
832 int ceph_monc_get_version_async(struct ceph_mon_client *monc, const char *what,
833 ceph_monc_callback_t cb, u64 private_data)
835 struct ceph_mon_generic_request *req;
837 req = __ceph_monc_get_version(monc, what, cb, private_data);
841 put_generic_request(req);
844 EXPORT_SYMBOL(ceph_monc_get_version_async);
846 static void handle_command_ack(struct ceph_mon_client *monc,
847 struct ceph_msg *msg)
849 struct ceph_mon_generic_request *req;
850 void *p = msg->front.iov_base;
851 void *const end = p + msg->front_alloc_len;
852 u64 tid = le64_to_cpu(msg->hdr.tid);
854 dout("%s msg %p tid %llu\n", __func__, msg, tid);
856 ceph_decode_need(&p, end, sizeof(struct ceph_mon_request_header) +
858 p += sizeof(struct ceph_mon_request_header);
860 mutex_lock(&monc->mutex);
861 req = lookup_generic_request(&monc->generic_request_tree, tid);
863 mutex_unlock(&monc->mutex);
867 req->result = ceph_decode_32(&p);
868 __finish_generic_request(req);
869 mutex_unlock(&monc->mutex);
871 complete_generic_request(req);
875 pr_err("corrupt mon_command ack, tid %llu\n", tid);
879 int ceph_monc_blacklist_add(struct ceph_mon_client *monc,
880 struct ceph_entity_addr *client_addr)
882 struct ceph_mon_generic_request *req;
883 struct ceph_mon_command *h;
887 req = alloc_generic_request(monc, GFP_NOIO);
891 req->request = ceph_msg_new(CEPH_MSG_MON_COMMAND, 256, GFP_NOIO, true);
895 req->reply = ceph_msg_new(CEPH_MSG_MON_COMMAND_ACK, 512, GFP_NOIO,
900 mutex_lock(&monc->mutex);
901 register_generic_request(req);
902 h = req->request->front.iov_base;
903 h->monhdr.have_version = 0;
904 h->monhdr.session_mon = cpu_to_le16(-1);
905 h->monhdr.session_mon_tid = 0;
906 h->fsid = monc->monmap->fsid;
907 h->num_strs = cpu_to_le32(1);
908 len = sprintf(h->str, "{ \"prefix\": \"osd blacklist\", \
909 \"blacklistop\": \"add\", \
910 \"addr\": \"%pISpc/%u\" }",
911 &client_addr->in_addr, le32_to_cpu(client_addr->nonce));
912 h->str_len = cpu_to_le32(len);
913 send_generic_request(monc, req);
914 mutex_unlock(&monc->mutex);
916 ret = wait_generic_request(req);
918 put_generic_request(req);
921 EXPORT_SYMBOL(ceph_monc_blacklist_add);
924 * Resend pending generic requests.
926 static void __resend_generic_request(struct ceph_mon_client *monc)
928 struct ceph_mon_generic_request *req;
931 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
932 req = rb_entry(p, struct ceph_mon_generic_request, node);
933 ceph_msg_revoke(req->request);
934 ceph_msg_revoke_incoming(req->reply);
935 ceph_con_send(&monc->con, ceph_msg_get(req->request));
940 * Delayed work. If we haven't mounted yet, retry. Otherwise,
941 * renew/retry subscription as needed (in case it is timing out, or we
942 * got an ENOMEM). And keep the monitor connection alive.
944 static void delayed_work(struct work_struct *work)
946 struct ceph_mon_client *monc =
947 container_of(work, struct ceph_mon_client, delayed_work.work);
949 dout("monc delayed_work\n");
950 mutex_lock(&monc->mutex);
952 dout("%s continuing hunt\n", __func__);
953 reopen_session(monc);
955 int is_auth = ceph_auth_is_authenticated(monc->auth);
956 if (ceph_con_keepalive_expired(&monc->con,
957 CEPH_MONC_PING_TIMEOUT)) {
958 dout("monc keepalive timeout\n");
960 reopen_session(monc);
963 if (!monc->hunting) {
964 ceph_con_keepalive(&monc->con);
965 __validate_auth(monc);
969 !(monc->con.peer_features & CEPH_FEATURE_MON_STATEFUL_SUB)) {
970 unsigned long now = jiffies;
972 dout("%s renew subs? now %lu renew after %lu\n",
973 __func__, now, monc->sub_renew_after);
974 if (time_after_eq(now, monc->sub_renew_after))
975 __send_subscribe(monc);
978 __schedule_delayed(monc);
979 mutex_unlock(&monc->mutex);
983 * On startup, we build a temporary monmap populated with the IPs
984 * provided by mount(2).
986 static int build_initial_monmap(struct ceph_mon_client *monc)
988 struct ceph_options *opt = monc->client->options;
989 struct ceph_entity_addr *mon_addr = opt->mon_addr;
990 int num_mon = opt->num_mon;
993 /* build initial monmap */
994 monc->monmap = kzalloc(sizeof(*monc->monmap) +
995 num_mon*sizeof(monc->monmap->mon_inst[0]),
999 for (i = 0; i < num_mon; i++) {
1000 monc->monmap->mon_inst[i].addr = mon_addr[i];
1001 monc->monmap->mon_inst[i].addr.nonce = 0;
1002 monc->monmap->mon_inst[i].name.type =
1003 CEPH_ENTITY_TYPE_MON;
1004 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
1006 monc->monmap->num_mon = num_mon;
1010 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
1015 memset(monc, 0, sizeof(*monc));
1017 monc->monmap = NULL;
1018 mutex_init(&monc->mutex);
1020 err = build_initial_monmap(monc);
1025 /* authentication */
1026 monc->auth = ceph_auth_init(cl->options->name,
1028 if (IS_ERR(monc->auth)) {
1029 err = PTR_ERR(monc->auth);
1032 monc->auth->want_keys =
1033 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
1034 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
1038 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
1039 sizeof(struct ceph_mon_subscribe_ack),
1041 if (!monc->m_subscribe_ack)
1044 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 128,
1046 if (!monc->m_subscribe)
1047 goto out_subscribe_ack;
1049 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096,
1051 if (!monc->m_auth_reply)
1054 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_KERNEL, true);
1055 monc->pending_auth = 0;
1057 goto out_auth_reply;
1059 ceph_con_init(&monc->con, monc, &mon_con_ops,
1060 &monc->client->msgr);
1063 monc->had_a_connection = false;
1064 monc->hunt_mult = 1;
1066 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
1067 monc->generic_request_tree = RB_ROOT;
1070 monc->fs_cluster_id = CEPH_FS_CLUSTER_ID_NONE;
1075 ceph_msg_put(monc->m_auth_reply);
1077 ceph_msg_put(monc->m_subscribe);
1079 ceph_msg_put(monc->m_subscribe_ack);
1081 ceph_auth_destroy(monc->auth);
1083 kfree(monc->monmap);
1087 EXPORT_SYMBOL(ceph_monc_init);
1089 void ceph_monc_stop(struct ceph_mon_client *monc)
1092 cancel_delayed_work_sync(&monc->delayed_work);
1094 mutex_lock(&monc->mutex);
1095 __close_session(monc);
1097 mutex_unlock(&monc->mutex);
1100 * flush msgr queue before we destroy ourselves to ensure that:
1101 * - any work that references our embedded con is finished.
1102 * - any osd_client or other work that may reference an authorizer
1103 * finishes before we shut down the auth subsystem.
1107 ceph_auth_destroy(monc->auth);
1109 WARN_ON(!RB_EMPTY_ROOT(&monc->generic_request_tree));
1111 ceph_msg_put(monc->m_auth);
1112 ceph_msg_put(monc->m_auth_reply);
1113 ceph_msg_put(monc->m_subscribe);
1114 ceph_msg_put(monc->m_subscribe_ack);
1116 kfree(monc->monmap);
1118 EXPORT_SYMBOL(ceph_monc_stop);
1120 static void finish_hunting(struct ceph_mon_client *monc)
1122 if (monc->hunting) {
1123 dout("%s found mon%d\n", __func__, monc->cur_mon);
1124 monc->hunting = false;
1125 monc->had_a_connection = true;
1126 monc->hunt_mult /= 2; /* reduce by 50% */
1127 if (monc->hunt_mult < 1)
1128 monc->hunt_mult = 1;
1132 static void handle_auth_reply(struct ceph_mon_client *monc,
1133 struct ceph_msg *msg)
1138 mutex_lock(&monc->mutex);
1139 was_auth = ceph_auth_is_authenticated(monc->auth);
1140 monc->pending_auth = 0;
1141 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
1143 monc->m_auth->front.iov_base,
1144 monc->m_auth->front_alloc_len);
1146 __send_prepared_auth_request(monc, ret);
1150 finish_hunting(monc);
1153 monc->client->auth_err = ret;
1154 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
1155 dout("authenticated, starting session\n");
1157 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
1158 monc->client->msgr.inst.name.num =
1159 cpu_to_le64(monc->auth->global_id);
1161 __send_subscribe(monc);
1162 __resend_generic_request(monc);
1164 pr_info("mon%d %s session established\n", monc->cur_mon,
1165 ceph_pr_addr(&monc->con.peer_addr.in_addr));
1169 mutex_unlock(&monc->mutex);
1170 if (monc->client->auth_err < 0)
1171 wake_up_all(&monc->client->auth_wq);
1174 static int __validate_auth(struct ceph_mon_client *monc)
1178 if (monc->pending_auth)
1181 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
1182 monc->m_auth->front_alloc_len);
1184 return ret; /* either an error, or no need to authenticate */
1185 __send_prepared_auth_request(monc, ret);
1189 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
1193 mutex_lock(&monc->mutex);
1194 ret = __validate_auth(monc);
1195 mutex_unlock(&monc->mutex);
1198 EXPORT_SYMBOL(ceph_monc_validate_auth);
1201 * handle incoming message
1203 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1205 struct ceph_mon_client *monc = con->private;
1206 int type = le16_to_cpu(msg->hdr.type);
1212 case CEPH_MSG_AUTH_REPLY:
1213 handle_auth_reply(monc, msg);
1216 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1217 handle_subscribe_ack(monc, msg);
1220 case CEPH_MSG_STATFS_REPLY:
1221 handle_statfs_reply(monc, msg);
1224 case CEPH_MSG_MON_GET_VERSION_REPLY:
1225 handle_get_version_reply(monc, msg);
1228 case CEPH_MSG_MON_COMMAND_ACK:
1229 handle_command_ack(monc, msg);
1232 case CEPH_MSG_MON_MAP:
1233 ceph_monc_handle_map(monc, msg);
1236 case CEPH_MSG_OSD_MAP:
1237 ceph_osdc_handle_map(&monc->client->osdc, msg);
1241 /* can the chained handler handle it? */
1242 if (monc->client->extra_mon_dispatch &&
1243 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1246 pr_err("received unknown message type %d %s\n", type,
1247 ceph_msg_type_name(type));
1253 * Allocate memory for incoming message
1255 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1256 struct ceph_msg_header *hdr,
1259 struct ceph_mon_client *monc = con->private;
1260 int type = le16_to_cpu(hdr->type);
1261 int front_len = le32_to_cpu(hdr->front_len);
1262 struct ceph_msg *m = NULL;
1267 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1268 m = ceph_msg_get(monc->m_subscribe_ack);
1270 case CEPH_MSG_STATFS_REPLY:
1271 case CEPH_MSG_MON_COMMAND_ACK:
1272 return get_generic_reply(con, hdr, skip);
1273 case CEPH_MSG_AUTH_REPLY:
1274 m = ceph_msg_get(monc->m_auth_reply);
1276 case CEPH_MSG_MON_GET_VERSION_REPLY:
1277 if (le64_to_cpu(hdr->tid) != 0)
1278 return get_generic_reply(con, hdr, skip);
1281 * Older OSDs don't set reply tid even if the orignal
1282 * request had a non-zero tid. Workaround this weirdness
1283 * by falling through to the allocate case.
1285 case CEPH_MSG_MON_MAP:
1286 case CEPH_MSG_MDS_MAP:
1287 case CEPH_MSG_OSD_MAP:
1288 case CEPH_MSG_FS_MAP_USER:
1289 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1291 return NULL; /* ENOMEM--return skip == 0 */
1296 pr_info("alloc_msg unknown type %d\n", type);
1298 } else if (front_len > m->front_alloc_len) {
1299 pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1300 front_len, m->front_alloc_len,
1301 (unsigned int)con->peer_name.type,
1302 le64_to_cpu(con->peer_name.num));
1304 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1311 * If the monitor connection resets, pick a new monitor and resubmit
1312 * any pending requests.
1314 static void mon_fault(struct ceph_connection *con)
1316 struct ceph_mon_client *monc = con->private;
1318 mutex_lock(&monc->mutex);
1319 dout("%s mon%d\n", __func__, monc->cur_mon);
1320 if (monc->cur_mon >= 0) {
1321 if (!monc->hunting) {
1322 dout("%s hunting for new mon\n", __func__);
1323 reopen_session(monc);
1324 __schedule_delayed(monc);
1326 dout("%s already hunting\n", __func__);
1329 mutex_unlock(&monc->mutex);
1333 * We can ignore refcounting on the connection struct, as all references
1334 * will come from the messenger workqueue, which is drained prior to
1335 * mon_client destruction.
1337 static struct ceph_connection *con_get(struct ceph_connection *con)
1342 static void con_put(struct ceph_connection *con)
1346 static const struct ceph_connection_operations mon_con_ops = {
1349 .dispatch = dispatch,
1351 .alloc_msg = mon_alloc_msg,