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 * 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) {
302 monc->sub_renew_after = monc->sub_renew_sent +
303 (seconds >> 1) * HZ - 1;
304 dout("%s sent %lu duration %d renew after %lu\n", __func__,
305 monc->sub_renew_sent, seconds, monc->sub_renew_after);
306 monc->sub_renew_sent = 0;
308 dout("%s sent %lu renew after %lu, ignoring\n", __func__,
309 monc->sub_renew_sent, monc->sub_renew_after);
311 mutex_unlock(&monc->mutex);
314 pr_err("got corrupt subscribe-ack msg\n");
319 * Register interest in a map
321 * @sub: one of CEPH_SUB_*
322 * @epoch: X for "every map since X", or 0 for "just the latest"
324 static bool __ceph_monc_want_map(struct ceph_mon_client *monc, int sub,
325 u32 epoch, bool continuous)
327 __le64 start = cpu_to_le64(epoch);
328 u8 flags = !continuous ? CEPH_SUBSCRIBE_ONETIME : 0;
330 dout("%s %s epoch %u continuous %d\n", __func__, ceph_sub_str[sub],
333 if (monc->subs[sub].want &&
334 monc->subs[sub].item.start == start &&
335 monc->subs[sub].item.flags == flags)
338 monc->subs[sub].item.start = start;
339 monc->subs[sub].item.flags = flags;
340 monc->subs[sub].want = true;
345 bool ceph_monc_want_map(struct ceph_mon_client *monc, int sub, u32 epoch,
350 mutex_lock(&monc->mutex);
351 need_request = __ceph_monc_want_map(monc, sub, epoch, continuous);
352 mutex_unlock(&monc->mutex);
356 EXPORT_SYMBOL(ceph_monc_want_map);
359 * Keep track of which maps we have
361 * @sub: one of CEPH_SUB_*
363 static void __ceph_monc_got_map(struct ceph_mon_client *monc, int sub,
366 dout("%s %s epoch %u\n", __func__, ceph_sub_str[sub], epoch);
368 if (monc->subs[sub].want) {
369 if (monc->subs[sub].item.flags & CEPH_SUBSCRIBE_ONETIME)
370 monc->subs[sub].want = false;
372 monc->subs[sub].item.start = cpu_to_le64(epoch + 1);
375 monc->subs[sub].have = epoch;
378 void ceph_monc_got_map(struct ceph_mon_client *monc, int sub, u32 epoch)
380 mutex_lock(&monc->mutex);
381 __ceph_monc_got_map(monc, sub, epoch);
382 mutex_unlock(&monc->mutex);
384 EXPORT_SYMBOL(ceph_monc_got_map);
386 void ceph_monc_renew_subs(struct ceph_mon_client *monc)
388 mutex_lock(&monc->mutex);
389 __send_subscribe(monc);
390 mutex_unlock(&monc->mutex);
392 EXPORT_SYMBOL(ceph_monc_renew_subs);
395 * Wait for an osdmap with a given epoch.
397 * @epoch: epoch to wait for
398 * @timeout: in jiffies, 0 means "wait forever"
400 int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
401 unsigned long timeout)
403 unsigned long started = jiffies;
406 mutex_lock(&monc->mutex);
407 while (monc->subs[CEPH_SUB_OSDMAP].have < epoch) {
408 mutex_unlock(&monc->mutex);
410 if (timeout && time_after_eq(jiffies, started + timeout))
413 ret = wait_event_interruptible_timeout(monc->client->auth_wq,
414 monc->subs[CEPH_SUB_OSDMAP].have >= epoch,
415 ceph_timeout_jiffies(timeout));
419 mutex_lock(&monc->mutex);
422 mutex_unlock(&monc->mutex);
425 EXPORT_SYMBOL(ceph_monc_wait_osdmap);
428 * Open a session with a random monitor. Request monmap and osdmap,
429 * which are waited upon in __ceph_open_session().
431 int ceph_monc_open_session(struct ceph_mon_client *monc)
433 mutex_lock(&monc->mutex);
434 __ceph_monc_want_map(monc, CEPH_SUB_MONMAP, 0, true);
435 __ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, 0, false);
436 __open_session(monc);
437 __schedule_delayed(monc);
438 mutex_unlock(&monc->mutex);
441 EXPORT_SYMBOL(ceph_monc_open_session);
443 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
444 struct ceph_msg *msg)
446 struct ceph_client *client = monc->client;
447 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
450 mutex_lock(&monc->mutex);
452 dout("handle_monmap\n");
453 p = msg->front.iov_base;
454 end = p + msg->front.iov_len;
456 monmap = ceph_monmap_decode(p, end);
457 if (IS_ERR(monmap)) {
458 pr_err("problem decoding monmap, %d\n",
459 (int)PTR_ERR(monmap));
463 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
468 client->monc.monmap = monmap;
471 __ceph_monc_got_map(monc, CEPH_SUB_MONMAP, monc->monmap->epoch);
472 client->have_fsid = true;
475 mutex_unlock(&monc->mutex);
476 wake_up_all(&client->auth_wq);
480 * generic requests (currently statfs, mon_get_version)
482 DEFINE_RB_FUNCS(generic_request, struct ceph_mon_generic_request, tid, node)
484 static void release_generic_request(struct kref *kref)
486 struct ceph_mon_generic_request *req =
487 container_of(kref, struct ceph_mon_generic_request, kref);
489 dout("%s greq %p request %p reply %p\n", __func__, req, req->request,
491 WARN_ON(!RB_EMPTY_NODE(&req->node));
494 ceph_msg_put(req->reply);
496 ceph_msg_put(req->request);
501 static void put_generic_request(struct ceph_mon_generic_request *req)
504 kref_put(&req->kref, release_generic_request);
507 static void get_generic_request(struct ceph_mon_generic_request *req)
509 kref_get(&req->kref);
512 static struct ceph_mon_generic_request *
513 alloc_generic_request(struct ceph_mon_client *monc, gfp_t gfp)
515 struct ceph_mon_generic_request *req;
517 req = kzalloc(sizeof(*req), gfp);
522 kref_init(&req->kref);
523 RB_CLEAR_NODE(&req->node);
524 init_completion(&req->completion);
526 dout("%s greq %p\n", __func__, req);
530 static void register_generic_request(struct ceph_mon_generic_request *req)
532 struct ceph_mon_client *monc = req->monc;
536 get_generic_request(req);
537 req->tid = ++monc->last_tid;
538 insert_generic_request(&monc->generic_request_tree, req);
541 static void send_generic_request(struct ceph_mon_client *monc,
542 struct ceph_mon_generic_request *req)
546 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
547 req->request->hdr.tid = cpu_to_le64(req->tid);
548 ceph_con_send(&monc->con, ceph_msg_get(req->request));
551 static void __finish_generic_request(struct ceph_mon_generic_request *req)
553 struct ceph_mon_client *monc = req->monc;
555 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
556 erase_generic_request(&monc->generic_request_tree, req);
558 ceph_msg_revoke(req->request);
559 ceph_msg_revoke_incoming(req->reply);
562 static void finish_generic_request(struct ceph_mon_generic_request *req)
564 __finish_generic_request(req);
565 put_generic_request(req);
568 static void complete_generic_request(struct ceph_mon_generic_request *req)
570 if (req->complete_cb)
571 req->complete_cb(req);
573 complete_all(&req->completion);
574 put_generic_request(req);
577 void cancel_generic_request(struct ceph_mon_generic_request *req)
579 struct ceph_mon_client *monc = req->monc;
580 struct ceph_mon_generic_request *lookup_req;
582 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
584 mutex_lock(&monc->mutex);
585 lookup_req = lookup_generic_request(&monc->generic_request_tree,
588 WARN_ON(lookup_req != req);
589 finish_generic_request(req);
592 mutex_unlock(&monc->mutex);
595 static int wait_generic_request(struct ceph_mon_generic_request *req)
599 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
600 ret = wait_for_completion_interruptible(&req->completion);
602 cancel_generic_request(req);
604 ret = req->result; /* completed */
609 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
610 struct ceph_msg_header *hdr,
613 struct ceph_mon_client *monc = con->private;
614 struct ceph_mon_generic_request *req;
615 u64 tid = le64_to_cpu(hdr->tid);
618 mutex_lock(&monc->mutex);
619 req = lookup_generic_request(&monc->generic_request_tree, tid);
621 dout("get_generic_reply %lld dne\n", tid);
625 dout("get_generic_reply %lld got %p\n", tid, req->reply);
627 m = ceph_msg_get(req->reply);
629 * we don't need to track the connection reading into
630 * this reply because we only have one open connection
634 mutex_unlock(&monc->mutex);
641 static void handle_statfs_reply(struct ceph_mon_client *monc,
642 struct ceph_msg *msg)
644 struct ceph_mon_generic_request *req;
645 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
646 u64 tid = le64_to_cpu(msg->hdr.tid);
648 dout("%s msg %p tid %llu\n", __func__, msg, tid);
650 if (msg->front.iov_len != sizeof(*reply))
653 mutex_lock(&monc->mutex);
654 req = lookup_generic_request(&monc->generic_request_tree, tid);
656 mutex_unlock(&monc->mutex);
661 *req->u.st = reply->st; /* struct */
662 __finish_generic_request(req);
663 mutex_unlock(&monc->mutex);
665 complete_generic_request(req);
669 pr_err("corrupt statfs reply, tid %llu\n", tid);
674 * Do a synchronous statfs().
676 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
678 struct ceph_mon_generic_request *req;
679 struct ceph_mon_statfs *h;
682 req = alloc_generic_request(monc, GFP_NOFS);
686 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
691 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 64, GFP_NOFS, true);
697 mutex_lock(&monc->mutex);
698 register_generic_request(req);
699 /* fill out request */
700 h = req->request->front.iov_base;
701 h->monhdr.have_version = 0;
702 h->monhdr.session_mon = cpu_to_le16(-1);
703 h->monhdr.session_mon_tid = 0;
704 h->fsid = monc->monmap->fsid;
705 send_generic_request(monc, req);
706 mutex_unlock(&monc->mutex);
708 ret = wait_generic_request(req);
710 put_generic_request(req);
713 EXPORT_SYMBOL(ceph_monc_do_statfs);
715 static void handle_get_version_reply(struct ceph_mon_client *monc,
716 struct ceph_msg *msg)
718 struct ceph_mon_generic_request *req;
719 u64 tid = le64_to_cpu(msg->hdr.tid);
720 void *p = msg->front.iov_base;
721 void *end = p + msg->front_alloc_len;
724 dout("%s msg %p tid %llu\n", __func__, msg, tid);
726 ceph_decode_need(&p, end, 2*sizeof(u64), bad);
727 handle = ceph_decode_64(&p);
728 if (tid != 0 && tid != handle)
731 mutex_lock(&monc->mutex);
732 req = lookup_generic_request(&monc->generic_request_tree, handle);
734 mutex_unlock(&monc->mutex);
739 req->u.newest = ceph_decode_64(&p);
740 __finish_generic_request(req);
741 mutex_unlock(&monc->mutex);
743 complete_generic_request(req);
747 pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
751 static struct ceph_mon_generic_request *
752 __ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
753 ceph_monc_callback_t cb, u64 private_data)
755 struct ceph_mon_generic_request *req;
757 req = alloc_generic_request(monc, GFP_NOIO);
761 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
762 sizeof(u64) + sizeof(u32) + strlen(what),
767 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 32, GFP_NOIO,
772 req->complete_cb = cb;
773 req->private_data = private_data;
775 mutex_lock(&monc->mutex);
776 register_generic_request(req);
778 void *p = req->request->front.iov_base;
779 void *const end = p + req->request->front_alloc_len;
781 ceph_encode_64(&p, req->tid); /* handle */
782 ceph_encode_string(&p, end, what, strlen(what));
785 send_generic_request(monc, req);
786 mutex_unlock(&monc->mutex);
791 put_generic_request(req);
792 return ERR_PTR(-ENOMEM);
796 * Send MMonGetVersion and wait for the reply.
798 * @what: one of "mdsmap", "osdmap" or "monmap"
800 int ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
803 struct ceph_mon_generic_request *req;
806 req = __ceph_monc_get_version(monc, what, NULL, 0);
810 ret = wait_generic_request(req);
812 *newest = req->u.newest;
814 put_generic_request(req);
817 EXPORT_SYMBOL(ceph_monc_get_version);
820 * Send MMonGetVersion,
822 * @what: one of "mdsmap", "osdmap" or "monmap"
824 int ceph_monc_get_version_async(struct ceph_mon_client *monc, const char *what,
825 ceph_monc_callback_t cb, u64 private_data)
827 struct ceph_mon_generic_request *req;
829 req = __ceph_monc_get_version(monc, what, cb, private_data);
833 put_generic_request(req);
836 EXPORT_SYMBOL(ceph_monc_get_version_async);
839 * Resend pending generic requests.
841 static void __resend_generic_request(struct ceph_mon_client *monc)
843 struct ceph_mon_generic_request *req;
846 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
847 req = rb_entry(p, struct ceph_mon_generic_request, node);
848 ceph_msg_revoke(req->request);
849 ceph_msg_revoke_incoming(req->reply);
850 ceph_con_send(&monc->con, ceph_msg_get(req->request));
855 * Delayed work. If we haven't mounted yet, retry. Otherwise,
856 * renew/retry subscription as needed (in case it is timing out, or we
857 * got an ENOMEM). And keep the monitor connection alive.
859 static void delayed_work(struct work_struct *work)
861 struct ceph_mon_client *monc =
862 container_of(work, struct ceph_mon_client, delayed_work.work);
864 dout("monc delayed_work\n");
865 mutex_lock(&monc->mutex);
867 dout("%s continuing hunt\n", __func__);
868 reopen_session(monc);
870 int is_auth = ceph_auth_is_authenticated(monc->auth);
871 if (ceph_con_keepalive_expired(&monc->con,
872 CEPH_MONC_PING_TIMEOUT)) {
873 dout("monc keepalive timeout\n");
875 reopen_session(monc);
878 if (!monc->hunting) {
879 ceph_con_keepalive(&monc->con);
880 __validate_auth(monc);
884 unsigned long now = jiffies;
886 dout("%s renew subs? now %lu renew after %lu\n",
887 __func__, now, monc->sub_renew_after);
888 if (time_after_eq(now, monc->sub_renew_after))
889 __send_subscribe(monc);
892 __schedule_delayed(monc);
893 mutex_unlock(&monc->mutex);
897 * On startup, we build a temporary monmap populated with the IPs
898 * provided by mount(2).
900 static int build_initial_monmap(struct ceph_mon_client *monc)
902 struct ceph_options *opt = monc->client->options;
903 struct ceph_entity_addr *mon_addr = opt->mon_addr;
904 int num_mon = opt->num_mon;
907 /* build initial monmap */
908 monc->monmap = kzalloc(sizeof(*monc->monmap) +
909 num_mon*sizeof(monc->monmap->mon_inst[0]),
913 for (i = 0; i < num_mon; i++) {
914 monc->monmap->mon_inst[i].addr = mon_addr[i];
915 monc->monmap->mon_inst[i].addr.nonce = 0;
916 monc->monmap->mon_inst[i].name.type =
917 CEPH_ENTITY_TYPE_MON;
918 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
920 monc->monmap->num_mon = num_mon;
924 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
929 memset(monc, 0, sizeof(*monc));
932 mutex_init(&monc->mutex);
934 err = build_initial_monmap(monc);
940 monc->auth = ceph_auth_init(cl->options->name,
942 if (IS_ERR(monc->auth)) {
943 err = PTR_ERR(monc->auth);
946 monc->auth->want_keys =
947 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
948 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
952 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
953 sizeof(struct ceph_mon_subscribe_ack),
955 if (!monc->m_subscribe_ack)
958 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 128, GFP_NOFS,
960 if (!monc->m_subscribe)
961 goto out_subscribe_ack;
963 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
965 if (!monc->m_auth_reply)
968 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
969 monc->pending_auth = 0;
973 ceph_con_init(&monc->con, monc, &mon_con_ops,
974 &monc->client->msgr);
977 monc->had_a_connection = false;
980 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
981 monc->generic_request_tree = RB_ROOT;
984 monc->fs_cluster_id = CEPH_FS_CLUSTER_ID_NONE;
989 ceph_msg_put(monc->m_auth_reply);
991 ceph_msg_put(monc->m_subscribe);
993 ceph_msg_put(monc->m_subscribe_ack);
995 ceph_auth_destroy(monc->auth);
1001 EXPORT_SYMBOL(ceph_monc_init);
1003 void ceph_monc_stop(struct ceph_mon_client *monc)
1006 cancel_delayed_work_sync(&monc->delayed_work);
1008 mutex_lock(&monc->mutex);
1009 __close_session(monc);
1011 mutex_unlock(&monc->mutex);
1014 * flush msgr queue before we destroy ourselves to ensure that:
1015 * - any work that references our embedded con is finished.
1016 * - any osd_client or other work that may reference an authorizer
1017 * finishes before we shut down the auth subsystem.
1021 ceph_auth_destroy(monc->auth);
1023 WARN_ON(!RB_EMPTY_ROOT(&monc->generic_request_tree));
1025 ceph_msg_put(monc->m_auth);
1026 ceph_msg_put(monc->m_auth_reply);
1027 ceph_msg_put(monc->m_subscribe);
1028 ceph_msg_put(monc->m_subscribe_ack);
1030 kfree(monc->monmap);
1032 EXPORT_SYMBOL(ceph_monc_stop);
1034 static void finish_hunting(struct ceph_mon_client *monc)
1036 if (monc->hunting) {
1037 dout("%s found mon%d\n", __func__, monc->cur_mon);
1038 monc->hunting = false;
1039 monc->had_a_connection = true;
1040 monc->hunt_mult /= 2; /* reduce by 50% */
1041 if (monc->hunt_mult < 1)
1042 monc->hunt_mult = 1;
1046 static void handle_auth_reply(struct ceph_mon_client *monc,
1047 struct ceph_msg *msg)
1052 mutex_lock(&monc->mutex);
1053 was_auth = ceph_auth_is_authenticated(monc->auth);
1054 monc->pending_auth = 0;
1055 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
1057 monc->m_auth->front.iov_base,
1058 monc->m_auth->front_alloc_len);
1060 __send_prepared_auth_request(monc, ret);
1064 finish_hunting(monc);
1067 monc->client->auth_err = ret;
1068 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
1069 dout("authenticated, starting session\n");
1071 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
1072 monc->client->msgr.inst.name.num =
1073 cpu_to_le64(monc->auth->global_id);
1075 __send_subscribe(monc);
1076 __resend_generic_request(monc);
1078 pr_info("mon%d %s session established\n", monc->cur_mon,
1079 ceph_pr_addr(&monc->con.peer_addr.in_addr));
1083 mutex_unlock(&monc->mutex);
1084 if (monc->client->auth_err < 0)
1085 wake_up_all(&monc->client->auth_wq);
1088 static int __validate_auth(struct ceph_mon_client *monc)
1092 if (monc->pending_auth)
1095 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
1096 monc->m_auth->front_alloc_len);
1098 return ret; /* either an error, or no need to authenticate */
1099 __send_prepared_auth_request(monc, ret);
1103 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
1107 mutex_lock(&monc->mutex);
1108 ret = __validate_auth(monc);
1109 mutex_unlock(&monc->mutex);
1112 EXPORT_SYMBOL(ceph_monc_validate_auth);
1115 * handle incoming message
1117 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1119 struct ceph_mon_client *monc = con->private;
1120 int type = le16_to_cpu(msg->hdr.type);
1126 case CEPH_MSG_AUTH_REPLY:
1127 handle_auth_reply(monc, msg);
1130 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1131 handle_subscribe_ack(monc, msg);
1134 case CEPH_MSG_STATFS_REPLY:
1135 handle_statfs_reply(monc, msg);
1138 case CEPH_MSG_MON_GET_VERSION_REPLY:
1139 handle_get_version_reply(monc, msg);
1142 case CEPH_MSG_MON_MAP:
1143 ceph_monc_handle_map(monc, msg);
1146 case CEPH_MSG_OSD_MAP:
1147 ceph_osdc_handle_map(&monc->client->osdc, msg);
1151 /* can the chained handler handle it? */
1152 if (monc->client->extra_mon_dispatch &&
1153 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1156 pr_err("received unknown message type %d %s\n", type,
1157 ceph_msg_type_name(type));
1163 * Allocate memory for incoming message
1165 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1166 struct ceph_msg_header *hdr,
1169 struct ceph_mon_client *monc = con->private;
1170 int type = le16_to_cpu(hdr->type);
1171 int front_len = le32_to_cpu(hdr->front_len);
1172 struct ceph_msg *m = NULL;
1177 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1178 m = ceph_msg_get(monc->m_subscribe_ack);
1180 case CEPH_MSG_STATFS_REPLY:
1181 return get_generic_reply(con, hdr, skip);
1182 case CEPH_MSG_AUTH_REPLY:
1183 m = ceph_msg_get(monc->m_auth_reply);
1185 case CEPH_MSG_MON_GET_VERSION_REPLY:
1186 if (le64_to_cpu(hdr->tid) != 0)
1187 return get_generic_reply(con, hdr, skip);
1190 * Older OSDs don't set reply tid even if the orignal
1191 * request had a non-zero tid. Workaround this weirdness
1192 * by falling through to the allocate case.
1194 case CEPH_MSG_MON_MAP:
1195 case CEPH_MSG_MDS_MAP:
1196 case CEPH_MSG_OSD_MAP:
1197 case CEPH_MSG_FS_MAP_USER:
1198 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1200 return NULL; /* ENOMEM--return skip == 0 */
1205 pr_info("alloc_msg unknown type %d\n", type);
1207 } else if (front_len > m->front_alloc_len) {
1208 pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1209 front_len, m->front_alloc_len,
1210 (unsigned int)con->peer_name.type,
1211 le64_to_cpu(con->peer_name.num));
1213 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1220 * If the monitor connection resets, pick a new monitor and resubmit
1221 * any pending requests.
1223 static void mon_fault(struct ceph_connection *con)
1225 struct ceph_mon_client *monc = con->private;
1227 mutex_lock(&monc->mutex);
1228 dout("%s mon%d\n", __func__, monc->cur_mon);
1229 if (monc->cur_mon >= 0) {
1230 if (!monc->hunting) {
1231 dout("%s hunting for new mon\n", __func__);
1232 reopen_session(monc);
1233 __schedule_delayed(monc);
1235 dout("%s already hunting\n", __func__);
1238 mutex_unlock(&monc->mutex);
1242 * We can ignore refcounting on the connection struct, as all references
1243 * will come from the messenger workqueue, which is drained prior to
1244 * mon_client destruction.
1246 static struct ceph_connection *con_get(struct ceph_connection *con)
1251 static void con_put(struct ceph_connection *con)
1255 static const struct ceph_connection_operations mon_con_ops = {
1258 .dispatch = dispatch,
1260 .alloc_msg = mon_alloc_msg,