1 #include "ceph_debug.h"
3 #include <linux/types.h>
4 #include <linux/slab.h>
5 #include <linux/random.h>
6 #include <linux/sched.h>
8 #include "mon_client.h"
14 * Interact with Ceph monitor cluster. Handle requests for new map
15 * versions, and periodically resend as needed. Also implement
16 * statfs() and umount().
18 * A small cluster of Ceph "monitors" are responsible for managing critical
19 * cluster configuration and state information. An odd number (e.g., 3, 5)
20 * of cmon daemons use a modified version of the Paxos part-time parliament
21 * algorithm to manage the MDS map (mds cluster membership), OSD map, and
22 * list of clients who have mounted the file system.
24 * We maintain an open, active session with a monitor at all times in order to
25 * receive timely MDSMap updates. We periodically send a keepalive byte on the
26 * TCP socket to ensure we detect a failure. If the connection does break, we
27 * randomly hunt for a new monitor. Once the connection is reestablished, we
28 * resend any outstanding requests.
31 static const struct ceph_connection_operations mon_con_ops;
33 static int __validate_auth(struct ceph_mon_client *monc);
36 * Decode a monmap blob (e.g., during mount).
38 struct ceph_monmap *ceph_monmap_decode(void *p, void *end)
40 struct ceph_monmap *m = NULL;
42 struct ceph_fsid fsid;
47 ceph_decode_32_safe(&p, end, len, bad);
48 ceph_decode_need(&p, end, len, bad);
50 dout("monmap_decode %p %p len %d\n", p, end, (int)(end-p));
52 ceph_decode_16_safe(&p, end, version, bad);
54 ceph_decode_need(&p, end, sizeof(fsid) + 2*sizeof(u32), bad);
55 ceph_decode_copy(&p, &fsid, sizeof(fsid));
56 epoch = ceph_decode_32(&p);
58 num_mon = ceph_decode_32(&p);
59 ceph_decode_need(&p, end, num_mon*sizeof(m->mon_inst[0]), bad);
61 if (num_mon >= CEPH_MAX_MON)
63 m = kmalloc(sizeof(*m) + sizeof(m->mon_inst[0])*num_mon, GFP_NOFS);
65 return ERR_PTR(-ENOMEM);
69 ceph_decode_copy(&p, m->mon_inst, num_mon*sizeof(m->mon_inst[0]));
70 for (i = 0; i < num_mon; i++)
71 ceph_decode_addr(&m->mon_inst[i].addr);
73 dout("monmap_decode epoch %d, num_mon %d\n", m->epoch,
75 for (i = 0; i < m->num_mon; i++)
76 dout("monmap_decode mon%d is %s\n", i,
77 pr_addr(&m->mon_inst[i].addr.in_addr));
81 dout("monmap_decode failed with %d\n", err);
87 * return true if *addr is included in the monmap.
89 int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr)
93 for (i = 0; i < m->num_mon; i++)
94 if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0)
100 * Send an auth request.
102 static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len)
104 monc->pending_auth = 1;
105 monc->m_auth->front.iov_len = len;
106 monc->m_auth->hdr.front_len = cpu_to_le32(len);
107 ceph_con_revoke(monc->con, monc->m_auth);
108 ceph_msg_get(monc->m_auth); /* keep our ref */
109 ceph_con_send(monc->con, monc->m_auth);
113 * Close monitor session, if any.
115 static void __close_session(struct ceph_mon_client *monc)
118 dout("__close_session closing mon%d\n", monc->cur_mon);
119 ceph_con_revoke(monc->con, monc->m_auth);
120 ceph_con_close(monc->con);
122 monc->pending_auth = 0;
123 ceph_auth_reset(monc->auth);
128 * Open a session with a (new) monitor.
130 static int __open_session(struct ceph_mon_client *monc)
135 if (monc->cur_mon < 0) {
136 get_random_bytes(&r, 1);
137 monc->cur_mon = r % monc->monmap->num_mon;
138 dout("open_session num=%d r=%d -> mon%d\n",
139 monc->monmap->num_mon, r, monc->cur_mon);
141 monc->sub_renew_after = jiffies; /* i.e., expired */
142 monc->want_next_osdmap = !!monc->want_next_osdmap;
144 dout("open_session mon%d opening\n", monc->cur_mon);
145 monc->con->peer_name.type = CEPH_ENTITY_TYPE_MON;
146 monc->con->peer_name.num = cpu_to_le64(monc->cur_mon);
147 ceph_con_open(monc->con,
148 &monc->monmap->mon_inst[monc->cur_mon].addr);
150 /* initiatiate authentication handshake */
151 ret = ceph_auth_build_hello(monc->auth,
152 monc->m_auth->front.iov_base,
153 monc->m_auth->front_max);
154 __send_prepared_auth_request(monc, ret);
156 dout("open_session mon%d already open\n", monc->cur_mon);
161 static bool __sub_expired(struct ceph_mon_client *monc)
163 return time_after_eq(jiffies, monc->sub_renew_after);
167 * Reschedule delayed work timer.
169 static void __schedule_delayed(struct ceph_mon_client *monc)
173 if (monc->cur_mon < 0 || __sub_expired(monc))
177 dout("__schedule_delayed after %u\n", delay);
178 schedule_delayed_work(&monc->delayed_work, delay);
182 * Send subscribe request for mdsmap and/or osdmap.
184 static void __send_subscribe(struct ceph_mon_client *monc)
186 dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n",
187 (unsigned)monc->sub_sent, __sub_expired(monc),
188 monc->want_next_osdmap);
189 if ((__sub_expired(monc) && !monc->sub_sent) ||
190 monc->want_next_osdmap == 1) {
191 struct ceph_msg *msg;
192 struct ceph_mon_subscribe_item *i;
195 msg = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS);
199 p = msg->front.iov_base;
200 end = p + msg->front.iov_len;
202 dout("__send_subscribe to 'mdsmap' %u+\n",
203 (unsigned)monc->have_mdsmap);
204 if (monc->want_next_osdmap) {
205 dout("__send_subscribe to 'osdmap' %u\n",
206 (unsigned)monc->have_osdmap);
207 ceph_encode_32(&p, 3);
208 ceph_encode_string(&p, end, "osdmap", 6);
210 i->have = cpu_to_le64(monc->have_osdmap);
213 monc->want_next_osdmap = 2; /* requested */
215 ceph_encode_32(&p, 2);
217 ceph_encode_string(&p, end, "mdsmap", 6);
219 i->have = cpu_to_le64(monc->have_mdsmap);
222 ceph_encode_string(&p, end, "monmap", 6);
228 msg->front.iov_len = p - msg->front.iov_base;
229 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
230 ceph_con_send(monc->con, msg);
232 monc->sub_sent = jiffies | 1; /* never 0 */
236 static void handle_subscribe_ack(struct ceph_mon_client *monc,
237 struct ceph_msg *msg)
240 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
242 if (msg->front.iov_len < sizeof(*h))
244 seconds = le32_to_cpu(h->duration);
246 mutex_lock(&monc->mutex);
248 pr_info("mon%d %s session established\n",
249 monc->cur_mon, pr_addr(&monc->con->peer_addr.in_addr));
250 monc->hunting = false;
252 dout("handle_subscribe_ack after %d seconds\n", seconds);
253 monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1;
255 mutex_unlock(&monc->mutex);
258 pr_err("got corrupt subscribe-ack msg\n");
263 * Keep track of which maps we have
265 int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got)
267 mutex_lock(&monc->mutex);
268 monc->have_mdsmap = got;
269 mutex_unlock(&monc->mutex);
273 int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got)
275 mutex_lock(&monc->mutex);
276 monc->have_osdmap = got;
277 monc->want_next_osdmap = 0;
278 mutex_unlock(&monc->mutex);
283 * Register interest in the next osdmap
285 void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
287 dout("request_next_osdmap have %u\n", monc->have_osdmap);
288 mutex_lock(&monc->mutex);
289 if (!monc->want_next_osdmap)
290 monc->want_next_osdmap = 1;
291 if (monc->want_next_osdmap < 2)
292 __send_subscribe(monc);
293 mutex_unlock(&monc->mutex);
299 int ceph_monc_open_session(struct ceph_mon_client *monc)
302 monc->con = kmalloc(sizeof(*monc->con), GFP_KERNEL);
305 ceph_con_init(monc->client->msgr, monc->con);
306 monc->con->private = monc;
307 monc->con->ops = &mon_con_ops;
310 mutex_lock(&monc->mutex);
311 __open_session(monc);
312 __schedule_delayed(monc);
313 mutex_unlock(&monc->mutex);
318 * The monitor responds with mount ack indicate mount success. The
319 * included client ticket allows the client to talk to MDSs and OSDs.
321 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
322 struct ceph_msg *msg)
324 struct ceph_client *client = monc->client;
325 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
328 mutex_lock(&monc->mutex);
330 dout("handle_monmap\n");
331 p = msg->front.iov_base;
332 end = p + msg->front.iov_len;
334 monmap = ceph_monmap_decode(p, end);
335 if (IS_ERR(monmap)) {
336 pr_err("problem decoding monmap, %d\n",
337 (int)PTR_ERR(monmap));
341 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
346 client->monc.monmap = monmap;
350 mutex_unlock(&monc->mutex);
351 wake_up(&client->auth_wq);
357 static struct ceph_mon_generic_request *__lookup_generic_req(
358 struct ceph_mon_client *monc, u64 tid)
360 struct ceph_mon_generic_request *req;
361 struct rb_node *n = monc->generic_request_tree.rb_node;
364 req = rb_entry(n, struct ceph_mon_generic_request, node);
367 else if (tid > req->tid)
375 static void __insert_generic_request(struct ceph_mon_client *monc,
376 struct ceph_mon_generic_request *new)
378 struct rb_node **p = &monc->generic_request_tree.rb_node;
379 struct rb_node *parent = NULL;
380 struct ceph_mon_generic_request *req = NULL;
384 req = rb_entry(parent, struct ceph_mon_generic_request, node);
385 if (new->tid < req->tid)
387 else if (new->tid > req->tid)
393 rb_link_node(&new->node, parent, p);
394 rb_insert_color(&new->node, &monc->generic_request_tree);
397 static void release_generic_request(struct kref *kref)
399 struct ceph_mon_generic_request *req =
400 container_of(kref, struct ceph_mon_generic_request, kref);
403 ceph_msg_put(req->reply);
405 ceph_msg_put(req->request);
408 static void put_generic_request(struct ceph_mon_generic_request *req)
410 kref_put(&req->kref, release_generic_request);
413 static void get_generic_request(struct ceph_mon_generic_request *req)
415 kref_get(&req->kref);
418 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
419 struct ceph_msg_header *hdr,
422 struct ceph_mon_client *monc = con->private;
423 struct ceph_mon_generic_request *req;
424 u64 tid = le64_to_cpu(hdr->tid);
427 mutex_lock(&monc->mutex);
428 req = __lookup_generic_req(monc, tid);
430 dout("get_generic_reply %lld dne\n", tid);
434 dout("get_generic_reply %lld got %p\n", tid, req->reply);
435 m = ceph_msg_get(req->reply);
437 * we don't need to track the connection reading into
438 * this reply because we only have one open connection
442 mutex_unlock(&monc->mutex);
446 static void handle_statfs_reply(struct ceph_mon_client *monc,
447 struct ceph_msg *msg)
449 struct ceph_mon_generic_request *req;
450 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
451 u64 tid = le64_to_cpu(msg->hdr.tid);
453 if (msg->front.iov_len != sizeof(*reply))
455 dout("handle_statfs_reply %p tid %llu\n", msg, tid);
457 mutex_lock(&monc->mutex);
458 req = __lookup_generic_req(monc, tid);
460 *(struct ceph_statfs *)req->buf = reply->st;
462 get_generic_request(req);
464 mutex_unlock(&monc->mutex);
466 complete(&req->completion);
467 put_generic_request(req);
472 pr_err("corrupt generic reply, no tid\n");
477 * Do a synchronous statfs().
479 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
481 struct ceph_mon_generic_request *req;
482 struct ceph_mon_statfs *h;
485 req = kzalloc(sizeof(*req), GFP_NOFS);
489 kref_init(&req->kref);
491 init_completion(&req->completion);
494 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS);
497 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS);
501 /* fill out request */
502 h = req->request->front.iov_base;
503 h->monhdr.have_version = 0;
504 h->monhdr.session_mon = cpu_to_le16(-1);
505 h->monhdr.session_mon_tid = 0;
506 h->fsid = monc->monmap->fsid;
508 /* register request */
509 mutex_lock(&monc->mutex);
510 req->tid = ++monc->last_tid;
511 req->request->hdr.tid = cpu_to_le64(req->tid);
512 __insert_generic_request(monc, req);
513 monc->num_generic_requests++;
514 mutex_unlock(&monc->mutex);
516 /* send request and wait */
517 ceph_con_send(monc->con, ceph_msg_get(req->request));
518 err = wait_for_completion_interruptible(&req->completion);
520 mutex_lock(&monc->mutex);
521 rb_erase(&req->node, &monc->generic_request_tree);
522 monc->num_generic_requests--;
523 mutex_unlock(&monc->mutex);
529 kref_put(&req->kref, release_generic_request);
534 * Resend pending statfs requests.
536 static void __resend_generic_request(struct ceph_mon_client *monc)
538 struct ceph_mon_generic_request *req;
541 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
542 req = rb_entry(p, struct ceph_mon_generic_request, node);
543 ceph_con_revoke(monc->con, req->request);
544 ceph_con_send(monc->con, ceph_msg_get(req->request));
549 * Delayed work. If we haven't mounted yet, retry. Otherwise,
550 * renew/retry subscription as needed (in case it is timing out, or we
551 * got an ENOMEM). And keep the monitor connection alive.
553 static void delayed_work(struct work_struct *work)
555 struct ceph_mon_client *monc =
556 container_of(work, struct ceph_mon_client, delayed_work.work);
558 dout("monc delayed_work\n");
559 mutex_lock(&monc->mutex);
561 __close_session(monc);
562 __open_session(monc); /* continue hunting */
564 ceph_con_keepalive(monc->con);
566 __validate_auth(monc);
568 if (monc->auth->ops->is_authenticated(monc->auth))
569 __send_subscribe(monc);
571 __schedule_delayed(monc);
572 mutex_unlock(&monc->mutex);
576 * On startup, we build a temporary monmap populated with the IPs
577 * provided by mount(2).
579 static int build_initial_monmap(struct ceph_mon_client *monc)
581 struct ceph_mount_args *args = monc->client->mount_args;
582 struct ceph_entity_addr *mon_addr = args->mon_addr;
583 int num_mon = args->num_mon;
586 /* build initial monmap */
587 monc->monmap = kzalloc(sizeof(*monc->monmap) +
588 num_mon*sizeof(monc->monmap->mon_inst[0]),
592 for (i = 0; i < num_mon; i++) {
593 monc->monmap->mon_inst[i].addr = mon_addr[i];
594 monc->monmap->mon_inst[i].addr.nonce = 0;
595 monc->monmap->mon_inst[i].name.type =
596 CEPH_ENTITY_TYPE_MON;
597 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
599 monc->monmap->num_mon = num_mon;
600 monc->have_fsid = false;
602 /* release addr memory */
603 kfree(args->mon_addr);
604 args->mon_addr = NULL;
609 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
614 memset(monc, 0, sizeof(*monc));
617 mutex_init(&monc->mutex);
619 err = build_initial_monmap(monc);
626 monc->auth = ceph_auth_init(cl->mount_args->name,
627 cl->mount_args->secret);
628 if (IS_ERR(monc->auth))
629 return PTR_ERR(monc->auth);
630 monc->auth->want_keys =
631 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
632 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
636 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
637 sizeof(struct ceph_mon_subscribe_ack),
639 if (!monc->m_subscribe_ack)
642 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS);
643 if (!monc->m_auth_reply)
644 goto out_subscribe_ack;
646 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS);
647 monc->pending_auth = 0;
652 monc->hunting = true;
653 monc->sub_renew_after = jiffies;
656 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
657 monc->generic_request_tree = RB_ROOT;
658 monc->num_generic_requests = 0;
661 monc->have_mdsmap = 0;
662 monc->have_osdmap = 0;
663 monc->want_next_osdmap = 1;
667 ceph_msg_put(monc->m_auth_reply);
669 ceph_msg_put(monc->m_subscribe_ack);
676 void ceph_monc_stop(struct ceph_mon_client *monc)
679 cancel_delayed_work_sync(&monc->delayed_work);
681 mutex_lock(&monc->mutex);
682 __close_session(monc);
684 monc->con->private = NULL;
685 monc->con->ops->put(monc->con);
688 mutex_unlock(&monc->mutex);
690 ceph_auth_destroy(monc->auth);
692 ceph_msg_put(monc->m_auth);
693 ceph_msg_put(monc->m_auth_reply);
694 ceph_msg_put(monc->m_subscribe_ack);
699 static void handle_auth_reply(struct ceph_mon_client *monc,
700 struct ceph_msg *msg)
704 mutex_lock(&monc->mutex);
705 monc->pending_auth = 0;
706 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
708 monc->m_auth->front.iov_base,
709 monc->m_auth->front_max);
711 monc->client->auth_err = ret;
712 wake_up(&monc->client->auth_wq);
713 } else if (ret > 0) {
714 __send_prepared_auth_request(monc, ret);
715 } else if (monc->auth->ops->is_authenticated(monc->auth)) {
716 dout("authenticated, starting session\n");
718 monc->client->msgr->inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
719 monc->client->msgr->inst.name.num = monc->auth->global_id;
721 __send_subscribe(monc);
722 __resend_generic_request(monc);
724 mutex_unlock(&monc->mutex);
727 static int __validate_auth(struct ceph_mon_client *monc)
731 if (monc->pending_auth)
734 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
735 monc->m_auth->front_max);
737 return ret; /* either an error, or no need to authenticate */
738 __send_prepared_auth_request(monc, ret);
742 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
746 mutex_lock(&monc->mutex);
747 ret = __validate_auth(monc);
748 mutex_unlock(&monc->mutex);
753 * handle incoming message
755 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
757 struct ceph_mon_client *monc = con->private;
758 int type = le16_to_cpu(msg->hdr.type);
764 case CEPH_MSG_AUTH_REPLY:
765 handle_auth_reply(monc, msg);
768 case CEPH_MSG_MON_SUBSCRIBE_ACK:
769 handle_subscribe_ack(monc, msg);
772 case CEPH_MSG_STATFS_REPLY:
773 handle_statfs_reply(monc, msg);
776 case CEPH_MSG_MON_MAP:
777 ceph_monc_handle_map(monc, msg);
780 case CEPH_MSG_MDS_MAP:
781 ceph_mdsc_handle_map(&monc->client->mdsc, msg);
784 case CEPH_MSG_OSD_MAP:
785 ceph_osdc_handle_map(&monc->client->osdc, msg);
789 pr_err("received unknown message type %d %s\n", type,
790 ceph_msg_type_name(type));
796 * Allocate memory for incoming message
798 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
799 struct ceph_msg_header *hdr,
802 struct ceph_mon_client *monc = con->private;
803 int type = le16_to_cpu(hdr->type);
804 int front_len = le32_to_cpu(hdr->front_len);
805 struct ceph_msg *m = NULL;
810 case CEPH_MSG_MON_SUBSCRIBE_ACK:
811 m = ceph_msg_get(monc->m_subscribe_ack);
813 case CEPH_MSG_STATFS_REPLY:
814 return get_generic_reply(con, hdr, skip);
815 case CEPH_MSG_AUTH_REPLY:
816 m = ceph_msg_get(monc->m_auth_reply);
818 case CEPH_MSG_MON_MAP:
819 case CEPH_MSG_MDS_MAP:
820 case CEPH_MSG_OSD_MAP:
821 m = ceph_msg_new(type, front_len, GFP_NOFS);
826 pr_info("alloc_msg unknown type %d\n", type);
833 * If the monitor connection resets, pick a new monitor and resubmit
834 * any pending requests.
836 static void mon_fault(struct ceph_connection *con)
838 struct ceph_mon_client *monc = con->private;
844 mutex_lock(&monc->mutex);
848 if (monc->con && !monc->hunting)
849 pr_info("mon%d %s session lost, "
850 "hunting for new mon\n", monc->cur_mon,
851 pr_addr(&monc->con->peer_addr.in_addr));
853 __close_session(monc);
854 if (!monc->hunting) {
856 monc->hunting = true;
857 __open_session(monc);
859 /* already hunting, let's wait a bit */
860 __schedule_delayed(monc);
863 mutex_unlock(&monc->mutex);
866 static const struct ceph_connection_operations mon_con_ops = {
869 .dispatch = dispatch,
871 .alloc_msg = mon_alloc_msg,