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 = monc->m_subscribe;
192 struct ceph_mon_subscribe_item *i;
195 p = msg->front.iov_base;
196 end = p + msg->front_max;
198 dout("__send_subscribe to 'mdsmap' %u+\n",
199 (unsigned)monc->have_mdsmap);
200 if (monc->want_next_osdmap) {
201 dout("__send_subscribe to 'osdmap' %u\n",
202 (unsigned)monc->have_osdmap);
203 ceph_encode_32(&p, 3);
204 ceph_encode_string(&p, end, "osdmap", 6);
206 i->have = cpu_to_le64(monc->have_osdmap);
209 monc->want_next_osdmap = 2; /* requested */
211 ceph_encode_32(&p, 2);
213 ceph_encode_string(&p, end, "mdsmap", 6);
215 i->have = cpu_to_le64(monc->have_mdsmap);
218 ceph_encode_string(&p, end, "monmap", 6);
224 msg->front.iov_len = p - msg->front.iov_base;
225 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
226 ceph_con_revoke(monc->con, msg);
227 ceph_con_send(monc->con, ceph_msg_get(msg));
229 monc->sub_sent = jiffies | 1; /* never 0 */
233 static void handle_subscribe_ack(struct ceph_mon_client *monc,
234 struct ceph_msg *msg)
237 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
239 if (msg->front.iov_len < sizeof(*h))
241 seconds = le32_to_cpu(h->duration);
243 mutex_lock(&monc->mutex);
245 pr_info("mon%d %s session established\n",
246 monc->cur_mon, pr_addr(&monc->con->peer_addr.in_addr));
247 monc->hunting = false;
249 dout("handle_subscribe_ack after %d seconds\n", seconds);
250 monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1;
252 mutex_unlock(&monc->mutex);
255 pr_err("got corrupt subscribe-ack msg\n");
260 * Keep track of which maps we have
262 int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got)
264 mutex_lock(&monc->mutex);
265 monc->have_mdsmap = got;
266 mutex_unlock(&monc->mutex);
270 int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got)
272 mutex_lock(&monc->mutex);
273 monc->have_osdmap = got;
274 monc->want_next_osdmap = 0;
275 mutex_unlock(&monc->mutex);
280 * Register interest in the next osdmap
282 void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
284 dout("request_next_osdmap have %u\n", monc->have_osdmap);
285 mutex_lock(&monc->mutex);
286 if (!monc->want_next_osdmap)
287 monc->want_next_osdmap = 1;
288 if (monc->want_next_osdmap < 2)
289 __send_subscribe(monc);
290 mutex_unlock(&monc->mutex);
296 int ceph_monc_open_session(struct ceph_mon_client *monc)
299 monc->con = kmalloc(sizeof(*monc->con), GFP_KERNEL);
302 ceph_con_init(monc->client->msgr, monc->con);
303 monc->con->private = monc;
304 monc->con->ops = &mon_con_ops;
307 mutex_lock(&monc->mutex);
308 __open_session(monc);
309 __schedule_delayed(monc);
310 mutex_unlock(&monc->mutex);
315 * The monitor responds with mount ack indicate mount success. The
316 * included client ticket allows the client to talk to MDSs and OSDs.
318 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
319 struct ceph_msg *msg)
321 struct ceph_client *client = monc->client;
322 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
325 mutex_lock(&monc->mutex);
327 dout("handle_monmap\n");
328 p = msg->front.iov_base;
329 end = p + msg->front.iov_len;
331 monmap = ceph_monmap_decode(p, end);
332 if (IS_ERR(monmap)) {
333 pr_err("problem decoding monmap, %d\n",
334 (int)PTR_ERR(monmap));
338 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
343 client->monc.monmap = monmap;
347 mutex_unlock(&monc->mutex);
348 wake_up_all(&client->auth_wq);
354 static struct ceph_mon_generic_request *__lookup_generic_req(
355 struct ceph_mon_client *monc, u64 tid)
357 struct ceph_mon_generic_request *req;
358 struct rb_node *n = monc->generic_request_tree.rb_node;
361 req = rb_entry(n, struct ceph_mon_generic_request, node);
364 else if (tid > req->tid)
372 static void __insert_generic_request(struct ceph_mon_client *monc,
373 struct ceph_mon_generic_request *new)
375 struct rb_node **p = &monc->generic_request_tree.rb_node;
376 struct rb_node *parent = NULL;
377 struct ceph_mon_generic_request *req = NULL;
381 req = rb_entry(parent, struct ceph_mon_generic_request, node);
382 if (new->tid < req->tid)
384 else if (new->tid > req->tid)
390 rb_link_node(&new->node, parent, p);
391 rb_insert_color(&new->node, &monc->generic_request_tree);
394 static void release_generic_request(struct kref *kref)
396 struct ceph_mon_generic_request *req =
397 container_of(kref, struct ceph_mon_generic_request, kref);
400 ceph_msg_put(req->reply);
402 ceph_msg_put(req->request);
407 static void put_generic_request(struct ceph_mon_generic_request *req)
409 kref_put(&req->kref, release_generic_request);
412 static void get_generic_request(struct ceph_mon_generic_request *req)
414 kref_get(&req->kref);
417 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
418 struct ceph_msg_header *hdr,
421 struct ceph_mon_client *monc = con->private;
422 struct ceph_mon_generic_request *req;
423 u64 tid = le64_to_cpu(hdr->tid);
426 mutex_lock(&monc->mutex);
427 req = __lookup_generic_req(monc, tid);
429 dout("get_generic_reply %lld dne\n", tid);
433 dout("get_generic_reply %lld got %p\n", tid, req->reply);
434 m = ceph_msg_get(req->reply);
436 * we don't need to track the connection reading into
437 * this reply because we only have one open connection
441 mutex_unlock(&monc->mutex);
445 static void handle_statfs_reply(struct ceph_mon_client *monc,
446 struct ceph_msg *msg)
448 struct ceph_mon_generic_request *req;
449 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
450 u64 tid = le64_to_cpu(msg->hdr.tid);
452 if (msg->front.iov_len != sizeof(*reply))
454 dout("handle_statfs_reply %p tid %llu\n", msg, tid);
456 mutex_lock(&monc->mutex);
457 req = __lookup_generic_req(monc, tid);
459 *(struct ceph_statfs *)req->buf = reply->st;
461 get_generic_request(req);
463 mutex_unlock(&monc->mutex);
465 complete_all(&req->completion);
466 put_generic_request(req);
471 pr_err("corrupt generic reply, no tid\n");
476 * Do a synchronous statfs().
478 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
480 struct ceph_mon_generic_request *req;
481 struct ceph_mon_statfs *h;
484 req = kzalloc(sizeof(*req), GFP_NOFS);
488 kref_init(&req->kref);
490 init_completion(&req->completion);
493 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS);
496 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS);
500 /* fill out request */
501 h = req->request->front.iov_base;
502 h->monhdr.have_version = 0;
503 h->monhdr.session_mon = cpu_to_le16(-1);
504 h->monhdr.session_mon_tid = 0;
505 h->fsid = monc->monmap->fsid;
507 /* register request */
508 mutex_lock(&monc->mutex);
509 req->tid = ++monc->last_tid;
510 req->request->hdr.tid = cpu_to_le64(req->tid);
511 __insert_generic_request(monc, req);
512 monc->num_generic_requests++;
513 mutex_unlock(&monc->mutex);
515 /* send request and wait */
516 ceph_con_send(monc->con, ceph_msg_get(req->request));
517 err = wait_for_completion_interruptible(&req->completion);
519 mutex_lock(&monc->mutex);
520 rb_erase(&req->node, &monc->generic_request_tree);
521 monc->num_generic_requests--;
522 mutex_unlock(&monc->mutex);
528 kref_put(&req->kref, release_generic_request);
533 * Resend pending statfs requests.
535 static void __resend_generic_request(struct ceph_mon_client *monc)
537 struct ceph_mon_generic_request *req;
540 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
541 req = rb_entry(p, struct ceph_mon_generic_request, node);
542 ceph_con_revoke(monc->con, req->request);
543 ceph_con_send(monc->con, ceph_msg_get(req->request));
548 * Delayed work. If we haven't mounted yet, retry. Otherwise,
549 * renew/retry subscription as needed (in case it is timing out, or we
550 * got an ENOMEM). And keep the monitor connection alive.
552 static void delayed_work(struct work_struct *work)
554 struct ceph_mon_client *monc =
555 container_of(work, struct ceph_mon_client, delayed_work.work);
557 dout("monc delayed_work\n");
558 mutex_lock(&monc->mutex);
560 __close_session(monc);
561 __open_session(monc); /* continue hunting */
563 ceph_con_keepalive(monc->con);
565 __validate_auth(monc);
567 if (monc->auth->ops->is_authenticated(monc->auth))
568 __send_subscribe(monc);
570 __schedule_delayed(monc);
571 mutex_unlock(&monc->mutex);
575 * On startup, we build a temporary monmap populated with the IPs
576 * provided by mount(2).
578 static int build_initial_monmap(struct ceph_mon_client *monc)
580 struct ceph_mount_args *args = monc->client->mount_args;
581 struct ceph_entity_addr *mon_addr = args->mon_addr;
582 int num_mon = args->num_mon;
585 /* build initial monmap */
586 monc->monmap = kzalloc(sizeof(*monc->monmap) +
587 num_mon*sizeof(monc->monmap->mon_inst[0]),
591 for (i = 0; i < num_mon; i++) {
592 monc->monmap->mon_inst[i].addr = mon_addr[i];
593 monc->monmap->mon_inst[i].addr.nonce = 0;
594 monc->monmap->mon_inst[i].name.type =
595 CEPH_ENTITY_TYPE_MON;
596 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
598 monc->monmap->num_mon = num_mon;
599 monc->have_fsid = false;
601 /* release addr memory */
602 kfree(args->mon_addr);
603 args->mon_addr = NULL;
608 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
613 memset(monc, 0, sizeof(*monc));
616 mutex_init(&monc->mutex);
618 err = build_initial_monmap(monc);
625 monc->auth = ceph_auth_init(cl->mount_args->name,
626 cl->mount_args->secret);
627 if (IS_ERR(monc->auth))
628 return PTR_ERR(monc->auth);
629 monc->auth->want_keys =
630 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
631 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
635 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
636 sizeof(struct ceph_mon_subscribe_ack),
638 if (!monc->m_subscribe_ack)
641 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS);
642 if (!monc->m_subscribe)
643 goto out_subscribe_ack;
645 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS);
646 if (!monc->m_auth_reply)
649 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS);
650 monc->pending_auth = 0;
655 monc->hunting = true;
656 monc->sub_renew_after = jiffies;
659 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
660 monc->generic_request_tree = RB_ROOT;
661 monc->num_generic_requests = 0;
664 monc->have_mdsmap = 0;
665 monc->have_osdmap = 0;
666 monc->want_next_osdmap = 1;
670 ceph_msg_put(monc->m_auth_reply);
672 ceph_msg_put(monc->m_subscribe);
674 ceph_msg_put(monc->m_subscribe_ack);
681 void ceph_monc_stop(struct ceph_mon_client *monc)
684 cancel_delayed_work_sync(&monc->delayed_work);
686 mutex_lock(&monc->mutex);
687 __close_session(monc);
689 monc->con->private = NULL;
690 monc->con->ops->put(monc->con);
693 mutex_unlock(&monc->mutex);
695 ceph_auth_destroy(monc->auth);
697 ceph_msg_put(monc->m_auth);
698 ceph_msg_put(monc->m_auth_reply);
699 ceph_msg_put(monc->m_subscribe);
700 ceph_msg_put(monc->m_subscribe_ack);
705 static void handle_auth_reply(struct ceph_mon_client *monc,
706 struct ceph_msg *msg)
711 mutex_lock(&monc->mutex);
713 was_auth = monc->auth->ops->is_authenticated(monc->auth);
714 monc->pending_auth = 0;
715 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
717 monc->m_auth->front.iov_base,
718 monc->m_auth->front_max);
720 monc->client->auth_err = ret;
721 wake_up_all(&monc->client->auth_wq);
722 } else if (ret > 0) {
723 __send_prepared_auth_request(monc, ret);
724 } else if (!was_auth && monc->auth->ops->is_authenticated(monc->auth)) {
725 dout("authenticated, starting session\n");
727 monc->client->msgr->inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
728 monc->client->msgr->inst.name.num =
729 cpu_to_le64(monc->auth->global_id);
731 __send_subscribe(monc);
732 __resend_generic_request(monc);
734 mutex_unlock(&monc->mutex);
737 static int __validate_auth(struct ceph_mon_client *monc)
741 if (monc->pending_auth)
744 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
745 monc->m_auth->front_max);
747 return ret; /* either an error, or no need to authenticate */
748 __send_prepared_auth_request(monc, ret);
752 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
756 mutex_lock(&monc->mutex);
757 ret = __validate_auth(monc);
758 mutex_unlock(&monc->mutex);
763 * handle incoming message
765 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
767 struct ceph_mon_client *monc = con->private;
768 int type = le16_to_cpu(msg->hdr.type);
774 case CEPH_MSG_AUTH_REPLY:
775 handle_auth_reply(monc, msg);
778 case CEPH_MSG_MON_SUBSCRIBE_ACK:
779 handle_subscribe_ack(monc, msg);
782 case CEPH_MSG_STATFS_REPLY:
783 handle_statfs_reply(monc, msg);
786 case CEPH_MSG_MON_MAP:
787 ceph_monc_handle_map(monc, msg);
790 case CEPH_MSG_MDS_MAP:
791 ceph_mdsc_handle_map(&monc->client->mdsc, msg);
794 case CEPH_MSG_OSD_MAP:
795 ceph_osdc_handle_map(&monc->client->osdc, msg);
799 pr_err("received unknown message type %d %s\n", type,
800 ceph_msg_type_name(type));
806 * Allocate memory for incoming message
808 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
809 struct ceph_msg_header *hdr,
812 struct ceph_mon_client *monc = con->private;
813 int type = le16_to_cpu(hdr->type);
814 int front_len = le32_to_cpu(hdr->front_len);
815 struct ceph_msg *m = NULL;
820 case CEPH_MSG_MON_SUBSCRIBE_ACK:
821 m = ceph_msg_get(monc->m_subscribe_ack);
823 case CEPH_MSG_STATFS_REPLY:
824 return get_generic_reply(con, hdr, skip);
825 case CEPH_MSG_AUTH_REPLY:
826 m = ceph_msg_get(monc->m_auth_reply);
828 case CEPH_MSG_MON_MAP:
829 case CEPH_MSG_MDS_MAP:
830 case CEPH_MSG_OSD_MAP:
831 m = ceph_msg_new(type, front_len, GFP_NOFS);
836 pr_info("alloc_msg unknown type %d\n", type);
843 * If the monitor connection resets, pick a new monitor and resubmit
844 * any pending requests.
846 static void mon_fault(struct ceph_connection *con)
848 struct ceph_mon_client *monc = con->private;
854 mutex_lock(&monc->mutex);
858 if (monc->con && !monc->hunting)
859 pr_info("mon%d %s session lost, "
860 "hunting for new mon\n", monc->cur_mon,
861 pr_addr(&monc->con->peer_addr.in_addr));
863 __close_session(monc);
864 if (!monc->hunting) {
866 monc->hunting = true;
867 __open_session(monc);
869 /* already hunting, let's wait a bit */
870 __schedule_delayed(monc);
873 mutex_unlock(&monc->mutex);
876 static const struct ceph_connection_operations mon_con_ops = {
879 .dispatch = dispatch,
881 .alloc_msg = mon_alloc_msg,