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 const static 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_msg_get(monc->m_auth); /* keep our ref */
108 ceph_con_send(monc->con, monc->m_auth);
112 * Close monitor session, if any.
114 static void __close_session(struct ceph_mon_client *monc)
117 dout("__close_session closing mon%d\n", monc->cur_mon);
118 ceph_con_revoke(monc->con, monc->m_auth);
119 ceph_con_close(monc->con);
121 monc->pending_auth = 0;
122 ceph_auth_reset(monc->auth);
127 * Open a session with a (new) monitor.
129 static int __open_session(struct ceph_mon_client *monc)
134 if (monc->cur_mon < 0) {
135 get_random_bytes(&r, 1);
136 monc->cur_mon = r % monc->monmap->num_mon;
137 dout("open_session num=%d r=%d -> mon%d\n",
138 monc->monmap->num_mon, r, monc->cur_mon);
140 monc->sub_renew_after = jiffies; /* i.e., expired */
141 monc->want_next_osdmap = !!monc->want_next_osdmap;
143 dout("open_session mon%d opening\n", monc->cur_mon);
144 monc->con->peer_name.type = CEPH_ENTITY_TYPE_MON;
145 monc->con->peer_name.num = cpu_to_le64(monc->cur_mon);
146 ceph_con_open(monc->con,
147 &monc->monmap->mon_inst[monc->cur_mon].addr);
149 /* initiatiate authentication handshake */
150 ret = ceph_auth_build_hello(monc->auth,
151 monc->m_auth->front.iov_base,
152 monc->m_auth->front_max);
153 __send_prepared_auth_request(monc, ret);
155 dout("open_session mon%d already open\n", monc->cur_mon);
160 static bool __sub_expired(struct ceph_mon_client *monc)
162 return time_after_eq(jiffies, monc->sub_renew_after);
166 * Reschedule delayed work timer.
168 static void __schedule_delayed(struct ceph_mon_client *monc)
172 if (monc->cur_mon < 0 || __sub_expired(monc))
176 dout("__schedule_delayed after %u\n", delay);
177 schedule_delayed_work(&monc->delayed_work, delay);
181 * Send subscribe request for mdsmap and/or osdmap.
183 static void __send_subscribe(struct ceph_mon_client *monc)
185 dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n",
186 (unsigned)monc->sub_sent, __sub_expired(monc),
187 monc->want_next_osdmap);
188 if ((__sub_expired(monc) && !monc->sub_sent) ||
189 monc->want_next_osdmap == 1) {
190 struct ceph_msg *msg;
191 struct ceph_mon_subscribe_item *i;
194 msg = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96);
198 p = msg->front.iov_base;
199 end = p + msg->front.iov_len;
201 dout("__send_subscribe to 'mdsmap' %u+\n",
202 (unsigned)monc->have_mdsmap);
203 if (monc->want_next_osdmap) {
204 dout("__send_subscribe to 'osdmap' %u\n",
205 (unsigned)monc->have_osdmap);
206 ceph_encode_32(&p, 3);
207 ceph_encode_string(&p, end, "osdmap", 6);
209 i->have = cpu_to_le64(monc->have_osdmap);
212 monc->want_next_osdmap = 2; /* requested */
214 ceph_encode_32(&p, 2);
216 ceph_encode_string(&p, end, "mdsmap", 6);
218 i->have = cpu_to_le64(monc->have_mdsmap);
221 ceph_encode_string(&p, end, "monmap", 6);
227 msg->front.iov_len = p - msg->front.iov_base;
228 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
229 ceph_con_send(monc->con, msg);
231 monc->sub_sent = jiffies | 1; /* never 0 */
235 static void handle_subscribe_ack(struct ceph_mon_client *monc,
236 struct ceph_msg *msg)
239 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
241 if (msg->front.iov_len < sizeof(*h))
243 seconds = le32_to_cpu(h->duration);
245 mutex_lock(&monc->mutex);
247 pr_info("mon%d %s session established\n",
248 monc->cur_mon, pr_addr(&monc->con->peer_addr.in_addr));
249 monc->hunting = false;
251 dout("handle_subscribe_ack after %d seconds\n", seconds);
252 monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1;
254 mutex_unlock(&monc->mutex);
257 pr_err("got corrupt subscribe-ack msg\n");
262 * Keep track of which maps we have
264 int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got)
266 mutex_lock(&monc->mutex);
267 monc->have_mdsmap = got;
268 mutex_unlock(&monc->mutex);
272 int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got)
274 mutex_lock(&monc->mutex);
275 monc->have_osdmap = got;
276 monc->want_next_osdmap = 0;
277 mutex_unlock(&monc->mutex);
282 * Register interest in the next osdmap
284 void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
286 dout("request_next_osdmap have %u\n", monc->have_osdmap);
287 mutex_lock(&monc->mutex);
288 if (!monc->want_next_osdmap)
289 monc->want_next_osdmap = 1;
290 if (monc->want_next_osdmap < 2)
291 __send_subscribe(monc);
292 mutex_unlock(&monc->mutex);
298 int ceph_monc_open_session(struct ceph_mon_client *monc)
301 monc->con = kmalloc(sizeof(*monc->con), GFP_KERNEL);
304 ceph_con_init(monc->client->msgr, monc->con);
305 monc->con->private = monc;
306 monc->con->ops = &mon_con_ops;
309 mutex_lock(&monc->mutex);
310 __open_session(monc);
311 __schedule_delayed(monc);
312 mutex_unlock(&monc->mutex);
317 * The monitor responds with mount ack indicate mount success. The
318 * included client ticket allows the client to talk to MDSs and OSDs.
320 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
321 struct ceph_msg *msg)
323 struct ceph_client *client = monc->client;
324 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
327 mutex_lock(&monc->mutex);
329 dout("handle_monmap\n");
330 p = msg->front.iov_base;
331 end = p + msg->front.iov_len;
333 monmap = ceph_monmap_decode(p, end);
334 if (IS_ERR(monmap)) {
335 pr_err("problem decoding monmap, %d\n",
336 (int)PTR_ERR(monmap));
340 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
345 client->monc.monmap = monmap;
349 mutex_unlock(&monc->mutex);
350 wake_up(&client->auth_wq);
356 static struct ceph_mon_generic_request *__lookup_generic_req(
357 struct ceph_mon_client *monc, u64 tid)
359 struct ceph_mon_generic_request *req;
360 struct rb_node *n = monc->generic_request_tree.rb_node;
363 req = rb_entry(n, struct ceph_mon_generic_request, node);
366 else if (tid > req->tid)
374 static void __insert_generic_request(struct ceph_mon_client *monc,
375 struct ceph_mon_generic_request *new)
377 struct rb_node **p = &monc->generic_request_tree.rb_node;
378 struct rb_node *parent = NULL;
379 struct ceph_mon_generic_request *req = NULL;
383 req = rb_entry(parent, struct ceph_mon_generic_request, node);
384 if (new->tid < req->tid)
386 else if (new->tid > req->tid)
392 rb_link_node(&new->node, parent, p);
393 rb_insert_color(&new->node, &monc->generic_request_tree);
396 static void release_generic_request(struct kref *kref)
398 struct ceph_mon_generic_request *req =
399 container_of(kref, struct ceph_mon_generic_request, kref);
402 ceph_msg_put(req->reply);
404 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(&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 = kmalloc(sizeof(*req), GFP_NOFS);
488 memset(req, 0, sizeof(*req));
489 kref_init(&req->kref);
491 init_completion(&req->completion);
494 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h));
497 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024);
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_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));
637 if (!monc->m_subscribe_ack)
640 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096);
641 if (!monc->m_auth_reply)
642 goto out_subscribe_ack;
644 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096);
645 monc->pending_auth = 0;
650 monc->hunting = true;
651 monc->sub_renew_after = jiffies;
654 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
655 monc->generic_request_tree = RB_ROOT;
656 monc->num_generic_requests = 0;
659 monc->have_mdsmap = 0;
660 monc->have_osdmap = 0;
661 monc->want_next_osdmap = 1;
665 ceph_msg_put(monc->m_auth_reply);
667 ceph_msg_put(monc->m_subscribe_ack);
674 void ceph_monc_stop(struct ceph_mon_client *monc)
677 cancel_delayed_work_sync(&monc->delayed_work);
679 mutex_lock(&monc->mutex);
680 __close_session(monc);
682 monc->con->private = NULL;
683 monc->con->ops->put(monc->con);
686 mutex_unlock(&monc->mutex);
688 ceph_auth_destroy(monc->auth);
690 ceph_msg_put(monc->m_auth);
691 ceph_msg_put(monc->m_auth_reply);
692 ceph_msg_put(monc->m_subscribe_ack);
697 static void handle_auth_reply(struct ceph_mon_client *monc,
698 struct ceph_msg *msg)
702 mutex_lock(&monc->mutex);
703 monc->pending_auth = 0;
704 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
706 monc->m_auth->front.iov_base,
707 monc->m_auth->front_max);
709 monc->client->auth_err = ret;
710 wake_up(&monc->client->auth_wq);
711 } else if (ret > 0) {
712 __send_prepared_auth_request(monc, ret);
713 } else if (monc->auth->ops->is_authenticated(monc->auth)) {
714 dout("authenticated, starting session\n");
716 monc->client->msgr->inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
717 monc->client->msgr->inst.name.num = monc->auth->global_id;
719 __send_subscribe(monc);
720 __resend_generic_request(monc);
722 mutex_unlock(&monc->mutex);
725 static int __validate_auth(struct ceph_mon_client *monc)
729 if (monc->pending_auth)
732 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
733 monc->m_auth->front_max);
735 return ret; /* either an error, or no need to authenticate */
736 __send_prepared_auth_request(monc, ret);
740 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
744 mutex_lock(&monc->mutex);
745 ret = __validate_auth(monc);
746 mutex_unlock(&monc->mutex);
751 * handle incoming message
753 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
755 struct ceph_mon_client *monc = con->private;
756 int type = le16_to_cpu(msg->hdr.type);
762 case CEPH_MSG_AUTH_REPLY:
763 handle_auth_reply(monc, msg);
766 case CEPH_MSG_MON_SUBSCRIBE_ACK:
767 handle_subscribe_ack(monc, msg);
770 case CEPH_MSG_STATFS_REPLY:
771 handle_statfs_reply(monc, msg);
774 case CEPH_MSG_MON_MAP:
775 ceph_monc_handle_map(monc, msg);
778 case CEPH_MSG_MDS_MAP:
779 ceph_mdsc_handle_map(&monc->client->mdsc, msg);
782 case CEPH_MSG_OSD_MAP:
783 ceph_osdc_handle_map(&monc->client->osdc, msg);
787 pr_err("received unknown message type %d %s\n", type,
788 ceph_msg_type_name(type));
794 * Allocate memory for incoming message
796 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
797 struct ceph_msg_header *hdr,
800 struct ceph_mon_client *monc = con->private;
801 int type = le16_to_cpu(hdr->type);
802 int front_len = le32_to_cpu(hdr->front_len);
803 struct ceph_msg *m = NULL;
808 case CEPH_MSG_MON_SUBSCRIBE_ACK:
809 m = ceph_msg_get(monc->m_subscribe_ack);
811 case CEPH_MSG_STATFS_REPLY:
812 return get_generic_reply(con, hdr, skip);
813 case CEPH_MSG_AUTH_REPLY:
814 m = ceph_msg_get(monc->m_auth_reply);
816 case CEPH_MSG_MON_MAP:
817 case CEPH_MSG_MDS_MAP:
818 case CEPH_MSG_OSD_MAP:
819 m = ceph_msg_new(type, front_len);
824 pr_info("alloc_msg unknown type %d\n", type);
831 * If the monitor connection resets, pick a new monitor and resubmit
832 * any pending requests.
834 static void mon_fault(struct ceph_connection *con)
836 struct ceph_mon_client *monc = con->private;
842 mutex_lock(&monc->mutex);
846 if (monc->con && !monc->hunting)
847 pr_info("mon%d %s session lost, "
848 "hunting for new mon\n", monc->cur_mon,
849 pr_addr(&monc->con->peer_addr.in_addr));
851 __close_session(monc);
852 if (!monc->hunting) {
854 monc->hunting = true;
855 __open_session(monc);
857 /* already hunting, let's wait a bit */
858 __schedule_delayed(monc);
861 mutex_unlock(&monc->mutex);
864 const static struct ceph_connection_operations mon_con_ops = {
867 .dispatch = dispatch,
869 .alloc_msg = mon_alloc_msg,