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 * Open a session with a (new) monitor.
133 static int __open_session(struct ceph_mon_client *monc)
138 if (monc->cur_mon < 0) {
139 get_random_bytes(&r, 1);
140 monc->cur_mon = r % monc->monmap->num_mon;
141 dout("open_session num=%d r=%d -> mon%d\n",
142 monc->monmap->num_mon, r, monc->cur_mon);
144 monc->sub_renew_after = jiffies; /* i.e., expired */
145 monc->want_next_osdmap = !!monc->want_next_osdmap;
147 dout("open_session mon%d opening\n", monc->cur_mon);
148 ceph_con_open(&monc->con,
149 CEPH_ENTITY_TYPE_MON, monc->cur_mon,
150 &monc->monmap->mon_inst[monc->cur_mon].addr);
152 /* initiatiate authentication handshake */
153 ret = ceph_auth_build_hello(monc->auth,
154 monc->m_auth->front.iov_base,
155 monc->m_auth->front_alloc_len);
156 __send_prepared_auth_request(monc, ret);
158 dout("open_session mon%d already open\n", monc->cur_mon);
163 static bool __sub_expired(struct ceph_mon_client *monc)
165 return time_after_eq(jiffies, monc->sub_renew_after);
169 * Reschedule delayed work timer.
171 static void __schedule_delayed(struct ceph_mon_client *monc)
175 if (monc->cur_mon < 0 || __sub_expired(monc))
179 dout("__schedule_delayed after %u\n", delay);
180 schedule_delayed_work(&monc->delayed_work, delay);
184 * Send subscribe request for mdsmap and/or osdmap.
186 static void __send_subscribe(struct ceph_mon_client *monc)
188 dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n",
189 (unsigned int)monc->sub_sent, __sub_expired(monc),
190 monc->want_next_osdmap);
191 if ((__sub_expired(monc) && !monc->sub_sent) ||
192 monc->want_next_osdmap == 1) {
193 struct ceph_msg *msg = monc->m_subscribe;
194 struct ceph_mon_subscribe_item *i;
198 p = msg->front.iov_base;
199 end = p + msg->front_alloc_len;
201 num = 1 + !!monc->want_next_osdmap + !!monc->want_mdsmap;
202 ceph_encode_32(&p, num);
204 if (monc->want_next_osdmap) {
205 dout("__send_subscribe to 'osdmap' %u\n",
206 (unsigned int)monc->have_osdmap);
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 if (monc->want_mdsmap) {
215 dout("__send_subscribe to 'mdsmap' %u+\n",
216 (unsigned int)monc->have_mdsmap);
217 ceph_encode_string(&p, end, "mdsmap", 6);
219 i->have = cpu_to_le64(monc->have_mdsmap);
223 ceph_encode_string(&p, end, "monmap", 6);
229 msg->front.iov_len = p - msg->front.iov_base;
230 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
231 ceph_msg_revoke(msg);
232 ceph_con_send(&monc->con, ceph_msg_get(msg));
234 monc->sub_sent = jiffies | 1; /* never 0 */
238 static void handle_subscribe_ack(struct ceph_mon_client *monc,
239 struct ceph_msg *msg)
241 unsigned int seconds;
242 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
244 if (msg->front.iov_len < sizeof(*h))
246 seconds = le32_to_cpu(h->duration);
248 mutex_lock(&monc->mutex);
250 pr_info("mon%d %s session established\n",
252 ceph_pr_addr(&monc->con.peer_addr.in_addr));
253 monc->hunting = false;
255 dout("handle_subscribe_ack after %d seconds\n", seconds);
256 monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1;
258 mutex_unlock(&monc->mutex);
261 pr_err("got corrupt subscribe-ack msg\n");
266 * Keep track of which maps we have
268 int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got)
270 mutex_lock(&monc->mutex);
271 monc->have_mdsmap = got;
272 mutex_unlock(&monc->mutex);
275 EXPORT_SYMBOL(ceph_monc_got_mdsmap);
277 int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got)
279 mutex_lock(&monc->mutex);
280 monc->have_osdmap = got;
281 monc->want_next_osdmap = 0;
282 mutex_unlock(&monc->mutex);
287 * Register interest in the next osdmap
289 void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
291 dout("request_next_osdmap have %u\n", monc->have_osdmap);
292 mutex_lock(&monc->mutex);
293 if (!monc->want_next_osdmap)
294 monc->want_next_osdmap = 1;
295 if (monc->want_next_osdmap < 2)
296 __send_subscribe(monc);
297 mutex_unlock(&monc->mutex);
299 EXPORT_SYMBOL(ceph_monc_request_next_osdmap);
301 int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
302 unsigned long timeout)
304 unsigned long started = jiffies;
307 mutex_lock(&monc->mutex);
308 while (monc->have_osdmap < epoch) {
309 mutex_unlock(&monc->mutex);
311 if (timeout != 0 && time_after_eq(jiffies, started + timeout))
314 ret = wait_event_interruptible_timeout(monc->client->auth_wq,
315 monc->have_osdmap >= epoch, timeout);
319 mutex_lock(&monc->mutex);
322 mutex_unlock(&monc->mutex);
325 EXPORT_SYMBOL(ceph_monc_wait_osdmap);
330 int ceph_monc_open_session(struct ceph_mon_client *monc)
332 mutex_lock(&monc->mutex);
333 __open_session(monc);
334 __schedule_delayed(monc);
335 mutex_unlock(&monc->mutex);
338 EXPORT_SYMBOL(ceph_monc_open_session);
341 * We require the fsid and global_id in order to initialize our
344 static bool have_debugfs_info(struct ceph_mon_client *monc)
346 dout("have_debugfs_info fsid %d globalid %lld\n",
347 (int)monc->client->have_fsid, monc->auth->global_id);
348 return monc->client->have_fsid && monc->auth->global_id > 0;
352 * The monitor responds with mount ack indicate mount success. The
353 * included client ticket allows the client to talk to MDSs and OSDs.
355 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
356 struct ceph_msg *msg)
358 struct ceph_client *client = monc->client;
359 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
361 int had_debugfs_info, init_debugfs = 0;
363 mutex_lock(&monc->mutex);
365 had_debugfs_info = have_debugfs_info(monc);
367 dout("handle_monmap\n");
368 p = msg->front.iov_base;
369 end = p + msg->front.iov_len;
371 monmap = ceph_monmap_decode(p, end);
372 if (IS_ERR(monmap)) {
373 pr_err("problem decoding monmap, %d\n",
374 (int)PTR_ERR(monmap));
378 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
383 client->monc.monmap = monmap;
386 if (!client->have_fsid) {
387 client->have_fsid = true;
388 if (!had_debugfs_info && have_debugfs_info(monc)) {
389 pr_info("client%lld fsid %pU\n",
390 ceph_client_id(monc->client),
391 &monc->client->fsid);
394 mutex_unlock(&monc->mutex);
398 * do debugfs initialization without mutex to avoid
399 * creating a locking dependency
401 ceph_debugfs_client_init(monc->client);
407 mutex_unlock(&monc->mutex);
409 wake_up_all(&client->auth_wq);
413 * generic requests (e.g., statfs, poolop)
415 static struct ceph_mon_generic_request *__lookup_generic_req(
416 struct ceph_mon_client *monc, u64 tid)
418 struct ceph_mon_generic_request *req;
419 struct rb_node *n = monc->generic_request_tree.rb_node;
422 req = rb_entry(n, struct ceph_mon_generic_request, node);
425 else if (tid > req->tid)
433 static void __insert_generic_request(struct ceph_mon_client *monc,
434 struct ceph_mon_generic_request *new)
436 struct rb_node **p = &monc->generic_request_tree.rb_node;
437 struct rb_node *parent = NULL;
438 struct ceph_mon_generic_request *req = NULL;
442 req = rb_entry(parent, struct ceph_mon_generic_request, node);
443 if (new->tid < req->tid)
445 else if (new->tid > req->tid)
451 rb_link_node(&new->node, parent, p);
452 rb_insert_color(&new->node, &monc->generic_request_tree);
455 static void release_generic_request(struct kref *kref)
457 struct ceph_mon_generic_request *req =
458 container_of(kref, struct ceph_mon_generic_request, kref);
461 ceph_msg_put(req->reply);
463 ceph_msg_put(req->request);
468 static void put_generic_request(struct ceph_mon_generic_request *req)
470 kref_put(&req->kref, release_generic_request);
473 static void get_generic_request(struct ceph_mon_generic_request *req)
475 kref_get(&req->kref);
478 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
479 struct ceph_msg_header *hdr,
482 struct ceph_mon_client *monc = con->private;
483 struct ceph_mon_generic_request *req;
484 u64 tid = le64_to_cpu(hdr->tid);
487 mutex_lock(&monc->mutex);
488 req = __lookup_generic_req(monc, tid);
490 dout("get_generic_reply %lld dne\n", tid);
494 dout("get_generic_reply %lld got %p\n", tid, req->reply);
496 m = ceph_msg_get(req->reply);
498 * we don't need to track the connection reading into
499 * this reply because we only have one open connection
503 mutex_unlock(&monc->mutex);
507 static int __do_generic_request(struct ceph_mon_client *monc, u64 tid,
508 struct ceph_mon_generic_request *req)
512 /* register request */
513 req->tid = tid != 0 ? tid : ++monc->last_tid;
514 req->request->hdr.tid = cpu_to_le64(req->tid);
515 __insert_generic_request(monc, req);
516 monc->num_generic_requests++;
517 ceph_con_send(&monc->con, ceph_msg_get(req->request));
518 mutex_unlock(&monc->mutex);
520 err = wait_for_completion_interruptible(&req->completion);
522 mutex_lock(&monc->mutex);
523 rb_erase(&req->node, &monc->generic_request_tree);
524 monc->num_generic_requests--;
531 static int do_generic_request(struct ceph_mon_client *monc,
532 struct ceph_mon_generic_request *req)
536 mutex_lock(&monc->mutex);
537 err = __do_generic_request(monc, 0, req);
538 mutex_unlock(&monc->mutex);
546 static void handle_statfs_reply(struct ceph_mon_client *monc,
547 struct ceph_msg *msg)
549 struct ceph_mon_generic_request *req;
550 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
551 u64 tid = le64_to_cpu(msg->hdr.tid);
553 if (msg->front.iov_len != sizeof(*reply))
555 dout("handle_statfs_reply %p tid %llu\n", msg, tid);
557 mutex_lock(&monc->mutex);
558 req = __lookup_generic_req(monc, tid);
560 *(struct ceph_statfs *)req->buf = reply->st;
562 get_generic_request(req);
564 mutex_unlock(&monc->mutex);
566 complete_all(&req->completion);
567 put_generic_request(req);
572 pr_err("corrupt generic reply, tid %llu\n", tid);
577 * Do a synchronous statfs().
579 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
581 struct ceph_mon_generic_request *req;
582 struct ceph_mon_statfs *h;
585 req = kzalloc(sizeof(*req), GFP_NOFS);
589 kref_init(&req->kref);
591 req->buf_len = sizeof(*buf);
592 init_completion(&req->completion);
595 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
599 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
604 /* fill out request */
605 h = req->request->front.iov_base;
606 h->monhdr.have_version = 0;
607 h->monhdr.session_mon = cpu_to_le16(-1);
608 h->monhdr.session_mon_tid = 0;
609 h->fsid = monc->monmap->fsid;
611 err = do_generic_request(monc, req);
614 kref_put(&req->kref, release_generic_request);
617 EXPORT_SYMBOL(ceph_monc_do_statfs);
619 static void handle_get_version_reply(struct ceph_mon_client *monc,
620 struct ceph_msg *msg)
622 struct ceph_mon_generic_request *req;
623 u64 tid = le64_to_cpu(msg->hdr.tid);
624 void *p = msg->front.iov_base;
625 void *end = p + msg->front_alloc_len;
628 dout("%s %p tid %llu\n", __func__, msg, tid);
630 ceph_decode_need(&p, end, 2*sizeof(u64), bad);
631 handle = ceph_decode_64(&p);
632 if (tid != 0 && tid != handle)
635 mutex_lock(&monc->mutex);
636 req = __lookup_generic_req(monc, handle);
638 *(u64 *)req->buf = ceph_decode_64(&p);
640 get_generic_request(req);
642 mutex_unlock(&monc->mutex);
644 complete_all(&req->completion);
645 put_generic_request(req);
650 pr_err("corrupt mon_get_version reply\n");
655 * Send MMonGetVersion and wait for the reply.
657 * @what: one of "mdsmap", "osdmap" or "monmap"
659 int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what,
662 struct ceph_mon_generic_request *req;
667 req = kzalloc(sizeof(*req), GFP_NOFS);
671 kref_init(&req->kref);
673 req->buf_len = sizeof(*newest);
674 init_completion(&req->completion);
676 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
677 sizeof(u64) + sizeof(u32) + strlen(what),
684 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024,
691 p = req->request->front.iov_base;
692 end = p + req->request->front_alloc_len;
694 /* fill out request */
695 mutex_lock(&monc->mutex);
696 tid = ++monc->last_tid;
697 ceph_encode_64(&p, tid); /* handle */
698 ceph_encode_string(&p, end, what, strlen(what));
700 err = __do_generic_request(monc, tid, req);
702 mutex_unlock(&monc->mutex);
704 kref_put(&req->kref, release_generic_request);
707 EXPORT_SYMBOL(ceph_monc_do_get_version);
712 static int get_poolop_reply_buf(const char *src, size_t src_len,
713 char *dst, size_t dst_len)
717 if (src_len != sizeof(u32) + dst_len)
720 buf_len = le32_to_cpu(*(u32 *)src);
721 if (buf_len != dst_len)
724 memcpy(dst, src + sizeof(u32), dst_len);
728 static void handle_poolop_reply(struct ceph_mon_client *monc,
729 struct ceph_msg *msg)
731 struct ceph_mon_generic_request *req;
732 struct ceph_mon_poolop_reply *reply = msg->front.iov_base;
733 u64 tid = le64_to_cpu(msg->hdr.tid);
735 if (msg->front.iov_len < sizeof(*reply))
737 dout("handle_poolop_reply %p tid %llu\n", msg, tid);
739 mutex_lock(&monc->mutex);
740 req = __lookup_generic_req(monc, tid);
743 get_poolop_reply_buf(msg->front.iov_base + sizeof(*reply),
744 msg->front.iov_len - sizeof(*reply),
745 req->buf, req->buf_len) < 0) {
746 mutex_unlock(&monc->mutex);
749 req->result = le32_to_cpu(reply->reply_code);
750 get_generic_request(req);
752 mutex_unlock(&monc->mutex);
754 complete(&req->completion);
755 put_generic_request(req);
760 pr_err("corrupt generic reply, tid %llu\n", tid);
765 * Do a synchronous pool op.
767 static int do_poolop(struct ceph_mon_client *monc, u32 op,
768 u32 pool, u64 snapid,
771 struct ceph_mon_generic_request *req;
772 struct ceph_mon_poolop *h;
775 req = kzalloc(sizeof(*req), GFP_NOFS);
779 kref_init(&req->kref);
782 init_completion(&req->completion);
785 req->request = ceph_msg_new(CEPH_MSG_POOLOP, sizeof(*h), GFP_NOFS,
789 req->reply = ceph_msg_new(CEPH_MSG_POOLOP_REPLY, 1024, GFP_NOFS,
794 /* fill out request */
795 req->request->hdr.version = cpu_to_le16(2);
796 h = req->request->front.iov_base;
797 h->monhdr.have_version = 0;
798 h->monhdr.session_mon = cpu_to_le16(-1);
799 h->monhdr.session_mon_tid = 0;
800 h->fsid = monc->monmap->fsid;
801 h->pool = cpu_to_le32(pool);
802 h->op = cpu_to_le32(op);
804 h->snapid = cpu_to_le64(snapid);
807 err = do_generic_request(monc, req);
810 kref_put(&req->kref, release_generic_request);
814 int ceph_monc_create_snapid(struct ceph_mon_client *monc,
815 u32 pool, u64 *snapid)
817 return do_poolop(monc, POOL_OP_CREATE_UNMANAGED_SNAP,
818 pool, 0, (char *)snapid, sizeof(*snapid));
821 EXPORT_SYMBOL(ceph_monc_create_snapid);
823 int ceph_monc_delete_snapid(struct ceph_mon_client *monc,
824 u32 pool, u64 snapid)
826 return do_poolop(monc, POOL_OP_CREATE_UNMANAGED_SNAP,
827 pool, snapid, NULL, 0);
832 * Resend pending generic requests.
834 static void __resend_generic_request(struct ceph_mon_client *monc)
836 struct ceph_mon_generic_request *req;
839 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
840 req = rb_entry(p, struct ceph_mon_generic_request, node);
841 ceph_msg_revoke(req->request);
842 ceph_msg_revoke_incoming(req->reply);
843 ceph_con_send(&monc->con, ceph_msg_get(req->request));
848 * Delayed work. If we haven't mounted yet, retry. Otherwise,
849 * renew/retry subscription as needed (in case it is timing out, or we
850 * got an ENOMEM). And keep the monitor connection alive.
852 static void delayed_work(struct work_struct *work)
854 struct ceph_mon_client *monc =
855 container_of(work, struct ceph_mon_client, delayed_work.work);
857 dout("monc delayed_work\n");
858 mutex_lock(&monc->mutex);
860 __close_session(monc);
861 __open_session(monc); /* continue hunting */
863 ceph_con_keepalive(&monc->con);
865 __validate_auth(monc);
867 if (ceph_auth_is_authenticated(monc->auth))
868 __send_subscribe(monc);
870 __schedule_delayed(monc);
871 mutex_unlock(&monc->mutex);
875 * On startup, we build a temporary monmap populated with the IPs
876 * provided by mount(2).
878 static int build_initial_monmap(struct ceph_mon_client *monc)
880 struct ceph_options *opt = monc->client->options;
881 struct ceph_entity_addr *mon_addr = opt->mon_addr;
882 int num_mon = opt->num_mon;
885 /* build initial monmap */
886 monc->monmap = kzalloc(sizeof(*monc->monmap) +
887 num_mon*sizeof(monc->monmap->mon_inst[0]),
891 for (i = 0; i < num_mon; i++) {
892 monc->monmap->mon_inst[i].addr = mon_addr[i];
893 monc->monmap->mon_inst[i].addr.nonce = 0;
894 monc->monmap->mon_inst[i].name.type =
895 CEPH_ENTITY_TYPE_MON;
896 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
898 monc->monmap->num_mon = num_mon;
902 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
907 memset(monc, 0, sizeof(*monc));
910 mutex_init(&monc->mutex);
912 err = build_initial_monmap(monc);
918 monc->auth = ceph_auth_init(cl->options->name,
920 if (IS_ERR(monc->auth)) {
921 err = PTR_ERR(monc->auth);
924 monc->auth->want_keys =
925 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
926 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
930 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
931 sizeof(struct ceph_mon_subscribe_ack),
933 if (!monc->m_subscribe_ack)
936 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
938 if (!monc->m_subscribe)
939 goto out_subscribe_ack;
941 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
943 if (!monc->m_auth_reply)
946 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
947 monc->pending_auth = 0;
951 ceph_con_init(&monc->con, monc, &mon_con_ops,
952 &monc->client->msgr);
955 monc->hunting = true;
956 monc->sub_renew_after = jiffies;
959 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
960 monc->generic_request_tree = RB_ROOT;
961 monc->num_generic_requests = 0;
964 monc->have_mdsmap = 0;
965 monc->have_osdmap = 0;
966 monc->want_next_osdmap = 1;
970 ceph_msg_put(monc->m_auth_reply);
972 ceph_msg_put(monc->m_subscribe);
974 ceph_msg_put(monc->m_subscribe_ack);
976 ceph_auth_destroy(monc->auth);
982 EXPORT_SYMBOL(ceph_monc_init);
984 void ceph_monc_stop(struct ceph_mon_client *monc)
987 cancel_delayed_work_sync(&monc->delayed_work);
989 mutex_lock(&monc->mutex);
990 __close_session(monc);
992 mutex_unlock(&monc->mutex);
995 * flush msgr queue before we destroy ourselves to ensure that:
996 * - any work that references our embedded con is finished.
997 * - any osd_client or other work that may reference an authorizer
998 * finishes before we shut down the auth subsystem.
1002 ceph_auth_destroy(monc->auth);
1004 ceph_msg_put(monc->m_auth);
1005 ceph_msg_put(monc->m_auth_reply);
1006 ceph_msg_put(monc->m_subscribe);
1007 ceph_msg_put(monc->m_subscribe_ack);
1009 kfree(monc->monmap);
1011 EXPORT_SYMBOL(ceph_monc_stop);
1013 static void handle_auth_reply(struct ceph_mon_client *monc,
1014 struct ceph_msg *msg)
1018 int had_debugfs_info, init_debugfs = 0;
1020 mutex_lock(&monc->mutex);
1021 had_debugfs_info = have_debugfs_info(monc);
1022 was_auth = ceph_auth_is_authenticated(monc->auth);
1023 monc->pending_auth = 0;
1024 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
1026 monc->m_auth->front.iov_base,
1027 monc->m_auth->front_alloc_len);
1029 monc->client->auth_err = ret;
1030 wake_up_all(&monc->client->auth_wq);
1031 } else if (ret > 0) {
1032 __send_prepared_auth_request(monc, ret);
1033 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
1034 dout("authenticated, starting session\n");
1036 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
1037 monc->client->msgr.inst.name.num =
1038 cpu_to_le64(monc->auth->global_id);
1040 __send_subscribe(monc);
1041 __resend_generic_request(monc);
1044 if (!had_debugfs_info && have_debugfs_info(monc)) {
1045 pr_info("client%lld fsid %pU\n",
1046 ceph_client_id(monc->client),
1047 &monc->client->fsid);
1050 mutex_unlock(&monc->mutex);
1054 * do debugfs initialization without mutex to avoid
1055 * creating a locking dependency
1057 ceph_debugfs_client_init(monc->client);
1061 static int __validate_auth(struct ceph_mon_client *monc)
1065 if (monc->pending_auth)
1068 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
1069 monc->m_auth->front_alloc_len);
1071 return ret; /* either an error, or no need to authenticate */
1072 __send_prepared_auth_request(monc, ret);
1076 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
1080 mutex_lock(&monc->mutex);
1081 ret = __validate_auth(monc);
1082 mutex_unlock(&monc->mutex);
1085 EXPORT_SYMBOL(ceph_monc_validate_auth);
1088 * handle incoming message
1090 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1092 struct ceph_mon_client *monc = con->private;
1093 int type = le16_to_cpu(msg->hdr.type);
1099 case CEPH_MSG_AUTH_REPLY:
1100 handle_auth_reply(monc, msg);
1103 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1104 handle_subscribe_ack(monc, msg);
1107 case CEPH_MSG_STATFS_REPLY:
1108 handle_statfs_reply(monc, msg);
1111 case CEPH_MSG_MON_GET_VERSION_REPLY:
1112 handle_get_version_reply(monc, msg);
1115 case CEPH_MSG_POOLOP_REPLY:
1116 handle_poolop_reply(monc, msg);
1119 case CEPH_MSG_MON_MAP:
1120 ceph_monc_handle_map(monc, msg);
1123 case CEPH_MSG_OSD_MAP:
1124 ceph_osdc_handle_map(&monc->client->osdc, msg);
1128 /* can the chained handler handle it? */
1129 if (monc->client->extra_mon_dispatch &&
1130 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1133 pr_err("received unknown message type %d %s\n", type,
1134 ceph_msg_type_name(type));
1140 * Allocate memory for incoming message
1142 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1143 struct ceph_msg_header *hdr,
1146 struct ceph_mon_client *monc = con->private;
1147 int type = le16_to_cpu(hdr->type);
1148 int front_len = le32_to_cpu(hdr->front_len);
1149 struct ceph_msg *m = NULL;
1154 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1155 m = ceph_msg_get(monc->m_subscribe_ack);
1157 case CEPH_MSG_POOLOP_REPLY:
1158 case CEPH_MSG_STATFS_REPLY:
1159 return get_generic_reply(con, hdr, skip);
1160 case CEPH_MSG_AUTH_REPLY:
1161 m = ceph_msg_get(monc->m_auth_reply);
1163 case CEPH_MSG_MON_GET_VERSION_REPLY:
1164 if (le64_to_cpu(hdr->tid) != 0)
1165 return get_generic_reply(con, hdr, skip);
1168 * Older OSDs don't set reply tid even if the orignal
1169 * request had a non-zero tid. Workaround this weirdness
1170 * by falling through to the allocate case.
1172 case CEPH_MSG_MON_MAP:
1173 case CEPH_MSG_MDS_MAP:
1174 case CEPH_MSG_OSD_MAP:
1175 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1177 return NULL; /* ENOMEM--return skip == 0 */
1182 pr_info("alloc_msg unknown type %d\n", type);
1184 } else if (front_len > m->front_alloc_len) {
1185 pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1186 front_len, m->front_alloc_len,
1187 (unsigned int)con->peer_name.type,
1188 le64_to_cpu(con->peer_name.num));
1190 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1197 * If the monitor connection resets, pick a new monitor and resubmit
1198 * any pending requests.
1200 static void mon_fault(struct ceph_connection *con)
1202 struct ceph_mon_client *monc = con->private;
1207 dout("mon_fault\n");
1208 mutex_lock(&monc->mutex);
1213 pr_info("mon%d %s session lost, "
1214 "hunting for new mon\n", monc->cur_mon,
1215 ceph_pr_addr(&monc->con.peer_addr.in_addr));
1217 __close_session(monc);
1218 if (!monc->hunting) {
1220 monc->hunting = true;
1221 __open_session(monc);
1223 /* already hunting, let's wait a bit */
1224 __schedule_delayed(monc);
1227 mutex_unlock(&monc->mutex);
1231 * We can ignore refcounting on the connection struct, as all references
1232 * will come from the messenger workqueue, which is drained prior to
1233 * mon_client destruction.
1235 static struct ceph_connection *con_get(struct ceph_connection *con)
1240 static void con_put(struct ceph_connection *con)
1244 static const struct ceph_connection_operations mon_con_ops = {
1247 .dispatch = dispatch,
1249 .alloc_msg = mon_alloc_msg,