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);
303 int ceph_monc_open_session(struct ceph_mon_client *monc)
305 mutex_lock(&monc->mutex);
306 __open_session(monc);
307 __schedule_delayed(monc);
308 mutex_unlock(&monc->mutex);
311 EXPORT_SYMBOL(ceph_monc_open_session);
314 * We require the fsid and global_id in order to initialize our
317 static bool have_debugfs_info(struct ceph_mon_client *monc)
319 dout("have_debugfs_info fsid %d globalid %lld\n",
320 (int)monc->client->have_fsid, monc->auth->global_id);
321 return monc->client->have_fsid && monc->auth->global_id > 0;
325 * The monitor responds with mount ack indicate mount success. The
326 * included client ticket allows the client to talk to MDSs and OSDs.
328 static void ceph_monc_handle_map(struct ceph_mon_client *monc,
329 struct ceph_msg *msg)
331 struct ceph_client *client = monc->client;
332 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
334 int had_debugfs_info, init_debugfs = 0;
336 mutex_lock(&monc->mutex);
338 had_debugfs_info = have_debugfs_info(monc);
340 dout("handle_monmap\n");
341 p = msg->front.iov_base;
342 end = p + msg->front.iov_len;
344 monmap = ceph_monmap_decode(p, end);
345 if (IS_ERR(monmap)) {
346 pr_err("problem decoding monmap, %d\n",
347 (int)PTR_ERR(monmap));
351 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
356 client->monc.monmap = monmap;
359 if (!client->have_fsid) {
360 client->have_fsid = true;
361 if (!had_debugfs_info && have_debugfs_info(monc)) {
362 pr_info("client%lld fsid %pU\n",
363 ceph_client_id(monc->client),
364 &monc->client->fsid);
367 mutex_unlock(&monc->mutex);
371 * do debugfs initialization without mutex to avoid
372 * creating a locking dependency
374 ceph_debugfs_client_init(monc->client);
380 mutex_unlock(&monc->mutex);
382 wake_up_all(&client->auth_wq);
386 * generic requests (e.g., statfs, poolop)
388 static struct ceph_mon_generic_request *__lookup_generic_req(
389 struct ceph_mon_client *monc, u64 tid)
391 struct ceph_mon_generic_request *req;
392 struct rb_node *n = monc->generic_request_tree.rb_node;
395 req = rb_entry(n, struct ceph_mon_generic_request, node);
398 else if (tid > req->tid)
406 static void __insert_generic_request(struct ceph_mon_client *monc,
407 struct ceph_mon_generic_request *new)
409 struct rb_node **p = &monc->generic_request_tree.rb_node;
410 struct rb_node *parent = NULL;
411 struct ceph_mon_generic_request *req = NULL;
415 req = rb_entry(parent, struct ceph_mon_generic_request, node);
416 if (new->tid < req->tid)
418 else if (new->tid > req->tid)
424 rb_link_node(&new->node, parent, p);
425 rb_insert_color(&new->node, &monc->generic_request_tree);
428 static void release_generic_request(struct kref *kref)
430 struct ceph_mon_generic_request *req =
431 container_of(kref, struct ceph_mon_generic_request, kref);
434 ceph_msg_put(req->reply);
436 ceph_msg_put(req->request);
441 static void put_generic_request(struct ceph_mon_generic_request *req)
443 kref_put(&req->kref, release_generic_request);
446 static void get_generic_request(struct ceph_mon_generic_request *req)
448 kref_get(&req->kref);
451 static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
452 struct ceph_msg_header *hdr,
455 struct ceph_mon_client *monc = con->private;
456 struct ceph_mon_generic_request *req;
457 u64 tid = le64_to_cpu(hdr->tid);
460 mutex_lock(&monc->mutex);
461 req = __lookup_generic_req(monc, tid);
463 dout("get_generic_reply %lld dne\n", tid);
467 dout("get_generic_reply %lld got %p\n", tid, req->reply);
469 m = ceph_msg_get(req->reply);
471 * we don't need to track the connection reading into
472 * this reply because we only have one open connection
476 mutex_unlock(&monc->mutex);
480 static int __do_generic_request(struct ceph_mon_client *monc, u64 tid,
481 struct ceph_mon_generic_request *req)
485 /* register request */
486 req->tid = tid != 0 ? tid : ++monc->last_tid;
487 req->request->hdr.tid = cpu_to_le64(req->tid);
488 __insert_generic_request(monc, req);
489 monc->num_generic_requests++;
490 ceph_con_send(&monc->con, ceph_msg_get(req->request));
491 mutex_unlock(&monc->mutex);
493 err = wait_for_completion_interruptible(&req->completion);
495 mutex_lock(&monc->mutex);
496 rb_erase(&req->node, &monc->generic_request_tree);
497 monc->num_generic_requests--;
504 static int do_generic_request(struct ceph_mon_client *monc,
505 struct ceph_mon_generic_request *req)
509 mutex_lock(&monc->mutex);
510 err = __do_generic_request(monc, 0, req);
511 mutex_unlock(&monc->mutex);
519 static void handle_statfs_reply(struct ceph_mon_client *monc,
520 struct ceph_msg *msg)
522 struct ceph_mon_generic_request *req;
523 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
524 u64 tid = le64_to_cpu(msg->hdr.tid);
526 if (msg->front.iov_len != sizeof(*reply))
528 dout("handle_statfs_reply %p tid %llu\n", msg, tid);
530 mutex_lock(&monc->mutex);
531 req = __lookup_generic_req(monc, tid);
533 *(struct ceph_statfs *)req->buf = reply->st;
535 get_generic_request(req);
537 mutex_unlock(&monc->mutex);
539 complete_all(&req->completion);
540 put_generic_request(req);
545 pr_err("corrupt generic reply, tid %llu\n", tid);
550 * Do a synchronous statfs().
552 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
554 struct ceph_mon_generic_request *req;
555 struct ceph_mon_statfs *h;
558 req = kzalloc(sizeof(*req), GFP_NOFS);
562 kref_init(&req->kref);
564 req->buf_len = sizeof(*buf);
565 init_completion(&req->completion);
568 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
572 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
577 /* fill out request */
578 h = req->request->front.iov_base;
579 h->monhdr.have_version = 0;
580 h->monhdr.session_mon = cpu_to_le16(-1);
581 h->monhdr.session_mon_tid = 0;
582 h->fsid = monc->monmap->fsid;
584 err = do_generic_request(monc, req);
587 kref_put(&req->kref, release_generic_request);
590 EXPORT_SYMBOL(ceph_monc_do_statfs);
592 static void handle_get_version_reply(struct ceph_mon_client *monc,
593 struct ceph_msg *msg)
595 struct ceph_mon_generic_request *req;
596 u64 tid = le64_to_cpu(msg->hdr.tid);
597 void *p = msg->front.iov_base;
598 void *end = p + msg->front_alloc_len;
601 dout("%s %p tid %llu\n", __func__, msg, tid);
603 ceph_decode_need(&p, end, 2*sizeof(u64), bad);
604 handle = ceph_decode_64(&p);
605 if (tid != 0 && tid != handle)
608 mutex_lock(&monc->mutex);
609 req = __lookup_generic_req(monc, handle);
611 *(u64 *)req->buf = ceph_decode_64(&p);
613 get_generic_request(req);
615 mutex_unlock(&monc->mutex);
617 complete_all(&req->completion);
618 put_generic_request(req);
623 pr_err("corrupt mon_get_version reply\n");
628 * Send MMonGetVersion and wait for the reply.
630 * @what: one of "mdsmap", "osdmap" or "monmap"
632 int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what,
635 struct ceph_mon_generic_request *req;
640 req = kzalloc(sizeof(*req), GFP_NOFS);
644 kref_init(&req->kref);
646 req->buf_len = sizeof(*newest);
647 init_completion(&req->completion);
649 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
650 sizeof(u64) + sizeof(u32) + strlen(what),
657 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024,
664 p = req->request->front.iov_base;
665 end = p + req->request->front_alloc_len;
667 /* fill out request */
668 mutex_lock(&monc->mutex);
669 tid = ++monc->last_tid;
670 ceph_encode_64(&p, tid); /* handle */
671 ceph_encode_string(&p, end, what, strlen(what));
673 err = __do_generic_request(monc, tid, req);
675 mutex_unlock(&monc->mutex);
677 kref_put(&req->kref, release_generic_request);
680 EXPORT_SYMBOL(ceph_monc_do_get_version);
685 static int get_poolop_reply_buf(const char *src, size_t src_len,
686 char *dst, size_t dst_len)
690 if (src_len != sizeof(u32) + dst_len)
693 buf_len = le32_to_cpu(*(u32 *)src);
694 if (buf_len != dst_len)
697 memcpy(dst, src + sizeof(u32), dst_len);
701 static void handle_poolop_reply(struct ceph_mon_client *monc,
702 struct ceph_msg *msg)
704 struct ceph_mon_generic_request *req;
705 struct ceph_mon_poolop_reply *reply = msg->front.iov_base;
706 u64 tid = le64_to_cpu(msg->hdr.tid);
708 if (msg->front.iov_len < sizeof(*reply))
710 dout("handle_poolop_reply %p tid %llu\n", msg, tid);
712 mutex_lock(&monc->mutex);
713 req = __lookup_generic_req(monc, tid);
716 get_poolop_reply_buf(msg->front.iov_base + sizeof(*reply),
717 msg->front.iov_len - sizeof(*reply),
718 req->buf, req->buf_len) < 0) {
719 mutex_unlock(&monc->mutex);
722 req->result = le32_to_cpu(reply->reply_code);
723 get_generic_request(req);
725 mutex_unlock(&monc->mutex);
727 complete(&req->completion);
728 put_generic_request(req);
733 pr_err("corrupt generic reply, tid %llu\n", tid);
738 * Do a synchronous pool op.
740 static int do_poolop(struct ceph_mon_client *monc, u32 op,
741 u32 pool, u64 snapid,
744 struct ceph_mon_generic_request *req;
745 struct ceph_mon_poolop *h;
748 req = kzalloc(sizeof(*req), GFP_NOFS);
752 kref_init(&req->kref);
755 init_completion(&req->completion);
758 req->request = ceph_msg_new(CEPH_MSG_POOLOP, sizeof(*h), GFP_NOFS,
762 req->reply = ceph_msg_new(CEPH_MSG_POOLOP_REPLY, 1024, GFP_NOFS,
767 /* fill out request */
768 req->request->hdr.version = cpu_to_le16(2);
769 h = req->request->front.iov_base;
770 h->monhdr.have_version = 0;
771 h->monhdr.session_mon = cpu_to_le16(-1);
772 h->monhdr.session_mon_tid = 0;
773 h->fsid = monc->monmap->fsid;
774 h->pool = cpu_to_le32(pool);
775 h->op = cpu_to_le32(op);
777 h->snapid = cpu_to_le64(snapid);
780 err = do_generic_request(monc, req);
783 kref_put(&req->kref, release_generic_request);
787 int ceph_monc_create_snapid(struct ceph_mon_client *monc,
788 u32 pool, u64 *snapid)
790 return do_poolop(monc, POOL_OP_CREATE_UNMANAGED_SNAP,
791 pool, 0, (char *)snapid, sizeof(*snapid));
794 EXPORT_SYMBOL(ceph_monc_create_snapid);
796 int ceph_monc_delete_snapid(struct ceph_mon_client *monc,
797 u32 pool, u64 snapid)
799 return do_poolop(monc, POOL_OP_CREATE_UNMANAGED_SNAP,
800 pool, snapid, NULL, 0);
805 * Resend pending generic requests.
807 static void __resend_generic_request(struct ceph_mon_client *monc)
809 struct ceph_mon_generic_request *req;
812 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
813 req = rb_entry(p, struct ceph_mon_generic_request, node);
814 ceph_msg_revoke(req->request);
815 ceph_msg_revoke_incoming(req->reply);
816 ceph_con_send(&monc->con, ceph_msg_get(req->request));
821 * Delayed work. If we haven't mounted yet, retry. Otherwise,
822 * renew/retry subscription as needed (in case it is timing out, or we
823 * got an ENOMEM). And keep the monitor connection alive.
825 static void delayed_work(struct work_struct *work)
827 struct ceph_mon_client *monc =
828 container_of(work, struct ceph_mon_client, delayed_work.work);
830 dout("monc delayed_work\n");
831 mutex_lock(&monc->mutex);
833 __close_session(monc);
834 __open_session(monc); /* continue hunting */
836 ceph_con_keepalive(&monc->con);
838 __validate_auth(monc);
840 if (ceph_auth_is_authenticated(monc->auth))
841 __send_subscribe(monc);
843 __schedule_delayed(monc);
844 mutex_unlock(&monc->mutex);
848 * On startup, we build a temporary monmap populated with the IPs
849 * provided by mount(2).
851 static int build_initial_monmap(struct ceph_mon_client *monc)
853 struct ceph_options *opt = monc->client->options;
854 struct ceph_entity_addr *mon_addr = opt->mon_addr;
855 int num_mon = opt->num_mon;
858 /* build initial monmap */
859 monc->monmap = kzalloc(sizeof(*monc->monmap) +
860 num_mon*sizeof(monc->monmap->mon_inst[0]),
864 for (i = 0; i < num_mon; i++) {
865 monc->monmap->mon_inst[i].addr = mon_addr[i];
866 monc->monmap->mon_inst[i].addr.nonce = 0;
867 monc->monmap->mon_inst[i].name.type =
868 CEPH_ENTITY_TYPE_MON;
869 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
871 monc->monmap->num_mon = num_mon;
875 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
880 memset(monc, 0, sizeof(*monc));
883 mutex_init(&monc->mutex);
885 err = build_initial_monmap(monc);
891 monc->auth = ceph_auth_init(cl->options->name,
893 if (IS_ERR(monc->auth)) {
894 err = PTR_ERR(monc->auth);
897 monc->auth->want_keys =
898 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
899 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
903 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
904 sizeof(struct ceph_mon_subscribe_ack),
906 if (!monc->m_subscribe_ack)
909 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
911 if (!monc->m_subscribe)
912 goto out_subscribe_ack;
914 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
916 if (!monc->m_auth_reply)
919 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
920 monc->pending_auth = 0;
924 ceph_con_init(&monc->con, monc, &mon_con_ops,
925 &monc->client->msgr);
928 monc->hunting = true;
929 monc->sub_renew_after = jiffies;
932 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
933 monc->generic_request_tree = RB_ROOT;
934 monc->num_generic_requests = 0;
937 monc->have_mdsmap = 0;
938 monc->have_osdmap = 0;
939 monc->want_next_osdmap = 1;
943 ceph_msg_put(monc->m_auth_reply);
945 ceph_msg_put(monc->m_subscribe);
947 ceph_msg_put(monc->m_subscribe_ack);
949 ceph_auth_destroy(monc->auth);
955 EXPORT_SYMBOL(ceph_monc_init);
957 void ceph_monc_stop(struct ceph_mon_client *monc)
960 cancel_delayed_work_sync(&monc->delayed_work);
962 mutex_lock(&monc->mutex);
963 __close_session(monc);
965 mutex_unlock(&monc->mutex);
968 * flush msgr queue before we destroy ourselves to ensure that:
969 * - any work that references our embedded con is finished.
970 * - any osd_client or other work that may reference an authorizer
971 * finishes before we shut down the auth subsystem.
975 ceph_auth_destroy(monc->auth);
977 ceph_msg_put(monc->m_auth);
978 ceph_msg_put(monc->m_auth_reply);
979 ceph_msg_put(monc->m_subscribe);
980 ceph_msg_put(monc->m_subscribe_ack);
984 EXPORT_SYMBOL(ceph_monc_stop);
986 static void handle_auth_reply(struct ceph_mon_client *monc,
987 struct ceph_msg *msg)
991 int had_debugfs_info, init_debugfs = 0;
993 mutex_lock(&monc->mutex);
994 had_debugfs_info = have_debugfs_info(monc);
995 was_auth = ceph_auth_is_authenticated(monc->auth);
996 monc->pending_auth = 0;
997 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
999 monc->m_auth->front.iov_base,
1000 monc->m_auth->front_alloc_len);
1002 monc->client->auth_err = ret;
1003 wake_up_all(&monc->client->auth_wq);
1004 } else if (ret > 0) {
1005 __send_prepared_auth_request(monc, ret);
1006 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
1007 dout("authenticated, starting session\n");
1009 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
1010 monc->client->msgr.inst.name.num =
1011 cpu_to_le64(monc->auth->global_id);
1013 __send_subscribe(monc);
1014 __resend_generic_request(monc);
1017 if (!had_debugfs_info && have_debugfs_info(monc)) {
1018 pr_info("client%lld fsid %pU\n",
1019 ceph_client_id(monc->client),
1020 &monc->client->fsid);
1023 mutex_unlock(&monc->mutex);
1027 * do debugfs initialization without mutex to avoid
1028 * creating a locking dependency
1030 ceph_debugfs_client_init(monc->client);
1034 static int __validate_auth(struct ceph_mon_client *monc)
1038 if (monc->pending_auth)
1041 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
1042 monc->m_auth->front_alloc_len);
1044 return ret; /* either an error, or no need to authenticate */
1045 __send_prepared_auth_request(monc, ret);
1049 int ceph_monc_validate_auth(struct ceph_mon_client *monc)
1053 mutex_lock(&monc->mutex);
1054 ret = __validate_auth(monc);
1055 mutex_unlock(&monc->mutex);
1058 EXPORT_SYMBOL(ceph_monc_validate_auth);
1061 * handle incoming message
1063 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1065 struct ceph_mon_client *monc = con->private;
1066 int type = le16_to_cpu(msg->hdr.type);
1072 case CEPH_MSG_AUTH_REPLY:
1073 handle_auth_reply(monc, msg);
1076 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1077 handle_subscribe_ack(monc, msg);
1080 case CEPH_MSG_STATFS_REPLY:
1081 handle_statfs_reply(monc, msg);
1084 case CEPH_MSG_MON_GET_VERSION_REPLY:
1085 handle_get_version_reply(monc, msg);
1088 case CEPH_MSG_POOLOP_REPLY:
1089 handle_poolop_reply(monc, msg);
1092 case CEPH_MSG_MON_MAP:
1093 ceph_monc_handle_map(monc, msg);
1096 case CEPH_MSG_OSD_MAP:
1097 ceph_osdc_handle_map(&monc->client->osdc, msg);
1101 /* can the chained handler handle it? */
1102 if (monc->client->extra_mon_dispatch &&
1103 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1106 pr_err("received unknown message type %d %s\n", type,
1107 ceph_msg_type_name(type));
1113 * Allocate memory for incoming message
1115 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1116 struct ceph_msg_header *hdr,
1119 struct ceph_mon_client *monc = con->private;
1120 int type = le16_to_cpu(hdr->type);
1121 int front_len = le32_to_cpu(hdr->front_len);
1122 struct ceph_msg *m = NULL;
1127 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1128 m = ceph_msg_get(monc->m_subscribe_ack);
1130 case CEPH_MSG_POOLOP_REPLY:
1131 case CEPH_MSG_STATFS_REPLY:
1132 return get_generic_reply(con, hdr, skip);
1133 case CEPH_MSG_AUTH_REPLY:
1134 m = ceph_msg_get(monc->m_auth_reply);
1136 case CEPH_MSG_MON_GET_VERSION_REPLY:
1137 if (le64_to_cpu(hdr->tid) != 0)
1138 return get_generic_reply(con, hdr, skip);
1141 * Older OSDs don't set reply tid even if the orignal
1142 * request had a non-zero tid. Workaround this weirdness
1143 * by falling through to the allocate case.
1145 case CEPH_MSG_MON_MAP:
1146 case CEPH_MSG_MDS_MAP:
1147 case CEPH_MSG_OSD_MAP:
1148 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1150 return NULL; /* ENOMEM--return skip == 0 */
1155 pr_info("alloc_msg unknown type %d\n", type);
1162 * If the monitor connection resets, pick a new monitor and resubmit
1163 * any pending requests.
1165 static void mon_fault(struct ceph_connection *con)
1167 struct ceph_mon_client *monc = con->private;
1172 dout("mon_fault\n");
1173 mutex_lock(&monc->mutex);
1178 pr_info("mon%d %s session lost, "
1179 "hunting for new mon\n", monc->cur_mon,
1180 ceph_pr_addr(&monc->con.peer_addr.in_addr));
1182 __close_session(monc);
1183 if (!monc->hunting) {
1185 monc->hunting = true;
1186 __open_session(monc);
1188 /* already hunting, let's wait a bit */
1189 __schedule_delayed(monc);
1192 mutex_unlock(&monc->mutex);
1196 * We can ignore refcounting on the connection struct, as all references
1197 * will come from the messenger workqueue, which is drained prior to
1198 * mon_client destruction.
1200 static struct ceph_connection *con_get(struct ceph_connection *con)
1205 static void con_put(struct ceph_connection *con)
1209 static const struct ceph_connection_operations mon_con_ops = {
1212 .dispatch = dispatch,
1214 .alloc_msg = mon_alloc_msg,