1 #include "ceph_debug.h"
3 #include <linux/types.h>
4 #include <linux/random.h>
5 #include <linux/sched.h>
7 #include "mon_client.h"
13 * Interact with Ceph monitor cluster. Handle requests for new map
14 * versions, and periodically resend as needed. Also implement
15 * statfs() and umount().
17 * A small cluster of Ceph "monitors" are responsible for managing critical
18 * cluster configuration and state information. An odd number (e.g., 3, 5)
19 * of cmon daemons use a modified version of the Paxos part-time parliament
20 * algorithm to manage the MDS map (mds cluster membership), OSD map, and
21 * list of clients who have mounted the file system.
23 * We maintain an open, active session with a monitor at all times in order to
24 * receive timely MDSMap updates. We periodically send a keepalive byte on the
25 * TCP socket to ensure we detect a failure. If the connection does break, we
26 * randomly hunt for a new monitor. Once the connection is reestablished, we
27 * resend any outstanding requests.
30 const static struct ceph_connection_operations mon_con_ops
;
33 * Decode a monmap blob (e.g., during mount).
35 struct ceph_monmap
*ceph_monmap_decode(void *p
, void *end
)
37 struct ceph_monmap
*m
= NULL
;
39 struct ceph_fsid fsid
;
44 ceph_decode_32_safe(&p
, end
, len
, bad
);
45 ceph_decode_need(&p
, end
, len
, bad
);
47 dout("monmap_decode %p %p len %d\n", p
, end
, (int)(end
-p
));
49 ceph_decode_16_safe(&p
, end
, version
, bad
);
51 ceph_decode_need(&p
, end
, sizeof(fsid
) + 2*sizeof(u32
), bad
);
52 ceph_decode_copy(&p
, &fsid
, sizeof(fsid
));
53 epoch
= ceph_decode_32(&p
);
55 num_mon
= ceph_decode_32(&p
);
56 ceph_decode_need(&p
, end
, num_mon
*sizeof(m
->mon_inst
[0]), bad
);
58 if (num_mon
>= CEPH_MAX_MON
)
60 m
= kmalloc(sizeof(*m
) + sizeof(m
->mon_inst
[0])*num_mon
, GFP_NOFS
);
62 return ERR_PTR(-ENOMEM
);
66 ceph_decode_copy(&p
, m
->mon_inst
, num_mon
*sizeof(m
->mon_inst
[0]));
67 for (i
= 0; i
< num_mon
; i
++)
68 ceph_decode_addr(&m
->mon_inst
[i
].addr
);
70 dout("monmap_decode epoch %d, num_mon %d\n", m
->epoch
,
72 for (i
= 0; i
< m
->num_mon
; i
++)
73 dout("monmap_decode mon%d is %s\n", i
,
74 pr_addr(&m
->mon_inst
[i
].addr
.in_addr
));
78 dout("monmap_decode failed with %d\n", err
);
84 * return true if *addr is included in the monmap.
86 int ceph_monmap_contains(struct ceph_monmap
*m
, struct ceph_entity_addr
*addr
)
90 for (i
= 0; i
< m
->num_mon
; i
++)
91 if (ceph_entity_addr_equal(addr
, &m
->mon_inst
[i
].addr
))
97 * Close monitor session, if any.
99 static void __close_session(struct ceph_mon_client
*monc
)
102 dout("__close_session closing mon%d\n", monc
->cur_mon
);
103 ceph_con_revoke(monc
->con
, monc
->m_auth
);
104 ceph_con_close(monc
->con
);
106 ceph_auth_reset(monc
->auth
);
111 * Open a session with a (new) monitor.
113 static int __open_session(struct ceph_mon_client
*monc
)
118 if (monc
->cur_mon
< 0) {
119 get_random_bytes(&r
, 1);
120 monc
->cur_mon
= r
% monc
->monmap
->num_mon
;
121 dout("open_session num=%d r=%d -> mon%d\n",
122 monc
->monmap
->num_mon
, r
, monc
->cur_mon
);
124 monc
->sub_renew_after
= jiffies
; /* i.e., expired */
125 monc
->want_next_osdmap
= !!monc
->want_next_osdmap
;
127 dout("open_session mon%d opening\n", monc
->cur_mon
);
128 monc
->con
->peer_name
.type
= CEPH_ENTITY_TYPE_MON
;
129 monc
->con
->peer_name
.num
= cpu_to_le64(monc
->cur_mon
);
130 ceph_con_open(monc
->con
,
131 &monc
->monmap
->mon_inst
[monc
->cur_mon
].addr
);
133 /* initiatiate authentication handshake */
134 ret
= ceph_auth_build_hello(monc
->auth
,
135 monc
->m_auth
->front
.iov_base
,
136 monc
->m_auth
->front_max
);
137 monc
->m_auth
->front
.iov_len
= ret
;
138 monc
->m_auth
->hdr
.front_len
= cpu_to_le32(ret
);
139 ceph_msg_get(monc
->m_auth
); /* keep our ref */
140 ceph_con_send(monc
->con
, monc
->m_auth
);
142 dout("open_session mon%d already open\n", monc
->cur_mon
);
147 static bool __sub_expired(struct ceph_mon_client
*monc
)
149 return time_after_eq(jiffies
, monc
->sub_renew_after
);
153 * Reschedule delayed work timer.
155 static void __schedule_delayed(struct ceph_mon_client
*monc
)
159 if (monc
->cur_mon
< 0 || __sub_expired(monc
))
163 dout("__schedule_delayed after %u\n", delay
);
164 schedule_delayed_work(&monc
->delayed_work
, delay
);
168 * Send subscribe request for mdsmap and/or osdmap.
170 static void __send_subscribe(struct ceph_mon_client
*monc
)
172 dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n",
173 (unsigned)monc
->sub_sent
, __sub_expired(monc
),
174 monc
->want_next_osdmap
);
175 if ((__sub_expired(monc
) && !monc
->sub_sent
) ||
176 monc
->want_next_osdmap
== 1) {
177 struct ceph_msg
*msg
;
178 struct ceph_mon_subscribe_item
*i
;
181 msg
= ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE
, 96, 0, 0, NULL
);
185 p
= msg
->front
.iov_base
;
186 end
= p
+ msg
->front
.iov_len
;
188 dout("__send_subscribe to 'mdsmap' %u+\n",
189 (unsigned)monc
->have_mdsmap
);
190 if (monc
->want_next_osdmap
) {
191 dout("__send_subscribe to 'osdmap' %u\n",
192 (unsigned)monc
->have_osdmap
);
193 ceph_encode_32(&p
, 3);
194 ceph_encode_string(&p
, end
, "osdmap", 6);
196 i
->have
= cpu_to_le64(monc
->have_osdmap
);
199 monc
->want_next_osdmap
= 2; /* requested */
201 ceph_encode_32(&p
, 2);
203 ceph_encode_string(&p
, end
, "mdsmap", 6);
205 i
->have
= cpu_to_le64(monc
->have_mdsmap
);
208 ceph_encode_string(&p
, end
, "monmap", 6);
214 msg
->front
.iov_len
= p
- msg
->front
.iov_base
;
215 msg
->hdr
.front_len
= cpu_to_le32(msg
->front
.iov_len
);
216 ceph_con_send(monc
->con
, msg
);
218 monc
->sub_sent
= jiffies
| 1; /* never 0 */
222 static void handle_subscribe_ack(struct ceph_mon_client
*monc
,
223 struct ceph_msg
*msg
)
226 struct ceph_mon_subscribe_ack
*h
= msg
->front
.iov_base
;
228 if (msg
->front
.iov_len
< sizeof(*h
))
230 seconds
= le32_to_cpu(h
->duration
);
232 mutex_lock(&monc
->mutex
);
234 pr_info("mon%d %s session established\n",
235 monc
->cur_mon
, pr_addr(&monc
->con
->peer_addr
.in_addr
));
236 monc
->hunting
= false;
238 dout("handle_subscribe_ack after %d seconds\n", seconds
);
239 monc
->sub_renew_after
= monc
->sub_sent
+ (seconds
>> 1)*HZ
- 1;
241 mutex_unlock(&monc
->mutex
);
244 pr_err("got corrupt subscribe-ack msg\n");
249 * Keep track of which maps we have
251 int ceph_monc_got_mdsmap(struct ceph_mon_client
*monc
, u32 got
)
253 mutex_lock(&monc
->mutex
);
254 monc
->have_mdsmap
= got
;
255 mutex_unlock(&monc
->mutex
);
259 int ceph_monc_got_osdmap(struct ceph_mon_client
*monc
, u32 got
)
261 mutex_lock(&monc
->mutex
);
262 monc
->have_osdmap
= got
;
263 monc
->want_next_osdmap
= 0;
264 mutex_unlock(&monc
->mutex
);
269 * Register interest in the next osdmap
271 void ceph_monc_request_next_osdmap(struct ceph_mon_client
*monc
)
273 dout("request_next_osdmap have %u\n", monc
->have_osdmap
);
274 mutex_lock(&monc
->mutex
);
275 if (!monc
->want_next_osdmap
)
276 monc
->want_next_osdmap
= 1;
277 if (monc
->want_next_osdmap
< 2)
278 __send_subscribe(monc
);
279 mutex_unlock(&monc
->mutex
);
285 int ceph_monc_open_session(struct ceph_mon_client
*monc
)
288 monc
->con
= kmalloc(sizeof(*monc
->con
), GFP_KERNEL
);
291 ceph_con_init(monc
->client
->msgr
, monc
->con
);
292 monc
->con
->private = monc
;
293 monc
->con
->ops
= &mon_con_ops
;
296 mutex_lock(&monc
->mutex
);
297 __open_session(monc
);
298 __schedule_delayed(monc
);
299 mutex_unlock(&monc
->mutex
);
304 * The monitor responds with mount ack indicate mount success. The
305 * included client ticket allows the client to talk to MDSs and OSDs.
307 static void ceph_monc_handle_map(struct ceph_mon_client
*monc
,
308 struct ceph_msg
*msg
)
310 struct ceph_client
*client
= monc
->client
;
311 struct ceph_monmap
*monmap
= NULL
, *old
= monc
->monmap
;
314 mutex_lock(&monc
->mutex
);
316 dout("handle_monmap\n");
317 p
= msg
->front
.iov_base
;
318 end
= p
+ msg
->front
.iov_len
;
320 monmap
= ceph_monmap_decode(p
, end
);
321 if (IS_ERR(monmap
)) {
322 pr_err("problem decoding monmap, %d\n",
323 (int)PTR_ERR(monmap
));
327 if (ceph_check_fsid(monc
->client
, &monmap
->fsid
) < 0) {
332 client
->monc
.monmap
= monmap
;
336 mutex_unlock(&monc
->mutex
);
337 wake_up(&client
->mount_wq
);
343 static void handle_statfs_reply(struct ceph_mon_client
*monc
,
344 struct ceph_msg
*msg
)
346 struct ceph_mon_statfs_request
*req
;
347 struct ceph_mon_statfs_reply
*reply
= msg
->front
.iov_base
;
350 if (msg
->front
.iov_len
!= sizeof(*reply
))
352 tid
= le64_to_cpu(msg
->hdr
.tid
);
353 dout("handle_statfs_reply %p tid %llu\n", msg
, tid
);
355 mutex_lock(&monc
->mutex
);
356 req
= radix_tree_lookup(&monc
->statfs_request_tree
, tid
);
358 *req
->buf
= reply
->st
;
361 mutex_unlock(&monc
->mutex
);
363 complete(&req
->completion
);
367 pr_err("corrupt statfs reply, no tid\n");
372 * (re)send a statfs request
374 static int send_statfs(struct ceph_mon_client
*monc
,
375 struct ceph_mon_statfs_request
*req
)
377 struct ceph_msg
*msg
;
378 struct ceph_mon_statfs
*h
;
380 dout("send_statfs tid %llu\n", req
->tid
);
381 msg
= ceph_msg_new(CEPH_MSG_STATFS
, sizeof(*h
), 0, 0, NULL
);
385 msg
->hdr
.tid
= cpu_to_le64(req
->tid
);
386 h
= msg
->front
.iov_base
;
387 h
->monhdr
.have_version
= 0;
388 h
->monhdr
.session_mon
= cpu_to_le16(-1);
389 h
->monhdr
.session_mon_tid
= 0;
390 h
->fsid
= monc
->monmap
->fsid
;
391 ceph_con_send(monc
->con
, msg
);
396 * Do a synchronous statfs().
398 int ceph_monc_do_statfs(struct ceph_mon_client
*monc
, struct ceph_statfs
*buf
)
400 struct ceph_mon_statfs_request req
;
404 init_completion(&req
.completion
);
406 /* allocate memory for reply */
407 err
= ceph_msgpool_resv(&monc
->msgpool_statfs_reply
, 1);
411 /* register request */
412 mutex_lock(&monc
->mutex
);
413 req
.tid
= ++monc
->last_tid
;
414 req
.last_attempt
= jiffies
;
415 req
.delay
= BASE_DELAY_INTERVAL
;
416 if (radix_tree_insert(&monc
->statfs_request_tree
, req
.tid
, &req
) < 0) {
417 mutex_unlock(&monc
->mutex
);
418 pr_err("ENOMEM in do_statfs\n");
421 monc
->num_statfs_requests
++;
422 mutex_unlock(&monc
->mutex
);
424 /* send request and wait */
425 err
= send_statfs(monc
, &req
);
427 err
= wait_for_completion_interruptible(&req
.completion
);
429 mutex_lock(&monc
->mutex
);
430 radix_tree_delete(&monc
->statfs_request_tree
, req
.tid
);
431 monc
->num_statfs_requests
--;
432 ceph_msgpool_resv(&monc
->msgpool_statfs_reply
, -1);
433 mutex_unlock(&monc
->mutex
);
441 * Resend pending statfs requests.
443 static void __resend_statfs(struct ceph_mon_client
*monc
)
448 struct ceph_mon_statfs_request
*req
;
451 got
= radix_tree_gang_lookup(&monc
->statfs_request_tree
,
457 next_tid
= req
->tid
+ 1;
459 send_statfs(monc
, req
);
464 * Delayed work. If we haven't mounted yet, retry. Otherwise,
465 * renew/retry subscription as needed (in case it is timing out, or we
466 * got an ENOMEM). And keep the monitor connection alive.
468 static void delayed_work(struct work_struct
*work
)
470 struct ceph_mon_client
*monc
=
471 container_of(work
, struct ceph_mon_client
, delayed_work
.work
);
473 dout("monc delayed_work\n");
474 mutex_lock(&monc
->mutex
);
476 __close_session(monc
);
477 __open_session(monc
); /* continue hunting */
479 ceph_con_keepalive(monc
->con
);
480 if (monc
->auth
->ops
->is_authenticated(monc
->auth
))
481 __send_subscribe(monc
);
483 __schedule_delayed(monc
);
484 mutex_unlock(&monc
->mutex
);
488 * On startup, we build a temporary monmap populated with the IPs
489 * provided by mount(2).
491 static int build_initial_monmap(struct ceph_mon_client
*monc
)
493 struct ceph_mount_args
*args
= monc
->client
->mount_args
;
494 struct ceph_entity_addr
*mon_addr
= args
->mon_addr
;
495 int num_mon
= args
->num_mon
;
498 /* build initial monmap */
499 monc
->monmap
= kzalloc(sizeof(*monc
->monmap
) +
500 num_mon
*sizeof(monc
->monmap
->mon_inst
[0]),
504 for (i
= 0; i
< num_mon
; i
++) {
505 monc
->monmap
->mon_inst
[i
].addr
= mon_addr
[i
];
506 monc
->monmap
->mon_inst
[i
].addr
.erank
= 0;
507 monc
->monmap
->mon_inst
[i
].addr
.nonce
= 0;
508 monc
->monmap
->mon_inst
[i
].name
.type
=
509 CEPH_ENTITY_TYPE_MON
;
510 monc
->monmap
->mon_inst
[i
].name
.num
= cpu_to_le64(i
);
512 monc
->monmap
->num_mon
= num_mon
;
513 monc
->have_fsid
= false;
515 /* release addr memory */
516 kfree(args
->mon_addr
);
517 args
->mon_addr
= NULL
;
522 int ceph_monc_init(struct ceph_mon_client
*monc
, struct ceph_client
*cl
)
527 memset(monc
, 0, sizeof(*monc
));
530 mutex_init(&monc
->mutex
);
532 err
= build_initial_monmap(monc
);
539 monc
->auth
= ceph_auth_init(cl
->mount_args
->name
,
540 cl
->mount_args
->secret
);
541 if (IS_ERR(monc
->auth
))
542 return PTR_ERR(monc
->auth
);
543 monc
->auth
->want_keys
=
544 CEPH_ENTITY_TYPE_AUTH
| CEPH_ENTITY_TYPE_MON
|
545 CEPH_ENTITY_TYPE_OSD
| CEPH_ENTITY_TYPE_MDS
;
548 err
= ceph_msgpool_init(&monc
->msgpool_subscribe_ack
,
549 sizeof(struct ceph_mon_subscribe_ack
), 1, false);
552 err
= ceph_msgpool_init(&monc
->msgpool_statfs_reply
,
553 sizeof(struct ceph_mon_statfs_reply
), 0, false);
556 err
= ceph_msgpool_init(&monc
->msgpool_auth_reply
, 4096, 1, false);
560 monc
->m_auth
= ceph_msg_new(CEPH_MSG_AUTH
, 4096, 0, 0, NULL
);
561 if (IS_ERR(monc
->m_auth
)) {
562 err
= PTR_ERR(monc
->m_auth
);
568 monc
->hunting
= true;
569 monc
->sub_renew_after
= jiffies
;
572 INIT_DELAYED_WORK(&monc
->delayed_work
, delayed_work
);
573 INIT_RADIX_TREE(&monc
->statfs_request_tree
, GFP_NOFS
);
574 monc
->num_statfs_requests
= 0;
577 monc
->have_mdsmap
= 0;
578 monc
->have_osdmap
= 0;
579 monc
->want_next_osdmap
= 1;
583 ceph_msgpool_destroy(&monc
->msgpool_auth_reply
);
585 ceph_msgpool_destroy(&monc
->msgpool_subscribe_ack
);
587 ceph_msgpool_destroy(&monc
->msgpool_statfs_reply
);
594 void ceph_monc_stop(struct ceph_mon_client
*monc
)
597 cancel_delayed_work_sync(&monc
->delayed_work
);
599 mutex_lock(&monc
->mutex
);
600 __close_session(monc
);
602 monc
->con
->private = NULL
;
603 monc
->con
->ops
->put(monc
->con
);
606 mutex_unlock(&monc
->mutex
);
608 ceph_auth_destroy(monc
->auth
);
610 ceph_msg_put(monc
->m_auth
);
611 ceph_msgpool_destroy(&monc
->msgpool_subscribe_ack
);
612 ceph_msgpool_destroy(&monc
->msgpool_statfs_reply
);
613 ceph_msgpool_destroy(&monc
->msgpool_auth_reply
);
619 static void handle_auth_reply(struct ceph_mon_client
*monc
,
620 struct ceph_msg
*msg
)
624 mutex_lock(&monc
->mutex
);
625 ret
= ceph_handle_auth_reply(monc
->auth
, msg
->front
.iov_base
,
627 monc
->m_auth
->front
.iov_base
,
628 monc
->m_auth
->front_max
);
630 monc
->client
->mount_err
= ret
;
631 wake_up(&monc
->client
->mount_wq
);
632 } else if (ret
> 0) {
633 monc
->m_auth
->front
.iov_len
= ret
;
634 monc
->m_auth
->hdr
.front_len
= cpu_to_le32(ret
);
635 ceph_msg_get(monc
->m_auth
); /* keep our ref */
636 ceph_con_send(monc
->con
, monc
->m_auth
);
637 } else if (monc
->auth
->ops
->is_authenticated(monc
->auth
)) {
638 dout("authenticated, starting session\n");
640 monc
->client
->msgr
->inst
.name
.type
= CEPH_ENTITY_TYPE_CLIENT
;
641 monc
->client
->msgr
->inst
.name
.num
= monc
->auth
->global_id
;
643 __send_subscribe(monc
);
644 __resend_statfs(monc
);
646 mutex_unlock(&monc
->mutex
);
650 * handle incoming message
652 static void dispatch(struct ceph_connection
*con
, struct ceph_msg
*msg
)
654 struct ceph_mon_client
*monc
= con
->private;
655 int type
= le16_to_cpu(msg
->hdr
.type
);
661 case CEPH_MSG_AUTH_REPLY
:
662 handle_auth_reply(monc
, msg
);
665 case CEPH_MSG_MON_SUBSCRIBE_ACK
:
666 handle_subscribe_ack(monc
, msg
);
669 case CEPH_MSG_STATFS_REPLY
:
670 handle_statfs_reply(monc
, msg
);
673 case CEPH_MSG_MON_MAP
:
674 ceph_monc_handle_map(monc
, msg
);
677 case CEPH_MSG_MDS_MAP
:
678 ceph_mdsc_handle_map(&monc
->client
->mdsc
, msg
);
681 case CEPH_MSG_OSD_MAP
:
682 ceph_osdc_handle_map(&monc
->client
->osdc
, msg
);
686 pr_err("received unknown message type %d %s\n", type
,
687 ceph_msg_type_name(type
));
693 * Allocate memory for incoming message
695 static struct ceph_msg
*mon_alloc_msg(struct ceph_connection
*con
,
696 struct ceph_msg_header
*hdr
)
698 struct ceph_mon_client
*monc
= con
->private;
699 int type
= le16_to_cpu(hdr
->type
);
700 int front
= le32_to_cpu(hdr
->front_len
);
703 case CEPH_MSG_MON_SUBSCRIBE_ACK
:
704 return ceph_msgpool_get(&monc
->msgpool_subscribe_ack
, front
);
705 case CEPH_MSG_STATFS_REPLY
:
706 return ceph_msgpool_get(&monc
->msgpool_statfs_reply
, front
);
707 case CEPH_MSG_AUTH_REPLY
:
708 return ceph_msgpool_get(&monc
->msgpool_auth_reply
, front
);
710 return ceph_alloc_msg(con
, hdr
);
714 * If the monitor connection resets, pick a new monitor and resubmit
715 * any pending requests.
717 static void mon_fault(struct ceph_connection
*con
)
719 struct ceph_mon_client
*monc
= con
->private;
725 mutex_lock(&monc
->mutex
);
729 if (monc
->con
&& !monc
->hunting
)
730 pr_info("mon%d %s session lost, "
731 "hunting for new mon\n", monc
->cur_mon
,
732 pr_addr(&monc
->con
->peer_addr
.in_addr
));
734 __close_session(monc
);
735 if (!monc
->hunting
) {
737 monc
->hunting
= true;
738 __open_session(monc
);
740 /* already hunting, let's wait a bit */
741 __schedule_delayed(monc
);
744 mutex_unlock(&monc
->mutex
);
747 const static struct ceph_connection_operations mon_con_ops
= {
750 .dispatch
= dispatch
,
752 .alloc_msg
= mon_alloc_msg
,
753 .alloc_middle
= ceph_alloc_middle
,