Loading...
1// SPDX-License-Identifier: GPL-2.0
2#include <linux/ceph/ceph_debug.h>
3
4#include <linux/module.h>
5#include <linux/types.h>
6#include <linux/slab.h>
7#include <linux/random.h>
8#include <linux/sched.h>
9
10#include <linux/ceph/ceph_features.h>
11#include <linux/ceph/mon_client.h>
12#include <linux/ceph/libceph.h>
13#include <linux/ceph/debugfs.h>
14#include <linux/ceph/decode.h>
15#include <linux/ceph/auth.h>
16
17/*
18 * Interact with Ceph monitor cluster. Handle requests for new map
19 * versions, and periodically resend as needed. Also implement
20 * statfs() and umount().
21 *
22 * A small cluster of Ceph "monitors" are responsible for managing critical
23 * cluster configuration and state information. An odd number (e.g., 3, 5)
24 * of cmon daemons use a modified version of the Paxos part-time parliament
25 * algorithm to manage the MDS map (mds cluster membership), OSD map, and
26 * list of clients who have mounted the file system.
27 *
28 * We maintain an open, active session with a monitor at all times in order to
29 * receive timely MDSMap updates. We periodically send a keepalive byte on the
30 * TCP socket to ensure we detect a failure. If the connection does break, we
31 * randomly hunt for a new monitor. Once the connection is reestablished, we
32 * resend any outstanding requests.
33 */
34
35static const struct ceph_connection_operations mon_con_ops;
36
37static int __validate_auth(struct ceph_mon_client *monc);
38
39static int decode_mon_info(void **p, void *end, bool msgr2,
40 struct ceph_entity_addr *addr)
41{
42 void *mon_info_end;
43 u32 struct_len;
44 u8 struct_v;
45 int ret;
46
47 ret = ceph_start_decoding(p, end, 1, "mon_info_t", &struct_v,
48 &struct_len);
49 if (ret)
50 return ret;
51
52 mon_info_end = *p + struct_len;
53 ceph_decode_skip_string(p, end, e_inval); /* skip mon name */
54 ret = ceph_decode_entity_addrvec(p, end, msgr2, addr);
55 if (ret)
56 return ret;
57
58 *p = mon_info_end;
59 return 0;
60
61e_inval:
62 return -EINVAL;
63}
64
65/*
66 * Decode a monmap blob (e.g., during mount).
67 *
68 * Assume MonMap v3 (i.e. encoding with MONNAMES and MONENC).
69 */
70static struct ceph_monmap *ceph_monmap_decode(void **p, void *end, bool msgr2)
71{
72 struct ceph_monmap *monmap = NULL;
73 struct ceph_fsid fsid;
74 u32 struct_len;
75 int blob_len;
76 int num_mon;
77 u8 struct_v;
78 u32 epoch;
79 int ret;
80 int i;
81
82 ceph_decode_32_safe(p, end, blob_len, e_inval);
83 ceph_decode_need(p, end, blob_len, e_inval);
84
85 ret = ceph_start_decoding(p, end, 6, "monmap", &struct_v, &struct_len);
86 if (ret)
87 goto fail;
88
89 dout("%s struct_v %d\n", __func__, struct_v);
90 ceph_decode_copy_safe(p, end, &fsid, sizeof(fsid), e_inval);
91 ceph_decode_32_safe(p, end, epoch, e_inval);
92 if (struct_v >= 6) {
93 u32 feat_struct_len;
94 u8 feat_struct_v;
95
96 *p += sizeof(struct ceph_timespec); /* skip last_changed */
97 *p += sizeof(struct ceph_timespec); /* skip created */
98
99 ret = ceph_start_decoding(p, end, 1, "mon_feature_t",
100 &feat_struct_v, &feat_struct_len);
101 if (ret)
102 goto fail;
103
104 *p += feat_struct_len; /* skip persistent_features */
105
106 ret = ceph_start_decoding(p, end, 1, "mon_feature_t",
107 &feat_struct_v, &feat_struct_len);
108 if (ret)
109 goto fail;
110
111 *p += feat_struct_len; /* skip optional_features */
112 }
113 ceph_decode_32_safe(p, end, num_mon, e_inval);
114
115 dout("%s fsid %pU epoch %u num_mon %d\n", __func__, &fsid, epoch,
116 num_mon);
117 if (num_mon > CEPH_MAX_MON)
118 goto e_inval;
119
120 monmap = kmalloc(struct_size(monmap, mon_inst, num_mon), GFP_NOIO);
121 if (!monmap) {
122 ret = -ENOMEM;
123 goto fail;
124 }
125 monmap->fsid = fsid;
126 monmap->epoch = epoch;
127 monmap->num_mon = num_mon;
128
129 /* legacy_mon_addr map or mon_info map */
130 for (i = 0; i < num_mon; i++) {
131 struct ceph_entity_inst *inst = &monmap->mon_inst[i];
132
133 ceph_decode_skip_string(p, end, e_inval); /* skip mon name */
134 inst->name.type = CEPH_ENTITY_TYPE_MON;
135 inst->name.num = cpu_to_le64(i);
136
137 if (struct_v >= 6)
138 ret = decode_mon_info(p, end, msgr2, &inst->addr);
139 else
140 ret = ceph_decode_entity_addr(p, end, &inst->addr);
141 if (ret)
142 goto fail;
143
144 dout("%s mon%d addr %s\n", __func__, i,
145 ceph_pr_addr(&inst->addr));
146 }
147
148 return monmap;
149
150e_inval:
151 ret = -EINVAL;
152fail:
153 kfree(monmap);
154 return ERR_PTR(ret);
155}
156
157/*
158 * return true if *addr is included in the monmap.
159 */
160int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr)
161{
162 int i;
163
164 for (i = 0; i < m->num_mon; i++) {
165 if (ceph_addr_equal_no_type(addr, &m->mon_inst[i].addr))
166 return 1;
167 }
168
169 return 0;
170}
171
172/*
173 * Send an auth request.
174 */
175static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len)
176{
177 monc->pending_auth = 1;
178 monc->m_auth->front.iov_len = len;
179 monc->m_auth->hdr.front_len = cpu_to_le32(len);
180 ceph_msg_revoke(monc->m_auth);
181 ceph_msg_get(monc->m_auth); /* keep our ref */
182 ceph_con_send(&monc->con, monc->m_auth);
183}
184
185/*
186 * Close monitor session, if any.
187 */
188static void __close_session(struct ceph_mon_client *monc)
189{
190 dout("__close_session closing mon%d\n", monc->cur_mon);
191 ceph_msg_revoke(monc->m_auth);
192 ceph_msg_revoke_incoming(monc->m_auth_reply);
193 ceph_msg_revoke(monc->m_subscribe);
194 ceph_msg_revoke_incoming(monc->m_subscribe_ack);
195 ceph_con_close(&monc->con);
196
197 monc->pending_auth = 0;
198 ceph_auth_reset(monc->auth);
199}
200
201/*
202 * Pick a new monitor at random and set cur_mon. If we are repicking
203 * (i.e. cur_mon is already set), be sure to pick a different one.
204 */
205static void pick_new_mon(struct ceph_mon_client *monc)
206{
207 int old_mon = monc->cur_mon;
208
209 BUG_ON(monc->monmap->num_mon < 1);
210
211 if (monc->monmap->num_mon == 1) {
212 monc->cur_mon = 0;
213 } else {
214 int max = monc->monmap->num_mon;
215 int o = -1;
216 int n;
217
218 if (monc->cur_mon >= 0) {
219 if (monc->cur_mon < monc->monmap->num_mon)
220 o = monc->cur_mon;
221 if (o >= 0)
222 max--;
223 }
224
225 n = get_random_u32_below(max);
226 if (o >= 0 && n >= o)
227 n++;
228
229 monc->cur_mon = n;
230 }
231
232 dout("%s mon%d -> mon%d out of %d mons\n", __func__, old_mon,
233 monc->cur_mon, monc->monmap->num_mon);
234}
235
236/*
237 * Open a session with a new monitor.
238 */
239static void __open_session(struct ceph_mon_client *monc)
240{
241 int ret;
242
243 pick_new_mon(monc);
244
245 monc->hunting = true;
246 if (monc->had_a_connection) {
247 monc->hunt_mult *= CEPH_MONC_HUNT_BACKOFF;
248 if (monc->hunt_mult > CEPH_MONC_HUNT_MAX_MULT)
249 monc->hunt_mult = CEPH_MONC_HUNT_MAX_MULT;
250 }
251
252 monc->sub_renew_after = jiffies; /* i.e., expired */
253 monc->sub_renew_sent = 0;
254
255 dout("%s opening mon%d\n", __func__, monc->cur_mon);
256 ceph_con_open(&monc->con, CEPH_ENTITY_TYPE_MON, monc->cur_mon,
257 &monc->monmap->mon_inst[monc->cur_mon].addr);
258
259 /*
260 * Queue a keepalive to ensure that in case of an early fault
261 * the messenger doesn't put us into STANDBY state and instead
262 * retries. This also ensures that our timestamp is valid by
263 * the time we finish hunting and delayed_work() checks it.
264 */
265 ceph_con_keepalive(&monc->con);
266 if (ceph_msgr2(monc->client)) {
267 monc->pending_auth = 1;
268 return;
269 }
270
271 /* initiate authentication handshake */
272 ret = ceph_auth_build_hello(monc->auth,
273 monc->m_auth->front.iov_base,
274 monc->m_auth->front_alloc_len);
275 BUG_ON(ret <= 0);
276 __send_prepared_auth_request(monc, ret);
277}
278
279static void reopen_session(struct ceph_mon_client *monc)
280{
281 if (!monc->hunting)
282 pr_info("mon%d %s session lost, hunting for new mon\n",
283 monc->cur_mon, ceph_pr_addr(&monc->con.peer_addr));
284
285 __close_session(monc);
286 __open_session(monc);
287}
288
289void ceph_monc_reopen_session(struct ceph_mon_client *monc)
290{
291 mutex_lock(&monc->mutex);
292 reopen_session(monc);
293 mutex_unlock(&monc->mutex);
294}
295
296static void un_backoff(struct ceph_mon_client *monc)
297{
298 monc->hunt_mult /= 2; /* reduce by 50% */
299 if (monc->hunt_mult < 1)
300 monc->hunt_mult = 1;
301 dout("%s hunt_mult now %d\n", __func__, monc->hunt_mult);
302}
303
304/*
305 * Reschedule delayed work timer.
306 */
307static void __schedule_delayed(struct ceph_mon_client *monc)
308{
309 unsigned long delay;
310
311 if (monc->hunting)
312 delay = CEPH_MONC_HUNT_INTERVAL * monc->hunt_mult;
313 else
314 delay = CEPH_MONC_PING_INTERVAL;
315
316 dout("__schedule_delayed after %lu\n", delay);
317 mod_delayed_work(system_wq, &monc->delayed_work,
318 round_jiffies_relative(delay));
319}
320
321const char *ceph_sub_str[] = {
322 [CEPH_SUB_MONMAP] = "monmap",
323 [CEPH_SUB_OSDMAP] = "osdmap",
324 [CEPH_SUB_FSMAP] = "fsmap.user",
325 [CEPH_SUB_MDSMAP] = "mdsmap",
326};
327
328/*
329 * Send subscribe request for one or more maps, according to
330 * monc->subs.
331 */
332static void __send_subscribe(struct ceph_mon_client *monc)
333{
334 struct ceph_msg *msg = monc->m_subscribe;
335 void *p = msg->front.iov_base;
336 void *const end = p + msg->front_alloc_len;
337 int num = 0;
338 int i;
339
340 dout("%s sent %lu\n", __func__, monc->sub_renew_sent);
341
342 BUG_ON(monc->cur_mon < 0);
343
344 if (!monc->sub_renew_sent)
345 monc->sub_renew_sent = jiffies | 1; /* never 0 */
346
347 msg->hdr.version = cpu_to_le16(2);
348
349 for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
350 if (monc->subs[i].want)
351 num++;
352 }
353 BUG_ON(num < 1); /* monmap sub is always there */
354 ceph_encode_32(&p, num);
355 for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
356 char buf[32];
357 int len;
358
359 if (!monc->subs[i].want)
360 continue;
361
362 len = sprintf(buf, "%s", ceph_sub_str[i]);
363 if (i == CEPH_SUB_MDSMAP &&
364 monc->fs_cluster_id != CEPH_FS_CLUSTER_ID_NONE)
365 len += sprintf(buf + len, ".%d", monc->fs_cluster_id);
366
367 dout("%s %s start %llu flags 0x%x\n", __func__, buf,
368 le64_to_cpu(monc->subs[i].item.start),
369 monc->subs[i].item.flags);
370 ceph_encode_string(&p, end, buf, len);
371 memcpy(p, &monc->subs[i].item, sizeof(monc->subs[i].item));
372 p += sizeof(monc->subs[i].item);
373 }
374
375 BUG_ON(p > end);
376 msg->front.iov_len = p - msg->front.iov_base;
377 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
378 ceph_msg_revoke(msg);
379 ceph_con_send(&monc->con, ceph_msg_get(msg));
380}
381
382static void handle_subscribe_ack(struct ceph_mon_client *monc,
383 struct ceph_msg *msg)
384{
385 unsigned int seconds;
386 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
387
388 if (msg->front.iov_len < sizeof(*h))
389 goto bad;
390 seconds = le32_to_cpu(h->duration);
391
392 mutex_lock(&monc->mutex);
393 if (monc->sub_renew_sent) {
394 /*
395 * This is only needed for legacy (infernalis or older)
396 * MONs -- see delayed_work().
397 */
398 monc->sub_renew_after = monc->sub_renew_sent +
399 (seconds >> 1) * HZ - 1;
400 dout("%s sent %lu duration %d renew after %lu\n", __func__,
401 monc->sub_renew_sent, seconds, monc->sub_renew_after);
402 monc->sub_renew_sent = 0;
403 } else {
404 dout("%s sent %lu renew after %lu, ignoring\n", __func__,
405 monc->sub_renew_sent, monc->sub_renew_after);
406 }
407 mutex_unlock(&monc->mutex);
408 return;
409bad:
410 pr_err("got corrupt subscribe-ack msg\n");
411 ceph_msg_dump(msg);
412}
413
414/*
415 * Register interest in a map
416 *
417 * @sub: one of CEPH_SUB_*
418 * @epoch: X for "every map since X", or 0 for "just the latest"
419 */
420static bool __ceph_monc_want_map(struct ceph_mon_client *monc, int sub,
421 u32 epoch, bool continuous)
422{
423 __le64 start = cpu_to_le64(epoch);
424 u8 flags = !continuous ? CEPH_SUBSCRIBE_ONETIME : 0;
425
426 dout("%s %s epoch %u continuous %d\n", __func__, ceph_sub_str[sub],
427 epoch, continuous);
428
429 if (monc->subs[sub].want &&
430 monc->subs[sub].item.start == start &&
431 monc->subs[sub].item.flags == flags)
432 return false;
433
434 monc->subs[sub].item.start = start;
435 monc->subs[sub].item.flags = flags;
436 monc->subs[sub].want = true;
437
438 return true;
439}
440
441bool ceph_monc_want_map(struct ceph_mon_client *monc, int sub, u32 epoch,
442 bool continuous)
443{
444 bool need_request;
445
446 mutex_lock(&monc->mutex);
447 need_request = __ceph_monc_want_map(monc, sub, epoch, continuous);
448 mutex_unlock(&monc->mutex);
449
450 return need_request;
451}
452EXPORT_SYMBOL(ceph_monc_want_map);
453
454/*
455 * Keep track of which maps we have
456 *
457 * @sub: one of CEPH_SUB_*
458 */
459static void __ceph_monc_got_map(struct ceph_mon_client *monc, int sub,
460 u32 epoch)
461{
462 dout("%s %s epoch %u\n", __func__, ceph_sub_str[sub], epoch);
463
464 if (monc->subs[sub].want) {
465 if (monc->subs[sub].item.flags & CEPH_SUBSCRIBE_ONETIME)
466 monc->subs[sub].want = false;
467 else
468 monc->subs[sub].item.start = cpu_to_le64(epoch + 1);
469 }
470
471 monc->subs[sub].have = epoch;
472}
473
474void ceph_monc_got_map(struct ceph_mon_client *monc, int sub, u32 epoch)
475{
476 mutex_lock(&monc->mutex);
477 __ceph_monc_got_map(monc, sub, epoch);
478 mutex_unlock(&monc->mutex);
479}
480EXPORT_SYMBOL(ceph_monc_got_map);
481
482void ceph_monc_renew_subs(struct ceph_mon_client *monc)
483{
484 mutex_lock(&monc->mutex);
485 __send_subscribe(monc);
486 mutex_unlock(&monc->mutex);
487}
488EXPORT_SYMBOL(ceph_monc_renew_subs);
489
490/*
491 * Wait for an osdmap with a given epoch.
492 *
493 * @epoch: epoch to wait for
494 * @timeout: in jiffies, 0 means "wait forever"
495 */
496int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
497 unsigned long timeout)
498{
499 unsigned long started = jiffies;
500 long ret;
501
502 mutex_lock(&monc->mutex);
503 while (monc->subs[CEPH_SUB_OSDMAP].have < epoch) {
504 mutex_unlock(&monc->mutex);
505
506 if (timeout && time_after_eq(jiffies, started + timeout))
507 return -ETIMEDOUT;
508
509 ret = wait_event_interruptible_timeout(monc->client->auth_wq,
510 monc->subs[CEPH_SUB_OSDMAP].have >= epoch,
511 ceph_timeout_jiffies(timeout));
512 if (ret < 0)
513 return ret;
514
515 mutex_lock(&monc->mutex);
516 }
517
518 mutex_unlock(&monc->mutex);
519 return 0;
520}
521EXPORT_SYMBOL(ceph_monc_wait_osdmap);
522
523/*
524 * Open a session with a random monitor. Request monmap and osdmap,
525 * which are waited upon in __ceph_open_session().
526 */
527int ceph_monc_open_session(struct ceph_mon_client *monc)
528{
529 mutex_lock(&monc->mutex);
530 __ceph_monc_want_map(monc, CEPH_SUB_MONMAP, 0, true);
531 __ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, 0, false);
532 __open_session(monc);
533 __schedule_delayed(monc);
534 mutex_unlock(&monc->mutex);
535 return 0;
536}
537EXPORT_SYMBOL(ceph_monc_open_session);
538
539static void ceph_monc_handle_map(struct ceph_mon_client *monc,
540 struct ceph_msg *msg)
541{
542 struct ceph_client *client = monc->client;
543 struct ceph_monmap *monmap;
544 void *p, *end;
545
546 mutex_lock(&monc->mutex);
547
548 dout("handle_monmap\n");
549 p = msg->front.iov_base;
550 end = p + msg->front.iov_len;
551
552 monmap = ceph_monmap_decode(&p, end, ceph_msgr2(client));
553 if (IS_ERR(monmap)) {
554 pr_err("problem decoding monmap, %d\n",
555 (int)PTR_ERR(monmap));
556 ceph_msg_dump(msg);
557 goto out;
558 }
559
560 if (ceph_check_fsid(client, &monmap->fsid) < 0) {
561 kfree(monmap);
562 goto out;
563 }
564
565 kfree(monc->monmap);
566 monc->monmap = monmap;
567
568 __ceph_monc_got_map(monc, CEPH_SUB_MONMAP, monc->monmap->epoch);
569 client->have_fsid = true;
570
571out:
572 mutex_unlock(&monc->mutex);
573 wake_up_all(&client->auth_wq);
574}
575
576/*
577 * generic requests (currently statfs, mon_get_version)
578 */
579DEFINE_RB_FUNCS(generic_request, struct ceph_mon_generic_request, tid, node)
580
581static void release_generic_request(struct kref *kref)
582{
583 struct ceph_mon_generic_request *req =
584 container_of(kref, struct ceph_mon_generic_request, kref);
585
586 dout("%s greq %p request %p reply %p\n", __func__, req, req->request,
587 req->reply);
588 WARN_ON(!RB_EMPTY_NODE(&req->node));
589
590 if (req->reply)
591 ceph_msg_put(req->reply);
592 if (req->request)
593 ceph_msg_put(req->request);
594
595 kfree(req);
596}
597
598static void put_generic_request(struct ceph_mon_generic_request *req)
599{
600 if (req)
601 kref_put(&req->kref, release_generic_request);
602}
603
604static void get_generic_request(struct ceph_mon_generic_request *req)
605{
606 kref_get(&req->kref);
607}
608
609static struct ceph_mon_generic_request *
610alloc_generic_request(struct ceph_mon_client *monc, gfp_t gfp)
611{
612 struct ceph_mon_generic_request *req;
613
614 req = kzalloc(sizeof(*req), gfp);
615 if (!req)
616 return NULL;
617
618 req->monc = monc;
619 kref_init(&req->kref);
620 RB_CLEAR_NODE(&req->node);
621 init_completion(&req->completion);
622
623 dout("%s greq %p\n", __func__, req);
624 return req;
625}
626
627static void register_generic_request(struct ceph_mon_generic_request *req)
628{
629 struct ceph_mon_client *monc = req->monc;
630
631 WARN_ON(req->tid);
632
633 get_generic_request(req);
634 req->tid = ++monc->last_tid;
635 insert_generic_request(&monc->generic_request_tree, req);
636}
637
638static void send_generic_request(struct ceph_mon_client *monc,
639 struct ceph_mon_generic_request *req)
640{
641 WARN_ON(!req->tid);
642
643 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
644 req->request->hdr.tid = cpu_to_le64(req->tid);
645 ceph_con_send(&monc->con, ceph_msg_get(req->request));
646}
647
648static void __finish_generic_request(struct ceph_mon_generic_request *req)
649{
650 struct ceph_mon_client *monc = req->monc;
651
652 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
653 erase_generic_request(&monc->generic_request_tree, req);
654
655 ceph_msg_revoke(req->request);
656 ceph_msg_revoke_incoming(req->reply);
657}
658
659static void finish_generic_request(struct ceph_mon_generic_request *req)
660{
661 __finish_generic_request(req);
662 put_generic_request(req);
663}
664
665static void complete_generic_request(struct ceph_mon_generic_request *req)
666{
667 if (req->complete_cb)
668 req->complete_cb(req);
669 else
670 complete_all(&req->completion);
671 put_generic_request(req);
672}
673
674static void cancel_generic_request(struct ceph_mon_generic_request *req)
675{
676 struct ceph_mon_client *monc = req->monc;
677 struct ceph_mon_generic_request *lookup_req;
678
679 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
680
681 mutex_lock(&monc->mutex);
682 lookup_req = lookup_generic_request(&monc->generic_request_tree,
683 req->tid);
684 if (lookup_req) {
685 WARN_ON(lookup_req != req);
686 finish_generic_request(req);
687 }
688
689 mutex_unlock(&monc->mutex);
690}
691
692static int wait_generic_request(struct ceph_mon_generic_request *req)
693{
694 int ret;
695
696 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
697 ret = wait_for_completion_interruptible(&req->completion);
698 if (ret)
699 cancel_generic_request(req);
700 else
701 ret = req->result; /* completed */
702
703 return ret;
704}
705
706static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
707 struct ceph_msg_header *hdr,
708 int *skip)
709{
710 struct ceph_mon_client *monc = con->private;
711 struct ceph_mon_generic_request *req;
712 u64 tid = le64_to_cpu(hdr->tid);
713 struct ceph_msg *m;
714
715 mutex_lock(&monc->mutex);
716 req = lookup_generic_request(&monc->generic_request_tree, tid);
717 if (!req) {
718 dout("get_generic_reply %lld dne\n", tid);
719 *skip = 1;
720 m = NULL;
721 } else {
722 dout("get_generic_reply %lld got %p\n", tid, req->reply);
723 *skip = 0;
724 m = ceph_msg_get(req->reply);
725 /*
726 * we don't need to track the connection reading into
727 * this reply because we only have one open connection
728 * at a time, ever.
729 */
730 }
731 mutex_unlock(&monc->mutex);
732 return m;
733}
734
735/*
736 * statfs
737 */
738static void handle_statfs_reply(struct ceph_mon_client *monc,
739 struct ceph_msg *msg)
740{
741 struct ceph_mon_generic_request *req;
742 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
743 u64 tid = le64_to_cpu(msg->hdr.tid);
744
745 dout("%s msg %p tid %llu\n", __func__, msg, tid);
746
747 if (msg->front.iov_len != sizeof(*reply))
748 goto bad;
749
750 mutex_lock(&monc->mutex);
751 req = lookup_generic_request(&monc->generic_request_tree, tid);
752 if (!req) {
753 mutex_unlock(&monc->mutex);
754 return;
755 }
756
757 req->result = 0;
758 *req->u.st = reply->st; /* struct */
759 __finish_generic_request(req);
760 mutex_unlock(&monc->mutex);
761
762 complete_generic_request(req);
763 return;
764
765bad:
766 pr_err("corrupt statfs reply, tid %llu\n", tid);
767 ceph_msg_dump(msg);
768}
769
770/*
771 * Do a synchronous statfs().
772 */
773int ceph_monc_do_statfs(struct ceph_mon_client *monc, u64 data_pool,
774 struct ceph_statfs *buf)
775{
776 struct ceph_mon_generic_request *req;
777 struct ceph_mon_statfs *h;
778 int ret = -ENOMEM;
779
780 req = alloc_generic_request(monc, GFP_NOFS);
781 if (!req)
782 goto out;
783
784 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
785 true);
786 if (!req->request)
787 goto out;
788
789 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 64, GFP_NOFS, true);
790 if (!req->reply)
791 goto out;
792
793 req->u.st = buf;
794 req->request->hdr.version = cpu_to_le16(2);
795
796 mutex_lock(&monc->mutex);
797 register_generic_request(req);
798 /* fill out request */
799 h = req->request->front.iov_base;
800 h->monhdr.have_version = 0;
801 h->monhdr.session_mon = cpu_to_le16(-1);
802 h->monhdr.session_mon_tid = 0;
803 h->fsid = monc->monmap->fsid;
804 h->contains_data_pool = (data_pool != CEPH_NOPOOL);
805 h->data_pool = cpu_to_le64(data_pool);
806 send_generic_request(monc, req);
807 mutex_unlock(&monc->mutex);
808
809 ret = wait_generic_request(req);
810out:
811 put_generic_request(req);
812 return ret;
813}
814EXPORT_SYMBOL(ceph_monc_do_statfs);
815
816static void handle_get_version_reply(struct ceph_mon_client *monc,
817 struct ceph_msg *msg)
818{
819 struct ceph_mon_generic_request *req;
820 u64 tid = le64_to_cpu(msg->hdr.tid);
821 void *p = msg->front.iov_base;
822 void *end = p + msg->front_alloc_len;
823 u64 handle;
824
825 dout("%s msg %p tid %llu\n", __func__, msg, tid);
826
827 ceph_decode_need(&p, end, 2*sizeof(u64), bad);
828 handle = ceph_decode_64(&p);
829 if (tid != 0 && tid != handle)
830 goto bad;
831
832 mutex_lock(&monc->mutex);
833 req = lookup_generic_request(&monc->generic_request_tree, handle);
834 if (!req) {
835 mutex_unlock(&monc->mutex);
836 return;
837 }
838
839 req->result = 0;
840 req->u.newest = ceph_decode_64(&p);
841 __finish_generic_request(req);
842 mutex_unlock(&monc->mutex);
843
844 complete_generic_request(req);
845 return;
846
847bad:
848 pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
849 ceph_msg_dump(msg);
850}
851
852static struct ceph_mon_generic_request *
853__ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
854 ceph_monc_callback_t cb, u64 private_data)
855{
856 struct ceph_mon_generic_request *req;
857
858 req = alloc_generic_request(monc, GFP_NOIO);
859 if (!req)
860 goto err_put_req;
861
862 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
863 sizeof(u64) + sizeof(u32) + strlen(what),
864 GFP_NOIO, true);
865 if (!req->request)
866 goto err_put_req;
867
868 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 32, GFP_NOIO,
869 true);
870 if (!req->reply)
871 goto err_put_req;
872
873 req->complete_cb = cb;
874 req->private_data = private_data;
875
876 mutex_lock(&monc->mutex);
877 register_generic_request(req);
878 {
879 void *p = req->request->front.iov_base;
880 void *const end = p + req->request->front_alloc_len;
881
882 ceph_encode_64(&p, req->tid); /* handle */
883 ceph_encode_string(&p, end, what, strlen(what));
884 WARN_ON(p != end);
885 }
886 send_generic_request(monc, req);
887 mutex_unlock(&monc->mutex);
888
889 return req;
890
891err_put_req:
892 put_generic_request(req);
893 return ERR_PTR(-ENOMEM);
894}
895
896/*
897 * Send MMonGetVersion and wait for the reply.
898 *
899 * @what: one of "mdsmap", "osdmap" or "monmap"
900 */
901int ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
902 u64 *newest)
903{
904 struct ceph_mon_generic_request *req;
905 int ret;
906
907 req = __ceph_monc_get_version(monc, what, NULL, 0);
908 if (IS_ERR(req))
909 return PTR_ERR(req);
910
911 ret = wait_generic_request(req);
912 if (!ret)
913 *newest = req->u.newest;
914
915 put_generic_request(req);
916 return ret;
917}
918EXPORT_SYMBOL(ceph_monc_get_version);
919
920/*
921 * Send MMonGetVersion,
922 *
923 * @what: one of "mdsmap", "osdmap" or "monmap"
924 */
925int ceph_monc_get_version_async(struct ceph_mon_client *monc, const char *what,
926 ceph_monc_callback_t cb, u64 private_data)
927{
928 struct ceph_mon_generic_request *req;
929
930 req = __ceph_monc_get_version(monc, what, cb, private_data);
931 if (IS_ERR(req))
932 return PTR_ERR(req);
933
934 put_generic_request(req);
935 return 0;
936}
937EXPORT_SYMBOL(ceph_monc_get_version_async);
938
939static void handle_command_ack(struct ceph_mon_client *monc,
940 struct ceph_msg *msg)
941{
942 struct ceph_mon_generic_request *req;
943 void *p = msg->front.iov_base;
944 void *const end = p + msg->front_alloc_len;
945 u64 tid = le64_to_cpu(msg->hdr.tid);
946
947 dout("%s msg %p tid %llu\n", __func__, msg, tid);
948
949 ceph_decode_need(&p, end, sizeof(struct ceph_mon_request_header) +
950 sizeof(u32), bad);
951 p += sizeof(struct ceph_mon_request_header);
952
953 mutex_lock(&monc->mutex);
954 req = lookup_generic_request(&monc->generic_request_tree, tid);
955 if (!req) {
956 mutex_unlock(&monc->mutex);
957 return;
958 }
959
960 req->result = ceph_decode_32(&p);
961 __finish_generic_request(req);
962 mutex_unlock(&monc->mutex);
963
964 complete_generic_request(req);
965 return;
966
967bad:
968 pr_err("corrupt mon_command ack, tid %llu\n", tid);
969 ceph_msg_dump(msg);
970}
971
972static __printf(2, 0)
973int do_mon_command_vargs(struct ceph_mon_client *monc, const char *fmt,
974 va_list ap)
975{
976 struct ceph_mon_generic_request *req;
977 struct ceph_mon_command *h;
978 int ret = -ENOMEM;
979 int len;
980
981 req = alloc_generic_request(monc, GFP_NOIO);
982 if (!req)
983 goto out;
984
985 req->request = ceph_msg_new(CEPH_MSG_MON_COMMAND, 256, GFP_NOIO, true);
986 if (!req->request)
987 goto out;
988
989 req->reply = ceph_msg_new(CEPH_MSG_MON_COMMAND_ACK, 512, GFP_NOIO,
990 true);
991 if (!req->reply)
992 goto out;
993
994 mutex_lock(&monc->mutex);
995 register_generic_request(req);
996 h = req->request->front.iov_base;
997 h->monhdr.have_version = 0;
998 h->monhdr.session_mon = cpu_to_le16(-1);
999 h->monhdr.session_mon_tid = 0;
1000 h->fsid = monc->monmap->fsid;
1001 h->num_strs = cpu_to_le32(1);
1002 len = vsprintf(h->str, fmt, ap);
1003 h->str_len = cpu_to_le32(len);
1004 send_generic_request(monc, req);
1005 mutex_unlock(&monc->mutex);
1006
1007 ret = wait_generic_request(req);
1008out:
1009 put_generic_request(req);
1010 return ret;
1011}
1012
1013static __printf(2, 3)
1014int do_mon_command(struct ceph_mon_client *monc, const char *fmt, ...)
1015{
1016 va_list ap;
1017 int ret;
1018
1019 va_start(ap, fmt);
1020 ret = do_mon_command_vargs(monc, fmt, ap);
1021 va_end(ap);
1022 return ret;
1023}
1024
1025int ceph_monc_blocklist_add(struct ceph_mon_client *monc,
1026 struct ceph_entity_addr *client_addr)
1027{
1028 int ret;
1029
1030 ret = do_mon_command(monc,
1031 "{ \"prefix\": \"osd blocklist\", \
1032 \"blocklistop\": \"add\", \
1033 \"addr\": \"%pISpc/%u\" }",
1034 &client_addr->in_addr,
1035 le32_to_cpu(client_addr->nonce));
1036 if (ret == -EINVAL) {
1037 /*
1038 * The monitor returns EINVAL on an unrecognized command.
1039 * Try the legacy command -- it is exactly the same except
1040 * for the name.
1041 */
1042 ret = do_mon_command(monc,
1043 "{ \"prefix\": \"osd blacklist\", \
1044 \"blacklistop\": \"add\", \
1045 \"addr\": \"%pISpc/%u\" }",
1046 &client_addr->in_addr,
1047 le32_to_cpu(client_addr->nonce));
1048 }
1049 if (ret)
1050 return ret;
1051
1052 /*
1053 * Make sure we have the osdmap that includes the blocklist
1054 * entry. This is needed to ensure that the OSDs pick up the
1055 * new blocklist before processing any future requests from
1056 * this client.
1057 */
1058 return ceph_wait_for_latest_osdmap(monc->client, 0);
1059}
1060EXPORT_SYMBOL(ceph_monc_blocklist_add);
1061
1062/*
1063 * Resend pending generic requests.
1064 */
1065static void __resend_generic_request(struct ceph_mon_client *monc)
1066{
1067 struct ceph_mon_generic_request *req;
1068 struct rb_node *p;
1069
1070 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
1071 req = rb_entry(p, struct ceph_mon_generic_request, node);
1072 ceph_msg_revoke(req->request);
1073 ceph_msg_revoke_incoming(req->reply);
1074 ceph_con_send(&monc->con, ceph_msg_get(req->request));
1075 }
1076}
1077
1078/*
1079 * Delayed work. If we haven't mounted yet, retry. Otherwise,
1080 * renew/retry subscription as needed (in case it is timing out, or we
1081 * got an ENOMEM). And keep the monitor connection alive.
1082 */
1083static void delayed_work(struct work_struct *work)
1084{
1085 struct ceph_mon_client *monc =
1086 container_of(work, struct ceph_mon_client, delayed_work.work);
1087
1088 mutex_lock(&monc->mutex);
1089 dout("%s mon%d\n", __func__, monc->cur_mon);
1090 if (monc->cur_mon < 0) {
1091 goto out;
1092 }
1093
1094 if (monc->hunting) {
1095 dout("%s continuing hunt\n", __func__);
1096 reopen_session(monc);
1097 } else {
1098 int is_auth = ceph_auth_is_authenticated(monc->auth);
1099
1100 dout("%s is_authed %d\n", __func__, is_auth);
1101 if (ceph_con_keepalive_expired(&monc->con,
1102 CEPH_MONC_PING_TIMEOUT)) {
1103 dout("monc keepalive timeout\n");
1104 is_auth = 0;
1105 reopen_session(monc);
1106 }
1107
1108 if (!monc->hunting) {
1109 ceph_con_keepalive(&monc->con);
1110 __validate_auth(monc);
1111 un_backoff(monc);
1112 }
1113
1114 if (is_auth &&
1115 !(monc->con.peer_features & CEPH_FEATURE_MON_STATEFUL_SUB)) {
1116 unsigned long now = jiffies;
1117
1118 dout("%s renew subs? now %lu renew after %lu\n",
1119 __func__, now, monc->sub_renew_after);
1120 if (time_after_eq(now, monc->sub_renew_after))
1121 __send_subscribe(monc);
1122 }
1123 }
1124 __schedule_delayed(monc);
1125
1126out:
1127 mutex_unlock(&monc->mutex);
1128}
1129
1130/*
1131 * On startup, we build a temporary monmap populated with the IPs
1132 * provided by mount(2).
1133 */
1134static int build_initial_monmap(struct ceph_mon_client *monc)
1135{
1136 __le32 my_type = ceph_msgr2(monc->client) ?
1137 CEPH_ENTITY_ADDR_TYPE_MSGR2 : CEPH_ENTITY_ADDR_TYPE_LEGACY;
1138 struct ceph_options *opt = monc->client->options;
1139 int num_mon = opt->num_mon;
1140 int i;
1141
1142 /* build initial monmap */
1143 monc->monmap = kzalloc(struct_size(monc->monmap, mon_inst, num_mon),
1144 GFP_KERNEL);
1145 if (!monc->monmap)
1146 return -ENOMEM;
1147 monc->monmap->num_mon = num_mon;
1148
1149 for (i = 0; i < num_mon; i++) {
1150 struct ceph_entity_inst *inst = &monc->monmap->mon_inst[i];
1151
1152 memcpy(&inst->addr.in_addr, &opt->mon_addr[i].in_addr,
1153 sizeof(inst->addr.in_addr));
1154 inst->addr.type = my_type;
1155 inst->addr.nonce = 0;
1156 inst->name.type = CEPH_ENTITY_TYPE_MON;
1157 inst->name.num = cpu_to_le64(i);
1158 }
1159 return 0;
1160}
1161
1162int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
1163{
1164 int err;
1165
1166 dout("init\n");
1167 memset(monc, 0, sizeof(*monc));
1168 monc->client = cl;
1169 mutex_init(&monc->mutex);
1170
1171 err = build_initial_monmap(monc);
1172 if (err)
1173 goto out;
1174
1175 /* connection */
1176 /* authentication */
1177 monc->auth = ceph_auth_init(cl->options->name, cl->options->key,
1178 cl->options->con_modes);
1179 if (IS_ERR(monc->auth)) {
1180 err = PTR_ERR(monc->auth);
1181 goto out_monmap;
1182 }
1183 monc->auth->want_keys =
1184 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
1185 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
1186
1187 /* msgs */
1188 err = -ENOMEM;
1189 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
1190 sizeof(struct ceph_mon_subscribe_ack),
1191 GFP_KERNEL, true);
1192 if (!monc->m_subscribe_ack)
1193 goto out_auth;
1194
1195 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 128,
1196 GFP_KERNEL, true);
1197 if (!monc->m_subscribe)
1198 goto out_subscribe_ack;
1199
1200 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096,
1201 GFP_KERNEL, true);
1202 if (!monc->m_auth_reply)
1203 goto out_subscribe;
1204
1205 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_KERNEL, true);
1206 monc->pending_auth = 0;
1207 if (!monc->m_auth)
1208 goto out_auth_reply;
1209
1210 ceph_con_init(&monc->con, monc, &mon_con_ops,
1211 &monc->client->msgr);
1212
1213 monc->cur_mon = -1;
1214 monc->had_a_connection = false;
1215 monc->hunt_mult = 1;
1216
1217 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
1218 monc->generic_request_tree = RB_ROOT;
1219 monc->last_tid = 0;
1220
1221 monc->fs_cluster_id = CEPH_FS_CLUSTER_ID_NONE;
1222
1223 return 0;
1224
1225out_auth_reply:
1226 ceph_msg_put(monc->m_auth_reply);
1227out_subscribe:
1228 ceph_msg_put(monc->m_subscribe);
1229out_subscribe_ack:
1230 ceph_msg_put(monc->m_subscribe_ack);
1231out_auth:
1232 ceph_auth_destroy(monc->auth);
1233out_monmap:
1234 kfree(monc->monmap);
1235out:
1236 return err;
1237}
1238EXPORT_SYMBOL(ceph_monc_init);
1239
1240void ceph_monc_stop(struct ceph_mon_client *monc)
1241{
1242 dout("stop\n");
1243
1244 mutex_lock(&monc->mutex);
1245 __close_session(monc);
1246 monc->hunting = false;
1247 monc->cur_mon = -1;
1248 mutex_unlock(&monc->mutex);
1249
1250 cancel_delayed_work_sync(&monc->delayed_work);
1251
1252 /*
1253 * flush msgr queue before we destroy ourselves to ensure that:
1254 * - any work that references our embedded con is finished.
1255 * - any osd_client or other work that may reference an authorizer
1256 * finishes before we shut down the auth subsystem.
1257 */
1258 ceph_msgr_flush();
1259
1260 ceph_auth_destroy(monc->auth);
1261
1262 WARN_ON(!RB_EMPTY_ROOT(&monc->generic_request_tree));
1263
1264 ceph_msg_put(monc->m_auth);
1265 ceph_msg_put(monc->m_auth_reply);
1266 ceph_msg_put(monc->m_subscribe);
1267 ceph_msg_put(monc->m_subscribe_ack);
1268
1269 kfree(monc->monmap);
1270}
1271EXPORT_SYMBOL(ceph_monc_stop);
1272
1273static void finish_hunting(struct ceph_mon_client *monc)
1274{
1275 if (monc->hunting) {
1276 dout("%s found mon%d\n", __func__, monc->cur_mon);
1277 monc->hunting = false;
1278 monc->had_a_connection = true;
1279 un_backoff(monc);
1280 __schedule_delayed(monc);
1281 }
1282}
1283
1284static void finish_auth(struct ceph_mon_client *monc, int auth_err,
1285 bool was_authed)
1286{
1287 dout("%s auth_err %d was_authed %d\n", __func__, auth_err, was_authed);
1288 WARN_ON(auth_err > 0);
1289
1290 monc->pending_auth = 0;
1291 if (auth_err) {
1292 monc->client->auth_err = auth_err;
1293 wake_up_all(&monc->client->auth_wq);
1294 return;
1295 }
1296
1297 if (!was_authed && ceph_auth_is_authenticated(monc->auth)) {
1298 dout("%s authenticated, starting session global_id %llu\n",
1299 __func__, monc->auth->global_id);
1300
1301 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
1302 monc->client->msgr.inst.name.num =
1303 cpu_to_le64(monc->auth->global_id);
1304
1305 __send_subscribe(monc);
1306 __resend_generic_request(monc);
1307
1308 pr_info("mon%d %s session established\n", monc->cur_mon,
1309 ceph_pr_addr(&monc->con.peer_addr));
1310 }
1311}
1312
1313static void handle_auth_reply(struct ceph_mon_client *monc,
1314 struct ceph_msg *msg)
1315{
1316 bool was_authed;
1317 int ret;
1318
1319 mutex_lock(&monc->mutex);
1320 was_authed = ceph_auth_is_authenticated(monc->auth);
1321 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
1322 msg->front.iov_len,
1323 monc->m_auth->front.iov_base,
1324 monc->m_auth->front_alloc_len);
1325 if (ret > 0) {
1326 __send_prepared_auth_request(monc, ret);
1327 } else {
1328 finish_auth(monc, ret, was_authed);
1329 finish_hunting(monc);
1330 }
1331 mutex_unlock(&monc->mutex);
1332}
1333
1334static int __validate_auth(struct ceph_mon_client *monc)
1335{
1336 int ret;
1337
1338 if (monc->pending_auth)
1339 return 0;
1340
1341 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
1342 monc->m_auth->front_alloc_len);
1343 if (ret <= 0)
1344 return ret; /* either an error, or no need to authenticate */
1345 __send_prepared_auth_request(monc, ret);
1346 return 0;
1347}
1348
1349int ceph_monc_validate_auth(struct ceph_mon_client *monc)
1350{
1351 int ret;
1352
1353 mutex_lock(&monc->mutex);
1354 ret = __validate_auth(monc);
1355 mutex_unlock(&monc->mutex);
1356 return ret;
1357}
1358EXPORT_SYMBOL(ceph_monc_validate_auth);
1359
1360static int mon_get_auth_request(struct ceph_connection *con,
1361 void *buf, int *buf_len,
1362 void **authorizer, int *authorizer_len)
1363{
1364 struct ceph_mon_client *monc = con->private;
1365 int ret;
1366
1367 mutex_lock(&monc->mutex);
1368 ret = ceph_auth_get_request(monc->auth, buf, *buf_len);
1369 mutex_unlock(&monc->mutex);
1370 if (ret < 0)
1371 return ret;
1372
1373 *buf_len = ret;
1374 *authorizer = NULL;
1375 *authorizer_len = 0;
1376 return 0;
1377}
1378
1379static int mon_handle_auth_reply_more(struct ceph_connection *con,
1380 void *reply, int reply_len,
1381 void *buf, int *buf_len,
1382 void **authorizer, int *authorizer_len)
1383{
1384 struct ceph_mon_client *monc = con->private;
1385 int ret;
1386
1387 mutex_lock(&monc->mutex);
1388 ret = ceph_auth_handle_reply_more(monc->auth, reply, reply_len,
1389 buf, *buf_len);
1390 mutex_unlock(&monc->mutex);
1391 if (ret < 0)
1392 return ret;
1393
1394 *buf_len = ret;
1395 *authorizer = NULL;
1396 *authorizer_len = 0;
1397 return 0;
1398}
1399
1400static int mon_handle_auth_done(struct ceph_connection *con,
1401 u64 global_id, void *reply, int reply_len,
1402 u8 *session_key, int *session_key_len,
1403 u8 *con_secret, int *con_secret_len)
1404{
1405 struct ceph_mon_client *monc = con->private;
1406 bool was_authed;
1407 int ret;
1408
1409 mutex_lock(&monc->mutex);
1410 WARN_ON(!monc->hunting);
1411 was_authed = ceph_auth_is_authenticated(monc->auth);
1412 ret = ceph_auth_handle_reply_done(monc->auth, global_id,
1413 reply, reply_len,
1414 session_key, session_key_len,
1415 con_secret, con_secret_len);
1416 finish_auth(monc, ret, was_authed);
1417 if (!ret)
1418 finish_hunting(monc);
1419 mutex_unlock(&monc->mutex);
1420 return 0;
1421}
1422
1423static int mon_handle_auth_bad_method(struct ceph_connection *con,
1424 int used_proto, int result,
1425 const int *allowed_protos, int proto_cnt,
1426 const int *allowed_modes, int mode_cnt)
1427{
1428 struct ceph_mon_client *monc = con->private;
1429 bool was_authed;
1430
1431 mutex_lock(&monc->mutex);
1432 WARN_ON(!monc->hunting);
1433 was_authed = ceph_auth_is_authenticated(monc->auth);
1434 ceph_auth_handle_bad_method(monc->auth, used_proto, result,
1435 allowed_protos, proto_cnt,
1436 allowed_modes, mode_cnt);
1437 finish_auth(monc, -EACCES, was_authed);
1438 mutex_unlock(&monc->mutex);
1439 return 0;
1440}
1441
1442/*
1443 * handle incoming message
1444 */
1445static void mon_dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1446{
1447 struct ceph_mon_client *monc = con->private;
1448 int type = le16_to_cpu(msg->hdr.type);
1449
1450 switch (type) {
1451 case CEPH_MSG_AUTH_REPLY:
1452 handle_auth_reply(monc, msg);
1453 break;
1454
1455 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1456 handle_subscribe_ack(monc, msg);
1457 break;
1458
1459 case CEPH_MSG_STATFS_REPLY:
1460 handle_statfs_reply(monc, msg);
1461 break;
1462
1463 case CEPH_MSG_MON_GET_VERSION_REPLY:
1464 handle_get_version_reply(monc, msg);
1465 break;
1466
1467 case CEPH_MSG_MON_COMMAND_ACK:
1468 handle_command_ack(monc, msg);
1469 break;
1470
1471 case CEPH_MSG_MON_MAP:
1472 ceph_monc_handle_map(monc, msg);
1473 break;
1474
1475 case CEPH_MSG_OSD_MAP:
1476 ceph_osdc_handle_map(&monc->client->osdc, msg);
1477 break;
1478
1479 default:
1480 /* can the chained handler handle it? */
1481 if (monc->client->extra_mon_dispatch &&
1482 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1483 break;
1484
1485 pr_err("received unknown message type %d %s\n", type,
1486 ceph_msg_type_name(type));
1487 }
1488 ceph_msg_put(msg);
1489}
1490
1491/*
1492 * Allocate memory for incoming message
1493 */
1494static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1495 struct ceph_msg_header *hdr,
1496 int *skip)
1497{
1498 struct ceph_mon_client *monc = con->private;
1499 int type = le16_to_cpu(hdr->type);
1500 int front_len = le32_to_cpu(hdr->front_len);
1501 struct ceph_msg *m = NULL;
1502
1503 *skip = 0;
1504
1505 switch (type) {
1506 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1507 m = ceph_msg_get(monc->m_subscribe_ack);
1508 break;
1509 case CEPH_MSG_STATFS_REPLY:
1510 case CEPH_MSG_MON_COMMAND_ACK:
1511 return get_generic_reply(con, hdr, skip);
1512 case CEPH_MSG_AUTH_REPLY:
1513 m = ceph_msg_get(monc->m_auth_reply);
1514 break;
1515 case CEPH_MSG_MON_GET_VERSION_REPLY:
1516 if (le64_to_cpu(hdr->tid) != 0)
1517 return get_generic_reply(con, hdr, skip);
1518
1519 /*
1520 * Older OSDs don't set reply tid even if the original
1521 * request had a non-zero tid. Work around this weirdness
1522 * by allocating a new message.
1523 */
1524 fallthrough;
1525 case CEPH_MSG_MON_MAP:
1526 case CEPH_MSG_MDS_MAP:
1527 case CEPH_MSG_OSD_MAP:
1528 case CEPH_MSG_FS_MAP_USER:
1529 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1530 if (!m)
1531 return NULL; /* ENOMEM--return skip == 0 */
1532 break;
1533 }
1534
1535 if (!m) {
1536 pr_info("alloc_msg unknown type %d\n", type);
1537 *skip = 1;
1538 } else if (front_len > m->front_alloc_len) {
1539 pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1540 front_len, m->front_alloc_len,
1541 (unsigned int)con->peer_name.type,
1542 le64_to_cpu(con->peer_name.num));
1543 ceph_msg_put(m);
1544 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1545 }
1546
1547 return m;
1548}
1549
1550/*
1551 * If the monitor connection resets, pick a new monitor and resubmit
1552 * any pending requests.
1553 */
1554static void mon_fault(struct ceph_connection *con)
1555{
1556 struct ceph_mon_client *monc = con->private;
1557
1558 mutex_lock(&monc->mutex);
1559 dout("%s mon%d\n", __func__, monc->cur_mon);
1560 if (monc->cur_mon >= 0) {
1561 if (!monc->hunting) {
1562 dout("%s hunting for new mon\n", __func__);
1563 reopen_session(monc);
1564 __schedule_delayed(monc);
1565 } else {
1566 dout("%s already hunting\n", __func__);
1567 }
1568 }
1569 mutex_unlock(&monc->mutex);
1570}
1571
1572/*
1573 * We can ignore refcounting on the connection struct, as all references
1574 * will come from the messenger workqueue, which is drained prior to
1575 * mon_client destruction.
1576 */
1577static struct ceph_connection *mon_get_con(struct ceph_connection *con)
1578{
1579 return con;
1580}
1581
1582static void mon_put_con(struct ceph_connection *con)
1583{
1584}
1585
1586static const struct ceph_connection_operations mon_con_ops = {
1587 .get = mon_get_con,
1588 .put = mon_put_con,
1589 .alloc_msg = mon_alloc_msg,
1590 .dispatch = mon_dispatch,
1591 .fault = mon_fault,
1592 .get_auth_request = mon_get_auth_request,
1593 .handle_auth_reply_more = mon_handle_auth_reply_more,
1594 .handle_auth_done = mon_handle_auth_done,
1595 .handle_auth_bad_method = mon_handle_auth_bad_method,
1596};
1// SPDX-License-Identifier: GPL-2.0
2#include <linux/ceph/ceph_debug.h>
3
4#include <linux/module.h>
5#include <linux/types.h>
6#include <linux/slab.h>
7#include <linux/random.h>
8#include <linux/sched.h>
9
10#include <linux/ceph/ceph_features.h>
11#include <linux/ceph/mon_client.h>
12#include <linux/ceph/libceph.h>
13#include <linux/ceph/debugfs.h>
14#include <linux/ceph/decode.h>
15#include <linux/ceph/auth.h>
16
17/*
18 * Interact with Ceph monitor cluster. Handle requests for new map
19 * versions, and periodically resend as needed. Also implement
20 * statfs() and umount().
21 *
22 * A small cluster of Ceph "monitors" are responsible for managing critical
23 * cluster configuration and state information. An odd number (e.g., 3, 5)
24 * of cmon daemons use a modified version of the Paxos part-time parliament
25 * algorithm to manage the MDS map (mds cluster membership), OSD map, and
26 * list of clients who have mounted the file system.
27 *
28 * We maintain an open, active session with a monitor at all times in order to
29 * receive timely MDSMap updates. We periodically send a keepalive byte on the
30 * TCP socket to ensure we detect a failure. If the connection does break, we
31 * randomly hunt for a new monitor. Once the connection is reestablished, we
32 * resend any outstanding requests.
33 */
34
35static const struct ceph_connection_operations mon_con_ops;
36
37static int __validate_auth(struct ceph_mon_client *monc);
38
39/*
40 * Decode a monmap blob (e.g., during mount).
41 */
42struct ceph_monmap *ceph_monmap_decode(void *p, void *end)
43{
44 struct ceph_monmap *m = NULL;
45 int i, err = -EINVAL;
46 struct ceph_fsid fsid;
47 u32 epoch, num_mon;
48 u32 len;
49
50 ceph_decode_32_safe(&p, end, len, bad);
51 ceph_decode_need(&p, end, len, bad);
52
53 dout("monmap_decode %p %p len %d\n", p, end, (int)(end-p));
54 p += sizeof(u16); /* skip version */
55
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);
59
60 num_mon = ceph_decode_32(&p);
61 ceph_decode_need(&p, end, num_mon*sizeof(m->mon_inst[0]), bad);
62
63 if (num_mon > CEPH_MAX_MON)
64 goto bad;
65 m = kmalloc(sizeof(*m) + sizeof(m->mon_inst[0])*num_mon, GFP_NOFS);
66 if (m == NULL)
67 return ERR_PTR(-ENOMEM);
68 m->fsid = fsid;
69 m->epoch = epoch;
70 m->num_mon = num_mon;
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);
74
75 dout("monmap_decode epoch %d, num_mon %d\n", m->epoch,
76 m->num_mon);
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));
80 return m;
81
82bad:
83 dout("monmap_decode failed with %d\n", err);
84 kfree(m);
85 return ERR_PTR(err);
86}
87
88/*
89 * return true if *addr is included in the monmap.
90 */
91int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr)
92{
93 int i;
94
95 for (i = 0; i < m->num_mon; i++)
96 if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0)
97 return 1;
98 return 0;
99}
100
101/*
102 * Send an auth request.
103 */
104static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len)
105{
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);
112}
113
114/*
115 * Close monitor session, if any.
116 */
117static void __close_session(struct ceph_mon_client *monc)
118{
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);
125
126 monc->pending_auth = 0;
127 ceph_auth_reset(monc->auth);
128}
129
130/*
131 * Pick a new monitor at random and set cur_mon. If we are repicking
132 * (i.e. cur_mon is already set), be sure to pick a different one.
133 */
134static void pick_new_mon(struct ceph_mon_client *monc)
135{
136 int old_mon = monc->cur_mon;
137
138 BUG_ON(monc->monmap->num_mon < 1);
139
140 if (monc->monmap->num_mon == 1) {
141 monc->cur_mon = 0;
142 } else {
143 int max = monc->monmap->num_mon;
144 int o = -1;
145 int n;
146
147 if (monc->cur_mon >= 0) {
148 if (monc->cur_mon < monc->monmap->num_mon)
149 o = monc->cur_mon;
150 if (o >= 0)
151 max--;
152 }
153
154 n = prandom_u32() % max;
155 if (o >= 0 && n >= o)
156 n++;
157
158 monc->cur_mon = n;
159 }
160
161 dout("%s mon%d -> mon%d out of %d mons\n", __func__, old_mon,
162 monc->cur_mon, monc->monmap->num_mon);
163}
164
165/*
166 * Open a session with a new monitor.
167 */
168static void __open_session(struct ceph_mon_client *monc)
169{
170 int ret;
171
172 pick_new_mon(monc);
173
174 monc->hunting = true;
175 if (monc->had_a_connection) {
176 monc->hunt_mult *= CEPH_MONC_HUNT_BACKOFF;
177 if (monc->hunt_mult > CEPH_MONC_HUNT_MAX_MULT)
178 monc->hunt_mult = CEPH_MONC_HUNT_MAX_MULT;
179 }
180
181 monc->sub_renew_after = jiffies; /* i.e., expired */
182 monc->sub_renew_sent = 0;
183
184 dout("%s opening mon%d\n", __func__, monc->cur_mon);
185 ceph_con_open(&monc->con, CEPH_ENTITY_TYPE_MON, monc->cur_mon,
186 &monc->monmap->mon_inst[monc->cur_mon].addr);
187
188 /*
189 * send an initial keepalive to ensure our timestamp is valid
190 * by the time we are in an OPENED state
191 */
192 ceph_con_keepalive(&monc->con);
193
194 /* initiate authentication handshake */
195 ret = ceph_auth_build_hello(monc->auth,
196 monc->m_auth->front.iov_base,
197 monc->m_auth->front_alloc_len);
198 BUG_ON(ret <= 0);
199 __send_prepared_auth_request(monc, ret);
200}
201
202static void reopen_session(struct ceph_mon_client *monc)
203{
204 if (!monc->hunting)
205 pr_info("mon%d %s session lost, hunting for new mon\n",
206 monc->cur_mon, ceph_pr_addr(&monc->con.peer_addr.in_addr));
207
208 __close_session(monc);
209 __open_session(monc);
210}
211
212static void un_backoff(struct ceph_mon_client *monc)
213{
214 monc->hunt_mult /= 2; /* reduce by 50% */
215 if (monc->hunt_mult < 1)
216 monc->hunt_mult = 1;
217 dout("%s hunt_mult now %d\n", __func__, monc->hunt_mult);
218}
219
220/*
221 * Reschedule delayed work timer.
222 */
223static void __schedule_delayed(struct ceph_mon_client *monc)
224{
225 unsigned long delay;
226
227 if (monc->hunting)
228 delay = CEPH_MONC_HUNT_INTERVAL * monc->hunt_mult;
229 else
230 delay = CEPH_MONC_PING_INTERVAL;
231
232 dout("__schedule_delayed after %lu\n", delay);
233 mod_delayed_work(system_wq, &monc->delayed_work,
234 round_jiffies_relative(delay));
235}
236
237const char *ceph_sub_str[] = {
238 [CEPH_SUB_MONMAP] = "monmap",
239 [CEPH_SUB_OSDMAP] = "osdmap",
240 [CEPH_SUB_FSMAP] = "fsmap.user",
241 [CEPH_SUB_MDSMAP] = "mdsmap",
242};
243
244/*
245 * Send subscribe request for one or more maps, according to
246 * monc->subs.
247 */
248static void __send_subscribe(struct ceph_mon_client *monc)
249{
250 struct ceph_msg *msg = monc->m_subscribe;
251 void *p = msg->front.iov_base;
252 void *const end = p + msg->front_alloc_len;
253 int num = 0;
254 int i;
255
256 dout("%s sent %lu\n", __func__, monc->sub_renew_sent);
257
258 BUG_ON(monc->cur_mon < 0);
259
260 if (!monc->sub_renew_sent)
261 monc->sub_renew_sent = jiffies | 1; /* never 0 */
262
263 msg->hdr.version = cpu_to_le16(2);
264
265 for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
266 if (monc->subs[i].want)
267 num++;
268 }
269 BUG_ON(num < 1); /* monmap sub is always there */
270 ceph_encode_32(&p, num);
271 for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
272 char buf[32];
273 int len;
274
275 if (!monc->subs[i].want)
276 continue;
277
278 len = sprintf(buf, "%s", ceph_sub_str[i]);
279 if (i == CEPH_SUB_MDSMAP &&
280 monc->fs_cluster_id != CEPH_FS_CLUSTER_ID_NONE)
281 len += sprintf(buf + len, ".%d", monc->fs_cluster_id);
282
283 dout("%s %s start %llu flags 0x%x\n", __func__, buf,
284 le64_to_cpu(monc->subs[i].item.start),
285 monc->subs[i].item.flags);
286 ceph_encode_string(&p, end, buf, len);
287 memcpy(p, &monc->subs[i].item, sizeof(monc->subs[i].item));
288 p += sizeof(monc->subs[i].item);
289 }
290
291 BUG_ON(p > end);
292 msg->front.iov_len = p - msg->front.iov_base;
293 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
294 ceph_msg_revoke(msg);
295 ceph_con_send(&monc->con, ceph_msg_get(msg));
296}
297
298static void handle_subscribe_ack(struct ceph_mon_client *monc,
299 struct ceph_msg *msg)
300{
301 unsigned int seconds;
302 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
303
304 if (msg->front.iov_len < sizeof(*h))
305 goto bad;
306 seconds = le32_to_cpu(h->duration);
307
308 mutex_lock(&monc->mutex);
309 if (monc->sub_renew_sent) {
310 /*
311 * This is only needed for legacy (infernalis or older)
312 * MONs -- see delayed_work().
313 */
314 monc->sub_renew_after = monc->sub_renew_sent +
315 (seconds >> 1) * HZ - 1;
316 dout("%s sent %lu duration %d renew after %lu\n", __func__,
317 monc->sub_renew_sent, seconds, monc->sub_renew_after);
318 monc->sub_renew_sent = 0;
319 } else {
320 dout("%s sent %lu renew after %lu, ignoring\n", __func__,
321 monc->sub_renew_sent, monc->sub_renew_after);
322 }
323 mutex_unlock(&monc->mutex);
324 return;
325bad:
326 pr_err("got corrupt subscribe-ack msg\n");
327 ceph_msg_dump(msg);
328}
329
330/*
331 * Register interest in a map
332 *
333 * @sub: one of CEPH_SUB_*
334 * @epoch: X for "every map since X", or 0 for "just the latest"
335 */
336static bool __ceph_monc_want_map(struct ceph_mon_client *monc, int sub,
337 u32 epoch, bool continuous)
338{
339 __le64 start = cpu_to_le64(epoch);
340 u8 flags = !continuous ? CEPH_SUBSCRIBE_ONETIME : 0;
341
342 dout("%s %s epoch %u continuous %d\n", __func__, ceph_sub_str[sub],
343 epoch, continuous);
344
345 if (monc->subs[sub].want &&
346 monc->subs[sub].item.start == start &&
347 monc->subs[sub].item.flags == flags)
348 return false;
349
350 monc->subs[sub].item.start = start;
351 monc->subs[sub].item.flags = flags;
352 monc->subs[sub].want = true;
353
354 return true;
355}
356
357bool ceph_monc_want_map(struct ceph_mon_client *monc, int sub, u32 epoch,
358 bool continuous)
359{
360 bool need_request;
361
362 mutex_lock(&monc->mutex);
363 need_request = __ceph_monc_want_map(monc, sub, epoch, continuous);
364 mutex_unlock(&monc->mutex);
365
366 return need_request;
367}
368EXPORT_SYMBOL(ceph_monc_want_map);
369
370/*
371 * Keep track of which maps we have
372 *
373 * @sub: one of CEPH_SUB_*
374 */
375static void __ceph_monc_got_map(struct ceph_mon_client *monc, int sub,
376 u32 epoch)
377{
378 dout("%s %s epoch %u\n", __func__, ceph_sub_str[sub], epoch);
379
380 if (monc->subs[sub].want) {
381 if (monc->subs[sub].item.flags & CEPH_SUBSCRIBE_ONETIME)
382 monc->subs[sub].want = false;
383 else
384 monc->subs[sub].item.start = cpu_to_le64(epoch + 1);
385 }
386
387 monc->subs[sub].have = epoch;
388}
389
390void ceph_monc_got_map(struct ceph_mon_client *monc, int sub, u32 epoch)
391{
392 mutex_lock(&monc->mutex);
393 __ceph_monc_got_map(monc, sub, epoch);
394 mutex_unlock(&monc->mutex);
395}
396EXPORT_SYMBOL(ceph_monc_got_map);
397
398void ceph_monc_renew_subs(struct ceph_mon_client *monc)
399{
400 mutex_lock(&monc->mutex);
401 __send_subscribe(monc);
402 mutex_unlock(&monc->mutex);
403}
404EXPORT_SYMBOL(ceph_monc_renew_subs);
405
406/*
407 * Wait for an osdmap with a given epoch.
408 *
409 * @epoch: epoch to wait for
410 * @timeout: in jiffies, 0 means "wait forever"
411 */
412int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
413 unsigned long timeout)
414{
415 unsigned long started = jiffies;
416 long ret;
417
418 mutex_lock(&monc->mutex);
419 while (monc->subs[CEPH_SUB_OSDMAP].have < epoch) {
420 mutex_unlock(&monc->mutex);
421
422 if (timeout && time_after_eq(jiffies, started + timeout))
423 return -ETIMEDOUT;
424
425 ret = wait_event_interruptible_timeout(monc->client->auth_wq,
426 monc->subs[CEPH_SUB_OSDMAP].have >= epoch,
427 ceph_timeout_jiffies(timeout));
428 if (ret < 0)
429 return ret;
430
431 mutex_lock(&monc->mutex);
432 }
433
434 mutex_unlock(&monc->mutex);
435 return 0;
436}
437EXPORT_SYMBOL(ceph_monc_wait_osdmap);
438
439/*
440 * Open a session with a random monitor. Request monmap and osdmap,
441 * which are waited upon in __ceph_open_session().
442 */
443int ceph_monc_open_session(struct ceph_mon_client *monc)
444{
445 mutex_lock(&monc->mutex);
446 __ceph_monc_want_map(monc, CEPH_SUB_MONMAP, 0, true);
447 __ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, 0, false);
448 __open_session(monc);
449 __schedule_delayed(monc);
450 mutex_unlock(&monc->mutex);
451 return 0;
452}
453EXPORT_SYMBOL(ceph_monc_open_session);
454
455static void ceph_monc_handle_map(struct ceph_mon_client *monc,
456 struct ceph_msg *msg)
457{
458 struct ceph_client *client = monc->client;
459 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
460 void *p, *end;
461
462 mutex_lock(&monc->mutex);
463
464 dout("handle_monmap\n");
465 p = msg->front.iov_base;
466 end = p + msg->front.iov_len;
467
468 monmap = ceph_monmap_decode(p, end);
469 if (IS_ERR(monmap)) {
470 pr_err("problem decoding monmap, %d\n",
471 (int)PTR_ERR(monmap));
472 goto out;
473 }
474
475 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
476 kfree(monmap);
477 goto out;
478 }
479
480 client->monc.monmap = monmap;
481 kfree(old);
482
483 __ceph_monc_got_map(monc, CEPH_SUB_MONMAP, monc->monmap->epoch);
484 client->have_fsid = true;
485
486out:
487 mutex_unlock(&monc->mutex);
488 wake_up_all(&client->auth_wq);
489}
490
491/*
492 * generic requests (currently statfs, mon_get_version)
493 */
494DEFINE_RB_FUNCS(generic_request, struct ceph_mon_generic_request, tid, node)
495
496static void release_generic_request(struct kref *kref)
497{
498 struct ceph_mon_generic_request *req =
499 container_of(kref, struct ceph_mon_generic_request, kref);
500
501 dout("%s greq %p request %p reply %p\n", __func__, req, req->request,
502 req->reply);
503 WARN_ON(!RB_EMPTY_NODE(&req->node));
504
505 if (req->reply)
506 ceph_msg_put(req->reply);
507 if (req->request)
508 ceph_msg_put(req->request);
509
510 kfree(req);
511}
512
513static void put_generic_request(struct ceph_mon_generic_request *req)
514{
515 if (req)
516 kref_put(&req->kref, release_generic_request);
517}
518
519static void get_generic_request(struct ceph_mon_generic_request *req)
520{
521 kref_get(&req->kref);
522}
523
524static struct ceph_mon_generic_request *
525alloc_generic_request(struct ceph_mon_client *monc, gfp_t gfp)
526{
527 struct ceph_mon_generic_request *req;
528
529 req = kzalloc(sizeof(*req), gfp);
530 if (!req)
531 return NULL;
532
533 req->monc = monc;
534 kref_init(&req->kref);
535 RB_CLEAR_NODE(&req->node);
536 init_completion(&req->completion);
537
538 dout("%s greq %p\n", __func__, req);
539 return req;
540}
541
542static void register_generic_request(struct ceph_mon_generic_request *req)
543{
544 struct ceph_mon_client *monc = req->monc;
545
546 WARN_ON(req->tid);
547
548 get_generic_request(req);
549 req->tid = ++monc->last_tid;
550 insert_generic_request(&monc->generic_request_tree, req);
551}
552
553static void send_generic_request(struct ceph_mon_client *monc,
554 struct ceph_mon_generic_request *req)
555{
556 WARN_ON(!req->tid);
557
558 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
559 req->request->hdr.tid = cpu_to_le64(req->tid);
560 ceph_con_send(&monc->con, ceph_msg_get(req->request));
561}
562
563static void __finish_generic_request(struct ceph_mon_generic_request *req)
564{
565 struct ceph_mon_client *monc = req->monc;
566
567 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
568 erase_generic_request(&monc->generic_request_tree, req);
569
570 ceph_msg_revoke(req->request);
571 ceph_msg_revoke_incoming(req->reply);
572}
573
574static void finish_generic_request(struct ceph_mon_generic_request *req)
575{
576 __finish_generic_request(req);
577 put_generic_request(req);
578}
579
580static void complete_generic_request(struct ceph_mon_generic_request *req)
581{
582 if (req->complete_cb)
583 req->complete_cb(req);
584 else
585 complete_all(&req->completion);
586 put_generic_request(req);
587}
588
589static void cancel_generic_request(struct ceph_mon_generic_request *req)
590{
591 struct ceph_mon_client *monc = req->monc;
592 struct ceph_mon_generic_request *lookup_req;
593
594 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
595
596 mutex_lock(&monc->mutex);
597 lookup_req = lookup_generic_request(&monc->generic_request_tree,
598 req->tid);
599 if (lookup_req) {
600 WARN_ON(lookup_req != req);
601 finish_generic_request(req);
602 }
603
604 mutex_unlock(&monc->mutex);
605}
606
607static int wait_generic_request(struct ceph_mon_generic_request *req)
608{
609 int ret;
610
611 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
612 ret = wait_for_completion_interruptible(&req->completion);
613 if (ret)
614 cancel_generic_request(req);
615 else
616 ret = req->result; /* completed */
617
618 return ret;
619}
620
621static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
622 struct ceph_msg_header *hdr,
623 int *skip)
624{
625 struct ceph_mon_client *monc = con->private;
626 struct ceph_mon_generic_request *req;
627 u64 tid = le64_to_cpu(hdr->tid);
628 struct ceph_msg *m;
629
630 mutex_lock(&monc->mutex);
631 req = lookup_generic_request(&monc->generic_request_tree, tid);
632 if (!req) {
633 dout("get_generic_reply %lld dne\n", tid);
634 *skip = 1;
635 m = NULL;
636 } else {
637 dout("get_generic_reply %lld got %p\n", tid, req->reply);
638 *skip = 0;
639 m = ceph_msg_get(req->reply);
640 /*
641 * we don't need to track the connection reading into
642 * this reply because we only have one open connection
643 * at a time, ever.
644 */
645 }
646 mutex_unlock(&monc->mutex);
647 return m;
648}
649
650/*
651 * statfs
652 */
653static void handle_statfs_reply(struct ceph_mon_client *monc,
654 struct ceph_msg *msg)
655{
656 struct ceph_mon_generic_request *req;
657 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
658 u64 tid = le64_to_cpu(msg->hdr.tid);
659
660 dout("%s msg %p tid %llu\n", __func__, msg, tid);
661
662 if (msg->front.iov_len != sizeof(*reply))
663 goto bad;
664
665 mutex_lock(&monc->mutex);
666 req = lookup_generic_request(&monc->generic_request_tree, tid);
667 if (!req) {
668 mutex_unlock(&monc->mutex);
669 return;
670 }
671
672 req->result = 0;
673 *req->u.st = reply->st; /* struct */
674 __finish_generic_request(req);
675 mutex_unlock(&monc->mutex);
676
677 complete_generic_request(req);
678 return;
679
680bad:
681 pr_err("corrupt statfs reply, tid %llu\n", tid);
682 ceph_msg_dump(msg);
683}
684
685/*
686 * Do a synchronous statfs().
687 */
688int ceph_monc_do_statfs(struct ceph_mon_client *monc, u64 data_pool,
689 struct ceph_statfs *buf)
690{
691 struct ceph_mon_generic_request *req;
692 struct ceph_mon_statfs *h;
693 int ret = -ENOMEM;
694
695 req = alloc_generic_request(monc, GFP_NOFS);
696 if (!req)
697 goto out;
698
699 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
700 true);
701 if (!req->request)
702 goto out;
703
704 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 64, GFP_NOFS, true);
705 if (!req->reply)
706 goto out;
707
708 req->u.st = buf;
709 req->request->hdr.version = cpu_to_le16(2);
710
711 mutex_lock(&monc->mutex);
712 register_generic_request(req);
713 /* fill out request */
714 h = req->request->front.iov_base;
715 h->monhdr.have_version = 0;
716 h->monhdr.session_mon = cpu_to_le16(-1);
717 h->monhdr.session_mon_tid = 0;
718 h->fsid = monc->monmap->fsid;
719 h->contains_data_pool = (data_pool != CEPH_NOPOOL);
720 h->data_pool = cpu_to_le64(data_pool);
721 send_generic_request(monc, req);
722 mutex_unlock(&monc->mutex);
723
724 ret = wait_generic_request(req);
725out:
726 put_generic_request(req);
727 return ret;
728}
729EXPORT_SYMBOL(ceph_monc_do_statfs);
730
731static void handle_get_version_reply(struct ceph_mon_client *monc,
732 struct ceph_msg *msg)
733{
734 struct ceph_mon_generic_request *req;
735 u64 tid = le64_to_cpu(msg->hdr.tid);
736 void *p = msg->front.iov_base;
737 void *end = p + msg->front_alloc_len;
738 u64 handle;
739
740 dout("%s msg %p tid %llu\n", __func__, msg, tid);
741
742 ceph_decode_need(&p, end, 2*sizeof(u64), bad);
743 handle = ceph_decode_64(&p);
744 if (tid != 0 && tid != handle)
745 goto bad;
746
747 mutex_lock(&monc->mutex);
748 req = lookup_generic_request(&monc->generic_request_tree, handle);
749 if (!req) {
750 mutex_unlock(&monc->mutex);
751 return;
752 }
753
754 req->result = 0;
755 req->u.newest = ceph_decode_64(&p);
756 __finish_generic_request(req);
757 mutex_unlock(&monc->mutex);
758
759 complete_generic_request(req);
760 return;
761
762bad:
763 pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
764 ceph_msg_dump(msg);
765}
766
767static struct ceph_mon_generic_request *
768__ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
769 ceph_monc_callback_t cb, u64 private_data)
770{
771 struct ceph_mon_generic_request *req;
772
773 req = alloc_generic_request(monc, GFP_NOIO);
774 if (!req)
775 goto err_put_req;
776
777 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
778 sizeof(u64) + sizeof(u32) + strlen(what),
779 GFP_NOIO, true);
780 if (!req->request)
781 goto err_put_req;
782
783 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 32, GFP_NOIO,
784 true);
785 if (!req->reply)
786 goto err_put_req;
787
788 req->complete_cb = cb;
789 req->private_data = private_data;
790
791 mutex_lock(&monc->mutex);
792 register_generic_request(req);
793 {
794 void *p = req->request->front.iov_base;
795 void *const end = p + req->request->front_alloc_len;
796
797 ceph_encode_64(&p, req->tid); /* handle */
798 ceph_encode_string(&p, end, what, strlen(what));
799 WARN_ON(p != end);
800 }
801 send_generic_request(monc, req);
802 mutex_unlock(&monc->mutex);
803
804 return req;
805
806err_put_req:
807 put_generic_request(req);
808 return ERR_PTR(-ENOMEM);
809}
810
811/*
812 * Send MMonGetVersion and wait for the reply.
813 *
814 * @what: one of "mdsmap", "osdmap" or "monmap"
815 */
816int ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
817 u64 *newest)
818{
819 struct ceph_mon_generic_request *req;
820 int ret;
821
822 req = __ceph_monc_get_version(monc, what, NULL, 0);
823 if (IS_ERR(req))
824 return PTR_ERR(req);
825
826 ret = wait_generic_request(req);
827 if (!ret)
828 *newest = req->u.newest;
829
830 put_generic_request(req);
831 return ret;
832}
833EXPORT_SYMBOL(ceph_monc_get_version);
834
835/*
836 * Send MMonGetVersion,
837 *
838 * @what: one of "mdsmap", "osdmap" or "monmap"
839 */
840int ceph_monc_get_version_async(struct ceph_mon_client *monc, const char *what,
841 ceph_monc_callback_t cb, u64 private_data)
842{
843 struct ceph_mon_generic_request *req;
844
845 req = __ceph_monc_get_version(monc, what, cb, private_data);
846 if (IS_ERR(req))
847 return PTR_ERR(req);
848
849 put_generic_request(req);
850 return 0;
851}
852EXPORT_SYMBOL(ceph_monc_get_version_async);
853
854static void handle_command_ack(struct ceph_mon_client *monc,
855 struct ceph_msg *msg)
856{
857 struct ceph_mon_generic_request *req;
858 void *p = msg->front.iov_base;
859 void *const end = p + msg->front_alloc_len;
860 u64 tid = le64_to_cpu(msg->hdr.tid);
861
862 dout("%s msg %p tid %llu\n", __func__, msg, tid);
863
864 ceph_decode_need(&p, end, sizeof(struct ceph_mon_request_header) +
865 sizeof(u32), bad);
866 p += sizeof(struct ceph_mon_request_header);
867
868 mutex_lock(&monc->mutex);
869 req = lookup_generic_request(&monc->generic_request_tree, tid);
870 if (!req) {
871 mutex_unlock(&monc->mutex);
872 return;
873 }
874
875 req->result = ceph_decode_32(&p);
876 __finish_generic_request(req);
877 mutex_unlock(&monc->mutex);
878
879 complete_generic_request(req);
880 return;
881
882bad:
883 pr_err("corrupt mon_command ack, tid %llu\n", tid);
884 ceph_msg_dump(msg);
885}
886
887int ceph_monc_blacklist_add(struct ceph_mon_client *monc,
888 struct ceph_entity_addr *client_addr)
889{
890 struct ceph_mon_generic_request *req;
891 struct ceph_mon_command *h;
892 int ret = -ENOMEM;
893 int len;
894
895 req = alloc_generic_request(monc, GFP_NOIO);
896 if (!req)
897 goto out;
898
899 req->request = ceph_msg_new(CEPH_MSG_MON_COMMAND, 256, GFP_NOIO, true);
900 if (!req->request)
901 goto out;
902
903 req->reply = ceph_msg_new(CEPH_MSG_MON_COMMAND_ACK, 512, GFP_NOIO,
904 true);
905 if (!req->reply)
906 goto out;
907
908 mutex_lock(&monc->mutex);
909 register_generic_request(req);
910 h = req->request->front.iov_base;
911 h->monhdr.have_version = 0;
912 h->monhdr.session_mon = cpu_to_le16(-1);
913 h->monhdr.session_mon_tid = 0;
914 h->fsid = monc->monmap->fsid;
915 h->num_strs = cpu_to_le32(1);
916 len = sprintf(h->str, "{ \"prefix\": \"osd blacklist\", \
917 \"blacklistop\": \"add\", \
918 \"addr\": \"%pISpc/%u\" }",
919 &client_addr->in_addr, le32_to_cpu(client_addr->nonce));
920 h->str_len = cpu_to_le32(len);
921 send_generic_request(monc, req);
922 mutex_unlock(&monc->mutex);
923
924 ret = wait_generic_request(req);
925out:
926 put_generic_request(req);
927 return ret;
928}
929EXPORT_SYMBOL(ceph_monc_blacklist_add);
930
931/*
932 * Resend pending generic requests.
933 */
934static void __resend_generic_request(struct ceph_mon_client *monc)
935{
936 struct ceph_mon_generic_request *req;
937 struct rb_node *p;
938
939 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
940 req = rb_entry(p, struct ceph_mon_generic_request, node);
941 ceph_msg_revoke(req->request);
942 ceph_msg_revoke_incoming(req->reply);
943 ceph_con_send(&monc->con, ceph_msg_get(req->request));
944 }
945}
946
947/*
948 * Delayed work. If we haven't mounted yet, retry. Otherwise,
949 * renew/retry subscription as needed (in case it is timing out, or we
950 * got an ENOMEM). And keep the monitor connection alive.
951 */
952static void delayed_work(struct work_struct *work)
953{
954 struct ceph_mon_client *monc =
955 container_of(work, struct ceph_mon_client, delayed_work.work);
956
957 dout("monc delayed_work\n");
958 mutex_lock(&monc->mutex);
959 if (monc->hunting) {
960 dout("%s continuing hunt\n", __func__);
961 reopen_session(monc);
962 } else {
963 int is_auth = ceph_auth_is_authenticated(monc->auth);
964 if (ceph_con_keepalive_expired(&monc->con,
965 CEPH_MONC_PING_TIMEOUT)) {
966 dout("monc keepalive timeout\n");
967 is_auth = 0;
968 reopen_session(monc);
969 }
970
971 if (!monc->hunting) {
972 ceph_con_keepalive(&monc->con);
973 __validate_auth(monc);
974 un_backoff(monc);
975 }
976
977 if (is_auth &&
978 !(monc->con.peer_features & CEPH_FEATURE_MON_STATEFUL_SUB)) {
979 unsigned long now = jiffies;
980
981 dout("%s renew subs? now %lu renew after %lu\n",
982 __func__, now, monc->sub_renew_after);
983 if (time_after_eq(now, monc->sub_renew_after))
984 __send_subscribe(monc);
985 }
986 }
987 __schedule_delayed(monc);
988 mutex_unlock(&monc->mutex);
989}
990
991/*
992 * On startup, we build a temporary monmap populated with the IPs
993 * provided by mount(2).
994 */
995static int build_initial_monmap(struct ceph_mon_client *monc)
996{
997 struct ceph_options *opt = monc->client->options;
998 struct ceph_entity_addr *mon_addr = opt->mon_addr;
999 int num_mon = opt->num_mon;
1000 int i;
1001
1002 /* build initial monmap */
1003 monc->monmap = kzalloc(sizeof(*monc->monmap) +
1004 num_mon*sizeof(monc->monmap->mon_inst[0]),
1005 GFP_KERNEL);
1006 if (!monc->monmap)
1007 return -ENOMEM;
1008 for (i = 0; i < num_mon; i++) {
1009 monc->monmap->mon_inst[i].addr = mon_addr[i];
1010 monc->monmap->mon_inst[i].addr.nonce = 0;
1011 monc->monmap->mon_inst[i].name.type =
1012 CEPH_ENTITY_TYPE_MON;
1013 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
1014 }
1015 monc->monmap->num_mon = num_mon;
1016 return 0;
1017}
1018
1019int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
1020{
1021 int err = 0;
1022
1023 dout("init\n");
1024 memset(monc, 0, sizeof(*monc));
1025 monc->client = cl;
1026 monc->monmap = NULL;
1027 mutex_init(&monc->mutex);
1028
1029 err = build_initial_monmap(monc);
1030 if (err)
1031 goto out;
1032
1033 /* connection */
1034 /* authentication */
1035 monc->auth = ceph_auth_init(cl->options->name,
1036 cl->options->key);
1037 if (IS_ERR(monc->auth)) {
1038 err = PTR_ERR(monc->auth);
1039 goto out_monmap;
1040 }
1041 monc->auth->want_keys =
1042 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
1043 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
1044
1045 /* msgs */
1046 err = -ENOMEM;
1047 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
1048 sizeof(struct ceph_mon_subscribe_ack),
1049 GFP_KERNEL, true);
1050 if (!monc->m_subscribe_ack)
1051 goto out_auth;
1052
1053 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 128,
1054 GFP_KERNEL, true);
1055 if (!monc->m_subscribe)
1056 goto out_subscribe_ack;
1057
1058 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096,
1059 GFP_KERNEL, true);
1060 if (!monc->m_auth_reply)
1061 goto out_subscribe;
1062
1063 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_KERNEL, true);
1064 monc->pending_auth = 0;
1065 if (!monc->m_auth)
1066 goto out_auth_reply;
1067
1068 ceph_con_init(&monc->con, monc, &mon_con_ops,
1069 &monc->client->msgr);
1070
1071 monc->cur_mon = -1;
1072 monc->had_a_connection = false;
1073 monc->hunt_mult = 1;
1074
1075 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
1076 monc->generic_request_tree = RB_ROOT;
1077 monc->last_tid = 0;
1078
1079 monc->fs_cluster_id = CEPH_FS_CLUSTER_ID_NONE;
1080
1081 return 0;
1082
1083out_auth_reply:
1084 ceph_msg_put(monc->m_auth_reply);
1085out_subscribe:
1086 ceph_msg_put(monc->m_subscribe);
1087out_subscribe_ack:
1088 ceph_msg_put(monc->m_subscribe_ack);
1089out_auth:
1090 ceph_auth_destroy(monc->auth);
1091out_monmap:
1092 kfree(monc->monmap);
1093out:
1094 return err;
1095}
1096EXPORT_SYMBOL(ceph_monc_init);
1097
1098void ceph_monc_stop(struct ceph_mon_client *monc)
1099{
1100 dout("stop\n");
1101 cancel_delayed_work_sync(&monc->delayed_work);
1102
1103 mutex_lock(&monc->mutex);
1104 __close_session(monc);
1105 monc->cur_mon = -1;
1106 mutex_unlock(&monc->mutex);
1107
1108 /*
1109 * flush msgr queue before we destroy ourselves to ensure that:
1110 * - any work that references our embedded con is finished.
1111 * - any osd_client or other work that may reference an authorizer
1112 * finishes before we shut down the auth subsystem.
1113 */
1114 ceph_msgr_flush();
1115
1116 ceph_auth_destroy(monc->auth);
1117
1118 WARN_ON(!RB_EMPTY_ROOT(&monc->generic_request_tree));
1119
1120 ceph_msg_put(monc->m_auth);
1121 ceph_msg_put(monc->m_auth_reply);
1122 ceph_msg_put(monc->m_subscribe);
1123 ceph_msg_put(monc->m_subscribe_ack);
1124
1125 kfree(monc->monmap);
1126}
1127EXPORT_SYMBOL(ceph_monc_stop);
1128
1129static void finish_hunting(struct ceph_mon_client *monc)
1130{
1131 if (monc->hunting) {
1132 dout("%s found mon%d\n", __func__, monc->cur_mon);
1133 monc->hunting = false;
1134 monc->had_a_connection = true;
1135 un_backoff(monc);
1136 __schedule_delayed(monc);
1137 }
1138}
1139
1140static void handle_auth_reply(struct ceph_mon_client *monc,
1141 struct ceph_msg *msg)
1142{
1143 int ret;
1144 int was_auth = 0;
1145
1146 mutex_lock(&monc->mutex);
1147 was_auth = ceph_auth_is_authenticated(monc->auth);
1148 monc->pending_auth = 0;
1149 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
1150 msg->front.iov_len,
1151 monc->m_auth->front.iov_base,
1152 monc->m_auth->front_alloc_len);
1153 if (ret > 0) {
1154 __send_prepared_auth_request(monc, ret);
1155 goto out;
1156 }
1157
1158 finish_hunting(monc);
1159
1160 if (ret < 0) {
1161 monc->client->auth_err = ret;
1162 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
1163 dout("authenticated, starting session\n");
1164
1165 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
1166 monc->client->msgr.inst.name.num =
1167 cpu_to_le64(monc->auth->global_id);
1168
1169 __send_subscribe(monc);
1170 __resend_generic_request(monc);
1171
1172 pr_info("mon%d %s session established\n", monc->cur_mon,
1173 ceph_pr_addr(&monc->con.peer_addr.in_addr));
1174 }
1175
1176out:
1177 mutex_unlock(&monc->mutex);
1178 if (monc->client->auth_err < 0)
1179 wake_up_all(&monc->client->auth_wq);
1180}
1181
1182static int __validate_auth(struct ceph_mon_client *monc)
1183{
1184 int ret;
1185
1186 if (monc->pending_auth)
1187 return 0;
1188
1189 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
1190 monc->m_auth->front_alloc_len);
1191 if (ret <= 0)
1192 return ret; /* either an error, or no need to authenticate */
1193 __send_prepared_auth_request(monc, ret);
1194 return 0;
1195}
1196
1197int ceph_monc_validate_auth(struct ceph_mon_client *monc)
1198{
1199 int ret;
1200
1201 mutex_lock(&monc->mutex);
1202 ret = __validate_auth(monc);
1203 mutex_unlock(&monc->mutex);
1204 return ret;
1205}
1206EXPORT_SYMBOL(ceph_monc_validate_auth);
1207
1208/*
1209 * handle incoming message
1210 */
1211static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1212{
1213 struct ceph_mon_client *monc = con->private;
1214 int type = le16_to_cpu(msg->hdr.type);
1215
1216 if (!monc)
1217 return;
1218
1219 switch (type) {
1220 case CEPH_MSG_AUTH_REPLY:
1221 handle_auth_reply(monc, msg);
1222 break;
1223
1224 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1225 handle_subscribe_ack(monc, msg);
1226 break;
1227
1228 case CEPH_MSG_STATFS_REPLY:
1229 handle_statfs_reply(monc, msg);
1230 break;
1231
1232 case CEPH_MSG_MON_GET_VERSION_REPLY:
1233 handle_get_version_reply(monc, msg);
1234 break;
1235
1236 case CEPH_MSG_MON_COMMAND_ACK:
1237 handle_command_ack(monc, msg);
1238 break;
1239
1240 case CEPH_MSG_MON_MAP:
1241 ceph_monc_handle_map(monc, msg);
1242 break;
1243
1244 case CEPH_MSG_OSD_MAP:
1245 ceph_osdc_handle_map(&monc->client->osdc, msg);
1246 break;
1247
1248 default:
1249 /* can the chained handler handle it? */
1250 if (monc->client->extra_mon_dispatch &&
1251 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1252 break;
1253
1254 pr_err("received unknown message type %d %s\n", type,
1255 ceph_msg_type_name(type));
1256 }
1257 ceph_msg_put(msg);
1258}
1259
1260/*
1261 * Allocate memory for incoming message
1262 */
1263static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
1264 struct ceph_msg_header *hdr,
1265 int *skip)
1266{
1267 struct ceph_mon_client *monc = con->private;
1268 int type = le16_to_cpu(hdr->type);
1269 int front_len = le32_to_cpu(hdr->front_len);
1270 struct ceph_msg *m = NULL;
1271
1272 *skip = 0;
1273
1274 switch (type) {
1275 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1276 m = ceph_msg_get(monc->m_subscribe_ack);
1277 break;
1278 case CEPH_MSG_STATFS_REPLY:
1279 case CEPH_MSG_MON_COMMAND_ACK:
1280 return get_generic_reply(con, hdr, skip);
1281 case CEPH_MSG_AUTH_REPLY:
1282 m = ceph_msg_get(monc->m_auth_reply);
1283 break;
1284 case CEPH_MSG_MON_GET_VERSION_REPLY:
1285 if (le64_to_cpu(hdr->tid) != 0)
1286 return get_generic_reply(con, hdr, skip);
1287
1288 /*
1289 * Older OSDs don't set reply tid even if the orignal
1290 * request had a non-zero tid. Work around this weirdness
1291 * by allocating a new message.
1292 */
1293 /* fall through */
1294 case CEPH_MSG_MON_MAP:
1295 case CEPH_MSG_MDS_MAP:
1296 case CEPH_MSG_OSD_MAP:
1297 case CEPH_MSG_FS_MAP_USER:
1298 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1299 if (!m)
1300 return NULL; /* ENOMEM--return skip == 0 */
1301 break;
1302 }
1303
1304 if (!m) {
1305 pr_info("alloc_msg unknown type %d\n", type);
1306 *skip = 1;
1307 } else if (front_len > m->front_alloc_len) {
1308 pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1309 front_len, m->front_alloc_len,
1310 (unsigned int)con->peer_name.type,
1311 le64_to_cpu(con->peer_name.num));
1312 ceph_msg_put(m);
1313 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
1314 }
1315
1316 return m;
1317}
1318
1319/*
1320 * If the monitor connection resets, pick a new monitor and resubmit
1321 * any pending requests.
1322 */
1323static void mon_fault(struct ceph_connection *con)
1324{
1325 struct ceph_mon_client *monc = con->private;
1326
1327 mutex_lock(&monc->mutex);
1328 dout("%s mon%d\n", __func__, monc->cur_mon);
1329 if (monc->cur_mon >= 0) {
1330 if (!monc->hunting) {
1331 dout("%s hunting for new mon\n", __func__);
1332 reopen_session(monc);
1333 __schedule_delayed(monc);
1334 } else {
1335 dout("%s already hunting\n", __func__);
1336 }
1337 }
1338 mutex_unlock(&monc->mutex);
1339}
1340
1341/*
1342 * We can ignore refcounting on the connection struct, as all references
1343 * will come from the messenger workqueue, which is drained prior to
1344 * mon_client destruction.
1345 */
1346static struct ceph_connection *con_get(struct ceph_connection *con)
1347{
1348 return con;
1349}
1350
1351static void con_put(struct ceph_connection *con)
1352{
1353}
1354
1355static const struct ceph_connection_operations mon_con_ops = {
1356 .get = con_get,
1357 .put = con_put,
1358 .dispatch = dispatch,
1359 .fault = mon_fault,
1360 .alloc_msg = mon_alloc_msg,
1361};