blob: 500481003de455aa36be564f975f35aba2dfee2c [file] [log] [blame]
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001#include <linux/ceph/ceph_debug.h>
Sage Weilba75bb92009-10-06 11:31:11 -07002
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07003#include <linux/module.h>
Sage Weilba75bb92009-10-06 11:31:11 -07004#include <linux/types.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +09005#include <linux/slab.h>
Sage Weilba75bb92009-10-06 11:31:11 -07006#include <linux/random.h>
7#include <linux/sched.h>
8
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07009#include <linux/ceph/mon_client.h>
10#include <linux/ceph/libceph.h>
Sage Weilab434b62012-01-13 22:22:03 -080011#include <linux/ceph/debugfs.h>
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070012#include <linux/ceph/decode.h>
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070013#include <linux/ceph/auth.h>
Sage Weilba75bb92009-10-06 11:31:11 -070014
15/*
16 * Interact with Ceph monitor cluster. Handle requests for new map
17 * versions, and periodically resend as needed. Also implement
18 * statfs() and umount().
19 *
20 * A small cluster of Ceph "monitors" are responsible for managing critical
21 * cluster configuration and state information. An odd number (e.g., 3, 5)
22 * of cmon daemons use a modified version of the Paxos part-time parliament
23 * algorithm to manage the MDS map (mds cluster membership), OSD map, and
24 * list of clients who have mounted the file system.
25 *
26 * We maintain an open, active session with a monitor at all times in order to
27 * receive timely MDSMap updates. We periodically send a keepalive byte on the
28 * TCP socket to ensure we detect a failure. If the connection does break, we
29 * randomly hunt for a new monitor. Once the connection is reestablished, we
30 * resend any outstanding requests.
31 */
32
Tobias Klauser9e327892010-05-20 10:40:19 +020033static const struct ceph_connection_operations mon_con_ops;
Sage Weilba75bb92009-10-06 11:31:11 -070034
Sage Weil9bd2e6f2010-02-02 16:21:06 -080035static int __validate_auth(struct ceph_mon_client *monc);
36
Sage Weilba75bb92009-10-06 11:31:11 -070037/*
38 * Decode a monmap blob (e.g., during mount).
39 */
40struct ceph_monmap *ceph_monmap_decode(void *p, void *end)
41{
42 struct ceph_monmap *m = NULL;
43 int i, err = -EINVAL;
44 struct ceph_fsid fsid;
45 u32 epoch, num_mon;
46 u16 version;
Sage Weil4e7a5dc2009-11-18 16:19:57 -080047 u32 len;
48
49 ceph_decode_32_safe(&p, end, len, bad);
50 ceph_decode_need(&p, end, len, bad);
Sage Weilba75bb92009-10-06 11:31:11 -070051
52 dout("monmap_decode %p %p len %d\n", p, end, (int)(end-p));
53
54 ceph_decode_16_safe(&p, end, version, bad);
55
56 ceph_decode_need(&p, end, sizeof(fsid) + 2*sizeof(u32), bad);
57 ceph_decode_copy(&p, &fsid, sizeof(fsid));
Sage Weilc89136e2009-10-14 09:59:09 -070058 epoch = ceph_decode_32(&p);
Sage Weilba75bb92009-10-06 11:31:11 -070059
Sage Weilc89136e2009-10-14 09:59:09 -070060 num_mon = ceph_decode_32(&p);
Sage Weilba75bb92009-10-06 11:31:11 -070061 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]));
Sage Weil63f2d212009-11-03 15:17:56 -080072 for (i = 0; i < num_mon; i++)
73 ceph_decode_addr(&m->mon_inst[i].addr);
Sage Weilba75bb92009-10-06 11:31:11 -070074
Sage Weilba75bb92009-10-06 11:31:11 -070075 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,
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -070079 ceph_pr_addr(&m->mon_inst[i].addr.in_addr));
Sage Weilba75bb92009-10-06 11:31:11 -070080 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++)
Sage Weil103e2d32010-01-07 16:12:36 -080096 if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0)
Sage Weilba75bb92009-10-06 11:31:11 -070097 return 1;
98 return 0;
99}
100
101/*
Sage Weil5ce6e9d2010-02-15 16:22:28 -0800102 * 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);
Alex Elder6740a842012-06-01 14:56:43 -0500109 ceph_msg_revoke(monc->m_auth);
Sage Weil5ce6e9d2010-02-15 16:22:28 -0800110 ceph_msg_get(monc->m_auth); /* keep our ref */
Alex Elder67130932012-05-26 23:26:43 -0500111 ceph_con_send(&monc->con, monc->m_auth);
Sage Weil5ce6e9d2010-02-15 16:22:28 -0800112}
113
114/*
Sage Weilba75bb92009-10-06 11:31:11 -0700115 * Close monitor session, if any.
116 */
117static void __close_session(struct ceph_mon_client *monc)
118{
Sage Weilf6a2f5b2011-08-09 09:27:44 -0700119 dout("__close_session closing mon%d\n", monc->cur_mon);
Alex Elder6740a842012-06-01 14:56:43 -0500120 ceph_msg_revoke(monc->m_auth);
Sage Weil4f471e42012-07-30 18:16:40 -0700121 ceph_msg_revoke_incoming(monc->m_auth_reply);
122 ceph_msg_revoke(monc->m_subscribe);
123 ceph_msg_revoke_incoming(monc->m_subscribe_ack);
Alex Elder67130932012-05-26 23:26:43 -0500124 ceph_con_close(&monc->con);
Ilya Dryomov0e04dc22016-01-20 17:50:31 +0100125
Sage Weilf6a2f5b2011-08-09 09:27:44 -0700126 monc->pending_auth = 0;
127 ceph_auth_reset(monc->auth);
Sage Weilba75bb92009-10-06 11:31:11 -0700128}
129
130/*
Ilya Dryomov0e04dc22016-01-20 17:50:31 +0100131 * 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.
Sage Weilba75bb92009-10-06 11:31:11 -0700133 */
Ilya Dryomov0e04dc22016-01-20 17:50:31 +0100134static void pick_new_mon(struct ceph_mon_client *monc)
Sage Weilba75bb92009-10-06 11:31:11 -0700135{
Ilya Dryomov0e04dc22016-01-20 17:50:31 +0100136 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{
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800170 int ret;
Sage Weilba75bb92009-10-06 11:31:11 -0700171
Ilya Dryomov0e04dc22016-01-20 17:50:31 +0100172 pick_new_mon(monc);
Sage Weilba75bb92009-10-06 11:31:11 -0700173
Ilya Dryomov1752b502016-01-21 16:33:25 +0100174 monc->hunting = true;
Ilya Dryomov168b9092016-01-21 16:33:19 +0100175 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
Ilya Dryomov0e04dc22016-01-20 17:50:31 +0100181 monc->sub_renew_after = jiffies; /* i.e., expired */
182 monc->sub_renew_sent = 0;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800183
Ilya Dryomov0e04dc22016-01-20 17:50:31 +0100184 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);
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800187
Ilya Dryomov0e04dc22016-01-20 17:50:31 +0100188 /*
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);
Sage Weilba75bb92009-10-06 11:31:11 -0700200}
201
Ilya Dryomov1752b502016-01-21 16:33:25 +0100202static 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
Ilya Dryomov480179a2018-04-23 15:25:10 +0200212static 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
Sage Weilba75bb92009-10-06 11:31:11 -0700220/*
221 * Reschedule delayed work timer.
222 */
223static void __schedule_delayed(struct ceph_mon_client *monc)
224{
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800225 unsigned long delay;
Sage Weilba75bb92009-10-06 11:31:11 -0700226
Ilya Dryomov168b9092016-01-21 16:33:19 +0100227 if (monc->hunting)
228 delay = CEPH_MONC_HUNT_INTERVAL * monc->hunt_mult;
229 else
Ilya Dryomov58d81b12016-01-21 16:33:15 +0100230 delay = CEPH_MONC_PING_INTERVAL;
Ilya Dryomov168b9092016-01-21 16:33:19 +0100231
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800232 dout("__schedule_delayed after %lu\n", delay);
Ilya Dryomovbee3a372016-01-22 18:42:17 +0100233 mod_delayed_work(system_wq, &monc->delayed_work,
234 round_jiffies_relative(delay));
Sage Weilba75bb92009-10-06 11:31:11 -0700235}
236
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100237const char *ceph_sub_str[] = {
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100238 [CEPH_SUB_MONMAP] = "monmap",
239 [CEPH_SUB_OSDMAP] = "osdmap",
Yan, Zheng0cabbd92016-04-07 11:34:43 +0800240 [CEPH_SUB_FSMAP] = "fsmap.user",
241 [CEPH_SUB_MDSMAP] = "mdsmap",
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100242};
243
Sage Weilba75bb92009-10-06 11:31:11 -0700244/*
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100245 * Send subscribe request for one or more maps, according to
246 * monc->subs.
Sage Weilba75bb92009-10-06 11:31:11 -0700247 */
248static void __send_subscribe(struct ceph_mon_client *monc)
249{
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100250 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;
Sage Weilba75bb92009-10-06 11:31:11 -0700255
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100256 dout("%s sent %lu\n", __func__, monc->sub_renew_sent);
Sage Weilba75bb92009-10-06 11:31:11 -0700257
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100258 BUG_ON(monc->cur_mon < 0);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700259
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100260 if (!monc->sub_renew_sent)
261 monc->sub_renew_sent = jiffies | 1; /* never 0 */
Sage Weilba75bb92009-10-06 11:31:11 -0700262
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100263 msg->hdr.version = cpu_to_le16(2);
Sage Weilba75bb92009-10-06 11:31:11 -0700264
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100265 for (i = 0; i < ARRAY_SIZE(monc->subs); i++) {
266 if (monc->subs[i].want)
267 num++;
Sage Weilba75bb92009-10-06 11:31:11 -0700268 }
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100269 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++) {
Ilya Dryomov737cc81e2016-05-26 00:05:01 +0200272 char buf[32];
273 int len;
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100274
275 if (!monc->subs[i].want)
276 continue;
277
Ilya Dryomov737cc81e2016-05-26 00:05:01 +0200278 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,
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100284 le64_to_cpu(monc->subs[i].item.start),
285 monc->subs[i].item.flags);
Ilya Dryomov737cc81e2016-05-26 00:05:01 +0200286 ceph_encode_string(&p, end, buf, len);
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100287 memcpy(p, &monc->subs[i].item, sizeof(monc->subs[i].item));
288 p += sizeof(monc->subs[i].item);
289 }
290
Ilya Dryomov737cc81e2016-05-26 00:05:01 +0200291 BUG_ON(p > end);
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100292 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));
Sage Weilba75bb92009-10-06 11:31:11 -0700296}
297
298static void handle_subscribe_ack(struct ceph_mon_client *monc,
299 struct ceph_msg *msg)
300{
Eric Dumazet95c96172012-04-15 05:58:06 +0000301 unsigned int seconds;
Sage Weil07bd10f2009-10-14 17:26:40 -0700302 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
Sage Weilba75bb92009-10-06 11:31:11 -0700303
Sage Weil07bd10f2009-10-14 17:26:40 -0700304 if (msg->front.iov_len < sizeof(*h))
305 goto bad;
306 seconds = le32_to_cpu(h->duration);
307
Sage Weilba75bb92009-10-06 11:31:11 -0700308 mutex_lock(&monc->mutex);
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100309 if (monc->sub_renew_sent) {
310 monc->sub_renew_after = monc->sub_renew_sent +
311 (seconds >> 1) * HZ - 1;
312 dout("%s sent %lu duration %d renew after %lu\n", __func__,
313 monc->sub_renew_sent, seconds, monc->sub_renew_after);
314 monc->sub_renew_sent = 0;
315 } else {
316 dout("%s sent %lu renew after %lu, ignoring\n", __func__,
317 monc->sub_renew_sent, monc->sub_renew_after);
318 }
Sage Weilba75bb92009-10-06 11:31:11 -0700319 mutex_unlock(&monc->mutex);
320 return;
321bad:
322 pr_err("got corrupt subscribe-ack msg\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -0800323 ceph_msg_dump(msg);
Sage Weilba75bb92009-10-06 11:31:11 -0700324}
325
326/*
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100327 * Register interest in a map
328 *
329 * @sub: one of CEPH_SUB_*
330 * @epoch: X for "every map since X", or 0 for "just the latest"
Sage Weilba75bb92009-10-06 11:31:11 -0700331 */
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100332static bool __ceph_monc_want_map(struct ceph_mon_client *monc, int sub,
333 u32 epoch, bool continuous)
Sage Weilba75bb92009-10-06 11:31:11 -0700334{
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100335 __le64 start = cpu_to_le64(epoch);
336 u8 flags = !continuous ? CEPH_SUBSCRIBE_ONETIME : 0;
Sage Weilba75bb92009-10-06 11:31:11 -0700337
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100338 dout("%s %s epoch %u continuous %d\n", __func__, ceph_sub_str[sub],
339 epoch, continuous);
340
341 if (monc->subs[sub].want &&
342 monc->subs[sub].item.start == start &&
343 monc->subs[sub].item.flags == flags)
344 return false;
345
346 monc->subs[sub].item.start = start;
347 monc->subs[sub].item.flags = flags;
348 monc->subs[sub].want = true;
349
350 return true;
351}
352
353bool ceph_monc_want_map(struct ceph_mon_client *monc, int sub, u32 epoch,
354 bool continuous)
355{
356 bool need_request;
357
358 mutex_lock(&monc->mutex);
359 need_request = __ceph_monc_want_map(monc, sub, epoch, continuous);
360 mutex_unlock(&monc->mutex);
361
362 return need_request;
363}
364EXPORT_SYMBOL(ceph_monc_want_map);
365
366/*
367 * Keep track of which maps we have
368 *
369 * @sub: one of CEPH_SUB_*
370 */
371static void __ceph_monc_got_map(struct ceph_mon_client *monc, int sub,
372 u32 epoch)
373{
374 dout("%s %s epoch %u\n", __func__, ceph_sub_str[sub], epoch);
375
376 if (monc->subs[sub].want) {
377 if (monc->subs[sub].item.flags & CEPH_SUBSCRIBE_ONETIME)
378 monc->subs[sub].want = false;
379 else
380 monc->subs[sub].item.start = cpu_to_le64(epoch + 1);
381 }
382
383 monc->subs[sub].have = epoch;
384}
385
386void ceph_monc_got_map(struct ceph_mon_client *monc, int sub, u32 epoch)
Sage Weilba75bb92009-10-06 11:31:11 -0700387{
388 mutex_lock(&monc->mutex);
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100389 __ceph_monc_got_map(monc, sub, epoch);
Sage Weilba75bb92009-10-06 11:31:11 -0700390 mutex_unlock(&monc->mutex);
Sage Weilba75bb92009-10-06 11:31:11 -0700391}
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100392EXPORT_SYMBOL(ceph_monc_got_map);
Sage Weilba75bb92009-10-06 11:31:11 -0700393
Ilya Dryomov42c1b122016-04-28 16:07:25 +0200394void ceph_monc_renew_subs(struct ceph_mon_client *monc)
395{
396 mutex_lock(&monc->mutex);
397 __send_subscribe(monc);
398 mutex_unlock(&monc->mutex);
399}
400EXPORT_SYMBOL(ceph_monc_renew_subs);
401
Sage Weilba75bb92009-10-06 11:31:11 -0700402/*
Ilya Dryomova319bf52015-05-15 12:02:17 +0300403 * Wait for an osdmap with a given epoch.
404 *
405 * @epoch: epoch to wait for
406 * @timeout: in jiffies, 0 means "wait forever"
407 */
Ilya Dryomov6044cde2014-05-13 11:19:27 +0400408int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
409 unsigned long timeout)
410{
411 unsigned long started = jiffies;
Ilya Dryomov216639d2015-05-19 12:03:33 +0300412 long ret;
Ilya Dryomov6044cde2014-05-13 11:19:27 +0400413
414 mutex_lock(&monc->mutex);
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100415 while (monc->subs[CEPH_SUB_OSDMAP].have < epoch) {
Ilya Dryomov6044cde2014-05-13 11:19:27 +0400416 mutex_unlock(&monc->mutex);
417
Ilya Dryomova319bf52015-05-15 12:02:17 +0300418 if (timeout && time_after_eq(jiffies, started + timeout))
Ilya Dryomov6044cde2014-05-13 11:19:27 +0400419 return -ETIMEDOUT;
420
421 ret = wait_event_interruptible_timeout(monc->client->auth_wq,
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100422 monc->subs[CEPH_SUB_OSDMAP].have >= epoch,
423 ceph_timeout_jiffies(timeout));
Ilya Dryomov6044cde2014-05-13 11:19:27 +0400424 if (ret < 0)
425 return ret;
426
427 mutex_lock(&monc->mutex);
428 }
429
430 mutex_unlock(&monc->mutex);
431 return 0;
432}
433EXPORT_SYMBOL(ceph_monc_wait_osdmap);
Sage Weilba75bb92009-10-06 11:31:11 -0700434
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800435/*
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100436 * Open a session with a random monitor. Request monmap and osdmap,
437 * which are waited upon in __ceph_open_session().
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800438 */
439int ceph_monc_open_session(struct ceph_mon_client *monc)
Sage Weilba75bb92009-10-06 11:31:11 -0700440{
Sage Weilba75bb92009-10-06 11:31:11 -0700441 mutex_lock(&monc->mutex);
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100442 __ceph_monc_want_map(monc, CEPH_SUB_MONMAP, 0, true);
443 __ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, 0, false);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800444 __open_session(monc);
Sage Weilba75bb92009-10-06 11:31:11 -0700445 __schedule_delayed(monc);
446 mutex_unlock(&monc->mutex);
447 return 0;
448}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700449EXPORT_SYMBOL(ceph_monc_open_session);
Sage Weilba75bb92009-10-06 11:31:11 -0700450
Sage Weil07433042009-11-18 16:50:41 -0800451static void ceph_monc_handle_map(struct ceph_mon_client *monc,
452 struct ceph_msg *msg)
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800453{
454 struct ceph_client *client = monc->client;
455 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
456 void *p, *end;
457
458 mutex_lock(&monc->mutex);
459
460 dout("handle_monmap\n");
461 p = msg->front.iov_base;
462 end = p + msg->front.iov_len;
463
464 monmap = ceph_monmap_decode(p, end);
465 if (IS_ERR(monmap)) {
466 pr_err("problem decoding monmap, %d\n",
467 (int)PTR_ERR(monmap));
Sage Weild4a780c2009-12-11 08:55:23 -0800468 goto out;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800469 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800470
Sage Weil07433042009-11-18 16:50:41 -0800471 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800472 kfree(monmap);
Sage Weild4a780c2009-12-11 08:55:23 -0800473 goto out;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800474 }
475
476 client->monc.monmap = monmap;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800477 kfree(old);
478
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100479 __ceph_monc_got_map(monc, CEPH_SUB_MONMAP, monc->monmap->epoch);
Ilya Dryomov02ac9562016-01-06 12:56:21 +0300480 client->have_fsid = true;
Sage Weild1c338a2012-08-19 12:29:16 -0700481
Sage Weild4a780c2009-12-11 08:55:23 -0800482out:
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800483 mutex_unlock(&monc->mutex);
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700484 wake_up_all(&client->auth_wq);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800485}
Sage Weilba75bb92009-10-06 11:31:11 -0700486
Sage Weilba75bb92009-10-06 11:31:11 -0700487/*
Ilya Dryomov7a6fdeb2014-12-22 19:14:26 +0300488 * generic requests (currently statfs, mon_get_version)
Sage Weilba75bb92009-10-06 11:31:11 -0700489 */
Ilya Dryomovfcd00b62016-04-28 16:07:22 +0200490DEFINE_RB_FUNCS(generic_request, struct ceph_mon_generic_request, tid, node)
Sage Weil85ff03f2010-02-15 14:47:28 -0800491
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700492static void release_generic_request(struct kref *kref)
Sage Weil3143edd2010-03-24 21:43:33 -0700493{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700494 struct ceph_mon_generic_request *req =
495 container_of(kref, struct ceph_mon_generic_request, kref);
Sage Weil3143edd2010-03-24 21:43:33 -0700496
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200497 dout("%s greq %p request %p reply %p\n", __func__, req, req->request,
498 req->reply);
499 WARN_ON(!RB_EMPTY_NODE(&req->node));
500
Sage Weil3143edd2010-03-24 21:43:33 -0700501 if (req->reply)
502 ceph_msg_put(req->reply);
503 if (req->request)
504 ceph_msg_put(req->request);
Yehuda Sadeh20547562010-06-01 10:37:40 -0700505
506 kfree(req);
Sage Weil3143edd2010-03-24 21:43:33 -0700507}
508
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700509static void put_generic_request(struct ceph_mon_generic_request *req)
Sage Weil3143edd2010-03-24 21:43:33 -0700510{
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200511 if (req)
512 kref_put(&req->kref, release_generic_request);
Sage Weil3143edd2010-03-24 21:43:33 -0700513}
514
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700515static void get_generic_request(struct ceph_mon_generic_request *req)
Sage Weil3143edd2010-03-24 21:43:33 -0700516{
517 kref_get(&req->kref);
518}
519
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200520static struct ceph_mon_generic_request *
521alloc_generic_request(struct ceph_mon_client *monc, gfp_t gfp)
522{
523 struct ceph_mon_generic_request *req;
524
525 req = kzalloc(sizeof(*req), gfp);
526 if (!req)
527 return NULL;
528
529 req->monc = monc;
530 kref_init(&req->kref);
531 RB_CLEAR_NODE(&req->node);
532 init_completion(&req->completion);
533
534 dout("%s greq %p\n", __func__, req);
535 return req;
536}
537
538static void register_generic_request(struct ceph_mon_generic_request *req)
539{
540 struct ceph_mon_client *monc = req->monc;
541
542 WARN_ON(req->tid);
543
544 get_generic_request(req);
545 req->tid = ++monc->last_tid;
546 insert_generic_request(&monc->generic_request_tree, req);
547}
548
549static void send_generic_request(struct ceph_mon_client *monc,
550 struct ceph_mon_generic_request *req)
551{
552 WARN_ON(!req->tid);
553
554 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
555 req->request->hdr.tid = cpu_to_le64(req->tid);
556 ceph_con_send(&monc->con, ceph_msg_get(req->request));
557}
558
559static void __finish_generic_request(struct ceph_mon_generic_request *req)
560{
561 struct ceph_mon_client *monc = req->monc;
562
563 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
564 erase_generic_request(&monc->generic_request_tree, req);
565
566 ceph_msg_revoke(req->request);
567 ceph_msg_revoke_incoming(req->reply);
568}
569
570static void finish_generic_request(struct ceph_mon_generic_request *req)
571{
572 __finish_generic_request(req);
573 put_generic_request(req);
574}
575
576static void complete_generic_request(struct ceph_mon_generic_request *req)
577{
578 if (req->complete_cb)
579 req->complete_cb(req);
580 else
581 complete_all(&req->completion);
582 put_generic_request(req);
583}
584
Wei Yongjunf52ec332016-07-30 00:37:31 +0000585static void cancel_generic_request(struct ceph_mon_generic_request *req)
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200586{
587 struct ceph_mon_client *monc = req->monc;
588 struct ceph_mon_generic_request *lookup_req;
589
590 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
591
592 mutex_lock(&monc->mutex);
593 lookup_req = lookup_generic_request(&monc->generic_request_tree,
594 req->tid);
595 if (lookup_req) {
596 WARN_ON(lookup_req != req);
597 finish_generic_request(req);
598 }
599
600 mutex_unlock(&monc->mutex);
601}
602
603static int wait_generic_request(struct ceph_mon_generic_request *req)
604{
605 int ret;
606
607 dout("%s greq %p tid %llu\n", __func__, req, req->tid);
608 ret = wait_for_completion_interruptible(&req->completion);
609 if (ret)
610 cancel_generic_request(req);
611 else
612 ret = req->result; /* completed */
613
614 return ret;
615}
616
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700617static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
Sage Weil3143edd2010-03-24 21:43:33 -0700618 struct ceph_msg_header *hdr,
619 int *skip)
620{
621 struct ceph_mon_client *monc = con->private;
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700622 struct ceph_mon_generic_request *req;
Sage Weil3143edd2010-03-24 21:43:33 -0700623 u64 tid = le64_to_cpu(hdr->tid);
624 struct ceph_msg *m;
625
626 mutex_lock(&monc->mutex);
Ilya Dryomovfcd00b62016-04-28 16:07:22 +0200627 req = lookup_generic_request(&monc->generic_request_tree, tid);
Sage Weil3143edd2010-03-24 21:43:33 -0700628 if (!req) {
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700629 dout("get_generic_reply %lld dne\n", tid);
Sage Weil3143edd2010-03-24 21:43:33 -0700630 *skip = 1;
631 m = NULL;
632 } else {
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700633 dout("get_generic_reply %lld got %p\n", tid, req->reply);
Alex Elder1c20f2d2012-06-04 14:43:32 -0500634 *skip = 0;
Sage Weil3143edd2010-03-24 21:43:33 -0700635 m = ceph_msg_get(req->reply);
636 /*
637 * we don't need to track the connection reading into
638 * this reply because we only have one open connection
639 * at a time, ever.
640 */
641 }
642 mutex_unlock(&monc->mutex);
643 return m;
644}
645
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700646/*
647 * statfs
648 */
Sage Weilba75bb92009-10-06 11:31:11 -0700649static void handle_statfs_reply(struct ceph_mon_client *monc,
650 struct ceph_msg *msg)
651{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700652 struct ceph_mon_generic_request *req;
Sage Weilba75bb92009-10-06 11:31:11 -0700653 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
Sage Weil3143edd2010-03-24 21:43:33 -0700654 u64 tid = le64_to_cpu(msg->hdr.tid);
Sage Weilba75bb92009-10-06 11:31:11 -0700655
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200656 dout("%s msg %p tid %llu\n", __func__, msg, tid);
657
Sage Weilba75bb92009-10-06 11:31:11 -0700658 if (msg->front.iov_len != sizeof(*reply))
659 goto bad;
Sage Weilba75bb92009-10-06 11:31:11 -0700660
661 mutex_lock(&monc->mutex);
Ilya Dryomovfcd00b62016-04-28 16:07:22 +0200662 req = lookup_generic_request(&monc->generic_request_tree, tid);
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200663 if (!req) {
664 mutex_unlock(&monc->mutex);
665 return;
Sage Weilba75bb92009-10-06 11:31:11 -0700666 }
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200667
668 req->result = 0;
669 *req->u.st = reply->st; /* struct */
670 __finish_generic_request(req);
Sage Weilba75bb92009-10-06 11:31:11 -0700671 mutex_unlock(&monc->mutex);
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200672
673 complete_generic_request(req);
Sage Weilba75bb92009-10-06 11:31:11 -0700674 return;
675
676bad:
Ilya Dryomov7a6fdeb2014-12-22 19:14:26 +0300677 pr_err("corrupt statfs reply, tid %llu\n", tid);
Sage Weil9ec7cab2009-12-14 15:13:47 -0800678 ceph_msg_dump(msg);
Sage Weilba75bb92009-10-06 11:31:11 -0700679}
680
681/*
Sage Weilba75bb92009-10-06 11:31:11 -0700682 * Do a synchronous statfs().
683 */
684int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
685{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700686 struct ceph_mon_generic_request *req;
Sage Weil3143edd2010-03-24 21:43:33 -0700687 struct ceph_mon_statfs *h;
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200688 int ret = -ENOMEM;
Sage Weilba75bb92009-10-06 11:31:11 -0700689
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200690 req = alloc_generic_request(monc, GFP_NOFS);
Sage Weil3143edd2010-03-24 21:43:33 -0700691 if (!req)
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200692 goto out;
Sage Weilba75bb92009-10-06 11:31:11 -0700693
Sage Weilb61c2762011-08-09 15:03:46 -0700694 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
695 true);
Sage Weila79832f2010-04-01 16:06:19 -0700696 if (!req->request)
Sage Weil3143edd2010-03-24 21:43:33 -0700697 goto out;
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200698
699 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 64, GFP_NOFS, true);
Sage Weila79832f2010-04-01 16:06:19 -0700700 if (!req->reply)
Sage Weil3143edd2010-03-24 21:43:33 -0700701 goto out;
Sage Weil3143edd2010-03-24 21:43:33 -0700702
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200703 req->u.st = buf;
704
705 mutex_lock(&monc->mutex);
706 register_generic_request(req);
Sage Weil3143edd2010-03-24 21:43:33 -0700707 /* fill out request */
708 h = req->request->front.iov_base;
709 h->monhdr.have_version = 0;
710 h->monhdr.session_mon = cpu_to_le16(-1);
711 h->monhdr.session_mon_tid = 0;
712 h->fsid = monc->monmap->fsid;
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200713 send_generic_request(monc, req);
714 mutex_unlock(&monc->mutex);
Sage Weilba75bb92009-10-06 11:31:11 -0700715
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200716 ret = wait_generic_request(req);
Sage Weil3143edd2010-03-24 21:43:33 -0700717out:
Ilya Dryomovf6469122014-12-22 19:35:17 +0300718 put_generic_request(req);
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200719 return ret;
Sage Weilba75bb92009-10-06 11:31:11 -0700720}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700721EXPORT_SYMBOL(ceph_monc_do_statfs);
Sage Weilba75bb92009-10-06 11:31:11 -0700722
Ilya Dryomov513a8242014-05-13 11:19:26 +0400723static void handle_get_version_reply(struct ceph_mon_client *monc,
724 struct ceph_msg *msg)
725{
726 struct ceph_mon_generic_request *req;
727 u64 tid = le64_to_cpu(msg->hdr.tid);
728 void *p = msg->front.iov_base;
729 void *end = p + msg->front_alloc_len;
730 u64 handle;
731
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200732 dout("%s msg %p tid %llu\n", __func__, msg, tid);
Ilya Dryomov513a8242014-05-13 11:19:26 +0400733
734 ceph_decode_need(&p, end, 2*sizeof(u64), bad);
735 handle = ceph_decode_64(&p);
736 if (tid != 0 && tid != handle)
737 goto bad;
738
739 mutex_lock(&monc->mutex);
Ilya Dryomovfcd00b62016-04-28 16:07:22 +0200740 req = lookup_generic_request(&monc->generic_request_tree, handle);
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200741 if (!req) {
742 mutex_unlock(&monc->mutex);
743 return;
Ilya Dryomov513a8242014-05-13 11:19:26 +0400744 }
745
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200746 req->result = 0;
747 req->u.newest = ceph_decode_64(&p);
748 __finish_generic_request(req);
749 mutex_unlock(&monc->mutex);
750
751 complete_generic_request(req);
Ilya Dryomov513a8242014-05-13 11:19:26 +0400752 return;
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200753
Ilya Dryomov513a8242014-05-13 11:19:26 +0400754bad:
Ilya Dryomov7a6fdeb2014-12-22 19:14:26 +0300755 pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
Ilya Dryomov513a8242014-05-13 11:19:26 +0400756 ceph_msg_dump(msg);
757}
758
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200759static struct ceph_mon_generic_request *
760__ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
761 ceph_monc_callback_t cb, u64 private_data)
762{
763 struct ceph_mon_generic_request *req;
764
765 req = alloc_generic_request(monc, GFP_NOIO);
766 if (!req)
767 goto err_put_req;
768
769 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
770 sizeof(u64) + sizeof(u32) + strlen(what),
771 GFP_NOIO, true);
772 if (!req->request)
773 goto err_put_req;
774
775 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 32, GFP_NOIO,
776 true);
777 if (!req->reply)
778 goto err_put_req;
779
780 req->complete_cb = cb;
781 req->private_data = private_data;
782
783 mutex_lock(&monc->mutex);
784 register_generic_request(req);
785 {
786 void *p = req->request->front.iov_base;
787 void *const end = p + req->request->front_alloc_len;
788
789 ceph_encode_64(&p, req->tid); /* handle */
790 ceph_encode_string(&p, end, what, strlen(what));
791 WARN_ON(p != end);
792 }
793 send_generic_request(monc, req);
794 mutex_unlock(&monc->mutex);
795
796 return req;
797
798err_put_req:
799 put_generic_request(req);
800 return ERR_PTR(-ENOMEM);
801}
802
Ilya Dryomov513a8242014-05-13 11:19:26 +0400803/*
804 * Send MMonGetVersion and wait for the reply.
805 *
806 * @what: one of "mdsmap", "osdmap" or "monmap"
807 */
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200808int ceph_monc_get_version(struct ceph_mon_client *monc, const char *what,
809 u64 *newest)
Ilya Dryomov513a8242014-05-13 11:19:26 +0400810{
811 struct ceph_mon_generic_request *req;
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200812 int ret;
Ilya Dryomov513a8242014-05-13 11:19:26 +0400813
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200814 req = __ceph_monc_get_version(monc, what, NULL, 0);
815 if (IS_ERR(req))
816 return PTR_ERR(req);
Ilya Dryomov513a8242014-05-13 11:19:26 +0400817
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200818 ret = wait_generic_request(req);
819 if (!ret)
820 *newest = req->u.newest;
Ilya Dryomov513a8242014-05-13 11:19:26 +0400821
Ilya Dryomovf6469122014-12-22 19:35:17 +0300822 put_generic_request(req);
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200823 return ret;
Ilya Dryomov513a8242014-05-13 11:19:26 +0400824}
Ilya Dryomovd0b19702016-04-28 16:07:27 +0200825EXPORT_SYMBOL(ceph_monc_get_version);
826
827/*
828 * Send MMonGetVersion,
829 *
830 * @what: one of "mdsmap", "osdmap" or "monmap"
831 */
832int ceph_monc_get_version_async(struct ceph_mon_client *monc, const char *what,
833 ceph_monc_callback_t cb, u64 private_data)
834{
835 struct ceph_mon_generic_request *req;
836
837 req = __ceph_monc_get_version(monc, what, cb, private_data);
838 if (IS_ERR(req))
839 return PTR_ERR(req);
840
841 put_generic_request(req);
842 return 0;
843}
844EXPORT_SYMBOL(ceph_monc_get_version_async);
Ilya Dryomov513a8242014-05-13 11:19:26 +0400845
Douglas Fuller6305a3b2015-07-22 20:59:52 -0400846static void handle_command_ack(struct ceph_mon_client *monc,
847 struct ceph_msg *msg)
848{
849 struct ceph_mon_generic_request *req;
850 void *p = msg->front.iov_base;
851 void *const end = p + msg->front_alloc_len;
852 u64 tid = le64_to_cpu(msg->hdr.tid);
853
854 dout("%s msg %p tid %llu\n", __func__, msg, tid);
855
856 ceph_decode_need(&p, end, sizeof(struct ceph_mon_request_header) +
857 sizeof(u32), bad);
858 p += sizeof(struct ceph_mon_request_header);
859
860 mutex_lock(&monc->mutex);
861 req = lookup_generic_request(&monc->generic_request_tree, tid);
862 if (!req) {
863 mutex_unlock(&monc->mutex);
864 return;
865 }
866
867 req->result = ceph_decode_32(&p);
868 __finish_generic_request(req);
869 mutex_unlock(&monc->mutex);
870
871 complete_generic_request(req);
872 return;
873
874bad:
875 pr_err("corrupt mon_command ack, tid %llu\n", tid);
876 ceph_msg_dump(msg);
877}
878
879int ceph_monc_blacklist_add(struct ceph_mon_client *monc,
880 struct ceph_entity_addr *client_addr)
881{
882 struct ceph_mon_generic_request *req;
883 struct ceph_mon_command *h;
884 int ret = -ENOMEM;
885 int len;
886
887 req = alloc_generic_request(monc, GFP_NOIO);
888 if (!req)
889 goto out;
890
891 req->request = ceph_msg_new(CEPH_MSG_MON_COMMAND, 256, GFP_NOIO, true);
892 if (!req->request)
893 goto out;
894
895 req->reply = ceph_msg_new(CEPH_MSG_MON_COMMAND_ACK, 512, GFP_NOIO,
896 true);
897 if (!req->reply)
898 goto out;
899
900 mutex_lock(&monc->mutex);
901 register_generic_request(req);
902 h = req->request->front.iov_base;
903 h->monhdr.have_version = 0;
904 h->monhdr.session_mon = cpu_to_le16(-1);
905 h->monhdr.session_mon_tid = 0;
906 h->fsid = monc->monmap->fsid;
907 h->num_strs = cpu_to_le32(1);
908 len = sprintf(h->str, "{ \"prefix\": \"osd blacklist\", \
909 \"blacklistop\": \"add\", \
910 \"addr\": \"%pISpc/%u\" }",
911 &client_addr->in_addr, le32_to_cpu(client_addr->nonce));
912 h->str_len = cpu_to_le32(len);
913 send_generic_request(monc, req);
914 mutex_unlock(&monc->mutex);
915
916 ret = wait_generic_request(req);
917out:
918 put_generic_request(req);
919 return ret;
920}
921EXPORT_SYMBOL(ceph_monc_blacklist_add);
922
Sage Weilba75bb92009-10-06 11:31:11 -0700923/*
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700924 * Resend pending generic requests.
Sage Weilba75bb92009-10-06 11:31:11 -0700925 */
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700926static void __resend_generic_request(struct ceph_mon_client *monc)
Sage Weilba75bb92009-10-06 11:31:11 -0700927{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700928 struct ceph_mon_generic_request *req;
Sage Weil85ff03f2010-02-15 14:47:28 -0800929 struct rb_node *p;
Sage Weilba75bb92009-10-06 11:31:11 -0700930
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700931 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
932 req = rb_entry(p, struct ceph_mon_generic_request, node);
Alex Elder6740a842012-06-01 14:56:43 -0500933 ceph_msg_revoke(req->request);
Sage Weil4f471e42012-07-30 18:16:40 -0700934 ceph_msg_revoke_incoming(req->reply);
Alex Elder67130932012-05-26 23:26:43 -0500935 ceph_con_send(&monc->con, ceph_msg_get(req->request));
Sage Weilba75bb92009-10-06 11:31:11 -0700936 }
937}
938
939/*
940 * Delayed work. If we haven't mounted yet, retry. Otherwise,
941 * renew/retry subscription as needed (in case it is timing out, or we
942 * got an ENOMEM). And keep the monitor connection alive.
943 */
944static void delayed_work(struct work_struct *work)
945{
946 struct ceph_mon_client *monc =
947 container_of(work, struct ceph_mon_client, delayed_work.work);
948
949 dout("monc delayed_work\n");
950 mutex_lock(&monc->mutex);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800951 if (monc->hunting) {
Ilya Dryomov1752b502016-01-21 16:33:25 +0100952 dout("%s continuing hunt\n", __func__);
953 reopen_session(monc);
Sage Weilba75bb92009-10-06 11:31:11 -0700954 } else {
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800955 int is_auth = ceph_auth_is_authenticated(monc->auth);
956 if (ceph_con_keepalive_expired(&monc->con,
Ilya Dryomov58d81b12016-01-21 16:33:15 +0100957 CEPH_MONC_PING_TIMEOUT)) {
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800958 dout("monc keepalive timeout\n");
959 is_auth = 0;
Ilya Dryomov1752b502016-01-21 16:33:25 +0100960 reopen_session(monc);
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800961 }
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800962
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800963 if (!monc->hunting) {
964 ceph_con_keepalive(&monc->con);
965 __validate_auth(monc);
Ilya Dryomov480179a2018-04-23 15:25:10 +0200966 un_backoff(monc);
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800967 }
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800968
Ilya Dryomov82dcaba2016-01-19 16:19:06 +0100969 if (is_auth) {
970 unsigned long now = jiffies;
971
972 dout("%s renew subs? now %lu renew after %lu\n",
973 __func__, now, monc->sub_renew_after);
974 if (time_after_eq(now, monc->sub_renew_after))
975 __send_subscribe(monc);
976 }
Sage Weilba75bb92009-10-06 11:31:11 -0700977 }
Sage Weilba75bb92009-10-06 11:31:11 -0700978 __schedule_delayed(monc);
979 mutex_unlock(&monc->mutex);
980}
981
Sage Weil6b805182009-10-27 11:50:50 -0700982/*
983 * On startup, we build a temporary monmap populated with the IPs
984 * provided by mount(2).
985 */
986static int build_initial_monmap(struct ceph_mon_client *monc)
987{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700988 struct ceph_options *opt = monc->client->options;
989 struct ceph_entity_addr *mon_addr = opt->mon_addr;
990 int num_mon = opt->num_mon;
Sage Weil6b805182009-10-27 11:50:50 -0700991 int i;
992
993 /* build initial monmap */
994 monc->monmap = kzalloc(sizeof(*monc->monmap) +
995 num_mon*sizeof(monc->monmap->mon_inst[0]),
996 GFP_KERNEL);
997 if (!monc->monmap)
998 return -ENOMEM;
999 for (i = 0; i < num_mon; i++) {
1000 monc->monmap->mon_inst[i].addr = mon_addr[i];
Sage Weil6b805182009-10-27 11:50:50 -07001001 monc->monmap->mon_inst[i].addr.nonce = 0;
1002 monc->monmap->mon_inst[i].name.type =
1003 CEPH_ENTITY_TYPE_MON;
1004 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
1005 }
1006 monc->monmap->num_mon = num_mon;
Sage Weil6b805182009-10-27 11:50:50 -07001007 return 0;
1008}
1009
Sage Weilba75bb92009-10-06 11:31:11 -07001010int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
1011{
1012 int err = 0;
1013
1014 dout("init\n");
1015 memset(monc, 0, sizeof(*monc));
1016 monc->client = cl;
1017 monc->monmap = NULL;
1018 mutex_init(&monc->mutex);
1019
Sage Weil6b805182009-10-27 11:50:50 -07001020 err = build_initial_monmap(monc);
1021 if (err)
1022 goto out;
1023
Sage Weilf6a2f5b2011-08-09 09:27:44 -07001024 /* connection */
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001025 /* authentication */
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001026 monc->auth = ceph_auth_init(cl->options->name,
Tommi Virtanen8323c3a2011-03-25 16:32:57 -07001027 cl->options->key);
Noah Watkins49d92242011-09-12 14:51:58 -07001028 if (IS_ERR(monc->auth)) {
1029 err = PTR_ERR(monc->auth);
Alex Elder67130932012-05-26 23:26:43 -05001030 goto out_monmap;
Noah Watkins49d92242011-09-12 14:51:58 -07001031 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001032 monc->auth->want_keys =
1033 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
1034 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
1035
Sage Weil240ed682010-05-21 14:57:25 -07001036 /* msgs */
Sage Weila79832f2010-04-01 16:06:19 -07001037 err = -ENOMEM;
Sage Weil7c315c52010-03-24 21:52:30 -07001038 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
Yehuda Sadeh34d23762010-04-06 14:33:58 -07001039 sizeof(struct ceph_mon_subscribe_ack),
Sage Weilb61c2762011-08-09 15:03:46 -07001040 GFP_NOFS, true);
Sage Weila79832f2010-04-01 16:06:19 -07001041 if (!monc->m_subscribe_ack)
Noah Watkins49d92242011-09-12 14:51:58 -07001042 goto out_auth;
Sage Weil6694d6b2010-03-24 21:48:05 -07001043
Ilya Dryomov737cc81e2016-05-26 00:05:01 +02001044 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 128, GFP_NOFS,
Sage Weilb61c2762011-08-09 15:03:46 -07001045 true);
Sage Weil240ed682010-05-21 14:57:25 -07001046 if (!monc->m_subscribe)
1047 goto out_subscribe_ack;
1048
Sage Weilb61c2762011-08-09 15:03:46 -07001049 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
1050 true);
Sage Weila79832f2010-04-01 16:06:19 -07001051 if (!monc->m_auth_reply)
Sage Weil240ed682010-05-21 14:57:25 -07001052 goto out_subscribe;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001053
Sage Weilb61c2762011-08-09 15:03:46 -07001054 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08001055 monc->pending_auth = 0;
Sage Weila79832f2010-04-01 16:06:19 -07001056 if (!monc->m_auth)
Sage Weil6694d6b2010-03-24 21:48:05 -07001057 goto out_auth_reply;
Sage Weilba75bb92009-10-06 11:31:11 -07001058
Sage Weil735a72e2012-06-27 12:24:34 -07001059 ceph_con_init(&monc->con, monc, &mon_con_ops,
1060 &monc->client->msgr);
1061
Sage Weilba75bb92009-10-06 11:31:11 -07001062 monc->cur_mon = -1;
Ilya Dryomov168b9092016-01-21 16:33:19 +01001063 monc->had_a_connection = false;
1064 monc->hunt_mult = 1;
Sage Weilba75bb92009-10-06 11:31:11 -07001065
1066 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -07001067 monc->generic_request_tree = RB_ROOT;
Sage Weilba75bb92009-10-06 11:31:11 -07001068 monc->last_tid = 0;
1069
Ilya Dryomov737cc81e2016-05-26 00:05:01 +02001070 monc->fs_cluster_id = CEPH_FS_CLUSTER_ID_NONE;
1071
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001072 return 0;
1073
Sage Weil6694d6b2010-03-24 21:48:05 -07001074out_auth_reply:
1075 ceph_msg_put(monc->m_auth_reply);
Sage Weil240ed682010-05-21 14:57:25 -07001076out_subscribe:
1077 ceph_msg_put(monc->m_subscribe);
Sage Weil7c315c52010-03-24 21:52:30 -07001078out_subscribe_ack:
1079 ceph_msg_put(monc->m_subscribe_ack);
Noah Watkins49d92242011-09-12 14:51:58 -07001080out_auth:
1081 ceph_auth_destroy(monc->auth);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001082out_monmap:
1083 kfree(monc->monmap);
Sage Weilba75bb92009-10-06 11:31:11 -07001084out:
1085 return err;
1086}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001087EXPORT_SYMBOL(ceph_monc_init);
Sage Weilba75bb92009-10-06 11:31:11 -07001088
1089void ceph_monc_stop(struct ceph_mon_client *monc)
1090{
1091 dout("stop\n");
1092 cancel_delayed_work_sync(&monc->delayed_work);
1093
1094 mutex_lock(&monc->mutex);
1095 __close_session(monc);
Ilya Dryomov0e04dc22016-01-20 17:50:31 +01001096 monc->cur_mon = -1;
Sage Weilba75bb92009-10-06 11:31:11 -07001097 mutex_unlock(&monc->mutex);
1098
Sage Weilf3dea7e2012-06-10 20:43:56 -07001099 /*
1100 * flush msgr queue before we destroy ourselves to ensure that:
1101 * - any work that references our embedded con is finished.
1102 * - any osd_client or other work that may reference an authorizer
1103 * finishes before we shut down the auth subsystem.
1104 */
1105 ceph_msgr_flush();
1106
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001107 ceph_auth_destroy(monc->auth);
1108
Ilya Dryomovd0b19702016-04-28 16:07:27 +02001109 WARN_ON(!RB_EMPTY_ROOT(&monc->generic_request_tree));
1110
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001111 ceph_msg_put(monc->m_auth);
Sage Weil6694d6b2010-03-24 21:48:05 -07001112 ceph_msg_put(monc->m_auth_reply);
Sage Weil240ed682010-05-21 14:57:25 -07001113 ceph_msg_put(monc->m_subscribe);
Sage Weil7c315c52010-03-24 21:52:30 -07001114 ceph_msg_put(monc->m_subscribe_ack);
Sage Weilba75bb92009-10-06 11:31:11 -07001115
1116 kfree(monc->monmap);
1117}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001118EXPORT_SYMBOL(ceph_monc_stop);
Sage Weilba75bb92009-10-06 11:31:11 -07001119
Ilya Dryomov0f9af162016-01-08 21:17:22 +03001120static void finish_hunting(struct ceph_mon_client *monc)
1121{
1122 if (monc->hunting) {
1123 dout("%s found mon%d\n", __func__, monc->cur_mon);
1124 monc->hunting = false;
Ilya Dryomov168b9092016-01-21 16:33:19 +01001125 monc->had_a_connection = true;
Ilya Dryomov480179a2018-04-23 15:25:10 +02001126 un_backoff(monc);
Ilya Dryomov6044c692018-04-23 15:25:10 +02001127 __schedule_delayed(monc);
Ilya Dryomov0f9af162016-01-08 21:17:22 +03001128 }
1129}
1130
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001131static void handle_auth_reply(struct ceph_mon_client *monc,
1132 struct ceph_msg *msg)
1133{
1134 int ret;
Sage Weil09c4d6a2010-05-25 15:38:06 -07001135 int was_auth = 0;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001136
1137 mutex_lock(&monc->mutex);
Sage Weil27859f92013-03-25 10:26:14 -07001138 was_auth = ceph_auth_is_authenticated(monc->auth);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08001139 monc->pending_auth = 0;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001140 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
1141 msg->front.iov_len,
1142 monc->m_auth->front.iov_base,
Ilya Dryomov3cea4c32014-01-09 20:08:21 +02001143 monc->m_auth->front_alloc_len);
Ilya Dryomov0f9af162016-01-08 21:17:22 +03001144 if (ret > 0) {
1145 __send_prepared_auth_request(monc, ret);
1146 goto out;
1147 }
1148
1149 finish_hunting(monc);
1150
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001151 if (ret < 0) {
Sage Weil9bd2e6f2010-02-02 16:21:06 -08001152 monc->client->auth_err = ret;
Sage Weil27859f92013-03-25 10:26:14 -07001153 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001154 dout("authenticated, starting session\n");
Sage Weil07433042009-11-18 16:50:41 -08001155
Alex Elder15d98822012-05-26 23:26:43 -05001156 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
1157 monc->client->msgr.inst.name.num =
Yehuda Sadeh0cf55372010-06-11 15:57:06 -07001158 cpu_to_le64(monc->auth->global_id);
Sage Weil07433042009-11-18 16:50:41 -08001159
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001160 __send_subscribe(monc);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -07001161 __resend_generic_request(monc);
Ilya Dryomov0f9af162016-01-08 21:17:22 +03001162
1163 pr_info("mon%d %s session established\n", monc->cur_mon,
1164 ceph_pr_addr(&monc->con.peer_addr.in_addr));
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001165 }
Ilya Dryomov0f9af162016-01-08 21:17:22 +03001166
1167out:
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001168 mutex_unlock(&monc->mutex);
Ilya Dryomov0f9af162016-01-08 21:17:22 +03001169 if (monc->client->auth_err < 0)
1170 wake_up_all(&monc->client->auth_wq);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001171}
1172
Sage Weil9bd2e6f2010-02-02 16:21:06 -08001173static int __validate_auth(struct ceph_mon_client *monc)
1174{
1175 int ret;
1176
1177 if (monc->pending_auth)
1178 return 0;
1179
1180 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
Ilya Dryomov3cea4c32014-01-09 20:08:21 +02001181 monc->m_auth->front_alloc_len);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08001182 if (ret <= 0)
1183 return ret; /* either an error, or no need to authenticate */
1184 __send_prepared_auth_request(monc, ret);
1185 return 0;
1186}
1187
1188int ceph_monc_validate_auth(struct ceph_mon_client *monc)
1189{
1190 int ret;
1191
1192 mutex_lock(&monc->mutex);
1193 ret = __validate_auth(monc);
1194 mutex_unlock(&monc->mutex);
1195 return ret;
1196}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001197EXPORT_SYMBOL(ceph_monc_validate_auth);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08001198
Sage Weilba75bb92009-10-06 11:31:11 -07001199/*
1200 * handle incoming message
1201 */
1202static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1203{
1204 struct ceph_mon_client *monc = con->private;
1205 int type = le16_to_cpu(msg->hdr.type);
1206
1207 if (!monc)
1208 return;
1209
1210 switch (type) {
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001211 case CEPH_MSG_AUTH_REPLY:
1212 handle_auth_reply(monc, msg);
Sage Weilba75bb92009-10-06 11:31:11 -07001213 break;
1214
1215 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1216 handle_subscribe_ack(monc, msg);
1217 break;
1218
1219 case CEPH_MSG_STATFS_REPLY:
1220 handle_statfs_reply(monc, msg);
1221 break;
1222
Ilya Dryomov513a8242014-05-13 11:19:26 +04001223 case CEPH_MSG_MON_GET_VERSION_REPLY:
1224 handle_get_version_reply(monc, msg);
1225 break;
1226
Douglas Fuller6305a3b2015-07-22 20:59:52 -04001227 case CEPH_MSG_MON_COMMAND_ACK:
1228 handle_command_ack(monc, msg);
1229 break;
1230
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001231 case CEPH_MSG_MON_MAP:
1232 ceph_monc_handle_map(monc, msg);
1233 break;
1234
Sage Weilba75bb92009-10-06 11:31:11 -07001235 case CEPH_MSG_OSD_MAP:
1236 ceph_osdc_handle_map(&monc->client->osdc, msg);
1237 break;
1238
1239 default:
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001240 /* can the chained handler handle it? */
1241 if (monc->client->extra_mon_dispatch &&
1242 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1243 break;
1244
Sage Weilba75bb92009-10-06 11:31:11 -07001245 pr_err("received unknown message type %d %s\n", type,
1246 ceph_msg_type_name(type));
1247 }
1248 ceph_msg_put(msg);
1249}
1250
1251/*
1252 * Allocate memory for incoming message
1253 */
1254static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
Yehuda Sadeh24504182010-01-08 13:58:34 -08001255 struct ceph_msg_header *hdr,
1256 int *skip)
Sage Weilba75bb92009-10-06 11:31:11 -07001257{
1258 struct ceph_mon_client *monc = con->private;
1259 int type = le16_to_cpu(hdr->type);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001260 int front_len = le32_to_cpu(hdr->front_len);
Sage Weil5b3a4db2010-02-19 21:43:23 -08001261 struct ceph_msg *m = NULL;
Sage Weilba75bb92009-10-06 11:31:11 -07001262
Yehuda Sadeh24504182010-01-08 13:58:34 -08001263 *skip = 0;
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08001264
Sage Weilba75bb92009-10-06 11:31:11 -07001265 switch (type) {
Sage Weilba75bb92009-10-06 11:31:11 -07001266 case CEPH_MSG_MON_SUBSCRIBE_ACK:
Sage Weil7c315c52010-03-24 21:52:30 -07001267 m = ceph_msg_get(monc->m_subscribe_ack);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001268 break;
Sage Weilba75bb92009-10-06 11:31:11 -07001269 case CEPH_MSG_STATFS_REPLY:
Douglas Fuller6305a3b2015-07-22 20:59:52 -04001270 case CEPH_MSG_MON_COMMAND_ACK:
Yehuda Sadehf8c76f62010-04-22 15:40:37 -07001271 return get_generic_reply(con, hdr, skip);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001272 case CEPH_MSG_AUTH_REPLY:
Sage Weil6694d6b2010-03-24 21:48:05 -07001273 m = ceph_msg_get(monc->m_auth_reply);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001274 break;
Ilya Dryomov513a8242014-05-13 11:19:26 +04001275 case CEPH_MSG_MON_GET_VERSION_REPLY:
1276 if (le64_to_cpu(hdr->tid) != 0)
1277 return get_generic_reply(con, hdr, skip);
1278
1279 /*
1280 * Older OSDs don't set reply tid even if the orignal
1281 * request had a non-zero tid. Workaround this weirdness
1282 * by falling through to the allocate case.
1283 */
Sage Weil5b3a4db2010-02-19 21:43:23 -08001284 case CEPH_MSG_MON_MAP:
1285 case CEPH_MSG_MDS_MAP:
1286 case CEPH_MSG_OSD_MAP:
Yan, Zheng0cabbd92016-04-07 11:34:43 +08001287 case CEPH_MSG_FS_MAP_USER:
Sage Weilb61c2762011-08-09 15:03:46 -07001288 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
Alex Elder1c20f2d2012-06-04 14:43:32 -05001289 if (!m)
1290 return NULL; /* ENOMEM--return skip == 0 */
Sage Weil5b3a4db2010-02-19 21:43:23 -08001291 break;
Sage Weilba75bb92009-10-06 11:31:11 -07001292 }
Yehuda Sadeh24504182010-01-08 13:58:34 -08001293
Sage Weil5b3a4db2010-02-19 21:43:23 -08001294 if (!m) {
1295 pr_info("alloc_msg unknown type %d\n", type);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001296 *skip = 1;
Sage Weil73c3d482014-08-04 07:01:54 -07001297 } else if (front_len > m->front_alloc_len) {
Joe Perchesb9a67892014-09-09 21:17:29 -07001298 pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1299 front_len, m->front_alloc_len,
1300 (unsigned int)con->peer_name.type,
1301 le64_to_cpu(con->peer_name.num));
Sage Weil73c3d482014-08-04 07:01:54 -07001302 ceph_msg_put(m);
1303 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
Sage Weil5b3a4db2010-02-19 21:43:23 -08001304 }
Sage Weil73c3d482014-08-04 07:01:54 -07001305
Yehuda Sadeh24504182010-01-08 13:58:34 -08001306 return m;
Sage Weilba75bb92009-10-06 11:31:11 -07001307}
1308
1309/*
1310 * If the monitor connection resets, pick a new monitor and resubmit
1311 * any pending requests.
1312 */
1313static void mon_fault(struct ceph_connection *con)
1314{
1315 struct ceph_mon_client *monc = con->private;
1316
Sage Weilba75bb92009-10-06 11:31:11 -07001317 mutex_lock(&monc->mutex);
Ilya Dryomovb5d91702016-01-23 15:57:51 +01001318 dout("%s mon%d\n", __func__, monc->cur_mon);
1319 if (monc->cur_mon >= 0) {
1320 if (!monc->hunting) {
1321 dout("%s hunting for new mon\n", __func__);
1322 reopen_session(monc);
1323 __schedule_delayed(monc);
1324 } else {
1325 dout("%s already hunting\n", __func__);
1326 }
Sage Weilba75bb92009-10-06 11:31:11 -07001327 }
Sage Weilba75bb92009-10-06 11:31:11 -07001328 mutex_unlock(&monc->mutex);
1329}
1330
Sage Weilec87ef42012-05-31 20:27:50 -07001331/*
1332 * We can ignore refcounting on the connection struct, as all references
1333 * will come from the messenger workqueue, which is drained prior to
1334 * mon_client destruction.
1335 */
1336static struct ceph_connection *con_get(struct ceph_connection *con)
1337{
1338 return con;
1339}
1340
1341static void con_put(struct ceph_connection *con)
1342{
1343}
1344
Tobias Klauser9e327892010-05-20 10:40:19 +02001345static const struct ceph_connection_operations mon_con_ops = {
Sage Weilec87ef42012-05-31 20:27:50 -07001346 .get = con_get,
1347 .put = con_put,
Sage Weilba75bb92009-10-06 11:31:11 -07001348 .dispatch = dispatch,
1349 .fault = mon_fault,
1350 .alloc_msg = mon_alloc_msg,
Sage Weilba75bb92009-10-06 11:31:11 -07001351};