blob: 5ab737ce4f6de680e4ef131ddeb9aa3fba0f3fbe [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);
Sage Weilf6a2f5b2011-08-09 09:27:44 -0700125 monc->cur_mon = -1;
126 monc->pending_auth = 0;
127 ceph_auth_reset(monc->auth);
Sage Weilba75bb92009-10-06 11:31:11 -0700128}
129
130/*
131 * Open a session with a (new) monitor.
132 */
133static int __open_session(struct ceph_mon_client *monc)
134{
135 char r;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800136 int ret;
Sage Weilba75bb92009-10-06 11:31:11 -0700137
138 if (monc->cur_mon < 0) {
139 get_random_bytes(&r, 1);
140 monc->cur_mon = r % monc->monmap->num_mon;
141 dout("open_session num=%d r=%d -> mon%d\n",
142 monc->monmap->num_mon, r, monc->cur_mon);
143 monc->sub_sent = 0;
144 monc->sub_renew_after = jiffies; /* i.e., expired */
145 monc->want_next_osdmap = !!monc->want_next_osdmap;
146
Alex Elder20581c12012-05-26 23:26:43 -0500147 dout("open_session mon%d opening\n", monc->cur_mon);
Alex Elder67130932012-05-26 23:26:43 -0500148 ceph_con_open(&monc->con,
Sage Weilb7a9e5d2012-06-27 12:24:08 -0700149 CEPH_ENTITY_TYPE_MON, monc->cur_mon,
Sage Weilba75bb92009-10-06 11:31:11 -0700150 &monc->monmap->mon_inst[monc->cur_mon].addr);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800151
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800152 /* send an initial keepalive to ensure our timestamp is
153 * valid by the time we are in an OPENED state */
154 ceph_con_keepalive(&monc->con);
155
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800156 /* initiatiate authentication handshake */
157 ret = ceph_auth_build_hello(monc->auth,
158 monc->m_auth->front.iov_base,
Ilya Dryomov3cea4c32014-01-09 20:08:21 +0200159 monc->m_auth->front_alloc_len);
Sage Weil5ce6e9d2010-02-15 16:22:28 -0800160 __send_prepared_auth_request(monc, ret);
Sage Weilba75bb92009-10-06 11:31:11 -0700161 } else {
162 dout("open_session mon%d already open\n", monc->cur_mon);
163 }
164 return 0;
165}
166
167static bool __sub_expired(struct ceph_mon_client *monc)
168{
169 return time_after_eq(jiffies, monc->sub_renew_after);
170}
171
172/*
173 * Reschedule delayed work timer.
174 */
175static void __schedule_delayed(struct ceph_mon_client *monc)
176{
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800177 struct ceph_options *opt = monc->client->options;
178 unsigned long delay;
Sage Weilba75bb92009-10-06 11:31:11 -0700179
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800180 if (monc->cur_mon < 0 || __sub_expired(monc)) {
Sage Weilba75bb92009-10-06 11:31:11 -0700181 delay = 10 * HZ;
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800182 } else {
Sage Weilba75bb92009-10-06 11:31:11 -0700183 delay = 20 * HZ;
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800184 if (opt->monc_ping_timeout > 0)
185 delay = min(delay, opt->monc_ping_timeout / 3);
186 }
187 dout("__schedule_delayed after %lu\n", delay);
188 schedule_delayed_work(&monc->delayed_work,
189 round_jiffies_relative(delay));
Sage Weilba75bb92009-10-06 11:31:11 -0700190}
191
192/*
193 * Send subscribe request for mdsmap and/or osdmap.
194 */
195static void __send_subscribe(struct ceph_mon_client *monc)
196{
197 dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n",
Eric Dumazet95c96172012-04-15 05:58:06 +0000198 (unsigned int)monc->sub_sent, __sub_expired(monc),
Sage Weilba75bb92009-10-06 11:31:11 -0700199 monc->want_next_osdmap);
200 if ((__sub_expired(monc) && !monc->sub_sent) ||
201 monc->want_next_osdmap == 1) {
Sage Weil240ed682010-05-21 14:57:25 -0700202 struct ceph_msg *msg = monc->m_subscribe;
Sage Weilba75bb92009-10-06 11:31:11 -0700203 struct ceph_mon_subscribe_item *i;
204 void *p, *end;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700205 int num;
Sage Weilba75bb92009-10-06 11:31:11 -0700206
Sage Weilba75bb92009-10-06 11:31:11 -0700207 p = msg->front.iov_base;
Ilya Dryomov3cea4c32014-01-09 20:08:21 +0200208 end = p + msg->front_alloc_len;
Sage Weilba75bb92009-10-06 11:31:11 -0700209
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700210 num = 1 + !!monc->want_next_osdmap + !!monc->want_mdsmap;
211 ceph_encode_32(&p, num);
212
Sage Weilba75bb92009-10-06 11:31:11 -0700213 if (monc->want_next_osdmap) {
214 dout("__send_subscribe to 'osdmap' %u\n",
Eric Dumazet95c96172012-04-15 05:58:06 +0000215 (unsigned int)monc->have_osdmap);
Sage Weilba75bb92009-10-06 11:31:11 -0700216 ceph_encode_string(&p, end, "osdmap", 6);
217 i = p;
218 i->have = cpu_to_le64(monc->have_osdmap);
219 i->onetime = 1;
220 p += sizeof(*i);
221 monc->want_next_osdmap = 2; /* requested */
Sage Weilba75bb92009-10-06 11:31:11 -0700222 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700223 if (monc->want_mdsmap) {
224 dout("__send_subscribe to 'mdsmap' %u+\n",
Eric Dumazet95c96172012-04-15 05:58:06 +0000225 (unsigned int)monc->have_mdsmap);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700226 ceph_encode_string(&p, end, "mdsmap", 6);
227 i = p;
228 i->have = cpu_to_le64(monc->have_mdsmap);
229 i->onetime = 0;
230 p += sizeof(*i);
231 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800232 ceph_encode_string(&p, end, "monmap", 6);
233 i = p;
234 i->have = 0;
235 i->onetime = 0;
236 p += sizeof(*i);
Sage Weilba75bb92009-10-06 11:31:11 -0700237
238 msg->front.iov_len = p - msg->front.iov_base;
239 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
Alex Elder6740a842012-06-01 14:56:43 -0500240 ceph_msg_revoke(msg);
Alex Elder67130932012-05-26 23:26:43 -0500241 ceph_con_send(&monc->con, ceph_msg_get(msg));
Sage Weilba75bb92009-10-06 11:31:11 -0700242
243 monc->sub_sent = jiffies | 1; /* never 0 */
244 }
245}
246
247static void handle_subscribe_ack(struct ceph_mon_client *monc,
248 struct ceph_msg *msg)
249{
Eric Dumazet95c96172012-04-15 05:58:06 +0000250 unsigned int seconds;
Sage Weil07bd10f2009-10-14 17:26:40 -0700251 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
Sage Weilba75bb92009-10-06 11:31:11 -0700252
Sage Weil07bd10f2009-10-14 17:26:40 -0700253 if (msg->front.iov_len < sizeof(*h))
254 goto bad;
255 seconds = le32_to_cpu(h->duration);
256
Sage Weilba75bb92009-10-06 11:31:11 -0700257 mutex_lock(&monc->mutex);
258 if (monc->hunting) {
259 pr_info("mon%d %s session established\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700260 monc->cur_mon,
Alex Elder67130932012-05-26 23:26:43 -0500261 ceph_pr_addr(&monc->con.peer_addr.in_addr));
Sage Weilba75bb92009-10-06 11:31:11 -0700262 monc->hunting = false;
263 }
264 dout("handle_subscribe_ack after %d seconds\n", seconds);
Sage Weil0656d112009-10-08 10:25:46 -0700265 monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1;
Sage Weilba75bb92009-10-06 11:31:11 -0700266 monc->sub_sent = 0;
267 mutex_unlock(&monc->mutex);
268 return;
269bad:
270 pr_err("got corrupt subscribe-ack msg\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -0800271 ceph_msg_dump(msg);
Sage Weilba75bb92009-10-06 11:31:11 -0700272}
273
274/*
275 * Keep track of which maps we have
276 */
277int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got)
278{
279 mutex_lock(&monc->mutex);
280 monc->have_mdsmap = got;
281 mutex_unlock(&monc->mutex);
282 return 0;
283}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700284EXPORT_SYMBOL(ceph_monc_got_mdsmap);
Sage Weilba75bb92009-10-06 11:31:11 -0700285
286int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got)
287{
288 mutex_lock(&monc->mutex);
289 monc->have_osdmap = got;
290 monc->want_next_osdmap = 0;
291 mutex_unlock(&monc->mutex);
292 return 0;
293}
294
295/*
296 * Register interest in the next osdmap
297 */
298void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
299{
300 dout("request_next_osdmap have %u\n", monc->have_osdmap);
301 mutex_lock(&monc->mutex);
302 if (!monc->want_next_osdmap)
303 monc->want_next_osdmap = 1;
304 if (monc->want_next_osdmap < 2)
305 __send_subscribe(monc);
306 mutex_unlock(&monc->mutex);
307}
Ilya Dryomov6044cde2014-05-13 11:19:27 +0400308EXPORT_SYMBOL(ceph_monc_request_next_osdmap);
309
Ilya Dryomova319bf52015-05-15 12:02:17 +0300310/*
311 * Wait for an osdmap with a given epoch.
312 *
313 * @epoch: epoch to wait for
314 * @timeout: in jiffies, 0 means "wait forever"
315 */
Ilya Dryomov6044cde2014-05-13 11:19:27 +0400316int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
317 unsigned long timeout)
318{
319 unsigned long started = jiffies;
Ilya Dryomov216639d2015-05-19 12:03:33 +0300320 long ret;
Ilya Dryomov6044cde2014-05-13 11:19:27 +0400321
322 mutex_lock(&monc->mutex);
323 while (monc->have_osdmap < epoch) {
324 mutex_unlock(&monc->mutex);
325
Ilya Dryomova319bf52015-05-15 12:02:17 +0300326 if (timeout && time_after_eq(jiffies, started + timeout))
Ilya Dryomov6044cde2014-05-13 11:19:27 +0400327 return -ETIMEDOUT;
328
329 ret = wait_event_interruptible_timeout(monc->client->auth_wq,
Ilya Dryomova319bf52015-05-15 12:02:17 +0300330 monc->have_osdmap >= epoch,
331 ceph_timeout_jiffies(timeout));
Ilya Dryomov6044cde2014-05-13 11:19:27 +0400332 if (ret < 0)
333 return ret;
334
335 mutex_lock(&monc->mutex);
336 }
337
338 mutex_unlock(&monc->mutex);
339 return 0;
340}
341EXPORT_SYMBOL(ceph_monc_wait_osdmap);
Sage Weilba75bb92009-10-06 11:31:11 -0700342
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800343/*
Sage Weil50b885b2009-12-01 14:12:07 -0800344 *
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800345 */
346int ceph_monc_open_session(struct ceph_mon_client *monc)
Sage Weilba75bb92009-10-06 11:31:11 -0700347{
Sage Weilba75bb92009-10-06 11:31:11 -0700348 mutex_lock(&monc->mutex);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800349 __open_session(monc);
Sage Weilba75bb92009-10-06 11:31:11 -0700350 __schedule_delayed(monc);
351 mutex_unlock(&monc->mutex);
352 return 0;
353}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700354EXPORT_SYMBOL(ceph_monc_open_session);
Sage Weilba75bb92009-10-06 11:31:11 -0700355
Sage Weil07433042009-11-18 16:50:41 -0800356static void ceph_monc_handle_map(struct ceph_mon_client *monc,
357 struct ceph_msg *msg)
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800358{
359 struct ceph_client *client = monc->client;
360 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
361 void *p, *end;
362
363 mutex_lock(&monc->mutex);
364
365 dout("handle_monmap\n");
366 p = msg->front.iov_base;
367 end = p + msg->front.iov_len;
368
369 monmap = ceph_monmap_decode(p, end);
370 if (IS_ERR(monmap)) {
371 pr_err("problem decoding monmap, %d\n",
372 (int)PTR_ERR(monmap));
Sage Weild4a780c2009-12-11 08:55:23 -0800373 goto out;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800374 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800375
Sage Weil07433042009-11-18 16:50:41 -0800376 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800377 kfree(monmap);
Sage Weild4a780c2009-12-11 08:55:23 -0800378 goto out;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800379 }
380
381 client->monc.monmap = monmap;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800382 kfree(old);
383
Ilya Dryomov02ac9562016-01-06 12:56:21 +0300384 client->have_fsid = true;
Sage Weild1c338a2012-08-19 12:29:16 -0700385
Sage Weild4a780c2009-12-11 08:55:23 -0800386out:
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800387 mutex_unlock(&monc->mutex);
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700388 wake_up_all(&client->auth_wq);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800389}
Sage Weilba75bb92009-10-06 11:31:11 -0700390
Sage Weilba75bb92009-10-06 11:31:11 -0700391/*
Ilya Dryomov7a6fdeb2014-12-22 19:14:26 +0300392 * generic requests (currently statfs, mon_get_version)
Sage Weilba75bb92009-10-06 11:31:11 -0700393 */
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700394static struct ceph_mon_generic_request *__lookup_generic_req(
Sage Weil85ff03f2010-02-15 14:47:28 -0800395 struct ceph_mon_client *monc, u64 tid)
396{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700397 struct ceph_mon_generic_request *req;
398 struct rb_node *n = monc->generic_request_tree.rb_node;
Sage Weil85ff03f2010-02-15 14:47:28 -0800399
400 while (n) {
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700401 req = rb_entry(n, struct ceph_mon_generic_request, node);
Sage Weil85ff03f2010-02-15 14:47:28 -0800402 if (tid < req->tid)
403 n = n->rb_left;
404 else if (tid > req->tid)
405 n = n->rb_right;
406 else
407 return req;
408 }
409 return NULL;
410}
411
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700412static void __insert_generic_request(struct ceph_mon_client *monc,
413 struct ceph_mon_generic_request *new)
Sage Weil85ff03f2010-02-15 14:47:28 -0800414{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700415 struct rb_node **p = &monc->generic_request_tree.rb_node;
Sage Weil85ff03f2010-02-15 14:47:28 -0800416 struct rb_node *parent = NULL;
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700417 struct ceph_mon_generic_request *req = NULL;
Sage Weil85ff03f2010-02-15 14:47:28 -0800418
419 while (*p) {
420 parent = *p;
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700421 req = rb_entry(parent, struct ceph_mon_generic_request, node);
Sage Weil85ff03f2010-02-15 14:47:28 -0800422 if (new->tid < req->tid)
423 p = &(*p)->rb_left;
424 else if (new->tid > req->tid)
425 p = &(*p)->rb_right;
426 else
427 BUG();
428 }
429
430 rb_link_node(&new->node, parent, p);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700431 rb_insert_color(&new->node, &monc->generic_request_tree);
Sage Weil85ff03f2010-02-15 14:47:28 -0800432}
433
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700434static void release_generic_request(struct kref *kref)
Sage Weil3143edd2010-03-24 21:43:33 -0700435{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700436 struct ceph_mon_generic_request *req =
437 container_of(kref, struct ceph_mon_generic_request, kref);
Sage Weil3143edd2010-03-24 21:43:33 -0700438
439 if (req->reply)
440 ceph_msg_put(req->reply);
441 if (req->request)
442 ceph_msg_put(req->request);
Yehuda Sadeh20547562010-06-01 10:37:40 -0700443
444 kfree(req);
Sage Weil3143edd2010-03-24 21:43:33 -0700445}
446
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700447static void put_generic_request(struct ceph_mon_generic_request *req)
Sage Weil3143edd2010-03-24 21:43:33 -0700448{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700449 kref_put(&req->kref, release_generic_request);
Sage Weil3143edd2010-03-24 21:43:33 -0700450}
451
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700452static void get_generic_request(struct ceph_mon_generic_request *req)
Sage Weil3143edd2010-03-24 21:43:33 -0700453{
454 kref_get(&req->kref);
455}
456
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700457static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
Sage Weil3143edd2010-03-24 21:43:33 -0700458 struct ceph_msg_header *hdr,
459 int *skip)
460{
461 struct ceph_mon_client *monc = con->private;
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700462 struct ceph_mon_generic_request *req;
Sage Weil3143edd2010-03-24 21:43:33 -0700463 u64 tid = le64_to_cpu(hdr->tid);
464 struct ceph_msg *m;
465
466 mutex_lock(&monc->mutex);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700467 req = __lookup_generic_req(monc, tid);
Sage Weil3143edd2010-03-24 21:43:33 -0700468 if (!req) {
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700469 dout("get_generic_reply %lld dne\n", tid);
Sage Weil3143edd2010-03-24 21:43:33 -0700470 *skip = 1;
471 m = NULL;
472 } else {
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700473 dout("get_generic_reply %lld got %p\n", tid, req->reply);
Alex Elder1c20f2d2012-06-04 14:43:32 -0500474 *skip = 0;
Sage Weil3143edd2010-03-24 21:43:33 -0700475 m = ceph_msg_get(req->reply);
476 /*
477 * we don't need to track the connection reading into
478 * this reply because we only have one open connection
479 * at a time, ever.
480 */
481 }
482 mutex_unlock(&monc->mutex);
483 return m;
484}
485
Ilya Dryomov513a8242014-05-13 11:19:26 +0400486static int __do_generic_request(struct ceph_mon_client *monc, u64 tid,
487 struct ceph_mon_generic_request *req)
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700488{
489 int err;
490
491 /* register request */
Ilya Dryomov513a8242014-05-13 11:19:26 +0400492 req->tid = tid != 0 ? tid : ++monc->last_tid;
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700493 req->request->hdr.tid = cpu_to_le64(req->tid);
494 __insert_generic_request(monc, req);
495 monc->num_generic_requests++;
Alex Elder67130932012-05-26 23:26:43 -0500496 ceph_con_send(&monc->con, ceph_msg_get(req->request));
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700497 mutex_unlock(&monc->mutex);
498
499 err = wait_for_completion_interruptible(&req->completion);
500
501 mutex_lock(&monc->mutex);
502 rb_erase(&req->node, &monc->generic_request_tree);
503 monc->num_generic_requests--;
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700504
505 if (!err)
506 err = req->result;
507 return err;
508}
509
Ilya Dryomov513a8242014-05-13 11:19:26 +0400510static int do_generic_request(struct ceph_mon_client *monc,
511 struct ceph_mon_generic_request *req)
512{
513 int err;
514
515 mutex_lock(&monc->mutex);
516 err = __do_generic_request(monc, 0, req);
517 mutex_unlock(&monc->mutex);
518
519 return err;
520}
521
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700522/*
523 * statfs
524 */
Sage Weilba75bb92009-10-06 11:31:11 -0700525static void handle_statfs_reply(struct ceph_mon_client *monc,
526 struct ceph_msg *msg)
527{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700528 struct ceph_mon_generic_request *req;
Sage Weilba75bb92009-10-06 11:31:11 -0700529 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
Sage Weil3143edd2010-03-24 21:43:33 -0700530 u64 tid = le64_to_cpu(msg->hdr.tid);
Sage Weilba75bb92009-10-06 11:31:11 -0700531
532 if (msg->front.iov_len != sizeof(*reply))
533 goto bad;
Sage Weilba75bb92009-10-06 11:31:11 -0700534 dout("handle_statfs_reply %p tid %llu\n", msg, tid);
535
536 mutex_lock(&monc->mutex);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700537 req = __lookup_generic_req(monc, tid);
Sage Weilba75bb92009-10-06 11:31:11 -0700538 if (req) {
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700539 *(struct ceph_statfs *)req->buf = reply->st;
Sage Weilba75bb92009-10-06 11:31:11 -0700540 req->result = 0;
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700541 get_generic_request(req);
Sage Weilba75bb92009-10-06 11:31:11 -0700542 }
543 mutex_unlock(&monc->mutex);
Sage Weil3143edd2010-03-24 21:43:33 -0700544 if (req) {
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700545 complete_all(&req->completion);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700546 put_generic_request(req);
Sage Weil3143edd2010-03-24 21:43:33 -0700547 }
Sage Weilba75bb92009-10-06 11:31:11 -0700548 return;
549
550bad:
Ilya Dryomov7a6fdeb2014-12-22 19:14:26 +0300551 pr_err("corrupt statfs reply, tid %llu\n", tid);
Sage Weil9ec7cab2009-12-14 15:13:47 -0800552 ceph_msg_dump(msg);
Sage Weilba75bb92009-10-06 11:31:11 -0700553}
554
555/*
Sage Weilba75bb92009-10-06 11:31:11 -0700556 * Do a synchronous statfs().
557 */
558int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
559{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700560 struct ceph_mon_generic_request *req;
Sage Weil3143edd2010-03-24 21:43:33 -0700561 struct ceph_mon_statfs *h;
Sage Weilba75bb92009-10-06 11:31:11 -0700562 int err;
563
Julia Lawallcffe7b62010-05-13 22:07:29 +0200564 req = kzalloc(sizeof(*req), GFP_NOFS);
Sage Weil3143edd2010-03-24 21:43:33 -0700565 if (!req)
566 return -ENOMEM;
Sage Weilba75bb92009-10-06 11:31:11 -0700567
Sage Weil3143edd2010-03-24 21:43:33 -0700568 kref_init(&req->kref);
569 req->buf = buf;
570 init_completion(&req->completion);
571
Sage Weila79832f2010-04-01 16:06:19 -0700572 err = -ENOMEM;
Sage Weilb61c2762011-08-09 15:03:46 -0700573 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
574 true);
Sage Weila79832f2010-04-01 16:06:19 -0700575 if (!req->request)
Sage Weil3143edd2010-03-24 21:43:33 -0700576 goto out;
Sage Weilb61c2762011-08-09 15:03:46 -0700577 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
578 true);
Sage Weila79832f2010-04-01 16:06:19 -0700579 if (!req->reply)
Sage Weil3143edd2010-03-24 21:43:33 -0700580 goto out;
Sage Weil3143edd2010-03-24 21:43:33 -0700581
582 /* fill out request */
583 h = req->request->front.iov_base;
584 h->monhdr.have_version = 0;
585 h->monhdr.session_mon = cpu_to_le16(-1);
586 h->monhdr.session_mon_tid = 0;
587 h->fsid = monc->monmap->fsid;
Sage Weilba75bb92009-10-06 11:31:11 -0700588
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700589 err = do_generic_request(monc, req);
Sage Weil3143edd2010-03-24 21:43:33 -0700590
591out:
Ilya Dryomovf6469122014-12-22 19:35:17 +0300592 put_generic_request(req);
Sage Weilba75bb92009-10-06 11:31:11 -0700593 return err;
594}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700595EXPORT_SYMBOL(ceph_monc_do_statfs);
Sage Weilba75bb92009-10-06 11:31:11 -0700596
Ilya Dryomov513a8242014-05-13 11:19:26 +0400597static void handle_get_version_reply(struct ceph_mon_client *monc,
598 struct ceph_msg *msg)
599{
600 struct ceph_mon_generic_request *req;
601 u64 tid = le64_to_cpu(msg->hdr.tid);
602 void *p = msg->front.iov_base;
603 void *end = p + msg->front_alloc_len;
604 u64 handle;
605
606 dout("%s %p tid %llu\n", __func__, msg, tid);
607
608 ceph_decode_need(&p, end, 2*sizeof(u64), bad);
609 handle = ceph_decode_64(&p);
610 if (tid != 0 && tid != handle)
611 goto bad;
612
613 mutex_lock(&monc->mutex);
614 req = __lookup_generic_req(monc, handle);
615 if (req) {
616 *(u64 *)req->buf = ceph_decode_64(&p);
617 req->result = 0;
618 get_generic_request(req);
619 }
620 mutex_unlock(&monc->mutex);
621 if (req) {
622 complete_all(&req->completion);
623 put_generic_request(req);
624 }
625
626 return;
627bad:
Ilya Dryomov7a6fdeb2014-12-22 19:14:26 +0300628 pr_err("corrupt mon_get_version reply, tid %llu\n", tid);
Ilya Dryomov513a8242014-05-13 11:19:26 +0400629 ceph_msg_dump(msg);
630}
631
632/*
633 * Send MMonGetVersion and wait for the reply.
634 *
635 * @what: one of "mdsmap", "osdmap" or "monmap"
636 */
637int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what,
638 u64 *newest)
639{
640 struct ceph_mon_generic_request *req;
641 void *p, *end;
642 u64 tid;
643 int err;
644
645 req = kzalloc(sizeof(*req), GFP_NOFS);
646 if (!req)
647 return -ENOMEM;
648
649 kref_init(&req->kref);
650 req->buf = newest;
Ilya Dryomov513a8242014-05-13 11:19:26 +0400651 init_completion(&req->completion);
652
653 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
654 sizeof(u64) + sizeof(u32) + strlen(what),
655 GFP_NOFS, true);
656 if (!req->request) {
657 err = -ENOMEM;
658 goto out;
659 }
660
661 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024,
662 GFP_NOFS, true);
663 if (!req->reply) {
664 err = -ENOMEM;
665 goto out;
666 }
667
668 p = req->request->front.iov_base;
669 end = p + req->request->front_alloc_len;
670
671 /* fill out request */
672 mutex_lock(&monc->mutex);
673 tid = ++monc->last_tid;
674 ceph_encode_64(&p, tid); /* handle */
675 ceph_encode_string(&p, end, what, strlen(what));
676
677 err = __do_generic_request(monc, tid, req);
678
679 mutex_unlock(&monc->mutex);
680out:
Ilya Dryomovf6469122014-12-22 19:35:17 +0300681 put_generic_request(req);
Ilya Dryomov513a8242014-05-13 11:19:26 +0400682 return err;
683}
684EXPORT_SYMBOL(ceph_monc_do_get_version);
685
Sage Weilba75bb92009-10-06 11:31:11 -0700686/*
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700687 * Resend pending generic requests.
Sage Weilba75bb92009-10-06 11:31:11 -0700688 */
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700689static void __resend_generic_request(struct ceph_mon_client *monc)
Sage Weilba75bb92009-10-06 11:31:11 -0700690{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700691 struct ceph_mon_generic_request *req;
Sage Weil85ff03f2010-02-15 14:47:28 -0800692 struct rb_node *p;
Sage Weilba75bb92009-10-06 11:31:11 -0700693
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700694 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
695 req = rb_entry(p, struct ceph_mon_generic_request, node);
Alex Elder6740a842012-06-01 14:56:43 -0500696 ceph_msg_revoke(req->request);
Sage Weil4f471e42012-07-30 18:16:40 -0700697 ceph_msg_revoke_incoming(req->reply);
Alex Elder67130932012-05-26 23:26:43 -0500698 ceph_con_send(&monc->con, ceph_msg_get(req->request));
Sage Weilba75bb92009-10-06 11:31:11 -0700699 }
700}
701
702/*
703 * Delayed work. If we haven't mounted yet, retry. Otherwise,
704 * renew/retry subscription as needed (in case it is timing out, or we
705 * got an ENOMEM). And keep the monitor connection alive.
706 */
707static void delayed_work(struct work_struct *work)
708{
709 struct ceph_mon_client *monc =
710 container_of(work, struct ceph_mon_client, delayed_work.work);
711
712 dout("monc delayed_work\n");
713 mutex_lock(&monc->mutex);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800714 if (monc->hunting) {
715 __close_session(monc);
716 __open_session(monc); /* continue hunting */
Sage Weilba75bb92009-10-06 11:31:11 -0700717 } else {
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800718 struct ceph_options *opt = monc->client->options;
719 int is_auth = ceph_auth_is_authenticated(monc->auth);
720 if (ceph_con_keepalive_expired(&monc->con,
721 opt->monc_ping_timeout)) {
722 dout("monc keepalive timeout\n");
723 is_auth = 0;
724 __close_session(monc);
725 monc->hunting = true;
726 __open_session(monc);
727 }
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800728
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800729 if (!monc->hunting) {
730 ceph_con_keepalive(&monc->con);
731 __validate_auth(monc);
732 }
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800733
Yan, Zheng8b9558a2015-09-01 17:19:38 +0800734 if (is_auth)
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800735 __send_subscribe(monc);
Sage Weilba75bb92009-10-06 11:31:11 -0700736 }
Sage Weilba75bb92009-10-06 11:31:11 -0700737 __schedule_delayed(monc);
738 mutex_unlock(&monc->mutex);
739}
740
Sage Weil6b805182009-10-27 11:50:50 -0700741/*
742 * On startup, we build a temporary monmap populated with the IPs
743 * provided by mount(2).
744 */
745static int build_initial_monmap(struct ceph_mon_client *monc)
746{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700747 struct ceph_options *opt = monc->client->options;
748 struct ceph_entity_addr *mon_addr = opt->mon_addr;
749 int num_mon = opt->num_mon;
Sage Weil6b805182009-10-27 11:50:50 -0700750 int i;
751
752 /* build initial monmap */
753 monc->monmap = kzalloc(sizeof(*monc->monmap) +
754 num_mon*sizeof(monc->monmap->mon_inst[0]),
755 GFP_KERNEL);
756 if (!monc->monmap)
757 return -ENOMEM;
758 for (i = 0; i < num_mon; i++) {
759 monc->monmap->mon_inst[i].addr = mon_addr[i];
Sage Weil6b805182009-10-27 11:50:50 -0700760 monc->monmap->mon_inst[i].addr.nonce = 0;
761 monc->monmap->mon_inst[i].name.type =
762 CEPH_ENTITY_TYPE_MON;
763 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
764 }
765 monc->monmap->num_mon = num_mon;
Sage Weil6b805182009-10-27 11:50:50 -0700766 return 0;
767}
768
Sage Weilba75bb92009-10-06 11:31:11 -0700769int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
770{
771 int err = 0;
772
773 dout("init\n");
774 memset(monc, 0, sizeof(*monc));
775 monc->client = cl;
776 monc->monmap = NULL;
777 mutex_init(&monc->mutex);
778
Sage Weil6b805182009-10-27 11:50:50 -0700779 err = build_initial_monmap(monc);
780 if (err)
781 goto out;
782
Sage Weilf6a2f5b2011-08-09 09:27:44 -0700783 /* connection */
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800784 /* authentication */
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700785 monc->auth = ceph_auth_init(cl->options->name,
Tommi Virtanen8323c3a2011-03-25 16:32:57 -0700786 cl->options->key);
Noah Watkins49d92242011-09-12 14:51:58 -0700787 if (IS_ERR(monc->auth)) {
788 err = PTR_ERR(monc->auth);
Alex Elder67130932012-05-26 23:26:43 -0500789 goto out_monmap;
Noah Watkins49d92242011-09-12 14:51:58 -0700790 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800791 monc->auth->want_keys =
792 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
793 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
794
Sage Weil240ed682010-05-21 14:57:25 -0700795 /* msgs */
Sage Weila79832f2010-04-01 16:06:19 -0700796 err = -ENOMEM;
Sage Weil7c315c52010-03-24 21:52:30 -0700797 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700798 sizeof(struct ceph_mon_subscribe_ack),
Sage Weilb61c2762011-08-09 15:03:46 -0700799 GFP_NOFS, true);
Sage Weila79832f2010-04-01 16:06:19 -0700800 if (!monc->m_subscribe_ack)
Noah Watkins49d92242011-09-12 14:51:58 -0700801 goto out_auth;
Sage Weil6694d6b2010-03-24 21:48:05 -0700802
Sage Weilb61c2762011-08-09 15:03:46 -0700803 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
804 true);
Sage Weil240ed682010-05-21 14:57:25 -0700805 if (!monc->m_subscribe)
806 goto out_subscribe_ack;
807
Sage Weilb61c2762011-08-09 15:03:46 -0700808 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
809 true);
Sage Weila79832f2010-04-01 16:06:19 -0700810 if (!monc->m_auth_reply)
Sage Weil240ed682010-05-21 14:57:25 -0700811 goto out_subscribe;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800812
Sage Weilb61c2762011-08-09 15:03:46 -0700813 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800814 monc->pending_auth = 0;
Sage Weila79832f2010-04-01 16:06:19 -0700815 if (!monc->m_auth)
Sage Weil6694d6b2010-03-24 21:48:05 -0700816 goto out_auth_reply;
Sage Weilba75bb92009-10-06 11:31:11 -0700817
Sage Weil735a72e2012-06-27 12:24:34 -0700818 ceph_con_init(&monc->con, monc, &mon_con_ops,
819 &monc->client->msgr);
820
Sage Weilba75bb92009-10-06 11:31:11 -0700821 monc->cur_mon = -1;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800822 monc->hunting = true;
Sage Weilba75bb92009-10-06 11:31:11 -0700823 monc->sub_renew_after = jiffies;
824 monc->sub_sent = 0;
825
826 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700827 monc->generic_request_tree = RB_ROOT;
828 monc->num_generic_requests = 0;
Sage Weilba75bb92009-10-06 11:31:11 -0700829 monc->last_tid = 0;
830
831 monc->have_mdsmap = 0;
832 monc->have_osdmap = 0;
833 monc->want_next_osdmap = 1;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800834 return 0;
835
Sage Weil6694d6b2010-03-24 21:48:05 -0700836out_auth_reply:
837 ceph_msg_put(monc->m_auth_reply);
Sage Weil240ed682010-05-21 14:57:25 -0700838out_subscribe:
839 ceph_msg_put(monc->m_subscribe);
Sage Weil7c315c52010-03-24 21:52:30 -0700840out_subscribe_ack:
841 ceph_msg_put(monc->m_subscribe_ack);
Noah Watkins49d92242011-09-12 14:51:58 -0700842out_auth:
843 ceph_auth_destroy(monc->auth);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800844out_monmap:
845 kfree(monc->monmap);
Sage Weilba75bb92009-10-06 11:31:11 -0700846out:
847 return err;
848}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700849EXPORT_SYMBOL(ceph_monc_init);
Sage Weilba75bb92009-10-06 11:31:11 -0700850
851void ceph_monc_stop(struct ceph_mon_client *monc)
852{
853 dout("stop\n");
854 cancel_delayed_work_sync(&monc->delayed_work);
855
856 mutex_lock(&monc->mutex);
857 __close_session(monc);
Sage Weilf6a2f5b2011-08-09 09:27:44 -0700858
Sage Weilba75bb92009-10-06 11:31:11 -0700859 mutex_unlock(&monc->mutex);
860
Sage Weilf3dea7e2012-06-10 20:43:56 -0700861 /*
862 * flush msgr queue before we destroy ourselves to ensure that:
863 * - any work that references our embedded con is finished.
864 * - any osd_client or other work that may reference an authorizer
865 * finishes before we shut down the auth subsystem.
866 */
867 ceph_msgr_flush();
868
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800869 ceph_auth_destroy(monc->auth);
870
871 ceph_msg_put(monc->m_auth);
Sage Weil6694d6b2010-03-24 21:48:05 -0700872 ceph_msg_put(monc->m_auth_reply);
Sage Weil240ed682010-05-21 14:57:25 -0700873 ceph_msg_put(monc->m_subscribe);
Sage Weil7c315c52010-03-24 21:52:30 -0700874 ceph_msg_put(monc->m_subscribe_ack);
Sage Weilba75bb92009-10-06 11:31:11 -0700875
876 kfree(monc->monmap);
877}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700878EXPORT_SYMBOL(ceph_monc_stop);
Sage Weilba75bb92009-10-06 11:31:11 -0700879
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800880static void handle_auth_reply(struct ceph_mon_client *monc,
881 struct ceph_msg *msg)
882{
883 int ret;
Sage Weil09c4d6a2010-05-25 15:38:06 -0700884 int was_auth = 0;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800885
886 mutex_lock(&monc->mutex);
Sage Weil27859f92013-03-25 10:26:14 -0700887 was_auth = ceph_auth_is_authenticated(monc->auth);
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800888 monc->pending_auth = 0;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800889 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
890 msg->front.iov_len,
891 monc->m_auth->front.iov_base,
Ilya Dryomov3cea4c32014-01-09 20:08:21 +0200892 monc->m_auth->front_alloc_len);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800893 if (ret < 0) {
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800894 monc->client->auth_err = ret;
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700895 wake_up_all(&monc->client->auth_wq);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800896 } else if (ret > 0) {
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800897 __send_prepared_auth_request(monc, ret);
Sage Weil27859f92013-03-25 10:26:14 -0700898 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800899 dout("authenticated, starting session\n");
Sage Weil07433042009-11-18 16:50:41 -0800900
Alex Elder15d98822012-05-26 23:26:43 -0500901 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
902 monc->client->msgr.inst.name.num =
Yehuda Sadeh0cf55372010-06-11 15:57:06 -0700903 cpu_to_le64(monc->auth->global_id);
Sage Weil07433042009-11-18 16:50:41 -0800904
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800905 __send_subscribe(monc);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700906 __resend_generic_request(monc);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800907 }
908 mutex_unlock(&monc->mutex);
909}
910
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800911static int __validate_auth(struct ceph_mon_client *monc)
912{
913 int ret;
914
915 if (monc->pending_auth)
916 return 0;
917
918 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
Ilya Dryomov3cea4c32014-01-09 20:08:21 +0200919 monc->m_auth->front_alloc_len);
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800920 if (ret <= 0)
921 return ret; /* either an error, or no need to authenticate */
922 __send_prepared_auth_request(monc, ret);
923 return 0;
924}
925
926int ceph_monc_validate_auth(struct ceph_mon_client *monc)
927{
928 int ret;
929
930 mutex_lock(&monc->mutex);
931 ret = __validate_auth(monc);
932 mutex_unlock(&monc->mutex);
933 return ret;
934}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700935EXPORT_SYMBOL(ceph_monc_validate_auth);
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800936
Sage Weilba75bb92009-10-06 11:31:11 -0700937/*
938 * handle incoming message
939 */
940static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
941{
942 struct ceph_mon_client *monc = con->private;
943 int type = le16_to_cpu(msg->hdr.type);
944
945 if (!monc)
946 return;
947
948 switch (type) {
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800949 case CEPH_MSG_AUTH_REPLY:
950 handle_auth_reply(monc, msg);
Sage Weilba75bb92009-10-06 11:31:11 -0700951 break;
952
953 case CEPH_MSG_MON_SUBSCRIBE_ACK:
954 handle_subscribe_ack(monc, msg);
955 break;
956
957 case CEPH_MSG_STATFS_REPLY:
958 handle_statfs_reply(monc, msg);
959 break;
960
Ilya Dryomov513a8242014-05-13 11:19:26 +0400961 case CEPH_MSG_MON_GET_VERSION_REPLY:
962 handle_get_version_reply(monc, msg);
963 break;
964
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800965 case CEPH_MSG_MON_MAP:
966 ceph_monc_handle_map(monc, msg);
967 break;
968
Sage Weilba75bb92009-10-06 11:31:11 -0700969 case CEPH_MSG_OSD_MAP:
970 ceph_osdc_handle_map(&monc->client->osdc, msg);
971 break;
972
973 default:
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700974 /* can the chained handler handle it? */
975 if (monc->client->extra_mon_dispatch &&
976 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
977 break;
978
Sage Weilba75bb92009-10-06 11:31:11 -0700979 pr_err("received unknown message type %d %s\n", type,
980 ceph_msg_type_name(type));
981 }
982 ceph_msg_put(msg);
983}
984
985/*
986 * Allocate memory for incoming message
987 */
988static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
Yehuda Sadeh24504182010-01-08 13:58:34 -0800989 struct ceph_msg_header *hdr,
990 int *skip)
Sage Weilba75bb92009-10-06 11:31:11 -0700991{
992 struct ceph_mon_client *monc = con->private;
993 int type = le16_to_cpu(hdr->type);
Yehuda Sadeh24504182010-01-08 13:58:34 -0800994 int front_len = le32_to_cpu(hdr->front_len);
Sage Weil5b3a4db2010-02-19 21:43:23 -0800995 struct ceph_msg *m = NULL;
Sage Weilba75bb92009-10-06 11:31:11 -0700996
Yehuda Sadeh24504182010-01-08 13:58:34 -0800997 *skip = 0;
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -0800998
Sage Weilba75bb92009-10-06 11:31:11 -0700999 switch (type) {
Sage Weilba75bb92009-10-06 11:31:11 -07001000 case CEPH_MSG_MON_SUBSCRIBE_ACK:
Sage Weil7c315c52010-03-24 21:52:30 -07001001 m = ceph_msg_get(monc->m_subscribe_ack);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001002 break;
Sage Weilba75bb92009-10-06 11:31:11 -07001003 case CEPH_MSG_STATFS_REPLY:
Yehuda Sadehf8c76f62010-04-22 15:40:37 -07001004 return get_generic_reply(con, hdr, skip);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001005 case CEPH_MSG_AUTH_REPLY:
Sage Weil6694d6b2010-03-24 21:48:05 -07001006 m = ceph_msg_get(monc->m_auth_reply);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001007 break;
Ilya Dryomov513a8242014-05-13 11:19:26 +04001008 case CEPH_MSG_MON_GET_VERSION_REPLY:
1009 if (le64_to_cpu(hdr->tid) != 0)
1010 return get_generic_reply(con, hdr, skip);
1011
1012 /*
1013 * Older OSDs don't set reply tid even if the orignal
1014 * request had a non-zero tid. Workaround this weirdness
1015 * by falling through to the allocate case.
1016 */
Sage Weil5b3a4db2010-02-19 21:43:23 -08001017 case CEPH_MSG_MON_MAP:
1018 case CEPH_MSG_MDS_MAP:
1019 case CEPH_MSG_OSD_MAP:
Sage Weilb61c2762011-08-09 15:03:46 -07001020 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
Alex Elder1c20f2d2012-06-04 14:43:32 -05001021 if (!m)
1022 return NULL; /* ENOMEM--return skip == 0 */
Sage Weil5b3a4db2010-02-19 21:43:23 -08001023 break;
Sage Weilba75bb92009-10-06 11:31:11 -07001024 }
Yehuda Sadeh24504182010-01-08 13:58:34 -08001025
Sage Weil5b3a4db2010-02-19 21:43:23 -08001026 if (!m) {
1027 pr_info("alloc_msg unknown type %d\n", type);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001028 *skip = 1;
Sage Weil73c3d482014-08-04 07:01:54 -07001029 } else if (front_len > m->front_alloc_len) {
Joe Perchesb9a67892014-09-09 21:17:29 -07001030 pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n",
1031 front_len, m->front_alloc_len,
1032 (unsigned int)con->peer_name.type,
1033 le64_to_cpu(con->peer_name.num));
Sage Weil73c3d482014-08-04 07:01:54 -07001034 ceph_msg_put(m);
1035 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
Sage Weil5b3a4db2010-02-19 21:43:23 -08001036 }
Sage Weil73c3d482014-08-04 07:01:54 -07001037
Yehuda Sadeh24504182010-01-08 13:58:34 -08001038 return m;
Sage Weilba75bb92009-10-06 11:31:11 -07001039}
1040
1041/*
1042 * If the monitor connection resets, pick a new monitor and resubmit
1043 * any pending requests.
1044 */
1045static void mon_fault(struct ceph_connection *con)
1046{
1047 struct ceph_mon_client *monc = con->private;
1048
1049 if (!monc)
1050 return;
1051
1052 dout("mon_fault\n");
1053 mutex_lock(&monc->mutex);
1054 if (!con->private)
1055 goto out;
1056
Sage Weilf6a2f5b2011-08-09 09:27:44 -07001057 if (!monc->hunting)
Sage Weilba75bb92009-10-06 11:31:11 -07001058 pr_info("mon%d %s session lost, "
1059 "hunting for new mon\n", monc->cur_mon,
Alex Elder67130932012-05-26 23:26:43 -05001060 ceph_pr_addr(&monc->con.peer_addr.in_addr));
Sage Weilba75bb92009-10-06 11:31:11 -07001061
1062 __close_session(monc);
1063 if (!monc->hunting) {
1064 /* start hunting */
1065 monc->hunting = true;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001066 __open_session(monc);
Sage Weilba75bb92009-10-06 11:31:11 -07001067 } else {
1068 /* already hunting, let's wait a bit */
1069 __schedule_delayed(monc);
1070 }
1071out:
1072 mutex_unlock(&monc->mutex);
1073}
1074
Sage Weilec87ef42012-05-31 20:27:50 -07001075/*
1076 * We can ignore refcounting on the connection struct, as all references
1077 * will come from the messenger workqueue, which is drained prior to
1078 * mon_client destruction.
1079 */
1080static struct ceph_connection *con_get(struct ceph_connection *con)
1081{
1082 return con;
1083}
1084
1085static void con_put(struct ceph_connection *con)
1086{
1087}
1088
Tobias Klauser9e327892010-05-20 10:40:19 +02001089static const struct ceph_connection_operations mon_con_ops = {
Sage Weilec87ef42012-05-31 20:27:50 -07001090 .get = con_get,
1091 .put = con_put,
Sage Weilba75bb92009-10-06 11:31:11 -07001092 .dispatch = dispatch,
1093 .fault = mon_fault,
1094 .alloc_msg = mon_alloc_msg,
Sage Weilba75bb92009-10-06 11:31:11 -07001095};