blob: 812eb3b46c1f94d188079274457108cad4c5b54f [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
152 /* initiatiate authentication handshake */
153 ret = ceph_auth_build_hello(monc->auth,
154 monc->m_auth->front.iov_base,
155 monc->m_auth->front_max);
Sage Weil5ce6e9d2010-02-15 16:22:28 -0800156 __send_prepared_auth_request(monc, ret);
Sage Weilba75bb92009-10-06 11:31:11 -0700157 } else {
158 dout("open_session mon%d already open\n", monc->cur_mon);
159 }
160 return 0;
161}
162
163static bool __sub_expired(struct ceph_mon_client *monc)
164{
165 return time_after_eq(jiffies, monc->sub_renew_after);
166}
167
168/*
169 * Reschedule delayed work timer.
170 */
171static void __schedule_delayed(struct ceph_mon_client *monc)
172{
Eric Dumazet95c96172012-04-15 05:58:06 +0000173 unsigned int delay;
Sage Weilba75bb92009-10-06 11:31:11 -0700174
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800175 if (monc->cur_mon < 0 || __sub_expired(monc))
Sage Weilba75bb92009-10-06 11:31:11 -0700176 delay = 10 * HZ;
177 else
178 delay = 20 * HZ;
179 dout("__schedule_delayed after %u\n", delay);
180 schedule_delayed_work(&monc->delayed_work, delay);
181}
182
183/*
184 * Send subscribe request for mdsmap and/or osdmap.
185 */
186static void __send_subscribe(struct ceph_mon_client *monc)
187{
188 dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n",
Eric Dumazet95c96172012-04-15 05:58:06 +0000189 (unsigned int)monc->sub_sent, __sub_expired(monc),
Sage Weilba75bb92009-10-06 11:31:11 -0700190 monc->want_next_osdmap);
191 if ((__sub_expired(monc) && !monc->sub_sent) ||
192 monc->want_next_osdmap == 1) {
Sage Weil240ed682010-05-21 14:57:25 -0700193 struct ceph_msg *msg = monc->m_subscribe;
Sage Weilba75bb92009-10-06 11:31:11 -0700194 struct ceph_mon_subscribe_item *i;
195 void *p, *end;
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700196 int num;
Sage Weilba75bb92009-10-06 11:31:11 -0700197
Sage Weilba75bb92009-10-06 11:31:11 -0700198 p = msg->front.iov_base;
Sage Weil240ed682010-05-21 14:57:25 -0700199 end = p + msg->front_max;
Sage Weilba75bb92009-10-06 11:31:11 -0700200
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700201 num = 1 + !!monc->want_next_osdmap + !!monc->want_mdsmap;
202 ceph_encode_32(&p, num);
203
Sage Weilba75bb92009-10-06 11:31:11 -0700204 if (monc->want_next_osdmap) {
205 dout("__send_subscribe to 'osdmap' %u\n",
Eric Dumazet95c96172012-04-15 05:58:06 +0000206 (unsigned int)monc->have_osdmap);
Sage Weilba75bb92009-10-06 11:31:11 -0700207 ceph_encode_string(&p, end, "osdmap", 6);
208 i = p;
209 i->have = cpu_to_le64(monc->have_osdmap);
210 i->onetime = 1;
211 p += sizeof(*i);
212 monc->want_next_osdmap = 2; /* requested */
Sage Weilba75bb92009-10-06 11:31:11 -0700213 }
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700214 if (monc->want_mdsmap) {
215 dout("__send_subscribe to 'mdsmap' %u+\n",
Eric Dumazet95c96172012-04-15 05:58:06 +0000216 (unsigned int)monc->have_mdsmap);
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700217 ceph_encode_string(&p, end, "mdsmap", 6);
218 i = p;
219 i->have = cpu_to_le64(monc->have_mdsmap);
220 i->onetime = 0;
221 p += sizeof(*i);
222 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800223 ceph_encode_string(&p, end, "monmap", 6);
224 i = p;
225 i->have = 0;
226 i->onetime = 0;
227 p += sizeof(*i);
Sage Weilba75bb92009-10-06 11:31:11 -0700228
229 msg->front.iov_len = p - msg->front.iov_base;
230 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len);
Alex Elder6740a842012-06-01 14:56:43 -0500231 ceph_msg_revoke(msg);
Alex Elder67130932012-05-26 23:26:43 -0500232 ceph_con_send(&monc->con, ceph_msg_get(msg));
Sage Weilba75bb92009-10-06 11:31:11 -0700233
234 monc->sub_sent = jiffies | 1; /* never 0 */
235 }
236}
237
238static void handle_subscribe_ack(struct ceph_mon_client *monc,
239 struct ceph_msg *msg)
240{
Eric Dumazet95c96172012-04-15 05:58:06 +0000241 unsigned int seconds;
Sage Weil07bd10f2009-10-14 17:26:40 -0700242 struct ceph_mon_subscribe_ack *h = msg->front.iov_base;
Sage Weilba75bb92009-10-06 11:31:11 -0700243
Sage Weil07bd10f2009-10-14 17:26:40 -0700244 if (msg->front.iov_len < sizeof(*h))
245 goto bad;
246 seconds = le32_to_cpu(h->duration);
247
Sage Weilba75bb92009-10-06 11:31:11 -0700248 mutex_lock(&monc->mutex);
249 if (monc->hunting) {
250 pr_info("mon%d %s session established\n",
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700251 monc->cur_mon,
Alex Elder67130932012-05-26 23:26:43 -0500252 ceph_pr_addr(&monc->con.peer_addr.in_addr));
Sage Weilba75bb92009-10-06 11:31:11 -0700253 monc->hunting = false;
254 }
255 dout("handle_subscribe_ack after %d seconds\n", seconds);
Sage Weil0656d112009-10-08 10:25:46 -0700256 monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1;
Sage Weilba75bb92009-10-06 11:31:11 -0700257 monc->sub_sent = 0;
258 mutex_unlock(&monc->mutex);
259 return;
260bad:
261 pr_err("got corrupt subscribe-ack msg\n");
Sage Weil9ec7cab2009-12-14 15:13:47 -0800262 ceph_msg_dump(msg);
Sage Weilba75bb92009-10-06 11:31:11 -0700263}
264
265/*
266 * Keep track of which maps we have
267 */
268int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got)
269{
270 mutex_lock(&monc->mutex);
271 monc->have_mdsmap = got;
272 mutex_unlock(&monc->mutex);
273 return 0;
274}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700275EXPORT_SYMBOL(ceph_monc_got_mdsmap);
Sage Weilba75bb92009-10-06 11:31:11 -0700276
277int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got)
278{
279 mutex_lock(&monc->mutex);
280 monc->have_osdmap = got;
281 monc->want_next_osdmap = 0;
282 mutex_unlock(&monc->mutex);
283 return 0;
284}
285
286/*
287 * Register interest in the next osdmap
288 */
289void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc)
290{
291 dout("request_next_osdmap have %u\n", monc->have_osdmap);
292 mutex_lock(&monc->mutex);
293 if (!monc->want_next_osdmap)
294 monc->want_next_osdmap = 1;
295 if (monc->want_next_osdmap < 2)
296 __send_subscribe(monc);
297 mutex_unlock(&monc->mutex);
298}
299
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800300/*
Sage Weil50b885b2009-12-01 14:12:07 -0800301 *
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800302 */
303int ceph_monc_open_session(struct ceph_mon_client *monc)
Sage Weilba75bb92009-10-06 11:31:11 -0700304{
Sage Weilba75bb92009-10-06 11:31:11 -0700305 mutex_lock(&monc->mutex);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800306 __open_session(monc);
Sage Weilba75bb92009-10-06 11:31:11 -0700307 __schedule_delayed(monc);
308 mutex_unlock(&monc->mutex);
309 return 0;
310}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700311EXPORT_SYMBOL(ceph_monc_open_session);
Sage Weilba75bb92009-10-06 11:31:11 -0700312
313/*
Sage Weild1c338a2012-08-19 12:29:16 -0700314 * We require the fsid and global_id in order to initialize our
315 * debugfs dir.
316 */
317static bool have_debugfs_info(struct ceph_mon_client *monc)
318{
319 dout("have_debugfs_info fsid %d globalid %lld\n",
320 (int)monc->client->have_fsid, monc->auth->global_id);
321 return monc->client->have_fsid && monc->auth->global_id > 0;
322}
323
324/*
Sage Weilba75bb92009-10-06 11:31:11 -0700325 * The monitor responds with mount ack indicate mount success. The
326 * included client ticket allows the client to talk to MDSs and OSDs.
327 */
Sage Weil07433042009-11-18 16:50:41 -0800328static void ceph_monc_handle_map(struct ceph_mon_client *monc,
329 struct ceph_msg *msg)
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800330{
331 struct ceph_client *client = monc->client;
332 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
333 void *p, *end;
Sage Weild1c338a2012-08-19 12:29:16 -0700334 int had_debugfs_info, init_debugfs = 0;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800335
336 mutex_lock(&monc->mutex);
337
Sage Weild1c338a2012-08-19 12:29:16 -0700338 had_debugfs_info = have_debugfs_info(monc);
339
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800340 dout("handle_monmap\n");
341 p = msg->front.iov_base;
342 end = p + msg->front.iov_len;
343
344 monmap = ceph_monmap_decode(p, end);
345 if (IS_ERR(monmap)) {
346 pr_err("problem decoding monmap, %d\n",
347 (int)PTR_ERR(monmap));
Sage Weild4a780c2009-12-11 08:55:23 -0800348 goto out;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800349 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800350
Sage Weil07433042009-11-18 16:50:41 -0800351 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800352 kfree(monmap);
Sage Weild4a780c2009-12-11 08:55:23 -0800353 goto out;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800354 }
355
356 client->monc.monmap = monmap;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800357 kfree(old);
358
Sage Weilab434b62012-01-13 22:22:03 -0800359 if (!client->have_fsid) {
360 client->have_fsid = true;
Sage Weild1c338a2012-08-19 12:29:16 -0700361 if (!had_debugfs_info && have_debugfs_info(monc)) {
362 pr_info("client%lld fsid %pU\n",
363 ceph_client_id(monc->client),
364 &monc->client->fsid);
365 init_debugfs = 1;
366 }
Sage Weilab434b62012-01-13 22:22:03 -0800367 mutex_unlock(&monc->mutex);
Sage Weild1c338a2012-08-19 12:29:16 -0700368
369 if (init_debugfs) {
370 /*
371 * do debugfs initialization without mutex to avoid
372 * creating a locking dependency
373 */
374 ceph_debugfs_client_init(monc->client);
375 }
376
Sage Weilab434b62012-01-13 22:22:03 -0800377 goto out_unlocked;
378 }
Sage Weild4a780c2009-12-11 08:55:23 -0800379out:
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800380 mutex_unlock(&monc->mutex);
Sage Weilab434b62012-01-13 22:22:03 -0800381out_unlocked:
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700382 wake_up_all(&client->auth_wq);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800383}
Sage Weilba75bb92009-10-06 11:31:11 -0700384
Sage Weilba75bb92009-10-06 11:31:11 -0700385/*
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700386 * generic requests (e.g., statfs, poolop)
Sage Weilba75bb92009-10-06 11:31:11 -0700387 */
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700388static struct ceph_mon_generic_request *__lookup_generic_req(
Sage Weil85ff03f2010-02-15 14:47:28 -0800389 struct ceph_mon_client *monc, u64 tid)
390{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700391 struct ceph_mon_generic_request *req;
392 struct rb_node *n = monc->generic_request_tree.rb_node;
Sage Weil85ff03f2010-02-15 14:47:28 -0800393
394 while (n) {
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700395 req = rb_entry(n, struct ceph_mon_generic_request, node);
Sage Weil85ff03f2010-02-15 14:47:28 -0800396 if (tid < req->tid)
397 n = n->rb_left;
398 else if (tid > req->tid)
399 n = n->rb_right;
400 else
401 return req;
402 }
403 return NULL;
404}
405
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700406static void __insert_generic_request(struct ceph_mon_client *monc,
407 struct ceph_mon_generic_request *new)
Sage Weil85ff03f2010-02-15 14:47:28 -0800408{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700409 struct rb_node **p = &monc->generic_request_tree.rb_node;
Sage Weil85ff03f2010-02-15 14:47:28 -0800410 struct rb_node *parent = NULL;
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700411 struct ceph_mon_generic_request *req = NULL;
Sage Weil85ff03f2010-02-15 14:47:28 -0800412
413 while (*p) {
414 parent = *p;
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700415 req = rb_entry(parent, struct ceph_mon_generic_request, node);
Sage Weil85ff03f2010-02-15 14:47:28 -0800416 if (new->tid < req->tid)
417 p = &(*p)->rb_left;
418 else if (new->tid > req->tid)
419 p = &(*p)->rb_right;
420 else
421 BUG();
422 }
423
424 rb_link_node(&new->node, parent, p);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700425 rb_insert_color(&new->node, &monc->generic_request_tree);
Sage Weil85ff03f2010-02-15 14:47:28 -0800426}
427
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700428static void release_generic_request(struct kref *kref)
Sage Weil3143edd2010-03-24 21:43:33 -0700429{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700430 struct ceph_mon_generic_request *req =
431 container_of(kref, struct ceph_mon_generic_request, kref);
Sage Weil3143edd2010-03-24 21:43:33 -0700432
433 if (req->reply)
434 ceph_msg_put(req->reply);
435 if (req->request)
436 ceph_msg_put(req->request);
Yehuda Sadeh20547562010-06-01 10:37:40 -0700437
438 kfree(req);
Sage Weil3143edd2010-03-24 21:43:33 -0700439}
440
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700441static void put_generic_request(struct ceph_mon_generic_request *req)
Sage Weil3143edd2010-03-24 21:43:33 -0700442{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700443 kref_put(&req->kref, release_generic_request);
Sage Weil3143edd2010-03-24 21:43:33 -0700444}
445
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700446static void get_generic_request(struct ceph_mon_generic_request *req)
Sage Weil3143edd2010-03-24 21:43:33 -0700447{
448 kref_get(&req->kref);
449}
450
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700451static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
Sage Weil3143edd2010-03-24 21:43:33 -0700452 struct ceph_msg_header *hdr,
453 int *skip)
454{
455 struct ceph_mon_client *monc = con->private;
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700456 struct ceph_mon_generic_request *req;
Sage Weil3143edd2010-03-24 21:43:33 -0700457 u64 tid = le64_to_cpu(hdr->tid);
458 struct ceph_msg *m;
459
460 mutex_lock(&monc->mutex);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700461 req = __lookup_generic_req(monc, tid);
Sage Weil3143edd2010-03-24 21:43:33 -0700462 if (!req) {
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700463 dout("get_generic_reply %lld dne\n", tid);
Sage Weil3143edd2010-03-24 21:43:33 -0700464 *skip = 1;
465 m = NULL;
466 } else {
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700467 dout("get_generic_reply %lld got %p\n", tid, req->reply);
Alex Elder1c20f2d2012-06-04 14:43:32 -0500468 *skip = 0;
Sage Weil3143edd2010-03-24 21:43:33 -0700469 m = ceph_msg_get(req->reply);
470 /*
471 * we don't need to track the connection reading into
472 * this reply because we only have one open connection
473 * at a time, ever.
474 */
475 }
476 mutex_unlock(&monc->mutex);
477 return m;
478}
479
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700480static int do_generic_request(struct ceph_mon_client *monc,
481 struct ceph_mon_generic_request *req)
482{
483 int err;
484
485 /* register request */
486 mutex_lock(&monc->mutex);
487 req->tid = ++monc->last_tid;
488 req->request->hdr.tid = cpu_to_le64(req->tid);
489 __insert_generic_request(monc, req);
490 monc->num_generic_requests++;
Alex Elder67130932012-05-26 23:26:43 -0500491 ceph_con_send(&monc->con, ceph_msg_get(req->request));
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700492 mutex_unlock(&monc->mutex);
493
494 err = wait_for_completion_interruptible(&req->completion);
495
496 mutex_lock(&monc->mutex);
497 rb_erase(&req->node, &monc->generic_request_tree);
498 monc->num_generic_requests--;
499 mutex_unlock(&monc->mutex);
500
501 if (!err)
502 err = req->result;
503 return err;
504}
505
506/*
507 * statfs
508 */
Sage Weilba75bb92009-10-06 11:31:11 -0700509static void handle_statfs_reply(struct ceph_mon_client *monc,
510 struct ceph_msg *msg)
511{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700512 struct ceph_mon_generic_request *req;
Sage Weilba75bb92009-10-06 11:31:11 -0700513 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
Sage Weil3143edd2010-03-24 21:43:33 -0700514 u64 tid = le64_to_cpu(msg->hdr.tid);
Sage Weilba75bb92009-10-06 11:31:11 -0700515
516 if (msg->front.iov_len != sizeof(*reply))
517 goto bad;
Sage Weilba75bb92009-10-06 11:31:11 -0700518 dout("handle_statfs_reply %p tid %llu\n", msg, tid);
519
520 mutex_lock(&monc->mutex);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700521 req = __lookup_generic_req(monc, tid);
Sage Weilba75bb92009-10-06 11:31:11 -0700522 if (req) {
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700523 *(struct ceph_statfs *)req->buf = reply->st;
Sage Weilba75bb92009-10-06 11:31:11 -0700524 req->result = 0;
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700525 get_generic_request(req);
Sage Weilba75bb92009-10-06 11:31:11 -0700526 }
527 mutex_unlock(&monc->mutex);
Sage Weil3143edd2010-03-24 21:43:33 -0700528 if (req) {
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700529 complete_all(&req->completion);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700530 put_generic_request(req);
Sage Weil3143edd2010-03-24 21:43:33 -0700531 }
Sage Weilba75bb92009-10-06 11:31:11 -0700532 return;
533
534bad:
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700535 pr_err("corrupt generic reply, tid %llu\n", tid);
Sage Weil9ec7cab2009-12-14 15:13:47 -0800536 ceph_msg_dump(msg);
Sage Weilba75bb92009-10-06 11:31:11 -0700537}
538
539/*
Sage Weilba75bb92009-10-06 11:31:11 -0700540 * Do a synchronous statfs().
541 */
542int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
543{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700544 struct ceph_mon_generic_request *req;
Sage Weil3143edd2010-03-24 21:43:33 -0700545 struct ceph_mon_statfs *h;
Sage Weilba75bb92009-10-06 11:31:11 -0700546 int err;
547
Julia Lawallcffe7b62010-05-13 22:07:29 +0200548 req = kzalloc(sizeof(*req), GFP_NOFS);
Sage Weil3143edd2010-03-24 21:43:33 -0700549 if (!req)
550 return -ENOMEM;
Sage Weilba75bb92009-10-06 11:31:11 -0700551
Sage Weil3143edd2010-03-24 21:43:33 -0700552 kref_init(&req->kref);
553 req->buf = buf;
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700554 req->buf_len = sizeof(*buf);
Sage Weil3143edd2010-03-24 21:43:33 -0700555 init_completion(&req->completion);
556
Sage Weila79832f2010-04-01 16:06:19 -0700557 err = -ENOMEM;
Sage Weilb61c2762011-08-09 15:03:46 -0700558 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
559 true);
Sage Weila79832f2010-04-01 16:06:19 -0700560 if (!req->request)
Sage Weil3143edd2010-03-24 21:43:33 -0700561 goto out;
Sage Weilb61c2762011-08-09 15:03:46 -0700562 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
563 true);
Sage Weila79832f2010-04-01 16:06:19 -0700564 if (!req->reply)
Sage Weil3143edd2010-03-24 21:43:33 -0700565 goto out;
Sage Weil3143edd2010-03-24 21:43:33 -0700566
567 /* fill out request */
568 h = req->request->front.iov_base;
569 h->monhdr.have_version = 0;
570 h->monhdr.session_mon = cpu_to_le16(-1);
571 h->monhdr.session_mon_tid = 0;
572 h->fsid = monc->monmap->fsid;
Sage Weilba75bb92009-10-06 11:31:11 -0700573
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700574 err = do_generic_request(monc, req);
Sage Weil3143edd2010-03-24 21:43:33 -0700575
576out:
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700577 kref_put(&req->kref, release_generic_request);
Sage Weilba75bb92009-10-06 11:31:11 -0700578 return err;
579}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700580EXPORT_SYMBOL(ceph_monc_do_statfs);
Sage Weilba75bb92009-10-06 11:31:11 -0700581
582/*
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700583 * pool ops
584 */
585static int get_poolop_reply_buf(const char *src, size_t src_len,
586 char *dst, size_t dst_len)
587{
588 u32 buf_len;
589
590 if (src_len != sizeof(u32) + dst_len)
591 return -EINVAL;
592
593 buf_len = le32_to_cpu(*(u32 *)src);
594 if (buf_len != dst_len)
595 return -EINVAL;
596
597 memcpy(dst, src + sizeof(u32), dst_len);
598 return 0;
599}
600
601static void handle_poolop_reply(struct ceph_mon_client *monc,
602 struct ceph_msg *msg)
603{
604 struct ceph_mon_generic_request *req;
605 struct ceph_mon_poolop_reply *reply = msg->front.iov_base;
606 u64 tid = le64_to_cpu(msg->hdr.tid);
607
608 if (msg->front.iov_len < sizeof(*reply))
609 goto bad;
610 dout("handle_poolop_reply %p tid %llu\n", msg, tid);
611
612 mutex_lock(&monc->mutex);
613 req = __lookup_generic_req(monc, tid);
614 if (req) {
615 if (req->buf_len &&
616 get_poolop_reply_buf(msg->front.iov_base + sizeof(*reply),
617 msg->front.iov_len - sizeof(*reply),
618 req->buf, req->buf_len) < 0) {
619 mutex_unlock(&monc->mutex);
620 goto bad;
621 }
622 req->result = le32_to_cpu(reply->reply_code);
623 get_generic_request(req);
624 }
625 mutex_unlock(&monc->mutex);
626 if (req) {
627 complete(&req->completion);
628 put_generic_request(req);
629 }
630 return;
631
632bad:
633 pr_err("corrupt generic reply, tid %llu\n", tid);
634 ceph_msg_dump(msg);
635}
636
637/*
638 * Do a synchronous pool op.
639 */
Iulius Curt7698f2f2012-08-23 15:14:29 +0300640static int do_poolop(struct ceph_mon_client *monc, u32 op,
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700641 u32 pool, u64 snapid,
642 char *buf, int len)
643{
644 struct ceph_mon_generic_request *req;
645 struct ceph_mon_poolop *h;
646 int err;
647
648 req = kzalloc(sizeof(*req), GFP_NOFS);
649 if (!req)
650 return -ENOMEM;
651
652 kref_init(&req->kref);
653 req->buf = buf;
654 req->buf_len = len;
655 init_completion(&req->completion);
656
657 err = -ENOMEM;
Sage Weilb61c2762011-08-09 15:03:46 -0700658 req->request = ceph_msg_new(CEPH_MSG_POOLOP, sizeof(*h), GFP_NOFS,
659 true);
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700660 if (!req->request)
661 goto out;
Sage Weilb61c2762011-08-09 15:03:46 -0700662 req->reply = ceph_msg_new(CEPH_MSG_POOLOP_REPLY, 1024, GFP_NOFS,
663 true);
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700664 if (!req->reply)
665 goto out;
666
667 /* fill out request */
668 req->request->hdr.version = cpu_to_le16(2);
669 h = req->request->front.iov_base;
670 h->monhdr.have_version = 0;
671 h->monhdr.session_mon = cpu_to_le16(-1);
672 h->monhdr.session_mon_tid = 0;
673 h->fsid = monc->monmap->fsid;
674 h->pool = cpu_to_le32(pool);
675 h->op = cpu_to_le32(op);
676 h->auid = 0;
677 h->snapid = cpu_to_le64(snapid);
678 h->name_len = 0;
679
680 err = do_generic_request(monc, req);
681
682out:
683 kref_put(&req->kref, release_generic_request);
684 return err;
685}
686
687int ceph_monc_create_snapid(struct ceph_mon_client *monc,
688 u32 pool, u64 *snapid)
689{
Iulius Curt7698f2f2012-08-23 15:14:29 +0300690 return do_poolop(monc, POOL_OP_CREATE_UNMANAGED_SNAP,
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700691 pool, 0, (char *)snapid, sizeof(*snapid));
692
693}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700694EXPORT_SYMBOL(ceph_monc_create_snapid);
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700695
696int ceph_monc_delete_snapid(struct ceph_mon_client *monc,
697 u32 pool, u64 snapid)
698{
Iulius Curt7698f2f2012-08-23 15:14:29 +0300699 return do_poolop(monc, POOL_OP_CREATE_UNMANAGED_SNAP,
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700700 pool, snapid, 0, 0);
701
702}
703
704/*
705 * Resend pending generic requests.
Sage Weilba75bb92009-10-06 11:31:11 -0700706 */
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700707static void __resend_generic_request(struct ceph_mon_client *monc)
Sage Weilba75bb92009-10-06 11:31:11 -0700708{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700709 struct ceph_mon_generic_request *req;
Sage Weil85ff03f2010-02-15 14:47:28 -0800710 struct rb_node *p;
Sage Weilba75bb92009-10-06 11:31:11 -0700711
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700712 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
713 req = rb_entry(p, struct ceph_mon_generic_request, node);
Alex Elder6740a842012-06-01 14:56:43 -0500714 ceph_msg_revoke(req->request);
Sage Weil4f471e42012-07-30 18:16:40 -0700715 ceph_msg_revoke_incoming(req->reply);
Alex Elder67130932012-05-26 23:26:43 -0500716 ceph_con_send(&monc->con, ceph_msg_get(req->request));
Sage Weilba75bb92009-10-06 11:31:11 -0700717 }
718}
719
720/*
721 * Delayed work. If we haven't mounted yet, retry. Otherwise,
722 * renew/retry subscription as needed (in case it is timing out, or we
723 * got an ENOMEM). And keep the monitor connection alive.
724 */
725static void delayed_work(struct work_struct *work)
726{
727 struct ceph_mon_client *monc =
728 container_of(work, struct ceph_mon_client, delayed_work.work);
729
730 dout("monc delayed_work\n");
731 mutex_lock(&monc->mutex);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800732 if (monc->hunting) {
733 __close_session(monc);
734 __open_session(monc); /* continue hunting */
Sage Weilba75bb92009-10-06 11:31:11 -0700735 } else {
Alex Elder67130932012-05-26 23:26:43 -0500736 ceph_con_keepalive(&monc->con);
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800737
738 __validate_auth(monc);
739
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800740 if (monc->auth->ops->is_authenticated(monc->auth))
741 __send_subscribe(monc);
Sage Weilba75bb92009-10-06 11:31:11 -0700742 }
Sage Weilba75bb92009-10-06 11:31:11 -0700743 __schedule_delayed(monc);
744 mutex_unlock(&monc->mutex);
745}
746
Sage Weil6b805182009-10-27 11:50:50 -0700747/*
748 * On startup, we build a temporary monmap populated with the IPs
749 * provided by mount(2).
750 */
751static int build_initial_monmap(struct ceph_mon_client *monc)
752{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700753 struct ceph_options *opt = monc->client->options;
754 struct ceph_entity_addr *mon_addr = opt->mon_addr;
755 int num_mon = opt->num_mon;
Sage Weil6b805182009-10-27 11:50:50 -0700756 int i;
757
758 /* build initial monmap */
759 monc->monmap = kzalloc(sizeof(*monc->monmap) +
760 num_mon*sizeof(monc->monmap->mon_inst[0]),
761 GFP_KERNEL);
762 if (!monc->monmap)
763 return -ENOMEM;
764 for (i = 0; i < num_mon; i++) {
765 monc->monmap->mon_inst[i].addr = mon_addr[i];
Sage Weil6b805182009-10-27 11:50:50 -0700766 monc->monmap->mon_inst[i].addr.nonce = 0;
767 monc->monmap->mon_inst[i].name.type =
768 CEPH_ENTITY_TYPE_MON;
769 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
770 }
771 monc->monmap->num_mon = num_mon;
Sage Weil6b805182009-10-27 11:50:50 -0700772 return 0;
773}
774
Sage Weilba75bb92009-10-06 11:31:11 -0700775int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
776{
777 int err = 0;
778
779 dout("init\n");
780 memset(monc, 0, sizeof(*monc));
781 monc->client = cl;
782 monc->monmap = NULL;
783 mutex_init(&monc->mutex);
784
Sage Weil6b805182009-10-27 11:50:50 -0700785 err = build_initial_monmap(monc);
786 if (err)
787 goto out;
788
Sage Weilf6a2f5b2011-08-09 09:27:44 -0700789 /* connection */
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800790 /* authentication */
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700791 monc->auth = ceph_auth_init(cl->options->name,
Tommi Virtanen8323c3a2011-03-25 16:32:57 -0700792 cl->options->key);
Noah Watkins49d92242011-09-12 14:51:58 -0700793 if (IS_ERR(monc->auth)) {
794 err = PTR_ERR(monc->auth);
Alex Elder67130932012-05-26 23:26:43 -0500795 goto out_monmap;
Noah Watkins49d92242011-09-12 14:51:58 -0700796 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800797 monc->auth->want_keys =
798 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
799 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
800
Sage Weil240ed682010-05-21 14:57:25 -0700801 /* msgs */
Sage Weila79832f2010-04-01 16:06:19 -0700802 err = -ENOMEM;
Sage Weil7c315c52010-03-24 21:52:30 -0700803 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700804 sizeof(struct ceph_mon_subscribe_ack),
Sage Weilb61c2762011-08-09 15:03:46 -0700805 GFP_NOFS, true);
Sage Weila79832f2010-04-01 16:06:19 -0700806 if (!monc->m_subscribe_ack)
Noah Watkins49d92242011-09-12 14:51:58 -0700807 goto out_auth;
Sage Weil6694d6b2010-03-24 21:48:05 -0700808
Sage Weilb61c2762011-08-09 15:03:46 -0700809 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
810 true);
Sage Weil240ed682010-05-21 14:57:25 -0700811 if (!monc->m_subscribe)
812 goto out_subscribe_ack;
813
Sage Weilb61c2762011-08-09 15:03:46 -0700814 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
815 true);
Sage Weila79832f2010-04-01 16:06:19 -0700816 if (!monc->m_auth_reply)
Sage Weil240ed682010-05-21 14:57:25 -0700817 goto out_subscribe;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800818
Sage Weilb61c2762011-08-09 15:03:46 -0700819 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800820 monc->pending_auth = 0;
Sage Weila79832f2010-04-01 16:06:19 -0700821 if (!monc->m_auth)
Sage Weil6694d6b2010-03-24 21:48:05 -0700822 goto out_auth_reply;
Sage Weilba75bb92009-10-06 11:31:11 -0700823
Sage Weil735a72e2012-06-27 12:24:34 -0700824 ceph_con_init(&monc->con, monc, &mon_con_ops,
825 &monc->client->msgr);
826
Sage Weilba75bb92009-10-06 11:31:11 -0700827 monc->cur_mon = -1;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800828 monc->hunting = true;
Sage Weilba75bb92009-10-06 11:31:11 -0700829 monc->sub_renew_after = jiffies;
830 monc->sub_sent = 0;
831
832 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700833 monc->generic_request_tree = RB_ROOT;
834 monc->num_generic_requests = 0;
Sage Weilba75bb92009-10-06 11:31:11 -0700835 monc->last_tid = 0;
836
837 monc->have_mdsmap = 0;
838 monc->have_osdmap = 0;
839 monc->want_next_osdmap = 1;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800840 return 0;
841
Sage Weil6694d6b2010-03-24 21:48:05 -0700842out_auth_reply:
843 ceph_msg_put(monc->m_auth_reply);
Sage Weil240ed682010-05-21 14:57:25 -0700844out_subscribe:
845 ceph_msg_put(monc->m_subscribe);
Sage Weil7c315c52010-03-24 21:52:30 -0700846out_subscribe_ack:
847 ceph_msg_put(monc->m_subscribe_ack);
Noah Watkins49d92242011-09-12 14:51:58 -0700848out_auth:
849 ceph_auth_destroy(monc->auth);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800850out_monmap:
851 kfree(monc->monmap);
Sage Weilba75bb92009-10-06 11:31:11 -0700852out:
853 return err;
854}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700855EXPORT_SYMBOL(ceph_monc_init);
Sage Weilba75bb92009-10-06 11:31:11 -0700856
857void ceph_monc_stop(struct ceph_mon_client *monc)
858{
859 dout("stop\n");
860 cancel_delayed_work_sync(&monc->delayed_work);
861
862 mutex_lock(&monc->mutex);
863 __close_session(monc);
Sage Weilf6a2f5b2011-08-09 09:27:44 -0700864
Sage Weilba75bb92009-10-06 11:31:11 -0700865 mutex_unlock(&monc->mutex);
866
Sage Weilf3dea7e2012-06-10 20:43:56 -0700867 /*
868 * flush msgr queue before we destroy ourselves to ensure that:
869 * - any work that references our embedded con is finished.
870 * - any osd_client or other work that may reference an authorizer
871 * finishes before we shut down the auth subsystem.
872 */
873 ceph_msgr_flush();
874
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800875 ceph_auth_destroy(monc->auth);
876
877 ceph_msg_put(monc->m_auth);
Sage Weil6694d6b2010-03-24 21:48:05 -0700878 ceph_msg_put(monc->m_auth_reply);
Sage Weil240ed682010-05-21 14:57:25 -0700879 ceph_msg_put(monc->m_subscribe);
Sage Weil7c315c52010-03-24 21:52:30 -0700880 ceph_msg_put(monc->m_subscribe_ack);
Sage Weilba75bb92009-10-06 11:31:11 -0700881
882 kfree(monc->monmap);
883}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700884EXPORT_SYMBOL(ceph_monc_stop);
Sage Weilba75bb92009-10-06 11:31:11 -0700885
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800886static void handle_auth_reply(struct ceph_mon_client *monc,
887 struct ceph_msg *msg)
888{
889 int ret;
Sage Weil09c4d6a2010-05-25 15:38:06 -0700890 int was_auth = 0;
Sage Weild1c338a2012-08-19 12:29:16 -0700891 int had_debugfs_info, init_debugfs = 0;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800892
893 mutex_lock(&monc->mutex);
Sage Weild1c338a2012-08-19 12:29:16 -0700894 had_debugfs_info = have_debugfs_info(monc);
Sage Weil09c4d6a2010-05-25 15:38:06 -0700895 if (monc->auth->ops)
896 was_auth = monc->auth->ops->is_authenticated(monc->auth);
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800897 monc->pending_auth = 0;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800898 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
899 msg->front.iov_len,
900 monc->m_auth->front.iov_base,
901 monc->m_auth->front_max);
902 if (ret < 0) {
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800903 monc->client->auth_err = ret;
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700904 wake_up_all(&monc->client->auth_wq);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800905 } else if (ret > 0) {
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800906 __send_prepared_auth_request(monc, ret);
Sage Weil09c4d6a2010-05-25 15:38:06 -0700907 } else if (!was_auth && monc->auth->ops->is_authenticated(monc->auth)) {
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800908 dout("authenticated, starting session\n");
Sage Weil07433042009-11-18 16:50:41 -0800909
Alex Elder15d98822012-05-26 23:26:43 -0500910 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
911 monc->client->msgr.inst.name.num =
Yehuda Sadeh0cf55372010-06-11 15:57:06 -0700912 cpu_to_le64(monc->auth->global_id);
Sage Weil07433042009-11-18 16:50:41 -0800913
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800914 __send_subscribe(monc);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700915 __resend_generic_request(monc);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800916 }
Sage Weild1c338a2012-08-19 12:29:16 -0700917
918 if (!had_debugfs_info && have_debugfs_info(monc)) {
919 pr_info("client%lld fsid %pU\n",
920 ceph_client_id(monc->client),
921 &monc->client->fsid);
922 init_debugfs = 1;
923 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800924 mutex_unlock(&monc->mutex);
Sage Weild1c338a2012-08-19 12:29:16 -0700925
926 if (init_debugfs) {
927 /*
928 * do debugfs initialization without mutex to avoid
929 * creating a locking dependency
930 */
931 ceph_debugfs_client_init(monc->client);
932 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800933}
934
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800935static int __validate_auth(struct ceph_mon_client *monc)
936{
937 int ret;
938
939 if (monc->pending_auth)
940 return 0;
941
942 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
943 monc->m_auth->front_max);
944 if (ret <= 0)
945 return ret; /* either an error, or no need to authenticate */
946 __send_prepared_auth_request(monc, ret);
947 return 0;
948}
949
950int ceph_monc_validate_auth(struct ceph_mon_client *monc)
951{
952 int ret;
953
954 mutex_lock(&monc->mutex);
955 ret = __validate_auth(monc);
956 mutex_unlock(&monc->mutex);
957 return ret;
958}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700959EXPORT_SYMBOL(ceph_monc_validate_auth);
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800960
Sage Weilba75bb92009-10-06 11:31:11 -0700961/*
962 * handle incoming message
963 */
964static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
965{
966 struct ceph_mon_client *monc = con->private;
967 int type = le16_to_cpu(msg->hdr.type);
968
969 if (!monc)
970 return;
971
972 switch (type) {
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800973 case CEPH_MSG_AUTH_REPLY:
974 handle_auth_reply(monc, msg);
Sage Weilba75bb92009-10-06 11:31:11 -0700975 break;
976
977 case CEPH_MSG_MON_SUBSCRIBE_ACK:
978 handle_subscribe_ack(monc, msg);
979 break;
980
981 case CEPH_MSG_STATFS_REPLY:
982 handle_statfs_reply(monc, msg);
983 break;
984
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700985 case CEPH_MSG_POOLOP_REPLY:
986 handle_poolop_reply(monc, msg);
987 break;
988
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800989 case CEPH_MSG_MON_MAP:
990 ceph_monc_handle_map(monc, msg);
991 break;
992
Sage Weilba75bb92009-10-06 11:31:11 -0700993 case CEPH_MSG_OSD_MAP:
994 ceph_osdc_handle_map(&monc->client->osdc, msg);
995 break;
996
997 default:
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700998 /* can the chained handler handle it? */
999 if (monc->client->extra_mon_dispatch &&
1000 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1001 break;
1002
Sage Weilba75bb92009-10-06 11:31:11 -07001003 pr_err("received unknown message type %d %s\n", type,
1004 ceph_msg_type_name(type));
1005 }
1006 ceph_msg_put(msg);
1007}
1008
1009/*
1010 * Allocate memory for incoming message
1011 */
1012static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
Yehuda Sadeh24504182010-01-08 13:58:34 -08001013 struct ceph_msg_header *hdr,
1014 int *skip)
Sage Weilba75bb92009-10-06 11:31:11 -07001015{
1016 struct ceph_mon_client *monc = con->private;
1017 int type = le16_to_cpu(hdr->type);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001018 int front_len = le32_to_cpu(hdr->front_len);
Sage Weil5b3a4db2010-02-19 21:43:23 -08001019 struct ceph_msg *m = NULL;
Sage Weilba75bb92009-10-06 11:31:11 -07001020
Yehuda Sadeh24504182010-01-08 13:58:34 -08001021 *skip = 0;
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08001022
Sage Weilba75bb92009-10-06 11:31:11 -07001023 switch (type) {
Sage Weilba75bb92009-10-06 11:31:11 -07001024 case CEPH_MSG_MON_SUBSCRIBE_ACK:
Sage Weil7c315c52010-03-24 21:52:30 -07001025 m = ceph_msg_get(monc->m_subscribe_ack);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001026 break;
Yehuda Sadehe56fa102010-05-17 12:40:28 -07001027 case CEPH_MSG_POOLOP_REPLY:
Sage Weilba75bb92009-10-06 11:31:11 -07001028 case CEPH_MSG_STATFS_REPLY:
Yehuda Sadehf8c76f62010-04-22 15:40:37 -07001029 return get_generic_reply(con, hdr, skip);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001030 case CEPH_MSG_AUTH_REPLY:
Sage Weil6694d6b2010-03-24 21:48:05 -07001031 m = ceph_msg_get(monc->m_auth_reply);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001032 break;
Sage Weil5b3a4db2010-02-19 21:43:23 -08001033 case CEPH_MSG_MON_MAP:
1034 case CEPH_MSG_MDS_MAP:
1035 case CEPH_MSG_OSD_MAP:
Sage Weilb61c2762011-08-09 15:03:46 -07001036 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
Alex Elder1c20f2d2012-06-04 14:43:32 -05001037 if (!m)
1038 return NULL; /* ENOMEM--return skip == 0 */
Sage Weil5b3a4db2010-02-19 21:43:23 -08001039 break;
Sage Weilba75bb92009-10-06 11:31:11 -07001040 }
Yehuda Sadeh24504182010-01-08 13:58:34 -08001041
Sage Weil5b3a4db2010-02-19 21:43:23 -08001042 if (!m) {
1043 pr_info("alloc_msg unknown type %d\n", type);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001044 *skip = 1;
Sage Weil5b3a4db2010-02-19 21:43:23 -08001045 }
Yehuda Sadeh24504182010-01-08 13:58:34 -08001046 return m;
Sage Weilba75bb92009-10-06 11:31:11 -07001047}
1048
1049/*
1050 * If the monitor connection resets, pick a new monitor and resubmit
1051 * any pending requests.
1052 */
1053static void mon_fault(struct ceph_connection *con)
1054{
1055 struct ceph_mon_client *monc = con->private;
1056
1057 if (!monc)
1058 return;
1059
1060 dout("mon_fault\n");
1061 mutex_lock(&monc->mutex);
1062 if (!con->private)
1063 goto out;
1064
Sage Weilf6a2f5b2011-08-09 09:27:44 -07001065 if (!monc->hunting)
Sage Weilba75bb92009-10-06 11:31:11 -07001066 pr_info("mon%d %s session lost, "
1067 "hunting for new mon\n", monc->cur_mon,
Alex Elder67130932012-05-26 23:26:43 -05001068 ceph_pr_addr(&monc->con.peer_addr.in_addr));
Sage Weilba75bb92009-10-06 11:31:11 -07001069
1070 __close_session(monc);
1071 if (!monc->hunting) {
1072 /* start hunting */
1073 monc->hunting = true;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001074 __open_session(monc);
Sage Weilba75bb92009-10-06 11:31:11 -07001075 } else {
1076 /* already hunting, let's wait a bit */
1077 __schedule_delayed(monc);
1078 }
1079out:
1080 mutex_unlock(&monc->mutex);
1081}
1082
Sage Weilec87ef42012-05-31 20:27:50 -07001083/*
1084 * We can ignore refcounting on the connection struct, as all references
1085 * will come from the messenger workqueue, which is drained prior to
1086 * mon_client destruction.
1087 */
1088static struct ceph_connection *con_get(struct ceph_connection *con)
1089{
1090 return con;
1091}
1092
1093static void con_put(struct ceph_connection *con)
1094{
1095}
1096
Tobias Klauser9e327892010-05-20 10:40:19 +02001097static const struct ceph_connection_operations mon_con_ops = {
Sage Weilec87ef42012-05-31 20:27:50 -07001098 .get = con_get,
1099 .put = con_put,
Sage Weilba75bb92009-10-06 11:31:11 -07001100 .dispatch = dispatch,
1101 .fault = mon_fault,
1102 .alloc_msg = mon_alloc_msg,
Sage Weilba75bb92009-10-06 11:31:11 -07001103};