blob: 067d3af2eaf61be41d601dc4324fc1a1565ab27d [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,
Ilya Dryomov3cea4c32014-01-09 20:08:21 +0200155 monc->m_auth->front_alloc_len);
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;
Ilya Dryomov3cea4c32014-01-09 20:08:21 +0200199 end = p + msg->front_alloc_len;
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}
Ilya Dryomov6044cde2014-05-13 11:19:27 +0400299EXPORT_SYMBOL(ceph_monc_request_next_osdmap);
300
301int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch,
302 unsigned long timeout)
303{
304 unsigned long started = jiffies;
305 int ret;
306
307 mutex_lock(&monc->mutex);
308 while (monc->have_osdmap < epoch) {
309 mutex_unlock(&monc->mutex);
310
311 if (timeout != 0 && time_after_eq(jiffies, started + timeout))
312 return -ETIMEDOUT;
313
314 ret = wait_event_interruptible_timeout(monc->client->auth_wq,
315 monc->have_osdmap >= epoch, timeout);
316 if (ret < 0)
317 return ret;
318
319 mutex_lock(&monc->mutex);
320 }
321
322 mutex_unlock(&monc->mutex);
323 return 0;
324}
325EXPORT_SYMBOL(ceph_monc_wait_osdmap);
Sage Weilba75bb92009-10-06 11:31:11 -0700326
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800327/*
Sage Weil50b885b2009-12-01 14:12:07 -0800328 *
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800329 */
330int ceph_monc_open_session(struct ceph_mon_client *monc)
Sage Weilba75bb92009-10-06 11:31:11 -0700331{
Sage Weilba75bb92009-10-06 11:31:11 -0700332 mutex_lock(&monc->mutex);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800333 __open_session(monc);
Sage Weilba75bb92009-10-06 11:31:11 -0700334 __schedule_delayed(monc);
335 mutex_unlock(&monc->mutex);
336 return 0;
337}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700338EXPORT_SYMBOL(ceph_monc_open_session);
Sage Weilba75bb92009-10-06 11:31:11 -0700339
340/*
Sage Weild1c338a2012-08-19 12:29:16 -0700341 * We require the fsid and global_id in order to initialize our
342 * debugfs dir.
343 */
344static bool have_debugfs_info(struct ceph_mon_client *monc)
345{
346 dout("have_debugfs_info fsid %d globalid %lld\n",
347 (int)monc->client->have_fsid, monc->auth->global_id);
348 return monc->client->have_fsid && monc->auth->global_id > 0;
349}
350
351/*
Sage Weilba75bb92009-10-06 11:31:11 -0700352 * The monitor responds with mount ack indicate mount success. The
353 * included client ticket allows the client to talk to MDSs and OSDs.
354 */
Sage Weil07433042009-11-18 16:50:41 -0800355static void ceph_monc_handle_map(struct ceph_mon_client *monc,
356 struct ceph_msg *msg)
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800357{
358 struct ceph_client *client = monc->client;
359 struct ceph_monmap *monmap = NULL, *old = monc->monmap;
360 void *p, *end;
Sage Weild1c338a2012-08-19 12:29:16 -0700361 int had_debugfs_info, init_debugfs = 0;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800362
363 mutex_lock(&monc->mutex);
364
Sage Weild1c338a2012-08-19 12:29:16 -0700365 had_debugfs_info = have_debugfs_info(monc);
366
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800367 dout("handle_monmap\n");
368 p = msg->front.iov_base;
369 end = p + msg->front.iov_len;
370
371 monmap = ceph_monmap_decode(p, end);
372 if (IS_ERR(monmap)) {
373 pr_err("problem decoding monmap, %d\n",
374 (int)PTR_ERR(monmap));
Sage Weild4a780c2009-12-11 08:55:23 -0800375 goto out;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800376 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800377
Sage Weil07433042009-11-18 16:50:41 -0800378 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) {
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800379 kfree(monmap);
Sage Weild4a780c2009-12-11 08:55:23 -0800380 goto out;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800381 }
382
383 client->monc.monmap = monmap;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800384 kfree(old);
385
Sage Weilab434b62012-01-13 22:22:03 -0800386 if (!client->have_fsid) {
387 client->have_fsid = true;
Sage Weild1c338a2012-08-19 12:29:16 -0700388 if (!had_debugfs_info && have_debugfs_info(monc)) {
389 pr_info("client%lld fsid %pU\n",
390 ceph_client_id(monc->client),
391 &monc->client->fsid);
392 init_debugfs = 1;
393 }
Sage Weilab434b62012-01-13 22:22:03 -0800394 mutex_unlock(&monc->mutex);
Sage Weild1c338a2012-08-19 12:29:16 -0700395
396 if (init_debugfs) {
397 /*
398 * do debugfs initialization without mutex to avoid
399 * creating a locking dependency
400 */
401 ceph_debugfs_client_init(monc->client);
402 }
403
Sage Weilab434b62012-01-13 22:22:03 -0800404 goto out_unlocked;
405 }
Sage Weild4a780c2009-12-11 08:55:23 -0800406out:
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800407 mutex_unlock(&monc->mutex);
Sage Weilab434b62012-01-13 22:22:03 -0800408out_unlocked:
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700409 wake_up_all(&client->auth_wq);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800410}
Sage Weilba75bb92009-10-06 11:31:11 -0700411
Sage Weilba75bb92009-10-06 11:31:11 -0700412/*
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700413 * generic requests (e.g., statfs, poolop)
Sage Weilba75bb92009-10-06 11:31:11 -0700414 */
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700415static struct ceph_mon_generic_request *__lookup_generic_req(
Sage Weil85ff03f2010-02-15 14:47:28 -0800416 struct ceph_mon_client *monc, u64 tid)
417{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700418 struct ceph_mon_generic_request *req;
419 struct rb_node *n = monc->generic_request_tree.rb_node;
Sage Weil85ff03f2010-02-15 14:47:28 -0800420
421 while (n) {
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700422 req = rb_entry(n, struct ceph_mon_generic_request, node);
Sage Weil85ff03f2010-02-15 14:47:28 -0800423 if (tid < req->tid)
424 n = n->rb_left;
425 else if (tid > req->tid)
426 n = n->rb_right;
427 else
428 return req;
429 }
430 return NULL;
431}
432
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700433static void __insert_generic_request(struct ceph_mon_client *monc,
434 struct ceph_mon_generic_request *new)
Sage Weil85ff03f2010-02-15 14:47:28 -0800435{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700436 struct rb_node **p = &monc->generic_request_tree.rb_node;
Sage Weil85ff03f2010-02-15 14:47:28 -0800437 struct rb_node *parent = NULL;
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700438 struct ceph_mon_generic_request *req = NULL;
Sage Weil85ff03f2010-02-15 14:47:28 -0800439
440 while (*p) {
441 parent = *p;
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700442 req = rb_entry(parent, struct ceph_mon_generic_request, node);
Sage Weil85ff03f2010-02-15 14:47:28 -0800443 if (new->tid < req->tid)
444 p = &(*p)->rb_left;
445 else if (new->tid > req->tid)
446 p = &(*p)->rb_right;
447 else
448 BUG();
449 }
450
451 rb_link_node(&new->node, parent, p);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700452 rb_insert_color(&new->node, &monc->generic_request_tree);
Sage Weil85ff03f2010-02-15 14:47:28 -0800453}
454
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700455static void release_generic_request(struct kref *kref)
Sage Weil3143edd2010-03-24 21:43:33 -0700456{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700457 struct ceph_mon_generic_request *req =
458 container_of(kref, struct ceph_mon_generic_request, kref);
Sage Weil3143edd2010-03-24 21:43:33 -0700459
460 if (req->reply)
461 ceph_msg_put(req->reply);
462 if (req->request)
463 ceph_msg_put(req->request);
Yehuda Sadeh20547562010-06-01 10:37:40 -0700464
465 kfree(req);
Sage Weil3143edd2010-03-24 21:43:33 -0700466}
467
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700468static void put_generic_request(struct ceph_mon_generic_request *req)
Sage Weil3143edd2010-03-24 21:43:33 -0700469{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700470 kref_put(&req->kref, release_generic_request);
Sage Weil3143edd2010-03-24 21:43:33 -0700471}
472
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700473static void get_generic_request(struct ceph_mon_generic_request *req)
Sage Weil3143edd2010-03-24 21:43:33 -0700474{
475 kref_get(&req->kref);
476}
477
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700478static struct ceph_msg *get_generic_reply(struct ceph_connection *con,
Sage Weil3143edd2010-03-24 21:43:33 -0700479 struct ceph_msg_header *hdr,
480 int *skip)
481{
482 struct ceph_mon_client *monc = con->private;
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700483 struct ceph_mon_generic_request *req;
Sage Weil3143edd2010-03-24 21:43:33 -0700484 u64 tid = le64_to_cpu(hdr->tid);
485 struct ceph_msg *m;
486
487 mutex_lock(&monc->mutex);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700488 req = __lookup_generic_req(monc, tid);
Sage Weil3143edd2010-03-24 21:43:33 -0700489 if (!req) {
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700490 dout("get_generic_reply %lld dne\n", tid);
Sage Weil3143edd2010-03-24 21:43:33 -0700491 *skip = 1;
492 m = NULL;
493 } else {
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700494 dout("get_generic_reply %lld got %p\n", tid, req->reply);
Alex Elder1c20f2d2012-06-04 14:43:32 -0500495 *skip = 0;
Sage Weil3143edd2010-03-24 21:43:33 -0700496 m = ceph_msg_get(req->reply);
497 /*
498 * we don't need to track the connection reading into
499 * this reply because we only have one open connection
500 * at a time, ever.
501 */
502 }
503 mutex_unlock(&monc->mutex);
504 return m;
505}
506
Ilya Dryomov513a8242014-05-13 11:19:26 +0400507static int __do_generic_request(struct ceph_mon_client *monc, u64 tid,
508 struct ceph_mon_generic_request *req)
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700509{
510 int err;
511
512 /* register request */
Ilya Dryomov513a8242014-05-13 11:19:26 +0400513 req->tid = tid != 0 ? tid : ++monc->last_tid;
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700514 req->request->hdr.tid = cpu_to_le64(req->tid);
515 __insert_generic_request(monc, req);
516 monc->num_generic_requests++;
Alex Elder67130932012-05-26 23:26:43 -0500517 ceph_con_send(&monc->con, ceph_msg_get(req->request));
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700518 mutex_unlock(&monc->mutex);
519
520 err = wait_for_completion_interruptible(&req->completion);
521
522 mutex_lock(&monc->mutex);
523 rb_erase(&req->node, &monc->generic_request_tree);
524 monc->num_generic_requests--;
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700525
526 if (!err)
527 err = req->result;
528 return err;
529}
530
Ilya Dryomov513a8242014-05-13 11:19:26 +0400531static int do_generic_request(struct ceph_mon_client *monc,
532 struct ceph_mon_generic_request *req)
533{
534 int err;
535
536 mutex_lock(&monc->mutex);
537 err = __do_generic_request(monc, 0, req);
538 mutex_unlock(&monc->mutex);
539
540 return err;
541}
542
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700543/*
544 * statfs
545 */
Sage Weilba75bb92009-10-06 11:31:11 -0700546static void handle_statfs_reply(struct ceph_mon_client *monc,
547 struct ceph_msg *msg)
548{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700549 struct ceph_mon_generic_request *req;
Sage Weilba75bb92009-10-06 11:31:11 -0700550 struct ceph_mon_statfs_reply *reply = msg->front.iov_base;
Sage Weil3143edd2010-03-24 21:43:33 -0700551 u64 tid = le64_to_cpu(msg->hdr.tid);
Sage Weilba75bb92009-10-06 11:31:11 -0700552
553 if (msg->front.iov_len != sizeof(*reply))
554 goto bad;
Sage Weilba75bb92009-10-06 11:31:11 -0700555 dout("handle_statfs_reply %p tid %llu\n", msg, tid);
556
557 mutex_lock(&monc->mutex);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700558 req = __lookup_generic_req(monc, tid);
Sage Weilba75bb92009-10-06 11:31:11 -0700559 if (req) {
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700560 *(struct ceph_statfs *)req->buf = reply->st;
Sage Weilba75bb92009-10-06 11:31:11 -0700561 req->result = 0;
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700562 get_generic_request(req);
Sage Weilba75bb92009-10-06 11:31:11 -0700563 }
564 mutex_unlock(&monc->mutex);
Sage Weil3143edd2010-03-24 21:43:33 -0700565 if (req) {
Yehuda Sadeh03066f22010-07-27 13:11:08 -0700566 complete_all(&req->completion);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700567 put_generic_request(req);
Sage Weil3143edd2010-03-24 21:43:33 -0700568 }
Sage Weilba75bb92009-10-06 11:31:11 -0700569 return;
570
571bad:
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700572 pr_err("corrupt generic reply, tid %llu\n", tid);
Sage Weil9ec7cab2009-12-14 15:13:47 -0800573 ceph_msg_dump(msg);
Sage Weilba75bb92009-10-06 11:31:11 -0700574}
575
576/*
Sage Weilba75bb92009-10-06 11:31:11 -0700577 * Do a synchronous statfs().
578 */
579int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf)
580{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700581 struct ceph_mon_generic_request *req;
Sage Weil3143edd2010-03-24 21:43:33 -0700582 struct ceph_mon_statfs *h;
Sage Weilba75bb92009-10-06 11:31:11 -0700583 int err;
584
Julia Lawallcffe7b62010-05-13 22:07:29 +0200585 req = kzalloc(sizeof(*req), GFP_NOFS);
Sage Weil3143edd2010-03-24 21:43:33 -0700586 if (!req)
587 return -ENOMEM;
Sage Weilba75bb92009-10-06 11:31:11 -0700588
Sage Weil3143edd2010-03-24 21:43:33 -0700589 kref_init(&req->kref);
590 req->buf = buf;
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700591 req->buf_len = sizeof(*buf);
Sage Weil3143edd2010-03-24 21:43:33 -0700592 init_completion(&req->completion);
593
Sage Weila79832f2010-04-01 16:06:19 -0700594 err = -ENOMEM;
Sage Weilb61c2762011-08-09 15:03:46 -0700595 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS,
596 true);
Sage Weila79832f2010-04-01 16:06:19 -0700597 if (!req->request)
Sage Weil3143edd2010-03-24 21:43:33 -0700598 goto out;
Sage Weilb61c2762011-08-09 15:03:46 -0700599 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS,
600 true);
Sage Weila79832f2010-04-01 16:06:19 -0700601 if (!req->reply)
Sage Weil3143edd2010-03-24 21:43:33 -0700602 goto out;
Sage Weil3143edd2010-03-24 21:43:33 -0700603
604 /* fill out request */
605 h = req->request->front.iov_base;
606 h->monhdr.have_version = 0;
607 h->monhdr.session_mon = cpu_to_le16(-1);
608 h->monhdr.session_mon_tid = 0;
609 h->fsid = monc->monmap->fsid;
Sage Weilba75bb92009-10-06 11:31:11 -0700610
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700611 err = do_generic_request(monc, req);
Sage Weil3143edd2010-03-24 21:43:33 -0700612
613out:
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700614 kref_put(&req->kref, release_generic_request);
Sage Weilba75bb92009-10-06 11:31:11 -0700615 return err;
616}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700617EXPORT_SYMBOL(ceph_monc_do_statfs);
Sage Weilba75bb92009-10-06 11:31:11 -0700618
Ilya Dryomov513a8242014-05-13 11:19:26 +0400619static void handle_get_version_reply(struct ceph_mon_client *monc,
620 struct ceph_msg *msg)
621{
622 struct ceph_mon_generic_request *req;
623 u64 tid = le64_to_cpu(msg->hdr.tid);
624 void *p = msg->front.iov_base;
625 void *end = p + msg->front_alloc_len;
626 u64 handle;
627
628 dout("%s %p tid %llu\n", __func__, msg, tid);
629
630 ceph_decode_need(&p, end, 2*sizeof(u64), bad);
631 handle = ceph_decode_64(&p);
632 if (tid != 0 && tid != handle)
633 goto bad;
634
635 mutex_lock(&monc->mutex);
636 req = __lookup_generic_req(monc, handle);
637 if (req) {
638 *(u64 *)req->buf = ceph_decode_64(&p);
639 req->result = 0;
640 get_generic_request(req);
641 }
642 mutex_unlock(&monc->mutex);
643 if (req) {
644 complete_all(&req->completion);
645 put_generic_request(req);
646 }
647
648 return;
649bad:
650 pr_err("corrupt mon_get_version reply\n");
651 ceph_msg_dump(msg);
652}
653
654/*
655 * Send MMonGetVersion and wait for the reply.
656 *
657 * @what: one of "mdsmap", "osdmap" or "monmap"
658 */
659int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what,
660 u64 *newest)
661{
662 struct ceph_mon_generic_request *req;
663 void *p, *end;
664 u64 tid;
665 int err;
666
667 req = kzalloc(sizeof(*req), GFP_NOFS);
668 if (!req)
669 return -ENOMEM;
670
671 kref_init(&req->kref);
672 req->buf = newest;
673 req->buf_len = sizeof(*newest);
674 init_completion(&req->completion);
675
676 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION,
677 sizeof(u64) + sizeof(u32) + strlen(what),
678 GFP_NOFS, true);
679 if (!req->request) {
680 err = -ENOMEM;
681 goto out;
682 }
683
684 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024,
685 GFP_NOFS, true);
686 if (!req->reply) {
687 err = -ENOMEM;
688 goto out;
689 }
690
691 p = req->request->front.iov_base;
692 end = p + req->request->front_alloc_len;
693
694 /* fill out request */
695 mutex_lock(&monc->mutex);
696 tid = ++monc->last_tid;
697 ceph_encode_64(&p, tid); /* handle */
698 ceph_encode_string(&p, end, what, strlen(what));
699
700 err = __do_generic_request(monc, tid, req);
701
702 mutex_unlock(&monc->mutex);
703out:
704 kref_put(&req->kref, release_generic_request);
705 return err;
706}
707EXPORT_SYMBOL(ceph_monc_do_get_version);
708
Sage Weilba75bb92009-10-06 11:31:11 -0700709/*
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700710 * pool ops
711 */
712static int get_poolop_reply_buf(const char *src, size_t src_len,
713 char *dst, size_t dst_len)
714{
715 u32 buf_len;
716
717 if (src_len != sizeof(u32) + dst_len)
718 return -EINVAL;
719
720 buf_len = le32_to_cpu(*(u32 *)src);
721 if (buf_len != dst_len)
722 return -EINVAL;
723
724 memcpy(dst, src + sizeof(u32), dst_len);
725 return 0;
726}
727
728static void handle_poolop_reply(struct ceph_mon_client *monc,
729 struct ceph_msg *msg)
730{
731 struct ceph_mon_generic_request *req;
732 struct ceph_mon_poolop_reply *reply = msg->front.iov_base;
733 u64 tid = le64_to_cpu(msg->hdr.tid);
734
735 if (msg->front.iov_len < sizeof(*reply))
736 goto bad;
737 dout("handle_poolop_reply %p tid %llu\n", msg, tid);
738
739 mutex_lock(&monc->mutex);
740 req = __lookup_generic_req(monc, tid);
741 if (req) {
742 if (req->buf_len &&
743 get_poolop_reply_buf(msg->front.iov_base + sizeof(*reply),
744 msg->front.iov_len - sizeof(*reply),
745 req->buf, req->buf_len) < 0) {
746 mutex_unlock(&monc->mutex);
747 goto bad;
748 }
749 req->result = le32_to_cpu(reply->reply_code);
750 get_generic_request(req);
751 }
752 mutex_unlock(&monc->mutex);
753 if (req) {
754 complete(&req->completion);
755 put_generic_request(req);
756 }
757 return;
758
759bad:
760 pr_err("corrupt generic reply, tid %llu\n", tid);
761 ceph_msg_dump(msg);
762}
763
764/*
765 * Do a synchronous pool op.
766 */
Iulius Curt7698f2f2012-08-23 15:14:29 +0300767static int do_poolop(struct ceph_mon_client *monc, u32 op,
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700768 u32 pool, u64 snapid,
769 char *buf, int len)
770{
771 struct ceph_mon_generic_request *req;
772 struct ceph_mon_poolop *h;
773 int err;
774
775 req = kzalloc(sizeof(*req), GFP_NOFS);
776 if (!req)
777 return -ENOMEM;
778
779 kref_init(&req->kref);
780 req->buf = buf;
781 req->buf_len = len;
782 init_completion(&req->completion);
783
784 err = -ENOMEM;
Sage Weilb61c2762011-08-09 15:03:46 -0700785 req->request = ceph_msg_new(CEPH_MSG_POOLOP, sizeof(*h), GFP_NOFS,
786 true);
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700787 if (!req->request)
788 goto out;
Sage Weilb61c2762011-08-09 15:03:46 -0700789 req->reply = ceph_msg_new(CEPH_MSG_POOLOP_REPLY, 1024, GFP_NOFS,
790 true);
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700791 if (!req->reply)
792 goto out;
793
794 /* fill out request */
795 req->request->hdr.version = cpu_to_le16(2);
796 h = req->request->front.iov_base;
797 h->monhdr.have_version = 0;
798 h->monhdr.session_mon = cpu_to_le16(-1);
799 h->monhdr.session_mon_tid = 0;
800 h->fsid = monc->monmap->fsid;
801 h->pool = cpu_to_le32(pool);
802 h->op = cpu_to_le32(op);
803 h->auid = 0;
804 h->snapid = cpu_to_le64(snapid);
805 h->name_len = 0;
806
807 err = do_generic_request(monc, req);
808
809out:
810 kref_put(&req->kref, release_generic_request);
811 return err;
812}
813
814int ceph_monc_create_snapid(struct ceph_mon_client *monc,
815 u32 pool, u64 *snapid)
816{
Iulius Curt7698f2f2012-08-23 15:14:29 +0300817 return do_poolop(monc, POOL_OP_CREATE_UNMANAGED_SNAP,
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700818 pool, 0, (char *)snapid, sizeof(*snapid));
819
820}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700821EXPORT_SYMBOL(ceph_monc_create_snapid);
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700822
823int ceph_monc_delete_snapid(struct ceph_mon_client *monc,
824 u32 pool, u64 snapid)
825{
Iulius Curt7698f2f2012-08-23 15:14:29 +0300826 return do_poolop(monc, POOL_OP_CREATE_UNMANAGED_SNAP,
Alex Elder15417162013-02-19 12:25:56 -0600827 pool, snapid, NULL, 0);
Yehuda Sadehe56fa102010-05-17 12:40:28 -0700828
829}
830
831/*
832 * Resend pending generic requests.
Sage Weilba75bb92009-10-06 11:31:11 -0700833 */
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700834static void __resend_generic_request(struct ceph_mon_client *monc)
Sage Weilba75bb92009-10-06 11:31:11 -0700835{
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700836 struct ceph_mon_generic_request *req;
Sage Weil85ff03f2010-02-15 14:47:28 -0800837 struct rb_node *p;
Sage Weilba75bb92009-10-06 11:31:11 -0700838
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700839 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) {
840 req = rb_entry(p, struct ceph_mon_generic_request, node);
Alex Elder6740a842012-06-01 14:56:43 -0500841 ceph_msg_revoke(req->request);
Sage Weil4f471e42012-07-30 18:16:40 -0700842 ceph_msg_revoke_incoming(req->reply);
Alex Elder67130932012-05-26 23:26:43 -0500843 ceph_con_send(&monc->con, ceph_msg_get(req->request));
Sage Weilba75bb92009-10-06 11:31:11 -0700844 }
845}
846
847/*
848 * Delayed work. If we haven't mounted yet, retry. Otherwise,
849 * renew/retry subscription as needed (in case it is timing out, or we
850 * got an ENOMEM). And keep the monitor connection alive.
851 */
852static void delayed_work(struct work_struct *work)
853{
854 struct ceph_mon_client *monc =
855 container_of(work, struct ceph_mon_client, delayed_work.work);
856
857 dout("monc delayed_work\n");
858 mutex_lock(&monc->mutex);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800859 if (monc->hunting) {
860 __close_session(monc);
861 __open_session(monc); /* continue hunting */
Sage Weilba75bb92009-10-06 11:31:11 -0700862 } else {
Alex Elder67130932012-05-26 23:26:43 -0500863 ceph_con_keepalive(&monc->con);
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800864
865 __validate_auth(monc);
866
Sage Weil27859f92013-03-25 10:26:14 -0700867 if (ceph_auth_is_authenticated(monc->auth))
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800868 __send_subscribe(monc);
Sage Weilba75bb92009-10-06 11:31:11 -0700869 }
Sage Weilba75bb92009-10-06 11:31:11 -0700870 __schedule_delayed(monc);
871 mutex_unlock(&monc->mutex);
872}
873
Sage Weil6b805182009-10-27 11:50:50 -0700874/*
875 * On startup, we build a temporary monmap populated with the IPs
876 * provided by mount(2).
877 */
878static int build_initial_monmap(struct ceph_mon_client *monc)
879{
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700880 struct ceph_options *opt = monc->client->options;
881 struct ceph_entity_addr *mon_addr = opt->mon_addr;
882 int num_mon = opt->num_mon;
Sage Weil6b805182009-10-27 11:50:50 -0700883 int i;
884
885 /* build initial monmap */
886 monc->monmap = kzalloc(sizeof(*monc->monmap) +
887 num_mon*sizeof(monc->monmap->mon_inst[0]),
888 GFP_KERNEL);
889 if (!monc->monmap)
890 return -ENOMEM;
891 for (i = 0; i < num_mon; i++) {
892 monc->monmap->mon_inst[i].addr = mon_addr[i];
Sage Weil6b805182009-10-27 11:50:50 -0700893 monc->monmap->mon_inst[i].addr.nonce = 0;
894 monc->monmap->mon_inst[i].name.type =
895 CEPH_ENTITY_TYPE_MON;
896 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i);
897 }
898 monc->monmap->num_mon = num_mon;
Sage Weil6b805182009-10-27 11:50:50 -0700899 return 0;
900}
901
Sage Weilba75bb92009-10-06 11:31:11 -0700902int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl)
903{
904 int err = 0;
905
906 dout("init\n");
907 memset(monc, 0, sizeof(*monc));
908 monc->client = cl;
909 monc->monmap = NULL;
910 mutex_init(&monc->mutex);
911
Sage Weil6b805182009-10-27 11:50:50 -0700912 err = build_initial_monmap(monc);
913 if (err)
914 goto out;
915
Sage Weilf6a2f5b2011-08-09 09:27:44 -0700916 /* connection */
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800917 /* authentication */
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700918 monc->auth = ceph_auth_init(cl->options->name,
Tommi Virtanen8323c3a2011-03-25 16:32:57 -0700919 cl->options->key);
Noah Watkins49d92242011-09-12 14:51:58 -0700920 if (IS_ERR(monc->auth)) {
921 err = PTR_ERR(monc->auth);
Alex Elder67130932012-05-26 23:26:43 -0500922 goto out_monmap;
Noah Watkins49d92242011-09-12 14:51:58 -0700923 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800924 monc->auth->want_keys =
925 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON |
926 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS;
927
Sage Weil240ed682010-05-21 14:57:25 -0700928 /* msgs */
Sage Weila79832f2010-04-01 16:06:19 -0700929 err = -ENOMEM;
Sage Weil7c315c52010-03-24 21:52:30 -0700930 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK,
Yehuda Sadeh34d23762010-04-06 14:33:58 -0700931 sizeof(struct ceph_mon_subscribe_ack),
Sage Weilb61c2762011-08-09 15:03:46 -0700932 GFP_NOFS, true);
Sage Weila79832f2010-04-01 16:06:19 -0700933 if (!monc->m_subscribe_ack)
Noah Watkins49d92242011-09-12 14:51:58 -0700934 goto out_auth;
Sage Weil6694d6b2010-03-24 21:48:05 -0700935
Sage Weilb61c2762011-08-09 15:03:46 -0700936 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS,
937 true);
Sage Weil240ed682010-05-21 14:57:25 -0700938 if (!monc->m_subscribe)
939 goto out_subscribe_ack;
940
Sage Weilb61c2762011-08-09 15:03:46 -0700941 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS,
942 true);
Sage Weila79832f2010-04-01 16:06:19 -0700943 if (!monc->m_auth_reply)
Sage Weil240ed682010-05-21 14:57:25 -0700944 goto out_subscribe;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800945
Sage Weilb61c2762011-08-09 15:03:46 -0700946 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true);
Sage Weil9bd2e6f2010-02-02 16:21:06 -0800947 monc->pending_auth = 0;
Sage Weila79832f2010-04-01 16:06:19 -0700948 if (!monc->m_auth)
Sage Weil6694d6b2010-03-24 21:48:05 -0700949 goto out_auth_reply;
Sage Weilba75bb92009-10-06 11:31:11 -0700950
Sage Weil735a72e2012-06-27 12:24:34 -0700951 ceph_con_init(&monc->con, monc, &mon_con_ops,
952 &monc->client->msgr);
953
Sage Weilba75bb92009-10-06 11:31:11 -0700954 monc->cur_mon = -1;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800955 monc->hunting = true;
Sage Weilba75bb92009-10-06 11:31:11 -0700956 monc->sub_renew_after = jiffies;
957 monc->sub_sent = 0;
958
959 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -0700960 monc->generic_request_tree = RB_ROOT;
961 monc->num_generic_requests = 0;
Sage Weilba75bb92009-10-06 11:31:11 -0700962 monc->last_tid = 0;
963
964 monc->have_mdsmap = 0;
965 monc->have_osdmap = 0;
966 monc->want_next_osdmap = 1;
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800967 return 0;
968
Sage Weil6694d6b2010-03-24 21:48:05 -0700969out_auth_reply:
970 ceph_msg_put(monc->m_auth_reply);
Sage Weil240ed682010-05-21 14:57:25 -0700971out_subscribe:
972 ceph_msg_put(monc->m_subscribe);
Sage Weil7c315c52010-03-24 21:52:30 -0700973out_subscribe_ack:
974 ceph_msg_put(monc->m_subscribe_ack);
Noah Watkins49d92242011-09-12 14:51:58 -0700975out_auth:
976 ceph_auth_destroy(monc->auth);
Sage Weil4e7a5dc2009-11-18 16:19:57 -0800977out_monmap:
978 kfree(monc->monmap);
Sage Weilba75bb92009-10-06 11:31:11 -0700979out:
980 return err;
981}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -0700982EXPORT_SYMBOL(ceph_monc_init);
Sage Weilba75bb92009-10-06 11:31:11 -0700983
984void ceph_monc_stop(struct ceph_mon_client *monc)
985{
986 dout("stop\n");
987 cancel_delayed_work_sync(&monc->delayed_work);
988
989 mutex_lock(&monc->mutex);
990 __close_session(monc);
Sage Weilf6a2f5b2011-08-09 09:27:44 -0700991
Sage Weilba75bb92009-10-06 11:31:11 -0700992 mutex_unlock(&monc->mutex);
993
Sage Weilf3dea7e2012-06-10 20:43:56 -0700994 /*
995 * flush msgr queue before we destroy ourselves to ensure that:
996 * - any work that references our embedded con is finished.
997 * - any osd_client or other work that may reference an authorizer
998 * finishes before we shut down the auth subsystem.
999 */
1000 ceph_msgr_flush();
1001
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001002 ceph_auth_destroy(monc->auth);
1003
1004 ceph_msg_put(monc->m_auth);
Sage Weil6694d6b2010-03-24 21:48:05 -07001005 ceph_msg_put(monc->m_auth_reply);
Sage Weil240ed682010-05-21 14:57:25 -07001006 ceph_msg_put(monc->m_subscribe);
Sage Weil7c315c52010-03-24 21:52:30 -07001007 ceph_msg_put(monc->m_subscribe_ack);
Sage Weilba75bb92009-10-06 11:31:11 -07001008
1009 kfree(monc->monmap);
1010}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001011EXPORT_SYMBOL(ceph_monc_stop);
Sage Weilba75bb92009-10-06 11:31:11 -07001012
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001013static void handle_auth_reply(struct ceph_mon_client *monc,
1014 struct ceph_msg *msg)
1015{
1016 int ret;
Sage Weil09c4d6a2010-05-25 15:38:06 -07001017 int was_auth = 0;
Sage Weild1c338a2012-08-19 12:29:16 -07001018 int had_debugfs_info, init_debugfs = 0;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001019
1020 mutex_lock(&monc->mutex);
Sage Weild1c338a2012-08-19 12:29:16 -07001021 had_debugfs_info = have_debugfs_info(monc);
Sage Weil27859f92013-03-25 10:26:14 -07001022 was_auth = ceph_auth_is_authenticated(monc->auth);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08001023 monc->pending_auth = 0;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001024 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base,
1025 msg->front.iov_len,
1026 monc->m_auth->front.iov_base,
Ilya Dryomov3cea4c32014-01-09 20:08:21 +02001027 monc->m_auth->front_alloc_len);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001028 if (ret < 0) {
Sage Weil9bd2e6f2010-02-02 16:21:06 -08001029 monc->client->auth_err = ret;
Yehuda Sadeh03066f22010-07-27 13:11:08 -07001030 wake_up_all(&monc->client->auth_wq);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001031 } else if (ret > 0) {
Sage Weil9bd2e6f2010-02-02 16:21:06 -08001032 __send_prepared_auth_request(monc, ret);
Sage Weil27859f92013-03-25 10:26:14 -07001033 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) {
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001034 dout("authenticated, starting session\n");
Sage Weil07433042009-11-18 16:50:41 -08001035
Alex Elder15d98822012-05-26 23:26:43 -05001036 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT;
1037 monc->client->msgr.inst.name.num =
Yehuda Sadeh0cf55372010-06-11 15:57:06 -07001038 cpu_to_le64(monc->auth->global_id);
Sage Weil07433042009-11-18 16:50:41 -08001039
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001040 __send_subscribe(monc);
Yehuda Sadehf8c76f62010-04-22 15:40:37 -07001041 __resend_generic_request(monc);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001042 }
Sage Weild1c338a2012-08-19 12:29:16 -07001043
1044 if (!had_debugfs_info && have_debugfs_info(monc)) {
1045 pr_info("client%lld fsid %pU\n",
1046 ceph_client_id(monc->client),
1047 &monc->client->fsid);
1048 init_debugfs = 1;
1049 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001050 mutex_unlock(&monc->mutex);
Sage Weild1c338a2012-08-19 12:29:16 -07001051
1052 if (init_debugfs) {
1053 /*
1054 * do debugfs initialization without mutex to avoid
1055 * creating a locking dependency
1056 */
1057 ceph_debugfs_client_init(monc->client);
1058 }
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001059}
1060
Sage Weil9bd2e6f2010-02-02 16:21:06 -08001061static int __validate_auth(struct ceph_mon_client *monc)
1062{
1063 int ret;
1064
1065 if (monc->pending_auth)
1066 return 0;
1067
1068 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base,
Ilya Dryomov3cea4c32014-01-09 20:08:21 +02001069 monc->m_auth->front_alloc_len);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08001070 if (ret <= 0)
1071 return ret; /* either an error, or no need to authenticate */
1072 __send_prepared_auth_request(monc, ret);
1073 return 0;
1074}
1075
1076int ceph_monc_validate_auth(struct ceph_mon_client *monc)
1077{
1078 int ret;
1079
1080 mutex_lock(&monc->mutex);
1081 ret = __validate_auth(monc);
1082 mutex_unlock(&monc->mutex);
1083 return ret;
1084}
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001085EXPORT_SYMBOL(ceph_monc_validate_auth);
Sage Weil9bd2e6f2010-02-02 16:21:06 -08001086
Sage Weilba75bb92009-10-06 11:31:11 -07001087/*
1088 * handle incoming message
1089 */
1090static void dispatch(struct ceph_connection *con, struct ceph_msg *msg)
1091{
1092 struct ceph_mon_client *monc = con->private;
1093 int type = le16_to_cpu(msg->hdr.type);
1094
1095 if (!monc)
1096 return;
1097
1098 switch (type) {
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001099 case CEPH_MSG_AUTH_REPLY:
1100 handle_auth_reply(monc, msg);
Sage Weilba75bb92009-10-06 11:31:11 -07001101 break;
1102
1103 case CEPH_MSG_MON_SUBSCRIBE_ACK:
1104 handle_subscribe_ack(monc, msg);
1105 break;
1106
1107 case CEPH_MSG_STATFS_REPLY:
1108 handle_statfs_reply(monc, msg);
1109 break;
1110
Ilya Dryomov513a8242014-05-13 11:19:26 +04001111 case CEPH_MSG_MON_GET_VERSION_REPLY:
1112 handle_get_version_reply(monc, msg);
1113 break;
1114
Yehuda Sadehe56fa102010-05-17 12:40:28 -07001115 case CEPH_MSG_POOLOP_REPLY:
1116 handle_poolop_reply(monc, msg);
1117 break;
1118
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001119 case CEPH_MSG_MON_MAP:
1120 ceph_monc_handle_map(monc, msg);
1121 break;
1122
Sage Weilba75bb92009-10-06 11:31:11 -07001123 case CEPH_MSG_OSD_MAP:
1124 ceph_osdc_handle_map(&monc->client->osdc, msg);
1125 break;
1126
1127 default:
Yehuda Sadeh3d14c5d2010-04-06 15:14:15 -07001128 /* can the chained handler handle it? */
1129 if (monc->client->extra_mon_dispatch &&
1130 monc->client->extra_mon_dispatch(monc->client, msg) == 0)
1131 break;
1132
Sage Weilba75bb92009-10-06 11:31:11 -07001133 pr_err("received unknown message type %d %s\n", type,
1134 ceph_msg_type_name(type));
1135 }
1136 ceph_msg_put(msg);
1137}
1138
1139/*
1140 * Allocate memory for incoming message
1141 */
1142static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con,
Yehuda Sadeh24504182010-01-08 13:58:34 -08001143 struct ceph_msg_header *hdr,
1144 int *skip)
Sage Weilba75bb92009-10-06 11:31:11 -07001145{
1146 struct ceph_mon_client *monc = con->private;
1147 int type = le16_to_cpu(hdr->type);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001148 int front_len = le32_to_cpu(hdr->front_len);
Sage Weil5b3a4db2010-02-19 21:43:23 -08001149 struct ceph_msg *m = NULL;
Sage Weilba75bb92009-10-06 11:31:11 -07001150
Yehuda Sadeh24504182010-01-08 13:58:34 -08001151 *skip = 0;
Yehuda Sadeh0547a9b2010-01-11 14:47:13 -08001152
Sage Weilba75bb92009-10-06 11:31:11 -07001153 switch (type) {
Sage Weilba75bb92009-10-06 11:31:11 -07001154 case CEPH_MSG_MON_SUBSCRIBE_ACK:
Sage Weil7c315c52010-03-24 21:52:30 -07001155 m = ceph_msg_get(monc->m_subscribe_ack);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001156 break;
Yehuda Sadehe56fa102010-05-17 12:40:28 -07001157 case CEPH_MSG_POOLOP_REPLY:
Sage Weilba75bb92009-10-06 11:31:11 -07001158 case CEPH_MSG_STATFS_REPLY:
Yehuda Sadehf8c76f62010-04-22 15:40:37 -07001159 return get_generic_reply(con, hdr, skip);
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001160 case CEPH_MSG_AUTH_REPLY:
Sage Weil6694d6b2010-03-24 21:48:05 -07001161 m = ceph_msg_get(monc->m_auth_reply);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001162 break;
Ilya Dryomov513a8242014-05-13 11:19:26 +04001163 case CEPH_MSG_MON_GET_VERSION_REPLY:
1164 if (le64_to_cpu(hdr->tid) != 0)
1165 return get_generic_reply(con, hdr, skip);
1166
1167 /*
1168 * Older OSDs don't set reply tid even if the orignal
1169 * request had a non-zero tid. Workaround this weirdness
1170 * by falling through to the allocate case.
1171 */
Sage Weil5b3a4db2010-02-19 21:43:23 -08001172 case CEPH_MSG_MON_MAP:
1173 case CEPH_MSG_MDS_MAP:
1174 case CEPH_MSG_OSD_MAP:
Sage Weilb61c2762011-08-09 15:03:46 -07001175 m = ceph_msg_new(type, front_len, GFP_NOFS, false);
Alex Elder1c20f2d2012-06-04 14:43:32 -05001176 if (!m)
1177 return NULL; /* ENOMEM--return skip == 0 */
Sage Weil5b3a4db2010-02-19 21:43:23 -08001178 break;
Sage Weilba75bb92009-10-06 11:31:11 -07001179 }
Yehuda Sadeh24504182010-01-08 13:58:34 -08001180
Sage Weil5b3a4db2010-02-19 21:43:23 -08001181 if (!m) {
1182 pr_info("alloc_msg unknown type %d\n", type);
Yehuda Sadeh24504182010-01-08 13:58:34 -08001183 *skip = 1;
Sage Weil5b3a4db2010-02-19 21:43:23 -08001184 }
Yehuda Sadeh24504182010-01-08 13:58:34 -08001185 return m;
Sage Weilba75bb92009-10-06 11:31:11 -07001186}
1187
1188/*
1189 * If the monitor connection resets, pick a new monitor and resubmit
1190 * any pending requests.
1191 */
1192static void mon_fault(struct ceph_connection *con)
1193{
1194 struct ceph_mon_client *monc = con->private;
1195
1196 if (!monc)
1197 return;
1198
1199 dout("mon_fault\n");
1200 mutex_lock(&monc->mutex);
1201 if (!con->private)
1202 goto out;
1203
Sage Weilf6a2f5b2011-08-09 09:27:44 -07001204 if (!monc->hunting)
Sage Weilba75bb92009-10-06 11:31:11 -07001205 pr_info("mon%d %s session lost, "
1206 "hunting for new mon\n", monc->cur_mon,
Alex Elder67130932012-05-26 23:26:43 -05001207 ceph_pr_addr(&monc->con.peer_addr.in_addr));
Sage Weilba75bb92009-10-06 11:31:11 -07001208
1209 __close_session(monc);
1210 if (!monc->hunting) {
1211 /* start hunting */
1212 monc->hunting = true;
Sage Weil4e7a5dc2009-11-18 16:19:57 -08001213 __open_session(monc);
Sage Weilba75bb92009-10-06 11:31:11 -07001214 } else {
1215 /* already hunting, let's wait a bit */
1216 __schedule_delayed(monc);
1217 }
1218out:
1219 mutex_unlock(&monc->mutex);
1220}
1221
Sage Weilec87ef42012-05-31 20:27:50 -07001222/*
1223 * We can ignore refcounting on the connection struct, as all references
1224 * will come from the messenger workqueue, which is drained prior to
1225 * mon_client destruction.
1226 */
1227static struct ceph_connection *con_get(struct ceph_connection *con)
1228{
1229 return con;
1230}
1231
1232static void con_put(struct ceph_connection *con)
1233{
1234}
1235
Tobias Klauser9e327892010-05-20 10:40:19 +02001236static const struct ceph_connection_operations mon_con_ops = {
Sage Weilec87ef42012-05-31 20:27:50 -07001237 .get = con_get,
1238 .put = con_put,
Sage Weilba75bb92009-10-06 11:31:11 -07001239 .dispatch = dispatch,
1240 .fault = mon_fault,
1241 .alloc_msg = mon_alloc_msg,
Sage Weilba75bb92009-10-06 11:31:11 -07001242};