Commit | Line | Data |
---|---|---|
3d14c5d2 | 1 | #include <linux/ceph/ceph_debug.h> |
ba75bb98 | 2 | |
3d14c5d2 | 3 | #include <linux/module.h> |
ba75bb98 | 4 | #include <linux/types.h> |
5a0e3ad6 | 5 | #include <linux/slab.h> |
ba75bb98 SW |
6 | #include <linux/random.h> |
7 | #include <linux/sched.h> | |
8 | ||
3d14c5d2 YS |
9 | #include <linux/ceph/mon_client.h> |
10 | #include <linux/ceph/libceph.h> | |
ab434b60 | 11 | #include <linux/ceph/debugfs.h> |
3d14c5d2 | 12 | #include <linux/ceph/decode.h> |
3d14c5d2 | 13 | #include <linux/ceph/auth.h> |
ba75bb98 SW |
14 | |
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 | ||
9e32789f | 33 | static const struct ceph_connection_operations mon_con_ops; |
ba75bb98 | 34 | |
9bd2e6f8 SW |
35 | static int __validate_auth(struct ceph_mon_client *monc); |
36 | ||
ba75bb98 SW |
37 | /* |
38 | * Decode a monmap blob (e.g., during mount). | |
39 | */ | |
40 | struct 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; | |
4e7a5dcd SW |
47 | u32 len; |
48 | ||
49 | ceph_decode_32_safe(&p, end, len, bad); | |
50 | ceph_decode_need(&p, end, len, bad); | |
ba75bb98 SW |
51 | |
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)); | |
c89136ea | 58 | epoch = ceph_decode_32(&p); |
ba75bb98 | 59 | |
c89136ea | 60 | num_mon = ceph_decode_32(&p); |
ba75bb98 SW |
61 | ceph_decode_need(&p, end, num_mon*sizeof(m->mon_inst[0]), bad); |
62 | ||
63 | if (num_mon >= CEPH_MAX_MON) | |
64 | goto bad; | |
65 | m = kmalloc(sizeof(*m) + sizeof(m->mon_inst[0])*num_mon, GFP_NOFS); | |
66 | if (m == NULL) | |
67 | return ERR_PTR(-ENOMEM); | |
68 | m->fsid = fsid; | |
69 | m->epoch = epoch; | |
70 | m->num_mon = num_mon; | |
71 | ceph_decode_copy(&p, m->mon_inst, num_mon*sizeof(m->mon_inst[0])); | |
63f2d211 SW |
72 | for (i = 0; i < num_mon; i++) |
73 | ceph_decode_addr(&m->mon_inst[i].addr); | |
ba75bb98 | 74 | |
ba75bb98 SW |
75 | dout("monmap_decode epoch %d, num_mon %d\n", m->epoch, |
76 | m->num_mon); | |
77 | for (i = 0; i < m->num_mon; i++) | |
78 | dout("monmap_decode mon%d is %s\n", i, | |
3d14c5d2 | 79 | ceph_pr_addr(&m->mon_inst[i].addr.in_addr)); |
ba75bb98 SW |
80 | return m; |
81 | ||
82 | bad: | |
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 | */ | |
91 | int 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++) | |
103e2d3a | 96 | if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0) |
ba75bb98 SW |
97 | return 1; |
98 | return 0; | |
99 | } | |
100 | ||
5ce6e9db SW |
101 | /* |
102 | * Send an auth request. | |
103 | */ | |
104 | static 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); | |
6740a845 | 109 | ceph_msg_revoke(monc->m_auth); |
5ce6e9db | 110 | ceph_msg_get(monc->m_auth); /* keep our ref */ |
67130934 | 111 | ceph_con_send(&monc->con, monc->m_auth); |
5ce6e9db SW |
112 | } |
113 | ||
ba75bb98 SW |
114 | /* |
115 | * Close monitor session, if any. | |
116 | */ | |
117 | static void __close_session(struct ceph_mon_client *monc) | |
118 | { | |
f6a2f5be | 119 | dout("__close_session closing mon%d\n", monc->cur_mon); |
6740a845 | 120 | ceph_msg_revoke(monc->m_auth); |
4f471e4a SW |
121 | ceph_msg_revoke_incoming(monc->m_auth_reply); |
122 | ceph_msg_revoke(monc->m_subscribe); | |
123 | ceph_msg_revoke_incoming(monc->m_subscribe_ack); | |
67130934 | 124 | ceph_con_close(&monc->con); |
0e04dc26 | 125 | |
f6a2f5be SW |
126 | monc->pending_auth = 0; |
127 | ceph_auth_reset(monc->auth); | |
ba75bb98 SW |
128 | } |
129 | ||
130 | /* | |
0e04dc26 ID |
131 | * Pick a new monitor at random and set cur_mon. If we are repicking |
132 | * (i.e. cur_mon is already set), be sure to pick a different one. | |
ba75bb98 | 133 | */ |
0e04dc26 | 134 | static void pick_new_mon(struct ceph_mon_client *monc) |
ba75bb98 | 135 | { |
0e04dc26 | 136 | int old_mon = monc->cur_mon; |
ba75bb98 | 137 | |
0e04dc26 | 138 | BUG_ON(monc->monmap->num_mon < 1); |
ba75bb98 | 139 | |
0e04dc26 ID |
140 | if (monc->monmap->num_mon == 1) { |
141 | monc->cur_mon = 0; | |
142 | } else { | |
143 | int max = monc->monmap->num_mon; | |
144 | int o = -1; | |
145 | int n; | |
146 | ||
147 | if (monc->cur_mon >= 0) { | |
148 | if (monc->cur_mon < monc->monmap->num_mon) | |
149 | o = monc->cur_mon; | |
150 | if (o >= 0) | |
151 | max--; | |
152 | } | |
4e7a5dcd | 153 | |
0e04dc26 ID |
154 | n = prandom_u32() % max; |
155 | if (o >= 0 && n >= o) | |
156 | n++; | |
8b9558aa | 157 | |
0e04dc26 | 158 | monc->cur_mon = n; |
ba75bb98 | 159 | } |
0e04dc26 ID |
160 | |
161 | dout("%s mon%d -> mon%d out of %d mons\n", __func__, old_mon, | |
162 | monc->cur_mon, monc->monmap->num_mon); | |
163 | } | |
164 | ||
165 | /* | |
166 | * Open a session with a new monitor. | |
167 | */ | |
168 | static void __open_session(struct ceph_mon_client *monc) | |
169 | { | |
170 | int ret; | |
171 | ||
172 | pick_new_mon(monc); | |
173 | ||
1752b50c | 174 | monc->hunting = true; |
168b9090 ID |
175 | if (monc->had_a_connection) { |
176 | monc->hunt_mult *= CEPH_MONC_HUNT_BACKOFF; | |
177 | if (monc->hunt_mult > CEPH_MONC_HUNT_MAX_MULT) | |
178 | monc->hunt_mult = CEPH_MONC_HUNT_MAX_MULT; | |
179 | } | |
180 | ||
0e04dc26 ID |
181 | monc->sub_renew_after = jiffies; /* i.e., expired */ |
182 | monc->sub_renew_sent = 0; | |
183 | ||
184 | dout("%s opening mon%d\n", __func__, monc->cur_mon); | |
185 | ceph_con_open(&monc->con, CEPH_ENTITY_TYPE_MON, monc->cur_mon, | |
186 | &monc->monmap->mon_inst[monc->cur_mon].addr); | |
187 | ||
188 | /* | |
189 | * send an initial keepalive to ensure our timestamp is valid | |
190 | * by the time we are in an OPENED state | |
191 | */ | |
192 | ceph_con_keepalive(&monc->con); | |
193 | ||
194 | /* initiate authentication handshake */ | |
195 | ret = ceph_auth_build_hello(monc->auth, | |
196 | monc->m_auth->front.iov_base, | |
197 | monc->m_auth->front_alloc_len); | |
198 | BUG_ON(ret <= 0); | |
199 | __send_prepared_auth_request(monc, ret); | |
ba75bb98 SW |
200 | } |
201 | ||
1752b50c ID |
202 | static void reopen_session(struct ceph_mon_client *monc) |
203 | { | |
204 | if (!monc->hunting) | |
205 | pr_info("mon%d %s session lost, hunting for new mon\n", | |
206 | monc->cur_mon, ceph_pr_addr(&monc->con.peer_addr.in_addr)); | |
207 | ||
208 | __close_session(monc); | |
209 | __open_session(monc); | |
210 | } | |
211 | ||
ba75bb98 SW |
212 | /* |
213 | * Reschedule delayed work timer. | |
214 | */ | |
215 | static void __schedule_delayed(struct ceph_mon_client *monc) | |
216 | { | |
8b9558aa | 217 | unsigned long delay; |
ba75bb98 | 218 | |
168b9090 ID |
219 | if (monc->hunting) |
220 | delay = CEPH_MONC_HUNT_INTERVAL * monc->hunt_mult; | |
221 | else | |
58d81b12 | 222 | delay = CEPH_MONC_PING_INTERVAL; |
168b9090 | 223 | |
8b9558aa | 224 | dout("__schedule_delayed after %lu\n", delay); |
bee3a37c ID |
225 | mod_delayed_work(system_wq, &monc->delayed_work, |
226 | round_jiffies_relative(delay)); | |
ba75bb98 SW |
227 | } |
228 | ||
82dcabad ID |
229 | const char *ceph_sub_str[] = { |
230 | [CEPH_SUB_MDSMAP] = "mdsmap", | |
231 | [CEPH_SUB_MONMAP] = "monmap", | |
232 | [CEPH_SUB_OSDMAP] = "osdmap", | |
233 | }; | |
234 | ||
ba75bb98 | 235 | /* |
82dcabad ID |
236 | * Send subscribe request for one or more maps, according to |
237 | * monc->subs. | |
ba75bb98 SW |
238 | */ |
239 | static void __send_subscribe(struct ceph_mon_client *monc) | |
240 | { | |
82dcabad ID |
241 | struct ceph_msg *msg = monc->m_subscribe; |
242 | void *p = msg->front.iov_base; | |
243 | void *const end = p + msg->front_alloc_len; | |
244 | int num = 0; | |
245 | int i; | |
246 | ||
247 | dout("%s sent %lu\n", __func__, monc->sub_renew_sent); | |
248 | ||
249 | BUG_ON(monc->cur_mon < 0); | |
250 | ||
251 | if (!monc->sub_renew_sent) | |
252 | monc->sub_renew_sent = jiffies | 1; /* never 0 */ | |
253 | ||
254 | msg->hdr.version = cpu_to_le16(2); | |
255 | ||
256 | for (i = 0; i < ARRAY_SIZE(monc->subs); i++) { | |
257 | if (monc->subs[i].want) | |
258 | num++; | |
259 | } | |
260 | BUG_ON(num < 1); /* monmap sub is always there */ | |
261 | ceph_encode_32(&p, num); | |
262 | for (i = 0; i < ARRAY_SIZE(monc->subs); i++) { | |
263 | const char *s = ceph_sub_str[i]; | |
264 | ||
265 | if (!monc->subs[i].want) | |
266 | continue; | |
267 | ||
268 | dout("%s %s start %llu flags 0x%x\n", __func__, s, | |
269 | le64_to_cpu(monc->subs[i].item.start), | |
270 | monc->subs[i].item.flags); | |
271 | ceph_encode_string(&p, end, s, strlen(s)); | |
272 | memcpy(p, &monc->subs[i].item, sizeof(monc->subs[i].item)); | |
273 | p += sizeof(monc->subs[i].item); | |
ba75bb98 | 274 | } |
82dcabad ID |
275 | |
276 | BUG_ON(p != (end - 35 - (ARRAY_SIZE(monc->subs) - num) * 19)); | |
277 | msg->front.iov_len = p - msg->front.iov_base; | |
278 | msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); | |
279 | ceph_msg_revoke(msg); | |
280 | ceph_con_send(&monc->con, ceph_msg_get(msg)); | |
ba75bb98 SW |
281 | } |
282 | ||
283 | static void handle_subscribe_ack(struct ceph_mon_client *monc, | |
284 | struct ceph_msg *msg) | |
285 | { | |
95c96174 | 286 | unsigned int seconds; |
07bd10fb SW |
287 | struct ceph_mon_subscribe_ack *h = msg->front.iov_base; |
288 | ||
289 | if (msg->front.iov_len < sizeof(*h)) | |
290 | goto bad; | |
291 | seconds = le32_to_cpu(h->duration); | |
ba75bb98 | 292 | |
ba75bb98 | 293 | mutex_lock(&monc->mutex); |
82dcabad ID |
294 | if (monc->sub_renew_sent) { |
295 | monc->sub_renew_after = monc->sub_renew_sent + | |
296 | (seconds >> 1) * HZ - 1; | |
297 | dout("%s sent %lu duration %d renew after %lu\n", __func__, | |
298 | monc->sub_renew_sent, seconds, monc->sub_renew_after); | |
299 | monc->sub_renew_sent = 0; | |
300 | } else { | |
301 | dout("%s sent %lu renew after %lu, ignoring\n", __func__, | |
302 | monc->sub_renew_sent, monc->sub_renew_after); | |
303 | } | |
ba75bb98 SW |
304 | mutex_unlock(&monc->mutex); |
305 | return; | |
306 | bad: | |
307 | pr_err("got corrupt subscribe-ack msg\n"); | |
9ec7cab1 | 308 | ceph_msg_dump(msg); |
ba75bb98 SW |
309 | } |
310 | ||
311 | /* | |
82dcabad ID |
312 | * Register interest in a map |
313 | * | |
314 | * @sub: one of CEPH_SUB_* | |
315 | * @epoch: X for "every map since X", or 0 for "just the latest" | |
ba75bb98 | 316 | */ |
82dcabad ID |
317 | static bool __ceph_monc_want_map(struct ceph_mon_client *monc, int sub, |
318 | u32 epoch, bool continuous) | |
ba75bb98 | 319 | { |
82dcabad ID |
320 | __le64 start = cpu_to_le64(epoch); |
321 | u8 flags = !continuous ? CEPH_SUBSCRIBE_ONETIME : 0; | |
322 | ||
323 | dout("%s %s epoch %u continuous %d\n", __func__, ceph_sub_str[sub], | |
324 | epoch, continuous); | |
325 | ||
326 | if (monc->subs[sub].want && | |
327 | monc->subs[sub].item.start == start && | |
328 | monc->subs[sub].item.flags == flags) | |
329 | return false; | |
330 | ||
331 | monc->subs[sub].item.start = start; | |
332 | monc->subs[sub].item.flags = flags; | |
333 | monc->subs[sub].want = true; | |
334 | ||
335 | return true; | |
336 | } | |
337 | ||
338 | bool ceph_monc_want_map(struct ceph_mon_client *monc, int sub, u32 epoch, | |
339 | bool continuous) | |
340 | { | |
341 | bool need_request; | |
342 | ||
ba75bb98 | 343 | mutex_lock(&monc->mutex); |
82dcabad | 344 | need_request = __ceph_monc_want_map(monc, sub, epoch, continuous); |
ba75bb98 | 345 | mutex_unlock(&monc->mutex); |
82dcabad ID |
346 | |
347 | return need_request; | |
348 | } | |
349 | EXPORT_SYMBOL(ceph_monc_want_map); | |
350 | ||
351 | /* | |
352 | * Keep track of which maps we have | |
353 | * | |
354 | * @sub: one of CEPH_SUB_* | |
355 | */ | |
356 | static void __ceph_monc_got_map(struct ceph_mon_client *monc, int sub, | |
357 | u32 epoch) | |
358 | { | |
359 | dout("%s %s epoch %u\n", __func__, ceph_sub_str[sub], epoch); | |
360 | ||
361 | if (monc->subs[sub].want) { | |
362 | if (monc->subs[sub].item.flags & CEPH_SUBSCRIBE_ONETIME) | |
363 | monc->subs[sub].want = false; | |
364 | else | |
365 | monc->subs[sub].item.start = cpu_to_le64(epoch + 1); | |
366 | } | |
367 | ||
368 | monc->subs[sub].have = epoch; | |
ba75bb98 SW |
369 | } |
370 | ||
82dcabad | 371 | void ceph_monc_got_map(struct ceph_mon_client *monc, int sub, u32 epoch) |
ba75bb98 SW |
372 | { |
373 | mutex_lock(&monc->mutex); | |
82dcabad | 374 | __ceph_monc_got_map(monc, sub, epoch); |
ba75bb98 | 375 | mutex_unlock(&monc->mutex); |
ba75bb98 | 376 | } |
82dcabad | 377 | EXPORT_SYMBOL(ceph_monc_got_map); |
ba75bb98 | 378 | |
42c1b124 ID |
379 | void ceph_monc_renew_subs(struct ceph_mon_client *monc) |
380 | { | |
381 | mutex_lock(&monc->mutex); | |
382 | __send_subscribe(monc); | |
383 | mutex_unlock(&monc->mutex); | |
384 | } | |
385 | EXPORT_SYMBOL(ceph_monc_renew_subs); | |
386 | ||
ba75bb98 SW |
387 | /* |
388 | * Register interest in the next osdmap | |
389 | */ | |
390 | void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc) | |
391 | { | |
82dcabad | 392 | dout("%s have %u\n", __func__, monc->subs[CEPH_SUB_OSDMAP].have); |
ba75bb98 | 393 | mutex_lock(&monc->mutex); |
82dcabad ID |
394 | if (__ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, |
395 | monc->subs[CEPH_SUB_OSDMAP].have + 1, false)) | |
ba75bb98 SW |
396 | __send_subscribe(monc); |
397 | mutex_unlock(&monc->mutex); | |
398 | } | |
6044cde6 ID |
399 | EXPORT_SYMBOL(ceph_monc_request_next_osdmap); |
400 | ||
a319bf56 ID |
401 | /* |
402 | * Wait for an osdmap with a given epoch. | |
403 | * | |
404 | * @epoch: epoch to wait for | |
405 | * @timeout: in jiffies, 0 means "wait forever" | |
406 | */ | |
6044cde6 ID |
407 | int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch, |
408 | unsigned long timeout) | |
409 | { | |
410 | unsigned long started = jiffies; | |
216639dd | 411 | long ret; |
6044cde6 ID |
412 | |
413 | mutex_lock(&monc->mutex); | |
82dcabad | 414 | while (monc->subs[CEPH_SUB_OSDMAP].have < epoch) { |
6044cde6 ID |
415 | mutex_unlock(&monc->mutex); |
416 | ||
a319bf56 | 417 | if (timeout && time_after_eq(jiffies, started + timeout)) |
6044cde6 ID |
418 | return -ETIMEDOUT; |
419 | ||
420 | ret = wait_event_interruptible_timeout(monc->client->auth_wq, | |
82dcabad ID |
421 | monc->subs[CEPH_SUB_OSDMAP].have >= epoch, |
422 | ceph_timeout_jiffies(timeout)); | |
6044cde6 ID |
423 | if (ret < 0) |
424 | return ret; | |
425 | ||
426 | mutex_lock(&monc->mutex); | |
427 | } | |
428 | ||
429 | mutex_unlock(&monc->mutex); | |
430 | return 0; | |
431 | } | |
432 | EXPORT_SYMBOL(ceph_monc_wait_osdmap); | |
ba75bb98 | 433 | |
4e7a5dcd | 434 | /* |
82dcabad ID |
435 | * Open a session with a random monitor. Request monmap and osdmap, |
436 | * which are waited upon in __ceph_open_session(). | |
4e7a5dcd SW |
437 | */ |
438 | int ceph_monc_open_session(struct ceph_mon_client *monc) | |
ba75bb98 | 439 | { |
ba75bb98 | 440 | mutex_lock(&monc->mutex); |
82dcabad ID |
441 | __ceph_monc_want_map(monc, CEPH_SUB_MONMAP, 0, true); |
442 | __ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, 0, false); | |
4e7a5dcd | 443 | __open_session(monc); |
ba75bb98 SW |
444 | __schedule_delayed(monc); |
445 | mutex_unlock(&monc->mutex); | |
446 | return 0; | |
447 | } | |
3d14c5d2 | 448 | EXPORT_SYMBOL(ceph_monc_open_session); |
ba75bb98 | 449 | |
0743304d SW |
450 | static void ceph_monc_handle_map(struct ceph_mon_client *monc, |
451 | struct ceph_msg *msg) | |
4e7a5dcd SW |
452 | { |
453 | struct ceph_client *client = monc->client; | |
454 | struct ceph_monmap *monmap = NULL, *old = monc->monmap; | |
455 | void *p, *end; | |
456 | ||
457 | mutex_lock(&monc->mutex); | |
458 | ||
459 | dout("handle_monmap\n"); | |
460 | p = msg->front.iov_base; | |
461 | end = p + msg->front.iov_len; | |
462 | ||
463 | monmap = ceph_monmap_decode(p, end); | |
464 | if (IS_ERR(monmap)) { | |
465 | pr_err("problem decoding monmap, %d\n", | |
466 | (int)PTR_ERR(monmap)); | |
d4a780ce | 467 | goto out; |
4e7a5dcd | 468 | } |
0743304d SW |
469 | |
470 | if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) { | |
4e7a5dcd | 471 | kfree(monmap); |
d4a780ce | 472 | goto out; |
4e7a5dcd SW |
473 | } |
474 | ||
475 | client->monc.monmap = monmap; | |
4e7a5dcd SW |
476 | kfree(old); |
477 | ||
82dcabad | 478 | __ceph_monc_got_map(monc, CEPH_SUB_MONMAP, monc->monmap->epoch); |
02ac956c | 479 | client->have_fsid = true; |
d1c338a5 | 480 | |
d4a780ce | 481 | out: |
4e7a5dcd | 482 | mutex_unlock(&monc->mutex); |
03066f23 | 483 | wake_up_all(&client->auth_wq); |
4e7a5dcd SW |
484 | } |
485 | ||
ba75bb98 | 486 | /* |
7a6fdeb2 | 487 | * generic requests (currently statfs, mon_get_version) |
ba75bb98 | 488 | */ |
fcd00b68 | 489 | DEFINE_RB_FUNCS(generic_request, struct ceph_mon_generic_request, tid, node) |
85ff03f6 | 490 | |
f8c76f6f | 491 | static void release_generic_request(struct kref *kref) |
3143edd3 | 492 | { |
f8c76f6f YS |
493 | struct ceph_mon_generic_request *req = |
494 | container_of(kref, struct ceph_mon_generic_request, kref); | |
3143edd3 SW |
495 | |
496 | if (req->reply) | |
497 | ceph_msg_put(req->reply); | |
498 | if (req->request) | |
499 | ceph_msg_put(req->request); | |
20547567 YS |
500 | |
501 | kfree(req); | |
3143edd3 SW |
502 | } |
503 | ||
f8c76f6f | 504 | static void put_generic_request(struct ceph_mon_generic_request *req) |
3143edd3 | 505 | { |
f8c76f6f | 506 | kref_put(&req->kref, release_generic_request); |
3143edd3 SW |
507 | } |
508 | ||
f8c76f6f | 509 | static void get_generic_request(struct ceph_mon_generic_request *req) |
3143edd3 SW |
510 | { |
511 | kref_get(&req->kref); | |
512 | } | |
513 | ||
f8c76f6f | 514 | static struct ceph_msg *get_generic_reply(struct ceph_connection *con, |
3143edd3 SW |
515 | struct ceph_msg_header *hdr, |
516 | int *skip) | |
517 | { | |
518 | struct ceph_mon_client *monc = con->private; | |
f8c76f6f | 519 | struct ceph_mon_generic_request *req; |
3143edd3 SW |
520 | u64 tid = le64_to_cpu(hdr->tid); |
521 | struct ceph_msg *m; | |
522 | ||
523 | mutex_lock(&monc->mutex); | |
fcd00b68 | 524 | req = lookup_generic_request(&monc->generic_request_tree, tid); |
3143edd3 | 525 | if (!req) { |
f8c76f6f | 526 | dout("get_generic_reply %lld dne\n", tid); |
3143edd3 SW |
527 | *skip = 1; |
528 | m = NULL; | |
529 | } else { | |
f8c76f6f | 530 | dout("get_generic_reply %lld got %p\n", tid, req->reply); |
1c20f2d2 | 531 | *skip = 0; |
3143edd3 SW |
532 | m = ceph_msg_get(req->reply); |
533 | /* | |
534 | * we don't need to track the connection reading into | |
535 | * this reply because we only have one open connection | |
536 | * at a time, ever. | |
537 | */ | |
538 | } | |
539 | mutex_unlock(&monc->mutex); | |
540 | return m; | |
541 | } | |
542 | ||
513a8243 ID |
543 | static int __do_generic_request(struct ceph_mon_client *monc, u64 tid, |
544 | struct ceph_mon_generic_request *req) | |
e56fa10e YS |
545 | { |
546 | int err; | |
547 | ||
548 | /* register request */ | |
513a8243 | 549 | req->tid = tid != 0 ? tid : ++monc->last_tid; |
e56fa10e | 550 | req->request->hdr.tid = cpu_to_le64(req->tid); |
fcd00b68 | 551 | insert_generic_request(&monc->generic_request_tree, req); |
67130934 | 552 | ceph_con_send(&monc->con, ceph_msg_get(req->request)); |
e56fa10e YS |
553 | mutex_unlock(&monc->mutex); |
554 | ||
555 | err = wait_for_completion_interruptible(&req->completion); | |
556 | ||
557 | mutex_lock(&monc->mutex); | |
fcd00b68 | 558 | erase_generic_request(&monc->generic_request_tree, req); |
e56fa10e YS |
559 | |
560 | if (!err) | |
561 | err = req->result; | |
562 | return err; | |
563 | } | |
564 | ||
513a8243 ID |
565 | static int do_generic_request(struct ceph_mon_client *monc, |
566 | struct ceph_mon_generic_request *req) | |
567 | { | |
568 | int err; | |
569 | ||
570 | mutex_lock(&monc->mutex); | |
571 | err = __do_generic_request(monc, 0, req); | |
572 | mutex_unlock(&monc->mutex); | |
573 | ||
574 | return err; | |
575 | } | |
576 | ||
e56fa10e YS |
577 | /* |
578 | * statfs | |
579 | */ | |
ba75bb98 SW |
580 | static void handle_statfs_reply(struct ceph_mon_client *monc, |
581 | struct ceph_msg *msg) | |
582 | { | |
f8c76f6f | 583 | struct ceph_mon_generic_request *req; |
ba75bb98 | 584 | struct ceph_mon_statfs_reply *reply = msg->front.iov_base; |
3143edd3 | 585 | u64 tid = le64_to_cpu(msg->hdr.tid); |
ba75bb98 SW |
586 | |
587 | if (msg->front.iov_len != sizeof(*reply)) | |
588 | goto bad; | |
ba75bb98 SW |
589 | dout("handle_statfs_reply %p tid %llu\n", msg, tid); |
590 | ||
591 | mutex_lock(&monc->mutex); | |
fcd00b68 | 592 | req = lookup_generic_request(&monc->generic_request_tree, tid); |
ba75bb98 | 593 | if (req) { |
f8c76f6f | 594 | *(struct ceph_statfs *)req->buf = reply->st; |
ba75bb98 | 595 | req->result = 0; |
f8c76f6f | 596 | get_generic_request(req); |
ba75bb98 SW |
597 | } |
598 | mutex_unlock(&monc->mutex); | |
3143edd3 | 599 | if (req) { |
03066f23 | 600 | complete_all(&req->completion); |
f8c76f6f | 601 | put_generic_request(req); |
3143edd3 | 602 | } |
ba75bb98 SW |
603 | return; |
604 | ||
605 | bad: | |
7a6fdeb2 | 606 | pr_err("corrupt statfs reply, tid %llu\n", tid); |
9ec7cab1 | 607 | ceph_msg_dump(msg); |
ba75bb98 SW |
608 | } |
609 | ||
610 | /* | |
3143edd3 | 611 | * Do a synchronous statfs(). |
ba75bb98 | 612 | */ |
3143edd3 | 613 | int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf) |
ba75bb98 | 614 | { |
f8c76f6f | 615 | struct ceph_mon_generic_request *req; |
ba75bb98 | 616 | struct ceph_mon_statfs *h; |
3143edd3 SW |
617 | int err; |
618 | ||
cffe7b6d | 619 | req = kzalloc(sizeof(*req), GFP_NOFS); |
3143edd3 SW |
620 | if (!req) |
621 | return -ENOMEM; | |
622 | ||
3143edd3 | 623 | kref_init(&req->kref); |
fcd00b68 | 624 | RB_CLEAR_NODE(&req->node); |
3143edd3 SW |
625 | req->buf = buf; |
626 | init_completion(&req->completion); | |
ba75bb98 | 627 | |
a79832f2 | 628 | err = -ENOMEM; |
b61c2763 SW |
629 | req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS, |
630 | true); | |
a79832f2 | 631 | if (!req->request) |
3143edd3 | 632 | goto out; |
b61c2763 SW |
633 | req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS, |
634 | true); | |
a79832f2 | 635 | if (!req->reply) |
3143edd3 | 636 | goto out; |
3143edd3 SW |
637 | |
638 | /* fill out request */ | |
639 | h = req->request->front.iov_base; | |
13e38c8a SW |
640 | h->monhdr.have_version = 0; |
641 | h->monhdr.session_mon = cpu_to_le16(-1); | |
642 | h->monhdr.session_mon_tid = 0; | |
ba75bb98 | 643 | h->fsid = monc->monmap->fsid; |
ba75bb98 | 644 | |
e56fa10e | 645 | err = do_generic_request(monc, req); |
ba75bb98 | 646 | |
e56fa10e | 647 | out: |
f646912d | 648 | put_generic_request(req); |
e56fa10e YS |
649 | return err; |
650 | } | |
3d14c5d2 | 651 | EXPORT_SYMBOL(ceph_monc_do_statfs); |
e56fa10e | 652 | |
513a8243 ID |
653 | static void handle_get_version_reply(struct ceph_mon_client *monc, |
654 | struct ceph_msg *msg) | |
655 | { | |
656 | struct ceph_mon_generic_request *req; | |
657 | u64 tid = le64_to_cpu(msg->hdr.tid); | |
658 | void *p = msg->front.iov_base; | |
659 | void *end = p + msg->front_alloc_len; | |
660 | u64 handle; | |
661 | ||
662 | dout("%s %p tid %llu\n", __func__, msg, tid); | |
663 | ||
664 | ceph_decode_need(&p, end, 2*sizeof(u64), bad); | |
665 | handle = ceph_decode_64(&p); | |
666 | if (tid != 0 && tid != handle) | |
667 | goto bad; | |
668 | ||
669 | mutex_lock(&monc->mutex); | |
fcd00b68 | 670 | req = lookup_generic_request(&monc->generic_request_tree, handle); |
513a8243 ID |
671 | if (req) { |
672 | *(u64 *)req->buf = ceph_decode_64(&p); | |
673 | req->result = 0; | |
674 | get_generic_request(req); | |
675 | } | |
676 | mutex_unlock(&monc->mutex); | |
677 | if (req) { | |
678 | complete_all(&req->completion); | |
679 | put_generic_request(req); | |
680 | } | |
681 | ||
682 | return; | |
683 | bad: | |
7a6fdeb2 | 684 | pr_err("corrupt mon_get_version reply, tid %llu\n", tid); |
513a8243 ID |
685 | ceph_msg_dump(msg); |
686 | } | |
687 | ||
688 | /* | |
689 | * Send MMonGetVersion and wait for the reply. | |
690 | * | |
691 | * @what: one of "mdsmap", "osdmap" or "monmap" | |
692 | */ | |
693 | int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what, | |
694 | u64 *newest) | |
695 | { | |
696 | struct ceph_mon_generic_request *req; | |
697 | void *p, *end; | |
698 | u64 tid; | |
699 | int err; | |
700 | ||
701 | req = kzalloc(sizeof(*req), GFP_NOFS); | |
702 | if (!req) | |
703 | return -ENOMEM; | |
704 | ||
705 | kref_init(&req->kref); | |
fcd00b68 | 706 | RB_CLEAR_NODE(&req->node); |
513a8243 | 707 | req->buf = newest; |
513a8243 ID |
708 | init_completion(&req->completion); |
709 | ||
710 | req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION, | |
711 | sizeof(u64) + sizeof(u32) + strlen(what), | |
712 | GFP_NOFS, true); | |
713 | if (!req->request) { | |
714 | err = -ENOMEM; | |
715 | goto out; | |
716 | } | |
717 | ||
718 | req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024, | |
719 | GFP_NOFS, true); | |
720 | if (!req->reply) { | |
721 | err = -ENOMEM; | |
722 | goto out; | |
723 | } | |
724 | ||
725 | p = req->request->front.iov_base; | |
726 | end = p + req->request->front_alloc_len; | |
727 | ||
728 | /* fill out request */ | |
729 | mutex_lock(&monc->mutex); | |
730 | tid = ++monc->last_tid; | |
731 | ceph_encode_64(&p, tid); /* handle */ | |
732 | ceph_encode_string(&p, end, what, strlen(what)); | |
733 | ||
734 | err = __do_generic_request(monc, tid, req); | |
735 | ||
736 | mutex_unlock(&monc->mutex); | |
737 | out: | |
f646912d | 738 | put_generic_request(req); |
513a8243 ID |
739 | return err; |
740 | } | |
741 | EXPORT_SYMBOL(ceph_monc_do_get_version); | |
742 | ||
ba75bb98 | 743 | /* |
e56fa10e | 744 | * Resend pending generic requests. |
ba75bb98 | 745 | */ |
f8c76f6f | 746 | static void __resend_generic_request(struct ceph_mon_client *monc) |
ba75bb98 | 747 | { |
f8c76f6f | 748 | struct ceph_mon_generic_request *req; |
85ff03f6 | 749 | struct rb_node *p; |
ba75bb98 | 750 | |
f8c76f6f YS |
751 | for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) { |
752 | req = rb_entry(p, struct ceph_mon_generic_request, node); | |
6740a845 | 753 | ceph_msg_revoke(req->request); |
4f471e4a | 754 | ceph_msg_revoke_incoming(req->reply); |
67130934 | 755 | ceph_con_send(&monc->con, ceph_msg_get(req->request)); |
ba75bb98 SW |
756 | } |
757 | } | |
758 | ||
759 | /* | |
760 | * Delayed work. If we haven't mounted yet, retry. Otherwise, | |
761 | * renew/retry subscription as needed (in case it is timing out, or we | |
762 | * got an ENOMEM). And keep the monitor connection alive. | |
763 | */ | |
764 | static void delayed_work(struct work_struct *work) | |
765 | { | |
766 | struct ceph_mon_client *monc = | |
767 | container_of(work, struct ceph_mon_client, delayed_work.work); | |
768 | ||
769 | dout("monc delayed_work\n"); | |
770 | mutex_lock(&monc->mutex); | |
4e7a5dcd | 771 | if (monc->hunting) { |
1752b50c ID |
772 | dout("%s continuing hunt\n", __func__); |
773 | reopen_session(monc); | |
ba75bb98 | 774 | } else { |
8b9558aa YZ |
775 | int is_auth = ceph_auth_is_authenticated(monc->auth); |
776 | if (ceph_con_keepalive_expired(&monc->con, | |
58d81b12 | 777 | CEPH_MONC_PING_TIMEOUT)) { |
8b9558aa YZ |
778 | dout("monc keepalive timeout\n"); |
779 | is_auth = 0; | |
1752b50c | 780 | reopen_session(monc); |
8b9558aa | 781 | } |
9bd2e6f8 | 782 | |
8b9558aa YZ |
783 | if (!monc->hunting) { |
784 | ceph_con_keepalive(&monc->con); | |
785 | __validate_auth(monc); | |
786 | } | |
9bd2e6f8 | 787 | |
82dcabad ID |
788 | if (is_auth) { |
789 | unsigned long now = jiffies; | |
790 | ||
791 | dout("%s renew subs? now %lu renew after %lu\n", | |
792 | __func__, now, monc->sub_renew_after); | |
793 | if (time_after_eq(now, monc->sub_renew_after)) | |
794 | __send_subscribe(monc); | |
795 | } | |
ba75bb98 | 796 | } |
ba75bb98 SW |
797 | __schedule_delayed(monc); |
798 | mutex_unlock(&monc->mutex); | |
799 | } | |
800 | ||
6b805185 SW |
801 | /* |
802 | * On startup, we build a temporary monmap populated with the IPs | |
803 | * provided by mount(2). | |
804 | */ | |
805 | static int build_initial_monmap(struct ceph_mon_client *monc) | |
806 | { | |
3d14c5d2 YS |
807 | struct ceph_options *opt = monc->client->options; |
808 | struct ceph_entity_addr *mon_addr = opt->mon_addr; | |
809 | int num_mon = opt->num_mon; | |
6b805185 SW |
810 | int i; |
811 | ||
812 | /* build initial monmap */ | |
813 | monc->monmap = kzalloc(sizeof(*monc->monmap) + | |
814 | num_mon*sizeof(monc->monmap->mon_inst[0]), | |
815 | GFP_KERNEL); | |
816 | if (!monc->monmap) | |
817 | return -ENOMEM; | |
818 | for (i = 0; i < num_mon; i++) { | |
819 | monc->monmap->mon_inst[i].addr = mon_addr[i]; | |
6b805185 SW |
820 | monc->monmap->mon_inst[i].addr.nonce = 0; |
821 | monc->monmap->mon_inst[i].name.type = | |
822 | CEPH_ENTITY_TYPE_MON; | |
823 | monc->monmap->mon_inst[i].name.num = cpu_to_le64(i); | |
824 | } | |
825 | monc->monmap->num_mon = num_mon; | |
6b805185 SW |
826 | return 0; |
827 | } | |
828 | ||
ba75bb98 SW |
829 | int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl) |
830 | { | |
831 | int err = 0; | |
832 | ||
833 | dout("init\n"); | |
834 | memset(monc, 0, sizeof(*monc)); | |
835 | monc->client = cl; | |
836 | monc->monmap = NULL; | |
837 | mutex_init(&monc->mutex); | |
838 | ||
6b805185 SW |
839 | err = build_initial_monmap(monc); |
840 | if (err) | |
841 | goto out; | |
842 | ||
f6a2f5be | 843 | /* connection */ |
4e7a5dcd | 844 | /* authentication */ |
3d14c5d2 | 845 | monc->auth = ceph_auth_init(cl->options->name, |
8323c3aa | 846 | cl->options->key); |
49d9224c NW |
847 | if (IS_ERR(monc->auth)) { |
848 | err = PTR_ERR(monc->auth); | |
67130934 | 849 | goto out_monmap; |
49d9224c | 850 | } |
4e7a5dcd SW |
851 | monc->auth->want_keys = |
852 | CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON | | |
853 | CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS; | |
854 | ||
240ed68e | 855 | /* msgs */ |
a79832f2 | 856 | err = -ENOMEM; |
7c315c55 | 857 | monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK, |
34d23762 | 858 | sizeof(struct ceph_mon_subscribe_ack), |
b61c2763 | 859 | GFP_NOFS, true); |
a79832f2 | 860 | if (!monc->m_subscribe_ack) |
49d9224c | 861 | goto out_auth; |
6694d6b9 | 862 | |
b61c2763 SW |
863 | monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS, |
864 | true); | |
240ed68e SW |
865 | if (!monc->m_subscribe) |
866 | goto out_subscribe_ack; | |
867 | ||
b61c2763 SW |
868 | monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS, |
869 | true); | |
a79832f2 | 870 | if (!monc->m_auth_reply) |
240ed68e | 871 | goto out_subscribe; |
4e7a5dcd | 872 | |
b61c2763 | 873 | monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true); |
9bd2e6f8 | 874 | monc->pending_auth = 0; |
a79832f2 | 875 | if (!monc->m_auth) |
6694d6b9 | 876 | goto out_auth_reply; |
ba75bb98 | 877 | |
735a72ef SW |
878 | ceph_con_init(&monc->con, monc, &mon_con_ops, |
879 | &monc->client->msgr); | |
880 | ||
ba75bb98 | 881 | monc->cur_mon = -1; |
168b9090 ID |
882 | monc->had_a_connection = false; |
883 | monc->hunt_mult = 1; | |
ba75bb98 SW |
884 | |
885 | INIT_DELAYED_WORK(&monc->delayed_work, delayed_work); | |
f8c76f6f | 886 | monc->generic_request_tree = RB_ROOT; |
ba75bb98 SW |
887 | monc->last_tid = 0; |
888 | ||
4e7a5dcd SW |
889 | return 0; |
890 | ||
6694d6b9 SW |
891 | out_auth_reply: |
892 | ceph_msg_put(monc->m_auth_reply); | |
240ed68e SW |
893 | out_subscribe: |
894 | ceph_msg_put(monc->m_subscribe); | |
7c315c55 SW |
895 | out_subscribe_ack: |
896 | ceph_msg_put(monc->m_subscribe_ack); | |
49d9224c NW |
897 | out_auth: |
898 | ceph_auth_destroy(monc->auth); | |
4e7a5dcd SW |
899 | out_monmap: |
900 | kfree(monc->monmap); | |
ba75bb98 SW |
901 | out: |
902 | return err; | |
903 | } | |
3d14c5d2 | 904 | EXPORT_SYMBOL(ceph_monc_init); |
ba75bb98 SW |
905 | |
906 | void ceph_monc_stop(struct ceph_mon_client *monc) | |
907 | { | |
908 | dout("stop\n"); | |
909 | cancel_delayed_work_sync(&monc->delayed_work); | |
910 | ||
911 | mutex_lock(&monc->mutex); | |
912 | __close_session(monc); | |
0e04dc26 | 913 | monc->cur_mon = -1; |
ba75bb98 SW |
914 | mutex_unlock(&monc->mutex); |
915 | ||
f3dea7ed SW |
916 | /* |
917 | * flush msgr queue before we destroy ourselves to ensure that: | |
918 | * - any work that references our embedded con is finished. | |
919 | * - any osd_client or other work that may reference an authorizer | |
920 | * finishes before we shut down the auth subsystem. | |
921 | */ | |
922 | ceph_msgr_flush(); | |
923 | ||
4e7a5dcd SW |
924 | ceph_auth_destroy(monc->auth); |
925 | ||
926 | ceph_msg_put(monc->m_auth); | |
6694d6b9 | 927 | ceph_msg_put(monc->m_auth_reply); |
240ed68e | 928 | ceph_msg_put(monc->m_subscribe); |
7c315c55 | 929 | ceph_msg_put(monc->m_subscribe_ack); |
ba75bb98 SW |
930 | |
931 | kfree(monc->monmap); | |
932 | } | |
3d14c5d2 | 933 | EXPORT_SYMBOL(ceph_monc_stop); |
ba75bb98 | 934 | |
0f9af169 ID |
935 | static void finish_hunting(struct ceph_mon_client *monc) |
936 | { | |
937 | if (monc->hunting) { | |
938 | dout("%s found mon%d\n", __func__, monc->cur_mon); | |
939 | monc->hunting = false; | |
168b9090 ID |
940 | monc->had_a_connection = true; |
941 | monc->hunt_mult /= 2; /* reduce by 50% */ | |
942 | if (monc->hunt_mult < 1) | |
943 | monc->hunt_mult = 1; | |
0f9af169 ID |
944 | } |
945 | } | |
946 | ||
4e7a5dcd SW |
947 | static void handle_auth_reply(struct ceph_mon_client *monc, |
948 | struct ceph_msg *msg) | |
949 | { | |
950 | int ret; | |
09c4d6a7 | 951 | int was_auth = 0; |
4e7a5dcd SW |
952 | |
953 | mutex_lock(&monc->mutex); | |
27859f97 | 954 | was_auth = ceph_auth_is_authenticated(monc->auth); |
9bd2e6f8 | 955 | monc->pending_auth = 0; |
4e7a5dcd SW |
956 | ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base, |
957 | msg->front.iov_len, | |
958 | monc->m_auth->front.iov_base, | |
3cea4c30 | 959 | monc->m_auth->front_alloc_len); |
0f9af169 ID |
960 | if (ret > 0) { |
961 | __send_prepared_auth_request(monc, ret); | |
962 | goto out; | |
963 | } | |
964 | ||
965 | finish_hunting(monc); | |
966 | ||
4e7a5dcd | 967 | if (ret < 0) { |
9bd2e6f8 | 968 | monc->client->auth_err = ret; |
27859f97 | 969 | } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) { |
4e7a5dcd | 970 | dout("authenticated, starting session\n"); |
0743304d | 971 | |
15d9882c AE |
972 | monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT; |
973 | monc->client->msgr.inst.name.num = | |
0cf5537b | 974 | cpu_to_le64(monc->auth->global_id); |
0743304d | 975 | |
4e7a5dcd | 976 | __send_subscribe(monc); |
f8c76f6f | 977 | __resend_generic_request(monc); |
0f9af169 ID |
978 | |
979 | pr_info("mon%d %s session established\n", monc->cur_mon, | |
980 | ceph_pr_addr(&monc->con.peer_addr.in_addr)); | |
4e7a5dcd | 981 | } |
0f9af169 ID |
982 | |
983 | out: | |
4e7a5dcd | 984 | mutex_unlock(&monc->mutex); |
0f9af169 ID |
985 | if (monc->client->auth_err < 0) |
986 | wake_up_all(&monc->client->auth_wq); | |
4e7a5dcd SW |
987 | } |
988 | ||
9bd2e6f8 SW |
989 | static int __validate_auth(struct ceph_mon_client *monc) |
990 | { | |
991 | int ret; | |
992 | ||
993 | if (monc->pending_auth) | |
994 | return 0; | |
995 | ||
996 | ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base, | |
3cea4c30 | 997 | monc->m_auth->front_alloc_len); |
9bd2e6f8 SW |
998 | if (ret <= 0) |
999 | return ret; /* either an error, or no need to authenticate */ | |
1000 | __send_prepared_auth_request(monc, ret); | |
1001 | return 0; | |
1002 | } | |
1003 | ||
1004 | int ceph_monc_validate_auth(struct ceph_mon_client *monc) | |
1005 | { | |
1006 | int ret; | |
1007 | ||
1008 | mutex_lock(&monc->mutex); | |
1009 | ret = __validate_auth(monc); | |
1010 | mutex_unlock(&monc->mutex); | |
1011 | return ret; | |
1012 | } | |
3d14c5d2 | 1013 | EXPORT_SYMBOL(ceph_monc_validate_auth); |
9bd2e6f8 | 1014 | |
ba75bb98 SW |
1015 | /* |
1016 | * handle incoming message | |
1017 | */ | |
1018 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
1019 | { | |
1020 | struct ceph_mon_client *monc = con->private; | |
1021 | int type = le16_to_cpu(msg->hdr.type); | |
1022 | ||
1023 | if (!monc) | |
1024 | return; | |
1025 | ||
1026 | switch (type) { | |
4e7a5dcd SW |
1027 | case CEPH_MSG_AUTH_REPLY: |
1028 | handle_auth_reply(monc, msg); | |
ba75bb98 SW |
1029 | break; |
1030 | ||
1031 | case CEPH_MSG_MON_SUBSCRIBE_ACK: | |
1032 | handle_subscribe_ack(monc, msg); | |
1033 | break; | |
1034 | ||
1035 | case CEPH_MSG_STATFS_REPLY: | |
1036 | handle_statfs_reply(monc, msg); | |
1037 | break; | |
1038 | ||
513a8243 ID |
1039 | case CEPH_MSG_MON_GET_VERSION_REPLY: |
1040 | handle_get_version_reply(monc, msg); | |
1041 | break; | |
1042 | ||
4e7a5dcd SW |
1043 | case CEPH_MSG_MON_MAP: |
1044 | ceph_monc_handle_map(monc, msg); | |
1045 | break; | |
1046 | ||
ba75bb98 SW |
1047 | case CEPH_MSG_OSD_MAP: |
1048 | ceph_osdc_handle_map(&monc->client->osdc, msg); | |
1049 | break; | |
1050 | ||
1051 | default: | |
3d14c5d2 YS |
1052 | /* can the chained handler handle it? */ |
1053 | if (monc->client->extra_mon_dispatch && | |
1054 | monc->client->extra_mon_dispatch(monc->client, msg) == 0) | |
1055 | break; | |
1056 | ||
ba75bb98 SW |
1057 | pr_err("received unknown message type %d %s\n", type, |
1058 | ceph_msg_type_name(type)); | |
1059 | } | |
1060 | ceph_msg_put(msg); | |
1061 | } | |
1062 | ||
1063 | /* | |
1064 | * Allocate memory for incoming message | |
1065 | */ | |
1066 | static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con, | |
2450418c YS |
1067 | struct ceph_msg_header *hdr, |
1068 | int *skip) | |
ba75bb98 SW |
1069 | { |
1070 | struct ceph_mon_client *monc = con->private; | |
1071 | int type = le16_to_cpu(hdr->type); | |
2450418c | 1072 | int front_len = le32_to_cpu(hdr->front_len); |
5b3a4db3 | 1073 | struct ceph_msg *m = NULL; |
ba75bb98 | 1074 | |
2450418c | 1075 | *skip = 0; |
0547a9b3 | 1076 | |
ba75bb98 | 1077 | switch (type) { |
ba75bb98 | 1078 | case CEPH_MSG_MON_SUBSCRIBE_ACK: |
7c315c55 | 1079 | m = ceph_msg_get(monc->m_subscribe_ack); |
2450418c | 1080 | break; |
ba75bb98 | 1081 | case CEPH_MSG_STATFS_REPLY: |
f8c76f6f | 1082 | return get_generic_reply(con, hdr, skip); |
4e7a5dcd | 1083 | case CEPH_MSG_AUTH_REPLY: |
6694d6b9 | 1084 | m = ceph_msg_get(monc->m_auth_reply); |
2450418c | 1085 | break; |
513a8243 ID |
1086 | case CEPH_MSG_MON_GET_VERSION_REPLY: |
1087 | if (le64_to_cpu(hdr->tid) != 0) | |
1088 | return get_generic_reply(con, hdr, skip); | |
1089 | ||
1090 | /* | |
1091 | * Older OSDs don't set reply tid even if the orignal | |
1092 | * request had a non-zero tid. Workaround this weirdness | |
1093 | * by falling through to the allocate case. | |
1094 | */ | |
5b3a4db3 SW |
1095 | case CEPH_MSG_MON_MAP: |
1096 | case CEPH_MSG_MDS_MAP: | |
1097 | case CEPH_MSG_OSD_MAP: | |
b61c2763 | 1098 | m = ceph_msg_new(type, front_len, GFP_NOFS, false); |
1c20f2d2 AE |
1099 | if (!m) |
1100 | return NULL; /* ENOMEM--return skip == 0 */ | |
5b3a4db3 | 1101 | break; |
ba75bb98 | 1102 | } |
2450418c | 1103 | |
5b3a4db3 SW |
1104 | if (!m) { |
1105 | pr_info("alloc_msg unknown type %d\n", type); | |
2450418c | 1106 | *skip = 1; |
73c3d481 | 1107 | } else if (front_len > m->front_alloc_len) { |
b9a67899 JP |
1108 | pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n", |
1109 | front_len, m->front_alloc_len, | |
1110 | (unsigned int)con->peer_name.type, | |
1111 | le64_to_cpu(con->peer_name.num)); | |
73c3d481 SW |
1112 | ceph_msg_put(m); |
1113 | m = ceph_msg_new(type, front_len, GFP_NOFS, false); | |
5b3a4db3 | 1114 | } |
73c3d481 | 1115 | |
2450418c | 1116 | return m; |
ba75bb98 SW |
1117 | } |
1118 | ||
1119 | /* | |
1120 | * If the monitor connection resets, pick a new monitor and resubmit | |
1121 | * any pending requests. | |
1122 | */ | |
1123 | static void mon_fault(struct ceph_connection *con) | |
1124 | { | |
1125 | struct ceph_mon_client *monc = con->private; | |
1126 | ||
ba75bb98 | 1127 | mutex_lock(&monc->mutex); |
b5d91704 ID |
1128 | dout("%s mon%d\n", __func__, monc->cur_mon); |
1129 | if (monc->cur_mon >= 0) { | |
1130 | if (!monc->hunting) { | |
1131 | dout("%s hunting for new mon\n", __func__); | |
1132 | reopen_session(monc); | |
1133 | __schedule_delayed(monc); | |
1134 | } else { | |
1135 | dout("%s already hunting\n", __func__); | |
1136 | } | |
ba75bb98 | 1137 | } |
ba75bb98 SW |
1138 | mutex_unlock(&monc->mutex); |
1139 | } | |
1140 | ||
ec87ef43 SW |
1141 | /* |
1142 | * We can ignore refcounting on the connection struct, as all references | |
1143 | * will come from the messenger workqueue, which is drained prior to | |
1144 | * mon_client destruction. | |
1145 | */ | |
1146 | static struct ceph_connection *con_get(struct ceph_connection *con) | |
1147 | { | |
1148 | return con; | |
1149 | } | |
1150 | ||
1151 | static void con_put(struct ceph_connection *con) | |
1152 | { | |
1153 | } | |
1154 | ||
9e32789f | 1155 | static const struct ceph_connection_operations mon_con_ops = { |
ec87ef43 SW |
1156 | .get = con_get, |
1157 | .put = con_put, | |
ba75bb98 SW |
1158 | .dispatch = dispatch, |
1159 | .fault = mon_fault, | |
1160 | .alloc_msg = mon_alloc_msg, | |
ba75bb98 | 1161 | }; |