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); |
f6a2f5be SW |
125 | monc->cur_mon = -1; |
126 | monc->pending_auth = 0; | |
127 | ceph_auth_reset(monc->auth); | |
ba75bb98 SW |
128 | } |
129 | ||
130 | /* | |
131 | * Open a session with a (new) monitor. | |
132 | */ | |
133 | static int __open_session(struct ceph_mon_client *monc) | |
134 | { | |
135 | char r; | |
4e7a5dcd | 136 | int ret; |
ba75bb98 SW |
137 | |
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); | |
ba75bb98 | 143 | monc->sub_renew_after = jiffies; /* i.e., expired */ |
82dcabad | 144 | monc->sub_renew_sent = 0; |
ba75bb98 | 145 | |
20581c1f | 146 | dout("open_session mon%d opening\n", monc->cur_mon); |
67130934 | 147 | ceph_con_open(&monc->con, |
b7a9e5dd | 148 | CEPH_ENTITY_TYPE_MON, monc->cur_mon, |
ba75bb98 | 149 | &monc->monmap->mon_inst[monc->cur_mon].addr); |
4e7a5dcd | 150 | |
8b9558aa YZ |
151 | /* send an initial keepalive to ensure our timestamp is |
152 | * valid by the time we are in an OPENED state */ | |
153 | ceph_con_keepalive(&monc->con); | |
154 | ||
4e7a5dcd SW |
155 | /* initiatiate authentication handshake */ |
156 | ret = ceph_auth_build_hello(monc->auth, | |
157 | monc->m_auth->front.iov_base, | |
3cea4c30 | 158 | monc->m_auth->front_alloc_len); |
5ce6e9db | 159 | __send_prepared_auth_request(monc, ret); |
ba75bb98 SW |
160 | } else { |
161 | dout("open_session mon%d already open\n", monc->cur_mon); | |
162 | } | |
163 | return 0; | |
164 | } | |
165 | ||
166 | static bool __sub_expired(struct ceph_mon_client *monc) | |
167 | { | |
168 | return time_after_eq(jiffies, monc->sub_renew_after); | |
169 | } | |
170 | ||
171 | /* | |
172 | * Reschedule delayed work timer. | |
173 | */ | |
174 | static void __schedule_delayed(struct ceph_mon_client *monc) | |
175 | { | |
8b9558aa YZ |
176 | struct ceph_options *opt = monc->client->options; |
177 | unsigned long delay; | |
ba75bb98 | 178 | |
8b9558aa | 179 | if (monc->cur_mon < 0 || __sub_expired(monc)) { |
ba75bb98 | 180 | delay = 10 * HZ; |
8b9558aa | 181 | } else { |
ba75bb98 | 182 | delay = 20 * HZ; |
8b9558aa YZ |
183 | if (opt->monc_ping_timeout > 0) |
184 | delay = min(delay, opt->monc_ping_timeout / 3); | |
185 | } | |
186 | dout("__schedule_delayed after %lu\n", delay); | |
187 | schedule_delayed_work(&monc->delayed_work, | |
188 | round_jiffies_relative(delay)); | |
ba75bb98 SW |
189 | } |
190 | ||
82dcabad ID |
191 | const char *ceph_sub_str[] = { |
192 | [CEPH_SUB_MDSMAP] = "mdsmap", | |
193 | [CEPH_SUB_MONMAP] = "monmap", | |
194 | [CEPH_SUB_OSDMAP] = "osdmap", | |
195 | }; | |
196 | ||
ba75bb98 | 197 | /* |
82dcabad ID |
198 | * Send subscribe request for one or more maps, according to |
199 | * monc->subs. | |
ba75bb98 SW |
200 | */ |
201 | static void __send_subscribe(struct ceph_mon_client *monc) | |
202 | { | |
82dcabad ID |
203 | struct ceph_msg *msg = monc->m_subscribe; |
204 | void *p = msg->front.iov_base; | |
205 | void *const end = p + msg->front_alloc_len; | |
206 | int num = 0; | |
207 | int i; | |
208 | ||
209 | dout("%s sent %lu\n", __func__, monc->sub_renew_sent); | |
210 | ||
211 | BUG_ON(monc->cur_mon < 0); | |
212 | ||
213 | if (!monc->sub_renew_sent) | |
214 | monc->sub_renew_sent = jiffies | 1; /* never 0 */ | |
215 | ||
216 | msg->hdr.version = cpu_to_le16(2); | |
217 | ||
218 | for (i = 0; i < ARRAY_SIZE(monc->subs); i++) { | |
219 | if (monc->subs[i].want) | |
220 | num++; | |
221 | } | |
222 | BUG_ON(num < 1); /* monmap sub is always there */ | |
223 | ceph_encode_32(&p, num); | |
224 | for (i = 0; i < ARRAY_SIZE(monc->subs); i++) { | |
225 | const char *s = ceph_sub_str[i]; | |
226 | ||
227 | if (!monc->subs[i].want) | |
228 | continue; | |
229 | ||
230 | dout("%s %s start %llu flags 0x%x\n", __func__, s, | |
231 | le64_to_cpu(monc->subs[i].item.start), | |
232 | monc->subs[i].item.flags); | |
233 | ceph_encode_string(&p, end, s, strlen(s)); | |
234 | memcpy(p, &monc->subs[i].item, sizeof(monc->subs[i].item)); | |
235 | p += sizeof(monc->subs[i].item); | |
ba75bb98 | 236 | } |
82dcabad ID |
237 | |
238 | BUG_ON(p != (end - 35 - (ARRAY_SIZE(monc->subs) - num) * 19)); | |
239 | msg->front.iov_len = p - msg->front.iov_base; | |
240 | msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); | |
241 | ceph_msg_revoke(msg); | |
242 | ceph_con_send(&monc->con, ceph_msg_get(msg)); | |
ba75bb98 SW |
243 | } |
244 | ||
245 | static void handle_subscribe_ack(struct ceph_mon_client *monc, | |
246 | struct ceph_msg *msg) | |
247 | { | |
95c96174 | 248 | unsigned int seconds; |
07bd10fb SW |
249 | struct ceph_mon_subscribe_ack *h = msg->front.iov_base; |
250 | ||
251 | if (msg->front.iov_len < sizeof(*h)) | |
252 | goto bad; | |
253 | seconds = le32_to_cpu(h->duration); | |
ba75bb98 | 254 | |
ba75bb98 | 255 | mutex_lock(&monc->mutex); |
82dcabad ID |
256 | if (monc->sub_renew_sent) { |
257 | monc->sub_renew_after = monc->sub_renew_sent + | |
258 | (seconds >> 1) * HZ - 1; | |
259 | dout("%s sent %lu duration %d renew after %lu\n", __func__, | |
260 | monc->sub_renew_sent, seconds, monc->sub_renew_after); | |
261 | monc->sub_renew_sent = 0; | |
262 | } else { | |
263 | dout("%s sent %lu renew after %lu, ignoring\n", __func__, | |
264 | monc->sub_renew_sent, monc->sub_renew_after); | |
265 | } | |
ba75bb98 SW |
266 | mutex_unlock(&monc->mutex); |
267 | return; | |
268 | bad: | |
269 | pr_err("got corrupt subscribe-ack msg\n"); | |
9ec7cab1 | 270 | ceph_msg_dump(msg); |
ba75bb98 SW |
271 | } |
272 | ||
273 | /* | |
82dcabad ID |
274 | * Register interest in a map |
275 | * | |
276 | * @sub: one of CEPH_SUB_* | |
277 | * @epoch: X for "every map since X", or 0 for "just the latest" | |
ba75bb98 | 278 | */ |
82dcabad ID |
279 | static bool __ceph_monc_want_map(struct ceph_mon_client *monc, int sub, |
280 | u32 epoch, bool continuous) | |
ba75bb98 | 281 | { |
82dcabad ID |
282 | __le64 start = cpu_to_le64(epoch); |
283 | u8 flags = !continuous ? CEPH_SUBSCRIBE_ONETIME : 0; | |
284 | ||
285 | dout("%s %s epoch %u continuous %d\n", __func__, ceph_sub_str[sub], | |
286 | epoch, continuous); | |
287 | ||
288 | if (monc->subs[sub].want && | |
289 | monc->subs[sub].item.start == start && | |
290 | monc->subs[sub].item.flags == flags) | |
291 | return false; | |
292 | ||
293 | monc->subs[sub].item.start = start; | |
294 | monc->subs[sub].item.flags = flags; | |
295 | monc->subs[sub].want = true; | |
296 | ||
297 | return true; | |
298 | } | |
299 | ||
300 | bool ceph_monc_want_map(struct ceph_mon_client *monc, int sub, u32 epoch, | |
301 | bool continuous) | |
302 | { | |
303 | bool need_request; | |
304 | ||
ba75bb98 | 305 | mutex_lock(&monc->mutex); |
82dcabad | 306 | need_request = __ceph_monc_want_map(monc, sub, epoch, continuous); |
ba75bb98 | 307 | mutex_unlock(&monc->mutex); |
82dcabad ID |
308 | |
309 | return need_request; | |
310 | } | |
311 | EXPORT_SYMBOL(ceph_monc_want_map); | |
312 | ||
313 | /* | |
314 | * Keep track of which maps we have | |
315 | * | |
316 | * @sub: one of CEPH_SUB_* | |
317 | */ | |
318 | static void __ceph_monc_got_map(struct ceph_mon_client *monc, int sub, | |
319 | u32 epoch) | |
320 | { | |
321 | dout("%s %s epoch %u\n", __func__, ceph_sub_str[sub], epoch); | |
322 | ||
323 | if (monc->subs[sub].want) { | |
324 | if (monc->subs[sub].item.flags & CEPH_SUBSCRIBE_ONETIME) | |
325 | monc->subs[sub].want = false; | |
326 | else | |
327 | monc->subs[sub].item.start = cpu_to_le64(epoch + 1); | |
328 | } | |
329 | ||
330 | monc->subs[sub].have = epoch; | |
ba75bb98 SW |
331 | } |
332 | ||
82dcabad | 333 | void ceph_monc_got_map(struct ceph_mon_client *monc, int sub, u32 epoch) |
ba75bb98 SW |
334 | { |
335 | mutex_lock(&monc->mutex); | |
82dcabad | 336 | __ceph_monc_got_map(monc, sub, epoch); |
ba75bb98 | 337 | mutex_unlock(&monc->mutex); |
ba75bb98 | 338 | } |
82dcabad | 339 | EXPORT_SYMBOL(ceph_monc_got_map); |
ba75bb98 SW |
340 | |
341 | /* | |
342 | * Register interest in the next osdmap | |
343 | */ | |
344 | void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc) | |
345 | { | |
82dcabad | 346 | dout("%s have %u\n", __func__, monc->subs[CEPH_SUB_OSDMAP].have); |
ba75bb98 | 347 | mutex_lock(&monc->mutex); |
82dcabad ID |
348 | if (__ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, |
349 | monc->subs[CEPH_SUB_OSDMAP].have + 1, false)) | |
ba75bb98 SW |
350 | __send_subscribe(monc); |
351 | mutex_unlock(&monc->mutex); | |
352 | } | |
6044cde6 ID |
353 | EXPORT_SYMBOL(ceph_monc_request_next_osdmap); |
354 | ||
a319bf56 ID |
355 | /* |
356 | * Wait for an osdmap with a given epoch. | |
357 | * | |
358 | * @epoch: epoch to wait for | |
359 | * @timeout: in jiffies, 0 means "wait forever" | |
360 | */ | |
6044cde6 ID |
361 | int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch, |
362 | unsigned long timeout) | |
363 | { | |
364 | unsigned long started = jiffies; | |
216639dd | 365 | long ret; |
6044cde6 ID |
366 | |
367 | mutex_lock(&monc->mutex); | |
82dcabad | 368 | while (monc->subs[CEPH_SUB_OSDMAP].have < epoch) { |
6044cde6 ID |
369 | mutex_unlock(&monc->mutex); |
370 | ||
a319bf56 | 371 | if (timeout && time_after_eq(jiffies, started + timeout)) |
6044cde6 ID |
372 | return -ETIMEDOUT; |
373 | ||
374 | ret = wait_event_interruptible_timeout(monc->client->auth_wq, | |
82dcabad ID |
375 | monc->subs[CEPH_SUB_OSDMAP].have >= epoch, |
376 | ceph_timeout_jiffies(timeout)); | |
6044cde6 ID |
377 | if (ret < 0) |
378 | return ret; | |
379 | ||
380 | mutex_lock(&monc->mutex); | |
381 | } | |
382 | ||
383 | mutex_unlock(&monc->mutex); | |
384 | return 0; | |
385 | } | |
386 | EXPORT_SYMBOL(ceph_monc_wait_osdmap); | |
ba75bb98 | 387 | |
4e7a5dcd | 388 | /* |
82dcabad ID |
389 | * Open a session with a random monitor. Request monmap and osdmap, |
390 | * which are waited upon in __ceph_open_session(). | |
4e7a5dcd SW |
391 | */ |
392 | int ceph_monc_open_session(struct ceph_mon_client *monc) | |
ba75bb98 | 393 | { |
ba75bb98 | 394 | mutex_lock(&monc->mutex); |
82dcabad ID |
395 | __ceph_monc_want_map(monc, CEPH_SUB_MONMAP, 0, true); |
396 | __ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, 0, false); | |
4e7a5dcd | 397 | __open_session(monc); |
ba75bb98 SW |
398 | __schedule_delayed(monc); |
399 | mutex_unlock(&monc->mutex); | |
400 | return 0; | |
401 | } | |
3d14c5d2 | 402 | EXPORT_SYMBOL(ceph_monc_open_session); |
ba75bb98 | 403 | |
0743304d SW |
404 | static void ceph_monc_handle_map(struct ceph_mon_client *monc, |
405 | struct ceph_msg *msg) | |
4e7a5dcd SW |
406 | { |
407 | struct ceph_client *client = monc->client; | |
408 | struct ceph_monmap *monmap = NULL, *old = monc->monmap; | |
409 | void *p, *end; | |
410 | ||
411 | mutex_lock(&monc->mutex); | |
412 | ||
413 | dout("handle_monmap\n"); | |
414 | p = msg->front.iov_base; | |
415 | end = p + msg->front.iov_len; | |
416 | ||
417 | monmap = ceph_monmap_decode(p, end); | |
418 | if (IS_ERR(monmap)) { | |
419 | pr_err("problem decoding monmap, %d\n", | |
420 | (int)PTR_ERR(monmap)); | |
d4a780ce | 421 | goto out; |
4e7a5dcd | 422 | } |
0743304d SW |
423 | |
424 | if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) { | |
4e7a5dcd | 425 | kfree(monmap); |
d4a780ce | 426 | goto out; |
4e7a5dcd SW |
427 | } |
428 | ||
429 | client->monc.monmap = monmap; | |
4e7a5dcd SW |
430 | kfree(old); |
431 | ||
82dcabad | 432 | __ceph_monc_got_map(monc, CEPH_SUB_MONMAP, monc->monmap->epoch); |
02ac956c | 433 | client->have_fsid = true; |
d1c338a5 | 434 | |
d4a780ce | 435 | out: |
4e7a5dcd | 436 | mutex_unlock(&monc->mutex); |
03066f23 | 437 | wake_up_all(&client->auth_wq); |
4e7a5dcd SW |
438 | } |
439 | ||
ba75bb98 | 440 | /* |
7a6fdeb2 | 441 | * generic requests (currently statfs, mon_get_version) |
ba75bb98 | 442 | */ |
f8c76f6f | 443 | static struct ceph_mon_generic_request *__lookup_generic_req( |
85ff03f6 SW |
444 | struct ceph_mon_client *monc, u64 tid) |
445 | { | |
f8c76f6f YS |
446 | struct ceph_mon_generic_request *req; |
447 | struct rb_node *n = monc->generic_request_tree.rb_node; | |
85ff03f6 SW |
448 | |
449 | while (n) { | |
f8c76f6f | 450 | req = rb_entry(n, struct ceph_mon_generic_request, node); |
85ff03f6 SW |
451 | if (tid < req->tid) |
452 | n = n->rb_left; | |
453 | else if (tid > req->tid) | |
454 | n = n->rb_right; | |
455 | else | |
456 | return req; | |
457 | } | |
458 | return NULL; | |
459 | } | |
460 | ||
f8c76f6f YS |
461 | static void __insert_generic_request(struct ceph_mon_client *monc, |
462 | struct ceph_mon_generic_request *new) | |
85ff03f6 | 463 | { |
f8c76f6f | 464 | struct rb_node **p = &monc->generic_request_tree.rb_node; |
85ff03f6 | 465 | struct rb_node *parent = NULL; |
f8c76f6f | 466 | struct ceph_mon_generic_request *req = NULL; |
85ff03f6 SW |
467 | |
468 | while (*p) { | |
469 | parent = *p; | |
f8c76f6f | 470 | req = rb_entry(parent, struct ceph_mon_generic_request, node); |
85ff03f6 SW |
471 | if (new->tid < req->tid) |
472 | p = &(*p)->rb_left; | |
473 | else if (new->tid > req->tid) | |
474 | p = &(*p)->rb_right; | |
475 | else | |
476 | BUG(); | |
477 | } | |
478 | ||
479 | rb_link_node(&new->node, parent, p); | |
f8c76f6f | 480 | rb_insert_color(&new->node, &monc->generic_request_tree); |
85ff03f6 SW |
481 | } |
482 | ||
f8c76f6f | 483 | static void release_generic_request(struct kref *kref) |
3143edd3 | 484 | { |
f8c76f6f YS |
485 | struct ceph_mon_generic_request *req = |
486 | container_of(kref, struct ceph_mon_generic_request, kref); | |
3143edd3 SW |
487 | |
488 | if (req->reply) | |
489 | ceph_msg_put(req->reply); | |
490 | if (req->request) | |
491 | ceph_msg_put(req->request); | |
20547567 YS |
492 | |
493 | kfree(req); | |
3143edd3 SW |
494 | } |
495 | ||
f8c76f6f | 496 | static void put_generic_request(struct ceph_mon_generic_request *req) |
3143edd3 | 497 | { |
f8c76f6f | 498 | kref_put(&req->kref, release_generic_request); |
3143edd3 SW |
499 | } |
500 | ||
f8c76f6f | 501 | static void get_generic_request(struct ceph_mon_generic_request *req) |
3143edd3 SW |
502 | { |
503 | kref_get(&req->kref); | |
504 | } | |
505 | ||
f8c76f6f | 506 | static struct ceph_msg *get_generic_reply(struct ceph_connection *con, |
3143edd3 SW |
507 | struct ceph_msg_header *hdr, |
508 | int *skip) | |
509 | { | |
510 | struct ceph_mon_client *monc = con->private; | |
f8c76f6f | 511 | struct ceph_mon_generic_request *req; |
3143edd3 SW |
512 | u64 tid = le64_to_cpu(hdr->tid); |
513 | struct ceph_msg *m; | |
514 | ||
515 | mutex_lock(&monc->mutex); | |
f8c76f6f | 516 | req = __lookup_generic_req(monc, tid); |
3143edd3 | 517 | if (!req) { |
f8c76f6f | 518 | dout("get_generic_reply %lld dne\n", tid); |
3143edd3 SW |
519 | *skip = 1; |
520 | m = NULL; | |
521 | } else { | |
f8c76f6f | 522 | dout("get_generic_reply %lld got %p\n", tid, req->reply); |
1c20f2d2 | 523 | *skip = 0; |
3143edd3 SW |
524 | m = ceph_msg_get(req->reply); |
525 | /* | |
526 | * we don't need to track the connection reading into | |
527 | * this reply because we only have one open connection | |
528 | * at a time, ever. | |
529 | */ | |
530 | } | |
531 | mutex_unlock(&monc->mutex); | |
532 | return m; | |
533 | } | |
534 | ||
513a8243 ID |
535 | static int __do_generic_request(struct ceph_mon_client *monc, u64 tid, |
536 | struct ceph_mon_generic_request *req) | |
e56fa10e YS |
537 | { |
538 | int err; | |
539 | ||
540 | /* register request */ | |
513a8243 | 541 | req->tid = tid != 0 ? tid : ++monc->last_tid; |
e56fa10e YS |
542 | req->request->hdr.tid = cpu_to_le64(req->tid); |
543 | __insert_generic_request(monc, req); | |
544 | monc->num_generic_requests++; | |
67130934 | 545 | ceph_con_send(&monc->con, ceph_msg_get(req->request)); |
e56fa10e YS |
546 | mutex_unlock(&monc->mutex); |
547 | ||
548 | err = wait_for_completion_interruptible(&req->completion); | |
549 | ||
550 | mutex_lock(&monc->mutex); | |
551 | rb_erase(&req->node, &monc->generic_request_tree); | |
552 | monc->num_generic_requests--; | |
e56fa10e YS |
553 | |
554 | if (!err) | |
555 | err = req->result; | |
556 | return err; | |
557 | } | |
558 | ||
513a8243 ID |
559 | static int do_generic_request(struct ceph_mon_client *monc, |
560 | struct ceph_mon_generic_request *req) | |
561 | { | |
562 | int err; | |
563 | ||
564 | mutex_lock(&monc->mutex); | |
565 | err = __do_generic_request(monc, 0, req); | |
566 | mutex_unlock(&monc->mutex); | |
567 | ||
568 | return err; | |
569 | } | |
570 | ||
e56fa10e YS |
571 | /* |
572 | * statfs | |
573 | */ | |
ba75bb98 SW |
574 | static void handle_statfs_reply(struct ceph_mon_client *monc, |
575 | struct ceph_msg *msg) | |
576 | { | |
f8c76f6f | 577 | struct ceph_mon_generic_request *req; |
ba75bb98 | 578 | struct ceph_mon_statfs_reply *reply = msg->front.iov_base; |
3143edd3 | 579 | u64 tid = le64_to_cpu(msg->hdr.tid); |
ba75bb98 SW |
580 | |
581 | if (msg->front.iov_len != sizeof(*reply)) | |
582 | goto bad; | |
ba75bb98 SW |
583 | dout("handle_statfs_reply %p tid %llu\n", msg, tid); |
584 | ||
585 | mutex_lock(&monc->mutex); | |
f8c76f6f | 586 | req = __lookup_generic_req(monc, tid); |
ba75bb98 | 587 | if (req) { |
f8c76f6f | 588 | *(struct ceph_statfs *)req->buf = reply->st; |
ba75bb98 | 589 | req->result = 0; |
f8c76f6f | 590 | get_generic_request(req); |
ba75bb98 SW |
591 | } |
592 | mutex_unlock(&monc->mutex); | |
3143edd3 | 593 | if (req) { |
03066f23 | 594 | complete_all(&req->completion); |
f8c76f6f | 595 | put_generic_request(req); |
3143edd3 | 596 | } |
ba75bb98 SW |
597 | return; |
598 | ||
599 | bad: | |
7a6fdeb2 | 600 | pr_err("corrupt statfs reply, tid %llu\n", tid); |
9ec7cab1 | 601 | ceph_msg_dump(msg); |
ba75bb98 SW |
602 | } |
603 | ||
604 | /* | |
3143edd3 | 605 | * Do a synchronous statfs(). |
ba75bb98 | 606 | */ |
3143edd3 | 607 | int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf) |
ba75bb98 | 608 | { |
f8c76f6f | 609 | struct ceph_mon_generic_request *req; |
ba75bb98 | 610 | struct ceph_mon_statfs *h; |
3143edd3 SW |
611 | int err; |
612 | ||
cffe7b6d | 613 | req = kzalloc(sizeof(*req), GFP_NOFS); |
3143edd3 SW |
614 | if (!req) |
615 | return -ENOMEM; | |
616 | ||
3143edd3 SW |
617 | kref_init(&req->kref); |
618 | req->buf = buf; | |
619 | init_completion(&req->completion); | |
ba75bb98 | 620 | |
a79832f2 | 621 | err = -ENOMEM; |
b61c2763 SW |
622 | req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS, |
623 | true); | |
a79832f2 | 624 | if (!req->request) |
3143edd3 | 625 | goto out; |
b61c2763 SW |
626 | req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS, |
627 | true); | |
a79832f2 | 628 | if (!req->reply) |
3143edd3 | 629 | goto out; |
3143edd3 SW |
630 | |
631 | /* fill out request */ | |
632 | h = req->request->front.iov_base; | |
13e38c8a SW |
633 | h->monhdr.have_version = 0; |
634 | h->monhdr.session_mon = cpu_to_le16(-1); | |
635 | h->monhdr.session_mon_tid = 0; | |
ba75bb98 | 636 | h->fsid = monc->monmap->fsid; |
ba75bb98 | 637 | |
e56fa10e | 638 | err = do_generic_request(monc, req); |
ba75bb98 | 639 | |
e56fa10e | 640 | out: |
f646912d | 641 | put_generic_request(req); |
e56fa10e YS |
642 | return err; |
643 | } | |
3d14c5d2 | 644 | EXPORT_SYMBOL(ceph_monc_do_statfs); |
e56fa10e | 645 | |
513a8243 ID |
646 | static void handle_get_version_reply(struct ceph_mon_client *monc, |
647 | struct ceph_msg *msg) | |
648 | { | |
649 | struct ceph_mon_generic_request *req; | |
650 | u64 tid = le64_to_cpu(msg->hdr.tid); | |
651 | void *p = msg->front.iov_base; | |
652 | void *end = p + msg->front_alloc_len; | |
653 | u64 handle; | |
654 | ||
655 | dout("%s %p tid %llu\n", __func__, msg, tid); | |
656 | ||
657 | ceph_decode_need(&p, end, 2*sizeof(u64), bad); | |
658 | handle = ceph_decode_64(&p); | |
659 | if (tid != 0 && tid != handle) | |
660 | goto bad; | |
661 | ||
662 | mutex_lock(&monc->mutex); | |
663 | req = __lookup_generic_req(monc, handle); | |
664 | if (req) { | |
665 | *(u64 *)req->buf = ceph_decode_64(&p); | |
666 | req->result = 0; | |
667 | get_generic_request(req); | |
668 | } | |
669 | mutex_unlock(&monc->mutex); | |
670 | if (req) { | |
671 | complete_all(&req->completion); | |
672 | put_generic_request(req); | |
673 | } | |
674 | ||
675 | return; | |
676 | bad: | |
7a6fdeb2 | 677 | pr_err("corrupt mon_get_version reply, tid %llu\n", tid); |
513a8243 ID |
678 | ceph_msg_dump(msg); |
679 | } | |
680 | ||
681 | /* | |
682 | * Send MMonGetVersion and wait for the reply. | |
683 | * | |
684 | * @what: one of "mdsmap", "osdmap" or "monmap" | |
685 | */ | |
686 | int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what, | |
687 | u64 *newest) | |
688 | { | |
689 | struct ceph_mon_generic_request *req; | |
690 | void *p, *end; | |
691 | u64 tid; | |
692 | int err; | |
693 | ||
694 | req = kzalloc(sizeof(*req), GFP_NOFS); | |
695 | if (!req) | |
696 | return -ENOMEM; | |
697 | ||
698 | kref_init(&req->kref); | |
699 | req->buf = newest; | |
513a8243 ID |
700 | init_completion(&req->completion); |
701 | ||
702 | req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION, | |
703 | sizeof(u64) + sizeof(u32) + strlen(what), | |
704 | GFP_NOFS, true); | |
705 | if (!req->request) { | |
706 | err = -ENOMEM; | |
707 | goto out; | |
708 | } | |
709 | ||
710 | req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024, | |
711 | GFP_NOFS, true); | |
712 | if (!req->reply) { | |
713 | err = -ENOMEM; | |
714 | goto out; | |
715 | } | |
716 | ||
717 | p = req->request->front.iov_base; | |
718 | end = p + req->request->front_alloc_len; | |
719 | ||
720 | /* fill out request */ | |
721 | mutex_lock(&monc->mutex); | |
722 | tid = ++monc->last_tid; | |
723 | ceph_encode_64(&p, tid); /* handle */ | |
724 | ceph_encode_string(&p, end, what, strlen(what)); | |
725 | ||
726 | err = __do_generic_request(monc, tid, req); | |
727 | ||
728 | mutex_unlock(&monc->mutex); | |
729 | out: | |
f646912d | 730 | put_generic_request(req); |
513a8243 ID |
731 | return err; |
732 | } | |
733 | EXPORT_SYMBOL(ceph_monc_do_get_version); | |
734 | ||
ba75bb98 | 735 | /* |
e56fa10e | 736 | * Resend pending generic requests. |
ba75bb98 | 737 | */ |
f8c76f6f | 738 | static void __resend_generic_request(struct ceph_mon_client *monc) |
ba75bb98 | 739 | { |
f8c76f6f | 740 | struct ceph_mon_generic_request *req; |
85ff03f6 | 741 | struct rb_node *p; |
ba75bb98 | 742 | |
f8c76f6f YS |
743 | for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) { |
744 | req = rb_entry(p, struct ceph_mon_generic_request, node); | |
6740a845 | 745 | ceph_msg_revoke(req->request); |
4f471e4a | 746 | ceph_msg_revoke_incoming(req->reply); |
67130934 | 747 | ceph_con_send(&monc->con, ceph_msg_get(req->request)); |
ba75bb98 SW |
748 | } |
749 | } | |
750 | ||
751 | /* | |
752 | * Delayed work. If we haven't mounted yet, retry. Otherwise, | |
753 | * renew/retry subscription as needed (in case it is timing out, or we | |
754 | * got an ENOMEM). And keep the monitor connection alive. | |
755 | */ | |
756 | static void delayed_work(struct work_struct *work) | |
757 | { | |
758 | struct ceph_mon_client *monc = | |
759 | container_of(work, struct ceph_mon_client, delayed_work.work); | |
760 | ||
761 | dout("monc delayed_work\n"); | |
762 | mutex_lock(&monc->mutex); | |
4e7a5dcd SW |
763 | if (monc->hunting) { |
764 | __close_session(monc); | |
765 | __open_session(monc); /* continue hunting */ | |
ba75bb98 | 766 | } else { |
8b9558aa YZ |
767 | struct ceph_options *opt = monc->client->options; |
768 | int is_auth = ceph_auth_is_authenticated(monc->auth); | |
769 | if (ceph_con_keepalive_expired(&monc->con, | |
770 | opt->monc_ping_timeout)) { | |
771 | dout("monc keepalive timeout\n"); | |
772 | is_auth = 0; | |
773 | __close_session(monc); | |
774 | monc->hunting = true; | |
775 | __open_session(monc); | |
776 | } | |
9bd2e6f8 | 777 | |
8b9558aa YZ |
778 | if (!monc->hunting) { |
779 | ceph_con_keepalive(&monc->con); | |
780 | __validate_auth(monc); | |
781 | } | |
9bd2e6f8 | 782 | |
82dcabad ID |
783 | if (is_auth) { |
784 | unsigned long now = jiffies; | |
785 | ||
786 | dout("%s renew subs? now %lu renew after %lu\n", | |
787 | __func__, now, monc->sub_renew_after); | |
788 | if (time_after_eq(now, monc->sub_renew_after)) | |
789 | __send_subscribe(monc); | |
790 | } | |
ba75bb98 | 791 | } |
ba75bb98 SW |
792 | __schedule_delayed(monc); |
793 | mutex_unlock(&monc->mutex); | |
794 | } | |
795 | ||
6b805185 SW |
796 | /* |
797 | * On startup, we build a temporary monmap populated with the IPs | |
798 | * provided by mount(2). | |
799 | */ | |
800 | static int build_initial_monmap(struct ceph_mon_client *monc) | |
801 | { | |
3d14c5d2 YS |
802 | struct ceph_options *opt = monc->client->options; |
803 | struct ceph_entity_addr *mon_addr = opt->mon_addr; | |
804 | int num_mon = opt->num_mon; | |
6b805185 SW |
805 | int i; |
806 | ||
807 | /* build initial monmap */ | |
808 | monc->monmap = kzalloc(sizeof(*monc->monmap) + | |
809 | num_mon*sizeof(monc->monmap->mon_inst[0]), | |
810 | GFP_KERNEL); | |
811 | if (!monc->monmap) | |
812 | return -ENOMEM; | |
813 | for (i = 0; i < num_mon; i++) { | |
814 | monc->monmap->mon_inst[i].addr = mon_addr[i]; | |
6b805185 SW |
815 | monc->monmap->mon_inst[i].addr.nonce = 0; |
816 | monc->monmap->mon_inst[i].name.type = | |
817 | CEPH_ENTITY_TYPE_MON; | |
818 | monc->monmap->mon_inst[i].name.num = cpu_to_le64(i); | |
819 | } | |
820 | monc->monmap->num_mon = num_mon; | |
6b805185 SW |
821 | return 0; |
822 | } | |
823 | ||
ba75bb98 SW |
824 | int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl) |
825 | { | |
826 | int err = 0; | |
827 | ||
828 | dout("init\n"); | |
829 | memset(monc, 0, sizeof(*monc)); | |
830 | monc->client = cl; | |
831 | monc->monmap = NULL; | |
832 | mutex_init(&monc->mutex); | |
833 | ||
6b805185 SW |
834 | err = build_initial_monmap(monc); |
835 | if (err) | |
836 | goto out; | |
837 | ||
f6a2f5be | 838 | /* connection */ |
4e7a5dcd | 839 | /* authentication */ |
3d14c5d2 | 840 | monc->auth = ceph_auth_init(cl->options->name, |
8323c3aa | 841 | cl->options->key); |
49d9224c NW |
842 | if (IS_ERR(monc->auth)) { |
843 | err = PTR_ERR(monc->auth); | |
67130934 | 844 | goto out_monmap; |
49d9224c | 845 | } |
4e7a5dcd SW |
846 | monc->auth->want_keys = |
847 | CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON | | |
848 | CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS; | |
849 | ||
240ed68e | 850 | /* msgs */ |
a79832f2 | 851 | err = -ENOMEM; |
7c315c55 | 852 | monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK, |
34d23762 | 853 | sizeof(struct ceph_mon_subscribe_ack), |
b61c2763 | 854 | GFP_NOFS, true); |
a79832f2 | 855 | if (!monc->m_subscribe_ack) |
49d9224c | 856 | goto out_auth; |
6694d6b9 | 857 | |
b61c2763 SW |
858 | monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS, |
859 | true); | |
240ed68e SW |
860 | if (!monc->m_subscribe) |
861 | goto out_subscribe_ack; | |
862 | ||
b61c2763 SW |
863 | monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS, |
864 | true); | |
a79832f2 | 865 | if (!monc->m_auth_reply) |
240ed68e | 866 | goto out_subscribe; |
4e7a5dcd | 867 | |
b61c2763 | 868 | monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true); |
9bd2e6f8 | 869 | monc->pending_auth = 0; |
a79832f2 | 870 | if (!monc->m_auth) |
6694d6b9 | 871 | goto out_auth_reply; |
ba75bb98 | 872 | |
735a72ef SW |
873 | ceph_con_init(&monc->con, monc, &mon_con_ops, |
874 | &monc->client->msgr); | |
875 | ||
ba75bb98 | 876 | monc->cur_mon = -1; |
4e7a5dcd | 877 | monc->hunting = true; |
ba75bb98 | 878 | monc->sub_renew_after = jiffies; |
82dcabad | 879 | monc->sub_renew_sent = 0; |
ba75bb98 SW |
880 | |
881 | INIT_DELAYED_WORK(&monc->delayed_work, delayed_work); | |
f8c76f6f YS |
882 | monc->generic_request_tree = RB_ROOT; |
883 | monc->num_generic_requests = 0; | |
ba75bb98 SW |
884 | monc->last_tid = 0; |
885 | ||
4e7a5dcd SW |
886 | return 0; |
887 | ||
6694d6b9 SW |
888 | out_auth_reply: |
889 | ceph_msg_put(monc->m_auth_reply); | |
240ed68e SW |
890 | out_subscribe: |
891 | ceph_msg_put(monc->m_subscribe); | |
7c315c55 SW |
892 | out_subscribe_ack: |
893 | ceph_msg_put(monc->m_subscribe_ack); | |
49d9224c NW |
894 | out_auth: |
895 | ceph_auth_destroy(monc->auth); | |
4e7a5dcd SW |
896 | out_monmap: |
897 | kfree(monc->monmap); | |
ba75bb98 SW |
898 | out: |
899 | return err; | |
900 | } | |
3d14c5d2 | 901 | EXPORT_SYMBOL(ceph_monc_init); |
ba75bb98 SW |
902 | |
903 | void ceph_monc_stop(struct ceph_mon_client *monc) | |
904 | { | |
905 | dout("stop\n"); | |
906 | cancel_delayed_work_sync(&monc->delayed_work); | |
907 | ||
908 | mutex_lock(&monc->mutex); | |
909 | __close_session(monc); | |
f6a2f5be | 910 | |
ba75bb98 SW |
911 | mutex_unlock(&monc->mutex); |
912 | ||
f3dea7ed SW |
913 | /* |
914 | * flush msgr queue before we destroy ourselves to ensure that: | |
915 | * - any work that references our embedded con is finished. | |
916 | * - any osd_client or other work that may reference an authorizer | |
917 | * finishes before we shut down the auth subsystem. | |
918 | */ | |
919 | ceph_msgr_flush(); | |
920 | ||
4e7a5dcd SW |
921 | ceph_auth_destroy(monc->auth); |
922 | ||
923 | ceph_msg_put(monc->m_auth); | |
6694d6b9 | 924 | ceph_msg_put(monc->m_auth_reply); |
240ed68e | 925 | ceph_msg_put(monc->m_subscribe); |
7c315c55 | 926 | ceph_msg_put(monc->m_subscribe_ack); |
ba75bb98 SW |
927 | |
928 | kfree(monc->monmap); | |
929 | } | |
3d14c5d2 | 930 | EXPORT_SYMBOL(ceph_monc_stop); |
ba75bb98 | 931 | |
0f9af169 ID |
932 | static void finish_hunting(struct ceph_mon_client *monc) |
933 | { | |
934 | if (monc->hunting) { | |
935 | dout("%s found mon%d\n", __func__, monc->cur_mon); | |
936 | monc->hunting = false; | |
937 | } | |
938 | } | |
939 | ||
4e7a5dcd SW |
940 | static void handle_auth_reply(struct ceph_mon_client *monc, |
941 | struct ceph_msg *msg) | |
942 | { | |
943 | int ret; | |
09c4d6a7 | 944 | int was_auth = 0; |
4e7a5dcd SW |
945 | |
946 | mutex_lock(&monc->mutex); | |
27859f97 | 947 | was_auth = ceph_auth_is_authenticated(monc->auth); |
9bd2e6f8 | 948 | monc->pending_auth = 0; |
4e7a5dcd SW |
949 | ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base, |
950 | msg->front.iov_len, | |
951 | monc->m_auth->front.iov_base, | |
3cea4c30 | 952 | monc->m_auth->front_alloc_len); |
0f9af169 ID |
953 | if (ret > 0) { |
954 | __send_prepared_auth_request(monc, ret); | |
955 | goto out; | |
956 | } | |
957 | ||
958 | finish_hunting(monc); | |
959 | ||
4e7a5dcd | 960 | if (ret < 0) { |
9bd2e6f8 | 961 | monc->client->auth_err = ret; |
27859f97 | 962 | } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) { |
4e7a5dcd | 963 | dout("authenticated, starting session\n"); |
0743304d | 964 | |
15d9882c AE |
965 | monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT; |
966 | monc->client->msgr.inst.name.num = | |
0cf5537b | 967 | cpu_to_le64(monc->auth->global_id); |
0743304d | 968 | |
4e7a5dcd | 969 | __send_subscribe(monc); |
f8c76f6f | 970 | __resend_generic_request(monc); |
0f9af169 ID |
971 | |
972 | pr_info("mon%d %s session established\n", monc->cur_mon, | |
973 | ceph_pr_addr(&monc->con.peer_addr.in_addr)); | |
4e7a5dcd | 974 | } |
0f9af169 ID |
975 | |
976 | out: | |
4e7a5dcd | 977 | mutex_unlock(&monc->mutex); |
0f9af169 ID |
978 | if (monc->client->auth_err < 0) |
979 | wake_up_all(&monc->client->auth_wq); | |
4e7a5dcd SW |
980 | } |
981 | ||
9bd2e6f8 SW |
982 | static int __validate_auth(struct ceph_mon_client *monc) |
983 | { | |
984 | int ret; | |
985 | ||
986 | if (monc->pending_auth) | |
987 | return 0; | |
988 | ||
989 | ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base, | |
3cea4c30 | 990 | monc->m_auth->front_alloc_len); |
9bd2e6f8 SW |
991 | if (ret <= 0) |
992 | return ret; /* either an error, or no need to authenticate */ | |
993 | __send_prepared_auth_request(monc, ret); | |
994 | return 0; | |
995 | } | |
996 | ||
997 | int ceph_monc_validate_auth(struct ceph_mon_client *monc) | |
998 | { | |
999 | int ret; | |
1000 | ||
1001 | mutex_lock(&monc->mutex); | |
1002 | ret = __validate_auth(monc); | |
1003 | mutex_unlock(&monc->mutex); | |
1004 | return ret; | |
1005 | } | |
3d14c5d2 | 1006 | EXPORT_SYMBOL(ceph_monc_validate_auth); |
9bd2e6f8 | 1007 | |
ba75bb98 SW |
1008 | /* |
1009 | * handle incoming message | |
1010 | */ | |
1011 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
1012 | { | |
1013 | struct ceph_mon_client *monc = con->private; | |
1014 | int type = le16_to_cpu(msg->hdr.type); | |
1015 | ||
1016 | if (!monc) | |
1017 | return; | |
1018 | ||
1019 | switch (type) { | |
4e7a5dcd SW |
1020 | case CEPH_MSG_AUTH_REPLY: |
1021 | handle_auth_reply(monc, msg); | |
ba75bb98 SW |
1022 | break; |
1023 | ||
1024 | case CEPH_MSG_MON_SUBSCRIBE_ACK: | |
1025 | handle_subscribe_ack(monc, msg); | |
1026 | break; | |
1027 | ||
1028 | case CEPH_MSG_STATFS_REPLY: | |
1029 | handle_statfs_reply(monc, msg); | |
1030 | break; | |
1031 | ||
513a8243 ID |
1032 | case CEPH_MSG_MON_GET_VERSION_REPLY: |
1033 | handle_get_version_reply(monc, msg); | |
1034 | break; | |
1035 | ||
4e7a5dcd SW |
1036 | case CEPH_MSG_MON_MAP: |
1037 | ceph_monc_handle_map(monc, msg); | |
1038 | break; | |
1039 | ||
ba75bb98 SW |
1040 | case CEPH_MSG_OSD_MAP: |
1041 | ceph_osdc_handle_map(&monc->client->osdc, msg); | |
1042 | break; | |
1043 | ||
1044 | default: | |
3d14c5d2 YS |
1045 | /* can the chained handler handle it? */ |
1046 | if (monc->client->extra_mon_dispatch && | |
1047 | monc->client->extra_mon_dispatch(monc->client, msg) == 0) | |
1048 | break; | |
1049 | ||
ba75bb98 SW |
1050 | pr_err("received unknown message type %d %s\n", type, |
1051 | ceph_msg_type_name(type)); | |
1052 | } | |
1053 | ceph_msg_put(msg); | |
1054 | } | |
1055 | ||
1056 | /* | |
1057 | * Allocate memory for incoming message | |
1058 | */ | |
1059 | static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con, | |
2450418c YS |
1060 | struct ceph_msg_header *hdr, |
1061 | int *skip) | |
ba75bb98 SW |
1062 | { |
1063 | struct ceph_mon_client *monc = con->private; | |
1064 | int type = le16_to_cpu(hdr->type); | |
2450418c | 1065 | int front_len = le32_to_cpu(hdr->front_len); |
5b3a4db3 | 1066 | struct ceph_msg *m = NULL; |
ba75bb98 | 1067 | |
2450418c | 1068 | *skip = 0; |
0547a9b3 | 1069 | |
ba75bb98 | 1070 | switch (type) { |
ba75bb98 | 1071 | case CEPH_MSG_MON_SUBSCRIBE_ACK: |
7c315c55 | 1072 | m = ceph_msg_get(monc->m_subscribe_ack); |
2450418c | 1073 | break; |
ba75bb98 | 1074 | case CEPH_MSG_STATFS_REPLY: |
f8c76f6f | 1075 | return get_generic_reply(con, hdr, skip); |
4e7a5dcd | 1076 | case CEPH_MSG_AUTH_REPLY: |
6694d6b9 | 1077 | m = ceph_msg_get(monc->m_auth_reply); |
2450418c | 1078 | break; |
513a8243 ID |
1079 | case CEPH_MSG_MON_GET_VERSION_REPLY: |
1080 | if (le64_to_cpu(hdr->tid) != 0) | |
1081 | return get_generic_reply(con, hdr, skip); | |
1082 | ||
1083 | /* | |
1084 | * Older OSDs don't set reply tid even if the orignal | |
1085 | * request had a non-zero tid. Workaround this weirdness | |
1086 | * by falling through to the allocate case. | |
1087 | */ | |
5b3a4db3 SW |
1088 | case CEPH_MSG_MON_MAP: |
1089 | case CEPH_MSG_MDS_MAP: | |
1090 | case CEPH_MSG_OSD_MAP: | |
b61c2763 | 1091 | m = ceph_msg_new(type, front_len, GFP_NOFS, false); |
1c20f2d2 AE |
1092 | if (!m) |
1093 | return NULL; /* ENOMEM--return skip == 0 */ | |
5b3a4db3 | 1094 | break; |
ba75bb98 | 1095 | } |
2450418c | 1096 | |
5b3a4db3 SW |
1097 | if (!m) { |
1098 | pr_info("alloc_msg unknown type %d\n", type); | |
2450418c | 1099 | *skip = 1; |
73c3d481 | 1100 | } else if (front_len > m->front_alloc_len) { |
b9a67899 JP |
1101 | pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n", |
1102 | front_len, m->front_alloc_len, | |
1103 | (unsigned int)con->peer_name.type, | |
1104 | le64_to_cpu(con->peer_name.num)); | |
73c3d481 SW |
1105 | ceph_msg_put(m); |
1106 | m = ceph_msg_new(type, front_len, GFP_NOFS, false); | |
5b3a4db3 | 1107 | } |
73c3d481 | 1108 | |
2450418c | 1109 | return m; |
ba75bb98 SW |
1110 | } |
1111 | ||
1112 | /* | |
1113 | * If the monitor connection resets, pick a new monitor and resubmit | |
1114 | * any pending requests. | |
1115 | */ | |
1116 | static void mon_fault(struct ceph_connection *con) | |
1117 | { | |
1118 | struct ceph_mon_client *monc = con->private; | |
1119 | ||
1120 | if (!monc) | |
1121 | return; | |
1122 | ||
1123 | dout("mon_fault\n"); | |
1124 | mutex_lock(&monc->mutex); | |
1125 | if (!con->private) | |
1126 | goto out; | |
1127 | ||
f6a2f5be | 1128 | if (!monc->hunting) |
ba75bb98 SW |
1129 | pr_info("mon%d %s session lost, " |
1130 | "hunting for new mon\n", monc->cur_mon, | |
67130934 | 1131 | ceph_pr_addr(&monc->con.peer_addr.in_addr)); |
ba75bb98 SW |
1132 | |
1133 | __close_session(monc); | |
1134 | if (!monc->hunting) { | |
1135 | /* start hunting */ | |
1136 | monc->hunting = true; | |
4e7a5dcd | 1137 | __open_session(monc); |
ba75bb98 SW |
1138 | } else { |
1139 | /* already hunting, let's wait a bit */ | |
1140 | __schedule_delayed(monc); | |
1141 | } | |
1142 | out: | |
1143 | mutex_unlock(&monc->mutex); | |
1144 | } | |
1145 | ||
ec87ef43 SW |
1146 | /* |
1147 | * We can ignore refcounting on the connection struct, as all references | |
1148 | * will come from the messenger workqueue, which is drained prior to | |
1149 | * mon_client destruction. | |
1150 | */ | |
1151 | static struct ceph_connection *con_get(struct ceph_connection *con) | |
1152 | { | |
1153 | return con; | |
1154 | } | |
1155 | ||
1156 | static void con_put(struct ceph_connection *con) | |
1157 | { | |
1158 | } | |
1159 | ||
9e32789f | 1160 | static const struct ceph_connection_operations mon_con_ops = { |
ec87ef43 SW |
1161 | .get = con_get, |
1162 | .put = con_put, | |
ba75bb98 SW |
1163 | .dispatch = dispatch, |
1164 | .fault = mon_fault, | |
1165 | .alloc_msg = mon_alloc_msg, | |
ba75bb98 | 1166 | }; |