]>
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 | ||
220abf5a | 9 | #include <linux/ceph/ceph_features.h> |
3d14c5d2 YS |
10 | #include <linux/ceph/mon_client.h> |
11 | #include <linux/ceph/libceph.h> | |
ab434b60 | 12 | #include <linux/ceph/debugfs.h> |
3d14c5d2 | 13 | #include <linux/ceph/decode.h> |
3d14c5d2 | 14 | #include <linux/ceph/auth.h> |
ba75bb98 SW |
15 | |
16 | /* | |
17 | * Interact with Ceph monitor cluster. Handle requests for new map | |
18 | * versions, and periodically resend as needed. Also implement | |
19 | * statfs() and umount(). | |
20 | * | |
21 | * A small cluster of Ceph "monitors" are responsible for managing critical | |
22 | * cluster configuration and state information. An odd number (e.g., 3, 5) | |
23 | * of cmon daemons use a modified version of the Paxos part-time parliament | |
24 | * algorithm to manage the MDS map (mds cluster membership), OSD map, and | |
25 | * list of clients who have mounted the file system. | |
26 | * | |
27 | * We maintain an open, active session with a monitor at all times in order to | |
28 | * receive timely MDSMap updates. We periodically send a keepalive byte on the | |
29 | * TCP socket to ensure we detect a failure. If the connection does break, we | |
30 | * randomly hunt for a new monitor. Once the connection is reestablished, we | |
31 | * resend any outstanding requests. | |
32 | */ | |
33 | ||
9e32789f | 34 | static const struct ceph_connection_operations mon_con_ops; |
ba75bb98 | 35 | |
9bd2e6f8 SW |
36 | static int __validate_auth(struct ceph_mon_client *monc); |
37 | ||
ba75bb98 SW |
38 | /* |
39 | * Decode a monmap blob (e.g., during mount). | |
40 | */ | |
41 | struct ceph_monmap *ceph_monmap_decode(void *p, void *end) | |
42 | { | |
43 | struct ceph_monmap *m = NULL; | |
44 | int i, err = -EINVAL; | |
45 | struct ceph_fsid fsid; | |
46 | u32 epoch, num_mon; | |
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)); | |
f3b4e55d | 53 | p += sizeof(u16); /* skip version */ |
ba75bb98 SW |
54 | |
55 | ceph_decode_need(&p, end, sizeof(fsid) + 2*sizeof(u32), bad); | |
56 | ceph_decode_copy(&p, &fsid, sizeof(fsid)); | |
c89136ea | 57 | epoch = ceph_decode_32(&p); |
ba75bb98 | 58 | |
c89136ea | 59 | num_mon = ceph_decode_32(&p); |
ba75bb98 SW |
60 | ceph_decode_need(&p, end, num_mon*sizeof(m->mon_inst[0]), bad); |
61 | ||
62 | if (num_mon >= CEPH_MAX_MON) | |
63 | goto bad; | |
64 | m = kmalloc(sizeof(*m) + sizeof(m->mon_inst[0])*num_mon, GFP_NOFS); | |
65 | if (m == NULL) | |
66 | return ERR_PTR(-ENOMEM); | |
67 | m->fsid = fsid; | |
68 | m->epoch = epoch; | |
69 | m->num_mon = num_mon; | |
70 | ceph_decode_copy(&p, m->mon_inst, num_mon*sizeof(m->mon_inst[0])); | |
63f2d211 SW |
71 | for (i = 0; i < num_mon; i++) |
72 | ceph_decode_addr(&m->mon_inst[i].addr); | |
ba75bb98 | 73 | |
ba75bb98 SW |
74 | dout("monmap_decode epoch %d, num_mon %d\n", m->epoch, |
75 | m->num_mon); | |
76 | for (i = 0; i < m->num_mon; i++) | |
77 | dout("monmap_decode mon%d is %s\n", i, | |
3d14c5d2 | 78 | ceph_pr_addr(&m->mon_inst[i].addr.in_addr)); |
ba75bb98 SW |
79 | return m; |
80 | ||
81 | bad: | |
82 | dout("monmap_decode failed with %d\n", err); | |
83 | kfree(m); | |
84 | return ERR_PTR(err); | |
85 | } | |
86 | ||
87 | /* | |
88 | * return true if *addr is included in the monmap. | |
89 | */ | |
90 | int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr) | |
91 | { | |
92 | int i; | |
93 | ||
94 | for (i = 0; i < m->num_mon; i++) | |
103e2d3a | 95 | if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0) |
ba75bb98 SW |
96 | return 1; |
97 | return 0; | |
98 | } | |
99 | ||
5ce6e9db SW |
100 | /* |
101 | * Send an auth request. | |
102 | */ | |
103 | static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len) | |
104 | { | |
105 | monc->pending_auth = 1; | |
106 | monc->m_auth->front.iov_len = len; | |
107 | monc->m_auth->hdr.front_len = cpu_to_le32(len); | |
6740a845 | 108 | ceph_msg_revoke(monc->m_auth); |
5ce6e9db | 109 | ceph_msg_get(monc->m_auth); /* keep our ref */ |
67130934 | 110 | ceph_con_send(&monc->con, monc->m_auth); |
5ce6e9db SW |
111 | } |
112 | ||
ba75bb98 SW |
113 | /* |
114 | * Close monitor session, if any. | |
115 | */ | |
116 | static void __close_session(struct ceph_mon_client *monc) | |
117 | { | |
f6a2f5be | 118 | dout("__close_session closing mon%d\n", monc->cur_mon); |
6740a845 | 119 | ceph_msg_revoke(monc->m_auth); |
4f471e4a SW |
120 | ceph_msg_revoke_incoming(monc->m_auth_reply); |
121 | ceph_msg_revoke(monc->m_subscribe); | |
122 | ceph_msg_revoke_incoming(monc->m_subscribe_ack); | |
67130934 | 123 | ceph_con_close(&monc->con); |
0e04dc26 | 124 | |
f6a2f5be SW |
125 | monc->pending_auth = 0; |
126 | ceph_auth_reset(monc->auth); | |
ba75bb98 SW |
127 | } |
128 | ||
129 | /* | |
0e04dc26 ID |
130 | * Pick a new monitor at random and set cur_mon. If we are repicking |
131 | * (i.e. cur_mon is already set), be sure to pick a different one. | |
ba75bb98 | 132 | */ |
0e04dc26 | 133 | static void pick_new_mon(struct ceph_mon_client *monc) |
ba75bb98 | 134 | { |
0e04dc26 | 135 | int old_mon = monc->cur_mon; |
ba75bb98 | 136 | |
0e04dc26 | 137 | BUG_ON(monc->monmap->num_mon < 1); |
ba75bb98 | 138 | |
0e04dc26 ID |
139 | if (monc->monmap->num_mon == 1) { |
140 | monc->cur_mon = 0; | |
141 | } else { | |
142 | int max = monc->monmap->num_mon; | |
143 | int o = -1; | |
144 | int n; | |
145 | ||
146 | if (monc->cur_mon >= 0) { | |
147 | if (monc->cur_mon < monc->monmap->num_mon) | |
148 | o = monc->cur_mon; | |
149 | if (o >= 0) | |
150 | max--; | |
151 | } | |
4e7a5dcd | 152 | |
0e04dc26 ID |
153 | n = prandom_u32() % max; |
154 | if (o >= 0 && n >= o) | |
155 | n++; | |
8b9558aa | 156 | |
0e04dc26 | 157 | monc->cur_mon = n; |
ba75bb98 | 158 | } |
0e04dc26 ID |
159 | |
160 | dout("%s mon%d -> mon%d out of %d mons\n", __func__, old_mon, | |
161 | monc->cur_mon, monc->monmap->num_mon); | |
162 | } | |
163 | ||
164 | /* | |
165 | * Open a session with a new monitor. | |
166 | */ | |
167 | static void __open_session(struct ceph_mon_client *monc) | |
168 | { | |
169 | int ret; | |
170 | ||
171 | pick_new_mon(monc); | |
172 | ||
1752b50c | 173 | monc->hunting = true; |
168b9090 ID |
174 | if (monc->had_a_connection) { |
175 | monc->hunt_mult *= CEPH_MONC_HUNT_BACKOFF; | |
176 | if (monc->hunt_mult > CEPH_MONC_HUNT_MAX_MULT) | |
177 | monc->hunt_mult = CEPH_MONC_HUNT_MAX_MULT; | |
178 | } | |
179 | ||
0e04dc26 ID |
180 | monc->sub_renew_after = jiffies; /* i.e., expired */ |
181 | monc->sub_renew_sent = 0; | |
182 | ||
183 | dout("%s opening mon%d\n", __func__, monc->cur_mon); | |
184 | ceph_con_open(&monc->con, CEPH_ENTITY_TYPE_MON, monc->cur_mon, | |
185 | &monc->monmap->mon_inst[monc->cur_mon].addr); | |
186 | ||
187 | /* | |
188 | * send an initial keepalive to ensure our timestamp is valid | |
189 | * by the time we are in an OPENED state | |
190 | */ | |
191 | ceph_con_keepalive(&monc->con); | |
192 | ||
193 | /* initiate authentication handshake */ | |
194 | ret = ceph_auth_build_hello(monc->auth, | |
195 | monc->m_auth->front.iov_base, | |
196 | monc->m_auth->front_alloc_len); | |
197 | BUG_ON(ret <= 0); | |
198 | __send_prepared_auth_request(monc, ret); | |
ba75bb98 SW |
199 | } |
200 | ||
1752b50c ID |
201 | static void reopen_session(struct ceph_mon_client *monc) |
202 | { | |
203 | if (!monc->hunting) | |
204 | pr_info("mon%d %s session lost, hunting for new mon\n", | |
205 | monc->cur_mon, ceph_pr_addr(&monc->con.peer_addr.in_addr)); | |
206 | ||
207 | __close_session(monc); | |
208 | __open_session(monc); | |
209 | } | |
210 | ||
ba75bb98 SW |
211 | /* |
212 | * Reschedule delayed work timer. | |
213 | */ | |
214 | static void __schedule_delayed(struct ceph_mon_client *monc) | |
215 | { | |
8b9558aa | 216 | unsigned long delay; |
ba75bb98 | 217 | |
168b9090 ID |
218 | if (monc->hunting) |
219 | delay = CEPH_MONC_HUNT_INTERVAL * monc->hunt_mult; | |
220 | else | |
58d81b12 | 221 | delay = CEPH_MONC_PING_INTERVAL; |
168b9090 | 222 | |
8b9558aa | 223 | dout("__schedule_delayed after %lu\n", delay); |
bee3a37c ID |
224 | mod_delayed_work(system_wq, &monc->delayed_work, |
225 | round_jiffies_relative(delay)); | |
ba75bb98 SW |
226 | } |
227 | ||
82dcabad | 228 | const char *ceph_sub_str[] = { |
82dcabad ID |
229 | [CEPH_SUB_MONMAP] = "monmap", |
230 | [CEPH_SUB_OSDMAP] = "osdmap", | |
0cabbd94 YZ |
231 | [CEPH_SUB_FSMAP] = "fsmap.user", |
232 | [CEPH_SUB_MDSMAP] = "mdsmap", | |
82dcabad ID |
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++) { | |
737cc81e ID |
263 | char buf[32]; |
264 | int len; | |
82dcabad ID |
265 | |
266 | if (!monc->subs[i].want) | |
267 | continue; | |
268 | ||
737cc81e ID |
269 | len = sprintf(buf, "%s", ceph_sub_str[i]); |
270 | if (i == CEPH_SUB_MDSMAP && | |
271 | monc->fs_cluster_id != CEPH_FS_CLUSTER_ID_NONE) | |
272 | len += sprintf(buf + len, ".%d", monc->fs_cluster_id); | |
273 | ||
274 | dout("%s %s start %llu flags 0x%x\n", __func__, buf, | |
82dcabad ID |
275 | le64_to_cpu(monc->subs[i].item.start), |
276 | monc->subs[i].item.flags); | |
737cc81e | 277 | ceph_encode_string(&p, end, buf, len); |
82dcabad ID |
278 | memcpy(p, &monc->subs[i].item, sizeof(monc->subs[i].item)); |
279 | p += sizeof(monc->subs[i].item); | |
ba75bb98 | 280 | } |
82dcabad | 281 | |
737cc81e | 282 | BUG_ON(p > end); |
82dcabad ID |
283 | msg->front.iov_len = p - msg->front.iov_base; |
284 | msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); | |
285 | ceph_msg_revoke(msg); | |
286 | ceph_con_send(&monc->con, ceph_msg_get(msg)); | |
ba75bb98 SW |
287 | } |
288 | ||
289 | static void handle_subscribe_ack(struct ceph_mon_client *monc, | |
290 | struct ceph_msg *msg) | |
291 | { | |
95c96174 | 292 | unsigned int seconds; |
07bd10fb SW |
293 | struct ceph_mon_subscribe_ack *h = msg->front.iov_base; |
294 | ||
295 | if (msg->front.iov_len < sizeof(*h)) | |
296 | goto bad; | |
297 | seconds = le32_to_cpu(h->duration); | |
ba75bb98 | 298 | |
ba75bb98 | 299 | mutex_lock(&monc->mutex); |
82dcabad | 300 | if (monc->sub_renew_sent) { |
220abf5a ID |
301 | /* |
302 | * This is only needed for legacy (infernalis or older) | |
303 | * MONs -- see delayed_work(). | |
304 | */ | |
82dcabad ID |
305 | monc->sub_renew_after = monc->sub_renew_sent + |
306 | (seconds >> 1) * HZ - 1; | |
307 | dout("%s sent %lu duration %d renew after %lu\n", __func__, | |
308 | monc->sub_renew_sent, seconds, monc->sub_renew_after); | |
309 | monc->sub_renew_sent = 0; | |
310 | } else { | |
311 | dout("%s sent %lu renew after %lu, ignoring\n", __func__, | |
312 | monc->sub_renew_sent, monc->sub_renew_after); | |
313 | } | |
ba75bb98 SW |
314 | mutex_unlock(&monc->mutex); |
315 | return; | |
316 | bad: | |
317 | pr_err("got corrupt subscribe-ack msg\n"); | |
9ec7cab1 | 318 | ceph_msg_dump(msg); |
ba75bb98 SW |
319 | } |
320 | ||
321 | /* | |
82dcabad ID |
322 | * Register interest in a map |
323 | * | |
324 | * @sub: one of CEPH_SUB_* | |
325 | * @epoch: X for "every map since X", or 0 for "just the latest" | |
ba75bb98 | 326 | */ |
82dcabad ID |
327 | static bool __ceph_monc_want_map(struct ceph_mon_client *monc, int sub, |
328 | u32 epoch, bool continuous) | |
ba75bb98 | 329 | { |
82dcabad ID |
330 | __le64 start = cpu_to_le64(epoch); |
331 | u8 flags = !continuous ? CEPH_SUBSCRIBE_ONETIME : 0; | |
332 | ||
333 | dout("%s %s epoch %u continuous %d\n", __func__, ceph_sub_str[sub], | |
334 | epoch, continuous); | |
335 | ||
336 | if (monc->subs[sub].want && | |
337 | monc->subs[sub].item.start == start && | |
338 | monc->subs[sub].item.flags == flags) | |
339 | return false; | |
340 | ||
341 | monc->subs[sub].item.start = start; | |
342 | monc->subs[sub].item.flags = flags; | |
343 | monc->subs[sub].want = true; | |
344 | ||
345 | return true; | |
346 | } | |
347 | ||
348 | bool ceph_monc_want_map(struct ceph_mon_client *monc, int sub, u32 epoch, | |
349 | bool continuous) | |
350 | { | |
351 | bool need_request; | |
352 | ||
ba75bb98 | 353 | mutex_lock(&monc->mutex); |
82dcabad | 354 | need_request = __ceph_monc_want_map(monc, sub, epoch, continuous); |
ba75bb98 | 355 | mutex_unlock(&monc->mutex); |
82dcabad ID |
356 | |
357 | return need_request; | |
358 | } | |
359 | EXPORT_SYMBOL(ceph_monc_want_map); | |
360 | ||
361 | /* | |
362 | * Keep track of which maps we have | |
363 | * | |
364 | * @sub: one of CEPH_SUB_* | |
365 | */ | |
366 | static void __ceph_monc_got_map(struct ceph_mon_client *monc, int sub, | |
367 | u32 epoch) | |
368 | { | |
369 | dout("%s %s epoch %u\n", __func__, ceph_sub_str[sub], epoch); | |
370 | ||
371 | if (monc->subs[sub].want) { | |
372 | if (monc->subs[sub].item.flags & CEPH_SUBSCRIBE_ONETIME) | |
373 | monc->subs[sub].want = false; | |
374 | else | |
375 | monc->subs[sub].item.start = cpu_to_le64(epoch + 1); | |
376 | } | |
377 | ||
378 | monc->subs[sub].have = epoch; | |
ba75bb98 SW |
379 | } |
380 | ||
82dcabad | 381 | void ceph_monc_got_map(struct ceph_mon_client *monc, int sub, u32 epoch) |
ba75bb98 SW |
382 | { |
383 | mutex_lock(&monc->mutex); | |
82dcabad | 384 | __ceph_monc_got_map(monc, sub, epoch); |
ba75bb98 | 385 | mutex_unlock(&monc->mutex); |
ba75bb98 | 386 | } |
82dcabad | 387 | EXPORT_SYMBOL(ceph_monc_got_map); |
ba75bb98 | 388 | |
42c1b124 ID |
389 | void ceph_monc_renew_subs(struct ceph_mon_client *monc) |
390 | { | |
391 | mutex_lock(&monc->mutex); | |
392 | __send_subscribe(monc); | |
393 | mutex_unlock(&monc->mutex); | |
394 | } | |
395 | EXPORT_SYMBOL(ceph_monc_renew_subs); | |
396 | ||
a319bf56 ID |
397 | /* |
398 | * Wait for an osdmap with a given epoch. | |
399 | * | |
400 | * @epoch: epoch to wait for | |
401 | * @timeout: in jiffies, 0 means "wait forever" | |
402 | */ | |
6044cde6 ID |
403 | int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch, |
404 | unsigned long timeout) | |
405 | { | |
406 | unsigned long started = jiffies; | |
216639dd | 407 | long ret; |
6044cde6 ID |
408 | |
409 | mutex_lock(&monc->mutex); | |
82dcabad | 410 | while (monc->subs[CEPH_SUB_OSDMAP].have < epoch) { |
6044cde6 ID |
411 | mutex_unlock(&monc->mutex); |
412 | ||
a319bf56 | 413 | if (timeout && time_after_eq(jiffies, started + timeout)) |
6044cde6 ID |
414 | return -ETIMEDOUT; |
415 | ||
416 | ret = wait_event_interruptible_timeout(monc->client->auth_wq, | |
82dcabad ID |
417 | monc->subs[CEPH_SUB_OSDMAP].have >= epoch, |
418 | ceph_timeout_jiffies(timeout)); | |
6044cde6 ID |
419 | if (ret < 0) |
420 | return ret; | |
421 | ||
422 | mutex_lock(&monc->mutex); | |
423 | } | |
424 | ||
425 | mutex_unlock(&monc->mutex); | |
426 | return 0; | |
427 | } | |
428 | EXPORT_SYMBOL(ceph_monc_wait_osdmap); | |
ba75bb98 | 429 | |
4e7a5dcd | 430 | /* |
82dcabad ID |
431 | * Open a session with a random monitor. Request monmap and osdmap, |
432 | * which are waited upon in __ceph_open_session(). | |
4e7a5dcd SW |
433 | */ |
434 | int ceph_monc_open_session(struct ceph_mon_client *monc) | |
ba75bb98 | 435 | { |
ba75bb98 | 436 | mutex_lock(&monc->mutex); |
82dcabad ID |
437 | __ceph_monc_want_map(monc, CEPH_SUB_MONMAP, 0, true); |
438 | __ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, 0, false); | |
4e7a5dcd | 439 | __open_session(monc); |
ba75bb98 SW |
440 | __schedule_delayed(monc); |
441 | mutex_unlock(&monc->mutex); | |
442 | return 0; | |
443 | } | |
3d14c5d2 | 444 | EXPORT_SYMBOL(ceph_monc_open_session); |
ba75bb98 | 445 | |
0743304d SW |
446 | static void ceph_monc_handle_map(struct ceph_mon_client *monc, |
447 | struct ceph_msg *msg) | |
4e7a5dcd SW |
448 | { |
449 | struct ceph_client *client = monc->client; | |
450 | struct ceph_monmap *monmap = NULL, *old = monc->monmap; | |
451 | void *p, *end; | |
452 | ||
453 | mutex_lock(&monc->mutex); | |
454 | ||
455 | dout("handle_monmap\n"); | |
456 | p = msg->front.iov_base; | |
457 | end = p + msg->front.iov_len; | |
458 | ||
459 | monmap = ceph_monmap_decode(p, end); | |
460 | if (IS_ERR(monmap)) { | |
461 | pr_err("problem decoding monmap, %d\n", | |
462 | (int)PTR_ERR(monmap)); | |
d4a780ce | 463 | goto out; |
4e7a5dcd | 464 | } |
0743304d SW |
465 | |
466 | if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) { | |
4e7a5dcd | 467 | kfree(monmap); |
d4a780ce | 468 | goto out; |
4e7a5dcd SW |
469 | } |
470 | ||
471 | client->monc.monmap = monmap; | |
4e7a5dcd SW |
472 | kfree(old); |
473 | ||
82dcabad | 474 | __ceph_monc_got_map(monc, CEPH_SUB_MONMAP, monc->monmap->epoch); |
02ac956c | 475 | client->have_fsid = true; |
d1c338a5 | 476 | |
d4a780ce | 477 | out: |
4e7a5dcd | 478 | mutex_unlock(&monc->mutex); |
03066f23 | 479 | wake_up_all(&client->auth_wq); |
4e7a5dcd SW |
480 | } |
481 | ||
ba75bb98 | 482 | /* |
7a6fdeb2 | 483 | * generic requests (currently statfs, mon_get_version) |
ba75bb98 | 484 | */ |
fcd00b68 | 485 | DEFINE_RB_FUNCS(generic_request, struct ceph_mon_generic_request, tid, node) |
85ff03f6 | 486 | |
f8c76f6f | 487 | static void release_generic_request(struct kref *kref) |
3143edd3 | 488 | { |
f8c76f6f YS |
489 | struct ceph_mon_generic_request *req = |
490 | container_of(kref, struct ceph_mon_generic_request, kref); | |
3143edd3 | 491 | |
d0b19705 ID |
492 | dout("%s greq %p request %p reply %p\n", __func__, req, req->request, |
493 | req->reply); | |
494 | WARN_ON(!RB_EMPTY_NODE(&req->node)); | |
495 | ||
3143edd3 SW |
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 | { |
d0b19705 ID |
506 | if (req) |
507 | kref_put(&req->kref, release_generic_request); | |
3143edd3 SW |
508 | } |
509 | ||
f8c76f6f | 510 | static void get_generic_request(struct ceph_mon_generic_request *req) |
3143edd3 SW |
511 | { |
512 | kref_get(&req->kref); | |
513 | } | |
514 | ||
d0b19705 ID |
515 | static struct ceph_mon_generic_request * |
516 | alloc_generic_request(struct ceph_mon_client *monc, gfp_t gfp) | |
517 | { | |
518 | struct ceph_mon_generic_request *req; | |
519 | ||
520 | req = kzalloc(sizeof(*req), gfp); | |
521 | if (!req) | |
522 | return NULL; | |
523 | ||
524 | req->monc = monc; | |
525 | kref_init(&req->kref); | |
526 | RB_CLEAR_NODE(&req->node); | |
527 | init_completion(&req->completion); | |
528 | ||
529 | dout("%s greq %p\n", __func__, req); | |
530 | return req; | |
531 | } | |
532 | ||
533 | static void register_generic_request(struct ceph_mon_generic_request *req) | |
534 | { | |
535 | struct ceph_mon_client *monc = req->monc; | |
536 | ||
537 | WARN_ON(req->tid); | |
538 | ||
539 | get_generic_request(req); | |
540 | req->tid = ++monc->last_tid; | |
541 | insert_generic_request(&monc->generic_request_tree, req); | |
542 | } | |
543 | ||
544 | static void send_generic_request(struct ceph_mon_client *monc, | |
545 | struct ceph_mon_generic_request *req) | |
546 | { | |
547 | WARN_ON(!req->tid); | |
548 | ||
549 | dout("%s greq %p tid %llu\n", __func__, req, req->tid); | |
550 | req->request->hdr.tid = cpu_to_le64(req->tid); | |
551 | ceph_con_send(&monc->con, ceph_msg_get(req->request)); | |
552 | } | |
553 | ||
554 | static void __finish_generic_request(struct ceph_mon_generic_request *req) | |
555 | { | |
556 | struct ceph_mon_client *monc = req->monc; | |
557 | ||
558 | dout("%s greq %p tid %llu\n", __func__, req, req->tid); | |
559 | erase_generic_request(&monc->generic_request_tree, req); | |
560 | ||
561 | ceph_msg_revoke(req->request); | |
562 | ceph_msg_revoke_incoming(req->reply); | |
563 | } | |
564 | ||
565 | static void finish_generic_request(struct ceph_mon_generic_request *req) | |
566 | { | |
567 | __finish_generic_request(req); | |
568 | put_generic_request(req); | |
569 | } | |
570 | ||
571 | static void complete_generic_request(struct ceph_mon_generic_request *req) | |
572 | { | |
573 | if (req->complete_cb) | |
574 | req->complete_cb(req); | |
575 | else | |
576 | complete_all(&req->completion); | |
577 | put_generic_request(req); | |
578 | } | |
579 | ||
f52ec33c | 580 | static void cancel_generic_request(struct ceph_mon_generic_request *req) |
d0b19705 ID |
581 | { |
582 | struct ceph_mon_client *monc = req->monc; | |
583 | struct ceph_mon_generic_request *lookup_req; | |
584 | ||
585 | dout("%s greq %p tid %llu\n", __func__, req, req->tid); | |
586 | ||
587 | mutex_lock(&monc->mutex); | |
588 | lookup_req = lookup_generic_request(&monc->generic_request_tree, | |
589 | req->tid); | |
590 | if (lookup_req) { | |
591 | WARN_ON(lookup_req != req); | |
592 | finish_generic_request(req); | |
593 | } | |
594 | ||
595 | mutex_unlock(&monc->mutex); | |
596 | } | |
597 | ||
598 | static int wait_generic_request(struct ceph_mon_generic_request *req) | |
599 | { | |
600 | int ret; | |
601 | ||
602 | dout("%s greq %p tid %llu\n", __func__, req, req->tid); | |
603 | ret = wait_for_completion_interruptible(&req->completion); | |
604 | if (ret) | |
605 | cancel_generic_request(req); | |
606 | else | |
607 | ret = req->result; /* completed */ | |
608 | ||
609 | return ret; | |
610 | } | |
611 | ||
f8c76f6f | 612 | static struct ceph_msg *get_generic_reply(struct ceph_connection *con, |
3143edd3 SW |
613 | struct ceph_msg_header *hdr, |
614 | int *skip) | |
615 | { | |
616 | struct ceph_mon_client *monc = con->private; | |
f8c76f6f | 617 | struct ceph_mon_generic_request *req; |
3143edd3 SW |
618 | u64 tid = le64_to_cpu(hdr->tid); |
619 | struct ceph_msg *m; | |
620 | ||
621 | mutex_lock(&monc->mutex); | |
fcd00b68 | 622 | req = lookup_generic_request(&monc->generic_request_tree, tid); |
3143edd3 | 623 | if (!req) { |
f8c76f6f | 624 | dout("get_generic_reply %lld dne\n", tid); |
3143edd3 SW |
625 | *skip = 1; |
626 | m = NULL; | |
627 | } else { | |
f8c76f6f | 628 | dout("get_generic_reply %lld got %p\n", tid, req->reply); |
1c20f2d2 | 629 | *skip = 0; |
3143edd3 SW |
630 | m = ceph_msg_get(req->reply); |
631 | /* | |
632 | * we don't need to track the connection reading into | |
633 | * this reply because we only have one open connection | |
634 | * at a time, ever. | |
635 | */ | |
636 | } | |
637 | mutex_unlock(&monc->mutex); | |
638 | return m; | |
639 | } | |
640 | ||
e56fa10e YS |
641 | /* |
642 | * statfs | |
643 | */ | |
ba75bb98 SW |
644 | static void handle_statfs_reply(struct ceph_mon_client *monc, |
645 | struct ceph_msg *msg) | |
646 | { | |
f8c76f6f | 647 | struct ceph_mon_generic_request *req; |
ba75bb98 | 648 | struct ceph_mon_statfs_reply *reply = msg->front.iov_base; |
3143edd3 | 649 | u64 tid = le64_to_cpu(msg->hdr.tid); |
ba75bb98 | 650 | |
d0b19705 ID |
651 | dout("%s msg %p tid %llu\n", __func__, msg, tid); |
652 | ||
ba75bb98 SW |
653 | if (msg->front.iov_len != sizeof(*reply)) |
654 | goto bad; | |
ba75bb98 SW |
655 | |
656 | mutex_lock(&monc->mutex); | |
fcd00b68 | 657 | req = lookup_generic_request(&monc->generic_request_tree, tid); |
d0b19705 ID |
658 | if (!req) { |
659 | mutex_unlock(&monc->mutex); | |
660 | return; | |
ba75bb98 | 661 | } |
d0b19705 ID |
662 | |
663 | req->result = 0; | |
664 | *req->u.st = reply->st; /* struct */ | |
665 | __finish_generic_request(req); | |
ba75bb98 | 666 | mutex_unlock(&monc->mutex); |
d0b19705 ID |
667 | |
668 | complete_generic_request(req); | |
ba75bb98 SW |
669 | return; |
670 | ||
671 | bad: | |
7a6fdeb2 | 672 | pr_err("corrupt statfs reply, tid %llu\n", tid); |
9ec7cab1 | 673 | ceph_msg_dump(msg); |
ba75bb98 SW |
674 | } |
675 | ||
676 | /* | |
3143edd3 | 677 | * Do a synchronous statfs(). |
ba75bb98 | 678 | */ |
3143edd3 | 679 | int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf) |
ba75bb98 | 680 | { |
f8c76f6f | 681 | struct ceph_mon_generic_request *req; |
ba75bb98 | 682 | struct ceph_mon_statfs *h; |
d0b19705 | 683 | int ret = -ENOMEM; |
3143edd3 | 684 | |
d0b19705 | 685 | req = alloc_generic_request(monc, GFP_NOFS); |
3143edd3 | 686 | if (!req) |
d0b19705 | 687 | goto out; |
ba75bb98 | 688 | |
b61c2763 SW |
689 | req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS, |
690 | true); | |
a79832f2 | 691 | if (!req->request) |
3143edd3 | 692 | goto out; |
d0b19705 ID |
693 | |
694 | req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 64, GFP_NOFS, true); | |
a79832f2 | 695 | if (!req->reply) |
3143edd3 | 696 | goto out; |
3143edd3 | 697 | |
d0b19705 ID |
698 | req->u.st = buf; |
699 | ||
700 | mutex_lock(&monc->mutex); | |
701 | register_generic_request(req); | |
3143edd3 SW |
702 | /* fill out request */ |
703 | h = req->request->front.iov_base; | |
13e38c8a SW |
704 | h->monhdr.have_version = 0; |
705 | h->monhdr.session_mon = cpu_to_le16(-1); | |
706 | h->monhdr.session_mon_tid = 0; | |
ba75bb98 | 707 | h->fsid = monc->monmap->fsid; |
d0b19705 ID |
708 | send_generic_request(monc, req); |
709 | mutex_unlock(&monc->mutex); | |
ba75bb98 | 710 | |
d0b19705 | 711 | ret = wait_generic_request(req); |
e56fa10e | 712 | out: |
f646912d | 713 | put_generic_request(req); |
d0b19705 | 714 | return ret; |
e56fa10e | 715 | } |
3d14c5d2 | 716 | EXPORT_SYMBOL(ceph_monc_do_statfs); |
e56fa10e | 717 | |
513a8243 ID |
718 | static void handle_get_version_reply(struct ceph_mon_client *monc, |
719 | struct ceph_msg *msg) | |
720 | { | |
721 | struct ceph_mon_generic_request *req; | |
722 | u64 tid = le64_to_cpu(msg->hdr.tid); | |
723 | void *p = msg->front.iov_base; | |
724 | void *end = p + msg->front_alloc_len; | |
725 | u64 handle; | |
726 | ||
d0b19705 | 727 | dout("%s msg %p tid %llu\n", __func__, msg, tid); |
513a8243 ID |
728 | |
729 | ceph_decode_need(&p, end, 2*sizeof(u64), bad); | |
730 | handle = ceph_decode_64(&p); | |
731 | if (tid != 0 && tid != handle) | |
732 | goto bad; | |
733 | ||
734 | mutex_lock(&monc->mutex); | |
fcd00b68 | 735 | req = lookup_generic_request(&monc->generic_request_tree, handle); |
d0b19705 ID |
736 | if (!req) { |
737 | mutex_unlock(&monc->mutex); | |
738 | return; | |
513a8243 | 739 | } |
d0b19705 ID |
740 | |
741 | req->result = 0; | |
742 | req->u.newest = ceph_decode_64(&p); | |
743 | __finish_generic_request(req); | |
513a8243 | 744 | mutex_unlock(&monc->mutex); |
513a8243 | 745 | |
d0b19705 | 746 | complete_generic_request(req); |
513a8243 | 747 | return; |
d0b19705 | 748 | |
513a8243 | 749 | bad: |
7a6fdeb2 | 750 | pr_err("corrupt mon_get_version reply, tid %llu\n", tid); |
513a8243 ID |
751 | ceph_msg_dump(msg); |
752 | } | |
753 | ||
d0b19705 ID |
754 | static struct ceph_mon_generic_request * |
755 | __ceph_monc_get_version(struct ceph_mon_client *monc, const char *what, | |
756 | ceph_monc_callback_t cb, u64 private_data) | |
513a8243 ID |
757 | { |
758 | struct ceph_mon_generic_request *req; | |
513a8243 | 759 | |
d0b19705 | 760 | req = alloc_generic_request(monc, GFP_NOIO); |
513a8243 | 761 | if (!req) |
d0b19705 | 762 | goto err_put_req; |
513a8243 ID |
763 | |
764 | req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION, | |
765 | sizeof(u64) + sizeof(u32) + strlen(what), | |
d0b19705 ID |
766 | GFP_NOIO, true); |
767 | if (!req->request) | |
768 | goto err_put_req; | |
513a8243 | 769 | |
d0b19705 ID |
770 | req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 32, GFP_NOIO, |
771 | true); | |
772 | if (!req->reply) | |
773 | goto err_put_req; | |
513a8243 | 774 | |
d0b19705 ID |
775 | req->complete_cb = cb; |
776 | req->private_data = private_data; | |
513a8243 | 777 | |
513a8243 | 778 | mutex_lock(&monc->mutex); |
d0b19705 ID |
779 | register_generic_request(req); |
780 | { | |
781 | void *p = req->request->front.iov_base; | |
782 | void *const end = p + req->request->front_alloc_len; | |
783 | ||
784 | ceph_encode_64(&p, req->tid); /* handle */ | |
785 | ceph_encode_string(&p, end, what, strlen(what)); | |
786 | WARN_ON(p != end); | |
787 | } | |
788 | send_generic_request(monc, req); | |
789 | mutex_unlock(&monc->mutex); | |
513a8243 | 790 | |
d0b19705 | 791 | return req; |
513a8243 | 792 | |
d0b19705 | 793 | err_put_req: |
f646912d | 794 | put_generic_request(req); |
d0b19705 | 795 | return ERR_PTR(-ENOMEM); |
513a8243 | 796 | } |
d0b19705 ID |
797 | |
798 | /* | |
799 | * Send MMonGetVersion and wait for the reply. | |
800 | * | |
801 | * @what: one of "mdsmap", "osdmap" or "monmap" | |
802 | */ | |
803 | int ceph_monc_get_version(struct ceph_mon_client *monc, const char *what, | |
804 | u64 *newest) | |
805 | { | |
806 | struct ceph_mon_generic_request *req; | |
807 | int ret; | |
808 | ||
809 | req = __ceph_monc_get_version(monc, what, NULL, 0); | |
810 | if (IS_ERR(req)) | |
811 | return PTR_ERR(req); | |
812 | ||
813 | ret = wait_generic_request(req); | |
814 | if (!ret) | |
815 | *newest = req->u.newest; | |
816 | ||
817 | put_generic_request(req); | |
818 | return ret; | |
819 | } | |
820 | EXPORT_SYMBOL(ceph_monc_get_version); | |
821 | ||
822 | /* | |
823 | * Send MMonGetVersion, | |
824 | * | |
825 | * @what: one of "mdsmap", "osdmap" or "monmap" | |
826 | */ | |
827 | int ceph_monc_get_version_async(struct ceph_mon_client *monc, const char *what, | |
828 | ceph_monc_callback_t cb, u64 private_data) | |
829 | { | |
830 | struct ceph_mon_generic_request *req; | |
831 | ||
832 | req = __ceph_monc_get_version(monc, what, cb, private_data); | |
833 | if (IS_ERR(req)) | |
834 | return PTR_ERR(req); | |
835 | ||
836 | put_generic_request(req); | |
837 | return 0; | |
838 | } | |
839 | EXPORT_SYMBOL(ceph_monc_get_version_async); | |
513a8243 | 840 | |
6305a3b4 DF |
841 | static void handle_command_ack(struct ceph_mon_client *monc, |
842 | struct ceph_msg *msg) | |
843 | { | |
844 | struct ceph_mon_generic_request *req; | |
845 | void *p = msg->front.iov_base; | |
846 | void *const end = p + msg->front_alloc_len; | |
847 | u64 tid = le64_to_cpu(msg->hdr.tid); | |
848 | ||
849 | dout("%s msg %p tid %llu\n", __func__, msg, tid); | |
850 | ||
851 | ceph_decode_need(&p, end, sizeof(struct ceph_mon_request_header) + | |
852 | sizeof(u32), bad); | |
853 | p += sizeof(struct ceph_mon_request_header); | |
854 | ||
855 | mutex_lock(&monc->mutex); | |
856 | req = lookup_generic_request(&monc->generic_request_tree, tid); | |
857 | if (!req) { | |
858 | mutex_unlock(&monc->mutex); | |
859 | return; | |
860 | } | |
861 | ||
862 | req->result = ceph_decode_32(&p); | |
863 | __finish_generic_request(req); | |
864 | mutex_unlock(&monc->mutex); | |
865 | ||
866 | complete_generic_request(req); | |
867 | return; | |
868 | ||
869 | bad: | |
870 | pr_err("corrupt mon_command ack, tid %llu\n", tid); | |
871 | ceph_msg_dump(msg); | |
872 | } | |
873 | ||
874 | int ceph_monc_blacklist_add(struct ceph_mon_client *monc, | |
875 | struct ceph_entity_addr *client_addr) | |
876 | { | |
877 | struct ceph_mon_generic_request *req; | |
878 | struct ceph_mon_command *h; | |
879 | int ret = -ENOMEM; | |
880 | int len; | |
881 | ||
882 | req = alloc_generic_request(monc, GFP_NOIO); | |
883 | if (!req) | |
884 | goto out; | |
885 | ||
886 | req->request = ceph_msg_new(CEPH_MSG_MON_COMMAND, 256, GFP_NOIO, true); | |
887 | if (!req->request) | |
888 | goto out; | |
889 | ||
890 | req->reply = ceph_msg_new(CEPH_MSG_MON_COMMAND_ACK, 512, GFP_NOIO, | |
891 | true); | |
892 | if (!req->reply) | |
893 | goto out; | |
894 | ||
895 | mutex_lock(&monc->mutex); | |
896 | register_generic_request(req); | |
897 | h = req->request->front.iov_base; | |
898 | h->monhdr.have_version = 0; | |
899 | h->monhdr.session_mon = cpu_to_le16(-1); | |
900 | h->monhdr.session_mon_tid = 0; | |
901 | h->fsid = monc->monmap->fsid; | |
902 | h->num_strs = cpu_to_le32(1); | |
903 | len = sprintf(h->str, "{ \"prefix\": \"osd blacklist\", \ | |
904 | \"blacklistop\": \"add\", \ | |
905 | \"addr\": \"%pISpc/%u\" }", | |
906 | &client_addr->in_addr, le32_to_cpu(client_addr->nonce)); | |
907 | h->str_len = cpu_to_le32(len); | |
908 | send_generic_request(monc, req); | |
909 | mutex_unlock(&monc->mutex); | |
910 | ||
911 | ret = wait_generic_request(req); | |
912 | out: | |
913 | put_generic_request(req); | |
914 | return ret; | |
915 | } | |
916 | EXPORT_SYMBOL(ceph_monc_blacklist_add); | |
917 | ||
ba75bb98 | 918 | /* |
e56fa10e | 919 | * Resend pending generic requests. |
ba75bb98 | 920 | */ |
f8c76f6f | 921 | static void __resend_generic_request(struct ceph_mon_client *monc) |
ba75bb98 | 922 | { |
f8c76f6f | 923 | struct ceph_mon_generic_request *req; |
85ff03f6 | 924 | struct rb_node *p; |
ba75bb98 | 925 | |
f8c76f6f YS |
926 | for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) { |
927 | req = rb_entry(p, struct ceph_mon_generic_request, node); | |
6740a845 | 928 | ceph_msg_revoke(req->request); |
4f471e4a | 929 | ceph_msg_revoke_incoming(req->reply); |
67130934 | 930 | ceph_con_send(&monc->con, ceph_msg_get(req->request)); |
ba75bb98 SW |
931 | } |
932 | } | |
933 | ||
934 | /* | |
935 | * Delayed work. If we haven't mounted yet, retry. Otherwise, | |
936 | * renew/retry subscription as needed (in case it is timing out, or we | |
937 | * got an ENOMEM). And keep the monitor connection alive. | |
938 | */ | |
939 | static void delayed_work(struct work_struct *work) | |
940 | { | |
941 | struct ceph_mon_client *monc = | |
942 | container_of(work, struct ceph_mon_client, delayed_work.work); | |
943 | ||
944 | dout("monc delayed_work\n"); | |
945 | mutex_lock(&monc->mutex); | |
4e7a5dcd | 946 | if (monc->hunting) { |
1752b50c ID |
947 | dout("%s continuing hunt\n", __func__); |
948 | reopen_session(monc); | |
ba75bb98 | 949 | } else { |
8b9558aa YZ |
950 | int is_auth = ceph_auth_is_authenticated(monc->auth); |
951 | if (ceph_con_keepalive_expired(&monc->con, | |
58d81b12 | 952 | CEPH_MONC_PING_TIMEOUT)) { |
8b9558aa YZ |
953 | dout("monc keepalive timeout\n"); |
954 | is_auth = 0; | |
1752b50c | 955 | reopen_session(monc); |
8b9558aa | 956 | } |
9bd2e6f8 | 957 | |
8b9558aa YZ |
958 | if (!monc->hunting) { |
959 | ceph_con_keepalive(&monc->con); | |
960 | __validate_auth(monc); | |
961 | } | |
9bd2e6f8 | 962 | |
220abf5a ID |
963 | if (is_auth && |
964 | !(monc->con.peer_features & CEPH_FEATURE_MON_STATEFUL_SUB)) { | |
82dcabad ID |
965 | unsigned long now = jiffies; |
966 | ||
967 | dout("%s renew subs? now %lu renew after %lu\n", | |
968 | __func__, now, monc->sub_renew_after); | |
969 | if (time_after_eq(now, monc->sub_renew_after)) | |
970 | __send_subscribe(monc); | |
971 | } | |
ba75bb98 | 972 | } |
ba75bb98 SW |
973 | __schedule_delayed(monc); |
974 | mutex_unlock(&monc->mutex); | |
975 | } | |
976 | ||
6b805185 SW |
977 | /* |
978 | * On startup, we build a temporary monmap populated with the IPs | |
979 | * provided by mount(2). | |
980 | */ | |
981 | static int build_initial_monmap(struct ceph_mon_client *monc) | |
982 | { | |
3d14c5d2 YS |
983 | struct ceph_options *opt = monc->client->options; |
984 | struct ceph_entity_addr *mon_addr = opt->mon_addr; | |
985 | int num_mon = opt->num_mon; | |
6b805185 SW |
986 | int i; |
987 | ||
988 | /* build initial monmap */ | |
989 | monc->monmap = kzalloc(sizeof(*monc->monmap) + | |
990 | num_mon*sizeof(monc->monmap->mon_inst[0]), | |
991 | GFP_KERNEL); | |
992 | if (!monc->monmap) | |
993 | return -ENOMEM; | |
994 | for (i = 0; i < num_mon; i++) { | |
995 | monc->monmap->mon_inst[i].addr = mon_addr[i]; | |
6b805185 SW |
996 | monc->monmap->mon_inst[i].addr.nonce = 0; |
997 | monc->monmap->mon_inst[i].name.type = | |
998 | CEPH_ENTITY_TYPE_MON; | |
999 | monc->monmap->mon_inst[i].name.num = cpu_to_le64(i); | |
1000 | } | |
1001 | monc->monmap->num_mon = num_mon; | |
6b805185 SW |
1002 | return 0; |
1003 | } | |
1004 | ||
ba75bb98 SW |
1005 | int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl) |
1006 | { | |
1007 | int err = 0; | |
1008 | ||
1009 | dout("init\n"); | |
1010 | memset(monc, 0, sizeof(*monc)); | |
1011 | monc->client = cl; | |
1012 | monc->monmap = NULL; | |
1013 | mutex_init(&monc->mutex); | |
1014 | ||
6b805185 SW |
1015 | err = build_initial_monmap(monc); |
1016 | if (err) | |
1017 | goto out; | |
1018 | ||
f6a2f5be | 1019 | /* connection */ |
4e7a5dcd | 1020 | /* authentication */ |
3d14c5d2 | 1021 | monc->auth = ceph_auth_init(cl->options->name, |
8323c3aa | 1022 | cl->options->key); |
49d9224c NW |
1023 | if (IS_ERR(monc->auth)) { |
1024 | err = PTR_ERR(monc->auth); | |
67130934 | 1025 | goto out_monmap; |
49d9224c | 1026 | } |
4e7a5dcd SW |
1027 | monc->auth->want_keys = |
1028 | CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON | | |
1029 | CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS; | |
1030 | ||
240ed68e | 1031 | /* msgs */ |
a79832f2 | 1032 | err = -ENOMEM; |
7c315c55 | 1033 | monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK, |
34d23762 | 1034 | sizeof(struct ceph_mon_subscribe_ack), |
5418d0a2 | 1035 | GFP_KERNEL, true); |
a79832f2 | 1036 | if (!monc->m_subscribe_ack) |
49d9224c | 1037 | goto out_auth; |
6694d6b9 | 1038 | |
5418d0a2 ID |
1039 | monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 128, |
1040 | GFP_KERNEL, true); | |
240ed68e SW |
1041 | if (!monc->m_subscribe) |
1042 | goto out_subscribe_ack; | |
1043 | ||
5418d0a2 ID |
1044 | monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, |
1045 | GFP_KERNEL, true); | |
a79832f2 | 1046 | if (!monc->m_auth_reply) |
240ed68e | 1047 | goto out_subscribe; |
4e7a5dcd | 1048 | |
5418d0a2 | 1049 | monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_KERNEL, true); |
9bd2e6f8 | 1050 | monc->pending_auth = 0; |
a79832f2 | 1051 | if (!monc->m_auth) |
6694d6b9 | 1052 | goto out_auth_reply; |
ba75bb98 | 1053 | |
735a72ef SW |
1054 | ceph_con_init(&monc->con, monc, &mon_con_ops, |
1055 | &monc->client->msgr); | |
1056 | ||
ba75bb98 | 1057 | monc->cur_mon = -1; |
168b9090 ID |
1058 | monc->had_a_connection = false; |
1059 | monc->hunt_mult = 1; | |
ba75bb98 SW |
1060 | |
1061 | INIT_DELAYED_WORK(&monc->delayed_work, delayed_work); | |
f8c76f6f | 1062 | monc->generic_request_tree = RB_ROOT; |
ba75bb98 SW |
1063 | monc->last_tid = 0; |
1064 | ||
737cc81e ID |
1065 | monc->fs_cluster_id = CEPH_FS_CLUSTER_ID_NONE; |
1066 | ||
4e7a5dcd SW |
1067 | return 0; |
1068 | ||
6694d6b9 SW |
1069 | out_auth_reply: |
1070 | ceph_msg_put(monc->m_auth_reply); | |
240ed68e SW |
1071 | out_subscribe: |
1072 | ceph_msg_put(monc->m_subscribe); | |
7c315c55 SW |
1073 | out_subscribe_ack: |
1074 | ceph_msg_put(monc->m_subscribe_ack); | |
49d9224c NW |
1075 | out_auth: |
1076 | ceph_auth_destroy(monc->auth); | |
4e7a5dcd SW |
1077 | out_monmap: |
1078 | kfree(monc->monmap); | |
ba75bb98 SW |
1079 | out: |
1080 | return err; | |
1081 | } | |
3d14c5d2 | 1082 | EXPORT_SYMBOL(ceph_monc_init); |
ba75bb98 SW |
1083 | |
1084 | void ceph_monc_stop(struct ceph_mon_client *monc) | |
1085 | { | |
1086 | dout("stop\n"); | |
1087 | cancel_delayed_work_sync(&monc->delayed_work); | |
1088 | ||
1089 | mutex_lock(&monc->mutex); | |
1090 | __close_session(monc); | |
0e04dc26 | 1091 | monc->cur_mon = -1; |
ba75bb98 SW |
1092 | mutex_unlock(&monc->mutex); |
1093 | ||
f3dea7ed SW |
1094 | /* |
1095 | * flush msgr queue before we destroy ourselves to ensure that: | |
1096 | * - any work that references our embedded con is finished. | |
1097 | * - any osd_client or other work that may reference an authorizer | |
1098 | * finishes before we shut down the auth subsystem. | |
1099 | */ | |
1100 | ceph_msgr_flush(); | |
1101 | ||
4e7a5dcd SW |
1102 | ceph_auth_destroy(monc->auth); |
1103 | ||
d0b19705 ID |
1104 | WARN_ON(!RB_EMPTY_ROOT(&monc->generic_request_tree)); |
1105 | ||
4e7a5dcd | 1106 | ceph_msg_put(monc->m_auth); |
6694d6b9 | 1107 | ceph_msg_put(monc->m_auth_reply); |
240ed68e | 1108 | ceph_msg_put(monc->m_subscribe); |
7c315c55 | 1109 | ceph_msg_put(monc->m_subscribe_ack); |
ba75bb98 SW |
1110 | |
1111 | kfree(monc->monmap); | |
1112 | } | |
3d14c5d2 | 1113 | EXPORT_SYMBOL(ceph_monc_stop); |
ba75bb98 | 1114 | |
0f9af169 ID |
1115 | static void finish_hunting(struct ceph_mon_client *monc) |
1116 | { | |
1117 | if (monc->hunting) { | |
1118 | dout("%s found mon%d\n", __func__, monc->cur_mon); | |
1119 | monc->hunting = false; | |
168b9090 ID |
1120 | monc->had_a_connection = true; |
1121 | monc->hunt_mult /= 2; /* reduce by 50% */ | |
1122 | if (monc->hunt_mult < 1) | |
1123 | monc->hunt_mult = 1; | |
0f9af169 ID |
1124 | } |
1125 | } | |
1126 | ||
4e7a5dcd SW |
1127 | static void handle_auth_reply(struct ceph_mon_client *monc, |
1128 | struct ceph_msg *msg) | |
1129 | { | |
1130 | int ret; | |
09c4d6a7 | 1131 | int was_auth = 0; |
4e7a5dcd SW |
1132 | |
1133 | mutex_lock(&monc->mutex); | |
27859f97 | 1134 | was_auth = ceph_auth_is_authenticated(monc->auth); |
9bd2e6f8 | 1135 | monc->pending_auth = 0; |
4e7a5dcd SW |
1136 | ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base, |
1137 | msg->front.iov_len, | |
1138 | monc->m_auth->front.iov_base, | |
3cea4c30 | 1139 | monc->m_auth->front_alloc_len); |
0f9af169 ID |
1140 | if (ret > 0) { |
1141 | __send_prepared_auth_request(monc, ret); | |
1142 | goto out; | |
1143 | } | |
1144 | ||
1145 | finish_hunting(monc); | |
1146 | ||
4e7a5dcd | 1147 | if (ret < 0) { |
9bd2e6f8 | 1148 | monc->client->auth_err = ret; |
27859f97 | 1149 | } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) { |
4e7a5dcd | 1150 | dout("authenticated, starting session\n"); |
0743304d | 1151 | |
15d9882c AE |
1152 | monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT; |
1153 | monc->client->msgr.inst.name.num = | |
0cf5537b | 1154 | cpu_to_le64(monc->auth->global_id); |
0743304d | 1155 | |
4e7a5dcd | 1156 | __send_subscribe(monc); |
f8c76f6f | 1157 | __resend_generic_request(monc); |
0f9af169 ID |
1158 | |
1159 | pr_info("mon%d %s session established\n", monc->cur_mon, | |
1160 | ceph_pr_addr(&monc->con.peer_addr.in_addr)); | |
4e7a5dcd | 1161 | } |
0f9af169 ID |
1162 | |
1163 | out: | |
4e7a5dcd | 1164 | mutex_unlock(&monc->mutex); |
0f9af169 ID |
1165 | if (monc->client->auth_err < 0) |
1166 | wake_up_all(&monc->client->auth_wq); | |
4e7a5dcd SW |
1167 | } |
1168 | ||
9bd2e6f8 SW |
1169 | static int __validate_auth(struct ceph_mon_client *monc) |
1170 | { | |
1171 | int ret; | |
1172 | ||
1173 | if (monc->pending_auth) | |
1174 | return 0; | |
1175 | ||
1176 | ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base, | |
3cea4c30 | 1177 | monc->m_auth->front_alloc_len); |
9bd2e6f8 SW |
1178 | if (ret <= 0) |
1179 | return ret; /* either an error, or no need to authenticate */ | |
1180 | __send_prepared_auth_request(monc, ret); | |
1181 | return 0; | |
1182 | } | |
1183 | ||
1184 | int ceph_monc_validate_auth(struct ceph_mon_client *monc) | |
1185 | { | |
1186 | int ret; | |
1187 | ||
1188 | mutex_lock(&monc->mutex); | |
1189 | ret = __validate_auth(monc); | |
1190 | mutex_unlock(&monc->mutex); | |
1191 | return ret; | |
1192 | } | |
3d14c5d2 | 1193 | EXPORT_SYMBOL(ceph_monc_validate_auth); |
9bd2e6f8 | 1194 | |
ba75bb98 SW |
1195 | /* |
1196 | * handle incoming message | |
1197 | */ | |
1198 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
1199 | { | |
1200 | struct ceph_mon_client *monc = con->private; | |
1201 | int type = le16_to_cpu(msg->hdr.type); | |
1202 | ||
1203 | if (!monc) | |
1204 | return; | |
1205 | ||
1206 | switch (type) { | |
4e7a5dcd SW |
1207 | case CEPH_MSG_AUTH_REPLY: |
1208 | handle_auth_reply(monc, msg); | |
ba75bb98 SW |
1209 | break; |
1210 | ||
1211 | case CEPH_MSG_MON_SUBSCRIBE_ACK: | |
1212 | handle_subscribe_ack(monc, msg); | |
1213 | break; | |
1214 | ||
1215 | case CEPH_MSG_STATFS_REPLY: | |
1216 | handle_statfs_reply(monc, msg); | |
1217 | break; | |
1218 | ||
513a8243 ID |
1219 | case CEPH_MSG_MON_GET_VERSION_REPLY: |
1220 | handle_get_version_reply(monc, msg); | |
1221 | break; | |
1222 | ||
6305a3b4 DF |
1223 | case CEPH_MSG_MON_COMMAND_ACK: |
1224 | handle_command_ack(monc, msg); | |
1225 | break; | |
1226 | ||
4e7a5dcd SW |
1227 | case CEPH_MSG_MON_MAP: |
1228 | ceph_monc_handle_map(monc, msg); | |
1229 | break; | |
1230 | ||
ba75bb98 SW |
1231 | case CEPH_MSG_OSD_MAP: |
1232 | ceph_osdc_handle_map(&monc->client->osdc, msg); | |
1233 | break; | |
1234 | ||
1235 | default: | |
3d14c5d2 YS |
1236 | /* can the chained handler handle it? */ |
1237 | if (monc->client->extra_mon_dispatch && | |
1238 | monc->client->extra_mon_dispatch(monc->client, msg) == 0) | |
1239 | break; | |
1240 | ||
ba75bb98 SW |
1241 | pr_err("received unknown message type %d %s\n", type, |
1242 | ceph_msg_type_name(type)); | |
1243 | } | |
1244 | ceph_msg_put(msg); | |
1245 | } | |
1246 | ||
1247 | /* | |
1248 | * Allocate memory for incoming message | |
1249 | */ | |
1250 | static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con, | |
2450418c YS |
1251 | struct ceph_msg_header *hdr, |
1252 | int *skip) | |
ba75bb98 SW |
1253 | { |
1254 | struct ceph_mon_client *monc = con->private; | |
1255 | int type = le16_to_cpu(hdr->type); | |
2450418c | 1256 | int front_len = le32_to_cpu(hdr->front_len); |
5b3a4db3 | 1257 | struct ceph_msg *m = NULL; |
ba75bb98 | 1258 | |
2450418c | 1259 | *skip = 0; |
0547a9b3 | 1260 | |
ba75bb98 | 1261 | switch (type) { |
ba75bb98 | 1262 | case CEPH_MSG_MON_SUBSCRIBE_ACK: |
7c315c55 | 1263 | m = ceph_msg_get(monc->m_subscribe_ack); |
2450418c | 1264 | break; |
ba75bb98 | 1265 | case CEPH_MSG_STATFS_REPLY: |
6305a3b4 | 1266 | case CEPH_MSG_MON_COMMAND_ACK: |
f8c76f6f | 1267 | return get_generic_reply(con, hdr, skip); |
4e7a5dcd | 1268 | case CEPH_MSG_AUTH_REPLY: |
6694d6b9 | 1269 | m = ceph_msg_get(monc->m_auth_reply); |
2450418c | 1270 | break; |
513a8243 ID |
1271 | case CEPH_MSG_MON_GET_VERSION_REPLY: |
1272 | if (le64_to_cpu(hdr->tid) != 0) | |
1273 | return get_generic_reply(con, hdr, skip); | |
1274 | ||
1275 | /* | |
1276 | * Older OSDs don't set reply tid even if the orignal | |
1277 | * request had a non-zero tid. Workaround this weirdness | |
1278 | * by falling through to the allocate case. | |
1279 | */ | |
5b3a4db3 SW |
1280 | case CEPH_MSG_MON_MAP: |
1281 | case CEPH_MSG_MDS_MAP: | |
1282 | case CEPH_MSG_OSD_MAP: | |
0cabbd94 | 1283 | case CEPH_MSG_FS_MAP_USER: |
b61c2763 | 1284 | m = ceph_msg_new(type, front_len, GFP_NOFS, false); |
1c20f2d2 AE |
1285 | if (!m) |
1286 | return NULL; /* ENOMEM--return skip == 0 */ | |
5b3a4db3 | 1287 | break; |
ba75bb98 | 1288 | } |
2450418c | 1289 | |
5b3a4db3 SW |
1290 | if (!m) { |
1291 | pr_info("alloc_msg unknown type %d\n", type); | |
2450418c | 1292 | *skip = 1; |
73c3d481 | 1293 | } else if (front_len > m->front_alloc_len) { |
b9a67899 JP |
1294 | pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n", |
1295 | front_len, m->front_alloc_len, | |
1296 | (unsigned int)con->peer_name.type, | |
1297 | le64_to_cpu(con->peer_name.num)); | |
73c3d481 SW |
1298 | ceph_msg_put(m); |
1299 | m = ceph_msg_new(type, front_len, GFP_NOFS, false); | |
5b3a4db3 | 1300 | } |
73c3d481 | 1301 | |
2450418c | 1302 | return m; |
ba75bb98 SW |
1303 | } |
1304 | ||
1305 | /* | |
1306 | * If the monitor connection resets, pick a new monitor and resubmit | |
1307 | * any pending requests. | |
1308 | */ | |
1309 | static void mon_fault(struct ceph_connection *con) | |
1310 | { | |
1311 | struct ceph_mon_client *monc = con->private; | |
1312 | ||
ba75bb98 | 1313 | mutex_lock(&monc->mutex); |
b5d91704 ID |
1314 | dout("%s mon%d\n", __func__, monc->cur_mon); |
1315 | if (monc->cur_mon >= 0) { | |
1316 | if (!monc->hunting) { | |
1317 | dout("%s hunting for new mon\n", __func__); | |
1318 | reopen_session(monc); | |
1319 | __schedule_delayed(monc); | |
1320 | } else { | |
1321 | dout("%s already hunting\n", __func__); | |
1322 | } | |
ba75bb98 | 1323 | } |
ba75bb98 SW |
1324 | mutex_unlock(&monc->mutex); |
1325 | } | |
1326 | ||
ec87ef43 SW |
1327 | /* |
1328 | * We can ignore refcounting on the connection struct, as all references | |
1329 | * will come from the messenger workqueue, which is drained prior to | |
1330 | * mon_client destruction. | |
1331 | */ | |
1332 | static struct ceph_connection *con_get(struct ceph_connection *con) | |
1333 | { | |
1334 | return con; | |
1335 | } | |
1336 | ||
1337 | static void con_put(struct ceph_connection *con) | |
1338 | { | |
1339 | } | |
1340 | ||
9e32789f | 1341 | static const struct ceph_connection_operations mon_con_ops = { |
ec87ef43 SW |
1342 | .get = con_get, |
1343 | .put = con_put, | |
ba75bb98 SW |
1344 | .dispatch = dispatch, |
1345 | .fault = mon_fault, | |
1346 | .alloc_msg = mon_alloc_msg, | |
ba75bb98 | 1347 | }; |