]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
3d14c5d2 | 2 | #include <linux/ceph/ceph_debug.h> |
ba75bb98 | 3 | |
3d14c5d2 | 4 | #include <linux/module.h> |
ba75bb98 | 5 | #include <linux/types.h> |
5a0e3ad6 | 6 | #include <linux/slab.h> |
ba75bb98 SW |
7 | #include <linux/random.h> |
8 | #include <linux/sched.h> | |
9 | ||
220abf5a | 10 | #include <linux/ceph/ceph_features.h> |
3d14c5d2 YS |
11 | #include <linux/ceph/mon_client.h> |
12 | #include <linux/ceph/libceph.h> | |
ab434b60 | 13 | #include <linux/ceph/debugfs.h> |
3d14c5d2 | 14 | #include <linux/ceph/decode.h> |
3d14c5d2 | 15 | #include <linux/ceph/auth.h> |
ba75bb98 SW |
16 | |
17 | /* | |
18 | * Interact with Ceph monitor cluster. Handle requests for new map | |
19 | * versions, and periodically resend as needed. Also implement | |
20 | * statfs() and umount(). | |
21 | * | |
22 | * A small cluster of Ceph "monitors" are responsible for managing critical | |
23 | * cluster configuration and state information. An odd number (e.g., 3, 5) | |
24 | * of cmon daemons use a modified version of the Paxos part-time parliament | |
25 | * algorithm to manage the MDS map (mds cluster membership), OSD map, and | |
26 | * list of clients who have mounted the file system. | |
27 | * | |
28 | * We maintain an open, active session with a monitor at all times in order to | |
29 | * receive timely MDSMap updates. We periodically send a keepalive byte on the | |
30 | * TCP socket to ensure we detect a failure. If the connection does break, we | |
31 | * randomly hunt for a new monitor. Once the connection is reestablished, we | |
32 | * resend any outstanding requests. | |
33 | */ | |
34 | ||
9e32789f | 35 | static const struct ceph_connection_operations mon_con_ops; |
ba75bb98 | 36 | |
9bd2e6f8 SW |
37 | static int __validate_auth(struct ceph_mon_client *monc); |
38 | ||
ba75bb98 SW |
39 | /* |
40 | * Decode a monmap blob (e.g., during mount). | |
41 | */ | |
42 | struct ceph_monmap *ceph_monmap_decode(void *p, void *end) | |
43 | { | |
44 | struct ceph_monmap *m = NULL; | |
45 | int i, err = -EINVAL; | |
46 | struct ceph_fsid fsid; | |
47 | u32 epoch, num_mon; | |
4e7a5dcd SW |
48 | u32 len; |
49 | ||
50 | ceph_decode_32_safe(&p, end, len, bad); | |
51 | ceph_decode_need(&p, end, len, bad); | |
ba75bb98 SW |
52 | |
53 | dout("monmap_decode %p %p len %d\n", p, end, (int)(end-p)); | |
f3b4e55d | 54 | p += sizeof(u16); /* skip version */ |
ba75bb98 SW |
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 | ||
7377324e | 63 | if (num_mon > CEPH_MAX_MON) |
ba75bb98 | 64 | goto bad; |
acafe7e3 | 65 | m = kmalloc(struct_size(m, mon_inst, num_mon), GFP_NOFS); |
ba75bb98 SW |
66 | if (m == NULL) |
67 | return ERR_PTR(-ENOMEM); | |
68 | m->fsid = fsid; | |
69 | m->epoch = epoch; | |
70 | m->num_mon = num_mon; | |
71 | ceph_decode_copy(&p, m->mon_inst, num_mon*sizeof(m->mon_inst[0])); | |
63f2d211 SW |
72 | for (i = 0; i < num_mon; i++) |
73 | ceph_decode_addr(&m->mon_inst[i].addr); | |
ba75bb98 | 74 | |
ba75bb98 SW |
75 | dout("monmap_decode epoch %d, num_mon %d\n", m->epoch, |
76 | m->num_mon); | |
77 | for (i = 0; i < m->num_mon; i++) | |
78 | dout("monmap_decode mon%d is %s\n", i, | |
3d14c5d2 | 79 | ceph_pr_addr(&m->mon_inst[i].addr.in_addr)); |
ba75bb98 SW |
80 | return m; |
81 | ||
82 | bad: | |
83 | dout("monmap_decode failed with %d\n", err); | |
84 | kfree(m); | |
85 | return ERR_PTR(err); | |
86 | } | |
87 | ||
88 | /* | |
89 | * return true if *addr is included in the monmap. | |
90 | */ | |
91 | int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr) | |
92 | { | |
93 | int i; | |
94 | ||
95 | for (i = 0; i < m->num_mon; i++) | |
103e2d3a | 96 | if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0) |
ba75bb98 SW |
97 | return 1; |
98 | return 0; | |
99 | } | |
100 | ||
5ce6e9db SW |
101 | /* |
102 | * Send an auth request. | |
103 | */ | |
104 | static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len) | |
105 | { | |
106 | monc->pending_auth = 1; | |
107 | monc->m_auth->front.iov_len = len; | |
108 | monc->m_auth->hdr.front_len = cpu_to_le32(len); | |
6740a845 | 109 | ceph_msg_revoke(monc->m_auth); |
5ce6e9db | 110 | ceph_msg_get(monc->m_auth); /* keep our ref */ |
67130934 | 111 | ceph_con_send(&monc->con, monc->m_auth); |
5ce6e9db SW |
112 | } |
113 | ||
ba75bb98 SW |
114 | /* |
115 | * Close monitor session, if any. | |
116 | */ | |
117 | static void __close_session(struct ceph_mon_client *monc) | |
118 | { | |
f6a2f5be | 119 | dout("__close_session closing mon%d\n", monc->cur_mon); |
6740a845 | 120 | ceph_msg_revoke(monc->m_auth); |
4f471e4a SW |
121 | ceph_msg_revoke_incoming(monc->m_auth_reply); |
122 | ceph_msg_revoke(monc->m_subscribe); | |
123 | ceph_msg_revoke_incoming(monc->m_subscribe_ack); | |
67130934 | 124 | ceph_con_close(&monc->con); |
0e04dc26 | 125 | |
f6a2f5be SW |
126 | monc->pending_auth = 0; |
127 | ceph_auth_reset(monc->auth); | |
ba75bb98 SW |
128 | } |
129 | ||
130 | /* | |
0e04dc26 ID |
131 | * Pick a new monitor at random and set cur_mon. If we are repicking |
132 | * (i.e. cur_mon is already set), be sure to pick a different one. | |
ba75bb98 | 133 | */ |
0e04dc26 | 134 | static void pick_new_mon(struct ceph_mon_client *monc) |
ba75bb98 | 135 | { |
0e04dc26 | 136 | int old_mon = monc->cur_mon; |
ba75bb98 | 137 | |
0e04dc26 | 138 | BUG_ON(monc->monmap->num_mon < 1); |
ba75bb98 | 139 | |
0e04dc26 ID |
140 | if (monc->monmap->num_mon == 1) { |
141 | monc->cur_mon = 0; | |
142 | } else { | |
143 | int max = monc->monmap->num_mon; | |
144 | int o = -1; | |
145 | int n; | |
146 | ||
147 | if (monc->cur_mon >= 0) { | |
148 | if (monc->cur_mon < monc->monmap->num_mon) | |
149 | o = monc->cur_mon; | |
150 | if (o >= 0) | |
151 | max--; | |
152 | } | |
4e7a5dcd | 153 | |
0e04dc26 ID |
154 | n = prandom_u32() % max; |
155 | if (o >= 0 && n >= o) | |
156 | n++; | |
8b9558aa | 157 | |
0e04dc26 | 158 | monc->cur_mon = n; |
ba75bb98 | 159 | } |
0e04dc26 ID |
160 | |
161 | dout("%s mon%d -> mon%d out of %d mons\n", __func__, old_mon, | |
162 | monc->cur_mon, monc->monmap->num_mon); | |
163 | } | |
164 | ||
165 | /* | |
166 | * Open a session with a new monitor. | |
167 | */ | |
168 | static void __open_session(struct ceph_mon_client *monc) | |
169 | { | |
170 | int ret; | |
171 | ||
172 | pick_new_mon(monc); | |
173 | ||
1752b50c | 174 | monc->hunting = true; |
168b9090 ID |
175 | if (monc->had_a_connection) { |
176 | monc->hunt_mult *= CEPH_MONC_HUNT_BACKOFF; | |
177 | if (monc->hunt_mult > CEPH_MONC_HUNT_MAX_MULT) | |
178 | monc->hunt_mult = CEPH_MONC_HUNT_MAX_MULT; | |
179 | } | |
180 | ||
0e04dc26 ID |
181 | monc->sub_renew_after = jiffies; /* i.e., expired */ |
182 | monc->sub_renew_sent = 0; | |
183 | ||
184 | dout("%s opening mon%d\n", __func__, monc->cur_mon); | |
185 | ceph_con_open(&monc->con, CEPH_ENTITY_TYPE_MON, monc->cur_mon, | |
186 | &monc->monmap->mon_inst[monc->cur_mon].addr); | |
187 | ||
188 | /* | |
189 | * send an initial keepalive to ensure our timestamp is valid | |
190 | * by the time we are in an OPENED state | |
191 | */ | |
192 | ceph_con_keepalive(&monc->con); | |
193 | ||
194 | /* initiate authentication handshake */ | |
195 | ret = ceph_auth_build_hello(monc->auth, | |
196 | monc->m_auth->front.iov_base, | |
197 | monc->m_auth->front_alloc_len); | |
198 | BUG_ON(ret <= 0); | |
199 | __send_prepared_auth_request(monc, ret); | |
ba75bb98 SW |
200 | } |
201 | ||
1752b50c ID |
202 | static void reopen_session(struct ceph_mon_client *monc) |
203 | { | |
204 | if (!monc->hunting) | |
205 | pr_info("mon%d %s session lost, hunting for new mon\n", | |
206 | monc->cur_mon, ceph_pr_addr(&monc->con.peer_addr.in_addr)); | |
207 | ||
208 | __close_session(monc); | |
209 | __open_session(monc); | |
210 | } | |
211 | ||
facb9f6e ID |
212 | static void un_backoff(struct ceph_mon_client *monc) |
213 | { | |
214 | monc->hunt_mult /= 2; /* reduce by 50% */ | |
215 | if (monc->hunt_mult < 1) | |
216 | monc->hunt_mult = 1; | |
217 | dout("%s hunt_mult now %d\n", __func__, monc->hunt_mult); | |
218 | } | |
219 | ||
ba75bb98 SW |
220 | /* |
221 | * Reschedule delayed work timer. | |
222 | */ | |
223 | static void __schedule_delayed(struct ceph_mon_client *monc) | |
224 | { | |
8b9558aa | 225 | unsigned long delay; |
ba75bb98 | 226 | |
168b9090 ID |
227 | if (monc->hunting) |
228 | delay = CEPH_MONC_HUNT_INTERVAL * monc->hunt_mult; | |
229 | else | |
58d81b12 | 230 | delay = CEPH_MONC_PING_INTERVAL; |
168b9090 | 231 | |
8b9558aa | 232 | dout("__schedule_delayed after %lu\n", delay); |
bee3a37c ID |
233 | mod_delayed_work(system_wq, &monc->delayed_work, |
234 | round_jiffies_relative(delay)); | |
ba75bb98 SW |
235 | } |
236 | ||
82dcabad | 237 | const char *ceph_sub_str[] = { |
82dcabad ID |
238 | [CEPH_SUB_MONMAP] = "monmap", |
239 | [CEPH_SUB_OSDMAP] = "osdmap", | |
0cabbd94 YZ |
240 | [CEPH_SUB_FSMAP] = "fsmap.user", |
241 | [CEPH_SUB_MDSMAP] = "mdsmap", | |
82dcabad ID |
242 | }; |
243 | ||
ba75bb98 | 244 | /* |
82dcabad ID |
245 | * Send subscribe request for one or more maps, according to |
246 | * monc->subs. | |
ba75bb98 SW |
247 | */ |
248 | static void __send_subscribe(struct ceph_mon_client *monc) | |
249 | { | |
82dcabad ID |
250 | struct ceph_msg *msg = monc->m_subscribe; |
251 | void *p = msg->front.iov_base; | |
252 | void *const end = p + msg->front_alloc_len; | |
253 | int num = 0; | |
254 | int i; | |
255 | ||
256 | dout("%s sent %lu\n", __func__, monc->sub_renew_sent); | |
257 | ||
258 | BUG_ON(monc->cur_mon < 0); | |
259 | ||
260 | if (!monc->sub_renew_sent) | |
261 | monc->sub_renew_sent = jiffies | 1; /* never 0 */ | |
262 | ||
263 | msg->hdr.version = cpu_to_le16(2); | |
264 | ||
265 | for (i = 0; i < ARRAY_SIZE(monc->subs); i++) { | |
266 | if (monc->subs[i].want) | |
267 | num++; | |
268 | } | |
269 | BUG_ON(num < 1); /* monmap sub is always there */ | |
270 | ceph_encode_32(&p, num); | |
271 | for (i = 0; i < ARRAY_SIZE(monc->subs); i++) { | |
737cc81e ID |
272 | char buf[32]; |
273 | int len; | |
82dcabad ID |
274 | |
275 | if (!monc->subs[i].want) | |
276 | continue; | |
277 | ||
737cc81e ID |
278 | len = sprintf(buf, "%s", ceph_sub_str[i]); |
279 | if (i == CEPH_SUB_MDSMAP && | |
280 | monc->fs_cluster_id != CEPH_FS_CLUSTER_ID_NONE) | |
281 | len += sprintf(buf + len, ".%d", monc->fs_cluster_id); | |
282 | ||
283 | dout("%s %s start %llu flags 0x%x\n", __func__, buf, | |
82dcabad ID |
284 | le64_to_cpu(monc->subs[i].item.start), |
285 | monc->subs[i].item.flags); | |
737cc81e | 286 | ceph_encode_string(&p, end, buf, len); |
82dcabad ID |
287 | memcpy(p, &monc->subs[i].item, sizeof(monc->subs[i].item)); |
288 | p += sizeof(monc->subs[i].item); | |
ba75bb98 | 289 | } |
82dcabad | 290 | |
737cc81e | 291 | BUG_ON(p > end); |
82dcabad ID |
292 | msg->front.iov_len = p - msg->front.iov_base; |
293 | msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); | |
294 | ceph_msg_revoke(msg); | |
295 | ceph_con_send(&monc->con, ceph_msg_get(msg)); | |
ba75bb98 SW |
296 | } |
297 | ||
298 | static void handle_subscribe_ack(struct ceph_mon_client *monc, | |
299 | struct ceph_msg *msg) | |
300 | { | |
95c96174 | 301 | unsigned int seconds; |
07bd10fb SW |
302 | struct ceph_mon_subscribe_ack *h = msg->front.iov_base; |
303 | ||
304 | if (msg->front.iov_len < sizeof(*h)) | |
305 | goto bad; | |
306 | seconds = le32_to_cpu(h->duration); | |
ba75bb98 | 307 | |
ba75bb98 | 308 | mutex_lock(&monc->mutex); |
82dcabad | 309 | if (monc->sub_renew_sent) { |
220abf5a ID |
310 | /* |
311 | * This is only needed for legacy (infernalis or older) | |
312 | * MONs -- see delayed_work(). | |
313 | */ | |
82dcabad ID |
314 | monc->sub_renew_after = monc->sub_renew_sent + |
315 | (seconds >> 1) * HZ - 1; | |
316 | dout("%s sent %lu duration %d renew after %lu\n", __func__, | |
317 | monc->sub_renew_sent, seconds, monc->sub_renew_after); | |
318 | monc->sub_renew_sent = 0; | |
319 | } else { | |
320 | dout("%s sent %lu renew after %lu, ignoring\n", __func__, | |
321 | monc->sub_renew_sent, monc->sub_renew_after); | |
322 | } | |
ba75bb98 SW |
323 | mutex_unlock(&monc->mutex); |
324 | return; | |
325 | bad: | |
326 | pr_err("got corrupt subscribe-ack msg\n"); | |
9ec7cab1 | 327 | ceph_msg_dump(msg); |
ba75bb98 SW |
328 | } |
329 | ||
330 | /* | |
82dcabad ID |
331 | * Register interest in a map |
332 | * | |
333 | * @sub: one of CEPH_SUB_* | |
334 | * @epoch: X for "every map since X", or 0 for "just the latest" | |
ba75bb98 | 335 | */ |
82dcabad ID |
336 | static bool __ceph_monc_want_map(struct ceph_mon_client *monc, int sub, |
337 | u32 epoch, bool continuous) | |
ba75bb98 | 338 | { |
82dcabad ID |
339 | __le64 start = cpu_to_le64(epoch); |
340 | u8 flags = !continuous ? CEPH_SUBSCRIBE_ONETIME : 0; | |
341 | ||
342 | dout("%s %s epoch %u continuous %d\n", __func__, ceph_sub_str[sub], | |
343 | epoch, continuous); | |
344 | ||
345 | if (monc->subs[sub].want && | |
346 | monc->subs[sub].item.start == start && | |
347 | monc->subs[sub].item.flags == flags) | |
348 | return false; | |
349 | ||
350 | monc->subs[sub].item.start = start; | |
351 | monc->subs[sub].item.flags = flags; | |
352 | monc->subs[sub].want = true; | |
353 | ||
354 | return true; | |
355 | } | |
356 | ||
357 | bool ceph_monc_want_map(struct ceph_mon_client *monc, int sub, u32 epoch, | |
358 | bool continuous) | |
359 | { | |
360 | bool need_request; | |
361 | ||
ba75bb98 | 362 | mutex_lock(&monc->mutex); |
82dcabad | 363 | need_request = __ceph_monc_want_map(monc, sub, epoch, continuous); |
ba75bb98 | 364 | mutex_unlock(&monc->mutex); |
82dcabad ID |
365 | |
366 | return need_request; | |
367 | } | |
368 | EXPORT_SYMBOL(ceph_monc_want_map); | |
369 | ||
370 | /* | |
371 | * Keep track of which maps we have | |
372 | * | |
373 | * @sub: one of CEPH_SUB_* | |
374 | */ | |
375 | static void __ceph_monc_got_map(struct ceph_mon_client *monc, int sub, | |
376 | u32 epoch) | |
377 | { | |
378 | dout("%s %s epoch %u\n", __func__, ceph_sub_str[sub], epoch); | |
379 | ||
380 | if (monc->subs[sub].want) { | |
381 | if (monc->subs[sub].item.flags & CEPH_SUBSCRIBE_ONETIME) | |
382 | monc->subs[sub].want = false; | |
383 | else | |
384 | monc->subs[sub].item.start = cpu_to_le64(epoch + 1); | |
385 | } | |
386 | ||
387 | monc->subs[sub].have = epoch; | |
ba75bb98 SW |
388 | } |
389 | ||
82dcabad | 390 | void ceph_monc_got_map(struct ceph_mon_client *monc, int sub, u32 epoch) |
ba75bb98 SW |
391 | { |
392 | mutex_lock(&monc->mutex); | |
82dcabad | 393 | __ceph_monc_got_map(monc, sub, epoch); |
ba75bb98 | 394 | mutex_unlock(&monc->mutex); |
ba75bb98 | 395 | } |
82dcabad | 396 | EXPORT_SYMBOL(ceph_monc_got_map); |
ba75bb98 | 397 | |
42c1b124 ID |
398 | void ceph_monc_renew_subs(struct ceph_mon_client *monc) |
399 | { | |
400 | mutex_lock(&monc->mutex); | |
401 | __send_subscribe(monc); | |
402 | mutex_unlock(&monc->mutex); | |
403 | } | |
404 | EXPORT_SYMBOL(ceph_monc_renew_subs); | |
405 | ||
a319bf56 ID |
406 | /* |
407 | * Wait for an osdmap with a given epoch. | |
408 | * | |
409 | * @epoch: epoch to wait for | |
410 | * @timeout: in jiffies, 0 means "wait forever" | |
411 | */ | |
6044cde6 ID |
412 | int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch, |
413 | unsigned long timeout) | |
414 | { | |
415 | unsigned long started = jiffies; | |
216639dd | 416 | long ret; |
6044cde6 ID |
417 | |
418 | mutex_lock(&monc->mutex); | |
82dcabad | 419 | while (monc->subs[CEPH_SUB_OSDMAP].have < epoch) { |
6044cde6 ID |
420 | mutex_unlock(&monc->mutex); |
421 | ||
a319bf56 | 422 | if (timeout && time_after_eq(jiffies, started + timeout)) |
6044cde6 ID |
423 | return -ETIMEDOUT; |
424 | ||
425 | ret = wait_event_interruptible_timeout(monc->client->auth_wq, | |
82dcabad ID |
426 | monc->subs[CEPH_SUB_OSDMAP].have >= epoch, |
427 | ceph_timeout_jiffies(timeout)); | |
6044cde6 ID |
428 | if (ret < 0) |
429 | return ret; | |
430 | ||
431 | mutex_lock(&monc->mutex); | |
432 | } | |
433 | ||
434 | mutex_unlock(&monc->mutex); | |
435 | return 0; | |
436 | } | |
437 | EXPORT_SYMBOL(ceph_monc_wait_osdmap); | |
ba75bb98 | 438 | |
4e7a5dcd | 439 | /* |
82dcabad ID |
440 | * Open a session with a random monitor. Request monmap and osdmap, |
441 | * which are waited upon in __ceph_open_session(). | |
4e7a5dcd SW |
442 | */ |
443 | int ceph_monc_open_session(struct ceph_mon_client *monc) | |
ba75bb98 | 444 | { |
ba75bb98 | 445 | mutex_lock(&monc->mutex); |
82dcabad ID |
446 | __ceph_monc_want_map(monc, CEPH_SUB_MONMAP, 0, true); |
447 | __ceph_monc_want_map(monc, CEPH_SUB_OSDMAP, 0, false); | |
4e7a5dcd | 448 | __open_session(monc); |
ba75bb98 SW |
449 | __schedule_delayed(monc); |
450 | mutex_unlock(&monc->mutex); | |
451 | return 0; | |
452 | } | |
3d14c5d2 | 453 | EXPORT_SYMBOL(ceph_monc_open_session); |
ba75bb98 | 454 | |
0743304d SW |
455 | static void ceph_monc_handle_map(struct ceph_mon_client *monc, |
456 | struct ceph_msg *msg) | |
4e7a5dcd SW |
457 | { |
458 | struct ceph_client *client = monc->client; | |
459 | struct ceph_monmap *monmap = NULL, *old = monc->monmap; | |
460 | void *p, *end; | |
461 | ||
462 | mutex_lock(&monc->mutex); | |
463 | ||
464 | dout("handle_monmap\n"); | |
465 | p = msg->front.iov_base; | |
466 | end = p + msg->front.iov_len; | |
467 | ||
468 | monmap = ceph_monmap_decode(p, end); | |
469 | if (IS_ERR(monmap)) { | |
470 | pr_err("problem decoding monmap, %d\n", | |
471 | (int)PTR_ERR(monmap)); | |
d4a780ce | 472 | goto out; |
4e7a5dcd | 473 | } |
0743304d SW |
474 | |
475 | if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) { | |
4e7a5dcd | 476 | kfree(monmap); |
d4a780ce | 477 | goto out; |
4e7a5dcd SW |
478 | } |
479 | ||
480 | client->monc.monmap = monmap; | |
4e7a5dcd SW |
481 | kfree(old); |
482 | ||
82dcabad | 483 | __ceph_monc_got_map(monc, CEPH_SUB_MONMAP, monc->monmap->epoch); |
02ac956c | 484 | client->have_fsid = true; |
d1c338a5 | 485 | |
d4a780ce | 486 | out: |
4e7a5dcd | 487 | mutex_unlock(&monc->mutex); |
03066f23 | 488 | wake_up_all(&client->auth_wq); |
4e7a5dcd SW |
489 | } |
490 | ||
ba75bb98 | 491 | /* |
7a6fdeb2 | 492 | * generic requests (currently statfs, mon_get_version) |
ba75bb98 | 493 | */ |
fcd00b68 | 494 | DEFINE_RB_FUNCS(generic_request, struct ceph_mon_generic_request, tid, node) |
85ff03f6 | 495 | |
f8c76f6f | 496 | static void release_generic_request(struct kref *kref) |
3143edd3 | 497 | { |
f8c76f6f YS |
498 | struct ceph_mon_generic_request *req = |
499 | container_of(kref, struct ceph_mon_generic_request, kref); | |
3143edd3 | 500 | |
d0b19705 ID |
501 | dout("%s greq %p request %p reply %p\n", __func__, req, req->request, |
502 | req->reply); | |
503 | WARN_ON(!RB_EMPTY_NODE(&req->node)); | |
504 | ||
3143edd3 SW |
505 | if (req->reply) |
506 | ceph_msg_put(req->reply); | |
507 | if (req->request) | |
508 | ceph_msg_put(req->request); | |
20547567 YS |
509 | |
510 | kfree(req); | |
3143edd3 SW |
511 | } |
512 | ||
f8c76f6f | 513 | static void put_generic_request(struct ceph_mon_generic_request *req) |
3143edd3 | 514 | { |
d0b19705 ID |
515 | if (req) |
516 | kref_put(&req->kref, release_generic_request); | |
3143edd3 SW |
517 | } |
518 | ||
f8c76f6f | 519 | static void get_generic_request(struct ceph_mon_generic_request *req) |
3143edd3 SW |
520 | { |
521 | kref_get(&req->kref); | |
522 | } | |
523 | ||
d0b19705 ID |
524 | static struct ceph_mon_generic_request * |
525 | alloc_generic_request(struct ceph_mon_client *monc, gfp_t gfp) | |
526 | { | |
527 | struct ceph_mon_generic_request *req; | |
528 | ||
529 | req = kzalloc(sizeof(*req), gfp); | |
530 | if (!req) | |
531 | return NULL; | |
532 | ||
533 | req->monc = monc; | |
534 | kref_init(&req->kref); | |
535 | RB_CLEAR_NODE(&req->node); | |
536 | init_completion(&req->completion); | |
537 | ||
538 | dout("%s greq %p\n", __func__, req); | |
539 | return req; | |
540 | } | |
541 | ||
542 | static void register_generic_request(struct ceph_mon_generic_request *req) | |
543 | { | |
544 | struct ceph_mon_client *monc = req->monc; | |
545 | ||
546 | WARN_ON(req->tid); | |
547 | ||
548 | get_generic_request(req); | |
549 | req->tid = ++monc->last_tid; | |
550 | insert_generic_request(&monc->generic_request_tree, req); | |
551 | } | |
552 | ||
553 | static void send_generic_request(struct ceph_mon_client *monc, | |
554 | struct ceph_mon_generic_request *req) | |
555 | { | |
556 | WARN_ON(!req->tid); | |
557 | ||
558 | dout("%s greq %p tid %llu\n", __func__, req, req->tid); | |
559 | req->request->hdr.tid = cpu_to_le64(req->tid); | |
560 | ceph_con_send(&monc->con, ceph_msg_get(req->request)); | |
561 | } | |
562 | ||
563 | static void __finish_generic_request(struct ceph_mon_generic_request *req) | |
564 | { | |
565 | struct ceph_mon_client *monc = req->monc; | |
566 | ||
567 | dout("%s greq %p tid %llu\n", __func__, req, req->tid); | |
568 | erase_generic_request(&monc->generic_request_tree, req); | |
569 | ||
570 | ceph_msg_revoke(req->request); | |
571 | ceph_msg_revoke_incoming(req->reply); | |
572 | } | |
573 | ||
574 | static void finish_generic_request(struct ceph_mon_generic_request *req) | |
575 | { | |
576 | __finish_generic_request(req); | |
577 | put_generic_request(req); | |
578 | } | |
579 | ||
580 | static void complete_generic_request(struct ceph_mon_generic_request *req) | |
581 | { | |
582 | if (req->complete_cb) | |
583 | req->complete_cb(req); | |
584 | else | |
585 | complete_all(&req->completion); | |
586 | put_generic_request(req); | |
587 | } | |
588 | ||
f52ec33c | 589 | static void cancel_generic_request(struct ceph_mon_generic_request *req) |
d0b19705 ID |
590 | { |
591 | struct ceph_mon_client *monc = req->monc; | |
592 | struct ceph_mon_generic_request *lookup_req; | |
593 | ||
594 | dout("%s greq %p tid %llu\n", __func__, req, req->tid); | |
595 | ||
596 | mutex_lock(&monc->mutex); | |
597 | lookup_req = lookup_generic_request(&monc->generic_request_tree, | |
598 | req->tid); | |
599 | if (lookup_req) { | |
600 | WARN_ON(lookup_req != req); | |
601 | finish_generic_request(req); | |
602 | } | |
603 | ||
604 | mutex_unlock(&monc->mutex); | |
605 | } | |
606 | ||
607 | static int wait_generic_request(struct ceph_mon_generic_request *req) | |
608 | { | |
609 | int ret; | |
610 | ||
611 | dout("%s greq %p tid %llu\n", __func__, req, req->tid); | |
612 | ret = wait_for_completion_interruptible(&req->completion); | |
613 | if (ret) | |
614 | cancel_generic_request(req); | |
615 | else | |
616 | ret = req->result; /* completed */ | |
617 | ||
618 | return ret; | |
619 | } | |
620 | ||
f8c76f6f | 621 | static struct ceph_msg *get_generic_reply(struct ceph_connection *con, |
3143edd3 SW |
622 | struct ceph_msg_header *hdr, |
623 | int *skip) | |
624 | { | |
625 | struct ceph_mon_client *monc = con->private; | |
f8c76f6f | 626 | struct ceph_mon_generic_request *req; |
3143edd3 SW |
627 | u64 tid = le64_to_cpu(hdr->tid); |
628 | struct ceph_msg *m; | |
629 | ||
630 | mutex_lock(&monc->mutex); | |
fcd00b68 | 631 | req = lookup_generic_request(&monc->generic_request_tree, tid); |
3143edd3 | 632 | if (!req) { |
f8c76f6f | 633 | dout("get_generic_reply %lld dne\n", tid); |
3143edd3 SW |
634 | *skip = 1; |
635 | m = NULL; | |
636 | } else { | |
f8c76f6f | 637 | dout("get_generic_reply %lld got %p\n", tid, req->reply); |
1c20f2d2 | 638 | *skip = 0; |
3143edd3 SW |
639 | m = ceph_msg_get(req->reply); |
640 | /* | |
641 | * we don't need to track the connection reading into | |
642 | * this reply because we only have one open connection | |
643 | * at a time, ever. | |
644 | */ | |
645 | } | |
646 | mutex_unlock(&monc->mutex); | |
647 | return m; | |
648 | } | |
649 | ||
e56fa10e YS |
650 | /* |
651 | * statfs | |
652 | */ | |
ba75bb98 SW |
653 | static void handle_statfs_reply(struct ceph_mon_client *monc, |
654 | struct ceph_msg *msg) | |
655 | { | |
f8c76f6f | 656 | struct ceph_mon_generic_request *req; |
ba75bb98 | 657 | struct ceph_mon_statfs_reply *reply = msg->front.iov_base; |
3143edd3 | 658 | u64 tid = le64_to_cpu(msg->hdr.tid); |
ba75bb98 | 659 | |
d0b19705 ID |
660 | dout("%s msg %p tid %llu\n", __func__, msg, tid); |
661 | ||
ba75bb98 SW |
662 | if (msg->front.iov_len != sizeof(*reply)) |
663 | goto bad; | |
ba75bb98 SW |
664 | |
665 | mutex_lock(&monc->mutex); | |
fcd00b68 | 666 | req = lookup_generic_request(&monc->generic_request_tree, tid); |
d0b19705 ID |
667 | if (!req) { |
668 | mutex_unlock(&monc->mutex); | |
669 | return; | |
ba75bb98 | 670 | } |
d0b19705 ID |
671 | |
672 | req->result = 0; | |
673 | *req->u.st = reply->st; /* struct */ | |
674 | __finish_generic_request(req); | |
ba75bb98 | 675 | mutex_unlock(&monc->mutex); |
d0b19705 ID |
676 | |
677 | complete_generic_request(req); | |
ba75bb98 SW |
678 | return; |
679 | ||
680 | bad: | |
7a6fdeb2 | 681 | pr_err("corrupt statfs reply, tid %llu\n", tid); |
9ec7cab1 | 682 | ceph_msg_dump(msg); |
ba75bb98 SW |
683 | } |
684 | ||
685 | /* | |
3143edd3 | 686 | * Do a synchronous statfs(). |
ba75bb98 | 687 | */ |
06d74376 DF |
688 | int ceph_monc_do_statfs(struct ceph_mon_client *monc, u64 data_pool, |
689 | struct ceph_statfs *buf) | |
ba75bb98 | 690 | { |
f8c76f6f | 691 | struct ceph_mon_generic_request *req; |
ba75bb98 | 692 | struct ceph_mon_statfs *h; |
d0b19705 | 693 | int ret = -ENOMEM; |
3143edd3 | 694 | |
d0b19705 | 695 | req = alloc_generic_request(monc, GFP_NOFS); |
3143edd3 | 696 | if (!req) |
d0b19705 | 697 | goto out; |
ba75bb98 | 698 | |
b61c2763 SW |
699 | req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS, |
700 | true); | |
a79832f2 | 701 | if (!req->request) |
3143edd3 | 702 | goto out; |
d0b19705 ID |
703 | |
704 | req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 64, GFP_NOFS, true); | |
a79832f2 | 705 | if (!req->reply) |
3143edd3 | 706 | goto out; |
3143edd3 | 707 | |
d0b19705 | 708 | req->u.st = buf; |
06d74376 | 709 | req->request->hdr.version = cpu_to_le16(2); |
d0b19705 ID |
710 | |
711 | mutex_lock(&monc->mutex); | |
712 | register_generic_request(req); | |
3143edd3 SW |
713 | /* fill out request */ |
714 | h = req->request->front.iov_base; | |
13e38c8a SW |
715 | h->monhdr.have_version = 0; |
716 | h->monhdr.session_mon = cpu_to_le16(-1); | |
717 | h->monhdr.session_mon_tid = 0; | |
ba75bb98 | 718 | h->fsid = monc->monmap->fsid; |
06d74376 DF |
719 | h->contains_data_pool = (data_pool != CEPH_NOPOOL); |
720 | h->data_pool = cpu_to_le64(data_pool); | |
d0b19705 ID |
721 | send_generic_request(monc, req); |
722 | mutex_unlock(&monc->mutex); | |
ba75bb98 | 723 | |
d0b19705 | 724 | ret = wait_generic_request(req); |
e56fa10e | 725 | out: |
f646912d | 726 | put_generic_request(req); |
d0b19705 | 727 | return ret; |
e56fa10e | 728 | } |
3d14c5d2 | 729 | EXPORT_SYMBOL(ceph_monc_do_statfs); |
e56fa10e | 730 | |
513a8243 ID |
731 | static void handle_get_version_reply(struct ceph_mon_client *monc, |
732 | struct ceph_msg *msg) | |
733 | { | |
734 | struct ceph_mon_generic_request *req; | |
735 | u64 tid = le64_to_cpu(msg->hdr.tid); | |
736 | void *p = msg->front.iov_base; | |
737 | void *end = p + msg->front_alloc_len; | |
738 | u64 handle; | |
739 | ||
d0b19705 | 740 | dout("%s msg %p tid %llu\n", __func__, msg, tid); |
513a8243 ID |
741 | |
742 | ceph_decode_need(&p, end, 2*sizeof(u64), bad); | |
743 | handle = ceph_decode_64(&p); | |
744 | if (tid != 0 && tid != handle) | |
745 | goto bad; | |
746 | ||
747 | mutex_lock(&monc->mutex); | |
fcd00b68 | 748 | req = lookup_generic_request(&monc->generic_request_tree, handle); |
d0b19705 ID |
749 | if (!req) { |
750 | mutex_unlock(&monc->mutex); | |
751 | return; | |
513a8243 | 752 | } |
d0b19705 ID |
753 | |
754 | req->result = 0; | |
755 | req->u.newest = ceph_decode_64(&p); | |
756 | __finish_generic_request(req); | |
513a8243 | 757 | mutex_unlock(&monc->mutex); |
513a8243 | 758 | |
d0b19705 | 759 | complete_generic_request(req); |
513a8243 | 760 | return; |
d0b19705 | 761 | |
513a8243 | 762 | bad: |
7a6fdeb2 | 763 | pr_err("corrupt mon_get_version reply, tid %llu\n", tid); |
513a8243 ID |
764 | ceph_msg_dump(msg); |
765 | } | |
766 | ||
d0b19705 ID |
767 | static struct ceph_mon_generic_request * |
768 | __ceph_monc_get_version(struct ceph_mon_client *monc, const char *what, | |
769 | ceph_monc_callback_t cb, u64 private_data) | |
513a8243 ID |
770 | { |
771 | struct ceph_mon_generic_request *req; | |
513a8243 | 772 | |
d0b19705 | 773 | req = alloc_generic_request(monc, GFP_NOIO); |
513a8243 | 774 | if (!req) |
d0b19705 | 775 | goto err_put_req; |
513a8243 ID |
776 | |
777 | req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION, | |
778 | sizeof(u64) + sizeof(u32) + strlen(what), | |
d0b19705 ID |
779 | GFP_NOIO, true); |
780 | if (!req->request) | |
781 | goto err_put_req; | |
513a8243 | 782 | |
d0b19705 ID |
783 | req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 32, GFP_NOIO, |
784 | true); | |
785 | if (!req->reply) | |
786 | goto err_put_req; | |
513a8243 | 787 | |
d0b19705 ID |
788 | req->complete_cb = cb; |
789 | req->private_data = private_data; | |
513a8243 | 790 | |
513a8243 | 791 | mutex_lock(&monc->mutex); |
d0b19705 ID |
792 | register_generic_request(req); |
793 | { | |
794 | void *p = req->request->front.iov_base; | |
795 | void *const end = p + req->request->front_alloc_len; | |
796 | ||
797 | ceph_encode_64(&p, req->tid); /* handle */ | |
798 | ceph_encode_string(&p, end, what, strlen(what)); | |
799 | WARN_ON(p != end); | |
800 | } | |
801 | send_generic_request(monc, req); | |
802 | mutex_unlock(&monc->mutex); | |
513a8243 | 803 | |
d0b19705 | 804 | return req; |
513a8243 | 805 | |
d0b19705 | 806 | err_put_req: |
f646912d | 807 | put_generic_request(req); |
d0b19705 | 808 | return ERR_PTR(-ENOMEM); |
513a8243 | 809 | } |
d0b19705 ID |
810 | |
811 | /* | |
812 | * Send MMonGetVersion and wait for the reply. | |
813 | * | |
814 | * @what: one of "mdsmap", "osdmap" or "monmap" | |
815 | */ | |
816 | int ceph_monc_get_version(struct ceph_mon_client *monc, const char *what, | |
817 | u64 *newest) | |
818 | { | |
819 | struct ceph_mon_generic_request *req; | |
820 | int ret; | |
821 | ||
822 | req = __ceph_monc_get_version(monc, what, NULL, 0); | |
823 | if (IS_ERR(req)) | |
824 | return PTR_ERR(req); | |
825 | ||
826 | ret = wait_generic_request(req); | |
827 | if (!ret) | |
828 | *newest = req->u.newest; | |
829 | ||
830 | put_generic_request(req); | |
831 | return ret; | |
832 | } | |
833 | EXPORT_SYMBOL(ceph_monc_get_version); | |
834 | ||
835 | /* | |
836 | * Send MMonGetVersion, | |
837 | * | |
838 | * @what: one of "mdsmap", "osdmap" or "monmap" | |
839 | */ | |
840 | int ceph_monc_get_version_async(struct ceph_mon_client *monc, const char *what, | |
841 | ceph_monc_callback_t cb, u64 private_data) | |
842 | { | |
843 | struct ceph_mon_generic_request *req; | |
844 | ||
845 | req = __ceph_monc_get_version(monc, what, cb, private_data); | |
846 | if (IS_ERR(req)) | |
847 | return PTR_ERR(req); | |
848 | ||
849 | put_generic_request(req); | |
850 | return 0; | |
851 | } | |
852 | EXPORT_SYMBOL(ceph_monc_get_version_async); | |
513a8243 | 853 | |
6305a3b4 DF |
854 | static void handle_command_ack(struct ceph_mon_client *monc, |
855 | struct ceph_msg *msg) | |
856 | { | |
857 | struct ceph_mon_generic_request *req; | |
858 | void *p = msg->front.iov_base; | |
859 | void *const end = p + msg->front_alloc_len; | |
860 | u64 tid = le64_to_cpu(msg->hdr.tid); | |
861 | ||
862 | dout("%s msg %p tid %llu\n", __func__, msg, tid); | |
863 | ||
864 | ceph_decode_need(&p, end, sizeof(struct ceph_mon_request_header) + | |
865 | sizeof(u32), bad); | |
866 | p += sizeof(struct ceph_mon_request_header); | |
867 | ||
868 | mutex_lock(&monc->mutex); | |
869 | req = lookup_generic_request(&monc->generic_request_tree, tid); | |
870 | if (!req) { | |
871 | mutex_unlock(&monc->mutex); | |
872 | return; | |
873 | } | |
874 | ||
875 | req->result = ceph_decode_32(&p); | |
876 | __finish_generic_request(req); | |
877 | mutex_unlock(&monc->mutex); | |
878 | ||
879 | complete_generic_request(req); | |
880 | return; | |
881 | ||
882 | bad: | |
883 | pr_err("corrupt mon_command ack, tid %llu\n", tid); | |
884 | ceph_msg_dump(msg); | |
885 | } | |
886 | ||
887 | int ceph_monc_blacklist_add(struct ceph_mon_client *monc, | |
888 | struct ceph_entity_addr *client_addr) | |
889 | { | |
890 | struct ceph_mon_generic_request *req; | |
891 | struct ceph_mon_command *h; | |
892 | int ret = -ENOMEM; | |
893 | int len; | |
894 | ||
895 | req = alloc_generic_request(monc, GFP_NOIO); | |
896 | if (!req) | |
897 | goto out; | |
898 | ||
899 | req->request = ceph_msg_new(CEPH_MSG_MON_COMMAND, 256, GFP_NOIO, true); | |
900 | if (!req->request) | |
901 | goto out; | |
902 | ||
903 | req->reply = ceph_msg_new(CEPH_MSG_MON_COMMAND_ACK, 512, GFP_NOIO, | |
904 | true); | |
905 | if (!req->reply) | |
906 | goto out; | |
907 | ||
908 | mutex_lock(&monc->mutex); | |
909 | register_generic_request(req); | |
910 | h = req->request->front.iov_base; | |
911 | h->monhdr.have_version = 0; | |
912 | h->monhdr.session_mon = cpu_to_le16(-1); | |
913 | h->monhdr.session_mon_tid = 0; | |
914 | h->fsid = monc->monmap->fsid; | |
915 | h->num_strs = cpu_to_le32(1); | |
916 | len = sprintf(h->str, "{ \"prefix\": \"osd blacklist\", \ | |
917 | \"blacklistop\": \"add\", \ | |
918 | \"addr\": \"%pISpc/%u\" }", | |
919 | &client_addr->in_addr, le32_to_cpu(client_addr->nonce)); | |
920 | h->str_len = cpu_to_le32(len); | |
921 | send_generic_request(monc, req); | |
922 | mutex_unlock(&monc->mutex); | |
923 | ||
924 | ret = wait_generic_request(req); | |
925 | out: | |
926 | put_generic_request(req); | |
927 | return ret; | |
928 | } | |
929 | EXPORT_SYMBOL(ceph_monc_blacklist_add); | |
930 | ||
ba75bb98 | 931 | /* |
e56fa10e | 932 | * Resend pending generic requests. |
ba75bb98 | 933 | */ |
f8c76f6f | 934 | static void __resend_generic_request(struct ceph_mon_client *monc) |
ba75bb98 | 935 | { |
f8c76f6f | 936 | struct ceph_mon_generic_request *req; |
85ff03f6 | 937 | struct rb_node *p; |
ba75bb98 | 938 | |
f8c76f6f YS |
939 | for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) { |
940 | req = rb_entry(p, struct ceph_mon_generic_request, node); | |
6740a845 | 941 | ceph_msg_revoke(req->request); |
4f471e4a | 942 | ceph_msg_revoke_incoming(req->reply); |
67130934 | 943 | ceph_con_send(&monc->con, ceph_msg_get(req->request)); |
ba75bb98 SW |
944 | } |
945 | } | |
946 | ||
947 | /* | |
948 | * Delayed work. If we haven't mounted yet, retry. Otherwise, | |
949 | * renew/retry subscription as needed (in case it is timing out, or we | |
950 | * got an ENOMEM). And keep the monitor connection alive. | |
951 | */ | |
952 | static void delayed_work(struct work_struct *work) | |
953 | { | |
954 | struct ceph_mon_client *monc = | |
955 | container_of(work, struct ceph_mon_client, delayed_work.work); | |
956 | ||
957 | dout("monc delayed_work\n"); | |
958 | mutex_lock(&monc->mutex); | |
4e7a5dcd | 959 | if (monc->hunting) { |
1752b50c ID |
960 | dout("%s continuing hunt\n", __func__); |
961 | reopen_session(monc); | |
ba75bb98 | 962 | } else { |
8b9558aa YZ |
963 | int is_auth = ceph_auth_is_authenticated(monc->auth); |
964 | if (ceph_con_keepalive_expired(&monc->con, | |
58d81b12 | 965 | CEPH_MONC_PING_TIMEOUT)) { |
8b9558aa YZ |
966 | dout("monc keepalive timeout\n"); |
967 | is_auth = 0; | |
1752b50c | 968 | reopen_session(monc); |
8b9558aa | 969 | } |
9bd2e6f8 | 970 | |
8b9558aa YZ |
971 | if (!monc->hunting) { |
972 | ceph_con_keepalive(&monc->con); | |
973 | __validate_auth(monc); | |
facb9f6e | 974 | un_backoff(monc); |
8b9558aa | 975 | } |
9bd2e6f8 | 976 | |
220abf5a ID |
977 | if (is_auth && |
978 | !(monc->con.peer_features & CEPH_FEATURE_MON_STATEFUL_SUB)) { | |
82dcabad ID |
979 | unsigned long now = jiffies; |
980 | ||
981 | dout("%s renew subs? now %lu renew after %lu\n", | |
982 | __func__, now, monc->sub_renew_after); | |
983 | if (time_after_eq(now, monc->sub_renew_after)) | |
984 | __send_subscribe(monc); | |
985 | } | |
ba75bb98 | 986 | } |
ba75bb98 SW |
987 | __schedule_delayed(monc); |
988 | mutex_unlock(&monc->mutex); | |
989 | } | |
990 | ||
6b805185 SW |
991 | /* |
992 | * On startup, we build a temporary monmap populated with the IPs | |
993 | * provided by mount(2). | |
994 | */ | |
995 | static int build_initial_monmap(struct ceph_mon_client *monc) | |
996 | { | |
3d14c5d2 YS |
997 | struct ceph_options *opt = monc->client->options; |
998 | struct ceph_entity_addr *mon_addr = opt->mon_addr; | |
999 | int num_mon = opt->num_mon; | |
6b805185 SW |
1000 | int i; |
1001 | ||
1002 | /* build initial monmap */ | |
acafe7e3 | 1003 | monc->monmap = kzalloc(struct_size(monc->monmap, mon_inst, num_mon), |
6b805185 SW |
1004 | GFP_KERNEL); |
1005 | if (!monc->monmap) | |
1006 | return -ENOMEM; | |
1007 | for (i = 0; i < num_mon; i++) { | |
1008 | monc->monmap->mon_inst[i].addr = mon_addr[i]; | |
6b805185 SW |
1009 | monc->monmap->mon_inst[i].addr.nonce = 0; |
1010 | monc->monmap->mon_inst[i].name.type = | |
1011 | CEPH_ENTITY_TYPE_MON; | |
1012 | monc->monmap->mon_inst[i].name.num = cpu_to_le64(i); | |
1013 | } | |
1014 | monc->monmap->num_mon = num_mon; | |
6b805185 SW |
1015 | return 0; |
1016 | } | |
1017 | ||
ba75bb98 SW |
1018 | int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl) |
1019 | { | |
1020 | int err = 0; | |
1021 | ||
1022 | dout("init\n"); | |
1023 | memset(monc, 0, sizeof(*monc)); | |
1024 | monc->client = cl; | |
1025 | monc->monmap = NULL; | |
1026 | mutex_init(&monc->mutex); | |
1027 | ||
6b805185 SW |
1028 | err = build_initial_monmap(monc); |
1029 | if (err) | |
1030 | goto out; | |
1031 | ||
f6a2f5be | 1032 | /* connection */ |
4e7a5dcd | 1033 | /* authentication */ |
3d14c5d2 | 1034 | monc->auth = ceph_auth_init(cl->options->name, |
8323c3aa | 1035 | cl->options->key); |
49d9224c NW |
1036 | if (IS_ERR(monc->auth)) { |
1037 | err = PTR_ERR(monc->auth); | |
67130934 | 1038 | goto out_monmap; |
49d9224c | 1039 | } |
4e7a5dcd SW |
1040 | monc->auth->want_keys = |
1041 | CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON | | |
1042 | CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS; | |
1043 | ||
240ed68e | 1044 | /* msgs */ |
a79832f2 | 1045 | err = -ENOMEM; |
7c315c55 | 1046 | monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK, |
34d23762 | 1047 | sizeof(struct ceph_mon_subscribe_ack), |
5418d0a2 | 1048 | GFP_KERNEL, true); |
a79832f2 | 1049 | if (!monc->m_subscribe_ack) |
49d9224c | 1050 | goto out_auth; |
6694d6b9 | 1051 | |
5418d0a2 ID |
1052 | monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 128, |
1053 | GFP_KERNEL, true); | |
240ed68e SW |
1054 | if (!monc->m_subscribe) |
1055 | goto out_subscribe_ack; | |
1056 | ||
5418d0a2 ID |
1057 | monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, |
1058 | GFP_KERNEL, true); | |
a79832f2 | 1059 | if (!monc->m_auth_reply) |
240ed68e | 1060 | goto out_subscribe; |
4e7a5dcd | 1061 | |
5418d0a2 | 1062 | monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_KERNEL, true); |
9bd2e6f8 | 1063 | monc->pending_auth = 0; |
a79832f2 | 1064 | if (!monc->m_auth) |
6694d6b9 | 1065 | goto out_auth_reply; |
ba75bb98 | 1066 | |
735a72ef SW |
1067 | ceph_con_init(&monc->con, monc, &mon_con_ops, |
1068 | &monc->client->msgr); | |
1069 | ||
ba75bb98 | 1070 | monc->cur_mon = -1; |
168b9090 ID |
1071 | monc->had_a_connection = false; |
1072 | monc->hunt_mult = 1; | |
ba75bb98 SW |
1073 | |
1074 | INIT_DELAYED_WORK(&monc->delayed_work, delayed_work); | |
f8c76f6f | 1075 | monc->generic_request_tree = RB_ROOT; |
ba75bb98 SW |
1076 | monc->last_tid = 0; |
1077 | ||
737cc81e ID |
1078 | monc->fs_cluster_id = CEPH_FS_CLUSTER_ID_NONE; |
1079 | ||
4e7a5dcd SW |
1080 | return 0; |
1081 | ||
6694d6b9 SW |
1082 | out_auth_reply: |
1083 | ceph_msg_put(monc->m_auth_reply); | |
240ed68e SW |
1084 | out_subscribe: |
1085 | ceph_msg_put(monc->m_subscribe); | |
7c315c55 SW |
1086 | out_subscribe_ack: |
1087 | ceph_msg_put(monc->m_subscribe_ack); | |
49d9224c NW |
1088 | out_auth: |
1089 | ceph_auth_destroy(monc->auth); | |
4e7a5dcd SW |
1090 | out_monmap: |
1091 | kfree(monc->monmap); | |
ba75bb98 SW |
1092 | out: |
1093 | return err; | |
1094 | } | |
3d14c5d2 | 1095 | EXPORT_SYMBOL(ceph_monc_init); |
ba75bb98 SW |
1096 | |
1097 | void ceph_monc_stop(struct ceph_mon_client *monc) | |
1098 | { | |
1099 | dout("stop\n"); | |
1100 | cancel_delayed_work_sync(&monc->delayed_work); | |
1101 | ||
1102 | mutex_lock(&monc->mutex); | |
1103 | __close_session(monc); | |
0e04dc26 | 1104 | monc->cur_mon = -1; |
ba75bb98 SW |
1105 | mutex_unlock(&monc->mutex); |
1106 | ||
f3dea7ed SW |
1107 | /* |
1108 | * flush msgr queue before we destroy ourselves to ensure that: | |
1109 | * - any work that references our embedded con is finished. | |
1110 | * - any osd_client or other work that may reference an authorizer | |
1111 | * finishes before we shut down the auth subsystem. | |
1112 | */ | |
1113 | ceph_msgr_flush(); | |
1114 | ||
4e7a5dcd SW |
1115 | ceph_auth_destroy(monc->auth); |
1116 | ||
d0b19705 ID |
1117 | WARN_ON(!RB_EMPTY_ROOT(&monc->generic_request_tree)); |
1118 | ||
4e7a5dcd | 1119 | ceph_msg_put(monc->m_auth); |
6694d6b9 | 1120 | ceph_msg_put(monc->m_auth_reply); |
240ed68e | 1121 | ceph_msg_put(monc->m_subscribe); |
7c315c55 | 1122 | ceph_msg_put(monc->m_subscribe_ack); |
ba75bb98 SW |
1123 | |
1124 | kfree(monc->monmap); | |
1125 | } | |
3d14c5d2 | 1126 | EXPORT_SYMBOL(ceph_monc_stop); |
ba75bb98 | 1127 | |
0f9af169 ID |
1128 | static void finish_hunting(struct ceph_mon_client *monc) |
1129 | { | |
1130 | if (monc->hunting) { | |
1131 | dout("%s found mon%d\n", __func__, monc->cur_mon); | |
1132 | monc->hunting = false; | |
168b9090 | 1133 | monc->had_a_connection = true; |
facb9f6e | 1134 | un_backoff(monc); |
7b4c443d | 1135 | __schedule_delayed(monc); |
0f9af169 ID |
1136 | } |
1137 | } | |
1138 | ||
4e7a5dcd SW |
1139 | static void handle_auth_reply(struct ceph_mon_client *monc, |
1140 | struct ceph_msg *msg) | |
1141 | { | |
1142 | int ret; | |
09c4d6a7 | 1143 | int was_auth = 0; |
4e7a5dcd SW |
1144 | |
1145 | mutex_lock(&monc->mutex); | |
27859f97 | 1146 | was_auth = ceph_auth_is_authenticated(monc->auth); |
9bd2e6f8 | 1147 | monc->pending_auth = 0; |
4e7a5dcd SW |
1148 | ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base, |
1149 | msg->front.iov_len, | |
1150 | monc->m_auth->front.iov_base, | |
3cea4c30 | 1151 | monc->m_auth->front_alloc_len); |
0f9af169 ID |
1152 | if (ret > 0) { |
1153 | __send_prepared_auth_request(monc, ret); | |
1154 | goto out; | |
1155 | } | |
1156 | ||
1157 | finish_hunting(monc); | |
1158 | ||
4e7a5dcd | 1159 | if (ret < 0) { |
9bd2e6f8 | 1160 | monc->client->auth_err = ret; |
27859f97 | 1161 | } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) { |
4e7a5dcd | 1162 | dout("authenticated, starting session\n"); |
0743304d | 1163 | |
15d9882c AE |
1164 | monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT; |
1165 | monc->client->msgr.inst.name.num = | |
0cf5537b | 1166 | cpu_to_le64(monc->auth->global_id); |
0743304d | 1167 | |
4e7a5dcd | 1168 | __send_subscribe(monc); |
f8c76f6f | 1169 | __resend_generic_request(monc); |
0f9af169 ID |
1170 | |
1171 | pr_info("mon%d %s session established\n", monc->cur_mon, | |
1172 | ceph_pr_addr(&monc->con.peer_addr.in_addr)); | |
4e7a5dcd | 1173 | } |
0f9af169 ID |
1174 | |
1175 | out: | |
4e7a5dcd | 1176 | mutex_unlock(&monc->mutex); |
0f9af169 ID |
1177 | if (monc->client->auth_err < 0) |
1178 | wake_up_all(&monc->client->auth_wq); | |
4e7a5dcd SW |
1179 | } |
1180 | ||
9bd2e6f8 SW |
1181 | static int __validate_auth(struct ceph_mon_client *monc) |
1182 | { | |
1183 | int ret; | |
1184 | ||
1185 | if (monc->pending_auth) | |
1186 | return 0; | |
1187 | ||
1188 | ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base, | |
3cea4c30 | 1189 | monc->m_auth->front_alloc_len); |
9bd2e6f8 SW |
1190 | if (ret <= 0) |
1191 | return ret; /* either an error, or no need to authenticate */ | |
1192 | __send_prepared_auth_request(monc, ret); | |
1193 | return 0; | |
1194 | } | |
1195 | ||
1196 | int ceph_monc_validate_auth(struct ceph_mon_client *monc) | |
1197 | { | |
1198 | int ret; | |
1199 | ||
1200 | mutex_lock(&monc->mutex); | |
1201 | ret = __validate_auth(monc); | |
1202 | mutex_unlock(&monc->mutex); | |
1203 | return ret; | |
1204 | } | |
3d14c5d2 | 1205 | EXPORT_SYMBOL(ceph_monc_validate_auth); |
9bd2e6f8 | 1206 | |
ba75bb98 SW |
1207 | /* |
1208 | * handle incoming message | |
1209 | */ | |
1210 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
1211 | { | |
1212 | struct ceph_mon_client *monc = con->private; | |
1213 | int type = le16_to_cpu(msg->hdr.type); | |
1214 | ||
1215 | if (!monc) | |
1216 | return; | |
1217 | ||
1218 | switch (type) { | |
4e7a5dcd SW |
1219 | case CEPH_MSG_AUTH_REPLY: |
1220 | handle_auth_reply(monc, msg); | |
ba75bb98 SW |
1221 | break; |
1222 | ||
1223 | case CEPH_MSG_MON_SUBSCRIBE_ACK: | |
1224 | handle_subscribe_ack(monc, msg); | |
1225 | break; | |
1226 | ||
1227 | case CEPH_MSG_STATFS_REPLY: | |
1228 | handle_statfs_reply(monc, msg); | |
1229 | break; | |
1230 | ||
513a8243 ID |
1231 | case CEPH_MSG_MON_GET_VERSION_REPLY: |
1232 | handle_get_version_reply(monc, msg); | |
1233 | break; | |
1234 | ||
6305a3b4 DF |
1235 | case CEPH_MSG_MON_COMMAND_ACK: |
1236 | handle_command_ack(monc, msg); | |
1237 | break; | |
1238 | ||
4e7a5dcd SW |
1239 | case CEPH_MSG_MON_MAP: |
1240 | ceph_monc_handle_map(monc, msg); | |
1241 | break; | |
1242 | ||
ba75bb98 SW |
1243 | case CEPH_MSG_OSD_MAP: |
1244 | ceph_osdc_handle_map(&monc->client->osdc, msg); | |
1245 | break; | |
1246 | ||
1247 | default: | |
3d14c5d2 YS |
1248 | /* can the chained handler handle it? */ |
1249 | if (monc->client->extra_mon_dispatch && | |
1250 | monc->client->extra_mon_dispatch(monc->client, msg) == 0) | |
1251 | break; | |
1252 | ||
ba75bb98 SW |
1253 | pr_err("received unknown message type %d %s\n", type, |
1254 | ceph_msg_type_name(type)); | |
1255 | } | |
1256 | ceph_msg_put(msg); | |
1257 | } | |
1258 | ||
1259 | /* | |
1260 | * Allocate memory for incoming message | |
1261 | */ | |
1262 | static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con, | |
2450418c YS |
1263 | struct ceph_msg_header *hdr, |
1264 | int *skip) | |
ba75bb98 SW |
1265 | { |
1266 | struct ceph_mon_client *monc = con->private; | |
1267 | int type = le16_to_cpu(hdr->type); | |
2450418c | 1268 | int front_len = le32_to_cpu(hdr->front_len); |
5b3a4db3 | 1269 | struct ceph_msg *m = NULL; |
ba75bb98 | 1270 | |
2450418c | 1271 | *skip = 0; |
0547a9b3 | 1272 | |
ba75bb98 | 1273 | switch (type) { |
ba75bb98 | 1274 | case CEPH_MSG_MON_SUBSCRIBE_ACK: |
7c315c55 | 1275 | m = ceph_msg_get(monc->m_subscribe_ack); |
2450418c | 1276 | break; |
ba75bb98 | 1277 | case CEPH_MSG_STATFS_REPLY: |
6305a3b4 | 1278 | case CEPH_MSG_MON_COMMAND_ACK: |
f8c76f6f | 1279 | return get_generic_reply(con, hdr, skip); |
4e7a5dcd | 1280 | case CEPH_MSG_AUTH_REPLY: |
6694d6b9 | 1281 | m = ceph_msg_get(monc->m_auth_reply); |
2450418c | 1282 | break; |
513a8243 ID |
1283 | case CEPH_MSG_MON_GET_VERSION_REPLY: |
1284 | if (le64_to_cpu(hdr->tid) != 0) | |
1285 | return get_generic_reply(con, hdr, skip); | |
1286 | ||
1287 | /* | |
1288 | * Older OSDs don't set reply tid even if the orignal | |
18370b36 GS |
1289 | * request had a non-zero tid. Work around this weirdness |
1290 | * by allocating a new message. | |
513a8243 | 1291 | */ |
18370b36 | 1292 | /* fall through */ |
5b3a4db3 SW |
1293 | case CEPH_MSG_MON_MAP: |
1294 | case CEPH_MSG_MDS_MAP: | |
1295 | case CEPH_MSG_OSD_MAP: | |
0cabbd94 | 1296 | case CEPH_MSG_FS_MAP_USER: |
b61c2763 | 1297 | m = ceph_msg_new(type, front_len, GFP_NOFS, false); |
1c20f2d2 AE |
1298 | if (!m) |
1299 | return NULL; /* ENOMEM--return skip == 0 */ | |
5b3a4db3 | 1300 | break; |
ba75bb98 | 1301 | } |
2450418c | 1302 | |
5b3a4db3 SW |
1303 | if (!m) { |
1304 | pr_info("alloc_msg unknown type %d\n", type); | |
2450418c | 1305 | *skip = 1; |
73c3d481 | 1306 | } else if (front_len > m->front_alloc_len) { |
b9a67899 JP |
1307 | pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n", |
1308 | front_len, m->front_alloc_len, | |
1309 | (unsigned int)con->peer_name.type, | |
1310 | le64_to_cpu(con->peer_name.num)); | |
73c3d481 SW |
1311 | ceph_msg_put(m); |
1312 | m = ceph_msg_new(type, front_len, GFP_NOFS, false); | |
5b3a4db3 | 1313 | } |
73c3d481 | 1314 | |
2450418c | 1315 | return m; |
ba75bb98 SW |
1316 | } |
1317 | ||
1318 | /* | |
1319 | * If the monitor connection resets, pick a new monitor and resubmit | |
1320 | * any pending requests. | |
1321 | */ | |
1322 | static void mon_fault(struct ceph_connection *con) | |
1323 | { | |
1324 | struct ceph_mon_client *monc = con->private; | |
1325 | ||
ba75bb98 | 1326 | mutex_lock(&monc->mutex); |
b5d91704 ID |
1327 | dout("%s mon%d\n", __func__, monc->cur_mon); |
1328 | if (monc->cur_mon >= 0) { | |
1329 | if (!monc->hunting) { | |
1330 | dout("%s hunting for new mon\n", __func__); | |
1331 | reopen_session(monc); | |
1332 | __schedule_delayed(monc); | |
1333 | } else { | |
1334 | dout("%s already hunting\n", __func__); | |
1335 | } | |
ba75bb98 | 1336 | } |
ba75bb98 SW |
1337 | mutex_unlock(&monc->mutex); |
1338 | } | |
1339 | ||
ec87ef43 SW |
1340 | /* |
1341 | * We can ignore refcounting on the connection struct, as all references | |
1342 | * will come from the messenger workqueue, which is drained prior to | |
1343 | * mon_client destruction. | |
1344 | */ | |
1345 | static struct ceph_connection *con_get(struct ceph_connection *con) | |
1346 | { | |
1347 | return con; | |
1348 | } | |
1349 | ||
1350 | static void con_put(struct ceph_connection *con) | |
1351 | { | |
1352 | } | |
1353 | ||
9e32789f | 1354 | static const struct ceph_connection_operations mon_con_ops = { |
ec87ef43 SW |
1355 | .get = con_get, |
1356 | .put = con_put, | |
ba75bb98 SW |
1357 | .dispatch = dispatch, |
1358 | .fault = mon_fault, | |
1359 | .alloc_msg = mon_alloc_msg, | |
ba75bb98 | 1360 | }; |