Commit | Line | Data |
---|---|---|
ba75bb98 SW |
1 | #include "ceph_debug.h" |
2 | ||
3 | #include <linux/types.h> | |
5a0e3ad6 | 4 | #include <linux/slab.h> |
ba75bb98 SW |
5 | #include <linux/random.h> |
6 | #include <linux/sched.h> | |
7 | ||
8 | #include "mon_client.h" | |
9 | #include "super.h" | |
4e7a5dcd | 10 | #include "auth.h" |
ba75bb98 SW |
11 | #include "decode.h" |
12 | ||
13 | /* | |
14 | * Interact with Ceph monitor cluster. Handle requests for new map | |
15 | * versions, and periodically resend as needed. Also implement | |
16 | * statfs() and umount(). | |
17 | * | |
18 | * A small cluster of Ceph "monitors" are responsible for managing critical | |
19 | * cluster configuration and state information. An odd number (e.g., 3, 5) | |
20 | * of cmon daemons use a modified version of the Paxos part-time parliament | |
21 | * algorithm to manage the MDS map (mds cluster membership), OSD map, and | |
22 | * list of clients who have mounted the file system. | |
23 | * | |
24 | * We maintain an open, active session with a monitor at all times in order to | |
25 | * receive timely MDSMap updates. We periodically send a keepalive byte on the | |
26 | * TCP socket to ensure we detect a failure. If the connection does break, we | |
27 | * randomly hunt for a new monitor. Once the connection is reestablished, we | |
28 | * resend any outstanding requests. | |
29 | */ | |
30 | ||
31 | const static struct ceph_connection_operations mon_con_ops; | |
32 | ||
9bd2e6f8 SW |
33 | static int __validate_auth(struct ceph_mon_client *monc); |
34 | ||
ba75bb98 SW |
35 | /* |
36 | * Decode a monmap blob (e.g., during mount). | |
37 | */ | |
38 | struct ceph_monmap *ceph_monmap_decode(void *p, void *end) | |
39 | { | |
40 | struct ceph_monmap *m = NULL; | |
41 | int i, err = -EINVAL; | |
42 | struct ceph_fsid fsid; | |
43 | u32 epoch, num_mon; | |
44 | u16 version; | |
4e7a5dcd SW |
45 | u32 len; |
46 | ||
47 | ceph_decode_32_safe(&p, end, len, bad); | |
48 | ceph_decode_need(&p, end, len, bad); | |
ba75bb98 SW |
49 | |
50 | dout("monmap_decode %p %p len %d\n", p, end, (int)(end-p)); | |
51 | ||
52 | ceph_decode_16_safe(&p, end, version, bad); | |
53 | ||
54 | ceph_decode_need(&p, end, sizeof(fsid) + 2*sizeof(u32), bad); | |
55 | ceph_decode_copy(&p, &fsid, sizeof(fsid)); | |
c89136ea | 56 | epoch = ceph_decode_32(&p); |
ba75bb98 | 57 | |
c89136ea | 58 | num_mon = ceph_decode_32(&p); |
ba75bb98 SW |
59 | ceph_decode_need(&p, end, num_mon*sizeof(m->mon_inst[0]), bad); |
60 | ||
61 | if (num_mon >= CEPH_MAX_MON) | |
62 | goto bad; | |
63 | m = kmalloc(sizeof(*m) + sizeof(m->mon_inst[0])*num_mon, GFP_NOFS); | |
64 | if (m == NULL) | |
65 | return ERR_PTR(-ENOMEM); | |
66 | m->fsid = fsid; | |
67 | m->epoch = epoch; | |
68 | m->num_mon = num_mon; | |
69 | ceph_decode_copy(&p, m->mon_inst, num_mon*sizeof(m->mon_inst[0])); | |
63f2d211 SW |
70 | for (i = 0; i < num_mon; i++) |
71 | ceph_decode_addr(&m->mon_inst[i].addr); | |
ba75bb98 | 72 | |
ba75bb98 SW |
73 | dout("monmap_decode epoch %d, num_mon %d\n", m->epoch, |
74 | m->num_mon); | |
75 | for (i = 0; i < m->num_mon; i++) | |
76 | dout("monmap_decode mon%d is %s\n", i, | |
77 | pr_addr(&m->mon_inst[i].addr.in_addr)); | |
78 | return m; | |
79 | ||
80 | bad: | |
81 | dout("monmap_decode failed with %d\n", err); | |
82 | kfree(m); | |
83 | return ERR_PTR(err); | |
84 | } | |
85 | ||
86 | /* | |
87 | * return true if *addr is included in the monmap. | |
88 | */ | |
89 | int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr) | |
90 | { | |
91 | int i; | |
92 | ||
93 | for (i = 0; i < m->num_mon; i++) | |
103e2d3a | 94 | if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0) |
ba75bb98 SW |
95 | return 1; |
96 | return 0; | |
97 | } | |
98 | ||
5ce6e9db SW |
99 | /* |
100 | * Send an auth request. | |
101 | */ | |
102 | static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len) | |
103 | { | |
104 | monc->pending_auth = 1; | |
105 | monc->m_auth->front.iov_len = len; | |
106 | monc->m_auth->hdr.front_len = cpu_to_le32(len); | |
107 | ceph_msg_get(monc->m_auth); /* keep our ref */ | |
108 | ceph_con_send(monc->con, monc->m_auth); | |
109 | } | |
110 | ||
ba75bb98 SW |
111 | /* |
112 | * Close monitor session, if any. | |
113 | */ | |
114 | static void __close_session(struct ceph_mon_client *monc) | |
115 | { | |
116 | if (monc->con) { | |
117 | dout("__close_session closing mon%d\n", monc->cur_mon); | |
4e7a5dcd | 118 | ceph_con_revoke(monc->con, monc->m_auth); |
ba75bb98 SW |
119 | ceph_con_close(monc->con); |
120 | monc->cur_mon = -1; | |
9bd2e6f8 | 121 | monc->pending_auth = 0; |
4e7a5dcd | 122 | ceph_auth_reset(monc->auth); |
ba75bb98 SW |
123 | } |
124 | } | |
125 | ||
126 | /* | |
127 | * Open a session with a (new) monitor. | |
128 | */ | |
129 | static int __open_session(struct ceph_mon_client *monc) | |
130 | { | |
131 | char r; | |
4e7a5dcd | 132 | int ret; |
ba75bb98 SW |
133 | |
134 | if (monc->cur_mon < 0) { | |
135 | get_random_bytes(&r, 1); | |
136 | monc->cur_mon = r % monc->monmap->num_mon; | |
137 | dout("open_session num=%d r=%d -> mon%d\n", | |
138 | monc->monmap->num_mon, r, monc->cur_mon); | |
139 | monc->sub_sent = 0; | |
140 | monc->sub_renew_after = jiffies; /* i.e., expired */ | |
141 | monc->want_next_osdmap = !!monc->want_next_osdmap; | |
142 | ||
143 | dout("open_session mon%d opening\n", monc->cur_mon); | |
144 | monc->con->peer_name.type = CEPH_ENTITY_TYPE_MON; | |
145 | monc->con->peer_name.num = cpu_to_le64(monc->cur_mon); | |
146 | ceph_con_open(monc->con, | |
147 | &monc->monmap->mon_inst[monc->cur_mon].addr); | |
4e7a5dcd SW |
148 | |
149 | /* initiatiate authentication handshake */ | |
150 | ret = ceph_auth_build_hello(monc->auth, | |
151 | monc->m_auth->front.iov_base, | |
152 | monc->m_auth->front_max); | |
5ce6e9db | 153 | __send_prepared_auth_request(monc, ret); |
ba75bb98 SW |
154 | } else { |
155 | dout("open_session mon%d already open\n", monc->cur_mon); | |
156 | } | |
157 | return 0; | |
158 | } | |
159 | ||
160 | static bool __sub_expired(struct ceph_mon_client *monc) | |
161 | { | |
162 | return time_after_eq(jiffies, monc->sub_renew_after); | |
163 | } | |
164 | ||
165 | /* | |
166 | * Reschedule delayed work timer. | |
167 | */ | |
168 | static void __schedule_delayed(struct ceph_mon_client *monc) | |
169 | { | |
170 | unsigned delay; | |
171 | ||
4e7a5dcd | 172 | if (monc->cur_mon < 0 || __sub_expired(monc)) |
ba75bb98 SW |
173 | delay = 10 * HZ; |
174 | else | |
175 | delay = 20 * HZ; | |
176 | dout("__schedule_delayed after %u\n", delay); | |
177 | schedule_delayed_work(&monc->delayed_work, delay); | |
178 | } | |
179 | ||
180 | /* | |
181 | * Send subscribe request for mdsmap and/or osdmap. | |
182 | */ | |
183 | static void __send_subscribe(struct ceph_mon_client *monc) | |
184 | { | |
185 | dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n", | |
186 | (unsigned)monc->sub_sent, __sub_expired(monc), | |
187 | monc->want_next_osdmap); | |
188 | if ((__sub_expired(monc) && !monc->sub_sent) || | |
189 | monc->want_next_osdmap == 1) { | |
190 | struct ceph_msg *msg; | |
191 | struct ceph_mon_subscribe_item *i; | |
192 | void *p, *end; | |
193 | ||
4e7a5dcd | 194 | msg = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, 0, 0, NULL); |
ba75bb98 SW |
195 | if (!msg) |
196 | return; | |
197 | ||
198 | p = msg->front.iov_base; | |
199 | end = p + msg->front.iov_len; | |
200 | ||
201 | dout("__send_subscribe to 'mdsmap' %u+\n", | |
202 | (unsigned)monc->have_mdsmap); | |
203 | if (monc->want_next_osdmap) { | |
204 | dout("__send_subscribe to 'osdmap' %u\n", | |
205 | (unsigned)monc->have_osdmap); | |
4e7a5dcd | 206 | ceph_encode_32(&p, 3); |
ba75bb98 SW |
207 | ceph_encode_string(&p, end, "osdmap", 6); |
208 | i = p; | |
209 | i->have = cpu_to_le64(monc->have_osdmap); | |
210 | i->onetime = 1; | |
211 | p += sizeof(*i); | |
212 | monc->want_next_osdmap = 2; /* requested */ | |
213 | } else { | |
4e7a5dcd | 214 | ceph_encode_32(&p, 2); |
ba75bb98 SW |
215 | } |
216 | ceph_encode_string(&p, end, "mdsmap", 6); | |
217 | i = p; | |
218 | i->have = cpu_to_le64(monc->have_mdsmap); | |
219 | i->onetime = 0; | |
220 | p += sizeof(*i); | |
4e7a5dcd SW |
221 | ceph_encode_string(&p, end, "monmap", 6); |
222 | i = p; | |
223 | i->have = 0; | |
224 | i->onetime = 0; | |
225 | p += sizeof(*i); | |
ba75bb98 SW |
226 | |
227 | msg->front.iov_len = p - msg->front.iov_base; | |
228 | msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); | |
229 | ceph_con_send(monc->con, msg); | |
230 | ||
231 | monc->sub_sent = jiffies | 1; /* never 0 */ | |
232 | } | |
233 | } | |
234 | ||
235 | static void handle_subscribe_ack(struct ceph_mon_client *monc, | |
236 | struct ceph_msg *msg) | |
237 | { | |
238 | unsigned seconds; | |
07bd10fb SW |
239 | struct ceph_mon_subscribe_ack *h = msg->front.iov_base; |
240 | ||
241 | if (msg->front.iov_len < sizeof(*h)) | |
242 | goto bad; | |
243 | seconds = le32_to_cpu(h->duration); | |
ba75bb98 | 244 | |
ba75bb98 SW |
245 | mutex_lock(&monc->mutex); |
246 | if (monc->hunting) { | |
247 | pr_info("mon%d %s session established\n", | |
248 | monc->cur_mon, pr_addr(&monc->con->peer_addr.in_addr)); | |
249 | monc->hunting = false; | |
250 | } | |
251 | dout("handle_subscribe_ack after %d seconds\n", seconds); | |
0656d11b | 252 | monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1; |
ba75bb98 SW |
253 | monc->sub_sent = 0; |
254 | mutex_unlock(&monc->mutex); | |
255 | return; | |
256 | bad: | |
257 | pr_err("got corrupt subscribe-ack msg\n"); | |
9ec7cab1 | 258 | ceph_msg_dump(msg); |
ba75bb98 SW |
259 | } |
260 | ||
261 | /* | |
262 | * Keep track of which maps we have | |
263 | */ | |
264 | int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got) | |
265 | { | |
266 | mutex_lock(&monc->mutex); | |
267 | monc->have_mdsmap = got; | |
268 | mutex_unlock(&monc->mutex); | |
269 | return 0; | |
270 | } | |
271 | ||
272 | int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got) | |
273 | { | |
274 | mutex_lock(&monc->mutex); | |
275 | monc->have_osdmap = got; | |
276 | monc->want_next_osdmap = 0; | |
277 | mutex_unlock(&monc->mutex); | |
278 | return 0; | |
279 | } | |
280 | ||
281 | /* | |
282 | * Register interest in the next osdmap | |
283 | */ | |
284 | void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc) | |
285 | { | |
286 | dout("request_next_osdmap have %u\n", monc->have_osdmap); | |
287 | mutex_lock(&monc->mutex); | |
288 | if (!monc->want_next_osdmap) | |
289 | monc->want_next_osdmap = 1; | |
290 | if (monc->want_next_osdmap < 2) | |
291 | __send_subscribe(monc); | |
292 | mutex_unlock(&monc->mutex); | |
293 | } | |
294 | ||
4e7a5dcd | 295 | /* |
50b885b9 | 296 | * |
4e7a5dcd SW |
297 | */ |
298 | int ceph_monc_open_session(struct ceph_mon_client *monc) | |
ba75bb98 SW |
299 | { |
300 | if (!monc->con) { | |
301 | monc->con = kmalloc(sizeof(*monc->con), GFP_KERNEL); | |
302 | if (!monc->con) | |
303 | return -ENOMEM; | |
304 | ceph_con_init(monc->client->msgr, monc->con); | |
305 | monc->con->private = monc; | |
306 | monc->con->ops = &mon_con_ops; | |
307 | } | |
308 | ||
309 | mutex_lock(&monc->mutex); | |
4e7a5dcd | 310 | __open_session(monc); |
ba75bb98 SW |
311 | __schedule_delayed(monc); |
312 | mutex_unlock(&monc->mutex); | |
313 | return 0; | |
314 | } | |
315 | ||
4e7a5dcd SW |
316 | /* |
317 | * The monitor responds with mount ack indicate mount success. The | |
318 | * included client ticket allows the client to talk to MDSs and OSDs. | |
319 | */ | |
0743304d SW |
320 | static void ceph_monc_handle_map(struct ceph_mon_client *monc, |
321 | struct ceph_msg *msg) | |
4e7a5dcd SW |
322 | { |
323 | struct ceph_client *client = monc->client; | |
324 | struct ceph_monmap *monmap = NULL, *old = monc->monmap; | |
325 | void *p, *end; | |
326 | ||
327 | mutex_lock(&monc->mutex); | |
328 | ||
329 | dout("handle_monmap\n"); | |
330 | p = msg->front.iov_base; | |
331 | end = p + msg->front.iov_len; | |
332 | ||
333 | monmap = ceph_monmap_decode(p, end); | |
334 | if (IS_ERR(monmap)) { | |
335 | pr_err("problem decoding monmap, %d\n", | |
336 | (int)PTR_ERR(monmap)); | |
d4a780ce | 337 | goto out; |
4e7a5dcd | 338 | } |
0743304d SW |
339 | |
340 | if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) { | |
4e7a5dcd | 341 | kfree(monmap); |
d4a780ce | 342 | goto out; |
4e7a5dcd SW |
343 | } |
344 | ||
345 | client->monc.monmap = monmap; | |
4e7a5dcd SW |
346 | kfree(old); |
347 | ||
d4a780ce | 348 | out: |
4e7a5dcd | 349 | mutex_unlock(&monc->mutex); |
9bd2e6f8 | 350 | wake_up(&client->auth_wq); |
4e7a5dcd SW |
351 | } |
352 | ||
ba75bb98 SW |
353 | /* |
354 | * statfs | |
355 | */ | |
85ff03f6 SW |
356 | static struct ceph_mon_statfs_request *__lookup_statfs( |
357 | struct ceph_mon_client *monc, u64 tid) | |
358 | { | |
359 | struct ceph_mon_statfs_request *req; | |
360 | struct rb_node *n = monc->statfs_request_tree.rb_node; | |
361 | ||
362 | while (n) { | |
363 | req = rb_entry(n, struct ceph_mon_statfs_request, node); | |
364 | if (tid < req->tid) | |
365 | n = n->rb_left; | |
366 | else if (tid > req->tid) | |
367 | n = n->rb_right; | |
368 | else | |
369 | return req; | |
370 | } | |
371 | return NULL; | |
372 | } | |
373 | ||
374 | static void __insert_statfs(struct ceph_mon_client *monc, | |
375 | struct ceph_mon_statfs_request *new) | |
376 | { | |
377 | struct rb_node **p = &monc->statfs_request_tree.rb_node; | |
378 | struct rb_node *parent = NULL; | |
379 | struct ceph_mon_statfs_request *req = NULL; | |
380 | ||
381 | while (*p) { | |
382 | parent = *p; | |
383 | req = rb_entry(parent, struct ceph_mon_statfs_request, node); | |
384 | if (new->tid < req->tid) | |
385 | p = &(*p)->rb_left; | |
386 | else if (new->tid > req->tid) | |
387 | p = &(*p)->rb_right; | |
388 | else | |
389 | BUG(); | |
390 | } | |
391 | ||
392 | rb_link_node(&new->node, parent, p); | |
393 | rb_insert_color(&new->node, &monc->statfs_request_tree); | |
394 | } | |
395 | ||
ba75bb98 SW |
396 | static void handle_statfs_reply(struct ceph_mon_client *monc, |
397 | struct ceph_msg *msg) | |
398 | { | |
399 | struct ceph_mon_statfs_request *req; | |
400 | struct ceph_mon_statfs_reply *reply = msg->front.iov_base; | |
401 | u64 tid; | |
402 | ||
403 | if (msg->front.iov_len != sizeof(*reply)) | |
404 | goto bad; | |
6df058c0 | 405 | tid = le64_to_cpu(msg->hdr.tid); |
ba75bb98 SW |
406 | dout("handle_statfs_reply %p tid %llu\n", msg, tid); |
407 | ||
408 | mutex_lock(&monc->mutex); | |
85ff03f6 | 409 | req = __lookup_statfs(monc, tid); |
ba75bb98 SW |
410 | if (req) { |
411 | *req->buf = reply->st; | |
412 | req->result = 0; | |
413 | } | |
414 | mutex_unlock(&monc->mutex); | |
415 | if (req) | |
416 | complete(&req->completion); | |
417 | return; | |
418 | ||
419 | bad: | |
420 | pr_err("corrupt statfs reply, no tid\n"); | |
9ec7cab1 | 421 | ceph_msg_dump(msg); |
ba75bb98 SW |
422 | } |
423 | ||
424 | /* | |
425 | * (re)send a statfs request | |
426 | */ | |
427 | static int send_statfs(struct ceph_mon_client *monc, | |
428 | struct ceph_mon_statfs_request *req) | |
429 | { | |
430 | struct ceph_msg *msg; | |
431 | struct ceph_mon_statfs *h; | |
ba75bb98 SW |
432 | |
433 | dout("send_statfs tid %llu\n", req->tid); | |
ba75bb98 SW |
434 | msg = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), 0, 0, NULL); |
435 | if (IS_ERR(msg)) | |
436 | return PTR_ERR(msg); | |
437 | req->request = msg; | |
6df058c0 | 438 | msg->hdr.tid = cpu_to_le64(req->tid); |
ba75bb98 | 439 | h = msg->front.iov_base; |
13e38c8a SW |
440 | h->monhdr.have_version = 0; |
441 | h->monhdr.session_mon = cpu_to_le16(-1); | |
442 | h->monhdr.session_mon_tid = 0; | |
ba75bb98 | 443 | h->fsid = monc->monmap->fsid; |
ba75bb98 SW |
444 | ceph_con_send(monc->con, msg); |
445 | return 0; | |
446 | } | |
447 | ||
448 | /* | |
449 | * Do a synchronous statfs(). | |
450 | */ | |
451 | int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf) | |
452 | { | |
453 | struct ceph_mon_statfs_request req; | |
454 | int err; | |
455 | ||
456 | req.buf = buf; | |
457 | init_completion(&req.completion); | |
458 | ||
459 | /* allocate memory for reply */ | |
460 | err = ceph_msgpool_resv(&monc->msgpool_statfs_reply, 1); | |
461 | if (err) | |
462 | return err; | |
463 | ||
464 | /* register request */ | |
465 | mutex_lock(&monc->mutex); | |
466 | req.tid = ++monc->last_tid; | |
467 | req.last_attempt = jiffies; | |
468 | req.delay = BASE_DELAY_INTERVAL; | |
85ff03f6 | 469 | __insert_statfs(monc, &req); |
ba75bb98 SW |
470 | monc->num_statfs_requests++; |
471 | mutex_unlock(&monc->mutex); | |
472 | ||
473 | /* send request and wait */ | |
474 | err = send_statfs(monc, &req); | |
475 | if (!err) | |
476 | err = wait_for_completion_interruptible(&req.completion); | |
477 | ||
478 | mutex_lock(&monc->mutex); | |
85ff03f6 | 479 | rb_erase(&req.node, &monc->statfs_request_tree); |
ba75bb98 SW |
480 | monc->num_statfs_requests--; |
481 | ceph_msgpool_resv(&monc->msgpool_statfs_reply, -1); | |
482 | mutex_unlock(&monc->mutex); | |
483 | ||
484 | if (!err) | |
485 | err = req.result; | |
486 | return err; | |
487 | } | |
488 | ||
489 | /* | |
490 | * Resend pending statfs requests. | |
491 | */ | |
492 | static void __resend_statfs(struct ceph_mon_client *monc) | |
493 | { | |
ba75bb98 | 494 | struct ceph_mon_statfs_request *req; |
85ff03f6 | 495 | struct rb_node *p; |
ba75bb98 | 496 | |
85ff03f6 SW |
497 | for (p = rb_first(&monc->statfs_request_tree); p; p = rb_next(p)) { |
498 | req = rb_entry(p, struct ceph_mon_statfs_request, node); | |
ba75bb98 SW |
499 | send_statfs(monc, req); |
500 | } | |
501 | } | |
502 | ||
503 | /* | |
504 | * Delayed work. If we haven't mounted yet, retry. Otherwise, | |
505 | * renew/retry subscription as needed (in case it is timing out, or we | |
506 | * got an ENOMEM). And keep the monitor connection alive. | |
507 | */ | |
508 | static void delayed_work(struct work_struct *work) | |
509 | { | |
510 | struct ceph_mon_client *monc = | |
511 | container_of(work, struct ceph_mon_client, delayed_work.work); | |
512 | ||
513 | dout("monc delayed_work\n"); | |
514 | mutex_lock(&monc->mutex); | |
4e7a5dcd SW |
515 | if (monc->hunting) { |
516 | __close_session(monc); | |
517 | __open_session(monc); /* continue hunting */ | |
ba75bb98 | 518 | } else { |
4e7a5dcd | 519 | ceph_con_keepalive(monc->con); |
9bd2e6f8 SW |
520 | |
521 | __validate_auth(monc); | |
522 | ||
4e7a5dcd SW |
523 | if (monc->auth->ops->is_authenticated(monc->auth)) |
524 | __send_subscribe(monc); | |
ba75bb98 | 525 | } |
ba75bb98 SW |
526 | __schedule_delayed(monc); |
527 | mutex_unlock(&monc->mutex); | |
528 | } | |
529 | ||
6b805185 SW |
530 | /* |
531 | * On startup, we build a temporary monmap populated with the IPs | |
532 | * provided by mount(2). | |
533 | */ | |
534 | static int build_initial_monmap(struct ceph_mon_client *monc) | |
535 | { | |
536 | struct ceph_mount_args *args = monc->client->mount_args; | |
537 | struct ceph_entity_addr *mon_addr = args->mon_addr; | |
538 | int num_mon = args->num_mon; | |
539 | int i; | |
540 | ||
541 | /* build initial monmap */ | |
542 | monc->monmap = kzalloc(sizeof(*monc->monmap) + | |
543 | num_mon*sizeof(monc->monmap->mon_inst[0]), | |
544 | GFP_KERNEL); | |
545 | if (!monc->monmap) | |
546 | return -ENOMEM; | |
547 | for (i = 0; i < num_mon; i++) { | |
548 | monc->monmap->mon_inst[i].addr = mon_addr[i]; | |
6b805185 SW |
549 | monc->monmap->mon_inst[i].addr.nonce = 0; |
550 | monc->monmap->mon_inst[i].name.type = | |
551 | CEPH_ENTITY_TYPE_MON; | |
552 | monc->monmap->mon_inst[i].name.num = cpu_to_le64(i); | |
553 | } | |
554 | monc->monmap->num_mon = num_mon; | |
4e7a5dcd | 555 | monc->have_fsid = false; |
6b805185 SW |
556 | |
557 | /* release addr memory */ | |
558 | kfree(args->mon_addr); | |
559 | args->mon_addr = NULL; | |
560 | args->num_mon = 0; | |
561 | return 0; | |
562 | } | |
563 | ||
ba75bb98 SW |
564 | int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl) |
565 | { | |
566 | int err = 0; | |
567 | ||
568 | dout("init\n"); | |
569 | memset(monc, 0, sizeof(*monc)); | |
570 | monc->client = cl; | |
571 | monc->monmap = NULL; | |
572 | mutex_init(&monc->mutex); | |
573 | ||
6b805185 SW |
574 | err = build_initial_monmap(monc); |
575 | if (err) | |
576 | goto out; | |
577 | ||
ba75bb98 SW |
578 | monc->con = NULL; |
579 | ||
4e7a5dcd SW |
580 | /* authentication */ |
581 | monc->auth = ceph_auth_init(cl->mount_args->name, | |
582 | cl->mount_args->secret); | |
583 | if (IS_ERR(monc->auth)) | |
584 | return PTR_ERR(monc->auth); | |
585 | monc->auth->want_keys = | |
586 | CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON | | |
587 | CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS; | |
588 | ||
ba75bb98 | 589 | /* msg pools */ |
07bd10fb SW |
590 | err = ceph_msgpool_init(&monc->msgpool_subscribe_ack, |
591 | sizeof(struct ceph_mon_subscribe_ack), 1, false); | |
ba75bb98 | 592 | if (err < 0) |
4e7a5dcd | 593 | goto out_monmap; |
ba75bb98 SW |
594 | err = ceph_msgpool_init(&monc->msgpool_statfs_reply, |
595 | sizeof(struct ceph_mon_statfs_reply), 0, false); | |
596 | if (err < 0) | |
4e7a5dcd SW |
597 | goto out_pool1; |
598 | err = ceph_msgpool_init(&monc->msgpool_auth_reply, 4096, 1, false); | |
599 | if (err < 0) | |
600 | goto out_pool2; | |
601 | ||
602 | monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, 0, 0, NULL); | |
9bd2e6f8 | 603 | monc->pending_auth = 0; |
4e7a5dcd SW |
604 | if (IS_ERR(monc->m_auth)) { |
605 | err = PTR_ERR(monc->m_auth); | |
606 | monc->m_auth = NULL; | |
607 | goto out_pool3; | |
608 | } | |
ba75bb98 SW |
609 | |
610 | monc->cur_mon = -1; | |
4e7a5dcd | 611 | monc->hunting = true; |
ba75bb98 SW |
612 | monc->sub_renew_after = jiffies; |
613 | monc->sub_sent = 0; | |
614 | ||
615 | INIT_DELAYED_WORK(&monc->delayed_work, delayed_work); | |
85ff03f6 | 616 | monc->statfs_request_tree = RB_ROOT; |
ba75bb98 SW |
617 | monc->num_statfs_requests = 0; |
618 | monc->last_tid = 0; | |
619 | ||
620 | monc->have_mdsmap = 0; | |
621 | monc->have_osdmap = 0; | |
622 | monc->want_next_osdmap = 1; | |
4e7a5dcd SW |
623 | return 0; |
624 | ||
625 | out_pool3: | |
626 | ceph_msgpool_destroy(&monc->msgpool_auth_reply); | |
627 | out_pool2: | |
628 | ceph_msgpool_destroy(&monc->msgpool_subscribe_ack); | |
629 | out_pool1: | |
630 | ceph_msgpool_destroy(&monc->msgpool_statfs_reply); | |
631 | out_monmap: | |
632 | kfree(monc->monmap); | |
ba75bb98 SW |
633 | out: |
634 | return err; | |
635 | } | |
636 | ||
637 | void ceph_monc_stop(struct ceph_mon_client *monc) | |
638 | { | |
639 | dout("stop\n"); | |
640 | cancel_delayed_work_sync(&monc->delayed_work); | |
641 | ||
642 | mutex_lock(&monc->mutex); | |
643 | __close_session(monc); | |
644 | if (monc->con) { | |
645 | monc->con->private = NULL; | |
646 | monc->con->ops->put(monc->con); | |
647 | monc->con = NULL; | |
648 | } | |
649 | mutex_unlock(&monc->mutex); | |
650 | ||
4e7a5dcd SW |
651 | ceph_auth_destroy(monc->auth); |
652 | ||
653 | ceph_msg_put(monc->m_auth); | |
ba75bb98 SW |
654 | ceph_msgpool_destroy(&monc->msgpool_subscribe_ack); |
655 | ceph_msgpool_destroy(&monc->msgpool_statfs_reply); | |
4e7a5dcd | 656 | ceph_msgpool_destroy(&monc->msgpool_auth_reply); |
ba75bb98 SW |
657 | |
658 | kfree(monc->monmap); | |
659 | } | |
660 | ||
4e7a5dcd SW |
661 | static void handle_auth_reply(struct ceph_mon_client *monc, |
662 | struct ceph_msg *msg) | |
663 | { | |
664 | int ret; | |
665 | ||
666 | mutex_lock(&monc->mutex); | |
9bd2e6f8 | 667 | monc->pending_auth = 0; |
4e7a5dcd SW |
668 | ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base, |
669 | msg->front.iov_len, | |
670 | monc->m_auth->front.iov_base, | |
671 | monc->m_auth->front_max); | |
672 | if (ret < 0) { | |
9bd2e6f8 SW |
673 | monc->client->auth_err = ret; |
674 | wake_up(&monc->client->auth_wq); | |
4e7a5dcd | 675 | } else if (ret > 0) { |
9bd2e6f8 | 676 | __send_prepared_auth_request(monc, ret); |
4e7a5dcd SW |
677 | } else if (monc->auth->ops->is_authenticated(monc->auth)) { |
678 | dout("authenticated, starting session\n"); | |
0743304d SW |
679 | |
680 | monc->client->msgr->inst.name.type = CEPH_ENTITY_TYPE_CLIENT; | |
681 | monc->client->msgr->inst.name.num = monc->auth->global_id; | |
682 | ||
4e7a5dcd SW |
683 | __send_subscribe(monc); |
684 | __resend_statfs(monc); | |
685 | } | |
686 | mutex_unlock(&monc->mutex); | |
687 | } | |
688 | ||
9bd2e6f8 SW |
689 | static int __validate_auth(struct ceph_mon_client *monc) |
690 | { | |
691 | int ret; | |
692 | ||
693 | if (monc->pending_auth) | |
694 | return 0; | |
695 | ||
696 | ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base, | |
697 | monc->m_auth->front_max); | |
698 | if (ret <= 0) | |
699 | return ret; /* either an error, or no need to authenticate */ | |
700 | __send_prepared_auth_request(monc, ret); | |
701 | return 0; | |
702 | } | |
703 | ||
704 | int ceph_monc_validate_auth(struct ceph_mon_client *monc) | |
705 | { | |
706 | int ret; | |
707 | ||
708 | mutex_lock(&monc->mutex); | |
709 | ret = __validate_auth(monc); | |
710 | mutex_unlock(&monc->mutex); | |
711 | return ret; | |
712 | } | |
713 | ||
ba75bb98 SW |
714 | /* |
715 | * handle incoming message | |
716 | */ | |
717 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
718 | { | |
719 | struct ceph_mon_client *monc = con->private; | |
720 | int type = le16_to_cpu(msg->hdr.type); | |
721 | ||
722 | if (!monc) | |
723 | return; | |
724 | ||
725 | switch (type) { | |
4e7a5dcd SW |
726 | case CEPH_MSG_AUTH_REPLY: |
727 | handle_auth_reply(monc, msg); | |
ba75bb98 SW |
728 | break; |
729 | ||
730 | case CEPH_MSG_MON_SUBSCRIBE_ACK: | |
731 | handle_subscribe_ack(monc, msg); | |
732 | break; | |
733 | ||
734 | case CEPH_MSG_STATFS_REPLY: | |
735 | handle_statfs_reply(monc, msg); | |
736 | break; | |
737 | ||
4e7a5dcd SW |
738 | case CEPH_MSG_MON_MAP: |
739 | ceph_monc_handle_map(monc, msg); | |
740 | break; | |
741 | ||
ba75bb98 SW |
742 | case CEPH_MSG_MDS_MAP: |
743 | ceph_mdsc_handle_map(&monc->client->mdsc, msg); | |
744 | break; | |
745 | ||
746 | case CEPH_MSG_OSD_MAP: | |
747 | ceph_osdc_handle_map(&monc->client->osdc, msg); | |
748 | break; | |
749 | ||
750 | default: | |
751 | pr_err("received unknown message type %d %s\n", type, | |
752 | ceph_msg_type_name(type)); | |
753 | } | |
754 | ceph_msg_put(msg); | |
755 | } | |
756 | ||
757 | /* | |
758 | * Allocate memory for incoming message | |
759 | */ | |
760 | static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con, | |
2450418c YS |
761 | struct ceph_msg_header *hdr, |
762 | int *skip) | |
ba75bb98 SW |
763 | { |
764 | struct ceph_mon_client *monc = con->private; | |
765 | int type = le16_to_cpu(hdr->type); | |
2450418c | 766 | int front_len = le32_to_cpu(hdr->front_len); |
5b3a4db3 | 767 | struct ceph_msg *m = NULL; |
ba75bb98 | 768 | |
2450418c | 769 | *skip = 0; |
0547a9b3 | 770 | |
ba75bb98 | 771 | switch (type) { |
ba75bb98 | 772 | case CEPH_MSG_MON_SUBSCRIBE_ACK: |
2450418c YS |
773 | m = ceph_msgpool_get(&monc->msgpool_subscribe_ack, front_len); |
774 | break; | |
ba75bb98 | 775 | case CEPH_MSG_STATFS_REPLY: |
2450418c YS |
776 | m = ceph_msgpool_get(&monc->msgpool_statfs_reply, front_len); |
777 | break; | |
4e7a5dcd | 778 | case CEPH_MSG_AUTH_REPLY: |
2450418c YS |
779 | m = ceph_msgpool_get(&monc->msgpool_auth_reply, front_len); |
780 | break; | |
5b3a4db3 SW |
781 | case CEPH_MSG_MON_MAP: |
782 | case CEPH_MSG_MDS_MAP: | |
783 | case CEPH_MSG_OSD_MAP: | |
784 | m = ceph_msg_new(type, front_len, 0, 0, NULL); | |
785 | break; | |
ba75bb98 | 786 | } |
2450418c | 787 | |
5b3a4db3 SW |
788 | if (!m) { |
789 | pr_info("alloc_msg unknown type %d\n", type); | |
2450418c | 790 | *skip = 1; |
5b3a4db3 | 791 | } |
2450418c | 792 | return m; |
ba75bb98 SW |
793 | } |
794 | ||
795 | /* | |
796 | * If the monitor connection resets, pick a new monitor and resubmit | |
797 | * any pending requests. | |
798 | */ | |
799 | static void mon_fault(struct ceph_connection *con) | |
800 | { | |
801 | struct ceph_mon_client *monc = con->private; | |
802 | ||
803 | if (!monc) | |
804 | return; | |
805 | ||
806 | dout("mon_fault\n"); | |
807 | mutex_lock(&monc->mutex); | |
808 | if (!con->private) | |
809 | goto out; | |
810 | ||
811 | if (monc->con && !monc->hunting) | |
812 | pr_info("mon%d %s session lost, " | |
813 | "hunting for new mon\n", monc->cur_mon, | |
814 | pr_addr(&monc->con->peer_addr.in_addr)); | |
815 | ||
816 | __close_session(monc); | |
817 | if (!monc->hunting) { | |
818 | /* start hunting */ | |
819 | monc->hunting = true; | |
4e7a5dcd | 820 | __open_session(monc); |
ba75bb98 SW |
821 | } else { |
822 | /* already hunting, let's wait a bit */ | |
823 | __schedule_delayed(monc); | |
824 | } | |
825 | out: | |
826 | mutex_unlock(&monc->mutex); | |
827 | } | |
828 | ||
829 | const static struct ceph_connection_operations mon_con_ops = { | |
830 | .get = ceph_con_get, | |
831 | .put = ceph_con_put, | |
832 | .dispatch = dispatch, | |
833 | .fault = mon_fault, | |
834 | .alloc_msg = mon_alloc_msg, | |
ba75bb98 | 835 | }; |