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