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 | ||
3143edd3 SW |
396 | static void release_statfs_request(struct kref *kref) |
397 | { | |
398 | struct ceph_mon_statfs_request *req = | |
399 | container_of(kref, struct ceph_mon_statfs_request, kref); | |
400 | ||
401 | if (req->reply) | |
402 | ceph_msg_put(req->reply); | |
403 | if (req->request) | |
404 | ceph_msg_put(req->request); | |
405 | } | |
406 | ||
407 | static void put_statfs_request(struct ceph_mon_statfs_request *req) | |
408 | { | |
409 | kref_put(&req->kref, release_statfs_request); | |
410 | } | |
411 | ||
412 | static void get_statfs_request(struct ceph_mon_statfs_request *req) | |
413 | { | |
414 | kref_get(&req->kref); | |
415 | } | |
416 | ||
417 | static struct ceph_msg *get_statfs_reply(struct ceph_connection *con, | |
418 | struct ceph_msg_header *hdr, | |
419 | int *skip) | |
420 | { | |
421 | struct ceph_mon_client *monc = con->private; | |
422 | struct ceph_mon_statfs_request *req; | |
423 | u64 tid = le64_to_cpu(hdr->tid); | |
424 | struct ceph_msg *m; | |
425 | ||
426 | mutex_lock(&monc->mutex); | |
427 | req = __lookup_statfs(monc, tid); | |
428 | if (!req) { | |
429 | dout("get_statfs_reply %lld dne\n", tid); | |
430 | *skip = 1; | |
431 | m = NULL; | |
432 | } else { | |
433 | dout("get_statfs_reply %lld got %p\n", tid, req->reply); | |
434 | m = ceph_msg_get(req->reply); | |
435 | /* | |
436 | * we don't need to track the connection reading into | |
437 | * this reply because we only have one open connection | |
438 | * at a time, ever. | |
439 | */ | |
440 | } | |
441 | mutex_unlock(&monc->mutex); | |
442 | return m; | |
443 | } | |
444 | ||
ba75bb98 SW |
445 | static void handle_statfs_reply(struct ceph_mon_client *monc, |
446 | struct ceph_msg *msg) | |
447 | { | |
448 | struct ceph_mon_statfs_request *req; | |
449 | struct ceph_mon_statfs_reply *reply = msg->front.iov_base; | |
3143edd3 | 450 | u64 tid = le64_to_cpu(msg->hdr.tid); |
ba75bb98 SW |
451 | |
452 | if (msg->front.iov_len != sizeof(*reply)) | |
453 | goto bad; | |
ba75bb98 SW |
454 | dout("handle_statfs_reply %p tid %llu\n", msg, tid); |
455 | ||
456 | mutex_lock(&monc->mutex); | |
85ff03f6 | 457 | req = __lookup_statfs(monc, tid); |
ba75bb98 SW |
458 | if (req) { |
459 | *req->buf = reply->st; | |
460 | req->result = 0; | |
3143edd3 | 461 | get_statfs_request(req); |
ba75bb98 SW |
462 | } |
463 | mutex_unlock(&monc->mutex); | |
3143edd3 | 464 | if (req) { |
ba75bb98 | 465 | complete(&req->completion); |
3143edd3 SW |
466 | put_statfs_request(req); |
467 | } | |
ba75bb98 SW |
468 | return; |
469 | ||
470 | bad: | |
471 | pr_err("corrupt statfs reply, no tid\n"); | |
9ec7cab1 | 472 | ceph_msg_dump(msg); |
ba75bb98 SW |
473 | } |
474 | ||
475 | /* | |
3143edd3 | 476 | * Do a synchronous statfs(). |
ba75bb98 | 477 | */ |
3143edd3 | 478 | int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf) |
ba75bb98 | 479 | { |
3143edd3 | 480 | struct ceph_mon_statfs_request *req; |
ba75bb98 | 481 | struct ceph_mon_statfs *h; |
3143edd3 SW |
482 | int err; |
483 | ||
484 | req = kmalloc(sizeof(*req), GFP_NOFS); | |
485 | if (!req) | |
486 | return -ENOMEM; | |
487 | ||
488 | memset(req, 0, sizeof(*req)); | |
489 | kref_init(&req->kref); | |
490 | req->buf = buf; | |
491 | init_completion(&req->completion); | |
ba75bb98 | 492 | |
3143edd3 SW |
493 | req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), 0, 0, NULL); |
494 | if (IS_ERR(req->request)) { | |
495 | err = PTR_ERR(req->request); | |
496 | goto out; | |
497 | } | |
498 | req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, 0, 0, NULL); | |
499 | if (IS_ERR(req->reply)) { | |
500 | err = PTR_ERR(req->reply); | |
501 | goto out; | |
502 | } | |
503 | ||
504 | /* fill out request */ | |
505 | h = req->request->front.iov_base; | |
13e38c8a SW |
506 | h->monhdr.have_version = 0; |
507 | h->monhdr.session_mon = cpu_to_le16(-1); | |
508 | h->monhdr.session_mon_tid = 0; | |
ba75bb98 | 509 | h->fsid = monc->monmap->fsid; |
ba75bb98 SW |
510 | |
511 | /* register request */ | |
512 | mutex_lock(&monc->mutex); | |
3143edd3 SW |
513 | req->tid = ++monc->last_tid; |
514 | req->request->hdr.tid = cpu_to_le64(req->tid); | |
515 | __insert_statfs(monc, req); | |
ba75bb98 SW |
516 | monc->num_statfs_requests++; |
517 | mutex_unlock(&monc->mutex); | |
518 | ||
519 | /* send request and wait */ | |
3143edd3 SW |
520 | ceph_con_send(monc->con, ceph_msg_get(req->request)); |
521 | err = wait_for_completion_interruptible(&req->completion); | |
ba75bb98 SW |
522 | |
523 | mutex_lock(&monc->mutex); | |
3143edd3 | 524 | rb_erase(&req->node, &monc->statfs_request_tree); |
ba75bb98 | 525 | monc->num_statfs_requests--; |
ba75bb98 SW |
526 | mutex_unlock(&monc->mutex); |
527 | ||
528 | if (!err) | |
3143edd3 SW |
529 | err = req->result; |
530 | ||
531 | out: | |
532 | kref_put(&req->kref, release_statfs_request); | |
ba75bb98 SW |
533 | return err; |
534 | } | |
535 | ||
536 | /* | |
537 | * Resend pending statfs requests. | |
538 | */ | |
539 | static void __resend_statfs(struct ceph_mon_client *monc) | |
540 | { | |
ba75bb98 | 541 | struct ceph_mon_statfs_request *req; |
85ff03f6 | 542 | struct rb_node *p; |
ba75bb98 | 543 | |
85ff03f6 SW |
544 | for (p = rb_first(&monc->statfs_request_tree); p; p = rb_next(p)) { |
545 | req = rb_entry(p, struct ceph_mon_statfs_request, node); | |
3143edd3 | 546 | ceph_con_send(monc->con, ceph_msg_get(req->request)); |
ba75bb98 SW |
547 | } |
548 | } | |
549 | ||
550 | /* | |
551 | * Delayed work. If we haven't mounted yet, retry. Otherwise, | |
552 | * renew/retry subscription as needed (in case it is timing out, or we | |
553 | * got an ENOMEM). And keep the monitor connection alive. | |
554 | */ | |
555 | static void delayed_work(struct work_struct *work) | |
556 | { | |
557 | struct ceph_mon_client *monc = | |
558 | container_of(work, struct ceph_mon_client, delayed_work.work); | |
559 | ||
560 | dout("monc delayed_work\n"); | |
561 | mutex_lock(&monc->mutex); | |
4e7a5dcd SW |
562 | if (monc->hunting) { |
563 | __close_session(monc); | |
564 | __open_session(monc); /* continue hunting */ | |
ba75bb98 | 565 | } else { |
4e7a5dcd | 566 | ceph_con_keepalive(monc->con); |
9bd2e6f8 SW |
567 | |
568 | __validate_auth(monc); | |
569 | ||
4e7a5dcd SW |
570 | if (monc->auth->ops->is_authenticated(monc->auth)) |
571 | __send_subscribe(monc); | |
ba75bb98 | 572 | } |
ba75bb98 SW |
573 | __schedule_delayed(monc); |
574 | mutex_unlock(&monc->mutex); | |
575 | } | |
576 | ||
6b805185 SW |
577 | /* |
578 | * On startup, we build a temporary monmap populated with the IPs | |
579 | * provided by mount(2). | |
580 | */ | |
581 | static int build_initial_monmap(struct ceph_mon_client *monc) | |
582 | { | |
583 | struct ceph_mount_args *args = monc->client->mount_args; | |
584 | struct ceph_entity_addr *mon_addr = args->mon_addr; | |
585 | int num_mon = args->num_mon; | |
586 | int i; | |
587 | ||
588 | /* build initial monmap */ | |
589 | monc->monmap = kzalloc(sizeof(*monc->monmap) + | |
590 | num_mon*sizeof(monc->monmap->mon_inst[0]), | |
591 | GFP_KERNEL); | |
592 | if (!monc->monmap) | |
593 | return -ENOMEM; | |
594 | for (i = 0; i < num_mon; i++) { | |
595 | monc->monmap->mon_inst[i].addr = mon_addr[i]; | |
6b805185 SW |
596 | monc->monmap->mon_inst[i].addr.nonce = 0; |
597 | monc->monmap->mon_inst[i].name.type = | |
598 | CEPH_ENTITY_TYPE_MON; | |
599 | monc->monmap->mon_inst[i].name.num = cpu_to_le64(i); | |
600 | } | |
601 | monc->monmap->num_mon = num_mon; | |
4e7a5dcd | 602 | monc->have_fsid = false; |
6b805185 SW |
603 | |
604 | /* release addr memory */ | |
605 | kfree(args->mon_addr); | |
606 | args->mon_addr = NULL; | |
607 | args->num_mon = 0; | |
608 | return 0; | |
609 | } | |
610 | ||
ba75bb98 SW |
611 | int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl) |
612 | { | |
613 | int err = 0; | |
614 | ||
615 | dout("init\n"); | |
616 | memset(monc, 0, sizeof(*monc)); | |
617 | monc->client = cl; | |
618 | monc->monmap = NULL; | |
619 | mutex_init(&monc->mutex); | |
620 | ||
6b805185 SW |
621 | err = build_initial_monmap(monc); |
622 | if (err) | |
623 | goto out; | |
624 | ||
ba75bb98 SW |
625 | monc->con = NULL; |
626 | ||
4e7a5dcd SW |
627 | /* authentication */ |
628 | monc->auth = ceph_auth_init(cl->mount_args->name, | |
629 | cl->mount_args->secret); | |
630 | if (IS_ERR(monc->auth)) | |
631 | return PTR_ERR(monc->auth); | |
632 | monc->auth->want_keys = | |
633 | CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON | | |
634 | CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS; | |
635 | ||
ba75bb98 | 636 | /* msg pools */ |
07bd10fb SW |
637 | err = ceph_msgpool_init(&monc->msgpool_subscribe_ack, |
638 | sizeof(struct ceph_mon_subscribe_ack), 1, false); | |
ba75bb98 | 639 | if (err < 0) |
4e7a5dcd | 640 | goto out_monmap; |
6694d6b9 SW |
641 | |
642 | monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, 0, 0, | |
643 | NULL); | |
644 | if (IS_ERR(monc->m_auth_reply)) { | |
645 | err = PTR_ERR(monc->m_auth_reply); | |
646 | monc->m_auth_reply = NULL; | |
3143edd3 | 647 | goto out_pool; |
6694d6b9 | 648 | } |
4e7a5dcd SW |
649 | |
650 | monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, 0, 0, NULL); | |
9bd2e6f8 | 651 | monc->pending_auth = 0; |
4e7a5dcd SW |
652 | if (IS_ERR(monc->m_auth)) { |
653 | err = PTR_ERR(monc->m_auth); | |
654 | monc->m_auth = NULL; | |
6694d6b9 | 655 | goto out_auth_reply; |
4e7a5dcd | 656 | } |
ba75bb98 SW |
657 | |
658 | monc->cur_mon = -1; | |
4e7a5dcd | 659 | monc->hunting = true; |
ba75bb98 SW |
660 | monc->sub_renew_after = jiffies; |
661 | monc->sub_sent = 0; | |
662 | ||
663 | INIT_DELAYED_WORK(&monc->delayed_work, delayed_work); | |
85ff03f6 | 664 | monc->statfs_request_tree = RB_ROOT; |
ba75bb98 SW |
665 | monc->num_statfs_requests = 0; |
666 | monc->last_tid = 0; | |
667 | ||
668 | monc->have_mdsmap = 0; | |
669 | monc->have_osdmap = 0; | |
670 | monc->want_next_osdmap = 1; | |
4e7a5dcd SW |
671 | return 0; |
672 | ||
6694d6b9 SW |
673 | out_auth_reply: |
674 | ceph_msg_put(monc->m_auth_reply); | |
3143edd3 | 675 | out_pool: |
4e7a5dcd | 676 | ceph_msgpool_destroy(&monc->msgpool_subscribe_ack); |
4e7a5dcd SW |
677 | out_monmap: |
678 | kfree(monc->monmap); | |
ba75bb98 SW |
679 | out: |
680 | return err; | |
681 | } | |
682 | ||
683 | void ceph_monc_stop(struct ceph_mon_client *monc) | |
684 | { | |
685 | dout("stop\n"); | |
686 | cancel_delayed_work_sync(&monc->delayed_work); | |
687 | ||
688 | mutex_lock(&monc->mutex); | |
689 | __close_session(monc); | |
690 | if (monc->con) { | |
691 | monc->con->private = NULL; | |
692 | monc->con->ops->put(monc->con); | |
693 | monc->con = NULL; | |
694 | } | |
695 | mutex_unlock(&monc->mutex); | |
696 | ||
4e7a5dcd SW |
697 | ceph_auth_destroy(monc->auth); |
698 | ||
699 | ceph_msg_put(monc->m_auth); | |
6694d6b9 | 700 | ceph_msg_put(monc->m_auth_reply); |
ba75bb98 | 701 | ceph_msgpool_destroy(&monc->msgpool_subscribe_ack); |
ba75bb98 SW |
702 | |
703 | kfree(monc->monmap); | |
704 | } | |
705 | ||
4e7a5dcd SW |
706 | static void handle_auth_reply(struct ceph_mon_client *monc, |
707 | struct ceph_msg *msg) | |
708 | { | |
709 | int ret; | |
710 | ||
711 | mutex_lock(&monc->mutex); | |
9bd2e6f8 | 712 | monc->pending_auth = 0; |
4e7a5dcd SW |
713 | ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base, |
714 | msg->front.iov_len, | |
715 | monc->m_auth->front.iov_base, | |
716 | monc->m_auth->front_max); | |
717 | if (ret < 0) { | |
9bd2e6f8 SW |
718 | monc->client->auth_err = ret; |
719 | wake_up(&monc->client->auth_wq); | |
4e7a5dcd | 720 | } else if (ret > 0) { |
9bd2e6f8 | 721 | __send_prepared_auth_request(monc, ret); |
4e7a5dcd SW |
722 | } else if (monc->auth->ops->is_authenticated(monc->auth)) { |
723 | dout("authenticated, starting session\n"); | |
0743304d SW |
724 | |
725 | monc->client->msgr->inst.name.type = CEPH_ENTITY_TYPE_CLIENT; | |
726 | monc->client->msgr->inst.name.num = monc->auth->global_id; | |
727 | ||
4e7a5dcd SW |
728 | __send_subscribe(monc); |
729 | __resend_statfs(monc); | |
730 | } | |
731 | mutex_unlock(&monc->mutex); | |
732 | } | |
733 | ||
9bd2e6f8 SW |
734 | static int __validate_auth(struct ceph_mon_client *monc) |
735 | { | |
736 | int ret; | |
737 | ||
738 | if (monc->pending_auth) | |
739 | return 0; | |
740 | ||
741 | ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base, | |
742 | monc->m_auth->front_max); | |
743 | if (ret <= 0) | |
744 | return ret; /* either an error, or no need to authenticate */ | |
745 | __send_prepared_auth_request(monc, ret); | |
746 | return 0; | |
747 | } | |
748 | ||
749 | int ceph_monc_validate_auth(struct ceph_mon_client *monc) | |
750 | { | |
751 | int ret; | |
752 | ||
753 | mutex_lock(&monc->mutex); | |
754 | ret = __validate_auth(monc); | |
755 | mutex_unlock(&monc->mutex); | |
756 | return ret; | |
757 | } | |
758 | ||
ba75bb98 SW |
759 | /* |
760 | * handle incoming message | |
761 | */ | |
762 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
763 | { | |
764 | struct ceph_mon_client *monc = con->private; | |
765 | int type = le16_to_cpu(msg->hdr.type); | |
766 | ||
767 | if (!monc) | |
768 | return; | |
769 | ||
770 | switch (type) { | |
4e7a5dcd SW |
771 | case CEPH_MSG_AUTH_REPLY: |
772 | handle_auth_reply(monc, msg); | |
ba75bb98 SW |
773 | break; |
774 | ||
775 | case CEPH_MSG_MON_SUBSCRIBE_ACK: | |
776 | handle_subscribe_ack(monc, msg); | |
777 | break; | |
778 | ||
779 | case CEPH_MSG_STATFS_REPLY: | |
780 | handle_statfs_reply(monc, msg); | |
781 | break; | |
782 | ||
4e7a5dcd SW |
783 | case CEPH_MSG_MON_MAP: |
784 | ceph_monc_handle_map(monc, msg); | |
785 | break; | |
786 | ||
ba75bb98 SW |
787 | case CEPH_MSG_MDS_MAP: |
788 | ceph_mdsc_handle_map(&monc->client->mdsc, msg); | |
789 | break; | |
790 | ||
791 | case CEPH_MSG_OSD_MAP: | |
792 | ceph_osdc_handle_map(&monc->client->osdc, msg); | |
793 | break; | |
794 | ||
795 | default: | |
796 | pr_err("received unknown message type %d %s\n", type, | |
797 | ceph_msg_type_name(type)); | |
798 | } | |
799 | ceph_msg_put(msg); | |
800 | } | |
801 | ||
802 | /* | |
803 | * Allocate memory for incoming message | |
804 | */ | |
805 | static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con, | |
2450418c YS |
806 | struct ceph_msg_header *hdr, |
807 | int *skip) | |
ba75bb98 SW |
808 | { |
809 | struct ceph_mon_client *monc = con->private; | |
810 | int type = le16_to_cpu(hdr->type); | |
2450418c | 811 | int front_len = le32_to_cpu(hdr->front_len); |
5b3a4db3 | 812 | struct ceph_msg *m = NULL; |
ba75bb98 | 813 | |
2450418c | 814 | *skip = 0; |
0547a9b3 | 815 | |
ba75bb98 | 816 | switch (type) { |
ba75bb98 | 817 | case CEPH_MSG_MON_SUBSCRIBE_ACK: |
2450418c YS |
818 | m = ceph_msgpool_get(&monc->msgpool_subscribe_ack, front_len); |
819 | break; | |
ba75bb98 | 820 | case CEPH_MSG_STATFS_REPLY: |
3143edd3 | 821 | return get_statfs_reply(con, hdr, skip); |
4e7a5dcd | 822 | case CEPH_MSG_AUTH_REPLY: |
6694d6b9 | 823 | m = ceph_msg_get(monc->m_auth_reply); |
2450418c | 824 | break; |
5b3a4db3 SW |
825 | case CEPH_MSG_MON_MAP: |
826 | case CEPH_MSG_MDS_MAP: | |
827 | case CEPH_MSG_OSD_MAP: | |
828 | m = ceph_msg_new(type, front_len, 0, 0, NULL); | |
829 | break; | |
ba75bb98 | 830 | } |
2450418c | 831 | |
5b3a4db3 SW |
832 | if (!m) { |
833 | pr_info("alloc_msg unknown type %d\n", type); | |
2450418c | 834 | *skip = 1; |
5b3a4db3 | 835 | } |
2450418c | 836 | return m; |
ba75bb98 SW |
837 | } |
838 | ||
839 | /* | |
840 | * If the monitor connection resets, pick a new monitor and resubmit | |
841 | * any pending requests. | |
842 | */ | |
843 | static void mon_fault(struct ceph_connection *con) | |
844 | { | |
845 | struct ceph_mon_client *monc = con->private; | |
846 | ||
847 | if (!monc) | |
848 | return; | |
849 | ||
850 | dout("mon_fault\n"); | |
851 | mutex_lock(&monc->mutex); | |
852 | if (!con->private) | |
853 | goto out; | |
854 | ||
855 | if (monc->con && !monc->hunting) | |
856 | pr_info("mon%d %s session lost, " | |
857 | "hunting for new mon\n", monc->cur_mon, | |
858 | pr_addr(&monc->con->peer_addr.in_addr)); | |
859 | ||
860 | __close_session(monc); | |
861 | if (!monc->hunting) { | |
862 | /* start hunting */ | |
863 | monc->hunting = true; | |
4e7a5dcd | 864 | __open_session(monc); |
ba75bb98 SW |
865 | } else { |
866 | /* already hunting, let's wait a bit */ | |
867 | __schedule_delayed(monc); | |
868 | } | |
869 | out: | |
870 | mutex_unlock(&monc->mutex); | |
871 | } | |
872 | ||
873 | const static struct ceph_connection_operations mon_con_ops = { | |
874 | .get = ceph_con_get, | |
875 | .put = ceph_con_put, | |
876 | .dispatch = dispatch, | |
877 | .fault = mon_fault, | |
878 | .alloc_msg = mon_alloc_msg, | |
ba75bb98 | 879 | }; |