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 | ||
9e32789f | 31 | static const struct ceph_connection_operations mon_con_ops; |
ba75bb98 | 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); | |
97069001 | 107 | ceph_con_revoke(monc->con, monc->m_auth); |
5ce6e9db SW |
108 | ceph_msg_get(monc->m_auth); /* keep our ref */ |
109 | ceph_con_send(monc->con, monc->m_auth); | |
110 | } | |
111 | ||
ba75bb98 SW |
112 | /* |
113 | * Close monitor session, if any. | |
114 | */ | |
115 | static void __close_session(struct ceph_mon_client *monc) | |
116 | { | |
117 | if (monc->con) { | |
118 | dout("__close_session closing mon%d\n", monc->cur_mon); | |
4e7a5dcd | 119 | ceph_con_revoke(monc->con, monc->m_auth); |
ba75bb98 SW |
120 | ceph_con_close(monc->con); |
121 | monc->cur_mon = -1; | |
9bd2e6f8 | 122 | monc->pending_auth = 0; |
4e7a5dcd | 123 | ceph_auth_reset(monc->auth); |
ba75bb98 SW |
124 | } |
125 | } | |
126 | ||
127 | /* | |
128 | * Open a session with a (new) monitor. | |
129 | */ | |
130 | static int __open_session(struct ceph_mon_client *monc) | |
131 | { | |
132 | char r; | |
4e7a5dcd | 133 | int ret; |
ba75bb98 SW |
134 | |
135 | if (monc->cur_mon < 0) { | |
136 | get_random_bytes(&r, 1); | |
137 | monc->cur_mon = r % monc->monmap->num_mon; | |
138 | dout("open_session num=%d r=%d -> mon%d\n", | |
139 | monc->monmap->num_mon, r, monc->cur_mon); | |
140 | monc->sub_sent = 0; | |
141 | monc->sub_renew_after = jiffies; /* i.e., expired */ | |
142 | monc->want_next_osdmap = !!monc->want_next_osdmap; | |
143 | ||
144 | dout("open_session mon%d opening\n", monc->cur_mon); | |
145 | monc->con->peer_name.type = CEPH_ENTITY_TYPE_MON; | |
146 | monc->con->peer_name.num = cpu_to_le64(monc->cur_mon); | |
147 | ceph_con_open(monc->con, | |
148 | &monc->monmap->mon_inst[monc->cur_mon].addr); | |
4e7a5dcd SW |
149 | |
150 | /* initiatiate authentication handshake */ | |
151 | ret = ceph_auth_build_hello(monc->auth, | |
152 | monc->m_auth->front.iov_base, | |
153 | monc->m_auth->front_max); | |
5ce6e9db | 154 | __send_prepared_auth_request(monc, ret); |
ba75bb98 SW |
155 | } else { |
156 | dout("open_session mon%d already open\n", monc->cur_mon); | |
157 | } | |
158 | return 0; | |
159 | } | |
160 | ||
161 | static bool __sub_expired(struct ceph_mon_client *monc) | |
162 | { | |
163 | return time_after_eq(jiffies, monc->sub_renew_after); | |
164 | } | |
165 | ||
166 | /* | |
167 | * Reschedule delayed work timer. | |
168 | */ | |
169 | static void __schedule_delayed(struct ceph_mon_client *monc) | |
170 | { | |
171 | unsigned delay; | |
172 | ||
4e7a5dcd | 173 | if (monc->cur_mon < 0 || __sub_expired(monc)) |
ba75bb98 SW |
174 | delay = 10 * HZ; |
175 | else | |
176 | delay = 20 * HZ; | |
177 | dout("__schedule_delayed after %u\n", delay); | |
178 | schedule_delayed_work(&monc->delayed_work, delay); | |
179 | } | |
180 | ||
181 | /* | |
182 | * Send subscribe request for mdsmap and/or osdmap. | |
183 | */ | |
184 | static void __send_subscribe(struct ceph_mon_client *monc) | |
185 | { | |
186 | dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n", | |
187 | (unsigned)monc->sub_sent, __sub_expired(monc), | |
188 | monc->want_next_osdmap); | |
189 | if ((__sub_expired(monc) && !monc->sub_sent) || | |
190 | monc->want_next_osdmap == 1) { | |
240ed68e | 191 | struct ceph_msg *msg = monc->m_subscribe; |
ba75bb98 SW |
192 | struct ceph_mon_subscribe_item *i; |
193 | void *p, *end; | |
194 | ||
ba75bb98 | 195 | p = msg->front.iov_base; |
240ed68e | 196 | end = p + msg->front_max; |
ba75bb98 SW |
197 | |
198 | dout("__send_subscribe to 'mdsmap' %u+\n", | |
199 | (unsigned)monc->have_mdsmap); | |
200 | if (monc->want_next_osdmap) { | |
201 | dout("__send_subscribe to 'osdmap' %u\n", | |
202 | (unsigned)monc->have_osdmap); | |
4e7a5dcd | 203 | ceph_encode_32(&p, 3); |
ba75bb98 SW |
204 | ceph_encode_string(&p, end, "osdmap", 6); |
205 | i = p; | |
206 | i->have = cpu_to_le64(monc->have_osdmap); | |
207 | i->onetime = 1; | |
208 | p += sizeof(*i); | |
209 | monc->want_next_osdmap = 2; /* requested */ | |
210 | } else { | |
4e7a5dcd | 211 | ceph_encode_32(&p, 2); |
ba75bb98 SW |
212 | } |
213 | ceph_encode_string(&p, end, "mdsmap", 6); | |
214 | i = p; | |
215 | i->have = cpu_to_le64(monc->have_mdsmap); | |
216 | i->onetime = 0; | |
217 | p += sizeof(*i); | |
4e7a5dcd SW |
218 | ceph_encode_string(&p, end, "monmap", 6); |
219 | i = p; | |
220 | i->have = 0; | |
221 | i->onetime = 0; | |
222 | p += sizeof(*i); | |
ba75bb98 SW |
223 | |
224 | msg->front.iov_len = p - msg->front.iov_base; | |
225 | msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); | |
240ed68e SW |
226 | ceph_con_revoke(monc->con, msg); |
227 | ceph_con_send(monc->con, ceph_msg_get(msg)); | |
ba75bb98 SW |
228 | |
229 | monc->sub_sent = jiffies | 1; /* never 0 */ | |
230 | } | |
231 | } | |
232 | ||
233 | static void handle_subscribe_ack(struct ceph_mon_client *monc, | |
234 | struct ceph_msg *msg) | |
235 | { | |
236 | unsigned seconds; | |
07bd10fb SW |
237 | struct ceph_mon_subscribe_ack *h = msg->front.iov_base; |
238 | ||
239 | if (msg->front.iov_len < sizeof(*h)) | |
240 | goto bad; | |
241 | seconds = le32_to_cpu(h->duration); | |
ba75bb98 | 242 | |
ba75bb98 SW |
243 | mutex_lock(&monc->mutex); |
244 | if (monc->hunting) { | |
245 | pr_info("mon%d %s session established\n", | |
246 | monc->cur_mon, pr_addr(&monc->con->peer_addr.in_addr)); | |
247 | monc->hunting = false; | |
248 | } | |
249 | dout("handle_subscribe_ack after %d seconds\n", seconds); | |
0656d11b | 250 | monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1; |
ba75bb98 SW |
251 | monc->sub_sent = 0; |
252 | mutex_unlock(&monc->mutex); | |
253 | return; | |
254 | bad: | |
255 | pr_err("got corrupt subscribe-ack msg\n"); | |
9ec7cab1 | 256 | ceph_msg_dump(msg); |
ba75bb98 SW |
257 | } |
258 | ||
259 | /* | |
260 | * Keep track of which maps we have | |
261 | */ | |
262 | int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got) | |
263 | { | |
264 | mutex_lock(&monc->mutex); | |
265 | monc->have_mdsmap = got; | |
266 | mutex_unlock(&monc->mutex); | |
267 | return 0; | |
268 | } | |
269 | ||
270 | int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got) | |
271 | { | |
272 | mutex_lock(&monc->mutex); | |
273 | monc->have_osdmap = got; | |
274 | monc->want_next_osdmap = 0; | |
275 | mutex_unlock(&monc->mutex); | |
276 | return 0; | |
277 | } | |
278 | ||
279 | /* | |
280 | * Register interest in the next osdmap | |
281 | */ | |
282 | void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc) | |
283 | { | |
284 | dout("request_next_osdmap have %u\n", monc->have_osdmap); | |
285 | mutex_lock(&monc->mutex); | |
286 | if (!monc->want_next_osdmap) | |
287 | monc->want_next_osdmap = 1; | |
288 | if (monc->want_next_osdmap < 2) | |
289 | __send_subscribe(monc); | |
290 | mutex_unlock(&monc->mutex); | |
291 | } | |
292 | ||
4e7a5dcd | 293 | /* |
50b885b9 | 294 | * |
4e7a5dcd SW |
295 | */ |
296 | int ceph_monc_open_session(struct ceph_mon_client *monc) | |
ba75bb98 SW |
297 | { |
298 | if (!monc->con) { | |
299 | monc->con = kmalloc(sizeof(*monc->con), GFP_KERNEL); | |
300 | if (!monc->con) | |
301 | return -ENOMEM; | |
302 | ceph_con_init(monc->client->msgr, monc->con); | |
303 | monc->con->private = monc; | |
304 | monc->con->ops = &mon_con_ops; | |
305 | } | |
306 | ||
307 | mutex_lock(&monc->mutex); | |
4e7a5dcd | 308 | __open_session(monc); |
ba75bb98 SW |
309 | __schedule_delayed(monc); |
310 | mutex_unlock(&monc->mutex); | |
311 | return 0; | |
312 | } | |
313 | ||
4e7a5dcd SW |
314 | /* |
315 | * The monitor responds with mount ack indicate mount success. The | |
316 | * included client ticket allows the client to talk to MDSs and OSDs. | |
317 | */ | |
0743304d SW |
318 | static void ceph_monc_handle_map(struct ceph_mon_client *monc, |
319 | struct ceph_msg *msg) | |
4e7a5dcd SW |
320 | { |
321 | struct ceph_client *client = monc->client; | |
322 | struct ceph_monmap *monmap = NULL, *old = monc->monmap; | |
323 | void *p, *end; | |
324 | ||
325 | mutex_lock(&monc->mutex); | |
326 | ||
327 | dout("handle_monmap\n"); | |
328 | p = msg->front.iov_base; | |
329 | end = p + msg->front.iov_len; | |
330 | ||
331 | monmap = ceph_monmap_decode(p, end); | |
332 | if (IS_ERR(monmap)) { | |
333 | pr_err("problem decoding monmap, %d\n", | |
334 | (int)PTR_ERR(monmap)); | |
d4a780ce | 335 | goto out; |
4e7a5dcd | 336 | } |
0743304d SW |
337 | |
338 | if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) { | |
4e7a5dcd | 339 | kfree(monmap); |
d4a780ce | 340 | goto out; |
4e7a5dcd SW |
341 | } |
342 | ||
343 | client->monc.monmap = monmap; | |
4e7a5dcd SW |
344 | kfree(old); |
345 | ||
d4a780ce | 346 | out: |
4e7a5dcd | 347 | mutex_unlock(&monc->mutex); |
03066f23 | 348 | wake_up_all(&client->auth_wq); |
4e7a5dcd SW |
349 | } |
350 | ||
ba75bb98 | 351 | /* |
e56fa10e | 352 | * generic requests (e.g., statfs, poolop) |
ba75bb98 | 353 | */ |
f8c76f6f | 354 | static struct ceph_mon_generic_request *__lookup_generic_req( |
85ff03f6 SW |
355 | struct ceph_mon_client *monc, u64 tid) |
356 | { | |
f8c76f6f YS |
357 | struct ceph_mon_generic_request *req; |
358 | struct rb_node *n = monc->generic_request_tree.rb_node; | |
85ff03f6 SW |
359 | |
360 | while (n) { | |
f8c76f6f | 361 | req = rb_entry(n, struct ceph_mon_generic_request, node); |
85ff03f6 SW |
362 | if (tid < req->tid) |
363 | n = n->rb_left; | |
364 | else if (tid > req->tid) | |
365 | n = n->rb_right; | |
366 | else | |
367 | return req; | |
368 | } | |
369 | return NULL; | |
370 | } | |
371 | ||
f8c76f6f YS |
372 | static void __insert_generic_request(struct ceph_mon_client *monc, |
373 | struct ceph_mon_generic_request *new) | |
85ff03f6 | 374 | { |
f8c76f6f | 375 | struct rb_node **p = &monc->generic_request_tree.rb_node; |
85ff03f6 | 376 | struct rb_node *parent = NULL; |
f8c76f6f | 377 | struct ceph_mon_generic_request *req = NULL; |
85ff03f6 SW |
378 | |
379 | while (*p) { | |
380 | parent = *p; | |
f8c76f6f | 381 | req = rb_entry(parent, struct ceph_mon_generic_request, node); |
85ff03f6 SW |
382 | if (new->tid < req->tid) |
383 | p = &(*p)->rb_left; | |
384 | else if (new->tid > req->tid) | |
385 | p = &(*p)->rb_right; | |
386 | else | |
387 | BUG(); | |
388 | } | |
389 | ||
390 | rb_link_node(&new->node, parent, p); | |
f8c76f6f | 391 | rb_insert_color(&new->node, &monc->generic_request_tree); |
85ff03f6 SW |
392 | } |
393 | ||
f8c76f6f | 394 | static void release_generic_request(struct kref *kref) |
3143edd3 | 395 | { |
f8c76f6f YS |
396 | struct ceph_mon_generic_request *req = |
397 | container_of(kref, struct ceph_mon_generic_request, kref); | |
3143edd3 SW |
398 | |
399 | if (req->reply) | |
400 | ceph_msg_put(req->reply); | |
401 | if (req->request) | |
402 | ceph_msg_put(req->request); | |
20547567 YS |
403 | |
404 | kfree(req); | |
3143edd3 SW |
405 | } |
406 | ||
f8c76f6f | 407 | static void put_generic_request(struct ceph_mon_generic_request *req) |
3143edd3 | 408 | { |
f8c76f6f | 409 | kref_put(&req->kref, release_generic_request); |
3143edd3 SW |
410 | } |
411 | ||
f8c76f6f | 412 | static void get_generic_request(struct ceph_mon_generic_request *req) |
3143edd3 SW |
413 | { |
414 | kref_get(&req->kref); | |
415 | } | |
416 | ||
f8c76f6f | 417 | static struct ceph_msg *get_generic_reply(struct ceph_connection *con, |
3143edd3 SW |
418 | struct ceph_msg_header *hdr, |
419 | int *skip) | |
420 | { | |
421 | struct ceph_mon_client *monc = con->private; | |
f8c76f6f | 422 | struct ceph_mon_generic_request *req; |
3143edd3 SW |
423 | u64 tid = le64_to_cpu(hdr->tid); |
424 | struct ceph_msg *m; | |
425 | ||
426 | mutex_lock(&monc->mutex); | |
f8c76f6f | 427 | req = __lookup_generic_req(monc, tid); |
3143edd3 | 428 | if (!req) { |
f8c76f6f | 429 | dout("get_generic_reply %lld dne\n", tid); |
3143edd3 SW |
430 | *skip = 1; |
431 | m = NULL; | |
432 | } else { | |
f8c76f6f | 433 | dout("get_generic_reply %lld got %p\n", tid, req->reply); |
3143edd3 SW |
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 | ||
e56fa10e YS |
445 | static int do_generic_request(struct ceph_mon_client *monc, |
446 | struct ceph_mon_generic_request *req) | |
447 | { | |
448 | int err; | |
449 | ||
450 | /* register request */ | |
451 | mutex_lock(&monc->mutex); | |
452 | req->tid = ++monc->last_tid; | |
453 | req->request->hdr.tid = cpu_to_le64(req->tid); | |
454 | __insert_generic_request(monc, req); | |
455 | monc->num_generic_requests++; | |
456 | ceph_con_send(monc->con, ceph_msg_get(req->request)); | |
457 | mutex_unlock(&monc->mutex); | |
458 | ||
459 | err = wait_for_completion_interruptible(&req->completion); | |
460 | ||
461 | mutex_lock(&monc->mutex); | |
462 | rb_erase(&req->node, &monc->generic_request_tree); | |
463 | monc->num_generic_requests--; | |
464 | mutex_unlock(&monc->mutex); | |
465 | ||
466 | if (!err) | |
467 | err = req->result; | |
468 | return err; | |
469 | } | |
470 | ||
471 | /* | |
472 | * statfs | |
473 | */ | |
ba75bb98 SW |
474 | static void handle_statfs_reply(struct ceph_mon_client *monc, |
475 | struct ceph_msg *msg) | |
476 | { | |
f8c76f6f | 477 | struct ceph_mon_generic_request *req; |
ba75bb98 | 478 | struct ceph_mon_statfs_reply *reply = msg->front.iov_base; |
3143edd3 | 479 | u64 tid = le64_to_cpu(msg->hdr.tid); |
ba75bb98 SW |
480 | |
481 | if (msg->front.iov_len != sizeof(*reply)) | |
482 | goto bad; | |
ba75bb98 SW |
483 | dout("handle_statfs_reply %p tid %llu\n", msg, tid); |
484 | ||
485 | mutex_lock(&monc->mutex); | |
f8c76f6f | 486 | req = __lookup_generic_req(monc, tid); |
ba75bb98 | 487 | if (req) { |
f8c76f6f | 488 | *(struct ceph_statfs *)req->buf = reply->st; |
ba75bb98 | 489 | req->result = 0; |
f8c76f6f | 490 | get_generic_request(req); |
ba75bb98 SW |
491 | } |
492 | mutex_unlock(&monc->mutex); | |
3143edd3 | 493 | if (req) { |
03066f23 | 494 | complete_all(&req->completion); |
f8c76f6f | 495 | put_generic_request(req); |
3143edd3 | 496 | } |
ba75bb98 SW |
497 | return; |
498 | ||
499 | bad: | |
e56fa10e | 500 | pr_err("corrupt generic reply, tid %llu\n", tid); |
9ec7cab1 | 501 | ceph_msg_dump(msg); |
ba75bb98 SW |
502 | } |
503 | ||
504 | /* | |
3143edd3 | 505 | * Do a synchronous statfs(). |
ba75bb98 | 506 | */ |
3143edd3 | 507 | int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf) |
ba75bb98 | 508 | { |
f8c76f6f | 509 | struct ceph_mon_generic_request *req; |
ba75bb98 | 510 | struct ceph_mon_statfs *h; |
3143edd3 SW |
511 | int err; |
512 | ||
cffe7b6d | 513 | req = kzalloc(sizeof(*req), GFP_NOFS); |
3143edd3 SW |
514 | if (!req) |
515 | return -ENOMEM; | |
516 | ||
3143edd3 SW |
517 | kref_init(&req->kref); |
518 | req->buf = buf; | |
e56fa10e | 519 | req->buf_len = sizeof(*buf); |
3143edd3 | 520 | init_completion(&req->completion); |
ba75bb98 | 521 | |
a79832f2 | 522 | err = -ENOMEM; |
34d23762 | 523 | req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS); |
a79832f2 | 524 | if (!req->request) |
3143edd3 | 525 | goto out; |
34d23762 | 526 | req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS); |
a79832f2 | 527 | if (!req->reply) |
3143edd3 | 528 | goto out; |
3143edd3 SW |
529 | |
530 | /* fill out request */ | |
531 | h = req->request->front.iov_base; | |
13e38c8a SW |
532 | h->monhdr.have_version = 0; |
533 | h->monhdr.session_mon = cpu_to_le16(-1); | |
534 | h->monhdr.session_mon_tid = 0; | |
ba75bb98 | 535 | h->fsid = monc->monmap->fsid; |
ba75bb98 | 536 | |
e56fa10e | 537 | err = do_generic_request(monc, req); |
ba75bb98 | 538 | |
e56fa10e YS |
539 | out: |
540 | kref_put(&req->kref, release_generic_request); | |
541 | return err; | |
542 | } | |
543 | ||
544 | /* | |
545 | * pool ops | |
546 | */ | |
547 | static int get_poolop_reply_buf(const char *src, size_t src_len, | |
548 | char *dst, size_t dst_len) | |
549 | { | |
550 | u32 buf_len; | |
551 | ||
552 | if (src_len != sizeof(u32) + dst_len) | |
553 | return -EINVAL; | |
554 | ||
555 | buf_len = le32_to_cpu(*(u32 *)src); | |
556 | if (buf_len != dst_len) | |
557 | return -EINVAL; | |
558 | ||
559 | memcpy(dst, src + sizeof(u32), dst_len); | |
560 | return 0; | |
561 | } | |
562 | ||
563 | static void handle_poolop_reply(struct ceph_mon_client *monc, | |
564 | struct ceph_msg *msg) | |
565 | { | |
566 | struct ceph_mon_generic_request *req; | |
567 | struct ceph_mon_poolop_reply *reply = msg->front.iov_base; | |
568 | u64 tid = le64_to_cpu(msg->hdr.tid); | |
569 | ||
570 | if (msg->front.iov_len < sizeof(*reply)) | |
571 | goto bad; | |
572 | dout("handle_poolop_reply %p tid %llu\n", msg, tid); | |
ba75bb98 SW |
573 | |
574 | mutex_lock(&monc->mutex); | |
e56fa10e YS |
575 | req = __lookup_generic_req(monc, tid); |
576 | if (req) { | |
577 | if (req->buf_len && | |
578 | get_poolop_reply_buf(msg->front.iov_base + sizeof(*reply), | |
579 | msg->front.iov_len - sizeof(*reply), | |
580 | req->buf, req->buf_len) < 0) { | |
581 | mutex_unlock(&monc->mutex); | |
582 | goto bad; | |
583 | } | |
584 | req->result = le32_to_cpu(reply->reply_code); | |
585 | get_generic_request(req); | |
586 | } | |
ba75bb98 | 587 | mutex_unlock(&monc->mutex); |
e56fa10e YS |
588 | if (req) { |
589 | complete(&req->completion); | |
590 | put_generic_request(req); | |
591 | } | |
592 | return; | |
ba75bb98 | 593 | |
e56fa10e YS |
594 | bad: |
595 | pr_err("corrupt generic reply, tid %llu\n", tid); | |
596 | ceph_msg_dump(msg); | |
597 | } | |
598 | ||
599 | /* | |
600 | * Do a synchronous pool op. | |
601 | */ | |
602 | int ceph_monc_do_poolop(struct ceph_mon_client *monc, u32 op, | |
603 | u32 pool, u64 snapid, | |
604 | char *buf, int len) | |
605 | { | |
606 | struct ceph_mon_generic_request *req; | |
607 | struct ceph_mon_poolop *h; | |
608 | int err; | |
609 | ||
610 | req = kzalloc(sizeof(*req), GFP_NOFS); | |
611 | if (!req) | |
612 | return -ENOMEM; | |
613 | ||
614 | kref_init(&req->kref); | |
615 | req->buf = buf; | |
616 | req->buf_len = len; | |
617 | init_completion(&req->completion); | |
618 | ||
619 | err = -ENOMEM; | |
620 | req->request = ceph_msg_new(CEPH_MSG_POOLOP, sizeof(*h), GFP_NOFS); | |
621 | if (!req->request) | |
622 | goto out; | |
623 | req->reply = ceph_msg_new(CEPH_MSG_POOLOP_REPLY, 1024, GFP_NOFS); | |
624 | if (!req->reply) | |
625 | goto out; | |
626 | ||
627 | /* fill out request */ | |
628 | req->request->hdr.version = cpu_to_le16(2); | |
629 | h = req->request->front.iov_base; | |
630 | h->monhdr.have_version = 0; | |
631 | h->monhdr.session_mon = cpu_to_le16(-1); | |
632 | h->monhdr.session_mon_tid = 0; | |
633 | h->fsid = monc->monmap->fsid; | |
634 | h->pool = cpu_to_le32(pool); | |
635 | h->op = cpu_to_le32(op); | |
636 | h->auid = 0; | |
637 | h->snapid = cpu_to_le64(snapid); | |
638 | h->name_len = 0; | |
639 | ||
640 | err = do_generic_request(monc, req); | |
3143edd3 SW |
641 | |
642 | out: | |
f8c76f6f | 643 | kref_put(&req->kref, release_generic_request); |
ba75bb98 SW |
644 | return err; |
645 | } | |
646 | ||
e56fa10e YS |
647 | int ceph_monc_create_snapid(struct ceph_mon_client *monc, |
648 | u32 pool, u64 *snapid) | |
649 | { | |
650 | return ceph_monc_do_poolop(monc, POOL_OP_CREATE_UNMANAGED_SNAP, | |
651 | pool, 0, (char *)snapid, sizeof(*snapid)); | |
652 | ||
653 | } | |
654 | ||
655 | int ceph_monc_delete_snapid(struct ceph_mon_client *monc, | |
656 | u32 pool, u64 snapid) | |
657 | { | |
658 | return ceph_monc_do_poolop(monc, POOL_OP_CREATE_UNMANAGED_SNAP, | |
659 | pool, snapid, 0, 0); | |
660 | ||
661 | } | |
662 | ||
ba75bb98 | 663 | /* |
e56fa10e | 664 | * Resend pending generic requests. |
ba75bb98 | 665 | */ |
f8c76f6f | 666 | static void __resend_generic_request(struct ceph_mon_client *monc) |
ba75bb98 | 667 | { |
f8c76f6f | 668 | struct ceph_mon_generic_request *req; |
85ff03f6 | 669 | struct rb_node *p; |
ba75bb98 | 670 | |
f8c76f6f YS |
671 | for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) { |
672 | req = rb_entry(p, struct ceph_mon_generic_request, node); | |
97069001 | 673 | ceph_con_revoke(monc->con, req->request); |
3143edd3 | 674 | ceph_con_send(monc->con, ceph_msg_get(req->request)); |
ba75bb98 SW |
675 | } |
676 | } | |
677 | ||
678 | /* | |
679 | * Delayed work. If we haven't mounted yet, retry. Otherwise, | |
680 | * renew/retry subscription as needed (in case it is timing out, or we | |
681 | * got an ENOMEM). And keep the monitor connection alive. | |
682 | */ | |
683 | static void delayed_work(struct work_struct *work) | |
684 | { | |
685 | struct ceph_mon_client *monc = | |
686 | container_of(work, struct ceph_mon_client, delayed_work.work); | |
687 | ||
688 | dout("monc delayed_work\n"); | |
689 | mutex_lock(&monc->mutex); | |
4e7a5dcd SW |
690 | if (monc->hunting) { |
691 | __close_session(monc); | |
692 | __open_session(monc); /* continue hunting */ | |
ba75bb98 | 693 | } else { |
4e7a5dcd | 694 | ceph_con_keepalive(monc->con); |
9bd2e6f8 SW |
695 | |
696 | __validate_auth(monc); | |
697 | ||
4e7a5dcd SW |
698 | if (monc->auth->ops->is_authenticated(monc->auth)) |
699 | __send_subscribe(monc); | |
ba75bb98 | 700 | } |
ba75bb98 SW |
701 | __schedule_delayed(monc); |
702 | mutex_unlock(&monc->mutex); | |
703 | } | |
704 | ||
6b805185 SW |
705 | /* |
706 | * On startup, we build a temporary monmap populated with the IPs | |
707 | * provided by mount(2). | |
708 | */ | |
709 | static int build_initial_monmap(struct ceph_mon_client *monc) | |
710 | { | |
711 | struct ceph_mount_args *args = monc->client->mount_args; | |
712 | struct ceph_entity_addr *mon_addr = args->mon_addr; | |
713 | int num_mon = args->num_mon; | |
714 | int i; | |
715 | ||
716 | /* build initial monmap */ | |
717 | monc->monmap = kzalloc(sizeof(*monc->monmap) + | |
718 | num_mon*sizeof(monc->monmap->mon_inst[0]), | |
719 | GFP_KERNEL); | |
720 | if (!monc->monmap) | |
721 | return -ENOMEM; | |
722 | for (i = 0; i < num_mon; i++) { | |
723 | monc->monmap->mon_inst[i].addr = mon_addr[i]; | |
6b805185 SW |
724 | monc->monmap->mon_inst[i].addr.nonce = 0; |
725 | monc->monmap->mon_inst[i].name.type = | |
726 | CEPH_ENTITY_TYPE_MON; | |
727 | monc->monmap->mon_inst[i].name.num = cpu_to_le64(i); | |
728 | } | |
729 | monc->monmap->num_mon = num_mon; | |
4e7a5dcd | 730 | monc->have_fsid = false; |
6b805185 SW |
731 | |
732 | /* release addr memory */ | |
733 | kfree(args->mon_addr); | |
734 | args->mon_addr = NULL; | |
735 | args->num_mon = 0; | |
736 | return 0; | |
737 | } | |
738 | ||
ba75bb98 SW |
739 | int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl) |
740 | { | |
741 | int err = 0; | |
742 | ||
743 | dout("init\n"); | |
744 | memset(monc, 0, sizeof(*monc)); | |
745 | monc->client = cl; | |
746 | monc->monmap = NULL; | |
747 | mutex_init(&monc->mutex); | |
748 | ||
6b805185 SW |
749 | err = build_initial_monmap(monc); |
750 | if (err) | |
751 | goto out; | |
752 | ||
ba75bb98 SW |
753 | monc->con = NULL; |
754 | ||
4e7a5dcd SW |
755 | /* authentication */ |
756 | monc->auth = ceph_auth_init(cl->mount_args->name, | |
757 | cl->mount_args->secret); | |
758 | if (IS_ERR(monc->auth)) | |
759 | return PTR_ERR(monc->auth); | |
760 | monc->auth->want_keys = | |
761 | CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON | | |
762 | CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS; | |
763 | ||
240ed68e | 764 | /* msgs */ |
a79832f2 | 765 | err = -ENOMEM; |
7c315c55 | 766 | monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK, |
34d23762 YS |
767 | sizeof(struct ceph_mon_subscribe_ack), |
768 | GFP_NOFS); | |
a79832f2 | 769 | if (!monc->m_subscribe_ack) |
4e7a5dcd | 770 | goto out_monmap; |
6694d6b9 | 771 | |
240ed68e SW |
772 | monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS); |
773 | if (!monc->m_subscribe) | |
774 | goto out_subscribe_ack; | |
775 | ||
34d23762 | 776 | monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS); |
a79832f2 | 777 | if (!monc->m_auth_reply) |
240ed68e | 778 | goto out_subscribe; |
4e7a5dcd | 779 | |
34d23762 | 780 | monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS); |
9bd2e6f8 | 781 | monc->pending_auth = 0; |
a79832f2 | 782 | if (!monc->m_auth) |
6694d6b9 | 783 | goto out_auth_reply; |
ba75bb98 SW |
784 | |
785 | monc->cur_mon = -1; | |
4e7a5dcd | 786 | monc->hunting = true; |
ba75bb98 SW |
787 | monc->sub_renew_after = jiffies; |
788 | monc->sub_sent = 0; | |
789 | ||
790 | INIT_DELAYED_WORK(&monc->delayed_work, delayed_work); | |
f8c76f6f YS |
791 | monc->generic_request_tree = RB_ROOT; |
792 | monc->num_generic_requests = 0; | |
ba75bb98 SW |
793 | monc->last_tid = 0; |
794 | ||
795 | monc->have_mdsmap = 0; | |
796 | monc->have_osdmap = 0; | |
797 | monc->want_next_osdmap = 1; | |
4e7a5dcd SW |
798 | return 0; |
799 | ||
6694d6b9 SW |
800 | out_auth_reply: |
801 | ceph_msg_put(monc->m_auth_reply); | |
240ed68e SW |
802 | out_subscribe: |
803 | ceph_msg_put(monc->m_subscribe); | |
7c315c55 SW |
804 | out_subscribe_ack: |
805 | ceph_msg_put(monc->m_subscribe_ack); | |
4e7a5dcd SW |
806 | out_monmap: |
807 | kfree(monc->monmap); | |
ba75bb98 SW |
808 | out: |
809 | return err; | |
810 | } | |
811 | ||
812 | void ceph_monc_stop(struct ceph_mon_client *monc) | |
813 | { | |
814 | dout("stop\n"); | |
815 | cancel_delayed_work_sync(&monc->delayed_work); | |
816 | ||
817 | mutex_lock(&monc->mutex); | |
818 | __close_session(monc); | |
819 | if (monc->con) { | |
820 | monc->con->private = NULL; | |
821 | monc->con->ops->put(monc->con); | |
822 | monc->con = NULL; | |
823 | } | |
824 | mutex_unlock(&monc->mutex); | |
825 | ||
4e7a5dcd SW |
826 | ceph_auth_destroy(monc->auth); |
827 | ||
828 | ceph_msg_put(monc->m_auth); | |
6694d6b9 | 829 | ceph_msg_put(monc->m_auth_reply); |
240ed68e | 830 | ceph_msg_put(monc->m_subscribe); |
7c315c55 | 831 | ceph_msg_put(monc->m_subscribe_ack); |
ba75bb98 SW |
832 | |
833 | kfree(monc->monmap); | |
834 | } | |
835 | ||
4e7a5dcd SW |
836 | static void handle_auth_reply(struct ceph_mon_client *monc, |
837 | struct ceph_msg *msg) | |
838 | { | |
839 | int ret; | |
09c4d6a7 | 840 | int was_auth = 0; |
4e7a5dcd SW |
841 | |
842 | mutex_lock(&monc->mutex); | |
09c4d6a7 SW |
843 | if (monc->auth->ops) |
844 | was_auth = monc->auth->ops->is_authenticated(monc->auth); | |
9bd2e6f8 | 845 | monc->pending_auth = 0; |
4e7a5dcd SW |
846 | ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base, |
847 | msg->front.iov_len, | |
848 | monc->m_auth->front.iov_base, | |
849 | monc->m_auth->front_max); | |
850 | if (ret < 0) { | |
9bd2e6f8 | 851 | monc->client->auth_err = ret; |
03066f23 | 852 | wake_up_all(&monc->client->auth_wq); |
4e7a5dcd | 853 | } else if (ret > 0) { |
9bd2e6f8 | 854 | __send_prepared_auth_request(monc, ret); |
09c4d6a7 | 855 | } else if (!was_auth && monc->auth->ops->is_authenticated(monc->auth)) { |
4e7a5dcd | 856 | dout("authenticated, starting session\n"); |
0743304d SW |
857 | |
858 | monc->client->msgr->inst.name.type = CEPH_ENTITY_TYPE_CLIENT; | |
0cf5537b YS |
859 | monc->client->msgr->inst.name.num = |
860 | cpu_to_le64(monc->auth->global_id); | |
0743304d | 861 | |
4e7a5dcd | 862 | __send_subscribe(monc); |
f8c76f6f | 863 | __resend_generic_request(monc); |
4e7a5dcd SW |
864 | } |
865 | mutex_unlock(&monc->mutex); | |
866 | } | |
867 | ||
9bd2e6f8 SW |
868 | static int __validate_auth(struct ceph_mon_client *monc) |
869 | { | |
870 | int ret; | |
871 | ||
872 | if (monc->pending_auth) | |
873 | return 0; | |
874 | ||
875 | ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base, | |
876 | monc->m_auth->front_max); | |
877 | if (ret <= 0) | |
878 | return ret; /* either an error, or no need to authenticate */ | |
879 | __send_prepared_auth_request(monc, ret); | |
880 | return 0; | |
881 | } | |
882 | ||
883 | int ceph_monc_validate_auth(struct ceph_mon_client *monc) | |
884 | { | |
885 | int ret; | |
886 | ||
887 | mutex_lock(&monc->mutex); | |
888 | ret = __validate_auth(monc); | |
889 | mutex_unlock(&monc->mutex); | |
890 | return ret; | |
891 | } | |
892 | ||
ba75bb98 SW |
893 | /* |
894 | * handle incoming message | |
895 | */ | |
896 | static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) | |
897 | { | |
898 | struct ceph_mon_client *monc = con->private; | |
899 | int type = le16_to_cpu(msg->hdr.type); | |
900 | ||
901 | if (!monc) | |
902 | return; | |
903 | ||
904 | switch (type) { | |
4e7a5dcd SW |
905 | case CEPH_MSG_AUTH_REPLY: |
906 | handle_auth_reply(monc, msg); | |
ba75bb98 SW |
907 | break; |
908 | ||
909 | case CEPH_MSG_MON_SUBSCRIBE_ACK: | |
910 | handle_subscribe_ack(monc, msg); | |
911 | break; | |
912 | ||
913 | case CEPH_MSG_STATFS_REPLY: | |
914 | handle_statfs_reply(monc, msg); | |
915 | break; | |
916 | ||
e56fa10e YS |
917 | case CEPH_MSG_POOLOP_REPLY: |
918 | handle_poolop_reply(monc, msg); | |
919 | break; | |
920 | ||
4e7a5dcd SW |
921 | case CEPH_MSG_MON_MAP: |
922 | ceph_monc_handle_map(monc, msg); | |
923 | break; | |
924 | ||
ba75bb98 SW |
925 | case CEPH_MSG_MDS_MAP: |
926 | ceph_mdsc_handle_map(&monc->client->mdsc, msg); | |
927 | break; | |
928 | ||
929 | case CEPH_MSG_OSD_MAP: | |
930 | ceph_osdc_handle_map(&monc->client->osdc, msg); | |
931 | break; | |
932 | ||
933 | default: | |
934 | pr_err("received unknown message type %d %s\n", type, | |
935 | ceph_msg_type_name(type)); | |
936 | } | |
937 | ceph_msg_put(msg); | |
938 | } | |
939 | ||
940 | /* | |
941 | * Allocate memory for incoming message | |
942 | */ | |
943 | static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con, | |
2450418c YS |
944 | struct ceph_msg_header *hdr, |
945 | int *skip) | |
ba75bb98 SW |
946 | { |
947 | struct ceph_mon_client *monc = con->private; | |
948 | int type = le16_to_cpu(hdr->type); | |
2450418c | 949 | int front_len = le32_to_cpu(hdr->front_len); |
5b3a4db3 | 950 | struct ceph_msg *m = NULL; |
ba75bb98 | 951 | |
2450418c | 952 | *skip = 0; |
0547a9b3 | 953 | |
ba75bb98 | 954 | switch (type) { |
ba75bb98 | 955 | case CEPH_MSG_MON_SUBSCRIBE_ACK: |
7c315c55 | 956 | m = ceph_msg_get(monc->m_subscribe_ack); |
2450418c | 957 | break; |
e56fa10e | 958 | case CEPH_MSG_POOLOP_REPLY: |
ba75bb98 | 959 | case CEPH_MSG_STATFS_REPLY: |
f8c76f6f | 960 | return get_generic_reply(con, hdr, skip); |
4e7a5dcd | 961 | case CEPH_MSG_AUTH_REPLY: |
6694d6b9 | 962 | m = ceph_msg_get(monc->m_auth_reply); |
2450418c | 963 | break; |
5b3a4db3 SW |
964 | case CEPH_MSG_MON_MAP: |
965 | case CEPH_MSG_MDS_MAP: | |
966 | case CEPH_MSG_OSD_MAP: | |
34d23762 | 967 | m = ceph_msg_new(type, front_len, GFP_NOFS); |
5b3a4db3 | 968 | break; |
ba75bb98 | 969 | } |
2450418c | 970 | |
5b3a4db3 SW |
971 | if (!m) { |
972 | pr_info("alloc_msg unknown type %d\n", type); | |
2450418c | 973 | *skip = 1; |
5b3a4db3 | 974 | } |
2450418c | 975 | return m; |
ba75bb98 SW |
976 | } |
977 | ||
978 | /* | |
979 | * If the monitor connection resets, pick a new monitor and resubmit | |
980 | * any pending requests. | |
981 | */ | |
982 | static void mon_fault(struct ceph_connection *con) | |
983 | { | |
984 | struct ceph_mon_client *monc = con->private; | |
985 | ||
986 | if (!monc) | |
987 | return; | |
988 | ||
989 | dout("mon_fault\n"); | |
990 | mutex_lock(&monc->mutex); | |
991 | if (!con->private) | |
992 | goto out; | |
993 | ||
994 | if (monc->con && !monc->hunting) | |
995 | pr_info("mon%d %s session lost, " | |
996 | "hunting for new mon\n", monc->cur_mon, | |
997 | pr_addr(&monc->con->peer_addr.in_addr)); | |
998 | ||
999 | __close_session(monc); | |
1000 | if (!monc->hunting) { | |
1001 | /* start hunting */ | |
1002 | monc->hunting = true; | |
4e7a5dcd | 1003 | __open_session(monc); |
ba75bb98 SW |
1004 | } else { |
1005 | /* already hunting, let's wait a bit */ | |
1006 | __schedule_delayed(monc); | |
1007 | } | |
1008 | out: | |
1009 | mutex_unlock(&monc->mutex); | |
1010 | } | |
1011 | ||
9e32789f | 1012 | static const struct ceph_connection_operations mon_con_ops = { |
ba75bb98 SW |
1013 | .get = ceph_con_get, |
1014 | .put = ceph_con_put, | |
1015 | .dispatch = dispatch, | |
1016 | .fault = mon_fault, | |
1017 | .alloc_msg = mon_alloc_msg, | |
ba75bb98 | 1018 | }; |