1 #include <linux/ceph/ceph_debug.h> 2 3 #include <linux/module.h> 4 #include <linux/types.h> 5 #include <linux/slab.h> 6 #include <linux/random.h> 7 #include <linux/sched.h> 8 9 #include <linux/ceph/mon_client.h> 10 #include <linux/ceph/libceph.h> 11 #include <linux/ceph/debugfs.h> 12 #include <linux/ceph/decode.h> 13 #include <linux/ceph/auth.h> 14 15 /* 16 * Interact with Ceph monitor cluster. Handle requests for new map 17 * versions, and periodically resend as needed. Also implement 18 * statfs() and umount(). 19 * 20 * A small cluster of Ceph "monitors" are responsible for managing critical 21 * cluster configuration and state information. An odd number (e.g., 3, 5) 22 * of cmon daemons use a modified version of the Paxos part-time parliament 23 * algorithm to manage the MDS map (mds cluster membership), OSD map, and 24 * list of clients who have mounted the file system. 25 * 26 * We maintain an open, active session with a monitor at all times in order to 27 * receive timely MDSMap updates. We periodically send a keepalive byte on the 28 * TCP socket to ensure we detect a failure. If the connection does break, we 29 * randomly hunt for a new monitor. Once the connection is reestablished, we 30 * resend any outstanding requests. 31 */ 32 33 static const struct ceph_connection_operations mon_con_ops; 34 35 static int __validate_auth(struct ceph_mon_client *monc); 36 37 /* 38 * Decode a monmap blob (e.g., during mount). 39 */ 40 struct ceph_monmap *ceph_monmap_decode(void *p, void *end) 41 { 42 struct ceph_monmap *m = NULL; 43 int i, err = -EINVAL; 44 struct ceph_fsid fsid; 45 u32 epoch, num_mon; 46 u16 version; 47 u32 len; 48 49 ceph_decode_32_safe(&p, end, len, bad); 50 ceph_decode_need(&p, end, len, bad); 51 52 dout("monmap_decode %p %p len %d\n", p, end, (int)(end-p)); 53 54 ceph_decode_16_safe(&p, end, version, bad); 55 56 ceph_decode_need(&p, end, sizeof(fsid) + 2*sizeof(u32), bad); 57 ceph_decode_copy(&p, &fsid, sizeof(fsid)); 58 epoch = ceph_decode_32(&p); 59 60 num_mon = ceph_decode_32(&p); 61 ceph_decode_need(&p, end, num_mon*sizeof(m->mon_inst[0]), bad); 62 63 if (num_mon >= CEPH_MAX_MON) 64 goto bad; 65 m = kmalloc(sizeof(*m) + sizeof(m->mon_inst[0])*num_mon, GFP_NOFS); 66 if (m == NULL) 67 return ERR_PTR(-ENOMEM); 68 m->fsid = fsid; 69 m->epoch = epoch; 70 m->num_mon = num_mon; 71 ceph_decode_copy(&p, m->mon_inst, num_mon*sizeof(m->mon_inst[0])); 72 for (i = 0; i < num_mon; i++) 73 ceph_decode_addr(&m->mon_inst[i].addr); 74 75 dout("monmap_decode epoch %d, num_mon %d\n", m->epoch, 76 m->num_mon); 77 for (i = 0; i < m->num_mon; i++) 78 dout("monmap_decode mon%d is %s\n", i, 79 ceph_pr_addr(&m->mon_inst[i].addr.in_addr)); 80 return m; 81 82 bad: 83 dout("monmap_decode failed with %d\n", err); 84 kfree(m); 85 return ERR_PTR(err); 86 } 87 88 /* 89 * return true if *addr is included in the monmap. 90 */ 91 int ceph_monmap_contains(struct ceph_monmap *m, struct ceph_entity_addr *addr) 92 { 93 int i; 94 95 for (i = 0; i < m->num_mon; i++) 96 if (memcmp(addr, &m->mon_inst[i].addr, sizeof(*addr)) == 0) 97 return 1; 98 return 0; 99 } 100 101 /* 102 * Send an auth request. 103 */ 104 static void __send_prepared_auth_request(struct ceph_mon_client *monc, int len) 105 { 106 monc->pending_auth = 1; 107 monc->m_auth->front.iov_len = len; 108 monc->m_auth->hdr.front_len = cpu_to_le32(len); 109 ceph_msg_revoke(monc->m_auth); 110 ceph_msg_get(monc->m_auth); /* keep our ref */ 111 ceph_con_send(&monc->con, monc->m_auth); 112 } 113 114 /* 115 * Close monitor session, if any. 116 */ 117 static void __close_session(struct ceph_mon_client *monc) 118 { 119 dout("__close_session closing mon%d\n", monc->cur_mon); 120 ceph_msg_revoke(monc->m_auth); 121 ceph_msg_revoke_incoming(monc->m_auth_reply); 122 ceph_msg_revoke(monc->m_subscribe); 123 ceph_msg_revoke_incoming(monc->m_subscribe_ack); 124 ceph_con_close(&monc->con); 125 monc->cur_mon = -1; 126 monc->pending_auth = 0; 127 ceph_auth_reset(monc->auth); 128 } 129 130 /* 131 * Open a session with a (new) monitor. 132 */ 133 static int __open_session(struct ceph_mon_client *monc) 134 { 135 char r; 136 int ret; 137 138 if (monc->cur_mon < 0) { 139 get_random_bytes(&r, 1); 140 monc->cur_mon = r % monc->monmap->num_mon; 141 dout("open_session num=%d r=%d -> mon%d\n", 142 monc->monmap->num_mon, r, monc->cur_mon); 143 monc->sub_sent = 0; 144 monc->sub_renew_after = jiffies; /* i.e., expired */ 145 monc->want_next_osdmap = !!monc->want_next_osdmap; 146 147 dout("open_session mon%d opening\n", monc->cur_mon); 148 ceph_con_open(&monc->con, 149 CEPH_ENTITY_TYPE_MON, monc->cur_mon, 150 &monc->monmap->mon_inst[monc->cur_mon].addr); 151 152 /* initiatiate authentication handshake */ 153 ret = ceph_auth_build_hello(monc->auth, 154 monc->m_auth->front.iov_base, 155 monc->m_auth->front_alloc_len); 156 __send_prepared_auth_request(monc, ret); 157 } else { 158 dout("open_session mon%d already open\n", monc->cur_mon); 159 } 160 return 0; 161 } 162 163 static bool __sub_expired(struct ceph_mon_client *monc) 164 { 165 return time_after_eq(jiffies, monc->sub_renew_after); 166 } 167 168 /* 169 * Reschedule delayed work timer. 170 */ 171 static void __schedule_delayed(struct ceph_mon_client *monc) 172 { 173 unsigned int delay; 174 175 if (monc->cur_mon < 0 || __sub_expired(monc)) 176 delay = 10 * HZ; 177 else 178 delay = 20 * HZ; 179 dout("__schedule_delayed after %u\n", delay); 180 schedule_delayed_work(&monc->delayed_work, delay); 181 } 182 183 /* 184 * Send subscribe request for mdsmap and/or osdmap. 185 */ 186 static void __send_subscribe(struct ceph_mon_client *monc) 187 { 188 dout("__send_subscribe sub_sent=%u exp=%u want_osd=%d\n", 189 (unsigned int)monc->sub_sent, __sub_expired(monc), 190 monc->want_next_osdmap); 191 if ((__sub_expired(monc) && !monc->sub_sent) || 192 monc->want_next_osdmap == 1) { 193 struct ceph_msg *msg = monc->m_subscribe; 194 struct ceph_mon_subscribe_item *i; 195 void *p, *end; 196 int num; 197 198 p = msg->front.iov_base; 199 end = p + msg->front_alloc_len; 200 201 num = 1 + !!monc->want_next_osdmap + !!monc->want_mdsmap; 202 ceph_encode_32(&p, num); 203 204 if (monc->want_next_osdmap) { 205 dout("__send_subscribe to 'osdmap' %u\n", 206 (unsigned int)monc->have_osdmap); 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 } 214 if (monc->want_mdsmap) { 215 dout("__send_subscribe to 'mdsmap' %u+\n", 216 (unsigned int)monc->have_mdsmap); 217 ceph_encode_string(&p, end, "mdsmap", 6); 218 i = p; 219 i->have = cpu_to_le64(monc->have_mdsmap); 220 i->onetime = 0; 221 p += sizeof(*i); 222 } 223 ceph_encode_string(&p, end, "monmap", 6); 224 i = p; 225 i->have = 0; 226 i->onetime = 0; 227 p += sizeof(*i); 228 229 msg->front.iov_len = p - msg->front.iov_base; 230 msg->hdr.front_len = cpu_to_le32(msg->front.iov_len); 231 ceph_msg_revoke(msg); 232 ceph_con_send(&monc->con, ceph_msg_get(msg)); 233 234 monc->sub_sent = jiffies | 1; /* never 0 */ 235 } 236 } 237 238 static void handle_subscribe_ack(struct ceph_mon_client *monc, 239 struct ceph_msg *msg) 240 { 241 unsigned int seconds; 242 struct ceph_mon_subscribe_ack *h = msg->front.iov_base; 243 244 if (msg->front.iov_len < sizeof(*h)) 245 goto bad; 246 seconds = le32_to_cpu(h->duration); 247 248 mutex_lock(&monc->mutex); 249 if (monc->hunting) { 250 pr_info("mon%d %s session established\n", 251 monc->cur_mon, 252 ceph_pr_addr(&monc->con.peer_addr.in_addr)); 253 monc->hunting = false; 254 } 255 dout("handle_subscribe_ack after %d seconds\n", seconds); 256 monc->sub_renew_after = monc->sub_sent + (seconds >> 1)*HZ - 1; 257 monc->sub_sent = 0; 258 mutex_unlock(&monc->mutex); 259 return; 260 bad: 261 pr_err("got corrupt subscribe-ack msg\n"); 262 ceph_msg_dump(msg); 263 } 264 265 /* 266 * Keep track of which maps we have 267 */ 268 int ceph_monc_got_mdsmap(struct ceph_mon_client *monc, u32 got) 269 { 270 mutex_lock(&monc->mutex); 271 monc->have_mdsmap = got; 272 mutex_unlock(&monc->mutex); 273 return 0; 274 } 275 EXPORT_SYMBOL(ceph_monc_got_mdsmap); 276 277 int ceph_monc_got_osdmap(struct ceph_mon_client *monc, u32 got) 278 { 279 mutex_lock(&monc->mutex); 280 monc->have_osdmap = got; 281 monc->want_next_osdmap = 0; 282 mutex_unlock(&monc->mutex); 283 return 0; 284 } 285 286 /* 287 * Register interest in the next osdmap 288 */ 289 void ceph_monc_request_next_osdmap(struct ceph_mon_client *monc) 290 { 291 dout("request_next_osdmap have %u\n", monc->have_osdmap); 292 mutex_lock(&monc->mutex); 293 if (!monc->want_next_osdmap) 294 monc->want_next_osdmap = 1; 295 if (monc->want_next_osdmap < 2) 296 __send_subscribe(monc); 297 mutex_unlock(&monc->mutex); 298 } 299 EXPORT_SYMBOL(ceph_monc_request_next_osdmap); 300 301 int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch, 302 unsigned long timeout) 303 { 304 unsigned long started = jiffies; 305 int ret; 306 307 mutex_lock(&monc->mutex); 308 while (monc->have_osdmap < epoch) { 309 mutex_unlock(&monc->mutex); 310 311 if (timeout != 0 && time_after_eq(jiffies, started + timeout)) 312 return -ETIMEDOUT; 313 314 ret = wait_event_interruptible_timeout(monc->client->auth_wq, 315 monc->have_osdmap >= epoch, timeout); 316 if (ret < 0) 317 return ret; 318 319 mutex_lock(&monc->mutex); 320 } 321 322 mutex_unlock(&monc->mutex); 323 return 0; 324 } 325 EXPORT_SYMBOL(ceph_monc_wait_osdmap); 326 327 /* 328 * 329 */ 330 int ceph_monc_open_session(struct ceph_mon_client *monc) 331 { 332 mutex_lock(&monc->mutex); 333 __open_session(monc); 334 __schedule_delayed(monc); 335 mutex_unlock(&monc->mutex); 336 return 0; 337 } 338 EXPORT_SYMBOL(ceph_monc_open_session); 339 340 /* 341 * We require the fsid and global_id in order to initialize our 342 * debugfs dir. 343 */ 344 static bool have_debugfs_info(struct ceph_mon_client *monc) 345 { 346 dout("have_debugfs_info fsid %d globalid %lld\n", 347 (int)monc->client->have_fsid, monc->auth->global_id); 348 return monc->client->have_fsid && monc->auth->global_id > 0; 349 } 350 351 /* 352 * The monitor responds with mount ack indicate mount success. The 353 * included client ticket allows the client to talk to MDSs and OSDs. 354 */ 355 static void ceph_monc_handle_map(struct ceph_mon_client *monc, 356 struct ceph_msg *msg) 357 { 358 struct ceph_client *client = monc->client; 359 struct ceph_monmap *monmap = NULL, *old = monc->monmap; 360 void *p, *end; 361 int had_debugfs_info, init_debugfs = 0; 362 363 mutex_lock(&monc->mutex); 364 365 had_debugfs_info = have_debugfs_info(monc); 366 367 dout("handle_monmap\n"); 368 p = msg->front.iov_base; 369 end = p + msg->front.iov_len; 370 371 monmap = ceph_monmap_decode(p, end); 372 if (IS_ERR(monmap)) { 373 pr_err("problem decoding monmap, %d\n", 374 (int)PTR_ERR(monmap)); 375 goto out; 376 } 377 378 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) { 379 kfree(monmap); 380 goto out; 381 } 382 383 client->monc.monmap = monmap; 384 kfree(old); 385 386 if (!client->have_fsid) { 387 client->have_fsid = true; 388 if (!had_debugfs_info && have_debugfs_info(monc)) { 389 pr_info("client%lld fsid %pU\n", 390 ceph_client_id(monc->client), 391 &monc->client->fsid); 392 init_debugfs = 1; 393 } 394 mutex_unlock(&monc->mutex); 395 396 if (init_debugfs) { 397 /* 398 * do debugfs initialization without mutex to avoid 399 * creating a locking dependency 400 */ 401 ceph_debugfs_client_init(monc->client); 402 } 403 404 goto out_unlocked; 405 } 406 out: 407 mutex_unlock(&monc->mutex); 408 out_unlocked: 409 wake_up_all(&client->auth_wq); 410 } 411 412 /* 413 * generic requests (currently statfs, mon_get_version) 414 */ 415 static struct ceph_mon_generic_request *__lookup_generic_req( 416 struct ceph_mon_client *monc, u64 tid) 417 { 418 struct ceph_mon_generic_request *req; 419 struct rb_node *n = monc->generic_request_tree.rb_node; 420 421 while (n) { 422 req = rb_entry(n, struct ceph_mon_generic_request, node); 423 if (tid < req->tid) 424 n = n->rb_left; 425 else if (tid > req->tid) 426 n = n->rb_right; 427 else 428 return req; 429 } 430 return NULL; 431 } 432 433 static void __insert_generic_request(struct ceph_mon_client *monc, 434 struct ceph_mon_generic_request *new) 435 { 436 struct rb_node **p = &monc->generic_request_tree.rb_node; 437 struct rb_node *parent = NULL; 438 struct ceph_mon_generic_request *req = NULL; 439 440 while (*p) { 441 parent = *p; 442 req = rb_entry(parent, struct ceph_mon_generic_request, node); 443 if (new->tid < req->tid) 444 p = &(*p)->rb_left; 445 else if (new->tid > req->tid) 446 p = &(*p)->rb_right; 447 else 448 BUG(); 449 } 450 451 rb_link_node(&new->node, parent, p); 452 rb_insert_color(&new->node, &monc->generic_request_tree); 453 } 454 455 static void release_generic_request(struct kref *kref) 456 { 457 struct ceph_mon_generic_request *req = 458 container_of(kref, struct ceph_mon_generic_request, kref); 459 460 if (req->reply) 461 ceph_msg_put(req->reply); 462 if (req->request) 463 ceph_msg_put(req->request); 464 465 kfree(req); 466 } 467 468 static void put_generic_request(struct ceph_mon_generic_request *req) 469 { 470 kref_put(&req->kref, release_generic_request); 471 } 472 473 static void get_generic_request(struct ceph_mon_generic_request *req) 474 { 475 kref_get(&req->kref); 476 } 477 478 static struct ceph_msg *get_generic_reply(struct ceph_connection *con, 479 struct ceph_msg_header *hdr, 480 int *skip) 481 { 482 struct ceph_mon_client *monc = con->private; 483 struct ceph_mon_generic_request *req; 484 u64 tid = le64_to_cpu(hdr->tid); 485 struct ceph_msg *m; 486 487 mutex_lock(&monc->mutex); 488 req = __lookup_generic_req(monc, tid); 489 if (!req) { 490 dout("get_generic_reply %lld dne\n", tid); 491 *skip = 1; 492 m = NULL; 493 } else { 494 dout("get_generic_reply %lld got %p\n", tid, req->reply); 495 *skip = 0; 496 m = ceph_msg_get(req->reply); 497 /* 498 * we don't need to track the connection reading into 499 * this reply because we only have one open connection 500 * at a time, ever. 501 */ 502 } 503 mutex_unlock(&monc->mutex); 504 return m; 505 } 506 507 static int __do_generic_request(struct ceph_mon_client *monc, u64 tid, 508 struct ceph_mon_generic_request *req) 509 { 510 int err; 511 512 /* register request */ 513 req->tid = tid != 0 ? tid : ++monc->last_tid; 514 req->request->hdr.tid = cpu_to_le64(req->tid); 515 __insert_generic_request(monc, req); 516 monc->num_generic_requests++; 517 ceph_con_send(&monc->con, ceph_msg_get(req->request)); 518 mutex_unlock(&monc->mutex); 519 520 err = wait_for_completion_interruptible(&req->completion); 521 522 mutex_lock(&monc->mutex); 523 rb_erase(&req->node, &monc->generic_request_tree); 524 monc->num_generic_requests--; 525 526 if (!err) 527 err = req->result; 528 return err; 529 } 530 531 static int do_generic_request(struct ceph_mon_client *monc, 532 struct ceph_mon_generic_request *req) 533 { 534 int err; 535 536 mutex_lock(&monc->mutex); 537 err = __do_generic_request(monc, 0, req); 538 mutex_unlock(&monc->mutex); 539 540 return err; 541 } 542 543 /* 544 * statfs 545 */ 546 static void handle_statfs_reply(struct ceph_mon_client *monc, 547 struct ceph_msg *msg) 548 { 549 struct ceph_mon_generic_request *req; 550 struct ceph_mon_statfs_reply *reply = msg->front.iov_base; 551 u64 tid = le64_to_cpu(msg->hdr.tid); 552 553 if (msg->front.iov_len != sizeof(*reply)) 554 goto bad; 555 dout("handle_statfs_reply %p tid %llu\n", msg, tid); 556 557 mutex_lock(&monc->mutex); 558 req = __lookup_generic_req(monc, tid); 559 if (req) { 560 *(struct ceph_statfs *)req->buf = reply->st; 561 req->result = 0; 562 get_generic_request(req); 563 } 564 mutex_unlock(&monc->mutex); 565 if (req) { 566 complete_all(&req->completion); 567 put_generic_request(req); 568 } 569 return; 570 571 bad: 572 pr_err("corrupt statfs reply, tid %llu\n", tid); 573 ceph_msg_dump(msg); 574 } 575 576 /* 577 * Do a synchronous statfs(). 578 */ 579 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf) 580 { 581 struct ceph_mon_generic_request *req; 582 struct ceph_mon_statfs *h; 583 int err; 584 585 req = kzalloc(sizeof(*req), GFP_NOFS); 586 if (!req) 587 return -ENOMEM; 588 589 kref_init(&req->kref); 590 req->buf = buf; 591 init_completion(&req->completion); 592 593 err = -ENOMEM; 594 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS, 595 true); 596 if (!req->request) 597 goto out; 598 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS, 599 true); 600 if (!req->reply) 601 goto out; 602 603 /* fill out request */ 604 h = req->request->front.iov_base; 605 h->monhdr.have_version = 0; 606 h->monhdr.session_mon = cpu_to_le16(-1); 607 h->monhdr.session_mon_tid = 0; 608 h->fsid = monc->monmap->fsid; 609 610 err = do_generic_request(monc, req); 611 612 out: 613 put_generic_request(req); 614 return err; 615 } 616 EXPORT_SYMBOL(ceph_monc_do_statfs); 617 618 static void handle_get_version_reply(struct ceph_mon_client *monc, 619 struct ceph_msg *msg) 620 { 621 struct ceph_mon_generic_request *req; 622 u64 tid = le64_to_cpu(msg->hdr.tid); 623 void *p = msg->front.iov_base; 624 void *end = p + msg->front_alloc_len; 625 u64 handle; 626 627 dout("%s %p tid %llu\n", __func__, msg, tid); 628 629 ceph_decode_need(&p, end, 2*sizeof(u64), bad); 630 handle = ceph_decode_64(&p); 631 if (tid != 0 && tid != handle) 632 goto bad; 633 634 mutex_lock(&monc->mutex); 635 req = __lookup_generic_req(monc, handle); 636 if (req) { 637 *(u64 *)req->buf = ceph_decode_64(&p); 638 req->result = 0; 639 get_generic_request(req); 640 } 641 mutex_unlock(&monc->mutex); 642 if (req) { 643 complete_all(&req->completion); 644 put_generic_request(req); 645 } 646 647 return; 648 bad: 649 pr_err("corrupt mon_get_version reply, tid %llu\n", tid); 650 ceph_msg_dump(msg); 651 } 652 653 /* 654 * Send MMonGetVersion and wait for the reply. 655 * 656 * @what: one of "mdsmap", "osdmap" or "monmap" 657 */ 658 int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what, 659 u64 *newest) 660 { 661 struct ceph_mon_generic_request *req; 662 void *p, *end; 663 u64 tid; 664 int err; 665 666 req = kzalloc(sizeof(*req), GFP_NOFS); 667 if (!req) 668 return -ENOMEM; 669 670 kref_init(&req->kref); 671 req->buf = newest; 672 init_completion(&req->completion); 673 674 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION, 675 sizeof(u64) + sizeof(u32) + strlen(what), 676 GFP_NOFS, true); 677 if (!req->request) { 678 err = -ENOMEM; 679 goto out; 680 } 681 682 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024, 683 GFP_NOFS, true); 684 if (!req->reply) { 685 err = -ENOMEM; 686 goto out; 687 } 688 689 p = req->request->front.iov_base; 690 end = p + req->request->front_alloc_len; 691 692 /* fill out request */ 693 mutex_lock(&monc->mutex); 694 tid = ++monc->last_tid; 695 ceph_encode_64(&p, tid); /* handle */ 696 ceph_encode_string(&p, end, what, strlen(what)); 697 698 err = __do_generic_request(monc, tid, req); 699 700 mutex_unlock(&monc->mutex); 701 out: 702 put_generic_request(req); 703 return err; 704 } 705 EXPORT_SYMBOL(ceph_monc_do_get_version); 706 707 /* 708 * Resend pending generic requests. 709 */ 710 static void __resend_generic_request(struct ceph_mon_client *monc) 711 { 712 struct ceph_mon_generic_request *req; 713 struct rb_node *p; 714 715 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) { 716 req = rb_entry(p, struct ceph_mon_generic_request, node); 717 ceph_msg_revoke(req->request); 718 ceph_msg_revoke_incoming(req->reply); 719 ceph_con_send(&monc->con, ceph_msg_get(req->request)); 720 } 721 } 722 723 /* 724 * Delayed work. If we haven't mounted yet, retry. Otherwise, 725 * renew/retry subscription as needed (in case it is timing out, or we 726 * got an ENOMEM). And keep the monitor connection alive. 727 */ 728 static void delayed_work(struct work_struct *work) 729 { 730 struct ceph_mon_client *monc = 731 container_of(work, struct ceph_mon_client, delayed_work.work); 732 733 dout("monc delayed_work\n"); 734 mutex_lock(&monc->mutex); 735 if (monc->hunting) { 736 __close_session(monc); 737 __open_session(monc); /* continue hunting */ 738 } else { 739 ceph_con_keepalive(&monc->con); 740 741 __validate_auth(monc); 742 743 if (ceph_auth_is_authenticated(monc->auth)) 744 __send_subscribe(monc); 745 } 746 __schedule_delayed(monc); 747 mutex_unlock(&monc->mutex); 748 } 749 750 /* 751 * On startup, we build a temporary monmap populated with the IPs 752 * provided by mount(2). 753 */ 754 static int build_initial_monmap(struct ceph_mon_client *monc) 755 { 756 struct ceph_options *opt = monc->client->options; 757 struct ceph_entity_addr *mon_addr = opt->mon_addr; 758 int num_mon = opt->num_mon; 759 int i; 760 761 /* build initial monmap */ 762 monc->monmap = kzalloc(sizeof(*monc->monmap) + 763 num_mon*sizeof(monc->monmap->mon_inst[0]), 764 GFP_KERNEL); 765 if (!monc->monmap) 766 return -ENOMEM; 767 for (i = 0; i < num_mon; i++) { 768 monc->monmap->mon_inst[i].addr = mon_addr[i]; 769 monc->monmap->mon_inst[i].addr.nonce = 0; 770 monc->monmap->mon_inst[i].name.type = 771 CEPH_ENTITY_TYPE_MON; 772 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i); 773 } 774 monc->monmap->num_mon = num_mon; 775 return 0; 776 } 777 778 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl) 779 { 780 int err = 0; 781 782 dout("init\n"); 783 memset(monc, 0, sizeof(*monc)); 784 monc->client = cl; 785 monc->monmap = NULL; 786 mutex_init(&monc->mutex); 787 788 err = build_initial_monmap(monc); 789 if (err) 790 goto out; 791 792 /* connection */ 793 /* authentication */ 794 monc->auth = ceph_auth_init(cl->options->name, 795 cl->options->key); 796 if (IS_ERR(monc->auth)) { 797 err = PTR_ERR(monc->auth); 798 goto out_monmap; 799 } 800 monc->auth->want_keys = 801 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON | 802 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS; 803 804 /* msgs */ 805 err = -ENOMEM; 806 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK, 807 sizeof(struct ceph_mon_subscribe_ack), 808 GFP_NOFS, true); 809 if (!monc->m_subscribe_ack) 810 goto out_auth; 811 812 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS, 813 true); 814 if (!monc->m_subscribe) 815 goto out_subscribe_ack; 816 817 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS, 818 true); 819 if (!monc->m_auth_reply) 820 goto out_subscribe; 821 822 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true); 823 monc->pending_auth = 0; 824 if (!monc->m_auth) 825 goto out_auth_reply; 826 827 ceph_con_init(&monc->con, monc, &mon_con_ops, 828 &monc->client->msgr); 829 830 monc->cur_mon = -1; 831 monc->hunting = true; 832 monc->sub_renew_after = jiffies; 833 monc->sub_sent = 0; 834 835 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work); 836 monc->generic_request_tree = RB_ROOT; 837 monc->num_generic_requests = 0; 838 monc->last_tid = 0; 839 840 monc->have_mdsmap = 0; 841 monc->have_osdmap = 0; 842 monc->want_next_osdmap = 1; 843 return 0; 844 845 out_auth_reply: 846 ceph_msg_put(monc->m_auth_reply); 847 out_subscribe: 848 ceph_msg_put(monc->m_subscribe); 849 out_subscribe_ack: 850 ceph_msg_put(monc->m_subscribe_ack); 851 out_auth: 852 ceph_auth_destroy(monc->auth); 853 out_monmap: 854 kfree(monc->monmap); 855 out: 856 return err; 857 } 858 EXPORT_SYMBOL(ceph_monc_init); 859 860 void ceph_monc_stop(struct ceph_mon_client *monc) 861 { 862 dout("stop\n"); 863 cancel_delayed_work_sync(&monc->delayed_work); 864 865 mutex_lock(&monc->mutex); 866 __close_session(monc); 867 868 mutex_unlock(&monc->mutex); 869 870 /* 871 * flush msgr queue before we destroy ourselves to ensure that: 872 * - any work that references our embedded con is finished. 873 * - any osd_client or other work that may reference an authorizer 874 * finishes before we shut down the auth subsystem. 875 */ 876 ceph_msgr_flush(); 877 878 ceph_auth_destroy(monc->auth); 879 880 ceph_msg_put(monc->m_auth); 881 ceph_msg_put(monc->m_auth_reply); 882 ceph_msg_put(monc->m_subscribe); 883 ceph_msg_put(monc->m_subscribe_ack); 884 885 kfree(monc->monmap); 886 } 887 EXPORT_SYMBOL(ceph_monc_stop); 888 889 static void handle_auth_reply(struct ceph_mon_client *monc, 890 struct ceph_msg *msg) 891 { 892 int ret; 893 int was_auth = 0; 894 int had_debugfs_info, init_debugfs = 0; 895 896 mutex_lock(&monc->mutex); 897 had_debugfs_info = have_debugfs_info(monc); 898 was_auth = ceph_auth_is_authenticated(monc->auth); 899 monc->pending_auth = 0; 900 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base, 901 msg->front.iov_len, 902 monc->m_auth->front.iov_base, 903 monc->m_auth->front_alloc_len); 904 if (ret < 0) { 905 monc->client->auth_err = ret; 906 wake_up_all(&monc->client->auth_wq); 907 } else if (ret > 0) { 908 __send_prepared_auth_request(monc, ret); 909 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) { 910 dout("authenticated, starting session\n"); 911 912 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT; 913 monc->client->msgr.inst.name.num = 914 cpu_to_le64(monc->auth->global_id); 915 916 __send_subscribe(monc); 917 __resend_generic_request(monc); 918 } 919 920 if (!had_debugfs_info && have_debugfs_info(monc)) { 921 pr_info("client%lld fsid %pU\n", 922 ceph_client_id(monc->client), 923 &monc->client->fsid); 924 init_debugfs = 1; 925 } 926 mutex_unlock(&monc->mutex); 927 928 if (init_debugfs) { 929 /* 930 * do debugfs initialization without mutex to avoid 931 * creating a locking dependency 932 */ 933 ceph_debugfs_client_init(monc->client); 934 } 935 } 936 937 static int __validate_auth(struct ceph_mon_client *monc) 938 { 939 int ret; 940 941 if (monc->pending_auth) 942 return 0; 943 944 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base, 945 monc->m_auth->front_alloc_len); 946 if (ret <= 0) 947 return ret; /* either an error, or no need to authenticate */ 948 __send_prepared_auth_request(monc, ret); 949 return 0; 950 } 951 952 int ceph_monc_validate_auth(struct ceph_mon_client *monc) 953 { 954 int ret; 955 956 mutex_lock(&monc->mutex); 957 ret = __validate_auth(monc); 958 mutex_unlock(&monc->mutex); 959 return ret; 960 } 961 EXPORT_SYMBOL(ceph_monc_validate_auth); 962 963 /* 964 * handle incoming message 965 */ 966 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) 967 { 968 struct ceph_mon_client *monc = con->private; 969 int type = le16_to_cpu(msg->hdr.type); 970 971 if (!monc) 972 return; 973 974 switch (type) { 975 case CEPH_MSG_AUTH_REPLY: 976 handle_auth_reply(monc, msg); 977 break; 978 979 case CEPH_MSG_MON_SUBSCRIBE_ACK: 980 handle_subscribe_ack(monc, msg); 981 break; 982 983 case CEPH_MSG_STATFS_REPLY: 984 handle_statfs_reply(monc, msg); 985 break; 986 987 case CEPH_MSG_MON_GET_VERSION_REPLY: 988 handle_get_version_reply(monc, msg); 989 break; 990 991 case CEPH_MSG_MON_MAP: 992 ceph_monc_handle_map(monc, msg); 993 break; 994 995 case CEPH_MSG_OSD_MAP: 996 ceph_osdc_handle_map(&monc->client->osdc, msg); 997 break; 998 999 default: 1000 /* can the chained handler handle it? */ 1001 if (monc->client->extra_mon_dispatch && 1002 monc->client->extra_mon_dispatch(monc->client, msg) == 0) 1003 break; 1004 1005 pr_err("received unknown message type %d %s\n", type, 1006 ceph_msg_type_name(type)); 1007 } 1008 ceph_msg_put(msg); 1009 } 1010 1011 /* 1012 * Allocate memory for incoming message 1013 */ 1014 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con, 1015 struct ceph_msg_header *hdr, 1016 int *skip) 1017 { 1018 struct ceph_mon_client *monc = con->private; 1019 int type = le16_to_cpu(hdr->type); 1020 int front_len = le32_to_cpu(hdr->front_len); 1021 struct ceph_msg *m = NULL; 1022 1023 *skip = 0; 1024 1025 switch (type) { 1026 case CEPH_MSG_MON_SUBSCRIBE_ACK: 1027 m = ceph_msg_get(monc->m_subscribe_ack); 1028 break; 1029 case CEPH_MSG_STATFS_REPLY: 1030 return get_generic_reply(con, hdr, skip); 1031 case CEPH_MSG_AUTH_REPLY: 1032 m = ceph_msg_get(monc->m_auth_reply); 1033 break; 1034 case CEPH_MSG_MON_GET_VERSION_REPLY: 1035 if (le64_to_cpu(hdr->tid) != 0) 1036 return get_generic_reply(con, hdr, skip); 1037 1038 /* 1039 * Older OSDs don't set reply tid even if the orignal 1040 * request had a non-zero tid. Workaround this weirdness 1041 * by falling through to the allocate case. 1042 */ 1043 case CEPH_MSG_MON_MAP: 1044 case CEPH_MSG_MDS_MAP: 1045 case CEPH_MSG_OSD_MAP: 1046 m = ceph_msg_new(type, front_len, GFP_NOFS, false); 1047 if (!m) 1048 return NULL; /* ENOMEM--return skip == 0 */ 1049 break; 1050 } 1051 1052 if (!m) { 1053 pr_info("alloc_msg unknown type %d\n", type); 1054 *skip = 1; 1055 } else if (front_len > m->front_alloc_len) { 1056 pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n", 1057 front_len, m->front_alloc_len, 1058 (unsigned int)con->peer_name.type, 1059 le64_to_cpu(con->peer_name.num)); 1060 ceph_msg_put(m); 1061 m = ceph_msg_new(type, front_len, GFP_NOFS, false); 1062 } 1063 1064 return m; 1065 } 1066 1067 /* 1068 * If the monitor connection resets, pick a new monitor and resubmit 1069 * any pending requests. 1070 */ 1071 static void mon_fault(struct ceph_connection *con) 1072 { 1073 struct ceph_mon_client *monc = con->private; 1074 1075 if (!monc) 1076 return; 1077 1078 dout("mon_fault\n"); 1079 mutex_lock(&monc->mutex); 1080 if (!con->private) 1081 goto out; 1082 1083 if (!monc->hunting) 1084 pr_info("mon%d %s session lost, " 1085 "hunting for new mon\n", monc->cur_mon, 1086 ceph_pr_addr(&monc->con.peer_addr.in_addr)); 1087 1088 __close_session(monc); 1089 if (!monc->hunting) { 1090 /* start hunting */ 1091 monc->hunting = true; 1092 __open_session(monc); 1093 } else { 1094 /* already hunting, let's wait a bit */ 1095 __schedule_delayed(monc); 1096 } 1097 out: 1098 mutex_unlock(&monc->mutex); 1099 } 1100 1101 /* 1102 * We can ignore refcounting on the connection struct, as all references 1103 * will come from the messenger workqueue, which is drained prior to 1104 * mon_client destruction. 1105 */ 1106 static struct ceph_connection *con_get(struct ceph_connection *con) 1107 { 1108 return con; 1109 } 1110 1111 static void con_put(struct ceph_connection *con) 1112 { 1113 } 1114 1115 static const struct ceph_connection_operations mon_con_ops = { 1116 .get = con_get, 1117 .put = con_put, 1118 .dispatch = dispatch, 1119 .fault = mon_fault, 1120 .alloc_msg = mon_alloc_msg, 1121 }; 1122