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 /* 302 * Wait for an osdmap with a given epoch. 303 * 304 * @epoch: epoch to wait for 305 * @timeout: in jiffies, 0 means "wait forever" 306 */ 307 int ceph_monc_wait_osdmap(struct ceph_mon_client *monc, u32 epoch, 308 unsigned long timeout) 309 { 310 unsigned long started = jiffies; 311 long ret; 312 313 mutex_lock(&monc->mutex); 314 while (monc->have_osdmap < epoch) { 315 mutex_unlock(&monc->mutex); 316 317 if (timeout && time_after_eq(jiffies, started + timeout)) 318 return -ETIMEDOUT; 319 320 ret = wait_event_interruptible_timeout(monc->client->auth_wq, 321 monc->have_osdmap >= epoch, 322 ceph_timeout_jiffies(timeout)); 323 if (ret < 0) 324 return ret; 325 326 mutex_lock(&monc->mutex); 327 } 328 329 mutex_unlock(&monc->mutex); 330 return 0; 331 } 332 EXPORT_SYMBOL(ceph_monc_wait_osdmap); 333 334 /* 335 * 336 */ 337 int ceph_monc_open_session(struct ceph_mon_client *monc) 338 { 339 mutex_lock(&monc->mutex); 340 __open_session(monc); 341 __schedule_delayed(monc); 342 mutex_unlock(&monc->mutex); 343 return 0; 344 } 345 EXPORT_SYMBOL(ceph_monc_open_session); 346 347 /* 348 * We require the fsid and global_id in order to initialize our 349 * debugfs dir. 350 */ 351 static bool have_debugfs_info(struct ceph_mon_client *monc) 352 { 353 dout("have_debugfs_info fsid %d globalid %lld\n", 354 (int)monc->client->have_fsid, monc->auth->global_id); 355 return monc->client->have_fsid && monc->auth->global_id > 0; 356 } 357 358 /* 359 * The monitor responds with mount ack indicate mount success. The 360 * included client ticket allows the client to talk to MDSs and OSDs. 361 */ 362 static void ceph_monc_handle_map(struct ceph_mon_client *monc, 363 struct ceph_msg *msg) 364 { 365 struct ceph_client *client = monc->client; 366 struct ceph_monmap *monmap = NULL, *old = monc->monmap; 367 void *p, *end; 368 int had_debugfs_info, init_debugfs = 0; 369 370 mutex_lock(&monc->mutex); 371 372 had_debugfs_info = have_debugfs_info(monc); 373 374 dout("handle_monmap\n"); 375 p = msg->front.iov_base; 376 end = p + msg->front.iov_len; 377 378 monmap = ceph_monmap_decode(p, end); 379 if (IS_ERR(monmap)) { 380 pr_err("problem decoding monmap, %d\n", 381 (int)PTR_ERR(monmap)); 382 goto out; 383 } 384 385 if (ceph_check_fsid(monc->client, &monmap->fsid) < 0) { 386 kfree(monmap); 387 goto out; 388 } 389 390 client->monc.monmap = monmap; 391 kfree(old); 392 393 if (!client->have_fsid) { 394 client->have_fsid = true; 395 if (!had_debugfs_info && have_debugfs_info(monc)) { 396 pr_info("client%lld fsid %pU\n", 397 ceph_client_id(monc->client), 398 &monc->client->fsid); 399 init_debugfs = 1; 400 } 401 mutex_unlock(&monc->mutex); 402 403 if (init_debugfs) { 404 /* 405 * do debugfs initialization without mutex to avoid 406 * creating a locking dependency 407 */ 408 ceph_debugfs_client_init(monc->client); 409 } 410 411 goto out_unlocked; 412 } 413 out: 414 mutex_unlock(&monc->mutex); 415 out_unlocked: 416 wake_up_all(&client->auth_wq); 417 } 418 419 /* 420 * generic requests (currently statfs, mon_get_version) 421 */ 422 static struct ceph_mon_generic_request *__lookup_generic_req( 423 struct ceph_mon_client *monc, u64 tid) 424 { 425 struct ceph_mon_generic_request *req; 426 struct rb_node *n = monc->generic_request_tree.rb_node; 427 428 while (n) { 429 req = rb_entry(n, struct ceph_mon_generic_request, node); 430 if (tid < req->tid) 431 n = n->rb_left; 432 else if (tid > req->tid) 433 n = n->rb_right; 434 else 435 return req; 436 } 437 return NULL; 438 } 439 440 static void __insert_generic_request(struct ceph_mon_client *monc, 441 struct ceph_mon_generic_request *new) 442 { 443 struct rb_node **p = &monc->generic_request_tree.rb_node; 444 struct rb_node *parent = NULL; 445 struct ceph_mon_generic_request *req = NULL; 446 447 while (*p) { 448 parent = *p; 449 req = rb_entry(parent, struct ceph_mon_generic_request, node); 450 if (new->tid < req->tid) 451 p = &(*p)->rb_left; 452 else if (new->tid > req->tid) 453 p = &(*p)->rb_right; 454 else 455 BUG(); 456 } 457 458 rb_link_node(&new->node, parent, p); 459 rb_insert_color(&new->node, &monc->generic_request_tree); 460 } 461 462 static void release_generic_request(struct kref *kref) 463 { 464 struct ceph_mon_generic_request *req = 465 container_of(kref, struct ceph_mon_generic_request, kref); 466 467 if (req->reply) 468 ceph_msg_put(req->reply); 469 if (req->request) 470 ceph_msg_put(req->request); 471 472 kfree(req); 473 } 474 475 static void put_generic_request(struct ceph_mon_generic_request *req) 476 { 477 kref_put(&req->kref, release_generic_request); 478 } 479 480 static void get_generic_request(struct ceph_mon_generic_request *req) 481 { 482 kref_get(&req->kref); 483 } 484 485 static struct ceph_msg *get_generic_reply(struct ceph_connection *con, 486 struct ceph_msg_header *hdr, 487 int *skip) 488 { 489 struct ceph_mon_client *monc = con->private; 490 struct ceph_mon_generic_request *req; 491 u64 tid = le64_to_cpu(hdr->tid); 492 struct ceph_msg *m; 493 494 mutex_lock(&monc->mutex); 495 req = __lookup_generic_req(monc, tid); 496 if (!req) { 497 dout("get_generic_reply %lld dne\n", tid); 498 *skip = 1; 499 m = NULL; 500 } else { 501 dout("get_generic_reply %lld got %p\n", tid, req->reply); 502 *skip = 0; 503 m = ceph_msg_get(req->reply); 504 /* 505 * we don't need to track the connection reading into 506 * this reply because we only have one open connection 507 * at a time, ever. 508 */ 509 } 510 mutex_unlock(&monc->mutex); 511 return m; 512 } 513 514 static int __do_generic_request(struct ceph_mon_client *monc, u64 tid, 515 struct ceph_mon_generic_request *req) 516 { 517 int err; 518 519 /* register request */ 520 req->tid = tid != 0 ? tid : ++monc->last_tid; 521 req->request->hdr.tid = cpu_to_le64(req->tid); 522 __insert_generic_request(monc, req); 523 monc->num_generic_requests++; 524 ceph_con_send(&monc->con, ceph_msg_get(req->request)); 525 mutex_unlock(&monc->mutex); 526 527 err = wait_for_completion_interruptible(&req->completion); 528 529 mutex_lock(&monc->mutex); 530 rb_erase(&req->node, &monc->generic_request_tree); 531 monc->num_generic_requests--; 532 533 if (!err) 534 err = req->result; 535 return err; 536 } 537 538 static int do_generic_request(struct ceph_mon_client *monc, 539 struct ceph_mon_generic_request *req) 540 { 541 int err; 542 543 mutex_lock(&monc->mutex); 544 err = __do_generic_request(monc, 0, req); 545 mutex_unlock(&monc->mutex); 546 547 return err; 548 } 549 550 /* 551 * statfs 552 */ 553 static void handle_statfs_reply(struct ceph_mon_client *monc, 554 struct ceph_msg *msg) 555 { 556 struct ceph_mon_generic_request *req; 557 struct ceph_mon_statfs_reply *reply = msg->front.iov_base; 558 u64 tid = le64_to_cpu(msg->hdr.tid); 559 560 if (msg->front.iov_len != sizeof(*reply)) 561 goto bad; 562 dout("handle_statfs_reply %p tid %llu\n", msg, tid); 563 564 mutex_lock(&monc->mutex); 565 req = __lookup_generic_req(monc, tid); 566 if (req) { 567 *(struct ceph_statfs *)req->buf = reply->st; 568 req->result = 0; 569 get_generic_request(req); 570 } 571 mutex_unlock(&monc->mutex); 572 if (req) { 573 complete_all(&req->completion); 574 put_generic_request(req); 575 } 576 return; 577 578 bad: 579 pr_err("corrupt statfs reply, tid %llu\n", tid); 580 ceph_msg_dump(msg); 581 } 582 583 /* 584 * Do a synchronous statfs(). 585 */ 586 int ceph_monc_do_statfs(struct ceph_mon_client *monc, struct ceph_statfs *buf) 587 { 588 struct ceph_mon_generic_request *req; 589 struct ceph_mon_statfs *h; 590 int err; 591 592 req = kzalloc(sizeof(*req), GFP_NOFS); 593 if (!req) 594 return -ENOMEM; 595 596 kref_init(&req->kref); 597 req->buf = buf; 598 init_completion(&req->completion); 599 600 err = -ENOMEM; 601 req->request = ceph_msg_new(CEPH_MSG_STATFS, sizeof(*h), GFP_NOFS, 602 true); 603 if (!req->request) 604 goto out; 605 req->reply = ceph_msg_new(CEPH_MSG_STATFS_REPLY, 1024, GFP_NOFS, 606 true); 607 if (!req->reply) 608 goto out; 609 610 /* fill out request */ 611 h = req->request->front.iov_base; 612 h->monhdr.have_version = 0; 613 h->monhdr.session_mon = cpu_to_le16(-1); 614 h->monhdr.session_mon_tid = 0; 615 h->fsid = monc->monmap->fsid; 616 617 err = do_generic_request(monc, req); 618 619 out: 620 put_generic_request(req); 621 return err; 622 } 623 EXPORT_SYMBOL(ceph_monc_do_statfs); 624 625 static void handle_get_version_reply(struct ceph_mon_client *monc, 626 struct ceph_msg *msg) 627 { 628 struct ceph_mon_generic_request *req; 629 u64 tid = le64_to_cpu(msg->hdr.tid); 630 void *p = msg->front.iov_base; 631 void *end = p + msg->front_alloc_len; 632 u64 handle; 633 634 dout("%s %p tid %llu\n", __func__, msg, tid); 635 636 ceph_decode_need(&p, end, 2*sizeof(u64), bad); 637 handle = ceph_decode_64(&p); 638 if (tid != 0 && tid != handle) 639 goto bad; 640 641 mutex_lock(&monc->mutex); 642 req = __lookup_generic_req(monc, handle); 643 if (req) { 644 *(u64 *)req->buf = ceph_decode_64(&p); 645 req->result = 0; 646 get_generic_request(req); 647 } 648 mutex_unlock(&monc->mutex); 649 if (req) { 650 complete_all(&req->completion); 651 put_generic_request(req); 652 } 653 654 return; 655 bad: 656 pr_err("corrupt mon_get_version reply, tid %llu\n", tid); 657 ceph_msg_dump(msg); 658 } 659 660 /* 661 * Send MMonGetVersion and wait for the reply. 662 * 663 * @what: one of "mdsmap", "osdmap" or "monmap" 664 */ 665 int ceph_monc_do_get_version(struct ceph_mon_client *monc, const char *what, 666 u64 *newest) 667 { 668 struct ceph_mon_generic_request *req; 669 void *p, *end; 670 u64 tid; 671 int err; 672 673 req = kzalloc(sizeof(*req), GFP_NOFS); 674 if (!req) 675 return -ENOMEM; 676 677 kref_init(&req->kref); 678 req->buf = newest; 679 init_completion(&req->completion); 680 681 req->request = ceph_msg_new(CEPH_MSG_MON_GET_VERSION, 682 sizeof(u64) + sizeof(u32) + strlen(what), 683 GFP_NOFS, true); 684 if (!req->request) { 685 err = -ENOMEM; 686 goto out; 687 } 688 689 req->reply = ceph_msg_new(CEPH_MSG_MON_GET_VERSION_REPLY, 1024, 690 GFP_NOFS, true); 691 if (!req->reply) { 692 err = -ENOMEM; 693 goto out; 694 } 695 696 p = req->request->front.iov_base; 697 end = p + req->request->front_alloc_len; 698 699 /* fill out request */ 700 mutex_lock(&monc->mutex); 701 tid = ++monc->last_tid; 702 ceph_encode_64(&p, tid); /* handle */ 703 ceph_encode_string(&p, end, what, strlen(what)); 704 705 err = __do_generic_request(monc, tid, req); 706 707 mutex_unlock(&monc->mutex); 708 out: 709 put_generic_request(req); 710 return err; 711 } 712 EXPORT_SYMBOL(ceph_monc_do_get_version); 713 714 /* 715 * Resend pending generic requests. 716 */ 717 static void __resend_generic_request(struct ceph_mon_client *monc) 718 { 719 struct ceph_mon_generic_request *req; 720 struct rb_node *p; 721 722 for (p = rb_first(&monc->generic_request_tree); p; p = rb_next(p)) { 723 req = rb_entry(p, struct ceph_mon_generic_request, node); 724 ceph_msg_revoke(req->request); 725 ceph_msg_revoke_incoming(req->reply); 726 ceph_con_send(&monc->con, ceph_msg_get(req->request)); 727 } 728 } 729 730 /* 731 * Delayed work. If we haven't mounted yet, retry. Otherwise, 732 * renew/retry subscription as needed (in case it is timing out, or we 733 * got an ENOMEM). And keep the monitor connection alive. 734 */ 735 static void delayed_work(struct work_struct *work) 736 { 737 struct ceph_mon_client *monc = 738 container_of(work, struct ceph_mon_client, delayed_work.work); 739 740 dout("monc delayed_work\n"); 741 mutex_lock(&monc->mutex); 742 if (monc->hunting) { 743 __close_session(monc); 744 __open_session(monc); /* continue hunting */ 745 } else { 746 ceph_con_keepalive(&monc->con); 747 748 __validate_auth(monc); 749 750 if (ceph_auth_is_authenticated(monc->auth)) 751 __send_subscribe(monc); 752 } 753 __schedule_delayed(monc); 754 mutex_unlock(&monc->mutex); 755 } 756 757 /* 758 * On startup, we build a temporary monmap populated with the IPs 759 * provided by mount(2). 760 */ 761 static int build_initial_monmap(struct ceph_mon_client *monc) 762 { 763 struct ceph_options *opt = monc->client->options; 764 struct ceph_entity_addr *mon_addr = opt->mon_addr; 765 int num_mon = opt->num_mon; 766 int i; 767 768 /* build initial monmap */ 769 monc->monmap = kzalloc(sizeof(*monc->monmap) + 770 num_mon*sizeof(monc->monmap->mon_inst[0]), 771 GFP_KERNEL); 772 if (!monc->monmap) 773 return -ENOMEM; 774 for (i = 0; i < num_mon; i++) { 775 monc->monmap->mon_inst[i].addr = mon_addr[i]; 776 monc->monmap->mon_inst[i].addr.nonce = 0; 777 monc->monmap->mon_inst[i].name.type = 778 CEPH_ENTITY_TYPE_MON; 779 monc->monmap->mon_inst[i].name.num = cpu_to_le64(i); 780 } 781 monc->monmap->num_mon = num_mon; 782 return 0; 783 } 784 785 int ceph_monc_init(struct ceph_mon_client *monc, struct ceph_client *cl) 786 { 787 int err = 0; 788 789 dout("init\n"); 790 memset(monc, 0, sizeof(*monc)); 791 monc->client = cl; 792 monc->monmap = NULL; 793 mutex_init(&monc->mutex); 794 795 err = build_initial_monmap(monc); 796 if (err) 797 goto out; 798 799 /* connection */ 800 /* authentication */ 801 monc->auth = ceph_auth_init(cl->options->name, 802 cl->options->key); 803 if (IS_ERR(monc->auth)) { 804 err = PTR_ERR(monc->auth); 805 goto out_monmap; 806 } 807 monc->auth->want_keys = 808 CEPH_ENTITY_TYPE_AUTH | CEPH_ENTITY_TYPE_MON | 809 CEPH_ENTITY_TYPE_OSD | CEPH_ENTITY_TYPE_MDS; 810 811 /* msgs */ 812 err = -ENOMEM; 813 monc->m_subscribe_ack = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE_ACK, 814 sizeof(struct ceph_mon_subscribe_ack), 815 GFP_NOFS, true); 816 if (!monc->m_subscribe_ack) 817 goto out_auth; 818 819 monc->m_subscribe = ceph_msg_new(CEPH_MSG_MON_SUBSCRIBE, 96, GFP_NOFS, 820 true); 821 if (!monc->m_subscribe) 822 goto out_subscribe_ack; 823 824 monc->m_auth_reply = ceph_msg_new(CEPH_MSG_AUTH_REPLY, 4096, GFP_NOFS, 825 true); 826 if (!monc->m_auth_reply) 827 goto out_subscribe; 828 829 monc->m_auth = ceph_msg_new(CEPH_MSG_AUTH, 4096, GFP_NOFS, true); 830 monc->pending_auth = 0; 831 if (!monc->m_auth) 832 goto out_auth_reply; 833 834 ceph_con_init(&monc->con, monc, &mon_con_ops, 835 &monc->client->msgr); 836 837 monc->cur_mon = -1; 838 monc->hunting = true; 839 monc->sub_renew_after = jiffies; 840 monc->sub_sent = 0; 841 842 INIT_DELAYED_WORK(&monc->delayed_work, delayed_work); 843 monc->generic_request_tree = RB_ROOT; 844 monc->num_generic_requests = 0; 845 monc->last_tid = 0; 846 847 monc->have_mdsmap = 0; 848 monc->have_osdmap = 0; 849 monc->want_next_osdmap = 1; 850 return 0; 851 852 out_auth_reply: 853 ceph_msg_put(monc->m_auth_reply); 854 out_subscribe: 855 ceph_msg_put(monc->m_subscribe); 856 out_subscribe_ack: 857 ceph_msg_put(monc->m_subscribe_ack); 858 out_auth: 859 ceph_auth_destroy(monc->auth); 860 out_monmap: 861 kfree(monc->monmap); 862 out: 863 return err; 864 } 865 EXPORT_SYMBOL(ceph_monc_init); 866 867 void ceph_monc_stop(struct ceph_mon_client *monc) 868 { 869 dout("stop\n"); 870 cancel_delayed_work_sync(&monc->delayed_work); 871 872 mutex_lock(&monc->mutex); 873 __close_session(monc); 874 875 mutex_unlock(&monc->mutex); 876 877 /* 878 * flush msgr queue before we destroy ourselves to ensure that: 879 * - any work that references our embedded con is finished. 880 * - any osd_client or other work that may reference an authorizer 881 * finishes before we shut down the auth subsystem. 882 */ 883 ceph_msgr_flush(); 884 885 ceph_auth_destroy(monc->auth); 886 887 ceph_msg_put(monc->m_auth); 888 ceph_msg_put(monc->m_auth_reply); 889 ceph_msg_put(monc->m_subscribe); 890 ceph_msg_put(monc->m_subscribe_ack); 891 892 kfree(monc->monmap); 893 } 894 EXPORT_SYMBOL(ceph_monc_stop); 895 896 static void handle_auth_reply(struct ceph_mon_client *monc, 897 struct ceph_msg *msg) 898 { 899 int ret; 900 int was_auth = 0; 901 int had_debugfs_info, init_debugfs = 0; 902 903 mutex_lock(&monc->mutex); 904 had_debugfs_info = have_debugfs_info(monc); 905 was_auth = ceph_auth_is_authenticated(monc->auth); 906 monc->pending_auth = 0; 907 ret = ceph_handle_auth_reply(monc->auth, msg->front.iov_base, 908 msg->front.iov_len, 909 monc->m_auth->front.iov_base, 910 monc->m_auth->front_alloc_len); 911 if (ret < 0) { 912 monc->client->auth_err = ret; 913 wake_up_all(&monc->client->auth_wq); 914 } else if (ret > 0) { 915 __send_prepared_auth_request(monc, ret); 916 } else if (!was_auth && ceph_auth_is_authenticated(monc->auth)) { 917 dout("authenticated, starting session\n"); 918 919 monc->client->msgr.inst.name.type = CEPH_ENTITY_TYPE_CLIENT; 920 monc->client->msgr.inst.name.num = 921 cpu_to_le64(monc->auth->global_id); 922 923 __send_subscribe(monc); 924 __resend_generic_request(monc); 925 } 926 927 if (!had_debugfs_info && have_debugfs_info(monc)) { 928 pr_info("client%lld fsid %pU\n", 929 ceph_client_id(monc->client), 930 &monc->client->fsid); 931 init_debugfs = 1; 932 } 933 mutex_unlock(&monc->mutex); 934 935 if (init_debugfs) { 936 /* 937 * do debugfs initialization without mutex to avoid 938 * creating a locking dependency 939 */ 940 ceph_debugfs_client_init(monc->client); 941 } 942 } 943 944 static int __validate_auth(struct ceph_mon_client *monc) 945 { 946 int ret; 947 948 if (monc->pending_auth) 949 return 0; 950 951 ret = ceph_build_auth(monc->auth, monc->m_auth->front.iov_base, 952 monc->m_auth->front_alloc_len); 953 if (ret <= 0) 954 return ret; /* either an error, or no need to authenticate */ 955 __send_prepared_auth_request(monc, ret); 956 return 0; 957 } 958 959 int ceph_monc_validate_auth(struct ceph_mon_client *monc) 960 { 961 int ret; 962 963 mutex_lock(&monc->mutex); 964 ret = __validate_auth(monc); 965 mutex_unlock(&monc->mutex); 966 return ret; 967 } 968 EXPORT_SYMBOL(ceph_monc_validate_auth); 969 970 /* 971 * handle incoming message 972 */ 973 static void dispatch(struct ceph_connection *con, struct ceph_msg *msg) 974 { 975 struct ceph_mon_client *monc = con->private; 976 int type = le16_to_cpu(msg->hdr.type); 977 978 if (!monc) 979 return; 980 981 switch (type) { 982 case CEPH_MSG_AUTH_REPLY: 983 handle_auth_reply(monc, msg); 984 break; 985 986 case CEPH_MSG_MON_SUBSCRIBE_ACK: 987 handle_subscribe_ack(monc, msg); 988 break; 989 990 case CEPH_MSG_STATFS_REPLY: 991 handle_statfs_reply(monc, msg); 992 break; 993 994 case CEPH_MSG_MON_GET_VERSION_REPLY: 995 handle_get_version_reply(monc, msg); 996 break; 997 998 case CEPH_MSG_MON_MAP: 999 ceph_monc_handle_map(monc, msg); 1000 break; 1001 1002 case CEPH_MSG_OSD_MAP: 1003 ceph_osdc_handle_map(&monc->client->osdc, msg); 1004 break; 1005 1006 default: 1007 /* can the chained handler handle it? */ 1008 if (monc->client->extra_mon_dispatch && 1009 monc->client->extra_mon_dispatch(monc->client, msg) == 0) 1010 break; 1011 1012 pr_err("received unknown message type %d %s\n", type, 1013 ceph_msg_type_name(type)); 1014 } 1015 ceph_msg_put(msg); 1016 } 1017 1018 /* 1019 * Allocate memory for incoming message 1020 */ 1021 static struct ceph_msg *mon_alloc_msg(struct ceph_connection *con, 1022 struct ceph_msg_header *hdr, 1023 int *skip) 1024 { 1025 struct ceph_mon_client *monc = con->private; 1026 int type = le16_to_cpu(hdr->type); 1027 int front_len = le32_to_cpu(hdr->front_len); 1028 struct ceph_msg *m = NULL; 1029 1030 *skip = 0; 1031 1032 switch (type) { 1033 case CEPH_MSG_MON_SUBSCRIBE_ACK: 1034 m = ceph_msg_get(monc->m_subscribe_ack); 1035 break; 1036 case CEPH_MSG_STATFS_REPLY: 1037 return get_generic_reply(con, hdr, skip); 1038 case CEPH_MSG_AUTH_REPLY: 1039 m = ceph_msg_get(monc->m_auth_reply); 1040 break; 1041 case CEPH_MSG_MON_GET_VERSION_REPLY: 1042 if (le64_to_cpu(hdr->tid) != 0) 1043 return get_generic_reply(con, hdr, skip); 1044 1045 /* 1046 * Older OSDs don't set reply tid even if the orignal 1047 * request had a non-zero tid. Workaround this weirdness 1048 * by falling through to the allocate case. 1049 */ 1050 case CEPH_MSG_MON_MAP: 1051 case CEPH_MSG_MDS_MAP: 1052 case CEPH_MSG_OSD_MAP: 1053 m = ceph_msg_new(type, front_len, GFP_NOFS, false); 1054 if (!m) 1055 return NULL; /* ENOMEM--return skip == 0 */ 1056 break; 1057 } 1058 1059 if (!m) { 1060 pr_info("alloc_msg unknown type %d\n", type); 1061 *skip = 1; 1062 } else if (front_len > m->front_alloc_len) { 1063 pr_warn("mon_alloc_msg front %d > prealloc %d (%u#%llu)\n", 1064 front_len, m->front_alloc_len, 1065 (unsigned int)con->peer_name.type, 1066 le64_to_cpu(con->peer_name.num)); 1067 ceph_msg_put(m); 1068 m = ceph_msg_new(type, front_len, GFP_NOFS, false); 1069 } 1070 1071 return m; 1072 } 1073 1074 /* 1075 * If the monitor connection resets, pick a new monitor and resubmit 1076 * any pending requests. 1077 */ 1078 static void mon_fault(struct ceph_connection *con) 1079 { 1080 struct ceph_mon_client *monc = con->private; 1081 1082 if (!monc) 1083 return; 1084 1085 dout("mon_fault\n"); 1086 mutex_lock(&monc->mutex); 1087 if (!con->private) 1088 goto out; 1089 1090 if (!monc->hunting) 1091 pr_info("mon%d %s session lost, " 1092 "hunting for new mon\n", monc->cur_mon, 1093 ceph_pr_addr(&monc->con.peer_addr.in_addr)); 1094 1095 __close_session(monc); 1096 if (!monc->hunting) { 1097 /* start hunting */ 1098 monc->hunting = true; 1099 __open_session(monc); 1100 } else { 1101 /* already hunting, let's wait a bit */ 1102 __schedule_delayed(monc); 1103 } 1104 out: 1105 mutex_unlock(&monc->mutex); 1106 } 1107 1108 /* 1109 * We can ignore refcounting on the connection struct, as all references 1110 * will come from the messenger workqueue, which is drained prior to 1111 * mon_client destruction. 1112 */ 1113 static struct ceph_connection *con_get(struct ceph_connection *con) 1114 { 1115 return con; 1116 } 1117 1118 static void con_put(struct ceph_connection *con) 1119 { 1120 } 1121 1122 static const struct ceph_connection_operations mon_con_ops = { 1123 .get = con_get, 1124 .put = con_put, 1125 .dispatch = dispatch, 1126 .fault = mon_fault, 1127 .alloc_msg = mon_alloc_msg, 1128 }; 1129