1 // SPDX-License-Identifier: GPL-2.0-only 2 /****************************************************************************** 3 ******************************************************************************* 4 ** 5 ** Copyright (C) 2005-2011 Red Hat, Inc. All rights reserved. 6 ** 7 ** 8 ******************************************************************************* 9 ******************************************************************************/ 10 11 #include "dlm_internal.h" 12 #include "lockspace.h" 13 #include "member.h" 14 #include "recoverd.h" 15 #include "recover.h" 16 #include "rcom.h" 17 #include "config.h" 18 #include "midcomms.h" 19 #include "lowcomms.h" 20 21 int dlm_slots_version(const struct dlm_header *h) 22 { 23 if ((le32_to_cpu(h->h_version) & 0x0000FFFF) < DLM_HEADER_SLOTS) 24 return 0; 25 return 1; 26 } 27 28 void dlm_slot_save(struct dlm_ls *ls, struct dlm_rcom *rc, 29 struct dlm_member *memb) 30 { 31 struct rcom_config *rf = (struct rcom_config *)rc->rc_buf; 32 33 if (!dlm_slots_version(&rc->rc_header)) 34 return; 35 36 memb->slot = le16_to_cpu(rf->rf_our_slot); 37 memb->generation = le32_to_cpu(rf->rf_generation); 38 } 39 40 void dlm_slots_copy_out(struct dlm_ls *ls, struct dlm_rcom *rc) 41 { 42 struct dlm_slot *slot; 43 struct rcom_slot *ro; 44 int i; 45 46 ro = (struct rcom_slot *)(rc->rc_buf + sizeof(struct rcom_config)); 47 48 /* ls_slots array is sparse, but not rcom_slots */ 49 50 for (i = 0; i < ls->ls_slots_size; i++) { 51 slot = &ls->ls_slots[i]; 52 if (!slot->nodeid) 53 continue; 54 ro->ro_nodeid = cpu_to_le32(slot->nodeid); 55 ro->ro_slot = cpu_to_le16(slot->slot); 56 ro++; 57 } 58 } 59 60 #define SLOT_DEBUG_LINE 128 61 62 static void log_slots(struct dlm_ls *ls, uint32_t gen, int num_slots, 63 struct rcom_slot *ro0, struct dlm_slot *array, 64 int array_size) 65 { 66 char line[SLOT_DEBUG_LINE]; 67 int len = SLOT_DEBUG_LINE - 1; 68 int pos = 0; 69 int ret, i; 70 71 memset(line, 0, sizeof(line)); 72 73 if (array) { 74 for (i = 0; i < array_size; i++) { 75 if (!array[i].nodeid) 76 continue; 77 78 ret = snprintf(line + pos, len - pos, " %d:%d", 79 array[i].slot, array[i].nodeid); 80 if (ret >= len - pos) 81 break; 82 pos += ret; 83 } 84 } else if (ro0) { 85 for (i = 0; i < num_slots; i++) { 86 ret = snprintf(line + pos, len - pos, " %d:%d", 87 ro0[i].ro_slot, ro0[i].ro_nodeid); 88 if (ret >= len - pos) 89 break; 90 pos += ret; 91 } 92 } 93 94 log_rinfo(ls, "generation %u slots %d%s", gen, num_slots, line); 95 } 96 97 int dlm_slots_copy_in(struct dlm_ls *ls) 98 { 99 struct dlm_member *memb; 100 struct dlm_rcom *rc = ls->ls_recover_buf; 101 struct rcom_config *rf = (struct rcom_config *)rc->rc_buf; 102 struct rcom_slot *ro0, *ro; 103 int our_nodeid = dlm_our_nodeid(); 104 int i, num_slots; 105 uint32_t gen; 106 107 if (!dlm_slots_version(&rc->rc_header)) 108 return -1; 109 110 gen = le32_to_cpu(rf->rf_generation); 111 if (gen <= ls->ls_generation) { 112 log_error(ls, "dlm_slots_copy_in gen %u old %u", 113 gen, ls->ls_generation); 114 } 115 ls->ls_generation = gen; 116 117 num_slots = le16_to_cpu(rf->rf_num_slots); 118 if (!num_slots) 119 return -1; 120 121 ro0 = (struct rcom_slot *)(rc->rc_buf + sizeof(struct rcom_config)); 122 123 log_slots(ls, gen, num_slots, ro0, NULL, 0); 124 125 list_for_each_entry(memb, &ls->ls_nodes, list) { 126 for (i = 0, ro = ro0; i < num_slots; i++, ro++) { 127 if (le32_to_cpu(ro->ro_nodeid) != memb->nodeid) 128 continue; 129 memb->slot = le16_to_cpu(ro->ro_slot); 130 memb->slot_prev = memb->slot; 131 break; 132 } 133 134 if (memb->nodeid == our_nodeid) { 135 if (ls->ls_slot && ls->ls_slot != memb->slot) { 136 log_error(ls, "dlm_slots_copy_in our slot " 137 "changed %d %d", ls->ls_slot, 138 memb->slot); 139 return -1; 140 } 141 142 if (!ls->ls_slot) 143 ls->ls_slot = memb->slot; 144 } 145 146 if (!memb->slot) { 147 log_error(ls, "dlm_slots_copy_in nodeid %d no slot", 148 memb->nodeid); 149 return -1; 150 } 151 } 152 153 return 0; 154 } 155 156 /* for any nodes that do not support slots, we will not have set memb->slot 157 in wait_status_all(), so memb->slot will remain -1, and we will not 158 assign slots or set ls_num_slots here */ 159 160 int dlm_slots_assign(struct dlm_ls *ls, int *num_slots, int *slots_size, 161 struct dlm_slot **slots_out, uint32_t *gen_out) 162 { 163 struct dlm_member *memb; 164 struct dlm_slot *array; 165 int our_nodeid = dlm_our_nodeid(); 166 int array_size, max_slots, i; 167 int need = 0; 168 int max = 0; 169 int num = 0; 170 uint32_t gen = 0; 171 172 /* our own memb struct will have slot -1 gen 0 */ 173 174 list_for_each_entry(memb, &ls->ls_nodes, list) { 175 if (memb->nodeid == our_nodeid) { 176 memb->slot = ls->ls_slot; 177 memb->generation = ls->ls_generation; 178 break; 179 } 180 } 181 182 list_for_each_entry(memb, &ls->ls_nodes, list) { 183 if (memb->generation > gen) 184 gen = memb->generation; 185 186 /* node doesn't support slots */ 187 188 if (memb->slot == -1) 189 return -1; 190 191 /* node needs a slot assigned */ 192 193 if (!memb->slot) 194 need++; 195 196 /* node has a slot assigned */ 197 198 num++; 199 200 if (!max || max < memb->slot) 201 max = memb->slot; 202 203 /* sanity check, once slot is assigned it shouldn't change */ 204 205 if (memb->slot_prev && memb->slot && memb->slot_prev != memb->slot) { 206 log_error(ls, "nodeid %d slot changed %d %d", 207 memb->nodeid, memb->slot_prev, memb->slot); 208 return -1; 209 } 210 memb->slot_prev = memb->slot; 211 } 212 213 array_size = max + need; 214 array = kcalloc(array_size, sizeof(*array), GFP_NOFS); 215 if (!array) 216 return -ENOMEM; 217 218 num = 0; 219 220 /* fill in slots (offsets) that are used */ 221 222 list_for_each_entry(memb, &ls->ls_nodes, list) { 223 if (!memb->slot) 224 continue; 225 226 if (memb->slot > array_size) { 227 log_error(ls, "invalid slot number %d", memb->slot); 228 kfree(array); 229 return -1; 230 } 231 232 array[memb->slot - 1].nodeid = memb->nodeid; 233 array[memb->slot - 1].slot = memb->slot; 234 num++; 235 } 236 237 /* assign new slots from unused offsets */ 238 239 list_for_each_entry(memb, &ls->ls_nodes, list) { 240 if (memb->slot) 241 continue; 242 243 for (i = 0; i < array_size; i++) { 244 if (array[i].nodeid) 245 continue; 246 247 memb->slot = i + 1; 248 memb->slot_prev = memb->slot; 249 array[i].nodeid = memb->nodeid; 250 array[i].slot = memb->slot; 251 num++; 252 253 if (!ls->ls_slot && memb->nodeid == our_nodeid) 254 ls->ls_slot = memb->slot; 255 break; 256 } 257 258 if (!memb->slot) { 259 log_error(ls, "no free slot found"); 260 kfree(array); 261 return -1; 262 } 263 } 264 265 gen++; 266 267 log_slots(ls, gen, num, NULL, array, array_size); 268 269 max_slots = (DLM_MAX_APP_BUFSIZE - sizeof(struct dlm_rcom) - 270 sizeof(struct rcom_config)) / sizeof(struct rcom_slot); 271 272 if (num > max_slots) { 273 log_error(ls, "num_slots %d exceeds max_slots %d", 274 num, max_slots); 275 kfree(array); 276 return -1; 277 } 278 279 *gen_out = gen; 280 *slots_out = array; 281 *slots_size = array_size; 282 *num_slots = num; 283 return 0; 284 } 285 286 static void add_ordered_member(struct dlm_ls *ls, struct dlm_member *new) 287 { 288 struct dlm_member *memb = NULL; 289 struct list_head *tmp; 290 struct list_head *newlist = &new->list; 291 struct list_head *head = &ls->ls_nodes; 292 293 list_for_each(tmp, head) { 294 memb = list_entry(tmp, struct dlm_member, list); 295 if (new->nodeid < memb->nodeid) 296 break; 297 } 298 299 if (!memb) 300 list_add_tail(newlist, head); 301 else { 302 /* FIXME: can use list macro here */ 303 newlist->prev = tmp->prev; 304 newlist->next = tmp; 305 tmp->prev->next = newlist; 306 tmp->prev = newlist; 307 } 308 } 309 310 static int add_remote_member(int nodeid) 311 { 312 int error; 313 314 if (nodeid == dlm_our_nodeid()) 315 return 0; 316 317 error = dlm_lowcomms_connect_node(nodeid); 318 if (error < 0) 319 return error; 320 321 dlm_midcomms_add_member(nodeid); 322 return 0; 323 } 324 325 static int dlm_add_member(struct dlm_ls *ls, struct dlm_config_node *node) 326 { 327 struct dlm_member *memb; 328 int error; 329 330 memb = kzalloc(sizeof(*memb), GFP_NOFS); 331 if (!memb) 332 return -ENOMEM; 333 334 memb->nodeid = node->nodeid; 335 memb->weight = node->weight; 336 memb->comm_seq = node->comm_seq; 337 338 error = add_remote_member(node->nodeid); 339 if (error < 0) { 340 kfree(memb); 341 return error; 342 } 343 344 add_ordered_member(ls, memb); 345 ls->ls_num_nodes++; 346 return 0; 347 } 348 349 static struct dlm_member *find_memb(struct list_head *head, int nodeid) 350 { 351 struct dlm_member *memb; 352 353 list_for_each_entry(memb, head, list) { 354 if (memb->nodeid == nodeid) 355 return memb; 356 } 357 return NULL; 358 } 359 360 int dlm_is_member(struct dlm_ls *ls, int nodeid) 361 { 362 if (find_memb(&ls->ls_nodes, nodeid)) 363 return 1; 364 return 0; 365 } 366 367 int dlm_is_removed(struct dlm_ls *ls, int nodeid) 368 { 369 if (find_memb(&ls->ls_nodes_gone, nodeid)) 370 return 1; 371 return 0; 372 } 373 374 static void clear_memb_list(struct list_head *head, 375 void (*after_del)(int nodeid)) 376 { 377 struct dlm_member *memb; 378 379 while (!list_empty(head)) { 380 memb = list_entry(head->next, struct dlm_member, list); 381 list_del(&memb->list); 382 if (after_del) 383 after_del(memb->nodeid); 384 kfree(memb); 385 } 386 } 387 388 static void remove_remote_member(int nodeid) 389 { 390 if (nodeid == dlm_our_nodeid()) 391 return; 392 393 dlm_midcomms_remove_member(nodeid); 394 } 395 396 void dlm_clear_members(struct dlm_ls *ls) 397 { 398 clear_memb_list(&ls->ls_nodes, remove_remote_member); 399 ls->ls_num_nodes = 0; 400 } 401 402 void dlm_clear_members_gone(struct dlm_ls *ls) 403 { 404 clear_memb_list(&ls->ls_nodes_gone, NULL); 405 } 406 407 static void make_member_array(struct dlm_ls *ls) 408 { 409 struct dlm_member *memb; 410 int i, w, x = 0, total = 0, all_zero = 0, *array; 411 412 kfree(ls->ls_node_array); 413 ls->ls_node_array = NULL; 414 415 list_for_each_entry(memb, &ls->ls_nodes, list) { 416 if (memb->weight) 417 total += memb->weight; 418 } 419 420 /* all nodes revert to weight of 1 if all have weight 0 */ 421 422 if (!total) { 423 total = ls->ls_num_nodes; 424 all_zero = 1; 425 } 426 427 ls->ls_total_weight = total; 428 array = kmalloc_array(total, sizeof(*array), GFP_NOFS); 429 if (!array) 430 return; 431 432 list_for_each_entry(memb, &ls->ls_nodes, list) { 433 if (!all_zero && !memb->weight) 434 continue; 435 436 if (all_zero) 437 w = 1; 438 else 439 w = memb->weight; 440 441 DLM_ASSERT(x < total, printk("total %d x %d\n", total, x);); 442 443 for (i = 0; i < w; i++) 444 array[x++] = memb->nodeid; 445 } 446 447 ls->ls_node_array = array; 448 } 449 450 /* send a status request to all members just to establish comms connections */ 451 452 static int ping_members(struct dlm_ls *ls, uint64_t seq) 453 { 454 struct dlm_member *memb; 455 int error = 0; 456 457 list_for_each_entry(memb, &ls->ls_nodes, list) { 458 if (dlm_recovery_stopped(ls)) { 459 error = -EINTR; 460 break; 461 } 462 error = dlm_rcom_status(ls, memb->nodeid, 0, seq); 463 if (error) 464 break; 465 } 466 if (error) 467 log_rinfo(ls, "ping_members aborted %d last nodeid %d", 468 error, ls->ls_recover_nodeid); 469 return error; 470 } 471 472 static void dlm_lsop_recover_prep(struct dlm_ls *ls) 473 { 474 if (!ls->ls_ops || !ls->ls_ops->recover_prep) 475 return; 476 ls->ls_ops->recover_prep(ls->ls_ops_arg); 477 } 478 479 static void dlm_lsop_recover_slot(struct dlm_ls *ls, struct dlm_member *memb) 480 { 481 struct dlm_slot slot; 482 uint32_t seq; 483 int error; 484 485 if (!ls->ls_ops || !ls->ls_ops->recover_slot) 486 return; 487 488 /* if there is no comms connection with this node 489 or the present comms connection is newer 490 than the one when this member was added, then 491 we consider the node to have failed (versus 492 being removed due to dlm_release_lockspace) */ 493 494 error = dlm_comm_seq(memb->nodeid, &seq); 495 496 if (!error && seq == memb->comm_seq) 497 return; 498 499 slot.nodeid = memb->nodeid; 500 slot.slot = memb->slot; 501 502 ls->ls_ops->recover_slot(ls->ls_ops_arg, &slot); 503 } 504 505 void dlm_lsop_recover_done(struct dlm_ls *ls) 506 { 507 struct dlm_member *memb; 508 struct dlm_slot *slots; 509 int i, num; 510 511 if (!ls->ls_ops || !ls->ls_ops->recover_done) 512 return; 513 514 num = ls->ls_num_nodes; 515 slots = kcalloc(num, sizeof(*slots), GFP_KERNEL); 516 if (!slots) 517 return; 518 519 i = 0; 520 list_for_each_entry(memb, &ls->ls_nodes, list) { 521 if (i == num) { 522 log_error(ls, "dlm_lsop_recover_done bad num %d", num); 523 goto out; 524 } 525 slots[i].nodeid = memb->nodeid; 526 slots[i].slot = memb->slot; 527 i++; 528 } 529 530 ls->ls_ops->recover_done(ls->ls_ops_arg, slots, num, 531 ls->ls_slot, ls->ls_generation); 532 out: 533 kfree(slots); 534 } 535 536 static struct dlm_config_node *find_config_node(struct dlm_recover *rv, 537 int nodeid) 538 { 539 int i; 540 541 for (i = 0; i < rv->nodes_count; i++) { 542 if (rv->nodes[i].nodeid == nodeid) 543 return &rv->nodes[i]; 544 } 545 return NULL; 546 } 547 548 int dlm_recover_members(struct dlm_ls *ls, struct dlm_recover *rv, int *neg_out) 549 { 550 struct dlm_member *memb, *safe; 551 struct dlm_config_node *node; 552 int i, error, neg = 0, low = -1; 553 554 /* previously removed members that we've not finished removing need to 555 * count as a negative change so the "neg" recovery steps will happen 556 * 557 * This functionality must report all member changes to lsops or 558 * midcomms layer and must never return before. 559 */ 560 561 list_for_each_entry(memb, &ls->ls_nodes_gone, list) { 562 log_rinfo(ls, "prev removed member %d", memb->nodeid); 563 neg++; 564 } 565 566 /* move departed members from ls_nodes to ls_nodes_gone */ 567 568 list_for_each_entry_safe(memb, safe, &ls->ls_nodes, list) { 569 node = find_config_node(rv, memb->nodeid); 570 if (node && !node->new) 571 continue; 572 573 if (!node) { 574 log_rinfo(ls, "remove member %d", memb->nodeid); 575 } else { 576 /* removed and re-added */ 577 log_rinfo(ls, "remove member %d comm_seq %u %u", 578 memb->nodeid, memb->comm_seq, node->comm_seq); 579 } 580 581 neg++; 582 list_move(&memb->list, &ls->ls_nodes_gone); 583 remove_remote_member(memb->nodeid); 584 ls->ls_num_nodes--; 585 dlm_lsop_recover_slot(ls, memb); 586 } 587 588 /* add new members to ls_nodes */ 589 590 for (i = 0; i < rv->nodes_count; i++) { 591 node = &rv->nodes[i]; 592 if (dlm_is_member(ls, node->nodeid)) 593 continue; 594 error = dlm_add_member(ls, node); 595 if (error) 596 return error; 597 598 log_rinfo(ls, "add member %d", node->nodeid); 599 } 600 601 list_for_each_entry(memb, &ls->ls_nodes, list) { 602 if (low == -1 || memb->nodeid < low) 603 low = memb->nodeid; 604 } 605 ls->ls_low_nodeid = low; 606 607 make_member_array(ls); 608 *neg_out = neg; 609 610 error = ping_members(ls, rv->seq); 611 log_rinfo(ls, "dlm_recover_members %d nodes", ls->ls_num_nodes); 612 return error; 613 } 614 615 /* Userspace guarantees that dlm_ls_stop() has completed on all nodes before 616 dlm_ls_start() is called on any of them to start the new recovery. */ 617 618 int dlm_ls_stop(struct dlm_ls *ls) 619 { 620 int new; 621 622 /* 623 * Prevent dlm_recv from being in the middle of something when we do 624 * the stop. This includes ensuring dlm_recv isn't processing a 625 * recovery message (rcom), while dlm_recoverd is aborting and 626 * resetting things from an in-progress recovery. i.e. we want 627 * dlm_recoverd to abort its recovery without worrying about dlm_recv 628 * processing an rcom at the same time. Stopping dlm_recv also makes 629 * it easy for dlm_receive_message() to check locking stopped and add a 630 * message to the requestqueue without races. 631 */ 632 633 write_lock_bh(&ls->ls_recv_active); 634 635 /* 636 * Abort any recovery that's in progress (see RECOVER_STOP, 637 * dlm_recovery_stopped()) and tell any other threads running in the 638 * dlm to quit any processing (see RUNNING, dlm_locking_stopped()). 639 */ 640 641 spin_lock_bh(&ls->ls_recover_lock); 642 set_bit(LSFL_RECOVER_STOP, &ls->ls_flags); 643 new = test_and_clear_bit(LSFL_RUNNING, &ls->ls_flags); 644 if (new) 645 timer_delete_sync(&ls->ls_timer); 646 ls->ls_recover_seq++; 647 648 /* activate requestqueue and stop processing */ 649 write_lock_bh(&ls->ls_requestqueue_lock); 650 set_bit(LSFL_RECV_MSG_BLOCKED, &ls->ls_flags); 651 write_unlock_bh(&ls->ls_requestqueue_lock); 652 spin_unlock_bh(&ls->ls_recover_lock); 653 654 /* 655 * Let dlm_recv run again, now any normal messages will be saved on the 656 * requestqueue for later. 657 */ 658 659 write_unlock_bh(&ls->ls_recv_active); 660 661 /* 662 * This in_recovery lock does two things: 663 * 1) Keeps this function from returning until all threads are out 664 * of locking routines and locking is truly stopped. 665 * 2) Keeps any new requests from being processed until it's unlocked 666 * when recovery is complete. 667 */ 668 669 if (new) { 670 set_bit(LSFL_RECOVER_DOWN, &ls->ls_flags); 671 wake_up_process(ls->ls_recoverd_task); 672 wait_event(ls->ls_recover_lock_wait, 673 test_bit(LSFL_RECOVER_LOCK, &ls->ls_flags)); 674 } 675 676 /* 677 * The recoverd suspend/resume makes sure that dlm_recoverd (if 678 * running) has noticed RECOVER_STOP above and quit processing the 679 * previous recovery. 680 */ 681 682 dlm_recoverd_suspend(ls); 683 684 spin_lock_bh(&ls->ls_recover_lock); 685 kfree(ls->ls_slots); 686 ls->ls_slots = NULL; 687 ls->ls_num_slots = 0; 688 ls->ls_slots_size = 0; 689 ls->ls_recover_status = 0; 690 spin_unlock_bh(&ls->ls_recover_lock); 691 692 dlm_recoverd_resume(ls); 693 694 if (!ls->ls_recover_begin) 695 ls->ls_recover_begin = jiffies; 696 697 /* call recover_prep ops only once and not multiple times 698 * for each possible dlm_ls_stop() when recovery is already 699 * stopped. 700 * 701 * If we successful was able to clear LSFL_RUNNING bit and 702 * it was set we know it is the first dlm_ls_stop() call. 703 */ 704 if (new) 705 dlm_lsop_recover_prep(ls); 706 707 return 0; 708 } 709 710 int dlm_ls_start(struct dlm_ls *ls) 711 { 712 struct dlm_recover *rv, *rv_old; 713 struct dlm_config_node *nodes = NULL; 714 int error, count; 715 716 rv = kzalloc(sizeof(*rv), GFP_NOFS); 717 if (!rv) 718 return -ENOMEM; 719 720 error = dlm_config_nodes(ls->ls_name, &nodes, &count); 721 if (error < 0) 722 goto fail_rv; 723 724 spin_lock_bh(&ls->ls_recover_lock); 725 726 /* the lockspace needs to be stopped before it can be started */ 727 728 if (!dlm_locking_stopped(ls)) { 729 spin_unlock_bh(&ls->ls_recover_lock); 730 log_error(ls, "start ignored: lockspace running"); 731 error = -EINVAL; 732 goto fail; 733 } 734 735 rv->nodes = nodes; 736 rv->nodes_count = count; 737 rv->seq = ++ls->ls_recover_seq; 738 rv_old = ls->ls_recover_args; 739 ls->ls_recover_args = rv; 740 spin_unlock_bh(&ls->ls_recover_lock); 741 742 if (rv_old) { 743 log_error(ls, "unused recovery %llx %d", 744 (unsigned long long)rv_old->seq, rv_old->nodes_count); 745 kfree(rv_old->nodes); 746 kfree(rv_old); 747 } 748 749 set_bit(LSFL_RECOVER_WORK, &ls->ls_flags); 750 wake_up_process(ls->ls_recoverd_task); 751 return 0; 752 753 fail: 754 kfree(nodes); 755 fail_rv: 756 kfree(rv); 757 return error; 758 } 759 760