1 // SPDX-License-Identifier: GPL-2.0-only 2 /****************************************************************************** 3 ******************************************************************************* 4 ** 5 ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 6 ** Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved. 7 ** 8 ** 9 ******************************************************************************* 10 ******************************************************************************/ 11 12 #include "dlm_internal.h" 13 #include "lockspace.h" 14 #include "member.h" 15 #include "lowcomms.h" 16 #include "midcomms.h" 17 #include "rcom.h" 18 #include "recover.h" 19 #include "dir.h" 20 #include "config.h" 21 #include "memory.h" 22 #include "lock.h" 23 #include "util.h" 24 25 static int rcom_response(struct dlm_ls *ls) 26 { 27 return test_bit(LSFL_RCOM_READY, &ls->ls_flags); 28 } 29 30 static int create_rcom(struct dlm_ls *ls, int to_nodeid, int type, int len, 31 struct dlm_rcom **rc_ret, struct dlm_mhandle **mh_ret) 32 { 33 struct dlm_rcom *rc; 34 struct dlm_mhandle *mh; 35 char *mb; 36 int mb_len = sizeof(struct dlm_rcom) + len; 37 38 mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, GFP_NOFS, &mb); 39 if (!mh) { 40 log_print("create_rcom to %d type %d len %d ENOBUFS", 41 to_nodeid, type, len); 42 return -ENOBUFS; 43 } 44 memset(mb, 0, mb_len); 45 46 rc = (struct dlm_rcom *) mb; 47 48 rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR); 49 rc->rc_header.h_lockspace = ls->ls_global_id; 50 rc->rc_header.h_nodeid = dlm_our_nodeid(); 51 rc->rc_header.h_length = mb_len; 52 rc->rc_header.h_cmd = DLM_RCOM; 53 54 rc->rc_type = type; 55 56 spin_lock(&ls->ls_recover_lock); 57 rc->rc_seq = ls->ls_recover_seq; 58 spin_unlock(&ls->ls_recover_lock); 59 60 *mh_ret = mh; 61 *rc_ret = rc; 62 return 0; 63 } 64 65 static void send_rcom(struct dlm_ls *ls, struct dlm_mhandle *mh, 66 struct dlm_rcom *rc) 67 { 68 dlm_rcom_out(rc); 69 dlm_lowcomms_commit_buffer(mh); 70 } 71 72 static void set_rcom_status(struct dlm_ls *ls, struct rcom_status *rs, 73 uint32_t flags) 74 { 75 rs->rs_flags = cpu_to_le32(flags); 76 } 77 78 /* When replying to a status request, a node also sends back its 79 configuration values. The requesting node then checks that the remote 80 node is configured the same way as itself. */ 81 82 static void set_rcom_config(struct dlm_ls *ls, struct rcom_config *rf, 83 uint32_t num_slots) 84 { 85 rf->rf_lvblen = cpu_to_le32(ls->ls_lvblen); 86 rf->rf_lsflags = cpu_to_le32(ls->ls_exflags); 87 88 rf->rf_our_slot = cpu_to_le16(ls->ls_slot); 89 rf->rf_num_slots = cpu_to_le16(num_slots); 90 rf->rf_generation = cpu_to_le32(ls->ls_generation); 91 } 92 93 static int check_rcom_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid) 94 { 95 struct rcom_config *rf = (struct rcom_config *) rc->rc_buf; 96 97 if ((rc->rc_header.h_version & 0xFFFF0000) != DLM_HEADER_MAJOR) { 98 log_error(ls, "version mismatch: %x nodeid %d: %x", 99 DLM_HEADER_MAJOR | DLM_HEADER_MINOR, nodeid, 100 rc->rc_header.h_version); 101 return -EPROTO; 102 } 103 104 if (le32_to_cpu(rf->rf_lvblen) != ls->ls_lvblen || 105 le32_to_cpu(rf->rf_lsflags) != ls->ls_exflags) { 106 log_error(ls, "config mismatch: %d,%x nodeid %d: %d,%x", 107 ls->ls_lvblen, ls->ls_exflags, nodeid, 108 le32_to_cpu(rf->rf_lvblen), 109 le32_to_cpu(rf->rf_lsflags)); 110 return -EPROTO; 111 } 112 return 0; 113 } 114 115 static void allow_sync_reply(struct dlm_ls *ls, uint64_t *new_seq) 116 { 117 spin_lock(&ls->ls_rcom_spin); 118 *new_seq = ++ls->ls_rcom_seq; 119 set_bit(LSFL_RCOM_WAIT, &ls->ls_flags); 120 spin_unlock(&ls->ls_rcom_spin); 121 } 122 123 static void disallow_sync_reply(struct dlm_ls *ls) 124 { 125 spin_lock(&ls->ls_rcom_spin); 126 clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags); 127 clear_bit(LSFL_RCOM_READY, &ls->ls_flags); 128 spin_unlock(&ls->ls_rcom_spin); 129 } 130 131 /* 132 * low nodeid gathers one slot value at a time from each node. 133 * it sets need_slots=0, and saves rf_our_slot returned from each 134 * rcom_config. 135 * 136 * other nodes gather all slot values at once from the low nodeid. 137 * they set need_slots=1, and ignore the rf_our_slot returned from each 138 * rcom_config. they use the rf_num_slots returned from the low 139 * node's rcom_config. 140 */ 141 142 int dlm_rcom_status(struct dlm_ls *ls, int nodeid, uint32_t status_flags) 143 { 144 struct dlm_rcom *rc; 145 struct dlm_mhandle *mh; 146 int error = 0; 147 148 ls->ls_recover_nodeid = nodeid; 149 150 if (nodeid == dlm_our_nodeid()) { 151 rc = ls->ls_recover_buf; 152 rc->rc_result = dlm_recover_status(ls); 153 goto out; 154 } 155 156 retry: 157 error = create_rcom(ls, nodeid, DLM_RCOM_STATUS, 158 sizeof(struct rcom_status), &rc, &mh); 159 if (error) 160 goto out; 161 162 set_rcom_status(ls, (struct rcom_status *)rc->rc_buf, status_flags); 163 164 allow_sync_reply(ls, &rc->rc_id); 165 memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size); 166 167 send_rcom(ls, mh, rc); 168 169 error = dlm_wait_function(ls, &rcom_response); 170 disallow_sync_reply(ls); 171 if (error == -ETIMEDOUT) 172 goto retry; 173 if (error) 174 goto out; 175 176 rc = ls->ls_recover_buf; 177 178 if (rc->rc_result == -ESRCH) { 179 /* we pretend the remote lockspace exists with 0 status */ 180 log_debug(ls, "remote node %d not ready", nodeid); 181 rc->rc_result = 0; 182 error = 0; 183 } else { 184 error = check_rcom_config(ls, rc, nodeid); 185 } 186 187 /* the caller looks at rc_result for the remote recovery status */ 188 out: 189 return error; 190 } 191 192 static void receive_rcom_status(struct dlm_ls *ls, struct dlm_rcom *rc_in) 193 { 194 struct dlm_rcom *rc; 195 struct dlm_mhandle *mh; 196 struct rcom_status *rs; 197 uint32_t status; 198 int nodeid = rc_in->rc_header.h_nodeid; 199 int len = sizeof(struct rcom_config); 200 int num_slots = 0; 201 int error; 202 203 if (!dlm_slots_version(&rc_in->rc_header)) { 204 status = dlm_recover_status(ls); 205 goto do_create; 206 } 207 208 rs = (struct rcom_status *)rc_in->rc_buf; 209 210 if (!(le32_to_cpu(rs->rs_flags) & DLM_RSF_NEED_SLOTS)) { 211 status = dlm_recover_status(ls); 212 goto do_create; 213 } 214 215 spin_lock(&ls->ls_recover_lock); 216 status = ls->ls_recover_status; 217 num_slots = ls->ls_num_slots; 218 spin_unlock(&ls->ls_recover_lock); 219 len += num_slots * sizeof(struct rcom_slot); 220 221 do_create: 222 error = create_rcom(ls, nodeid, DLM_RCOM_STATUS_REPLY, 223 len, &rc, &mh); 224 if (error) 225 return; 226 227 rc->rc_id = rc_in->rc_id; 228 rc->rc_seq_reply = rc_in->rc_seq; 229 rc->rc_result = status; 230 231 set_rcom_config(ls, (struct rcom_config *)rc->rc_buf, num_slots); 232 233 if (!num_slots) 234 goto do_send; 235 236 spin_lock(&ls->ls_recover_lock); 237 if (ls->ls_num_slots != num_slots) { 238 spin_unlock(&ls->ls_recover_lock); 239 log_debug(ls, "receive_rcom_status num_slots %d to %d", 240 num_slots, ls->ls_num_slots); 241 rc->rc_result = 0; 242 set_rcom_config(ls, (struct rcom_config *)rc->rc_buf, 0); 243 goto do_send; 244 } 245 246 dlm_slots_copy_out(ls, rc); 247 spin_unlock(&ls->ls_recover_lock); 248 249 do_send: 250 send_rcom(ls, mh, rc); 251 } 252 253 static void receive_sync_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in) 254 { 255 spin_lock(&ls->ls_rcom_spin); 256 if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) || 257 rc_in->rc_id != ls->ls_rcom_seq) { 258 log_debug(ls, "reject reply %d from %d seq %llx expect %llx", 259 rc_in->rc_type, rc_in->rc_header.h_nodeid, 260 (unsigned long long)rc_in->rc_id, 261 (unsigned long long)ls->ls_rcom_seq); 262 goto out; 263 } 264 memcpy(ls->ls_recover_buf, rc_in, rc_in->rc_header.h_length); 265 set_bit(LSFL_RCOM_READY, &ls->ls_flags); 266 clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags); 267 wake_up(&ls->ls_wait_general); 268 out: 269 spin_unlock(&ls->ls_rcom_spin); 270 } 271 272 int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name, int last_len) 273 { 274 struct dlm_rcom *rc; 275 struct dlm_mhandle *mh; 276 int error = 0; 277 278 ls->ls_recover_nodeid = nodeid; 279 280 retry: 281 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES, last_len, &rc, &mh); 282 if (error) 283 goto out; 284 memcpy(rc->rc_buf, last_name, last_len); 285 286 allow_sync_reply(ls, &rc->rc_id); 287 memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size); 288 289 send_rcom(ls, mh, rc); 290 291 error = dlm_wait_function(ls, &rcom_response); 292 disallow_sync_reply(ls); 293 if (error == -ETIMEDOUT) 294 goto retry; 295 out: 296 return error; 297 } 298 299 static void receive_rcom_names(struct dlm_ls *ls, struct dlm_rcom *rc_in) 300 { 301 struct dlm_rcom *rc; 302 struct dlm_mhandle *mh; 303 int error, inlen, outlen, nodeid; 304 305 nodeid = rc_in->rc_header.h_nodeid; 306 inlen = rc_in->rc_header.h_length - sizeof(struct dlm_rcom); 307 outlen = dlm_config.ci_buffer_size - sizeof(struct dlm_rcom); 308 309 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES_REPLY, outlen, &rc, &mh); 310 if (error) 311 return; 312 rc->rc_id = rc_in->rc_id; 313 rc->rc_seq_reply = rc_in->rc_seq; 314 315 dlm_copy_master_names(ls, rc_in->rc_buf, inlen, rc->rc_buf, outlen, 316 nodeid); 317 send_rcom(ls, mh, rc); 318 } 319 320 int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid) 321 { 322 struct dlm_rcom *rc; 323 struct dlm_mhandle *mh; 324 struct dlm_ls *ls = r->res_ls; 325 int error; 326 327 error = create_rcom(ls, dir_nodeid, DLM_RCOM_LOOKUP, r->res_length, 328 &rc, &mh); 329 if (error) 330 goto out; 331 memcpy(rc->rc_buf, r->res_name, r->res_length); 332 rc->rc_id = (unsigned long) r->res_id; 333 334 send_rcom(ls, mh, rc); 335 out: 336 return error; 337 } 338 339 static void receive_rcom_lookup(struct dlm_ls *ls, struct dlm_rcom *rc_in) 340 { 341 struct dlm_rcom *rc; 342 struct dlm_mhandle *mh; 343 int error, ret_nodeid, nodeid = rc_in->rc_header.h_nodeid; 344 int len = rc_in->rc_header.h_length - sizeof(struct dlm_rcom); 345 346 error = create_rcom(ls, nodeid, DLM_RCOM_LOOKUP_REPLY, 0, &rc, &mh); 347 if (error) 348 return; 349 350 /* Old code would send this special id to trigger a debug dump. */ 351 if (rc_in->rc_id == 0xFFFFFFFF) { 352 log_error(ls, "receive_rcom_lookup dump from %d", nodeid); 353 dlm_dump_rsb_name(ls, rc_in->rc_buf, len); 354 return; 355 } 356 357 error = dlm_master_lookup(ls, nodeid, rc_in->rc_buf, len, 358 DLM_LU_RECOVER_MASTER, &ret_nodeid, NULL); 359 if (error) 360 ret_nodeid = error; 361 rc->rc_result = ret_nodeid; 362 rc->rc_id = rc_in->rc_id; 363 rc->rc_seq_reply = rc_in->rc_seq; 364 365 send_rcom(ls, mh, rc); 366 } 367 368 static void receive_rcom_lookup_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in) 369 { 370 dlm_recover_master_reply(ls, rc_in); 371 } 372 373 static void pack_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb, 374 struct rcom_lock *rl) 375 { 376 memset(rl, 0, sizeof(*rl)); 377 378 rl->rl_ownpid = cpu_to_le32(lkb->lkb_ownpid); 379 rl->rl_lkid = cpu_to_le32(lkb->lkb_id); 380 rl->rl_exflags = cpu_to_le32(lkb->lkb_exflags); 381 rl->rl_flags = cpu_to_le32(lkb->lkb_flags); 382 rl->rl_lvbseq = cpu_to_le32(lkb->lkb_lvbseq); 383 rl->rl_rqmode = lkb->lkb_rqmode; 384 rl->rl_grmode = lkb->lkb_grmode; 385 rl->rl_status = lkb->lkb_status; 386 rl->rl_wait_type = cpu_to_le16(lkb->lkb_wait_type); 387 388 if (lkb->lkb_bastfn) 389 rl->rl_asts |= DLM_CB_BAST; 390 if (lkb->lkb_astfn) 391 rl->rl_asts |= DLM_CB_CAST; 392 393 rl->rl_namelen = cpu_to_le16(r->res_length); 394 memcpy(rl->rl_name, r->res_name, r->res_length); 395 396 /* FIXME: might we have an lvb without DLM_LKF_VALBLK set ? 397 If so, receive_rcom_lock_args() won't take this copy. */ 398 399 if (lkb->lkb_lvbptr) 400 memcpy(rl->rl_lvb, lkb->lkb_lvbptr, r->res_ls->ls_lvblen); 401 } 402 403 int dlm_send_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb) 404 { 405 struct dlm_ls *ls = r->res_ls; 406 struct dlm_rcom *rc; 407 struct dlm_mhandle *mh; 408 struct rcom_lock *rl; 409 int error, len = sizeof(struct rcom_lock); 410 411 if (lkb->lkb_lvbptr) 412 len += ls->ls_lvblen; 413 414 error = create_rcom(ls, r->res_nodeid, DLM_RCOM_LOCK, len, &rc, &mh); 415 if (error) 416 goto out; 417 418 rl = (struct rcom_lock *) rc->rc_buf; 419 pack_rcom_lock(r, lkb, rl); 420 rc->rc_id = (unsigned long) r; 421 422 send_rcom(ls, mh, rc); 423 out: 424 return error; 425 } 426 427 /* needs at least dlm_rcom + rcom_lock */ 428 static void receive_rcom_lock(struct dlm_ls *ls, struct dlm_rcom *rc_in) 429 { 430 struct dlm_rcom *rc; 431 struct dlm_mhandle *mh; 432 int error, nodeid = rc_in->rc_header.h_nodeid; 433 434 dlm_recover_master_copy(ls, rc_in); 435 436 error = create_rcom(ls, nodeid, DLM_RCOM_LOCK_REPLY, 437 sizeof(struct rcom_lock), &rc, &mh); 438 if (error) 439 return; 440 441 /* We send back the same rcom_lock struct we received, but 442 dlm_recover_master_copy() has filled in rl_remid and rl_result */ 443 444 memcpy(rc->rc_buf, rc_in->rc_buf, sizeof(struct rcom_lock)); 445 rc->rc_id = rc_in->rc_id; 446 rc->rc_seq_reply = rc_in->rc_seq; 447 448 send_rcom(ls, mh, rc); 449 } 450 451 /* If the lockspace doesn't exist then still send a status message 452 back; it's possible that it just doesn't have its global_id yet. */ 453 454 int dlm_send_ls_not_ready(int nodeid, struct dlm_rcom *rc_in) 455 { 456 struct dlm_rcom *rc; 457 struct rcom_config *rf; 458 struct dlm_mhandle *mh; 459 char *mb; 460 int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config); 461 462 mh = dlm_lowcomms_get_buffer(nodeid, mb_len, GFP_NOFS, &mb); 463 if (!mh) 464 return -ENOBUFS; 465 memset(mb, 0, mb_len); 466 467 rc = (struct dlm_rcom *) mb; 468 469 rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR); 470 rc->rc_header.h_lockspace = rc_in->rc_header.h_lockspace; 471 rc->rc_header.h_nodeid = dlm_our_nodeid(); 472 rc->rc_header.h_length = mb_len; 473 rc->rc_header.h_cmd = DLM_RCOM; 474 475 rc->rc_type = DLM_RCOM_STATUS_REPLY; 476 rc->rc_id = rc_in->rc_id; 477 rc->rc_seq_reply = rc_in->rc_seq; 478 rc->rc_result = -ESRCH; 479 480 rf = (struct rcom_config *) rc->rc_buf; 481 rf->rf_lvblen = cpu_to_le32(~0U); 482 483 dlm_rcom_out(rc); 484 dlm_lowcomms_commit_buffer(mh); 485 486 return 0; 487 } 488 489 /* 490 * Ignore messages for stage Y before we set 491 * recover_status bit for stage X: 492 * 493 * recover_status = 0 494 * 495 * dlm_recover_members() 496 * - send nothing 497 * - recv nothing 498 * - ignore NAMES, NAMES_REPLY 499 * - ignore LOOKUP, LOOKUP_REPLY 500 * - ignore LOCK, LOCK_REPLY 501 * 502 * recover_status |= NODES 503 * 504 * dlm_recover_members_wait() 505 * 506 * dlm_recover_directory() 507 * - send NAMES 508 * - recv NAMES_REPLY 509 * - ignore LOOKUP, LOOKUP_REPLY 510 * - ignore LOCK, LOCK_REPLY 511 * 512 * recover_status |= DIR 513 * 514 * dlm_recover_directory_wait() 515 * 516 * dlm_recover_masters() 517 * - send LOOKUP 518 * - recv LOOKUP_REPLY 519 * 520 * dlm_recover_locks() 521 * - send LOCKS 522 * - recv LOCKS_REPLY 523 * 524 * recover_status |= LOCKS 525 * 526 * dlm_recover_locks_wait() 527 * 528 * recover_status |= DONE 529 */ 530 531 /* Called by dlm_recv; corresponds to dlm_receive_message() but special 532 recovery-only comms are sent through here. */ 533 534 void dlm_receive_rcom(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid) 535 { 536 int lock_size = sizeof(struct dlm_rcom) + sizeof(struct rcom_lock); 537 int stop, reply = 0, names = 0, lookup = 0, lock = 0; 538 uint32_t status; 539 uint64_t seq; 540 541 switch (rc->rc_type) { 542 case DLM_RCOM_STATUS_REPLY: 543 reply = 1; 544 break; 545 case DLM_RCOM_NAMES: 546 names = 1; 547 break; 548 case DLM_RCOM_NAMES_REPLY: 549 names = 1; 550 reply = 1; 551 break; 552 case DLM_RCOM_LOOKUP: 553 lookup = 1; 554 break; 555 case DLM_RCOM_LOOKUP_REPLY: 556 lookup = 1; 557 reply = 1; 558 break; 559 case DLM_RCOM_LOCK: 560 lock = 1; 561 break; 562 case DLM_RCOM_LOCK_REPLY: 563 lock = 1; 564 reply = 1; 565 break; 566 } 567 568 spin_lock(&ls->ls_recover_lock); 569 status = ls->ls_recover_status; 570 stop = test_bit(LSFL_RECOVER_STOP, &ls->ls_flags); 571 seq = ls->ls_recover_seq; 572 spin_unlock(&ls->ls_recover_lock); 573 574 if (stop && (rc->rc_type != DLM_RCOM_STATUS)) 575 goto ignore; 576 577 if (reply && (rc->rc_seq_reply != seq)) 578 goto ignore; 579 580 if (!(status & DLM_RS_NODES) && (names || lookup || lock)) 581 goto ignore; 582 583 if (!(status & DLM_RS_DIR) && (lookup || lock)) 584 goto ignore; 585 586 switch (rc->rc_type) { 587 case DLM_RCOM_STATUS: 588 receive_rcom_status(ls, rc); 589 break; 590 591 case DLM_RCOM_NAMES: 592 receive_rcom_names(ls, rc); 593 break; 594 595 case DLM_RCOM_LOOKUP: 596 receive_rcom_lookup(ls, rc); 597 break; 598 599 case DLM_RCOM_LOCK: 600 if (rc->rc_header.h_length < lock_size) 601 goto Eshort; 602 receive_rcom_lock(ls, rc); 603 break; 604 605 case DLM_RCOM_STATUS_REPLY: 606 receive_sync_reply(ls, rc); 607 break; 608 609 case DLM_RCOM_NAMES_REPLY: 610 receive_sync_reply(ls, rc); 611 break; 612 613 case DLM_RCOM_LOOKUP_REPLY: 614 receive_rcom_lookup_reply(ls, rc); 615 break; 616 617 case DLM_RCOM_LOCK_REPLY: 618 if (rc->rc_header.h_length < lock_size) 619 goto Eshort; 620 dlm_recover_process_copy(ls, rc); 621 break; 622 623 default: 624 log_error(ls, "receive_rcom bad type %d", rc->rc_type); 625 } 626 return; 627 628 ignore: 629 log_limit(ls, "dlm_receive_rcom ignore msg %d " 630 "from %d %llu %llu recover seq %llu sts %x gen %u", 631 rc->rc_type, 632 nodeid, 633 (unsigned long long)rc->rc_seq, 634 (unsigned long long)rc->rc_seq_reply, 635 (unsigned long long)seq, 636 status, ls->ls_generation); 637 return; 638 Eshort: 639 log_error(ls, "recovery message %d from %d is too short", 640 rc->rc_type, nodeid); 641 } 642 643