1 /****************************************************************************** 2 ******************************************************************************* 3 ** 4 ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 5 ** Copyright (C) 2005-2008 Red Hat, Inc. All rights reserved. 6 ** 7 ** This copyrighted material is made available to anyone wishing to use, 8 ** modify, copy, or redistribute it subject to the terms and conditions 9 ** of the GNU General Public License v.2. 10 ** 11 ******************************************************************************* 12 ******************************************************************************/ 13 14 #include "dlm_internal.h" 15 #include "lockspace.h" 16 #include "member.h" 17 #include "lowcomms.h" 18 #include "midcomms.h" 19 #include "rcom.h" 20 #include "recover.h" 21 #include "dir.h" 22 #include "config.h" 23 #include "memory.h" 24 #include "lock.h" 25 #include "util.h" 26 27 28 static int rcom_response(struct dlm_ls *ls) 29 { 30 return test_bit(LSFL_RCOM_READY, &ls->ls_flags); 31 } 32 33 static int create_rcom(struct dlm_ls *ls, int to_nodeid, int type, int len, 34 struct dlm_rcom **rc_ret, struct dlm_mhandle **mh_ret) 35 { 36 struct dlm_rcom *rc; 37 struct dlm_mhandle *mh; 38 char *mb; 39 int mb_len = sizeof(struct dlm_rcom) + len; 40 41 mh = dlm_lowcomms_get_buffer(to_nodeid, mb_len, ls->ls_allocation, &mb); 42 if (!mh) { 43 log_print("create_rcom to %d type %d len %d ENOBUFS", 44 to_nodeid, type, len); 45 return -ENOBUFS; 46 } 47 memset(mb, 0, mb_len); 48 49 rc = (struct dlm_rcom *) mb; 50 51 rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR); 52 rc->rc_header.h_lockspace = ls->ls_global_id; 53 rc->rc_header.h_nodeid = dlm_our_nodeid(); 54 rc->rc_header.h_length = mb_len; 55 rc->rc_header.h_cmd = DLM_RCOM; 56 57 rc->rc_type = type; 58 59 spin_lock(&ls->ls_recover_lock); 60 rc->rc_seq = ls->ls_recover_seq; 61 spin_unlock(&ls->ls_recover_lock); 62 63 *mh_ret = mh; 64 *rc_ret = rc; 65 return 0; 66 } 67 68 static void send_rcom(struct dlm_ls *ls, struct dlm_mhandle *mh, 69 struct dlm_rcom *rc) 70 { 71 dlm_rcom_out(rc); 72 dlm_lowcomms_commit_buffer(mh); 73 } 74 75 /* When replying to a status request, a node also sends back its 76 configuration values. The requesting node then checks that the remote 77 node is configured the same way as itself. */ 78 79 static void make_config(struct dlm_ls *ls, struct rcom_config *rf) 80 { 81 rf->rf_lvblen = cpu_to_le32(ls->ls_lvblen); 82 rf->rf_lsflags = cpu_to_le32(ls->ls_exflags); 83 } 84 85 static int check_config(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid) 86 { 87 struct rcom_config *rf = (struct rcom_config *) rc->rc_buf; 88 size_t conf_size = sizeof(struct dlm_rcom) + sizeof(struct rcom_config); 89 90 if ((rc->rc_header.h_version & 0xFFFF0000) != DLM_HEADER_MAJOR) { 91 log_error(ls, "version mismatch: %x nodeid %d: %x", 92 DLM_HEADER_MAJOR | DLM_HEADER_MINOR, nodeid, 93 rc->rc_header.h_version); 94 return -EPROTO; 95 } 96 97 if (rc->rc_header.h_length < conf_size) { 98 log_error(ls, "config too short: %d nodeid %d", 99 rc->rc_header.h_length, nodeid); 100 return -EPROTO; 101 } 102 103 if (le32_to_cpu(rf->rf_lvblen) != ls->ls_lvblen || 104 le32_to_cpu(rf->rf_lsflags) != ls->ls_exflags) { 105 log_error(ls, "config mismatch: %d,%x nodeid %d: %d,%x", 106 ls->ls_lvblen, ls->ls_exflags, nodeid, 107 le32_to_cpu(rf->rf_lvblen), 108 le32_to_cpu(rf->rf_lsflags)); 109 return -EPROTO; 110 } 111 return 0; 112 } 113 114 static void allow_sync_reply(struct dlm_ls *ls, uint64_t *new_seq) 115 { 116 spin_lock(&ls->ls_rcom_spin); 117 *new_seq = ++ls->ls_rcom_seq; 118 set_bit(LSFL_RCOM_WAIT, &ls->ls_flags); 119 spin_unlock(&ls->ls_rcom_spin); 120 } 121 122 static void disallow_sync_reply(struct dlm_ls *ls) 123 { 124 spin_lock(&ls->ls_rcom_spin); 125 clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags); 126 clear_bit(LSFL_RCOM_READY, &ls->ls_flags); 127 spin_unlock(&ls->ls_rcom_spin); 128 } 129 130 int dlm_rcom_status(struct dlm_ls *ls, int nodeid) 131 { 132 struct dlm_rcom *rc; 133 struct dlm_mhandle *mh; 134 int error = 0; 135 136 ls->ls_recover_nodeid = nodeid; 137 138 if (nodeid == dlm_our_nodeid()) { 139 rc = ls->ls_recover_buf; 140 rc->rc_result = dlm_recover_status(ls); 141 goto out; 142 } 143 144 error = create_rcom(ls, nodeid, DLM_RCOM_STATUS, 0, &rc, &mh); 145 if (error) 146 goto out; 147 148 allow_sync_reply(ls, &rc->rc_id); 149 memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size); 150 151 send_rcom(ls, mh, rc); 152 153 error = dlm_wait_function(ls, &rcom_response); 154 disallow_sync_reply(ls); 155 if (error) 156 goto out; 157 158 rc = ls->ls_recover_buf; 159 160 if (rc->rc_result == -ESRCH) { 161 /* we pretend the remote lockspace exists with 0 status */ 162 log_debug(ls, "remote node %d not ready", nodeid); 163 rc->rc_result = 0; 164 } else 165 error = check_config(ls, rc, nodeid); 166 /* the caller looks at rc_result for the remote recovery status */ 167 out: 168 return error; 169 } 170 171 static void receive_rcom_status(struct dlm_ls *ls, struct dlm_rcom *rc_in) 172 { 173 struct dlm_rcom *rc; 174 struct dlm_mhandle *mh; 175 int error, nodeid = rc_in->rc_header.h_nodeid; 176 177 error = create_rcom(ls, nodeid, DLM_RCOM_STATUS_REPLY, 178 sizeof(struct rcom_config), &rc, &mh); 179 if (error) 180 return; 181 rc->rc_id = rc_in->rc_id; 182 rc->rc_seq_reply = rc_in->rc_seq; 183 rc->rc_result = dlm_recover_status(ls); 184 make_config(ls, (struct rcom_config *) rc->rc_buf); 185 186 send_rcom(ls, mh, rc); 187 } 188 189 static void receive_sync_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in) 190 { 191 spin_lock(&ls->ls_rcom_spin); 192 if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) || 193 rc_in->rc_id != ls->ls_rcom_seq) { 194 log_debug(ls, "reject reply %d from %d seq %llx expect %llx", 195 rc_in->rc_type, rc_in->rc_header.h_nodeid, 196 (unsigned long long)rc_in->rc_id, 197 (unsigned long long)ls->ls_rcom_seq); 198 goto out; 199 } 200 memcpy(ls->ls_recover_buf, rc_in, rc_in->rc_header.h_length); 201 set_bit(LSFL_RCOM_READY, &ls->ls_flags); 202 clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags); 203 wake_up(&ls->ls_wait_general); 204 out: 205 spin_unlock(&ls->ls_rcom_spin); 206 } 207 208 int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name, int last_len) 209 { 210 struct dlm_rcom *rc; 211 struct dlm_mhandle *mh; 212 int error = 0; 213 int max_size = dlm_config.ci_buffer_size - sizeof(struct dlm_rcom); 214 215 ls->ls_recover_nodeid = nodeid; 216 217 if (nodeid == dlm_our_nodeid()) { 218 dlm_copy_master_names(ls, last_name, last_len, 219 ls->ls_recover_buf->rc_buf, 220 max_size, nodeid); 221 goto out; 222 } 223 224 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES, last_len, &rc, &mh); 225 if (error) 226 goto out; 227 memcpy(rc->rc_buf, last_name, last_len); 228 229 allow_sync_reply(ls, &rc->rc_id); 230 memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size); 231 232 send_rcom(ls, mh, rc); 233 234 error = dlm_wait_function(ls, &rcom_response); 235 disallow_sync_reply(ls); 236 out: 237 return error; 238 } 239 240 static void receive_rcom_names(struct dlm_ls *ls, struct dlm_rcom *rc_in) 241 { 242 struct dlm_rcom *rc; 243 struct dlm_mhandle *mh; 244 int error, inlen, outlen, nodeid; 245 246 nodeid = rc_in->rc_header.h_nodeid; 247 inlen = rc_in->rc_header.h_length - sizeof(struct dlm_rcom); 248 outlen = dlm_config.ci_buffer_size - sizeof(struct dlm_rcom); 249 250 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES_REPLY, outlen, &rc, &mh); 251 if (error) 252 return; 253 rc->rc_id = rc_in->rc_id; 254 rc->rc_seq_reply = rc_in->rc_seq; 255 256 dlm_copy_master_names(ls, rc_in->rc_buf, inlen, rc->rc_buf, outlen, 257 nodeid); 258 send_rcom(ls, mh, rc); 259 } 260 261 int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid) 262 { 263 struct dlm_rcom *rc; 264 struct dlm_mhandle *mh; 265 struct dlm_ls *ls = r->res_ls; 266 int error; 267 268 error = create_rcom(ls, dir_nodeid, DLM_RCOM_LOOKUP, r->res_length, 269 &rc, &mh); 270 if (error) 271 goto out; 272 memcpy(rc->rc_buf, r->res_name, r->res_length); 273 rc->rc_id = (unsigned long) r; 274 275 send_rcom(ls, mh, rc); 276 out: 277 return error; 278 } 279 280 static void receive_rcom_lookup(struct dlm_ls *ls, struct dlm_rcom *rc_in) 281 { 282 struct dlm_rcom *rc; 283 struct dlm_mhandle *mh; 284 int error, ret_nodeid, nodeid = rc_in->rc_header.h_nodeid; 285 int len = rc_in->rc_header.h_length - sizeof(struct dlm_rcom); 286 287 error = create_rcom(ls, nodeid, DLM_RCOM_LOOKUP_REPLY, 0, &rc, &mh); 288 if (error) 289 return; 290 291 error = dlm_dir_lookup(ls, nodeid, rc_in->rc_buf, len, &ret_nodeid); 292 if (error) 293 ret_nodeid = error; 294 rc->rc_result = ret_nodeid; 295 rc->rc_id = rc_in->rc_id; 296 rc->rc_seq_reply = rc_in->rc_seq; 297 298 send_rcom(ls, mh, rc); 299 } 300 301 static void receive_rcom_lookup_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in) 302 { 303 dlm_recover_master_reply(ls, rc_in); 304 } 305 306 static void pack_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb, 307 struct rcom_lock *rl) 308 { 309 memset(rl, 0, sizeof(*rl)); 310 311 rl->rl_ownpid = cpu_to_le32(lkb->lkb_ownpid); 312 rl->rl_lkid = cpu_to_le32(lkb->lkb_id); 313 rl->rl_exflags = cpu_to_le32(lkb->lkb_exflags); 314 rl->rl_flags = cpu_to_le32(lkb->lkb_flags); 315 rl->rl_lvbseq = cpu_to_le32(lkb->lkb_lvbseq); 316 rl->rl_rqmode = lkb->lkb_rqmode; 317 rl->rl_grmode = lkb->lkb_grmode; 318 rl->rl_status = lkb->lkb_status; 319 rl->rl_wait_type = cpu_to_le16(lkb->lkb_wait_type); 320 321 if (lkb->lkb_bastfn) 322 rl->rl_asts |= AST_BAST; 323 if (lkb->lkb_astfn) 324 rl->rl_asts |= AST_COMP; 325 326 rl->rl_namelen = cpu_to_le16(r->res_length); 327 memcpy(rl->rl_name, r->res_name, r->res_length); 328 329 /* FIXME: might we have an lvb without DLM_LKF_VALBLK set ? 330 If so, receive_rcom_lock_args() won't take this copy. */ 331 332 if (lkb->lkb_lvbptr) 333 memcpy(rl->rl_lvb, lkb->lkb_lvbptr, r->res_ls->ls_lvblen); 334 } 335 336 int dlm_send_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb) 337 { 338 struct dlm_ls *ls = r->res_ls; 339 struct dlm_rcom *rc; 340 struct dlm_mhandle *mh; 341 struct rcom_lock *rl; 342 int error, len = sizeof(struct rcom_lock); 343 344 if (lkb->lkb_lvbptr) 345 len += ls->ls_lvblen; 346 347 error = create_rcom(ls, r->res_nodeid, DLM_RCOM_LOCK, len, &rc, &mh); 348 if (error) 349 goto out; 350 351 rl = (struct rcom_lock *) rc->rc_buf; 352 pack_rcom_lock(r, lkb, rl); 353 rc->rc_id = (unsigned long) r; 354 355 send_rcom(ls, mh, rc); 356 out: 357 return error; 358 } 359 360 /* needs at least dlm_rcom + rcom_lock */ 361 static void receive_rcom_lock(struct dlm_ls *ls, struct dlm_rcom *rc_in) 362 { 363 struct dlm_rcom *rc; 364 struct dlm_mhandle *mh; 365 int error, nodeid = rc_in->rc_header.h_nodeid; 366 367 dlm_recover_master_copy(ls, rc_in); 368 369 error = create_rcom(ls, nodeid, DLM_RCOM_LOCK_REPLY, 370 sizeof(struct rcom_lock), &rc, &mh); 371 if (error) 372 return; 373 374 /* We send back the same rcom_lock struct we received, but 375 dlm_recover_master_copy() has filled in rl_remid and rl_result */ 376 377 memcpy(rc->rc_buf, rc_in->rc_buf, sizeof(struct rcom_lock)); 378 rc->rc_id = rc_in->rc_id; 379 rc->rc_seq_reply = rc_in->rc_seq; 380 381 send_rcom(ls, mh, rc); 382 } 383 384 /* If the lockspace doesn't exist then still send a status message 385 back; it's possible that it just doesn't have its global_id yet. */ 386 387 int dlm_send_ls_not_ready(int nodeid, struct dlm_rcom *rc_in) 388 { 389 struct dlm_rcom *rc; 390 struct rcom_config *rf; 391 struct dlm_mhandle *mh; 392 char *mb; 393 int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config); 394 395 mh = dlm_lowcomms_get_buffer(nodeid, mb_len, GFP_NOFS, &mb); 396 if (!mh) 397 return -ENOBUFS; 398 memset(mb, 0, mb_len); 399 400 rc = (struct dlm_rcom *) mb; 401 402 rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR); 403 rc->rc_header.h_lockspace = rc_in->rc_header.h_lockspace; 404 rc->rc_header.h_nodeid = dlm_our_nodeid(); 405 rc->rc_header.h_length = mb_len; 406 rc->rc_header.h_cmd = DLM_RCOM; 407 408 rc->rc_type = DLM_RCOM_STATUS_REPLY; 409 rc->rc_id = rc_in->rc_id; 410 rc->rc_seq_reply = rc_in->rc_seq; 411 rc->rc_result = -ESRCH; 412 413 rf = (struct rcom_config *) rc->rc_buf; 414 rf->rf_lvblen = cpu_to_le32(~0U); 415 416 dlm_rcom_out(rc); 417 dlm_lowcomms_commit_buffer(mh); 418 419 return 0; 420 } 421 422 static int is_old_reply(struct dlm_ls *ls, struct dlm_rcom *rc) 423 { 424 uint64_t seq; 425 int rv = 0; 426 427 switch (rc->rc_type) { 428 case DLM_RCOM_STATUS_REPLY: 429 case DLM_RCOM_NAMES_REPLY: 430 case DLM_RCOM_LOOKUP_REPLY: 431 case DLM_RCOM_LOCK_REPLY: 432 spin_lock(&ls->ls_recover_lock); 433 seq = ls->ls_recover_seq; 434 spin_unlock(&ls->ls_recover_lock); 435 if (rc->rc_seq_reply != seq) { 436 log_debug(ls, "ignoring old reply %x from %d " 437 "seq_reply %llx expect %llx", 438 rc->rc_type, rc->rc_header.h_nodeid, 439 (unsigned long long)rc->rc_seq_reply, 440 (unsigned long long)seq); 441 rv = 1; 442 } 443 } 444 return rv; 445 } 446 447 /* Called by dlm_recv; corresponds to dlm_receive_message() but special 448 recovery-only comms are sent through here. */ 449 450 void dlm_receive_rcom(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid) 451 { 452 int lock_size = sizeof(struct dlm_rcom) + sizeof(struct rcom_lock); 453 454 if (dlm_recovery_stopped(ls) && (rc->rc_type != DLM_RCOM_STATUS)) { 455 log_debug(ls, "ignoring recovery message %x from %d", 456 rc->rc_type, nodeid); 457 goto out; 458 } 459 460 if (is_old_reply(ls, rc)) 461 goto out; 462 463 switch (rc->rc_type) { 464 case DLM_RCOM_STATUS: 465 receive_rcom_status(ls, rc); 466 break; 467 468 case DLM_RCOM_NAMES: 469 receive_rcom_names(ls, rc); 470 break; 471 472 case DLM_RCOM_LOOKUP: 473 receive_rcom_lookup(ls, rc); 474 break; 475 476 case DLM_RCOM_LOCK: 477 if (rc->rc_header.h_length < lock_size) 478 goto Eshort; 479 receive_rcom_lock(ls, rc); 480 break; 481 482 case DLM_RCOM_STATUS_REPLY: 483 receive_sync_reply(ls, rc); 484 break; 485 486 case DLM_RCOM_NAMES_REPLY: 487 receive_sync_reply(ls, rc); 488 break; 489 490 case DLM_RCOM_LOOKUP_REPLY: 491 receive_rcom_lookup_reply(ls, rc); 492 break; 493 494 case DLM_RCOM_LOCK_REPLY: 495 if (rc->rc_header.h_length < lock_size) 496 goto Eshort; 497 dlm_recover_process_copy(ls, rc); 498 break; 499 500 default: 501 log_error(ls, "receive_rcom bad type %d", rc->rc_type); 502 } 503 out: 504 return; 505 Eshort: 506 log_error(ls, "recovery message %x from %d is too short", 507 rc->rc_type, nodeid); 508 } 509 510