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 ls->ls_recover_buf->rc_header.h_length = 219 dlm_config.ci_buffer_size; 220 dlm_copy_master_names(ls, last_name, last_len, 221 ls->ls_recover_buf->rc_buf, 222 max_size, nodeid); 223 goto out; 224 } 225 226 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES, last_len, &rc, &mh); 227 if (error) 228 goto out; 229 memcpy(rc->rc_buf, last_name, last_len); 230 231 allow_sync_reply(ls, &rc->rc_id); 232 memset(ls->ls_recover_buf, 0, dlm_config.ci_buffer_size); 233 234 send_rcom(ls, mh, rc); 235 236 error = dlm_wait_function(ls, &rcom_response); 237 disallow_sync_reply(ls); 238 out: 239 return error; 240 } 241 242 static void receive_rcom_names(struct dlm_ls *ls, struct dlm_rcom *rc_in) 243 { 244 struct dlm_rcom *rc; 245 struct dlm_mhandle *mh; 246 int error, inlen, outlen, nodeid; 247 248 nodeid = rc_in->rc_header.h_nodeid; 249 inlen = rc_in->rc_header.h_length - sizeof(struct dlm_rcom); 250 outlen = dlm_config.ci_buffer_size - sizeof(struct dlm_rcom); 251 252 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES_REPLY, outlen, &rc, &mh); 253 if (error) 254 return; 255 rc->rc_id = rc_in->rc_id; 256 rc->rc_seq_reply = rc_in->rc_seq; 257 258 dlm_copy_master_names(ls, rc_in->rc_buf, inlen, rc->rc_buf, outlen, 259 nodeid); 260 send_rcom(ls, mh, rc); 261 } 262 263 int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid) 264 { 265 struct dlm_rcom *rc; 266 struct dlm_mhandle *mh; 267 struct dlm_ls *ls = r->res_ls; 268 int error; 269 270 error = create_rcom(ls, dir_nodeid, DLM_RCOM_LOOKUP, r->res_length, 271 &rc, &mh); 272 if (error) 273 goto out; 274 memcpy(rc->rc_buf, r->res_name, r->res_length); 275 rc->rc_id = (unsigned long) r; 276 277 send_rcom(ls, mh, rc); 278 out: 279 return error; 280 } 281 282 static void receive_rcom_lookup(struct dlm_ls *ls, struct dlm_rcom *rc_in) 283 { 284 struct dlm_rcom *rc; 285 struct dlm_mhandle *mh; 286 int error, ret_nodeid, nodeid = rc_in->rc_header.h_nodeid; 287 int len = rc_in->rc_header.h_length - sizeof(struct dlm_rcom); 288 289 error = create_rcom(ls, nodeid, DLM_RCOM_LOOKUP_REPLY, 0, &rc, &mh); 290 if (error) 291 return; 292 293 error = dlm_dir_lookup(ls, nodeid, rc_in->rc_buf, len, &ret_nodeid); 294 if (error) 295 ret_nodeid = error; 296 rc->rc_result = ret_nodeid; 297 rc->rc_id = rc_in->rc_id; 298 rc->rc_seq_reply = rc_in->rc_seq; 299 300 send_rcom(ls, mh, rc); 301 } 302 303 static void receive_rcom_lookup_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in) 304 { 305 dlm_recover_master_reply(ls, rc_in); 306 } 307 308 static void pack_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb, 309 struct rcom_lock *rl) 310 { 311 memset(rl, 0, sizeof(*rl)); 312 313 rl->rl_ownpid = cpu_to_le32(lkb->lkb_ownpid); 314 rl->rl_lkid = cpu_to_le32(lkb->lkb_id); 315 rl->rl_exflags = cpu_to_le32(lkb->lkb_exflags); 316 rl->rl_flags = cpu_to_le32(lkb->lkb_flags); 317 rl->rl_lvbseq = cpu_to_le32(lkb->lkb_lvbseq); 318 rl->rl_rqmode = lkb->lkb_rqmode; 319 rl->rl_grmode = lkb->lkb_grmode; 320 rl->rl_status = lkb->lkb_status; 321 rl->rl_wait_type = cpu_to_le16(lkb->lkb_wait_type); 322 323 if (lkb->lkb_bastfn) 324 rl->rl_asts |= AST_BAST; 325 if (lkb->lkb_astfn) 326 rl->rl_asts |= AST_COMP; 327 328 rl->rl_namelen = cpu_to_le16(r->res_length); 329 memcpy(rl->rl_name, r->res_name, r->res_length); 330 331 /* FIXME: might we have an lvb without DLM_LKF_VALBLK set ? 332 If so, receive_rcom_lock_args() won't take this copy. */ 333 334 if (lkb->lkb_lvbptr) 335 memcpy(rl->rl_lvb, lkb->lkb_lvbptr, r->res_ls->ls_lvblen); 336 } 337 338 int dlm_send_rcom_lock(struct dlm_rsb *r, struct dlm_lkb *lkb) 339 { 340 struct dlm_ls *ls = r->res_ls; 341 struct dlm_rcom *rc; 342 struct dlm_mhandle *mh; 343 struct rcom_lock *rl; 344 int error, len = sizeof(struct rcom_lock); 345 346 if (lkb->lkb_lvbptr) 347 len += ls->ls_lvblen; 348 349 error = create_rcom(ls, r->res_nodeid, DLM_RCOM_LOCK, len, &rc, &mh); 350 if (error) 351 goto out; 352 353 rl = (struct rcom_lock *) rc->rc_buf; 354 pack_rcom_lock(r, lkb, rl); 355 rc->rc_id = (unsigned long) r; 356 357 send_rcom(ls, mh, rc); 358 out: 359 return error; 360 } 361 362 /* needs at least dlm_rcom + rcom_lock */ 363 static void receive_rcom_lock(struct dlm_ls *ls, struct dlm_rcom *rc_in) 364 { 365 struct dlm_rcom *rc; 366 struct dlm_mhandle *mh; 367 int error, nodeid = rc_in->rc_header.h_nodeid; 368 369 dlm_recover_master_copy(ls, rc_in); 370 371 error = create_rcom(ls, nodeid, DLM_RCOM_LOCK_REPLY, 372 sizeof(struct rcom_lock), &rc, &mh); 373 if (error) 374 return; 375 376 /* We send back the same rcom_lock struct we received, but 377 dlm_recover_master_copy() has filled in rl_remid and rl_result */ 378 379 memcpy(rc->rc_buf, rc_in->rc_buf, sizeof(struct rcom_lock)); 380 rc->rc_id = rc_in->rc_id; 381 rc->rc_seq_reply = rc_in->rc_seq; 382 383 send_rcom(ls, mh, rc); 384 } 385 386 /* If the lockspace doesn't exist then still send a status message 387 back; it's possible that it just doesn't have its global_id yet. */ 388 389 int dlm_send_ls_not_ready(int nodeid, struct dlm_rcom *rc_in) 390 { 391 struct dlm_rcom *rc; 392 struct rcom_config *rf; 393 struct dlm_mhandle *mh; 394 char *mb; 395 int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config); 396 397 mh = dlm_lowcomms_get_buffer(nodeid, mb_len, GFP_NOFS, &mb); 398 if (!mh) 399 return -ENOBUFS; 400 memset(mb, 0, mb_len); 401 402 rc = (struct dlm_rcom *) mb; 403 404 rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR); 405 rc->rc_header.h_lockspace = rc_in->rc_header.h_lockspace; 406 rc->rc_header.h_nodeid = dlm_our_nodeid(); 407 rc->rc_header.h_length = mb_len; 408 rc->rc_header.h_cmd = DLM_RCOM; 409 410 rc->rc_type = DLM_RCOM_STATUS_REPLY; 411 rc->rc_id = rc_in->rc_id; 412 rc->rc_seq_reply = rc_in->rc_seq; 413 rc->rc_result = -ESRCH; 414 415 rf = (struct rcom_config *) rc->rc_buf; 416 rf->rf_lvblen = cpu_to_le32(~0U); 417 418 dlm_rcom_out(rc); 419 dlm_lowcomms_commit_buffer(mh); 420 421 return 0; 422 } 423 424 static int is_old_reply(struct dlm_ls *ls, struct dlm_rcom *rc) 425 { 426 uint64_t seq; 427 int rv = 0; 428 429 switch (rc->rc_type) { 430 case DLM_RCOM_STATUS_REPLY: 431 case DLM_RCOM_NAMES_REPLY: 432 case DLM_RCOM_LOOKUP_REPLY: 433 case DLM_RCOM_LOCK_REPLY: 434 spin_lock(&ls->ls_recover_lock); 435 seq = ls->ls_recover_seq; 436 spin_unlock(&ls->ls_recover_lock); 437 if (rc->rc_seq_reply != seq) { 438 log_debug(ls, "ignoring old reply %x from %d " 439 "seq_reply %llx expect %llx", 440 rc->rc_type, rc->rc_header.h_nodeid, 441 (unsigned long long)rc->rc_seq_reply, 442 (unsigned long long)seq); 443 rv = 1; 444 } 445 } 446 return rv; 447 } 448 449 /* Called by dlm_recv; corresponds to dlm_receive_message() but special 450 recovery-only comms are sent through here. */ 451 452 void dlm_receive_rcom(struct dlm_ls *ls, struct dlm_rcom *rc, int nodeid) 453 { 454 int lock_size = sizeof(struct dlm_rcom) + sizeof(struct rcom_lock); 455 456 if (dlm_recovery_stopped(ls) && (rc->rc_type != DLM_RCOM_STATUS)) { 457 log_debug(ls, "ignoring recovery message %x from %d", 458 rc->rc_type, nodeid); 459 goto out; 460 } 461 462 if (is_old_reply(ls, rc)) 463 goto out; 464 465 switch (rc->rc_type) { 466 case DLM_RCOM_STATUS: 467 receive_rcom_status(ls, rc); 468 break; 469 470 case DLM_RCOM_NAMES: 471 receive_rcom_names(ls, rc); 472 break; 473 474 case DLM_RCOM_LOOKUP: 475 receive_rcom_lookup(ls, rc); 476 break; 477 478 case DLM_RCOM_LOCK: 479 if (rc->rc_header.h_length < lock_size) 480 goto Eshort; 481 receive_rcom_lock(ls, rc); 482 break; 483 484 case DLM_RCOM_STATUS_REPLY: 485 receive_sync_reply(ls, rc); 486 break; 487 488 case DLM_RCOM_NAMES_REPLY: 489 receive_sync_reply(ls, rc); 490 break; 491 492 case DLM_RCOM_LOOKUP_REPLY: 493 receive_rcom_lookup_reply(ls, rc); 494 break; 495 496 case DLM_RCOM_LOCK_REPLY: 497 if (rc->rc_header.h_length < lock_size) 498 goto Eshort; 499 dlm_recover_process_copy(ls, rc); 500 break; 501 502 default: 503 log_error(ls, "receive_rcom bad type %d", rc->rc_type); 504 } 505 out: 506 return; 507 Eshort: 508 log_error(ls, "recovery message %x from %d is too short", 509 rc->rc_type, nodeid); 510 } 511 512