1 /****************************************************************************** 2 ******************************************************************************* 3 ** 4 ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 5 ** Copyright (C) 2005 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, GFP_KERNEL, &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 *mh_ret = mh; 60 *rc_ret = rc; 61 return 0; 62 } 63 64 static void send_rcom(struct dlm_ls *ls, struct dlm_mhandle *mh, 65 struct dlm_rcom *rc) 66 { 67 dlm_rcom_out(rc); 68 dlm_lowcomms_commit_buffer(mh); 69 } 70 71 /* When replying to a status request, a node also sends back its 72 configuration values. The requesting node then checks that the remote 73 node is configured the same way as itself. */ 74 75 static void make_config(struct dlm_ls *ls, struct rcom_config *rf) 76 { 77 rf->rf_lvblen = ls->ls_lvblen; 78 rf->rf_lsflags = ls->ls_exflags; 79 } 80 81 static int check_config(struct dlm_ls *ls, struct rcom_config *rf, int nodeid) 82 { 83 if (rf->rf_lvblen != ls->ls_lvblen || 84 rf->rf_lsflags != ls->ls_exflags) { 85 log_error(ls, "config mismatch: %d,%x nodeid %d: %d,%x", 86 ls->ls_lvblen, ls->ls_exflags, 87 nodeid, rf->rf_lvblen, rf->rf_lsflags); 88 return -EINVAL; 89 } 90 return 0; 91 } 92 93 static void allow_sync_reply(struct dlm_ls *ls, uint64_t *new_seq) 94 { 95 spin_lock(&ls->ls_rcom_spin); 96 *new_seq = ++ls->ls_rcom_seq; 97 set_bit(LSFL_RCOM_WAIT, &ls->ls_flags); 98 spin_unlock(&ls->ls_rcom_spin); 99 } 100 101 static void disallow_sync_reply(struct dlm_ls *ls) 102 { 103 spin_lock(&ls->ls_rcom_spin); 104 clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags); 105 clear_bit(LSFL_RCOM_READY, &ls->ls_flags); 106 spin_unlock(&ls->ls_rcom_spin); 107 } 108 109 int dlm_rcom_status(struct dlm_ls *ls, int nodeid) 110 { 111 struct dlm_rcom *rc; 112 struct dlm_mhandle *mh; 113 int error = 0; 114 115 ls->ls_recover_nodeid = nodeid; 116 117 if (nodeid == dlm_our_nodeid()) { 118 rc = (struct dlm_rcom *) ls->ls_recover_buf; 119 rc->rc_result = dlm_recover_status(ls); 120 goto out; 121 } 122 123 error = create_rcom(ls, nodeid, DLM_RCOM_STATUS, 0, &rc, &mh); 124 if (error) 125 goto out; 126 127 allow_sync_reply(ls, &rc->rc_id); 128 memset(ls->ls_recover_buf, 0, dlm_config.buffer_size); 129 130 send_rcom(ls, mh, rc); 131 132 error = dlm_wait_function(ls, &rcom_response); 133 disallow_sync_reply(ls); 134 if (error) 135 goto out; 136 137 rc = (struct dlm_rcom *) ls->ls_recover_buf; 138 139 if (rc->rc_result == -ESRCH) { 140 /* we pretend the remote lockspace exists with 0 status */ 141 log_debug(ls, "remote node %d not ready", nodeid); 142 rc->rc_result = 0; 143 } else 144 error = check_config(ls, (struct rcom_config *) rc->rc_buf, 145 nodeid); 146 /* the caller looks at rc_result for the remote recovery status */ 147 out: 148 return error; 149 } 150 151 static void receive_rcom_status(struct dlm_ls *ls, struct dlm_rcom *rc_in) 152 { 153 struct dlm_rcom *rc; 154 struct dlm_mhandle *mh; 155 int error, nodeid = rc_in->rc_header.h_nodeid; 156 157 error = create_rcom(ls, nodeid, DLM_RCOM_STATUS_REPLY, 158 sizeof(struct rcom_config), &rc, &mh); 159 if (error) 160 return; 161 rc->rc_id = rc_in->rc_id; 162 rc->rc_result = dlm_recover_status(ls); 163 make_config(ls, (struct rcom_config *) rc->rc_buf); 164 165 send_rcom(ls, mh, rc); 166 } 167 168 static void receive_sync_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in) 169 { 170 spin_lock(&ls->ls_rcom_spin); 171 if (!test_bit(LSFL_RCOM_WAIT, &ls->ls_flags) || 172 rc_in->rc_id != ls->ls_rcom_seq) { 173 log_debug(ls, "reject reply %d from %d seq %llx expect %llx", 174 rc_in->rc_type, rc_in->rc_header.h_nodeid, 175 (unsigned long long)rc_in->rc_id, 176 (unsigned long long)ls->ls_rcom_seq); 177 goto out; 178 } 179 memcpy(ls->ls_recover_buf, rc_in, rc_in->rc_header.h_length); 180 set_bit(LSFL_RCOM_READY, &ls->ls_flags); 181 clear_bit(LSFL_RCOM_WAIT, &ls->ls_flags); 182 wake_up(&ls->ls_wait_general); 183 out: 184 spin_unlock(&ls->ls_rcom_spin); 185 } 186 187 static void receive_rcom_status_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in) 188 { 189 receive_sync_reply(ls, rc_in); 190 } 191 192 int dlm_rcom_names(struct dlm_ls *ls, int nodeid, char *last_name, int last_len) 193 { 194 struct dlm_rcom *rc; 195 struct dlm_mhandle *mh; 196 int error = 0, len = sizeof(struct dlm_rcom); 197 198 ls->ls_recover_nodeid = nodeid; 199 200 if (nodeid == dlm_our_nodeid()) { 201 dlm_copy_master_names(ls, last_name, last_len, 202 ls->ls_recover_buf + len, 203 dlm_config.buffer_size - len, nodeid); 204 goto out; 205 } 206 207 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES, last_len, &rc, &mh); 208 if (error) 209 goto out; 210 memcpy(rc->rc_buf, last_name, last_len); 211 212 allow_sync_reply(ls, &rc->rc_id); 213 memset(ls->ls_recover_buf, 0, dlm_config.buffer_size); 214 215 send_rcom(ls, mh, rc); 216 217 error = dlm_wait_function(ls, &rcom_response); 218 disallow_sync_reply(ls); 219 out: 220 return error; 221 } 222 223 static void receive_rcom_names(struct dlm_ls *ls, struct dlm_rcom *rc_in) 224 { 225 struct dlm_rcom *rc; 226 struct dlm_mhandle *mh; 227 int error, inlen, outlen; 228 int nodeid = rc_in->rc_header.h_nodeid; 229 uint32_t status = dlm_recover_status(ls); 230 231 /* 232 * We can't run dlm_dir_rebuild_send (which uses ls_nodes) while 233 * dlm_recoverd is running ls_nodes_reconfig (which changes ls_nodes). 234 * It could only happen in rare cases where we get a late NAMES 235 * message from a previous instance of recovery. 236 */ 237 238 if (!(status & DLM_RS_NODES)) { 239 log_debug(ls, "ignoring RCOM_NAMES from %u", nodeid); 240 return; 241 } 242 243 nodeid = rc_in->rc_header.h_nodeid; 244 inlen = rc_in->rc_header.h_length - sizeof(struct dlm_rcom); 245 outlen = dlm_config.buffer_size - sizeof(struct dlm_rcom); 246 247 error = create_rcom(ls, nodeid, DLM_RCOM_NAMES_REPLY, outlen, &rc, &mh); 248 if (error) 249 return; 250 rc->rc_id = rc_in->rc_id; 251 252 dlm_copy_master_names(ls, rc_in->rc_buf, inlen, rc->rc_buf, outlen, 253 nodeid); 254 send_rcom(ls, mh, rc); 255 } 256 257 static void receive_rcom_names_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in) 258 { 259 receive_sync_reply(ls, rc_in); 260 } 261 262 int dlm_send_rcom_lookup(struct dlm_rsb *r, int dir_nodeid) 263 { 264 struct dlm_rcom *rc; 265 struct dlm_mhandle *mh; 266 struct dlm_ls *ls = r->res_ls; 267 int error; 268 269 error = create_rcom(ls, dir_nodeid, DLM_RCOM_LOOKUP, r->res_length, 270 &rc, &mh); 271 if (error) 272 goto out; 273 memcpy(rc->rc_buf, r->res_name, r->res_length); 274 rc->rc_id = (unsigned long) r; 275 276 send_rcom(ls, mh, rc); 277 out: 278 return error; 279 } 280 281 static void receive_rcom_lookup(struct dlm_ls *ls, struct dlm_rcom *rc_in) 282 { 283 struct dlm_rcom *rc; 284 struct dlm_mhandle *mh; 285 int error, ret_nodeid, nodeid = rc_in->rc_header.h_nodeid; 286 int len = rc_in->rc_header.h_length - sizeof(struct dlm_rcom); 287 288 error = create_rcom(ls, nodeid, DLM_RCOM_LOOKUP_REPLY, 0, &rc, &mh); 289 if (error) 290 return; 291 292 error = dlm_dir_lookup(ls, nodeid, rc_in->rc_buf, len, &ret_nodeid); 293 if (error) 294 ret_nodeid = error; 295 rc->rc_result = ret_nodeid; 296 rc->rc_id = rc_in->rc_id; 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 = lkb->lkb_ownpid; 312 rl->rl_lkid = lkb->lkb_id; 313 rl->rl_exflags = lkb->lkb_exflags; 314 rl->rl_flags = lkb->lkb_flags; 315 rl->rl_lvbseq = 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 = lkb->lkb_wait_type; 320 321 if (lkb->lkb_bastaddr) 322 rl->rl_asts |= AST_BAST; 323 if (lkb->lkb_astaddr) 324 rl->rl_asts |= AST_COMP; 325 326 rl->rl_namelen = 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 static void receive_rcom_lock(struct dlm_ls *ls, struct dlm_rcom *rc_in) 361 { 362 struct dlm_rcom *rc; 363 struct dlm_mhandle *mh; 364 int error, nodeid = rc_in->rc_header.h_nodeid; 365 366 dlm_recover_master_copy(ls, rc_in); 367 368 error = create_rcom(ls, nodeid, DLM_RCOM_LOCK_REPLY, 369 sizeof(struct rcom_lock), &rc, &mh); 370 if (error) 371 return; 372 373 /* We send back the same rcom_lock struct we received, but 374 dlm_recover_master_copy() has filled in rl_remid and rl_result */ 375 376 memcpy(rc->rc_buf, rc_in->rc_buf, sizeof(struct rcom_lock)); 377 rc->rc_id = rc_in->rc_id; 378 379 send_rcom(ls, mh, rc); 380 } 381 382 static void receive_rcom_lock_reply(struct dlm_ls *ls, struct dlm_rcom *rc_in) 383 { 384 uint32_t status = dlm_recover_status(ls); 385 386 if (!(status & DLM_RS_DIR)) { 387 log_debug(ls, "ignoring RCOM_LOCK_REPLY from %u", 388 rc_in->rc_header.h_nodeid); 389 return; 390 } 391 392 dlm_recover_process_copy(ls, rc_in); 393 } 394 395 static int send_ls_not_ready(int nodeid, struct dlm_rcom *rc_in) 396 { 397 struct dlm_rcom *rc; 398 struct rcom_config *rf; 399 struct dlm_mhandle *mh; 400 char *mb; 401 int mb_len = sizeof(struct dlm_rcom) + sizeof(struct rcom_config); 402 403 mh = dlm_lowcomms_get_buffer(nodeid, mb_len, GFP_KERNEL, &mb); 404 if (!mh) 405 return -ENOBUFS; 406 memset(mb, 0, mb_len); 407 408 rc = (struct dlm_rcom *) mb; 409 410 rc->rc_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR); 411 rc->rc_header.h_lockspace = rc_in->rc_header.h_lockspace; 412 rc->rc_header.h_nodeid = dlm_our_nodeid(); 413 rc->rc_header.h_length = mb_len; 414 rc->rc_header.h_cmd = DLM_RCOM; 415 416 rc->rc_type = DLM_RCOM_STATUS_REPLY; 417 rc->rc_id = rc_in->rc_id; 418 rc->rc_result = -ESRCH; 419 420 rf = (struct rcom_config *) rc->rc_buf; 421 rf->rf_lvblen = -1; 422 423 dlm_rcom_out(rc); 424 dlm_lowcomms_commit_buffer(mh); 425 426 return 0; 427 } 428 429 /* Called by dlm_recvd; corresponds to dlm_receive_message() but special 430 recovery-only comms are sent through here. */ 431 432 void dlm_receive_rcom(struct dlm_header *hd, int nodeid) 433 { 434 struct dlm_rcom *rc = (struct dlm_rcom *) hd; 435 struct dlm_ls *ls; 436 437 dlm_rcom_in(rc); 438 439 /* If the lockspace doesn't exist then still send a status message 440 back; it's possible that it just doesn't have its global_id yet. */ 441 442 ls = dlm_find_lockspace_global(hd->h_lockspace); 443 if (!ls) { 444 log_print("lockspace %x from %d type %x not found", 445 hd->h_lockspace, nodeid, rc->rc_type); 446 if (rc->rc_type == DLM_RCOM_STATUS) 447 send_ls_not_ready(nodeid, rc); 448 return; 449 } 450 451 if (dlm_recovery_stopped(ls) && (rc->rc_type != DLM_RCOM_STATUS)) { 452 log_error(ls, "ignoring recovery message %x from %d", 453 rc->rc_type, nodeid); 454 goto out; 455 } 456 457 if (nodeid != rc->rc_header.h_nodeid) { 458 log_error(ls, "bad rcom nodeid %d from %d", 459 rc->rc_header.h_nodeid, nodeid); 460 goto out; 461 } 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 receive_rcom_lock(ls, rc); 478 break; 479 480 case DLM_RCOM_STATUS_REPLY: 481 receive_rcom_status_reply(ls, rc); 482 break; 483 484 case DLM_RCOM_NAMES_REPLY: 485 receive_rcom_names_reply(ls, rc); 486 break; 487 488 case DLM_RCOM_LOOKUP_REPLY: 489 receive_rcom_lookup_reply(ls, rc); 490 break; 491 492 case DLM_RCOM_LOCK_REPLY: 493 receive_rcom_lock_reply(ls, rc); 494 break; 495 496 default: 497 DLM_ASSERT(0, printk("rc_type=%x\n", rc->rc_type);); 498 } 499 out: 500 dlm_put_lockspace(ls); 501 } 502 503