1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright 2012 Milan Jurik. All rights reserved. 25 * 26 * fme.c -- fault management exercise module 27 * 28 * this module provides the simulated fault management exercise. 29 */ 30 31 #include <stdio.h> 32 #include <stdlib.h> 33 #include <string.h> 34 #include <strings.h> 35 #include <ctype.h> 36 #include <alloca.h> 37 #include <libnvpair.h> 38 #include <sys/fm/protocol.h> 39 #include <fm/fmd_api.h> 40 #include "alloc.h" 41 #include "out.h" 42 #include "stats.h" 43 #include "stable.h" 44 #include "literals.h" 45 #include "lut.h" 46 #include "tree.h" 47 #include "ptree.h" 48 #include "itree.h" 49 #include "ipath.h" 50 #include "fme.h" 51 #include "evnv.h" 52 #include "eval.h" 53 #include "config.h" 54 #include "platform.h" 55 #include "esclex.h" 56 57 /* imported from eft.c... */ 58 extern hrtime_t Hesitate; 59 extern char *Serd_Override; 60 extern nv_alloc_t Eft_nv_hdl; 61 extern int Max_fme; 62 extern fmd_hdl_t *Hdl; 63 64 static int Istat_need_save; 65 static int Serd_need_save; 66 void istat_save(void); 67 void serd_save(void); 68 69 /* fme under construction is global so we can free it on module abort */ 70 static struct fme *Nfmep; 71 72 static int Undiag_reason = UD_VAL_UNKNOWN; 73 74 static int Nextid = 0; 75 76 static int Open_fme_count = 0; /* Count of open FMEs */ 77 78 /* list of fault management exercises underway */ 79 static struct fme { 80 struct fme *next; /* next exercise */ 81 unsigned long long ull; /* time when fme was created */ 82 int id; /* FME id */ 83 struct config *config; /* cooked configuration data */ 84 struct lut *eventtree; /* propagation tree for this FME */ 85 /* 86 * The initial error report that created this FME is kept in 87 * two forms. e0 points to the instance tree node and is used 88 * by fme_eval() as the starting point for the inference 89 * algorithm. e0r is the event handle FMD passed to us when 90 * the ereport first arrived and is used when setting timers, 91 * which are always relative to the time of this initial 92 * report. 93 */ 94 struct event *e0; 95 fmd_event_t *e0r; 96 97 id_t timer; /* for setting an fmd time-out */ 98 99 struct event *ecurrent; /* ereport under consideration */ 100 struct event *suspects; /* current suspect list */ 101 struct event *psuspects; /* previous suspect list */ 102 int nsuspects; /* count of suspects */ 103 int posted_suspects; /* true if we've posted a diagnosis */ 104 int uniqobs; /* number of unique events observed */ 105 int peek; /* just peeking, don't track suspects */ 106 int overflow; /* true if overflow FME */ 107 enum fme_state { 108 FME_NOTHING = 5000, /* not evaluated yet */ 109 FME_WAIT, /* need to wait for more info */ 110 FME_CREDIBLE, /* suspect list is credible */ 111 FME_DISPROVED, /* no valid suspects found */ 112 FME_DEFERRED /* don't know yet (k-count not met) */ 113 } state; 114 115 unsigned long long pull; /* time passed since created */ 116 unsigned long long wull; /* wait until this time for re-eval */ 117 struct event *observations; /* observation list */ 118 struct lut *globals; /* values of global variables */ 119 /* fmd interfacing */ 120 fmd_hdl_t *hdl; /* handle for talking with fmd */ 121 fmd_case_t *fmcase; /* what fmd 'case' we associate with */ 122 /* stats */ 123 struct stats *Rcount; 124 struct stats *Hcallcount; 125 struct stats *Rcallcount; 126 struct stats *Ccallcount; 127 struct stats *Ecallcount; 128 struct stats *Tcallcount; 129 struct stats *Marrowcount; 130 struct stats *diags; 131 } *FMElist, *EFMElist, *ClosedFMEs; 132 133 static struct case_list { 134 fmd_case_t *fmcase; 135 struct case_list *next; 136 } *Undiagablecaselist; 137 138 static void fme_eval(struct fme *fmep, fmd_event_t *ffep); 139 static enum fme_state hypothesise(struct fme *fmep, struct event *ep, 140 unsigned long long at_latest_by, unsigned long long *pdelay); 141 static struct node *eventprop_lookup(struct event *ep, const char *propname); 142 static struct node *pathstring2epnamenp(char *path); 143 static void publish_undiagnosable(fmd_hdl_t *hdl, fmd_event_t *ffep, 144 fmd_case_t *fmcase, nvlist_t *detector, char *arg); 145 static char *undiag_2reason_str(int ud, char *arg); 146 static const char *undiag_2defect_str(int ud); 147 static void restore_suspects(struct fme *fmep); 148 static void save_suspects(struct fme *fmep); 149 static void destroy_fme(struct fme *f); 150 static void fme_receive_report(fmd_hdl_t *hdl, fmd_event_t *ffep, 151 const char *eventstring, const struct ipath *ipp, nvlist_t *nvl); 152 static void istat_counter_reset_cb(struct istat_entry *entp, 153 struct stats *statp, const struct ipath *ipp); 154 static void istat_counter_topo_chg_cb(struct istat_entry *entp, 155 struct stats *statp, void *unused); 156 static void serd_reset_cb(struct serd_entry *entp, void *unused, 157 const struct ipath *ipp); 158 static void serd_topo_chg_cb(struct serd_entry *entp, void *unused, 159 void *unused2); 160 static void destroy_fme_bufs(struct fme *fp); 161 162 static struct fme * 163 alloc_fme(void) 164 { 165 struct fme *fmep; 166 167 fmep = MALLOC(sizeof (*fmep)); 168 bzero(fmep, sizeof (*fmep)); 169 return (fmep); 170 } 171 172 /* 173 * fme_ready -- called when all initialization of the FME (except for 174 * stats) has completed successfully. Adds the fme to global lists 175 * and establishes its stats. 176 */ 177 static struct fme * 178 fme_ready(struct fme *fmep) 179 { 180 char nbuf[100]; 181 182 Nfmep = NULL; /* don't need to free this on module abort now */ 183 184 if (EFMElist) { 185 EFMElist->next = fmep; 186 EFMElist = fmep; 187 } else 188 FMElist = EFMElist = fmep; 189 190 (void) sprintf(nbuf, "fme%d.Rcount", fmep->id); 191 fmep->Rcount = stats_new_counter(nbuf, "ereports received", 0); 192 (void) sprintf(nbuf, "fme%d.Hcall", fmep->id); 193 fmep->Hcallcount = stats_new_counter(nbuf, "calls to hypothesise()", 1); 194 (void) sprintf(nbuf, "fme%d.Rcall", fmep->id); 195 fmep->Rcallcount = stats_new_counter(nbuf, 196 "calls to requirements_test()", 1); 197 (void) sprintf(nbuf, "fme%d.Ccall", fmep->id); 198 fmep->Ccallcount = stats_new_counter(nbuf, "calls to causes_test()", 1); 199 (void) sprintf(nbuf, "fme%d.Ecall", fmep->id); 200 fmep->Ecallcount = 201 stats_new_counter(nbuf, "calls to effects_test()", 1); 202 (void) sprintf(nbuf, "fme%d.Tcall", fmep->id); 203 fmep->Tcallcount = stats_new_counter(nbuf, "calls to triggered()", 1); 204 (void) sprintf(nbuf, "fme%d.Marrow", fmep->id); 205 fmep->Marrowcount = stats_new_counter(nbuf, 206 "arrows marked by mark_arrows()", 1); 207 (void) sprintf(nbuf, "fme%d.diags", fmep->id); 208 fmep->diags = stats_new_counter(nbuf, "suspect lists diagnosed", 0); 209 210 out(O_ALTFP|O_VERB2, "newfme: config snapshot contains..."); 211 config_print(O_ALTFP|O_VERB2, fmep->config); 212 213 return (fmep); 214 } 215 216 extern void ipath_dummy_lut(struct arrow *); 217 extern struct lut *itree_create_dummy(const char *, const struct ipath *); 218 219 /* ARGSUSED */ 220 static void 221 set_needed_arrows(struct event *ep, struct event *ep2, struct fme *fmep) 222 { 223 struct bubble *bp; 224 struct arrowlist *ap; 225 226 for (bp = itree_next_bubble(ep, NULL); bp; 227 bp = itree_next_bubble(ep, bp)) { 228 if (bp->t != B_FROM) 229 continue; 230 for (ap = itree_next_arrow(bp, NULL); ap; 231 ap = itree_next_arrow(bp, ap)) { 232 ap->arrowp->pnode->u.arrow.needed = 1; 233 ipath_dummy_lut(ap->arrowp); 234 } 235 } 236 } 237 238 /* ARGSUSED */ 239 static void 240 unset_needed_arrows(struct event *ep, struct event *ep2, struct fme *fmep) 241 { 242 struct bubble *bp; 243 struct arrowlist *ap; 244 245 for (bp = itree_next_bubble(ep, NULL); bp; 246 bp = itree_next_bubble(ep, bp)) { 247 if (bp->t != B_FROM) 248 continue; 249 for (ap = itree_next_arrow(bp, NULL); ap; 250 ap = itree_next_arrow(bp, ap)) 251 ap->arrowp->pnode->u.arrow.needed = 0; 252 } 253 } 254 255 static void globals_destructor(void *left, void *right, void *arg); 256 static void clear_arrows(struct event *ep, struct event *ep2, struct fme *fmep); 257 258 static boolean_t 259 prune_propagations(const char *e0class, const struct ipath *e0ipp) 260 { 261 char nbuf[100]; 262 unsigned long long my_delay = TIMEVAL_EVENTUALLY; 263 extern struct lut *Usednames; 264 265 Nfmep = alloc_fme(); 266 Nfmep->id = Nextid; 267 Nfmep->state = FME_NOTHING; 268 Nfmep->eventtree = itree_create_dummy(e0class, e0ipp); 269 if ((Nfmep->e0 = 270 itree_lookup(Nfmep->eventtree, e0class, e0ipp)) == NULL) { 271 itree_free(Nfmep->eventtree); 272 FREE(Nfmep); 273 Nfmep = NULL; 274 return (B_FALSE); 275 } 276 Nfmep->ecurrent = Nfmep->observations = Nfmep->e0; 277 Nfmep->e0->count++; 278 279 (void) sprintf(nbuf, "fme%d.Rcount", Nfmep->id); 280 Nfmep->Rcount = stats_new_counter(nbuf, "ereports received", 0); 281 (void) sprintf(nbuf, "fme%d.Hcall", Nfmep->id); 282 Nfmep->Hcallcount = 283 stats_new_counter(nbuf, "calls to hypothesise()", 1); 284 (void) sprintf(nbuf, "fme%d.Rcall", Nfmep->id); 285 Nfmep->Rcallcount = stats_new_counter(nbuf, 286 "calls to requirements_test()", 1); 287 (void) sprintf(nbuf, "fme%d.Ccall", Nfmep->id); 288 Nfmep->Ccallcount = 289 stats_new_counter(nbuf, "calls to causes_test()", 1); 290 (void) sprintf(nbuf, "fme%d.Ecall", Nfmep->id); 291 Nfmep->Ecallcount = 292 stats_new_counter(nbuf, "calls to effects_test()", 1); 293 (void) sprintf(nbuf, "fme%d.Tcall", Nfmep->id); 294 Nfmep->Tcallcount = stats_new_counter(nbuf, "calls to triggered()", 1); 295 (void) sprintf(nbuf, "fme%d.Marrow", Nfmep->id); 296 Nfmep->Marrowcount = stats_new_counter(nbuf, 297 "arrows marked by mark_arrows()", 1); 298 (void) sprintf(nbuf, "fme%d.diags", Nfmep->id); 299 Nfmep->diags = stats_new_counter(nbuf, "suspect lists diagnosed", 0); 300 301 Nfmep->peek = 1; 302 lut_walk(Nfmep->eventtree, (lut_cb)unset_needed_arrows, (void *)Nfmep); 303 lut_free(Usednames, NULL, NULL); 304 Usednames = NULL; 305 lut_walk(Nfmep->eventtree, (lut_cb)clear_arrows, (void *)Nfmep); 306 (void) hypothesise(Nfmep, Nfmep->e0, Nfmep->ull, &my_delay); 307 itree_prune(Nfmep->eventtree); 308 lut_walk(Nfmep->eventtree, (lut_cb)set_needed_arrows, (void *)Nfmep); 309 310 stats_delete(Nfmep->Rcount); 311 stats_delete(Nfmep->Hcallcount); 312 stats_delete(Nfmep->Rcallcount); 313 stats_delete(Nfmep->Ccallcount); 314 stats_delete(Nfmep->Ecallcount); 315 stats_delete(Nfmep->Tcallcount); 316 stats_delete(Nfmep->Marrowcount); 317 stats_delete(Nfmep->diags); 318 itree_free(Nfmep->eventtree); 319 lut_free(Nfmep->globals, globals_destructor, NULL); 320 FREE(Nfmep); 321 return (B_TRUE); 322 } 323 324 static struct fme * 325 newfme(const char *e0class, const struct ipath *e0ipp, fmd_hdl_t *hdl, 326 fmd_case_t *fmcase, fmd_event_t *ffep, nvlist_t *nvl) 327 { 328 struct cfgdata *cfgdata; 329 int init_size; 330 extern int alloc_total(); 331 nvlist_t *detector = NULL; 332 char *pathstr; 333 char *arg; 334 335 /* 336 * First check if e0ipp is actually in the topology so we can give a 337 * more useful error message. 338 */ 339 ipathlastcomp(e0ipp); 340 pathstr = ipath2str(NULL, e0ipp); 341 cfgdata = config_snapshot(); 342 platform_units_translate(0, cfgdata->cooked, NULL, NULL, 343 &detector, pathstr); 344 FREE(pathstr); 345 structconfig_free(cfgdata->cooked); 346 config_free(cfgdata); 347 if (detector == NULL) { 348 /* See if class permits silent discard on unknown component. */ 349 if (lut_lookup(Ereportenames_discard, (void *)e0class, NULL)) { 350 out(O_ALTFP|O_VERB2, "Unable to map \"%s\" ereport " 351 "to component path, but silent discard allowed.", 352 e0class); 353 } else { 354 Undiag_reason = UD_VAL_BADEVENTPATH; 355 (void) nvlist_lookup_nvlist(nvl, FM_EREPORT_DETECTOR, 356 &detector); 357 arg = ipath2str(e0class, e0ipp); 358 publish_undiagnosable(hdl, ffep, fmcase, detector, arg); 359 FREE(arg); 360 } 361 return (NULL); 362 } 363 364 /* 365 * Next run a quick first pass of the rules with a dummy config. This 366 * allows us to prune those rules which can't possibly cause this 367 * ereport. 368 */ 369 if (!prune_propagations(e0class, e0ipp)) { 370 /* 371 * The fault class must have been in the rules or we would 372 * not have registered for it (and got a "nosub"), and the 373 * pathname must be in the topology or we would have failed the 374 * previous test. So to get here means the combination of 375 * class and pathname in the ereport must be invalid. 376 */ 377 Undiag_reason = UD_VAL_BADEVENTCLASS; 378 arg = ipath2str(e0class, e0ipp); 379 publish_undiagnosable(hdl, ffep, fmcase, detector, arg); 380 nvlist_free(detector); 381 FREE(arg); 382 return (NULL); 383 } 384 385 /* 386 * Now go ahead and create the real fme using the pruned rules. 387 */ 388 init_size = alloc_total(); 389 out(O_ALTFP|O_STAMP, "start config_snapshot using %d bytes", init_size); 390 nvlist_free(detector); 391 pathstr = ipath2str(NULL, e0ipp); 392 cfgdata = config_snapshot(); 393 platform_units_translate(0, cfgdata->cooked, NULL, NULL, 394 &detector, pathstr); 395 FREE(pathstr); 396 platform_save_config(hdl, fmcase); 397 out(O_ALTFP|O_STAMP, "config_snapshot added %d bytes", 398 alloc_total() - init_size); 399 400 Nfmep = alloc_fme(); 401 402 Nfmep->id = Nextid++; 403 Nfmep->config = cfgdata->cooked; 404 config_free(cfgdata); 405 Nfmep->posted_suspects = 0; 406 Nfmep->uniqobs = 0; 407 Nfmep->state = FME_NOTHING; 408 Nfmep->pull = 0ULL; 409 Nfmep->overflow = 0; 410 411 Nfmep->fmcase = fmcase; 412 Nfmep->hdl = hdl; 413 414 if ((Nfmep->eventtree = itree_create(Nfmep->config)) == NULL) { 415 Undiag_reason = UD_VAL_INSTFAIL; 416 arg = ipath2str(e0class, e0ipp); 417 publish_undiagnosable(hdl, ffep, fmcase, detector, arg); 418 nvlist_free(detector); 419 FREE(arg); 420 structconfig_free(Nfmep->config); 421 destroy_fme_bufs(Nfmep); 422 FREE(Nfmep); 423 Nfmep = NULL; 424 return (NULL); 425 } 426 427 itree_ptree(O_ALTFP|O_VERB2, Nfmep->eventtree); 428 429 if ((Nfmep->e0 = 430 itree_lookup(Nfmep->eventtree, e0class, e0ipp)) == NULL) { 431 Undiag_reason = UD_VAL_BADEVENTI; 432 arg = ipath2str(e0class, e0ipp); 433 publish_undiagnosable(hdl, ffep, fmcase, detector, arg); 434 nvlist_free(detector); 435 FREE(arg); 436 itree_free(Nfmep->eventtree); 437 structconfig_free(Nfmep->config); 438 destroy_fme_bufs(Nfmep); 439 FREE(Nfmep); 440 Nfmep = NULL; 441 return (NULL); 442 } 443 444 nvlist_free(detector); 445 return (fme_ready(Nfmep)); 446 } 447 448 void 449 fme_fini(void) 450 { 451 struct fme *sfp, *fp; 452 struct case_list *ucasep, *nextcasep; 453 454 ucasep = Undiagablecaselist; 455 while (ucasep != NULL) { 456 nextcasep = ucasep->next; 457 FREE(ucasep); 458 ucasep = nextcasep; 459 } 460 Undiagablecaselist = NULL; 461 462 /* clean up closed fmes */ 463 fp = ClosedFMEs; 464 while (fp != NULL) { 465 sfp = fp->next; 466 destroy_fme(fp); 467 fp = sfp; 468 } 469 ClosedFMEs = NULL; 470 471 fp = FMElist; 472 while (fp != NULL) { 473 sfp = fp->next; 474 destroy_fme(fp); 475 fp = sfp; 476 } 477 FMElist = EFMElist = NULL; 478 479 /* if we were in the middle of creating an fme, free it now */ 480 if (Nfmep) { 481 destroy_fme(Nfmep); 482 Nfmep = NULL; 483 } 484 } 485 486 /* 487 * Allocated space for a buffer name. 20 bytes allows for 488 * a ridiculous 9,999,999 unique observations. 489 */ 490 #define OBBUFNMSZ 20 491 492 /* 493 * serialize_observation 494 * 495 * Create a recoverable version of the current observation 496 * (f->ecurrent). We keep a serialized version of each unique 497 * observation in order that we may resume correctly the fme in the 498 * correct state if eft or fmd crashes and we're restarted. 499 */ 500 static void 501 serialize_observation(struct fme *fp, const char *cls, const struct ipath *ipp) 502 { 503 size_t pkdlen; 504 char tmpbuf[OBBUFNMSZ]; 505 char *pkd = NULL; 506 char *estr; 507 508 (void) snprintf(tmpbuf, OBBUFNMSZ, "observed%d", fp->uniqobs); 509 estr = ipath2str(cls, ipp); 510 fmd_buf_create(fp->hdl, fp->fmcase, tmpbuf, strlen(estr) + 1); 511 fmd_buf_write(fp->hdl, fp->fmcase, tmpbuf, (void *)estr, 512 strlen(estr) + 1); 513 FREE(estr); 514 515 if (fp->ecurrent != NULL && fp->ecurrent->nvp != NULL) { 516 (void) snprintf(tmpbuf, 517 OBBUFNMSZ, "observed%d.nvp", fp->uniqobs); 518 if (nvlist_xpack(fp->ecurrent->nvp, 519 &pkd, &pkdlen, NV_ENCODE_XDR, &Eft_nv_hdl) != 0) 520 out(O_DIE|O_SYS, "pack of observed nvl failed"); 521 fmd_buf_create(fp->hdl, fp->fmcase, tmpbuf, pkdlen); 522 fmd_buf_write(fp->hdl, fp->fmcase, tmpbuf, (void *)pkd, pkdlen); 523 FREE(pkd); 524 } 525 526 fp->uniqobs++; 527 fmd_buf_write(fp->hdl, fp->fmcase, WOBUF_NOBS, (void *)&fp->uniqobs, 528 sizeof (fp->uniqobs)); 529 } 530 531 /* 532 * init_fme_bufs -- We keep several bits of state about an fme for 533 * use if eft or fmd crashes and we're restarted. 534 */ 535 static void 536 init_fme_bufs(struct fme *fp) 537 { 538 fmd_buf_create(fp->hdl, fp->fmcase, WOBUF_PULL, sizeof (fp->pull)); 539 fmd_buf_write(fp->hdl, fp->fmcase, WOBUF_PULL, (void *)&fp->pull, 540 sizeof (fp->pull)); 541 542 fmd_buf_create(fp->hdl, fp->fmcase, WOBUF_ID, sizeof (fp->id)); 543 fmd_buf_write(fp->hdl, fp->fmcase, WOBUF_ID, (void *)&fp->id, 544 sizeof (fp->id)); 545 546 fmd_buf_create(fp->hdl, fp->fmcase, WOBUF_NOBS, sizeof (fp->uniqobs)); 547 fmd_buf_write(fp->hdl, fp->fmcase, WOBUF_NOBS, (void *)&fp->uniqobs, 548 sizeof (fp->uniqobs)); 549 550 fmd_buf_create(fp->hdl, fp->fmcase, WOBUF_POSTD, 551 sizeof (fp->posted_suspects)); 552 fmd_buf_write(fp->hdl, fp->fmcase, WOBUF_POSTD, 553 (void *)&fp->posted_suspects, sizeof (fp->posted_suspects)); 554 } 555 556 static void 557 destroy_fme_bufs(struct fme *fp) 558 { 559 char tmpbuf[OBBUFNMSZ]; 560 int o; 561 562 platform_restore_config(fp->hdl, fp->fmcase); 563 fmd_buf_destroy(fp->hdl, fp->fmcase, WOBUF_CFGLEN); 564 fmd_buf_destroy(fp->hdl, fp->fmcase, WOBUF_CFG); 565 fmd_buf_destroy(fp->hdl, fp->fmcase, WOBUF_PULL); 566 fmd_buf_destroy(fp->hdl, fp->fmcase, WOBUF_ID); 567 fmd_buf_destroy(fp->hdl, fp->fmcase, WOBUF_POSTD); 568 fmd_buf_destroy(fp->hdl, fp->fmcase, WOBUF_NOBS); 569 570 for (o = 0; o < fp->uniqobs; o++) { 571 (void) snprintf(tmpbuf, OBBUFNMSZ, "observed%d", o); 572 fmd_buf_destroy(fp->hdl, fp->fmcase, tmpbuf); 573 (void) snprintf(tmpbuf, OBBUFNMSZ, "observed%d.nvp", o); 574 fmd_buf_destroy(fp->hdl, fp->fmcase, tmpbuf); 575 } 576 } 577 578 /* 579 * reconstitute_observations -- convert a case's serialized observations 580 * back into struct events. Returns zero if all observations are 581 * successfully reconstituted. 582 */ 583 static int 584 reconstitute_observations(struct fme *fmep) 585 { 586 struct event *ep; 587 struct node *epnamenp = NULL; 588 size_t pkdlen; 589 char *pkd = NULL; 590 char *tmpbuf = alloca(OBBUFNMSZ); 591 char *sepptr; 592 char *estr; 593 int ocnt; 594 int elen; 595 596 for (ocnt = 0; ocnt < fmep->uniqobs; ocnt++) { 597 (void) snprintf(tmpbuf, OBBUFNMSZ, "observed%d", ocnt); 598 elen = fmd_buf_size(fmep->hdl, fmep->fmcase, tmpbuf); 599 if (elen == 0) { 600 out(O_ALTFP, 601 "reconstitute_observation: no %s buffer found.", 602 tmpbuf); 603 Undiag_reason = UD_VAL_MISSINGOBS; 604 break; 605 } 606 607 estr = MALLOC(elen); 608 fmd_buf_read(fmep->hdl, fmep->fmcase, tmpbuf, estr, elen); 609 sepptr = strchr(estr, '@'); 610 if (sepptr == NULL) { 611 out(O_ALTFP, 612 "reconstitute_observation: %s: " 613 "missing @ separator in %s.", 614 tmpbuf, estr); 615 Undiag_reason = UD_VAL_MISSINGPATH; 616 FREE(estr); 617 break; 618 } 619 620 *sepptr = '\0'; 621 if ((epnamenp = pathstring2epnamenp(sepptr + 1)) == NULL) { 622 out(O_ALTFP, 623 "reconstitute_observation: %s: " 624 "trouble converting path string \"%s\" " 625 "to internal representation.", 626 tmpbuf, sepptr + 1); 627 Undiag_reason = UD_VAL_MISSINGPATH; 628 FREE(estr); 629 break; 630 } 631 632 /* construct the event */ 633 ep = itree_lookup(fmep->eventtree, 634 stable(estr), ipath(epnamenp)); 635 if (ep == NULL) { 636 out(O_ALTFP, 637 "reconstitute_observation: %s: " 638 "lookup of \"%s\" in itree failed.", 639 tmpbuf, ipath2str(estr, ipath(epnamenp))); 640 Undiag_reason = UD_VAL_BADOBS; 641 tree_free(epnamenp); 642 FREE(estr); 643 break; 644 } 645 tree_free(epnamenp); 646 647 /* 648 * We may or may not have a saved nvlist for the observation 649 */ 650 (void) snprintf(tmpbuf, OBBUFNMSZ, "observed%d.nvp", ocnt); 651 pkdlen = fmd_buf_size(fmep->hdl, fmep->fmcase, tmpbuf); 652 if (pkdlen != 0) { 653 pkd = MALLOC(pkdlen); 654 fmd_buf_read(fmep->hdl, 655 fmep->fmcase, tmpbuf, pkd, pkdlen); 656 ASSERT(ep->nvp == NULL); 657 if (nvlist_xunpack(pkd, 658 pkdlen, &ep->nvp, &Eft_nv_hdl) != 0) 659 out(O_DIE|O_SYS, "pack of observed nvl failed"); 660 FREE(pkd); 661 } 662 663 if (ocnt == 0) 664 fmep->e0 = ep; 665 666 FREE(estr); 667 fmep->ecurrent = ep; 668 ep->count++; 669 670 /* link it into list of observations seen */ 671 ep->observations = fmep->observations; 672 fmep->observations = ep; 673 } 674 675 if (ocnt == fmep->uniqobs) { 676 (void) fme_ready(fmep); 677 return (0); 678 } 679 680 return (1); 681 } 682 683 /* 684 * restart_fme -- called during eft initialization. Reconstitutes 685 * an in-progress fme. 686 */ 687 void 688 fme_restart(fmd_hdl_t *hdl, fmd_case_t *inprogress) 689 { 690 nvlist_t *defect; 691 struct case_list *bad; 692 struct fme *fmep; 693 struct cfgdata *cfgdata; 694 size_t rawsz; 695 struct event *ep; 696 char *tmpbuf = alloca(OBBUFNMSZ); 697 char *sepptr; 698 char *estr; 699 int elen; 700 struct node *epnamenp = NULL; 701 int init_size; 702 extern int alloc_total(); 703 char *reason; 704 705 /* 706 * ignore solved or closed cases 707 */ 708 if (fmd_case_solved(hdl, inprogress) || 709 fmd_case_closed(hdl, inprogress)) 710 return; 711 712 fmep = alloc_fme(); 713 fmep->fmcase = inprogress; 714 fmep->hdl = hdl; 715 716 if (fmd_buf_size(hdl, inprogress, WOBUF_POSTD) == 0) { 717 out(O_ALTFP, "restart_fme: no saved posted status"); 718 Undiag_reason = UD_VAL_MISSINGINFO; 719 goto badcase; 720 } else { 721 fmd_buf_read(hdl, inprogress, WOBUF_POSTD, 722 (void *)&fmep->posted_suspects, 723 sizeof (fmep->posted_suspects)); 724 } 725 726 if (fmd_buf_size(hdl, inprogress, WOBUF_ID) == 0) { 727 out(O_ALTFP, "restart_fme: no saved id"); 728 Undiag_reason = UD_VAL_MISSINGINFO; 729 goto badcase; 730 } else { 731 fmd_buf_read(hdl, inprogress, WOBUF_ID, (void *)&fmep->id, 732 sizeof (fmep->id)); 733 } 734 if (Nextid <= fmep->id) 735 Nextid = fmep->id + 1; 736 737 out(O_ALTFP, "Replay FME %d", fmep->id); 738 739 if (fmd_buf_size(hdl, inprogress, WOBUF_CFGLEN) != sizeof (size_t)) { 740 out(O_ALTFP, "restart_fme: No config data"); 741 Undiag_reason = UD_VAL_MISSINGINFO; 742 goto badcase; 743 } 744 fmd_buf_read(hdl, inprogress, WOBUF_CFGLEN, (void *)&rawsz, 745 sizeof (size_t)); 746 747 if ((fmep->e0r = fmd_case_getprincipal(hdl, inprogress)) == NULL) { 748 out(O_ALTFP, "restart_fme: No event zero"); 749 Undiag_reason = UD_VAL_MISSINGZERO; 750 goto badcase; 751 } 752 753 if (fmd_buf_size(hdl, inprogress, WOBUF_PULL) == 0) { 754 out(O_ALTFP, "restart_fme: no saved wait time"); 755 Undiag_reason = UD_VAL_MISSINGINFO; 756 goto badcase; 757 } else { 758 fmd_buf_read(hdl, inprogress, WOBUF_PULL, (void *)&fmep->pull, 759 sizeof (fmep->pull)); 760 } 761 762 if (fmd_buf_size(hdl, inprogress, WOBUF_NOBS) == 0) { 763 out(O_ALTFP, "restart_fme: no count of observations"); 764 Undiag_reason = UD_VAL_MISSINGINFO; 765 goto badcase; 766 } else { 767 fmd_buf_read(hdl, inprogress, WOBUF_NOBS, 768 (void *)&fmep->uniqobs, sizeof (fmep->uniqobs)); 769 } 770 771 (void) snprintf(tmpbuf, OBBUFNMSZ, "observed0"); 772 elen = fmd_buf_size(fmep->hdl, fmep->fmcase, tmpbuf); 773 if (elen == 0) { 774 out(O_ALTFP, "reconstitute_observation: no %s buffer found.", 775 tmpbuf); 776 Undiag_reason = UD_VAL_MISSINGOBS; 777 goto badcase; 778 } 779 estr = MALLOC(elen); 780 fmd_buf_read(fmep->hdl, fmep->fmcase, tmpbuf, estr, elen); 781 sepptr = strchr(estr, '@'); 782 if (sepptr == NULL) { 783 out(O_ALTFP, "reconstitute_observation: %s: " 784 "missing @ separator in %s.", 785 tmpbuf, estr); 786 Undiag_reason = UD_VAL_MISSINGPATH; 787 FREE(estr); 788 goto badcase; 789 } 790 *sepptr = '\0'; 791 if ((epnamenp = pathstring2epnamenp(sepptr + 1)) == NULL) { 792 out(O_ALTFP, "reconstitute_observation: %s: " 793 "trouble converting path string \"%s\" " 794 "to internal representation.", tmpbuf, sepptr + 1); 795 Undiag_reason = UD_VAL_MISSINGPATH; 796 FREE(estr); 797 goto badcase; 798 } 799 (void) prune_propagations(stable(estr), ipath(epnamenp)); 800 tree_free(epnamenp); 801 FREE(estr); 802 803 init_size = alloc_total(); 804 out(O_ALTFP|O_STAMP, "start config_restore using %d bytes", init_size); 805 cfgdata = MALLOC(sizeof (struct cfgdata)); 806 cfgdata->cooked = NULL; 807 cfgdata->devcache = NULL; 808 cfgdata->devidcache = NULL; 809 cfgdata->tpcache = NULL; 810 cfgdata->cpucache = NULL; 811 cfgdata->raw_refcnt = 1; 812 813 if (rawsz > 0) { 814 if (fmd_buf_size(hdl, inprogress, WOBUF_CFG) != rawsz) { 815 out(O_ALTFP, "restart_fme: Config data size mismatch"); 816 Undiag_reason = UD_VAL_CFGMISMATCH; 817 goto badcase; 818 } 819 cfgdata->begin = MALLOC(rawsz); 820 cfgdata->end = cfgdata->nextfree = cfgdata->begin + rawsz; 821 fmd_buf_read(hdl, 822 inprogress, WOBUF_CFG, cfgdata->begin, rawsz); 823 } else { 824 cfgdata->begin = cfgdata->end = cfgdata->nextfree = NULL; 825 } 826 827 config_cook(cfgdata); 828 fmep->config = cfgdata->cooked; 829 config_free(cfgdata); 830 out(O_ALTFP|O_STAMP, "config_restore added %d bytes", 831 alloc_total() - init_size); 832 833 if ((fmep->eventtree = itree_create(fmep->config)) == NULL) { 834 /* case not properly saved or irretrievable */ 835 out(O_ALTFP, "restart_fme: NULL instance tree"); 836 Undiag_reason = UD_VAL_INSTFAIL; 837 goto badcase; 838 } 839 840 itree_ptree(O_ALTFP|O_VERB2, fmep->eventtree); 841 842 if (reconstitute_observations(fmep) != 0) 843 goto badcase; 844 845 out(O_ALTFP|O_NONL, "FME %d replay observations: ", fmep->id); 846 for (ep = fmep->observations; ep; ep = ep->observations) { 847 out(O_ALTFP|O_NONL, " "); 848 itree_pevent_brief(O_ALTFP|O_NONL, ep); 849 } 850 out(O_ALTFP, NULL); 851 852 Open_fme_count++; 853 854 /* give the diagnosis algorithm a shot at the new FME state */ 855 fme_eval(fmep, fmep->e0r); 856 return; 857 858 badcase: 859 if (fmep->eventtree != NULL) 860 itree_free(fmep->eventtree); 861 if (fmep->config) 862 structconfig_free(fmep->config); 863 destroy_fme_bufs(fmep); 864 FREE(fmep); 865 866 /* 867 * Since we're unable to restart the case, add it to the undiagable 868 * list and solve and close it as appropriate. 869 */ 870 bad = MALLOC(sizeof (struct case_list)); 871 bad->next = NULL; 872 873 if (Undiagablecaselist != NULL) 874 bad->next = Undiagablecaselist; 875 Undiagablecaselist = bad; 876 bad->fmcase = inprogress; 877 878 out(O_ALTFP|O_NONL, "[case %s (unable to restart), ", 879 fmd_case_uuid(hdl, bad->fmcase)); 880 881 if (fmd_case_solved(hdl, bad->fmcase)) { 882 out(O_ALTFP|O_NONL, "already solved, "); 883 } else { 884 out(O_ALTFP|O_NONL, "solving, "); 885 defect = fmd_nvl_create_fault(hdl, 886 undiag_2defect_str(Undiag_reason), 100, NULL, NULL, NULL); 887 reason = undiag_2reason_str(Undiag_reason, NULL); 888 (void) nvlist_add_string(defect, UNDIAG_REASON, reason); 889 FREE(reason); 890 fmd_case_add_suspect(hdl, bad->fmcase, defect); 891 fmd_case_solve(hdl, bad->fmcase); 892 Undiag_reason = UD_VAL_UNKNOWN; 893 } 894 895 if (fmd_case_closed(hdl, bad->fmcase)) { 896 out(O_ALTFP, "already closed ]"); 897 } else { 898 out(O_ALTFP, "closing ]"); 899 fmd_case_close(hdl, bad->fmcase); 900 } 901 } 902 903 /*ARGSUSED*/ 904 static void 905 globals_destructor(void *left, void *right, void *arg) 906 { 907 struct evalue *evp = (struct evalue *)right; 908 if (evp->t == NODEPTR) 909 tree_free((struct node *)(uintptr_t)evp->v); 910 evp->v = (uintptr_t)NULL; 911 FREE(evp); 912 } 913 914 void 915 destroy_fme(struct fme *f) 916 { 917 stats_delete(f->Rcount); 918 stats_delete(f->Hcallcount); 919 stats_delete(f->Rcallcount); 920 stats_delete(f->Ccallcount); 921 stats_delete(f->Ecallcount); 922 stats_delete(f->Tcallcount); 923 stats_delete(f->Marrowcount); 924 stats_delete(f->diags); 925 926 if (f->eventtree != NULL) 927 itree_free(f->eventtree); 928 if (f->config) 929 structconfig_free(f->config); 930 lut_free(f->globals, globals_destructor, NULL); 931 FREE(f); 932 } 933 934 static const char * 935 fme_state2str(enum fme_state s) 936 { 937 switch (s) { 938 case FME_NOTHING: return ("NOTHING"); 939 case FME_WAIT: return ("WAIT"); 940 case FME_CREDIBLE: return ("CREDIBLE"); 941 case FME_DISPROVED: return ("DISPROVED"); 942 case FME_DEFERRED: return ("DEFERRED"); 943 default: return ("UNKNOWN"); 944 } 945 } 946 947 static int 948 is_problem(enum nametype t) 949 { 950 return (t == N_FAULT || t == N_DEFECT || t == N_UPSET); 951 } 952 953 static int 954 is_defect(enum nametype t) 955 { 956 return (t == N_DEFECT); 957 } 958 959 static int 960 is_upset(enum nametype t) 961 { 962 return (t == N_UPSET); 963 } 964 965 static void 966 fme_print(int flags, struct fme *fmep) 967 { 968 struct event *ep; 969 970 out(flags, "Fault Management Exercise %d", fmep->id); 971 out(flags, "\t State: %s", fme_state2str(fmep->state)); 972 out(flags|O_NONL, "\t Start time: "); 973 ptree_timeval(flags|O_NONL, &fmep->ull); 974 out(flags, NULL); 975 if (fmep->wull) { 976 out(flags|O_NONL, "\t Wait time: "); 977 ptree_timeval(flags|O_NONL, &fmep->wull); 978 out(flags, NULL); 979 } 980 out(flags|O_NONL, "\t E0: "); 981 if (fmep->e0) 982 itree_pevent_brief(flags|O_NONL, fmep->e0); 983 else 984 out(flags|O_NONL, "NULL"); 985 out(flags, NULL); 986 out(flags|O_NONL, "\tObservations:"); 987 for (ep = fmep->observations; ep; ep = ep->observations) { 988 out(flags|O_NONL, " "); 989 itree_pevent_brief(flags|O_NONL, ep); 990 } 991 out(flags, NULL); 992 out(flags|O_NONL, "\tSuspect list:"); 993 for (ep = fmep->suspects; ep; ep = ep->suspects) { 994 out(flags|O_NONL, " "); 995 itree_pevent_brief(flags|O_NONL, ep); 996 } 997 out(flags, NULL); 998 if (fmep->eventtree != NULL) { 999 out(flags|O_VERB2, "\t Tree:"); 1000 itree_ptree(flags|O_VERB2, fmep->eventtree); 1001 } 1002 } 1003 1004 static struct node * 1005 pathstring2epnamenp(char *path) 1006 { 1007 char *sep = "/"; 1008 struct node *ret; 1009 char *ptr; 1010 1011 if ((ptr = strtok(path, sep)) == NULL) 1012 out(O_DIE, "pathstring2epnamenp: invalid empty class"); 1013 1014 ret = tree_iname(stable(ptr), NULL, 0); 1015 1016 while ((ptr = strtok(NULL, sep)) != NULL) 1017 ret = tree_name_append(ret, 1018 tree_iname(stable(ptr), NULL, 0)); 1019 1020 return (ret); 1021 } 1022 1023 /* 1024 * for a given upset sp, increment the corresponding SERD engine. if the 1025 * SERD engine trips, return the ename and ipp of the resulting ereport. 1026 * returns true if engine tripped and *enamep and *ippp were filled in. 1027 */ 1028 static int 1029 serd_eval(struct fme *fmep, fmd_hdl_t *hdl, fmd_event_t *ffep, 1030 fmd_case_t *fmcase, struct event *sp, const char **enamep, 1031 const struct ipath **ippp) 1032 { 1033 struct node *serdinst; 1034 char *serdname; 1035 char *serdresource; 1036 char *serdclass; 1037 struct node *nid; 1038 struct serd_entry *newentp; 1039 int i, serdn = -1, serdincrement = 1, len = 0; 1040 char *serdsuffix = NULL, *serdt = NULL; 1041 struct evalue *ep; 1042 1043 ASSERT(sp->t == N_UPSET); 1044 ASSERT(ffep != NULL); 1045 1046 if ((ep = (struct evalue *)lut_lookup(sp->serdprops, 1047 (void *)"n", (lut_cmp)strcmp)) != NULL) { 1048 ASSERT(ep->t == UINT64); 1049 serdn = (int)ep->v; 1050 } 1051 if ((ep = (struct evalue *)lut_lookup(sp->serdprops, 1052 (void *)"t", (lut_cmp)strcmp)) != NULL) { 1053 ASSERT(ep->t == STRING); 1054 serdt = (char *)(uintptr_t)ep->v; 1055 } 1056 if ((ep = (struct evalue *)lut_lookup(sp->serdprops, 1057 (void *)"suffix", (lut_cmp)strcmp)) != NULL) { 1058 ASSERT(ep->t == STRING); 1059 serdsuffix = (char *)(uintptr_t)ep->v; 1060 } 1061 if ((ep = (struct evalue *)lut_lookup(sp->serdprops, 1062 (void *)"increment", (lut_cmp)strcmp)) != NULL) { 1063 ASSERT(ep->t == UINT64); 1064 serdincrement = (int)ep->v; 1065 } 1066 1067 /* 1068 * obtain instanced SERD engine from the upset sp. from this 1069 * derive serdname, the string used to identify the SERD engine. 1070 */ 1071 serdinst = eventprop_lookup(sp, L_engine); 1072 1073 if (serdinst == NULL) 1074 return (-1); 1075 1076 len = strlen(serdinst->u.stmt.np->u.event.ename->u.name.s) + 1; 1077 if (serdsuffix != NULL) 1078 len += strlen(serdsuffix); 1079 serdclass = MALLOC(len); 1080 if (serdsuffix != NULL) 1081 (void) snprintf(serdclass, len, "%s%s", 1082 serdinst->u.stmt.np->u.event.ename->u.name.s, serdsuffix); 1083 else 1084 (void) snprintf(serdclass, len, "%s", 1085 serdinst->u.stmt.np->u.event.ename->u.name.s); 1086 serdresource = ipath2str(NULL, 1087 ipath(serdinst->u.stmt.np->u.event.epname)); 1088 len += strlen(serdresource) + 1; 1089 serdname = MALLOC(len); 1090 (void) snprintf(serdname, len, "%s@%s", serdclass, serdresource); 1091 FREE(serdresource); 1092 1093 /* handle serd engine "id" property, if there is one */ 1094 if ((nid = 1095 lut_lookup(serdinst->u.stmt.lutp, (void *)L_id, NULL)) != NULL) { 1096 struct evalue *gval; 1097 char suffixbuf[200]; 1098 char *suffix; 1099 char *nserdname; 1100 size_t nname; 1101 1102 out(O_ALTFP|O_NONL, "serd \"%s\" id: ", serdname); 1103 ptree_name_iter(O_ALTFP|O_NONL, nid); 1104 1105 ASSERTinfo(nid->t == T_GLOBID, ptree_nodetype2str(nid->t)); 1106 1107 if ((gval = lut_lookup(fmep->globals, 1108 (void *)nid->u.globid.s, NULL)) == NULL) { 1109 out(O_ALTFP, " undefined"); 1110 } else if (gval->t == UINT64) { 1111 out(O_ALTFP, " %llu", gval->v); 1112 (void) sprintf(suffixbuf, "%llu", gval->v); 1113 suffix = suffixbuf; 1114 } else { 1115 out(O_ALTFP, " \"%s\"", (char *)(uintptr_t)gval->v); 1116 suffix = (char *)(uintptr_t)gval->v; 1117 } 1118 1119 nname = strlen(serdname) + strlen(suffix) + 2; 1120 nserdname = MALLOC(nname); 1121 (void) snprintf(nserdname, nname, "%s:%s", serdname, suffix); 1122 FREE(serdname); 1123 serdname = nserdname; 1124 } 1125 1126 /* 1127 * if the engine is empty, and we have an override for n/t then 1128 * destroy and recreate it. 1129 */ 1130 if ((serdn != -1 || serdt != NULL) && fmd_serd_exists(hdl, serdname) && 1131 fmd_serd_empty(hdl, serdname)) 1132 fmd_serd_destroy(hdl, serdname); 1133 1134 if (!fmd_serd_exists(hdl, serdname)) { 1135 struct node *nN, *nT; 1136 const char *s; 1137 struct node *nodep; 1138 struct config *cp; 1139 char *path; 1140 uint_t nval; 1141 hrtime_t tval; 1142 int i; 1143 char *ptr; 1144 int got_n_override = 0, got_t_override = 0; 1145 1146 /* no SERD engine yet, so create it */ 1147 nodep = serdinst->u.stmt.np->u.event.epname; 1148 path = ipath2str(NULL, ipath(nodep)); 1149 cp = config_lookup(fmep->config, path, 0); 1150 FREE((void *)path); 1151 1152 /* 1153 * We allow serd paramaters to be overridden, either from 1154 * eft.conf file values (if Serd_Override is set) or from 1155 * driver properties (for "serd.io.device" engines). 1156 */ 1157 if (Serd_Override != NULL) { 1158 char *save_ptr, *ptr1, *ptr2, *ptr3; 1159 ptr3 = save_ptr = STRDUP(Serd_Override); 1160 while (*ptr3 != '\0') { 1161 ptr1 = strchr(ptr3, ','); 1162 *ptr1 = '\0'; 1163 if (strcmp(ptr3, serdclass) == 0) { 1164 ptr2 = strchr(ptr1 + 1, ','); 1165 *ptr2 = '\0'; 1166 nval = atoi(ptr1 + 1); 1167 out(O_ALTFP, "serd override %s_n %d", 1168 serdclass, nval); 1169 ptr3 = strchr(ptr2 + 1, ' '); 1170 if (ptr3) 1171 *ptr3 = '\0'; 1172 ptr = STRDUP(ptr2 + 1); 1173 out(O_ALTFP, "serd override %s_t %s", 1174 serdclass, ptr); 1175 got_n_override = 1; 1176 got_t_override = 1; 1177 break; 1178 } else { 1179 ptr2 = strchr(ptr1 + 1, ','); 1180 ptr3 = strchr(ptr2 + 1, ' '); 1181 if (ptr3 == NULL) 1182 break; 1183 } 1184 ptr3++; 1185 } 1186 FREE(save_ptr); 1187 } 1188 1189 if (cp && got_n_override == 0) { 1190 /* 1191 * convert serd engine class into property name 1192 */ 1193 char *prop_name = MALLOC(strlen(serdclass) + 3); 1194 for (i = 0; i < strlen(serdclass); i++) { 1195 if (serdclass[i] == '.') 1196 prop_name[i] = '_'; 1197 else 1198 prop_name[i] = serdclass[i]; 1199 } 1200 prop_name[i++] = '_'; 1201 prop_name[i++] = 'n'; 1202 prop_name[i] = '\0'; 1203 if (s = config_getprop(cp, prop_name)) { 1204 nval = atoi(s); 1205 out(O_ALTFP, "serd override %s_n %s", 1206 serdclass, s); 1207 got_n_override = 1; 1208 } 1209 prop_name[i - 1] = 't'; 1210 if (s = config_getprop(cp, prop_name)) { 1211 ptr = STRDUP(s); 1212 out(O_ALTFP, "serd override %s_t %s", 1213 serdclass, s); 1214 got_t_override = 1; 1215 } 1216 FREE(prop_name); 1217 } 1218 1219 if (serdn != -1 && got_n_override == 0) { 1220 nval = serdn; 1221 out(O_ALTFP, "serd override %s_n %d", serdclass, serdn); 1222 got_n_override = 1; 1223 } 1224 if (serdt != NULL && got_t_override == 0) { 1225 ptr = STRDUP(serdt); 1226 out(O_ALTFP, "serd override %s_t %s", serdclass, serdt); 1227 got_t_override = 1; 1228 } 1229 1230 if (!got_n_override) { 1231 nN = lut_lookup(serdinst->u.stmt.lutp, (void *)L_N, 1232 NULL); 1233 ASSERT(nN->t == T_NUM); 1234 nval = (uint_t)nN->u.ull; 1235 } 1236 if (!got_t_override) { 1237 nT = lut_lookup(serdinst->u.stmt.lutp, (void *)L_T, 1238 NULL); 1239 ASSERT(nT->t == T_TIMEVAL); 1240 tval = (hrtime_t)nT->u.ull; 1241 } else { 1242 const unsigned long long *ullp; 1243 const char *suffix; 1244 int len; 1245 1246 len = strspn(ptr, "0123456789"); 1247 suffix = stable(&ptr[len]); 1248 ullp = (unsigned long long *)lut_lookup(Timesuffixlut, 1249 (void *)suffix, NULL); 1250 ptr[len] = '\0'; 1251 tval = strtoull(ptr, NULL, 0) * (ullp ? *ullp : 1ll); 1252 FREE(ptr); 1253 } 1254 fmd_serd_create(hdl, serdname, nval, tval); 1255 } 1256 1257 newentp = MALLOC(sizeof (*newentp)); 1258 newentp->ename = stable(serdclass); 1259 FREE(serdclass); 1260 newentp->ipath = ipath(serdinst->u.stmt.np->u.event.epname); 1261 newentp->hdl = hdl; 1262 if (lut_lookup(SerdEngines, newentp, (lut_cmp)serd_cmp) == NULL) { 1263 SerdEngines = lut_add(SerdEngines, (void *)newentp, 1264 (void *)newentp, (lut_cmp)serd_cmp); 1265 Serd_need_save = 1; 1266 serd_save(); 1267 } else { 1268 FREE(newentp); 1269 } 1270 1271 1272 /* 1273 * increment SERD engine. if engine fires, reset serd 1274 * engine and return trip_strcode if required. 1275 */ 1276 for (i = 0; i < serdincrement; i++) { 1277 if (fmd_serd_record(hdl, serdname, ffep)) { 1278 fmd_case_add_serd(hdl, fmcase, serdname); 1279 fmd_serd_reset(hdl, serdname); 1280 1281 if (ippp) { 1282 struct node *tripinst = 1283 lut_lookup(serdinst->u.stmt.lutp, 1284 (void *)L_trip, NULL); 1285 ASSERT(tripinst != NULL); 1286 *enamep = tripinst->u.event.ename->u.name.s; 1287 *ippp = ipath(tripinst->u.event.epname); 1288 out(O_ALTFP|O_NONL, 1289 "[engine fired: %s, sending: ", serdname); 1290 ipath_print(O_ALTFP|O_NONL, *enamep, *ippp); 1291 out(O_ALTFP, "]"); 1292 } else { 1293 out(O_ALTFP, "[engine fired: %s, no trip]", 1294 serdname); 1295 } 1296 FREE(serdname); 1297 return (1); 1298 } 1299 } 1300 1301 FREE(serdname); 1302 return (0); 1303 } 1304 1305 /* 1306 * search a suspect list for upsets. feed each upset to serd_eval() and 1307 * build up tripped[], an array of ereports produced by the firing of 1308 * any SERD engines. then feed each ereport back into 1309 * fme_receive_report(). 1310 * 1311 * returns ntrip, the number of these ereports produced. 1312 */ 1313 static int 1314 upsets_eval(struct fme *fmep, fmd_event_t *ffep) 1315 { 1316 /* we build an array of tripped ereports that we send ourselves */ 1317 struct { 1318 const char *ename; 1319 const struct ipath *ipp; 1320 } *tripped; 1321 struct event *sp; 1322 int ntrip, nupset, i; 1323 1324 /* 1325 * count the number of upsets to determine the upper limit on 1326 * expected trip ereport strings. remember that one upset can 1327 * lead to at most one ereport. 1328 */ 1329 nupset = 0; 1330 for (sp = fmep->suspects; sp; sp = sp->suspects) { 1331 if (sp->t == N_UPSET) 1332 nupset++; 1333 } 1334 1335 if (nupset == 0) 1336 return (0); 1337 1338 /* 1339 * get to this point if we have upsets and expect some trip 1340 * ereports 1341 */ 1342 tripped = alloca(sizeof (*tripped) * nupset); 1343 bzero((void *)tripped, sizeof (*tripped) * nupset); 1344 1345 ntrip = 0; 1346 for (sp = fmep->suspects; sp; sp = sp->suspects) 1347 if (sp->t == N_UPSET && 1348 serd_eval(fmep, fmep->hdl, ffep, fmep->fmcase, sp, 1349 &tripped[ntrip].ename, &tripped[ntrip].ipp) == 1) 1350 ntrip++; 1351 1352 for (i = 0; i < ntrip; i++) { 1353 struct event *ep, *nep; 1354 struct fme *nfmep; 1355 fmd_case_t *fmcase; 1356 const struct ipath *ipp; 1357 const char *eventstring; 1358 int prev_verbose; 1359 unsigned long long my_delay = TIMEVAL_EVENTUALLY; 1360 enum fme_state state; 1361 1362 /* 1363 * First try and evaluate a case with the trip ereport plus 1364 * all the other ereports that cause the trip. If that fails 1365 * to evaluate then try again with just this ereport on its own. 1366 */ 1367 out(O_ALTFP|O_NONL, "fme_receive_report_serd: "); 1368 ipath_print(O_ALTFP|O_NONL, tripped[i].ename, tripped[i].ipp); 1369 out(O_ALTFP|O_STAMP, NULL); 1370 ep = fmep->e0; 1371 eventstring = ep->enode->u.event.ename->u.name.s; 1372 ipp = ep->ipp; 1373 1374 /* 1375 * create a duplicate fme and case 1376 */ 1377 fmcase = fmd_case_open(fmep->hdl, NULL); 1378 out(O_ALTFP|O_NONL, "duplicate fme for event ["); 1379 ipath_print(O_ALTFP|O_NONL, eventstring, ipp); 1380 out(O_ALTFP, " ]"); 1381 1382 if ((nfmep = newfme(eventstring, ipp, fmep->hdl, 1383 fmcase, ffep, ep->nvp)) == NULL) { 1384 out(O_ALTFP|O_NONL, "["); 1385 ipath_print(O_ALTFP|O_NONL, eventstring, ipp); 1386 out(O_ALTFP, " CANNOT DIAGNOSE]"); 1387 continue; 1388 } 1389 1390 Open_fme_count++; 1391 nfmep->pull = fmep->pull; 1392 init_fme_bufs(nfmep); 1393 out(O_ALTFP|O_NONL, "["); 1394 ipath_print(O_ALTFP|O_NONL, eventstring, ipp); 1395 out(O_ALTFP, " created FME%d, case %s]", nfmep->id, 1396 fmd_case_uuid(nfmep->hdl, nfmep->fmcase)); 1397 if (ffep) { 1398 fmd_case_setprincipal(nfmep->hdl, nfmep->fmcase, ffep); 1399 fmd_case_add_ereport(nfmep->hdl, nfmep->fmcase, ffep); 1400 nfmep->e0r = ffep; 1401 } 1402 1403 /* 1404 * add the original ereports 1405 */ 1406 for (ep = fmep->observations; ep; ep = ep->observations) { 1407 eventstring = ep->enode->u.event.ename->u.name.s; 1408 ipp = ep->ipp; 1409 out(O_ALTFP|O_NONL, "adding event ["); 1410 ipath_print(O_ALTFP|O_NONL, eventstring, ipp); 1411 out(O_ALTFP, " ]"); 1412 nep = itree_lookup(nfmep->eventtree, eventstring, ipp); 1413 if (nep->count++ == 0) { 1414 nep->observations = nfmep->observations; 1415 nfmep->observations = nep; 1416 serialize_observation(nfmep, eventstring, ipp); 1417 nep->nvp = evnv_dupnvl(ep->nvp); 1418 } 1419 if (ep->ffep && ep->ffep != ffep) 1420 fmd_case_add_ereport(nfmep->hdl, nfmep->fmcase, 1421 ep->ffep); 1422 stats_counter_bump(nfmep->Rcount); 1423 } 1424 1425 /* 1426 * add the serd trigger ereport 1427 */ 1428 if ((ep = itree_lookup(nfmep->eventtree, tripped[i].ename, 1429 tripped[i].ipp)) == NULL) { 1430 /* 1431 * The trigger ereport is not in the instance tree. It 1432 * was presumably removed by prune_propagations() as 1433 * this combination of events is not present in the 1434 * rules. 1435 */ 1436 out(O_ALTFP, "upsets_eval: e0 not in instance tree"); 1437 Undiag_reason = UD_VAL_BADEVENTI; 1438 goto retry_lone_ereport; 1439 } 1440 out(O_ALTFP|O_NONL, "adding event ["); 1441 ipath_print(O_ALTFP|O_NONL, tripped[i].ename, tripped[i].ipp); 1442 out(O_ALTFP, " ]"); 1443 nfmep->ecurrent = ep; 1444 ep->nvp = NULL; 1445 ep->count = 1; 1446 ep->observations = nfmep->observations; 1447 nfmep->observations = ep; 1448 1449 /* 1450 * just peek first. 1451 */ 1452 nfmep->peek = 1; 1453 prev_verbose = Verbose; 1454 if (Debug == 0) 1455 Verbose = 0; 1456 lut_walk(nfmep->eventtree, (lut_cb)clear_arrows, (void *)nfmep); 1457 state = hypothesise(nfmep, nfmep->e0, nfmep->ull, &my_delay); 1458 nfmep->peek = 0; 1459 Verbose = prev_verbose; 1460 if (state == FME_DISPROVED) { 1461 out(O_ALTFP, "upsets_eval: hypothesis disproved"); 1462 Undiag_reason = UD_VAL_UNSOLVD; 1463 retry_lone_ereport: 1464 /* 1465 * However the trigger ereport on its own might be 1466 * diagnosable, so check for that. Undo the new fme 1467 * and case we just created and call fme_receive_report. 1468 */ 1469 out(O_ALTFP|O_NONL, "["); 1470 ipath_print(O_ALTFP|O_NONL, tripped[i].ename, 1471 tripped[i].ipp); 1472 out(O_ALTFP, " retrying with just trigger ereport]"); 1473 itree_free(nfmep->eventtree); 1474 nfmep->eventtree = NULL; 1475 structconfig_free(nfmep->config); 1476 nfmep->config = NULL; 1477 destroy_fme_bufs(nfmep); 1478 fmd_case_close(nfmep->hdl, nfmep->fmcase); 1479 fme_receive_report(fmep->hdl, ffep, 1480 tripped[i].ename, tripped[i].ipp, NULL); 1481 continue; 1482 } 1483 1484 /* 1485 * and evaluate 1486 */ 1487 serialize_observation(nfmep, tripped[i].ename, tripped[i].ipp); 1488 fme_eval(nfmep, ffep); 1489 } 1490 1491 return (ntrip); 1492 } 1493 1494 /* 1495 * fme_receive_external_report -- call when an external ereport comes in 1496 * 1497 * this routine just converts the relevant information from the ereport 1498 * into a format used internally and passes it on to fme_receive_report(). 1499 */ 1500 void 1501 fme_receive_external_report(fmd_hdl_t *hdl, fmd_event_t *ffep, nvlist_t *nvl, 1502 const char *class) 1503 { 1504 struct node *epnamenp; 1505 fmd_case_t *fmcase; 1506 const struct ipath *ipp; 1507 nvlist_t *detector = NULL; 1508 1509 class = stable(class); 1510 1511 /* Get the component path from the ereport */ 1512 epnamenp = platform_getpath(nvl); 1513 1514 /* See if we ended up without a path. */ 1515 if (epnamenp == NULL) { 1516 /* See if class permits silent discard on unknown component. */ 1517 if (lut_lookup(Ereportenames_discard, (void *)class, NULL)) { 1518 out(O_ALTFP|O_VERB2, "Unable to map \"%s\" ereport " 1519 "to component path, but silent discard allowed.", 1520 class); 1521 } else { 1522 /* 1523 * XFILE: Failure to find a component is bad unless 1524 * 'discard_if_config_unknown=1' was specified in the 1525 * ereport definition. Indicate undiagnosable. 1526 */ 1527 Undiag_reason = UD_VAL_NOPATH; 1528 fmcase = fmd_case_open(hdl, NULL); 1529 1530 /* 1531 * We don't have a component path here (which means that 1532 * the detector was not in hc-scheme and couldn't be 1533 * converted to hc-scheme. Report the raw detector as 1534 * the suspect resource if there is one. 1535 */ 1536 (void) nvlist_lookup_nvlist(nvl, FM_EREPORT_DETECTOR, 1537 &detector); 1538 publish_undiagnosable(hdl, ffep, fmcase, detector, 1539 (char *)class); 1540 } 1541 return; 1542 } 1543 1544 ipp = ipath(epnamenp); 1545 tree_free(epnamenp); 1546 fme_receive_report(hdl, ffep, class, ipp, nvl); 1547 } 1548 1549 /*ARGSUSED*/ 1550 void 1551 fme_receive_repair_list(fmd_hdl_t *hdl, fmd_event_t *ffep, nvlist_t *nvl, 1552 const char *eventstring) 1553 { 1554 char *uuid; 1555 nvlist_t **nva; 1556 uint_t nvc; 1557 const struct ipath *ipp; 1558 1559 if (nvlist_lookup_string(nvl, FM_SUSPECT_UUID, &uuid) != 0 || 1560 nvlist_lookup_nvlist_array(nvl, FM_SUSPECT_FAULT_LIST, 1561 &nva, &nvc) != 0) { 1562 out(O_ALTFP, "No uuid or fault list for list.repaired event"); 1563 return; 1564 } 1565 1566 out(O_ALTFP, "Processing list.repaired from case %s", uuid); 1567 1568 while (nvc-- != 0) { 1569 /* 1570 * Reset any istat or serd engine associated with this path. 1571 */ 1572 char *path; 1573 1574 if ((ipp = platform_fault2ipath(*nva++)) == NULL) 1575 continue; 1576 1577 path = ipath2str(NULL, ipp); 1578 out(O_ALTFP, "fme_receive_repair_list: resetting state for %s", 1579 path); 1580 FREE(path); 1581 1582 lut_walk(Istats, (lut_cb)istat_counter_reset_cb, (void *)ipp); 1583 istat_save(); 1584 1585 lut_walk(SerdEngines, (lut_cb)serd_reset_cb, (void *)ipp); 1586 serd_save(); 1587 } 1588 } 1589 1590 /*ARGSUSED*/ 1591 void 1592 fme_receive_topology_change(void) 1593 { 1594 lut_walk(Istats, (lut_cb)istat_counter_topo_chg_cb, NULL); 1595 istat_save(); 1596 1597 lut_walk(SerdEngines, (lut_cb)serd_topo_chg_cb, NULL); 1598 serd_save(); 1599 } 1600 1601 static int mark_arrows(struct fme *fmep, struct event *ep, int mark, 1602 unsigned long long at_latest_by, unsigned long long *pdelay, int keep); 1603 1604 /* ARGSUSED */ 1605 static void 1606 clear_arrows(struct event *ep, struct event *ep2, struct fme *fmep) 1607 { 1608 struct bubble *bp; 1609 struct arrowlist *ap; 1610 1611 ep->cached_state = 0; 1612 ep->keep_in_tree = 0; 1613 for (bp = itree_next_bubble(ep, NULL); bp; 1614 bp = itree_next_bubble(ep, bp)) { 1615 if (bp->t != B_FROM) 1616 continue; 1617 bp->mark = 0; 1618 for (ap = itree_next_arrow(bp, NULL); ap; 1619 ap = itree_next_arrow(bp, ap)) 1620 ap->arrowp->mark = 0; 1621 } 1622 } 1623 1624 static void 1625 fme_receive_report(fmd_hdl_t *hdl, fmd_event_t *ffep, 1626 const char *eventstring, const struct ipath *ipp, nvlist_t *nvl) 1627 { 1628 struct event *ep; 1629 struct fme *fmep = NULL; 1630 struct fme *ofmep = NULL; 1631 struct fme *cfmep, *svfmep; 1632 int matched = 0; 1633 nvlist_t *defect; 1634 fmd_case_t *fmcase; 1635 char *reason; 1636 1637 out(O_ALTFP|O_NONL, "fme_receive_report: "); 1638 ipath_print(O_ALTFP|O_NONL, eventstring, ipp); 1639 out(O_ALTFP|O_STAMP, NULL); 1640 1641 /* decide which FME it goes to */ 1642 for (fmep = FMElist; fmep; fmep = fmep->next) { 1643 int prev_verbose; 1644 unsigned long long my_delay = TIMEVAL_EVENTUALLY; 1645 enum fme_state state; 1646 nvlist_t *pre_peek_nvp = NULL; 1647 1648 if (fmep->overflow) { 1649 if (!(fmd_case_closed(fmep->hdl, fmep->fmcase))) 1650 ofmep = fmep; 1651 1652 continue; 1653 } 1654 1655 /* 1656 * ignore solved or closed cases 1657 */ 1658 if (fmep->posted_suspects || 1659 fmd_case_solved(fmep->hdl, fmep->fmcase) || 1660 fmd_case_closed(fmep->hdl, fmep->fmcase)) 1661 continue; 1662 1663 /* look up event in event tree for this FME */ 1664 if ((ep = itree_lookup(fmep->eventtree, 1665 eventstring, ipp)) == NULL) 1666 continue; 1667 1668 /* note observation */ 1669 fmep->ecurrent = ep; 1670 if (ep->count++ == 0) { 1671 /* link it into list of observations seen */ 1672 ep->observations = fmep->observations; 1673 fmep->observations = ep; 1674 ep->nvp = evnv_dupnvl(nvl); 1675 } else { 1676 /* use new payload values for peek */ 1677 pre_peek_nvp = ep->nvp; 1678 ep->nvp = evnv_dupnvl(nvl); 1679 } 1680 1681 /* tell hypothesise() not to mess with suspect list */ 1682 fmep->peek = 1; 1683 1684 /* don't want this to be verbose (unless Debug is set) */ 1685 prev_verbose = Verbose; 1686 if (Debug == 0) 1687 Verbose = 0; 1688 1689 lut_walk(fmep->eventtree, (lut_cb)clear_arrows, (void *)fmep); 1690 state = hypothesise(fmep, fmep->e0, fmep->ull, &my_delay); 1691 1692 fmep->peek = 0; 1693 1694 /* put verbose flag back */ 1695 Verbose = prev_verbose; 1696 1697 if (state != FME_DISPROVED) { 1698 /* found an FME that explains the ereport */ 1699 matched++; 1700 out(O_ALTFP|O_NONL, "["); 1701 ipath_print(O_ALTFP|O_NONL, eventstring, ipp); 1702 out(O_ALTFP, " explained by FME%d]", fmep->id); 1703 1704 nvlist_free(pre_peek_nvp); 1705 1706 if (ep->count == 1) 1707 serialize_observation(fmep, eventstring, ipp); 1708 1709 if (ffep) { 1710 fmd_case_add_ereport(hdl, fmep->fmcase, ffep); 1711 ep->ffep = ffep; 1712 } 1713 1714 stats_counter_bump(fmep->Rcount); 1715 1716 /* re-eval FME */ 1717 fme_eval(fmep, ffep); 1718 } else { 1719 1720 /* not a match, undo noting of observation */ 1721 fmep->ecurrent = NULL; 1722 if (--ep->count == 0) { 1723 /* unlink it from observations */ 1724 fmep->observations = ep->observations; 1725 ep->observations = NULL; 1726 nvlist_free(ep->nvp); 1727 ep->nvp = NULL; 1728 } else { 1729 nvlist_free(ep->nvp); 1730 ep->nvp = pre_peek_nvp; 1731 } 1732 } 1733 } 1734 1735 if (matched) 1736 return; /* explained by at least one existing FME */ 1737 1738 /* clean up closed fmes */ 1739 cfmep = ClosedFMEs; 1740 while (cfmep != NULL) { 1741 svfmep = cfmep->next; 1742 destroy_fme(cfmep); 1743 cfmep = svfmep; 1744 } 1745 ClosedFMEs = NULL; 1746 1747 if (ofmep) { 1748 out(O_ALTFP|O_NONL, "["); 1749 ipath_print(O_ALTFP|O_NONL, eventstring, ipp); 1750 out(O_ALTFP, " ADDING TO OVERFLOW FME]"); 1751 if (ffep) 1752 fmd_case_add_ereport(hdl, ofmep->fmcase, ffep); 1753 1754 return; 1755 1756 } else if (Max_fme && (Open_fme_count >= Max_fme)) { 1757 out(O_ALTFP|O_NONL, "["); 1758 ipath_print(O_ALTFP|O_NONL, eventstring, ipp); 1759 out(O_ALTFP, " MAX OPEN FME REACHED]"); 1760 1761 fmcase = fmd_case_open(hdl, NULL); 1762 1763 /* Create overflow fme */ 1764 if ((fmep = newfme(eventstring, ipp, hdl, fmcase, ffep, 1765 nvl)) == NULL) { 1766 out(O_ALTFP|O_NONL, "["); 1767 ipath_print(O_ALTFP|O_NONL, eventstring, ipp); 1768 out(O_ALTFP, " CANNOT OPEN OVERFLOW FME]"); 1769 return; 1770 } 1771 1772 Open_fme_count++; 1773 1774 init_fme_bufs(fmep); 1775 fmep->overflow = B_TRUE; 1776 1777 if (ffep) 1778 fmd_case_add_ereport(hdl, fmep->fmcase, ffep); 1779 1780 Undiag_reason = UD_VAL_MAXFME; 1781 defect = fmd_nvl_create_fault(hdl, 1782 undiag_2defect_str(Undiag_reason), 100, NULL, NULL, NULL); 1783 reason = undiag_2reason_str(Undiag_reason, NULL); 1784 (void) nvlist_add_string(defect, UNDIAG_REASON, reason); 1785 FREE(reason); 1786 fmd_case_add_suspect(hdl, fmep->fmcase, defect); 1787 fmd_case_solve(hdl, fmep->fmcase); 1788 Undiag_reason = UD_VAL_UNKNOWN; 1789 return; 1790 } 1791 1792 /* open a case */ 1793 fmcase = fmd_case_open(hdl, NULL); 1794 1795 /* start a new FME */ 1796 if ((fmep = newfme(eventstring, ipp, hdl, fmcase, ffep, nvl)) == NULL) { 1797 out(O_ALTFP|O_NONL, "["); 1798 ipath_print(O_ALTFP|O_NONL, eventstring, ipp); 1799 out(O_ALTFP, " CANNOT DIAGNOSE]"); 1800 return; 1801 } 1802 1803 Open_fme_count++; 1804 1805 init_fme_bufs(fmep); 1806 1807 out(O_ALTFP|O_NONL, "["); 1808 ipath_print(O_ALTFP|O_NONL, eventstring, ipp); 1809 out(O_ALTFP, " created FME%d, case %s]", fmep->id, 1810 fmd_case_uuid(hdl, fmep->fmcase)); 1811 1812 ep = fmep->e0; 1813 ASSERT(ep != NULL); 1814 1815 /* note observation */ 1816 fmep->ecurrent = ep; 1817 if (ep->count++ == 0) { 1818 /* link it into list of observations seen */ 1819 ep->observations = fmep->observations; 1820 fmep->observations = ep; 1821 ep->nvp = evnv_dupnvl(nvl); 1822 serialize_observation(fmep, eventstring, ipp); 1823 } else { 1824 /* new payload overrides any previous */ 1825 nvlist_free(ep->nvp); 1826 ep->nvp = evnv_dupnvl(nvl); 1827 } 1828 1829 stats_counter_bump(fmep->Rcount); 1830 1831 if (ffep) { 1832 fmd_case_add_ereport(hdl, fmep->fmcase, ffep); 1833 fmd_case_setprincipal(hdl, fmep->fmcase, ffep); 1834 fmep->e0r = ffep; 1835 ep->ffep = ffep; 1836 } 1837 1838 /* give the diagnosis algorithm a shot at the new FME state */ 1839 fme_eval(fmep, ffep); 1840 } 1841 1842 void 1843 fme_status(int flags) 1844 { 1845 struct fme *fmep; 1846 1847 if (FMElist == NULL) { 1848 out(flags, "No fault management exercises underway."); 1849 return; 1850 } 1851 1852 for (fmep = FMElist; fmep; fmep = fmep->next) 1853 fme_print(flags, fmep); 1854 } 1855 1856 /* 1857 * "indent" routines used mostly for nicely formatted debug output, but also 1858 * for sanity checking for infinite recursion bugs. 1859 */ 1860 1861 #define MAX_INDENT 1024 1862 static const char *indent_s[MAX_INDENT]; 1863 static int current_indent; 1864 1865 static void 1866 indent_push(const char *s) 1867 { 1868 if (current_indent < MAX_INDENT) 1869 indent_s[current_indent++] = s; 1870 else 1871 out(O_DIE, "unexpected recursion depth (%d)", current_indent); 1872 } 1873 1874 static void 1875 indent_set(const char *s) 1876 { 1877 current_indent = 0; 1878 indent_push(s); 1879 } 1880 1881 static void 1882 indent_pop(void) 1883 { 1884 if (current_indent > 0) 1885 current_indent--; 1886 else 1887 out(O_DIE, "recursion underflow"); 1888 } 1889 1890 static void 1891 indent(void) 1892 { 1893 int i; 1894 if (!Verbose) 1895 return; 1896 for (i = 0; i < current_indent; i++) 1897 out(O_ALTFP|O_VERB|O_NONL, indent_s[i]); 1898 } 1899 1900 #define SLNEW 1 1901 #define SLCHANGED 2 1902 #define SLWAIT 3 1903 #define SLDISPROVED 4 1904 1905 static void 1906 print_suspects(int circumstance, struct fme *fmep) 1907 { 1908 struct event *ep; 1909 1910 out(O_ALTFP|O_NONL, "["); 1911 if (circumstance == SLCHANGED) { 1912 out(O_ALTFP|O_NONL, "FME%d diagnosis changed. state: %s, " 1913 "suspect list:", fmep->id, fme_state2str(fmep->state)); 1914 } else if (circumstance == SLWAIT) { 1915 out(O_ALTFP|O_NONL, "FME%d set wait timer %ld ", fmep->id, 1916 fmep->timer); 1917 ptree_timeval(O_ALTFP|O_NONL, &fmep->wull); 1918 } else if (circumstance == SLDISPROVED) { 1919 out(O_ALTFP|O_NONL, "FME%d DIAGNOSIS UNKNOWN", fmep->id); 1920 } else { 1921 out(O_ALTFP|O_NONL, "FME%d DIAGNOSIS PRODUCED:", fmep->id); 1922 } 1923 1924 if (circumstance == SLWAIT || circumstance == SLDISPROVED) { 1925 out(O_ALTFP, "]"); 1926 return; 1927 } 1928 1929 for (ep = fmep->suspects; ep; ep = ep->suspects) { 1930 out(O_ALTFP|O_NONL, " "); 1931 itree_pevent_brief(O_ALTFP|O_NONL, ep); 1932 } 1933 out(O_ALTFP, "]"); 1934 } 1935 1936 static struct node * 1937 eventprop_lookup(struct event *ep, const char *propname) 1938 { 1939 return (lut_lookup(ep->props, (void *)propname, NULL)); 1940 } 1941 1942 #define MAXDIGITIDX 23 1943 static char numbuf[MAXDIGITIDX + 1]; 1944 1945 static int 1946 node2uint(struct node *n, uint_t *valp) 1947 { 1948 struct evalue value; 1949 struct lut *globals = NULL; 1950 1951 if (n == NULL) 1952 return (1); 1953 1954 /* 1955 * check value.v since we are being asked to convert an unsigned 1956 * long long int to an unsigned int 1957 */ 1958 if (! eval_expr(n, NULL, NULL, &globals, NULL, NULL, 0, &value) || 1959 value.t != UINT64 || value.v > (1ULL << 32)) 1960 return (1); 1961 1962 *valp = (uint_t)value.v; 1963 1964 return (0); 1965 } 1966 1967 static nvlist_t * 1968 node2fmri(struct node *n) 1969 { 1970 nvlist_t **pa, *f, *p; 1971 struct node *nc; 1972 uint_t depth = 0; 1973 char *numstr, *nullbyte; 1974 char *failure; 1975 int err, i; 1976 1977 /* XXX do we need to be able to handle a non-T_NAME node? */ 1978 if (n == NULL || n->t != T_NAME) 1979 return (NULL); 1980 1981 for (nc = n; nc != NULL; nc = nc->u.name.next) { 1982 if (nc->u.name.child == NULL || nc->u.name.child->t != T_NUM) 1983 break; 1984 depth++; 1985 } 1986 1987 if (nc != NULL) { 1988 /* We bailed early, something went wrong */ 1989 return (NULL); 1990 } 1991 1992 if ((err = nvlist_xalloc(&f, NV_UNIQUE_NAME, &Eft_nv_hdl)) != 0) 1993 out(O_DIE|O_SYS, "alloc of fmri nvl failed"); 1994 pa = alloca(depth * sizeof (nvlist_t *)); 1995 for (i = 0; i < depth; i++) 1996 pa[i] = NULL; 1997 1998 err = nvlist_add_string(f, FM_FMRI_SCHEME, FM_FMRI_SCHEME_HC); 1999 err |= nvlist_add_uint8(f, FM_VERSION, FM_HC_SCHEME_VERSION); 2000 err |= nvlist_add_string(f, FM_FMRI_HC_ROOT, ""); 2001 err |= nvlist_add_uint32(f, FM_FMRI_HC_LIST_SZ, depth); 2002 if (err != 0) { 2003 failure = "basic construction of FMRI failed"; 2004 goto boom; 2005 } 2006 2007 numbuf[MAXDIGITIDX] = '\0'; 2008 nullbyte = &numbuf[MAXDIGITIDX]; 2009 i = 0; 2010 2011 for (nc = n; nc != NULL; nc = nc->u.name.next) { 2012 err = nvlist_xalloc(&p, NV_UNIQUE_NAME, &Eft_nv_hdl); 2013 if (err != 0) { 2014 failure = "alloc of an hc-pair failed"; 2015 goto boom; 2016 } 2017 err = nvlist_add_string(p, FM_FMRI_HC_NAME, nc->u.name.s); 2018 numstr = ulltostr(nc->u.name.child->u.ull, nullbyte); 2019 err |= nvlist_add_string(p, FM_FMRI_HC_ID, numstr); 2020 if (err != 0) { 2021 failure = "construction of an hc-pair failed"; 2022 goto boom; 2023 } 2024 pa[i++] = p; 2025 } 2026 2027 err = nvlist_add_nvlist_array(f, FM_FMRI_HC_LIST, pa, depth); 2028 if (err == 0) { 2029 for (i = 0; i < depth; i++) 2030 nvlist_free(pa[i]); 2031 return (f); 2032 } 2033 failure = "addition of hc-pair array to FMRI failed"; 2034 2035 boom: 2036 for (i = 0; i < depth; i++) 2037 nvlist_free(pa[i]); 2038 nvlist_free(f); 2039 out(O_DIE, "%s", failure); 2040 /*NOTREACHED*/ 2041 return (NULL); 2042 } 2043 2044 /* an ipath cache entry is an array of these, with s==NULL at the end */ 2045 struct ipath { 2046 const char *s; /* component name (in stable) */ 2047 int i; /* instance number */ 2048 }; 2049 2050 static nvlist_t * 2051 ipath2fmri(struct ipath *ipath) 2052 { 2053 nvlist_t **pa, *f, *p; 2054 uint_t depth = 0; 2055 char *numstr, *nullbyte; 2056 char *failure; 2057 int err, i; 2058 struct ipath *ipp; 2059 2060 for (ipp = ipath; ipp->s != NULL; ipp++) 2061 depth++; 2062 2063 if ((err = nvlist_xalloc(&f, NV_UNIQUE_NAME, &Eft_nv_hdl)) != 0) 2064 out(O_DIE|O_SYS, "alloc of fmri nvl failed"); 2065 pa = alloca(depth * sizeof (nvlist_t *)); 2066 for (i = 0; i < depth; i++) 2067 pa[i] = NULL; 2068 2069 err = nvlist_add_string(f, FM_FMRI_SCHEME, FM_FMRI_SCHEME_HC); 2070 err |= nvlist_add_uint8(f, FM_VERSION, FM_HC_SCHEME_VERSION); 2071 err |= nvlist_add_string(f, FM_FMRI_HC_ROOT, ""); 2072 err |= nvlist_add_uint32(f, FM_FMRI_HC_LIST_SZ, depth); 2073 if (err != 0) { 2074 failure = "basic construction of FMRI failed"; 2075 goto boom; 2076 } 2077 2078 numbuf[MAXDIGITIDX] = '\0'; 2079 nullbyte = &numbuf[MAXDIGITIDX]; 2080 i = 0; 2081 2082 for (ipp = ipath; ipp->s != NULL; ipp++) { 2083 err = nvlist_xalloc(&p, NV_UNIQUE_NAME, &Eft_nv_hdl); 2084 if (err != 0) { 2085 failure = "alloc of an hc-pair failed"; 2086 goto boom; 2087 } 2088 err = nvlist_add_string(p, FM_FMRI_HC_NAME, ipp->s); 2089 numstr = ulltostr(ipp->i, nullbyte); 2090 err |= nvlist_add_string(p, FM_FMRI_HC_ID, numstr); 2091 if (err != 0) { 2092 failure = "construction of an hc-pair failed"; 2093 goto boom; 2094 } 2095 pa[i++] = p; 2096 } 2097 2098 err = nvlist_add_nvlist_array(f, FM_FMRI_HC_LIST, pa, depth); 2099 if (err == 0) { 2100 for (i = 0; i < depth; i++) 2101 nvlist_free(pa[i]); 2102 return (f); 2103 } 2104 failure = "addition of hc-pair array to FMRI failed"; 2105 2106 boom: 2107 for (i = 0; i < depth; i++) 2108 nvlist_free(pa[i]); 2109 nvlist_free(f); 2110 out(O_DIE, "%s", failure); 2111 /*NOTREACHED*/ 2112 return (NULL); 2113 } 2114 2115 static uint8_t 2116 percentof(uint_t part, uint_t whole) 2117 { 2118 unsigned long long p = part * 1000; 2119 2120 return ((p / whole / 10) + (((p / whole % 10) >= 5) ? 1 : 0)); 2121 } 2122 2123 struct rsl { 2124 struct event *suspect; 2125 nvlist_t *asru; 2126 nvlist_t *fru; 2127 nvlist_t *rsrc; 2128 }; 2129 2130 static void publish_suspects(struct fme *fmep, struct rsl *srl); 2131 2132 /* 2133 * rslfree -- free internal members of struct rsl not expected to be 2134 * freed elsewhere. 2135 */ 2136 static void 2137 rslfree(struct rsl *freeme) 2138 { 2139 nvlist_free(freeme->asru); 2140 nvlist_free(freeme->fru); 2141 if (freeme->rsrc != freeme->asru) 2142 nvlist_free(freeme->rsrc); 2143 } 2144 2145 /* 2146 * rslcmp -- compare two rsl structures. Use the following 2147 * comparisons to establish cardinality: 2148 * 2149 * 1. Name of the suspect's class. (simple strcmp) 2150 * 2. Name of the suspect's ASRU. (trickier, since nvlist) 2151 * 2152 */ 2153 static int 2154 rslcmp(const void *a, const void *b) 2155 { 2156 struct rsl *r1 = (struct rsl *)a; 2157 struct rsl *r2 = (struct rsl *)b; 2158 int rv; 2159 2160 rv = strcmp(r1->suspect->enode->u.event.ename->u.name.s, 2161 r2->suspect->enode->u.event.ename->u.name.s); 2162 if (rv != 0) 2163 return (rv); 2164 2165 if (r1->rsrc == NULL && r2->rsrc == NULL) 2166 return (0); 2167 if (r1->rsrc == NULL) 2168 return (-1); 2169 if (r2->rsrc == NULL) 2170 return (1); 2171 return (evnv_cmpnvl(r1->rsrc, r2->rsrc, 0)); 2172 } 2173 2174 /* 2175 * get_resources -- for a given suspect, determine what ASRU, FRU and 2176 * RSRC nvlists should be advertised in the final suspect list. 2177 */ 2178 void 2179 get_resources(struct event *sp, struct rsl *rsrcs, struct config *croot) 2180 { 2181 struct node *asrudef, *frudef; 2182 nvlist_t *asru, *fru; 2183 nvlist_t *rsrc = NULL; 2184 char *pathstr; 2185 2186 /* 2187 * First find any ASRU and/or FRU defined in the 2188 * initial fault tree. 2189 */ 2190 asrudef = eventprop_lookup(sp, L_ASRU); 2191 frudef = eventprop_lookup(sp, L_FRU); 2192 2193 /* 2194 * Create FMRIs based on those definitions 2195 */ 2196 asru = node2fmri(asrudef); 2197 fru = node2fmri(frudef); 2198 pathstr = ipath2str(NULL, sp->ipp); 2199 2200 /* 2201 * Allow for platform translations of the FMRIs 2202 */ 2203 platform_units_translate(is_defect(sp->t), croot, &asru, &fru, &rsrc, 2204 pathstr); 2205 2206 FREE(pathstr); 2207 rsrcs->suspect = sp; 2208 rsrcs->asru = asru; 2209 rsrcs->fru = fru; 2210 rsrcs->rsrc = rsrc; 2211 } 2212 2213 /* 2214 * trim_suspects -- prior to publishing, we may need to remove some 2215 * suspects from the list. If we're auto-closing upsets, we don't 2216 * want any of those in the published list. If the ASRUs for multiple 2217 * defects resolve to the same ASRU (driver) we only want to publish 2218 * that as a single suspect. 2219 */ 2220 static int 2221 trim_suspects(struct fme *fmep, struct rsl *begin, struct rsl *begin2, 2222 fmd_event_t *ffep) 2223 { 2224 struct event *ep; 2225 struct rsl *rp = begin; 2226 struct rsl *rp2 = begin2; 2227 int mess_zero_count = 0; 2228 int serd_rval; 2229 uint_t messval; 2230 2231 /* remove any unwanted upsets and populate our array */ 2232 for (ep = fmep->psuspects; ep; ep = ep->psuspects) { 2233 if (is_upset(ep->t)) 2234 continue; 2235 serd_rval = serd_eval(fmep, fmep->hdl, ffep, fmep->fmcase, ep, 2236 NULL, NULL); 2237 if (serd_rval == 0) 2238 continue; 2239 if (node2uint(eventprop_lookup(ep, L_message), 2240 &messval) == 0 && messval == 0) { 2241 get_resources(ep, rp2, fmep->config); 2242 rp2++; 2243 mess_zero_count++; 2244 } else { 2245 get_resources(ep, rp, fmep->config); 2246 rp++; 2247 fmep->nsuspects++; 2248 } 2249 } 2250 return (mess_zero_count); 2251 } 2252 2253 /* 2254 * addpayloadprop -- add a payload prop to a problem 2255 */ 2256 static void 2257 addpayloadprop(const char *lhs, struct evalue *rhs, nvlist_t *fault) 2258 { 2259 nvlist_t *rsrc, *hcs; 2260 2261 ASSERT(fault != NULL); 2262 ASSERT(lhs != NULL); 2263 ASSERT(rhs != NULL); 2264 2265 if (nvlist_lookup_nvlist(fault, FM_FAULT_RESOURCE, &rsrc) != 0) 2266 out(O_DIE, "cannot add payloadprop \"%s\" to fault", lhs); 2267 2268 if (nvlist_lookup_nvlist(rsrc, FM_FMRI_HC_SPECIFIC, &hcs) != 0) { 2269 out(O_ALTFP|O_VERB2, "addpayloadprop: create hc_specific"); 2270 if (nvlist_xalloc(&hcs, NV_UNIQUE_NAME, &Eft_nv_hdl) != 0) 2271 out(O_DIE, 2272 "cannot add payloadprop \"%s\" to fault", lhs); 2273 if (nvlist_add_nvlist(rsrc, FM_FMRI_HC_SPECIFIC, hcs) != 0) 2274 out(O_DIE, 2275 "cannot add payloadprop \"%s\" to fault", lhs); 2276 nvlist_free(hcs); 2277 if (nvlist_lookup_nvlist(rsrc, FM_FMRI_HC_SPECIFIC, &hcs) != 0) 2278 out(O_DIE, 2279 "cannot add payloadprop \"%s\" to fault", lhs); 2280 } else 2281 out(O_ALTFP|O_VERB2, "addpayloadprop: reuse hc_specific"); 2282 2283 if (rhs->t == UINT64) { 2284 out(O_ALTFP|O_VERB2, "addpayloadprop: %s=%llu", lhs, rhs->v); 2285 2286 if (nvlist_add_uint64(hcs, lhs, rhs->v) != 0) 2287 out(O_DIE, 2288 "cannot add payloadprop \"%s\" to fault", lhs); 2289 } else { 2290 out(O_ALTFP|O_VERB2, "addpayloadprop: %s=\"%s\"", 2291 lhs, (char *)(uintptr_t)rhs->v); 2292 2293 if (nvlist_add_string(hcs, lhs, (char *)(uintptr_t)rhs->v) != 0) 2294 out(O_DIE, 2295 "cannot add payloadprop \"%s\" to fault", lhs); 2296 } 2297 } 2298 2299 static char *Istatbuf; 2300 static char *Istatbufptr; 2301 static int Istatsz; 2302 2303 /* 2304 * istataddsize -- calculate size of istat and add it to Istatsz 2305 */ 2306 /*ARGSUSED2*/ 2307 static void 2308 istataddsize(const struct istat_entry *lhs, struct stats *rhs, void *arg) 2309 { 2310 int val; 2311 2312 ASSERT(lhs != NULL); 2313 ASSERT(rhs != NULL); 2314 2315 if ((val = stats_counter_value(rhs)) == 0) 2316 return; /* skip zero-valued stats */ 2317 2318 /* count up the size of the stat name */ 2319 Istatsz += ipath2strlen(lhs->ename, lhs->ipath); 2320 Istatsz++; /* for the trailing NULL byte */ 2321 2322 /* count up the size of the stat value */ 2323 Istatsz += snprintf(NULL, 0, "%d", val); 2324 Istatsz++; /* for the trailing NULL byte */ 2325 } 2326 2327 /* 2328 * istat2str -- serialize an istat, writing result to *Istatbufptr 2329 */ 2330 /*ARGSUSED2*/ 2331 static void 2332 istat2str(const struct istat_entry *lhs, struct stats *rhs, void *arg) 2333 { 2334 char *str; 2335 int len; 2336 int val; 2337 2338 ASSERT(lhs != NULL); 2339 ASSERT(rhs != NULL); 2340 2341 if ((val = stats_counter_value(rhs)) == 0) 2342 return; /* skip zero-valued stats */ 2343 2344 /* serialize the stat name */ 2345 str = ipath2str(lhs->ename, lhs->ipath); 2346 len = strlen(str); 2347 2348 ASSERT(Istatbufptr + len + 1 < &Istatbuf[Istatsz]); 2349 (void) strlcpy(Istatbufptr, str, &Istatbuf[Istatsz] - Istatbufptr); 2350 Istatbufptr += len; 2351 FREE(str); 2352 *Istatbufptr++ = '\0'; 2353 2354 /* serialize the stat value */ 2355 Istatbufptr += snprintf(Istatbufptr, &Istatbuf[Istatsz] - Istatbufptr, 2356 "%d", val); 2357 *Istatbufptr++ = '\0'; 2358 2359 ASSERT(Istatbufptr <= &Istatbuf[Istatsz]); 2360 } 2361 2362 void 2363 istat_save() 2364 { 2365 if (Istat_need_save == 0) 2366 return; 2367 2368 /* figure out how big the serialzed info is */ 2369 Istatsz = 0; 2370 lut_walk(Istats, (lut_cb)istataddsize, NULL); 2371 2372 if (Istatsz == 0) { 2373 /* no stats to save */ 2374 fmd_buf_destroy(Hdl, NULL, WOBUF_ISTATS); 2375 return; 2376 } 2377 2378 /* create the serialized buffer */ 2379 Istatbufptr = Istatbuf = MALLOC(Istatsz); 2380 lut_walk(Istats, (lut_cb)istat2str, NULL); 2381 2382 /* clear out current saved stats */ 2383 fmd_buf_destroy(Hdl, NULL, WOBUF_ISTATS); 2384 2385 /* write out the new version */ 2386 fmd_buf_write(Hdl, NULL, WOBUF_ISTATS, Istatbuf, Istatsz); 2387 FREE(Istatbuf); 2388 2389 Istat_need_save = 0; 2390 } 2391 2392 int 2393 istat_cmp(struct istat_entry *ent1, struct istat_entry *ent2) 2394 { 2395 if (ent1->ename != ent2->ename) 2396 return (ent2->ename - ent1->ename); 2397 if (ent1->ipath != ent2->ipath) 2398 return ((char *)ent2->ipath - (char *)ent1->ipath); 2399 2400 return (0); 2401 } 2402 2403 /* 2404 * istat-verify -- verify the component associated with a stat still exists 2405 * 2406 * if the component no longer exists, this routine resets the stat and 2407 * returns 0. if the component still exists, it returns 1. 2408 */ 2409 static int 2410 istat_verify(struct node *snp, struct istat_entry *entp) 2411 { 2412 struct stats *statp; 2413 nvlist_t *fmri; 2414 2415 fmri = node2fmri(snp->u.event.epname); 2416 if (platform_path_exists(fmri)) { 2417 nvlist_free(fmri); 2418 return (1); 2419 } 2420 nvlist_free(fmri); 2421 2422 /* component no longer in system. zero out the associated stats */ 2423 if ((statp = (struct stats *) 2424 lut_lookup(Istats, entp, (lut_cmp)istat_cmp)) == NULL || 2425 stats_counter_value(statp) == 0) 2426 return (0); /* stat is already reset */ 2427 2428 Istat_need_save = 1; 2429 stats_counter_reset(statp); 2430 return (0); 2431 } 2432 2433 static void 2434 istat_bump(struct node *snp, int n) 2435 { 2436 struct stats *statp; 2437 struct istat_entry ent; 2438 2439 ASSERT(snp != NULL); 2440 ASSERTinfo(snp->t == T_EVENT, ptree_nodetype2str(snp->t)); 2441 ASSERT(snp->u.event.epname != NULL); 2442 2443 /* class name should be hoisted into a single stable entry */ 2444 ASSERT(snp->u.event.ename->u.name.next == NULL); 2445 ent.ename = snp->u.event.ename->u.name.s; 2446 ent.ipath = ipath(snp->u.event.epname); 2447 2448 if (!istat_verify(snp, &ent)) { 2449 /* component no longer exists in system, nothing to do */ 2450 return; 2451 } 2452 2453 if ((statp = (struct stats *) 2454 lut_lookup(Istats, &ent, (lut_cmp)istat_cmp)) == NULL) { 2455 /* need to create the counter */ 2456 int cnt = 0; 2457 struct node *np; 2458 char *sname; 2459 char *snamep; 2460 struct istat_entry *newentp; 2461 2462 /* count up the size of the stat name */ 2463 np = snp->u.event.ename; 2464 while (np != NULL) { 2465 cnt += strlen(np->u.name.s); 2466 cnt++; /* for the '.' or '@' */ 2467 np = np->u.name.next; 2468 } 2469 np = snp->u.event.epname; 2470 while (np != NULL) { 2471 cnt += snprintf(NULL, 0, "%s%llu", 2472 np->u.name.s, np->u.name.child->u.ull); 2473 cnt++; /* for the '/' or trailing NULL byte */ 2474 np = np->u.name.next; 2475 } 2476 2477 /* build the stat name */ 2478 snamep = sname = alloca(cnt); 2479 np = snp->u.event.ename; 2480 while (np != NULL) { 2481 snamep += snprintf(snamep, &sname[cnt] - snamep, 2482 "%s", np->u.name.s); 2483 np = np->u.name.next; 2484 if (np) 2485 *snamep++ = '.'; 2486 } 2487 *snamep++ = '@'; 2488 np = snp->u.event.epname; 2489 while (np != NULL) { 2490 snamep += snprintf(snamep, &sname[cnt] - snamep, 2491 "%s%llu", np->u.name.s, np->u.name.child->u.ull); 2492 np = np->u.name.next; 2493 if (np) 2494 *snamep++ = '/'; 2495 } 2496 *snamep++ = '\0'; 2497 2498 /* create the new stat & add it to our list */ 2499 newentp = MALLOC(sizeof (*newentp)); 2500 *newentp = ent; 2501 statp = stats_new_counter(NULL, sname, 0); 2502 Istats = lut_add(Istats, (void *)newentp, (void *)statp, 2503 (lut_cmp)istat_cmp); 2504 } 2505 2506 /* if n is non-zero, set that value instead of bumping */ 2507 if (n) { 2508 stats_counter_reset(statp); 2509 stats_counter_add(statp, n); 2510 } else 2511 stats_counter_bump(statp); 2512 Istat_need_save = 1; 2513 2514 ipath_print(O_ALTFP|O_VERB2, ent.ename, ent.ipath); 2515 out(O_ALTFP|O_VERB2, " %s to value %d", n ? "set" : "incremented", 2516 stats_counter_value(statp)); 2517 } 2518 2519 /*ARGSUSED*/ 2520 static void 2521 istat_destructor(void *left, void *right, void *arg) 2522 { 2523 struct istat_entry *entp = (struct istat_entry *)left; 2524 struct stats *statp = (struct stats *)right; 2525 FREE(entp); 2526 stats_delete(statp); 2527 } 2528 2529 /* 2530 * Callback used in a walk of the Istats to reset matching stat counters. 2531 */ 2532 static void 2533 istat_counter_reset_cb(struct istat_entry *entp, struct stats *statp, 2534 const struct ipath *ipp) 2535 { 2536 char *path; 2537 2538 if (entp->ipath == ipp) { 2539 path = ipath2str(entp->ename, ipp); 2540 out(O_ALTFP, "istat_counter_reset_cb: resetting %s", path); 2541 FREE(path); 2542 stats_counter_reset(statp); 2543 Istat_need_save = 1; 2544 } 2545 } 2546 2547 /*ARGSUSED*/ 2548 static void 2549 istat_counter_topo_chg_cb(struct istat_entry *entp, struct stats *statp, 2550 void *unused) 2551 { 2552 char *path; 2553 nvlist_t *fmri; 2554 2555 fmri = ipath2fmri((struct ipath *)(entp->ipath)); 2556 if (!platform_path_exists(fmri)) { 2557 path = ipath2str(entp->ename, entp->ipath); 2558 out(O_ALTFP, "istat_counter_topo_chg_cb: not present %s", path); 2559 FREE(path); 2560 stats_counter_reset(statp); 2561 Istat_need_save = 1; 2562 } 2563 nvlist_free(fmri); 2564 } 2565 2566 void 2567 istat_fini(void) 2568 { 2569 lut_free(Istats, istat_destructor, NULL); 2570 } 2571 2572 static char *Serdbuf; 2573 static char *Serdbufptr; 2574 static int Serdsz; 2575 2576 /* 2577 * serdaddsize -- calculate size of serd and add it to Serdsz 2578 */ 2579 /*ARGSUSED*/ 2580 static void 2581 serdaddsize(const struct serd_entry *lhs, struct stats *rhs, void *arg) 2582 { 2583 ASSERT(lhs != NULL); 2584 2585 /* count up the size of the stat name */ 2586 Serdsz += ipath2strlen(lhs->ename, lhs->ipath); 2587 Serdsz++; /* for the trailing NULL byte */ 2588 } 2589 2590 /* 2591 * serd2str -- serialize a serd engine, writing result to *Serdbufptr 2592 */ 2593 /*ARGSUSED*/ 2594 static void 2595 serd2str(const struct serd_entry *lhs, struct stats *rhs, void *arg) 2596 { 2597 char *str; 2598 int len; 2599 2600 ASSERT(lhs != NULL); 2601 2602 /* serialize the serd engine name */ 2603 str = ipath2str(lhs->ename, lhs->ipath); 2604 len = strlen(str); 2605 2606 ASSERT(Serdbufptr + len + 1 <= &Serdbuf[Serdsz]); 2607 (void) strlcpy(Serdbufptr, str, &Serdbuf[Serdsz] - Serdbufptr); 2608 Serdbufptr += len; 2609 FREE(str); 2610 *Serdbufptr++ = '\0'; 2611 ASSERT(Serdbufptr <= &Serdbuf[Serdsz]); 2612 } 2613 2614 void 2615 serd_save() 2616 { 2617 if (Serd_need_save == 0) 2618 return; 2619 2620 /* figure out how big the serialzed info is */ 2621 Serdsz = 0; 2622 lut_walk(SerdEngines, (lut_cb)serdaddsize, NULL); 2623 2624 if (Serdsz == 0) { 2625 /* no serd engines to save */ 2626 fmd_buf_destroy(Hdl, NULL, WOBUF_SERDS); 2627 return; 2628 } 2629 2630 /* create the serialized buffer */ 2631 Serdbufptr = Serdbuf = MALLOC(Serdsz); 2632 lut_walk(SerdEngines, (lut_cb)serd2str, NULL); 2633 2634 /* clear out current saved stats */ 2635 fmd_buf_destroy(Hdl, NULL, WOBUF_SERDS); 2636 2637 /* write out the new version */ 2638 fmd_buf_write(Hdl, NULL, WOBUF_SERDS, Serdbuf, Serdsz); 2639 FREE(Serdbuf); 2640 Serd_need_save = 0; 2641 } 2642 2643 int 2644 serd_cmp(struct serd_entry *ent1, struct serd_entry *ent2) 2645 { 2646 if (ent1->ename != ent2->ename) 2647 return (ent2->ename - ent1->ename); 2648 if (ent1->ipath != ent2->ipath) 2649 return ((char *)ent2->ipath - (char *)ent1->ipath); 2650 2651 return (0); 2652 } 2653 2654 void 2655 fme_serd_load(fmd_hdl_t *hdl) 2656 { 2657 int sz; 2658 char *sbuf; 2659 char *sepptr; 2660 char *ptr; 2661 struct serd_entry *newentp; 2662 struct node *epname; 2663 nvlist_t *fmri; 2664 char *namestring; 2665 2666 if ((sz = fmd_buf_size(hdl, NULL, WOBUF_SERDS)) == 0) 2667 return; 2668 sbuf = alloca(sz); 2669 fmd_buf_read(hdl, NULL, WOBUF_SERDS, sbuf, sz); 2670 ptr = sbuf; 2671 while (ptr < &sbuf[sz]) { 2672 sepptr = strchr(ptr, '@'); 2673 *sepptr = '\0'; 2674 namestring = ptr; 2675 sepptr++; 2676 ptr = sepptr; 2677 ptr += strlen(ptr); 2678 ptr++; /* move past the '\0' separating paths */ 2679 epname = pathstring2epnamenp(sepptr); 2680 fmri = node2fmri(epname); 2681 if (platform_path_exists(fmri)) { 2682 newentp = MALLOC(sizeof (*newentp)); 2683 newentp->hdl = hdl; 2684 newentp->ipath = ipath(epname); 2685 newentp->ename = stable(namestring); 2686 SerdEngines = lut_add(SerdEngines, (void *)newentp, 2687 (void *)newentp, (lut_cmp)serd_cmp); 2688 } else 2689 Serd_need_save = 1; 2690 tree_free(epname); 2691 nvlist_free(fmri); 2692 } 2693 /* save it back again in case some of the paths no longer exist */ 2694 serd_save(); 2695 } 2696 2697 /*ARGSUSED*/ 2698 static void 2699 serd_destructor(void *left, void *right, void *arg) 2700 { 2701 struct serd_entry *entp = (struct serd_entry *)left; 2702 FREE(entp); 2703 } 2704 2705 /* 2706 * Callback used in a walk of the SerdEngines to reset matching serd engines. 2707 */ 2708 /*ARGSUSED*/ 2709 static void 2710 serd_reset_cb(struct serd_entry *entp, void *unused, const struct ipath *ipp) 2711 { 2712 char *path; 2713 2714 if (entp->ipath == ipp) { 2715 path = ipath2str(entp->ename, ipp); 2716 out(O_ALTFP, "serd_reset_cb: resetting %s", path); 2717 fmd_serd_reset(entp->hdl, path); 2718 FREE(path); 2719 Serd_need_save = 1; 2720 } 2721 } 2722 2723 /*ARGSUSED*/ 2724 static void 2725 serd_topo_chg_cb(struct serd_entry *entp, void *unused, void *unused2) 2726 { 2727 char *path; 2728 nvlist_t *fmri; 2729 2730 fmri = ipath2fmri((struct ipath *)(entp->ipath)); 2731 if (!platform_path_exists(fmri)) { 2732 path = ipath2str(entp->ename, entp->ipath); 2733 out(O_ALTFP, "serd_topo_chg_cb: not present %s", path); 2734 fmd_serd_reset(entp->hdl, path); 2735 FREE(path); 2736 Serd_need_save = 1; 2737 } 2738 nvlist_free(fmri); 2739 } 2740 2741 void 2742 serd_fini(void) 2743 { 2744 lut_free(SerdEngines, serd_destructor, NULL); 2745 } 2746 2747 static void 2748 publish_suspects(struct fme *fmep, struct rsl *srl) 2749 { 2750 struct rsl *rp; 2751 nvlist_t *fault; 2752 uint8_t cert; 2753 uint_t *frs; 2754 uint_t frsum, fr; 2755 uint_t messval; 2756 uint_t retireval; 2757 uint_t responseval; 2758 struct node *snp; 2759 int frcnt, fridx; 2760 boolean_t allfaulty = B_TRUE; 2761 struct rsl *erl = srl + fmep->nsuspects - 1; 2762 2763 /* 2764 * sort the array 2765 */ 2766 qsort(srl, fmep->nsuspects, sizeof (struct rsl), rslcmp); 2767 2768 /* sum the fitrates */ 2769 frs = alloca(fmep->nsuspects * sizeof (uint_t)); 2770 fridx = frcnt = frsum = 0; 2771 2772 for (rp = srl; rp <= erl; rp++) { 2773 struct node *n; 2774 2775 n = eventprop_lookup(rp->suspect, L_FITrate); 2776 if (node2uint(n, &fr) != 0) { 2777 out(O_DEBUG|O_NONL, "event "); 2778 ipath_print(O_DEBUG|O_NONL, 2779 rp->suspect->enode->u.event.ename->u.name.s, 2780 rp->suspect->ipp); 2781 out(O_VERB, " has no FITrate (using 1)"); 2782 fr = 1; 2783 } else if (fr == 0) { 2784 out(O_DEBUG|O_NONL, "event "); 2785 ipath_print(O_DEBUG|O_NONL, 2786 rp->suspect->enode->u.event.ename->u.name.s, 2787 rp->suspect->ipp); 2788 out(O_VERB, " has zero FITrate (using 1)"); 2789 fr = 1; 2790 } 2791 2792 frs[fridx++] = fr; 2793 frsum += fr; 2794 frcnt++; 2795 } 2796 2797 /* Add them in reverse order of our sort, as fmd reverses order */ 2798 for (rp = erl; rp >= srl; rp--) { 2799 cert = percentof(frs[--fridx], frsum); 2800 fault = fmd_nvl_create_fault(fmep->hdl, 2801 rp->suspect->enode->u.event.ename->u.name.s, 2802 cert, 2803 rp->asru, 2804 rp->fru, 2805 rp->rsrc); 2806 if (fault == NULL) 2807 out(O_DIE, "fault creation failed"); 2808 /* if "message" property exists, add it to the fault */ 2809 if (node2uint(eventprop_lookup(rp->suspect, L_message), 2810 &messval) == 0) { 2811 2812 out(O_ALTFP, 2813 "[FME%d, %s adds message=%d to suspect list]", 2814 fmep->id, 2815 rp->suspect->enode->u.event.ename->u.name.s, 2816 messval); 2817 if (nvlist_add_boolean_value(fault, 2818 FM_SUSPECT_MESSAGE, 2819 (messval) ? B_TRUE : B_FALSE) != 0) { 2820 out(O_DIE, "cannot add no-message to fault"); 2821 } 2822 } 2823 2824 /* if "retire" property exists, add it to the fault */ 2825 if (node2uint(eventprop_lookup(rp->suspect, L_retire), 2826 &retireval) == 0) { 2827 2828 out(O_ALTFP, 2829 "[FME%d, %s adds retire=%d to suspect list]", 2830 fmep->id, 2831 rp->suspect->enode->u.event.ename->u.name.s, 2832 retireval); 2833 if (nvlist_add_boolean_value(fault, 2834 FM_SUSPECT_RETIRE, 2835 (retireval) ? B_TRUE : B_FALSE) != 0) { 2836 out(O_DIE, "cannot add no-retire to fault"); 2837 } 2838 } 2839 2840 /* if "response" property exists, add it to the fault */ 2841 if (node2uint(eventprop_lookup(rp->suspect, L_response), 2842 &responseval) == 0) { 2843 2844 out(O_ALTFP, 2845 "[FME%d, %s adds response=%d to suspect list]", 2846 fmep->id, 2847 rp->suspect->enode->u.event.ename->u.name.s, 2848 responseval); 2849 if (nvlist_add_boolean_value(fault, 2850 FM_SUSPECT_RESPONSE, 2851 (responseval) ? B_TRUE : B_FALSE) != 0) { 2852 out(O_DIE, "cannot add no-response to fault"); 2853 } 2854 } 2855 2856 /* add any payload properties */ 2857 lut_walk(rp->suspect->payloadprops, 2858 (lut_cb)addpayloadprop, (void *)fault); 2859 rslfree(rp); 2860 2861 /* 2862 * If "action" property exists, evaluate it; this must be done 2863 * before the allfaulty check below since some actions may 2864 * modify the asru to be used in fmd_nvl_fmri_has_fault. This 2865 * needs to be restructured if any new actions are introduced 2866 * that have effects that we do not want to be visible if 2867 * we decide not to publish in the dupclose check below. 2868 */ 2869 if ((snp = eventprop_lookup(rp->suspect, L_action)) != NULL) { 2870 struct evalue evalue; 2871 2872 out(O_ALTFP|O_NONL, 2873 "[FME%d, %s action ", fmep->id, 2874 rp->suspect->enode->u.event.ename->u.name.s); 2875 ptree_name_iter(O_ALTFP|O_NONL, snp); 2876 out(O_ALTFP, "]"); 2877 Action_nvl = fault; 2878 (void) eval_expr(snp, NULL, NULL, NULL, NULL, 2879 NULL, 0, &evalue); 2880 } 2881 2882 fmd_case_add_suspect(fmep->hdl, fmep->fmcase, fault); 2883 2884 /* 2885 * check if the asru is already marked as "faulty". 2886 */ 2887 if (allfaulty) { 2888 nvlist_t *asru; 2889 2890 out(O_ALTFP|O_VERB, "FME%d dup check ", fmep->id); 2891 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, rp->suspect); 2892 out(O_ALTFP|O_VERB|O_NONL, " "); 2893 if (nvlist_lookup_nvlist(fault, 2894 FM_FAULT_ASRU, &asru) != 0) { 2895 out(O_ALTFP|O_VERB, "NULL asru"); 2896 allfaulty = B_FALSE; 2897 } else if (fmd_nvl_fmri_has_fault(fmep->hdl, asru, 2898 FMD_HAS_FAULT_ASRU, NULL)) { 2899 out(O_ALTFP|O_VERB, "faulty"); 2900 } else { 2901 out(O_ALTFP|O_VERB, "not faulty"); 2902 allfaulty = B_FALSE; 2903 } 2904 } 2905 2906 } 2907 2908 if (!allfaulty) { 2909 /* 2910 * don't update the count stat if all asrus are already 2911 * present and unrepaired in the asru cache 2912 */ 2913 for (rp = erl; rp >= srl; rp--) { 2914 struct event *suspect = rp->suspect; 2915 2916 if (suspect == NULL) 2917 continue; 2918 2919 /* if "count" exists, increment the appropriate stat */ 2920 if ((snp = eventprop_lookup(suspect, 2921 L_count)) != NULL) { 2922 out(O_ALTFP|O_NONL, 2923 "[FME%d, %s count ", fmep->id, 2924 suspect->enode->u.event.ename->u.name.s); 2925 ptree_name_iter(O_ALTFP|O_NONL, snp); 2926 out(O_ALTFP, "]"); 2927 istat_bump(snp, 0); 2928 2929 } 2930 } 2931 istat_save(); /* write out any istat changes */ 2932 } 2933 } 2934 2935 static const char * 2936 undiag_2defect_str(int ud) 2937 { 2938 switch (ud) { 2939 case UD_VAL_MISSINGINFO: 2940 case UD_VAL_MISSINGOBS: 2941 case UD_VAL_MISSINGPATH: 2942 case UD_VAL_MISSINGZERO: 2943 case UD_VAL_BADOBS: 2944 case UD_VAL_CFGMISMATCH: 2945 return (UNDIAG_DEFECT_CHKPT); 2946 2947 case UD_VAL_BADEVENTI: 2948 case UD_VAL_BADEVENTPATH: 2949 case UD_VAL_BADEVENTCLASS: 2950 case UD_VAL_INSTFAIL: 2951 case UD_VAL_NOPATH: 2952 case UD_VAL_UNSOLVD: 2953 return (UNDIAG_DEFECT_FME); 2954 2955 case UD_VAL_MAXFME: 2956 return (UNDIAG_DEFECT_LIMIT); 2957 2958 case UD_VAL_UNKNOWN: 2959 default: 2960 return (UNDIAG_DEFECT_UNKNOWN); 2961 } 2962 } 2963 2964 static const char * 2965 undiag_2fault_str(int ud) 2966 { 2967 switch (ud) { 2968 case UD_VAL_BADEVENTI: 2969 case UD_VAL_BADEVENTPATH: 2970 case UD_VAL_BADEVENTCLASS: 2971 case UD_VAL_INSTFAIL: 2972 case UD_VAL_NOPATH: 2973 case UD_VAL_UNSOLVD: 2974 return (UNDIAG_FAULT_FME); 2975 default: 2976 return (NULL); 2977 } 2978 } 2979 2980 static char * 2981 undiag_2reason_str(int ud, char *arg) 2982 { 2983 const char *ptr; 2984 char *buf; 2985 int with_arg = 0; 2986 2987 switch (ud) { 2988 case UD_VAL_BADEVENTPATH: 2989 ptr = UD_STR_BADEVENTPATH; 2990 with_arg = 1; 2991 break; 2992 case UD_VAL_BADEVENTCLASS: 2993 ptr = UD_STR_BADEVENTCLASS; 2994 with_arg = 1; 2995 break; 2996 case UD_VAL_BADEVENTI: 2997 ptr = UD_STR_BADEVENTI; 2998 with_arg = 1; 2999 break; 3000 case UD_VAL_BADOBS: 3001 ptr = UD_STR_BADOBS; 3002 break; 3003 case UD_VAL_CFGMISMATCH: 3004 ptr = UD_STR_CFGMISMATCH; 3005 break; 3006 case UD_VAL_INSTFAIL: 3007 ptr = UD_STR_INSTFAIL; 3008 with_arg = 1; 3009 break; 3010 case UD_VAL_MAXFME: 3011 ptr = UD_STR_MAXFME; 3012 break; 3013 case UD_VAL_MISSINGINFO: 3014 ptr = UD_STR_MISSINGINFO; 3015 break; 3016 case UD_VAL_MISSINGOBS: 3017 ptr = UD_STR_MISSINGOBS; 3018 break; 3019 case UD_VAL_MISSINGPATH: 3020 ptr = UD_STR_MISSINGPATH; 3021 break; 3022 case UD_VAL_MISSINGZERO: 3023 ptr = UD_STR_MISSINGZERO; 3024 break; 3025 case UD_VAL_NOPATH: 3026 ptr = UD_STR_NOPATH; 3027 with_arg = 1; 3028 break; 3029 case UD_VAL_UNSOLVD: 3030 ptr = UD_STR_UNSOLVD; 3031 break; 3032 case UD_VAL_UNKNOWN: 3033 default: 3034 ptr = UD_STR_UNKNOWN; 3035 break; 3036 } 3037 if (with_arg) { 3038 buf = MALLOC(strlen(ptr) + strlen(arg) - 1); 3039 (void) sprintf(buf, ptr, arg); 3040 } else { 3041 buf = MALLOC(strlen(ptr) + 1); 3042 (void) sprintf(buf, ptr); 3043 } 3044 return (buf); 3045 } 3046 3047 static void 3048 publish_undiagnosable(fmd_hdl_t *hdl, fmd_event_t *ffep, fmd_case_t *fmcase, 3049 nvlist_t *detector, char *arg) 3050 { 3051 struct case_list *newcase; 3052 nvlist_t *defect, *fault; 3053 const char *faultstr; 3054 char *reason = undiag_2reason_str(Undiag_reason, arg); 3055 3056 out(O_ALTFP, 3057 "[undiagnosable ereport received, " 3058 "creating and closing a new case (%s)]", reason); 3059 3060 newcase = MALLOC(sizeof (struct case_list)); 3061 newcase->next = NULL; 3062 newcase->fmcase = fmcase; 3063 if (Undiagablecaselist != NULL) 3064 newcase->next = Undiagablecaselist; 3065 Undiagablecaselist = newcase; 3066 3067 if (ffep != NULL) 3068 fmd_case_add_ereport(hdl, newcase->fmcase, ffep); 3069 3070 /* add defect */ 3071 defect = fmd_nvl_create_fault(hdl, 3072 undiag_2defect_str(Undiag_reason), 50, NULL, NULL, detector); 3073 (void) nvlist_add_string(defect, UNDIAG_REASON, reason); 3074 (void) nvlist_add_boolean_value(defect, FM_SUSPECT_RETIRE, B_FALSE); 3075 (void) nvlist_add_boolean_value(defect, FM_SUSPECT_RESPONSE, B_FALSE); 3076 fmd_case_add_suspect(hdl, newcase->fmcase, defect); 3077 3078 /* add fault if appropriate */ 3079 faultstr = undiag_2fault_str(Undiag_reason); 3080 if (faultstr != NULL) { 3081 fault = fmd_nvl_create_fault(hdl, faultstr, 50, NULL, NULL, 3082 detector); 3083 (void) nvlist_add_string(fault, UNDIAG_REASON, reason); 3084 (void) nvlist_add_boolean_value(fault, FM_SUSPECT_RETIRE, 3085 B_FALSE); 3086 (void) nvlist_add_boolean_value(fault, FM_SUSPECT_RESPONSE, 3087 B_FALSE); 3088 fmd_case_add_suspect(hdl, newcase->fmcase, fault); 3089 } 3090 FREE(reason); 3091 3092 /* solve and close case */ 3093 fmd_case_solve(hdl, newcase->fmcase); 3094 fmd_case_close(hdl, newcase->fmcase); 3095 Undiag_reason = UD_VAL_UNKNOWN; 3096 } 3097 3098 static void 3099 fme_undiagnosable(struct fme *f) 3100 { 3101 nvlist_t *defect, *fault, *detector = NULL; 3102 struct event *ep; 3103 char *pathstr; 3104 const char *faultstr; 3105 char *reason = undiag_2reason_str(Undiag_reason, NULL); 3106 3107 out(O_ALTFP, "[solving/closing FME%d, case %s (%s)]", 3108 f->id, fmd_case_uuid(f->hdl, f->fmcase), reason); 3109 3110 for (ep = f->observations; ep; ep = ep->observations) { 3111 3112 if (ep->ffep != f->e0r) 3113 fmd_case_add_ereport(f->hdl, f->fmcase, ep->ffep); 3114 3115 pathstr = ipath2str(NULL, ipath(platform_getpath(ep->nvp))); 3116 platform_units_translate(0, f->config, NULL, NULL, &detector, 3117 pathstr); 3118 FREE(pathstr); 3119 3120 /* add defect */ 3121 defect = fmd_nvl_create_fault(f->hdl, 3122 undiag_2defect_str(Undiag_reason), 50 / f->uniqobs, 3123 NULL, NULL, detector); 3124 (void) nvlist_add_string(defect, UNDIAG_REASON, reason); 3125 (void) nvlist_add_boolean_value(defect, FM_SUSPECT_RETIRE, 3126 B_FALSE); 3127 (void) nvlist_add_boolean_value(defect, FM_SUSPECT_RESPONSE, 3128 B_FALSE); 3129 fmd_case_add_suspect(f->hdl, f->fmcase, defect); 3130 3131 /* add fault if appropriate */ 3132 faultstr = undiag_2fault_str(Undiag_reason); 3133 if (faultstr == NULL) 3134 continue; 3135 fault = fmd_nvl_create_fault(f->hdl, faultstr, 50 / f->uniqobs, 3136 NULL, NULL, detector); 3137 (void) nvlist_add_string(fault, UNDIAG_REASON, reason); 3138 (void) nvlist_add_boolean_value(fault, FM_SUSPECT_RETIRE, 3139 B_FALSE); 3140 (void) nvlist_add_boolean_value(fault, FM_SUSPECT_RESPONSE, 3141 B_FALSE); 3142 fmd_case_add_suspect(f->hdl, f->fmcase, fault); 3143 nvlist_free(detector); 3144 } 3145 FREE(reason); 3146 fmd_case_solve(f->hdl, f->fmcase); 3147 fmd_case_close(f->hdl, f->fmcase); 3148 Undiag_reason = UD_VAL_UNKNOWN; 3149 } 3150 3151 /* 3152 * fme_close_case 3153 * 3154 * Find the requested case amongst our fmes and close it. Free up 3155 * the related fme. 3156 */ 3157 void 3158 fme_close_case(fmd_hdl_t *hdl, fmd_case_t *fmcase) 3159 { 3160 struct case_list *ucasep, *prevcasep = NULL; 3161 struct fme *prev = NULL; 3162 struct fme *fmep; 3163 3164 for (ucasep = Undiagablecaselist; ucasep; ucasep = ucasep->next) { 3165 if (fmcase != ucasep->fmcase) { 3166 prevcasep = ucasep; 3167 continue; 3168 } 3169 3170 if (prevcasep == NULL) 3171 Undiagablecaselist = Undiagablecaselist->next; 3172 else 3173 prevcasep->next = ucasep->next; 3174 3175 FREE(ucasep); 3176 return; 3177 } 3178 3179 for (fmep = FMElist; fmep; fmep = fmep->next) { 3180 if (fmep->hdl == hdl && fmep->fmcase == fmcase) 3181 break; 3182 prev = fmep; 3183 } 3184 3185 if (fmep == NULL) { 3186 out(O_WARN, "Eft asked to close unrecognized case [%s].", 3187 fmd_case_uuid(hdl, fmcase)); 3188 return; 3189 } 3190 3191 if (EFMElist == fmep) 3192 EFMElist = prev; 3193 3194 if (prev == NULL) 3195 FMElist = FMElist->next; 3196 else 3197 prev->next = fmep->next; 3198 3199 fmep->next = NULL; 3200 3201 /* Get rid of any timer this fme has set */ 3202 if (fmep->wull != 0) 3203 fmd_timer_remove(fmep->hdl, fmep->timer); 3204 3205 if (ClosedFMEs == NULL) { 3206 ClosedFMEs = fmep; 3207 } else { 3208 fmep->next = ClosedFMEs; 3209 ClosedFMEs = fmep; 3210 } 3211 3212 Open_fme_count--; 3213 3214 /* See if we can close the overflow FME */ 3215 if (Open_fme_count <= Max_fme) { 3216 for (fmep = FMElist; fmep; fmep = fmep->next) { 3217 if (fmep->overflow && !(fmd_case_closed(fmep->hdl, 3218 fmep->fmcase))) 3219 break; 3220 } 3221 3222 if (fmep != NULL) 3223 fmd_case_close(fmep->hdl, fmep->fmcase); 3224 } 3225 } 3226 3227 /* 3228 * fme_set_timer() 3229 * If the time we need to wait for the given FME is less than the 3230 * current timer, kick that old timer out and establish a new one. 3231 */ 3232 static int 3233 fme_set_timer(struct fme *fmep, unsigned long long wull) 3234 { 3235 out(O_ALTFP|O_VERB|O_NONL, " fme_set_timer: request to wait "); 3236 ptree_timeval(O_ALTFP|O_VERB, &wull); 3237 3238 if (wull <= fmep->pull) { 3239 out(O_ALTFP|O_VERB|O_NONL, "already have waited at least "); 3240 ptree_timeval(O_ALTFP|O_VERB, &fmep->pull); 3241 out(O_ALTFP|O_VERB, NULL); 3242 /* we've waited at least wull already, don't need timer */ 3243 return (0); 3244 } 3245 3246 out(O_ALTFP|O_VERB|O_NONL, " currently "); 3247 if (fmep->wull != 0) { 3248 out(O_ALTFP|O_VERB|O_NONL, "waiting "); 3249 ptree_timeval(O_ALTFP|O_VERB, &fmep->wull); 3250 out(O_ALTFP|O_VERB, NULL); 3251 } else { 3252 out(O_ALTFP|O_VERB|O_NONL, "not waiting"); 3253 out(O_ALTFP|O_VERB, NULL); 3254 } 3255 3256 if (fmep->wull != 0) 3257 if (wull >= fmep->wull) 3258 /* New timer would fire later than established timer */ 3259 return (0); 3260 3261 if (fmep->wull != 0) { 3262 fmd_timer_remove(fmep->hdl, fmep->timer); 3263 } 3264 3265 fmep->timer = fmd_timer_install(fmep->hdl, (void *)fmep, 3266 fmep->e0r, wull); 3267 out(O_ALTFP|O_VERB, "timer set, id is %ld", fmep->timer); 3268 fmep->wull = wull; 3269 return (1); 3270 } 3271 3272 void 3273 fme_timer_fired(struct fme *fmep, id_t tid) 3274 { 3275 struct fme *ffmep = NULL; 3276 3277 for (ffmep = FMElist; ffmep; ffmep = ffmep->next) 3278 if (ffmep == fmep) 3279 break; 3280 3281 if (ffmep == NULL) { 3282 out(O_WARN, "Timer fired for an FME (%p) not in FMEs list.", 3283 (void *)fmep); 3284 return; 3285 } 3286 3287 out(O_ALTFP|O_VERB, "Timer fired %lx", tid); 3288 fmep->pull = fmep->wull; 3289 fmep->wull = 0; 3290 fmd_buf_write(fmep->hdl, fmep->fmcase, 3291 WOBUF_PULL, (void *)&fmep->pull, sizeof (fmep->pull)); 3292 3293 fme_eval(fmep, fmep->e0r); 3294 } 3295 3296 /* 3297 * Preserve the fme's suspect list in its psuspects list, NULLing the 3298 * suspects list in the meantime. 3299 */ 3300 static void 3301 save_suspects(struct fme *fmep) 3302 { 3303 struct event *ep; 3304 struct event *nextep; 3305 3306 /* zero out the previous suspect list */ 3307 for (ep = fmep->psuspects; ep; ep = nextep) { 3308 nextep = ep->psuspects; 3309 ep->psuspects = NULL; 3310 } 3311 fmep->psuspects = NULL; 3312 3313 /* zero out the suspect list, copying it to previous suspect list */ 3314 fmep->psuspects = fmep->suspects; 3315 for (ep = fmep->suspects; ep; ep = nextep) { 3316 nextep = ep->suspects; 3317 ep->psuspects = ep->suspects; 3318 ep->suspects = NULL; 3319 ep->is_suspect = 0; 3320 } 3321 fmep->suspects = NULL; 3322 fmep->nsuspects = 0; 3323 } 3324 3325 /* 3326 * Retrieve the fme's suspect list from its psuspects list. 3327 */ 3328 static void 3329 restore_suspects(struct fme *fmep) 3330 { 3331 struct event *ep; 3332 struct event *nextep; 3333 3334 fmep->nsuspects = 0; 3335 fmep->suspects = fmep->psuspects; 3336 for (ep = fmep->psuspects; ep; ep = nextep) { 3337 fmep->nsuspects++; 3338 nextep = ep->psuspects; 3339 ep->suspects = ep->psuspects; 3340 } 3341 } 3342 3343 /* 3344 * this is what we use to call the Emrys prototype code instead of main() 3345 */ 3346 static void 3347 fme_eval(struct fme *fmep, fmd_event_t *ffep) 3348 { 3349 struct event *ep; 3350 unsigned long long my_delay = TIMEVAL_EVENTUALLY; 3351 struct rsl *srl = NULL; 3352 struct rsl *srl2 = NULL; 3353 int mess_zero_count; 3354 int rpcnt; 3355 3356 save_suspects(fmep); 3357 3358 out(O_ALTFP, "Evaluate FME %d", fmep->id); 3359 indent_set(" "); 3360 3361 lut_walk(fmep->eventtree, (lut_cb)clear_arrows, (void *)fmep); 3362 fmep->state = hypothesise(fmep, fmep->e0, fmep->ull, &my_delay); 3363 3364 out(O_ALTFP|O_NONL, "FME%d state: %s, suspect list:", fmep->id, 3365 fme_state2str(fmep->state)); 3366 for (ep = fmep->suspects; ep; ep = ep->suspects) { 3367 out(O_ALTFP|O_NONL, " "); 3368 itree_pevent_brief(O_ALTFP|O_NONL, ep); 3369 } 3370 out(O_ALTFP, NULL); 3371 3372 switch (fmep->state) { 3373 case FME_CREDIBLE: 3374 print_suspects(SLNEW, fmep); 3375 (void) upsets_eval(fmep, ffep); 3376 3377 /* 3378 * we may have already posted suspects in upsets_eval() which 3379 * can recurse into fme_eval() again. If so then just return. 3380 */ 3381 if (fmep->posted_suspects) 3382 return; 3383 3384 stats_counter_bump(fmep->diags); 3385 rpcnt = fmep->nsuspects; 3386 save_suspects(fmep); 3387 3388 /* 3389 * create two lists, one for "message=1" faults and one for 3390 * "message=0" faults. If we have a mixture we will generate 3391 * two separate suspect lists. 3392 */ 3393 srl = MALLOC(rpcnt * sizeof (struct rsl)); 3394 bzero(srl, rpcnt * sizeof (struct rsl)); 3395 srl2 = MALLOC(rpcnt * sizeof (struct rsl)); 3396 bzero(srl2, rpcnt * sizeof (struct rsl)); 3397 mess_zero_count = trim_suspects(fmep, srl, srl2, ffep); 3398 3399 /* 3400 * If the resulting suspect list has no members, we're 3401 * done so simply close the case. Otherwise sort and publish. 3402 */ 3403 if (fmep->nsuspects == 0 && mess_zero_count == 0) { 3404 out(O_ALTFP, 3405 "[FME%d, case %s (all suspects are upsets)]", 3406 fmep->id, fmd_case_uuid(fmep->hdl, fmep->fmcase)); 3407 fmd_case_close(fmep->hdl, fmep->fmcase); 3408 } else if (fmep->nsuspects != 0 && mess_zero_count == 0) { 3409 publish_suspects(fmep, srl); 3410 out(O_ALTFP, "[solving FME%d, case %s]", fmep->id, 3411 fmd_case_uuid(fmep->hdl, fmep->fmcase)); 3412 fmd_case_solve(fmep->hdl, fmep->fmcase); 3413 } else if (fmep->nsuspects == 0 && mess_zero_count != 0) { 3414 fmep->nsuspects = mess_zero_count; 3415 publish_suspects(fmep, srl2); 3416 out(O_ALTFP, "[solving FME%d, case %s]", fmep->id, 3417 fmd_case_uuid(fmep->hdl, fmep->fmcase)); 3418 fmd_case_solve(fmep->hdl, fmep->fmcase); 3419 } else { 3420 struct event *obsp; 3421 struct fme *nfmep; 3422 3423 publish_suspects(fmep, srl); 3424 out(O_ALTFP, "[solving FME%d, case %s]", fmep->id, 3425 fmd_case_uuid(fmep->hdl, fmep->fmcase)); 3426 fmd_case_solve(fmep->hdl, fmep->fmcase); 3427 3428 /* 3429 * Got both message=0 and message=1 so create a 3430 * duplicate case. Also need a temporary duplicate fme 3431 * structure for use by publish_suspects(). 3432 */ 3433 nfmep = alloc_fme(); 3434 nfmep->id = Nextid++; 3435 nfmep->hdl = fmep->hdl; 3436 nfmep->nsuspects = mess_zero_count; 3437 nfmep->fmcase = fmd_case_open(fmep->hdl, NULL); 3438 out(O_ALTFP|O_STAMP, 3439 "[creating parallel FME%d, case %s]", nfmep->id, 3440 fmd_case_uuid(nfmep->hdl, nfmep->fmcase)); 3441 Open_fme_count++; 3442 if (ffep) { 3443 fmd_case_setprincipal(nfmep->hdl, 3444 nfmep->fmcase, ffep); 3445 fmd_case_add_ereport(nfmep->hdl, 3446 nfmep->fmcase, ffep); 3447 } 3448 for (obsp = fmep->observations; obsp; 3449 obsp = obsp->observations) 3450 if (obsp->ffep && obsp->ffep != ffep) 3451 fmd_case_add_ereport(nfmep->hdl, 3452 nfmep->fmcase, obsp->ffep); 3453 3454 publish_suspects(nfmep, srl2); 3455 out(O_ALTFP, "[solving FME%d, case %s]", nfmep->id, 3456 fmd_case_uuid(nfmep->hdl, nfmep->fmcase)); 3457 fmd_case_solve(nfmep->hdl, nfmep->fmcase); 3458 FREE(nfmep); 3459 } 3460 FREE(srl); 3461 FREE(srl2); 3462 restore_suspects(fmep); 3463 3464 fmep->posted_suspects = 1; 3465 fmd_buf_write(fmep->hdl, fmep->fmcase, 3466 WOBUF_POSTD, 3467 (void *)&fmep->posted_suspects, 3468 sizeof (fmep->posted_suspects)); 3469 3470 /* 3471 * Now the suspects have been posted, we can clear up 3472 * the instance tree as we won't be looking at it again. 3473 * Also cancel the timer as the case is now solved. 3474 */ 3475 if (fmep->wull != 0) { 3476 fmd_timer_remove(fmep->hdl, fmep->timer); 3477 fmep->wull = 0; 3478 } 3479 break; 3480 3481 case FME_WAIT: 3482 ASSERT(my_delay > fmep->ull); 3483 (void) fme_set_timer(fmep, my_delay); 3484 print_suspects(SLWAIT, fmep); 3485 itree_prune(fmep->eventtree); 3486 return; 3487 3488 case FME_DISPROVED: 3489 print_suspects(SLDISPROVED, fmep); 3490 Undiag_reason = UD_VAL_UNSOLVD; 3491 fme_undiagnosable(fmep); 3492 break; 3493 } 3494 3495 itree_free(fmep->eventtree); 3496 fmep->eventtree = NULL; 3497 structconfig_free(fmep->config); 3498 fmep->config = NULL; 3499 destroy_fme_bufs(fmep); 3500 } 3501 3502 static void indent(void); 3503 static int triggered(struct fme *fmep, struct event *ep, int mark); 3504 static enum fme_state effects_test(struct fme *fmep, 3505 struct event *fault_event, unsigned long long at_latest_by, 3506 unsigned long long *pdelay); 3507 static enum fme_state requirements_test(struct fme *fmep, struct event *ep, 3508 unsigned long long at_latest_by, unsigned long long *pdelay); 3509 static enum fme_state causes_test(struct fme *fmep, struct event *ep, 3510 unsigned long long at_latest_by, unsigned long long *pdelay); 3511 3512 static int 3513 checkconstraints(struct fme *fmep, struct arrow *arrowp) 3514 { 3515 struct constraintlist *ctp; 3516 struct evalue value; 3517 char *sep = ""; 3518 3519 if (arrowp->forever_false) { 3520 indent(); 3521 out(O_ALTFP|O_VERB|O_NONL, " Forever false constraint: "); 3522 for (ctp = arrowp->constraints; ctp != NULL; ctp = ctp->next) { 3523 out(O_ALTFP|O_VERB|O_NONL, sep); 3524 ptree(O_ALTFP|O_VERB|O_NONL, ctp->cnode, 1, 0); 3525 sep = ", "; 3526 } 3527 out(O_ALTFP|O_VERB, NULL); 3528 return (0); 3529 } 3530 if (arrowp->forever_true) { 3531 indent(); 3532 out(O_ALTFP|O_VERB|O_NONL, " Forever true constraint: "); 3533 for (ctp = arrowp->constraints; ctp != NULL; ctp = ctp->next) { 3534 out(O_ALTFP|O_VERB|O_NONL, sep); 3535 ptree(O_ALTFP|O_VERB|O_NONL, ctp->cnode, 1, 0); 3536 sep = ", "; 3537 } 3538 out(O_ALTFP|O_VERB, NULL); 3539 return (1); 3540 } 3541 3542 for (ctp = arrowp->constraints; ctp != NULL; ctp = ctp->next) { 3543 if (eval_expr(ctp->cnode, NULL, NULL, 3544 &fmep->globals, fmep->config, 3545 arrowp, 0, &value)) { 3546 /* evaluation successful */ 3547 if (value.t == UNDEFINED || value.v == 0) { 3548 /* known false */ 3549 arrowp->forever_false = 1; 3550 indent(); 3551 out(O_ALTFP|O_VERB|O_NONL, 3552 " False constraint: "); 3553 ptree(O_ALTFP|O_VERB|O_NONL, ctp->cnode, 1, 0); 3554 out(O_ALTFP|O_VERB, NULL); 3555 return (0); 3556 } 3557 } else { 3558 /* evaluation unsuccessful -- unknown value */ 3559 indent(); 3560 out(O_ALTFP|O_VERB|O_NONL, 3561 " Deferred constraint: "); 3562 ptree(O_ALTFP|O_VERB|O_NONL, ctp->cnode, 1, 0); 3563 out(O_ALTFP|O_VERB, NULL); 3564 return (1); 3565 } 3566 } 3567 /* known true */ 3568 arrowp->forever_true = 1; 3569 indent(); 3570 out(O_ALTFP|O_VERB|O_NONL, " True constraint: "); 3571 for (ctp = arrowp->constraints; ctp != NULL; ctp = ctp->next) { 3572 out(O_ALTFP|O_VERB|O_NONL, sep); 3573 ptree(O_ALTFP|O_VERB|O_NONL, ctp->cnode, 1, 0); 3574 sep = ", "; 3575 } 3576 out(O_ALTFP|O_VERB, NULL); 3577 return (1); 3578 } 3579 3580 static int 3581 triggered(struct fme *fmep, struct event *ep, int mark) 3582 { 3583 struct bubble *bp; 3584 struct arrowlist *ap; 3585 int count = 0; 3586 3587 stats_counter_bump(fmep->Tcallcount); 3588 for (bp = itree_next_bubble(ep, NULL); bp; 3589 bp = itree_next_bubble(ep, bp)) { 3590 if (bp->t != B_TO) 3591 continue; 3592 for (ap = itree_next_arrow(bp, NULL); ap; 3593 ap = itree_next_arrow(bp, ap)) { 3594 /* check count of marks against K in the bubble */ 3595 if ((ap->arrowp->mark & mark) && 3596 ++count >= bp->nork) 3597 return (1); 3598 } 3599 } 3600 return (0); 3601 } 3602 3603 static int 3604 mark_arrows(struct fme *fmep, struct event *ep, int mark, 3605 unsigned long long at_latest_by, unsigned long long *pdelay, int keep) 3606 { 3607 struct bubble *bp; 3608 struct arrowlist *ap; 3609 unsigned long long overall_delay = TIMEVAL_EVENTUALLY; 3610 unsigned long long my_delay; 3611 enum fme_state result; 3612 int retval = 0; 3613 3614 for (bp = itree_next_bubble(ep, NULL); bp; 3615 bp = itree_next_bubble(ep, bp)) { 3616 if (bp->t != B_FROM) 3617 continue; 3618 stats_counter_bump(fmep->Marrowcount); 3619 for (ap = itree_next_arrow(bp, NULL); ap; 3620 ap = itree_next_arrow(bp, ap)) { 3621 struct event *ep2 = ap->arrowp->head->myevent; 3622 /* 3623 * if we're clearing marks, we can avoid doing 3624 * all that work evaluating constraints. 3625 */ 3626 if (mark == 0) { 3627 if (ap->arrowp->arrow_marked == 0) 3628 continue; 3629 ap->arrowp->arrow_marked = 0; 3630 ap->arrowp->mark &= ~EFFECTS_COUNTER; 3631 if (keep && (ep2->cached_state & 3632 (WAIT_EFFECT|CREDIBLE_EFFECT|PARENT_WAIT))) 3633 ep2->keep_in_tree = 1; 3634 ep2->cached_state &= 3635 ~(WAIT_EFFECT|CREDIBLE_EFFECT|PARENT_WAIT); 3636 (void) mark_arrows(fmep, ep2, mark, 0, NULL, 3637 keep); 3638 continue; 3639 } 3640 ap->arrowp->arrow_marked = 1; 3641 if (ep2->cached_state & REQMNTS_DISPROVED) { 3642 indent(); 3643 out(O_ALTFP|O_VERB|O_NONL, 3644 " ALREADY DISPROVED "); 3645 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 3646 out(O_ALTFP|O_VERB, NULL); 3647 continue; 3648 } 3649 if (ep2->cached_state & WAIT_EFFECT) { 3650 indent(); 3651 out(O_ALTFP|O_VERB|O_NONL, 3652 " ALREADY EFFECTS WAIT "); 3653 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 3654 out(O_ALTFP|O_VERB, NULL); 3655 continue; 3656 } 3657 if (ep2->cached_state & CREDIBLE_EFFECT) { 3658 indent(); 3659 out(O_ALTFP|O_VERB|O_NONL, 3660 " ALREADY EFFECTS CREDIBLE "); 3661 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 3662 out(O_ALTFP|O_VERB, NULL); 3663 continue; 3664 } 3665 if ((ep2->cached_state & PARENT_WAIT) && 3666 (mark & PARENT_WAIT)) { 3667 indent(); 3668 out(O_ALTFP|O_VERB|O_NONL, 3669 " ALREADY PARENT EFFECTS WAIT "); 3670 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 3671 out(O_ALTFP|O_VERB, NULL); 3672 continue; 3673 } 3674 platform_set_payloadnvp(ep2->nvp); 3675 if (checkconstraints(fmep, ap->arrowp) == 0) { 3676 platform_set_payloadnvp(NULL); 3677 indent(); 3678 out(O_ALTFP|O_VERB|O_NONL, 3679 " CONSTRAINTS FAIL "); 3680 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 3681 out(O_ALTFP|O_VERB, NULL); 3682 continue; 3683 } 3684 platform_set_payloadnvp(NULL); 3685 ap->arrowp->mark |= EFFECTS_COUNTER; 3686 if (!triggered(fmep, ep2, EFFECTS_COUNTER)) { 3687 indent(); 3688 out(O_ALTFP|O_VERB|O_NONL, 3689 " K-COUNT NOT YET MET "); 3690 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 3691 out(O_ALTFP|O_VERB, NULL); 3692 continue; 3693 } 3694 ep2->cached_state &= ~PARENT_WAIT; 3695 /* 3696 * if we've reached an ereport and no propagation time 3697 * is specified, use the Hesitate value 3698 */ 3699 if (ep2->t == N_EREPORT && at_latest_by == 0ULL && 3700 ap->arrowp->maxdelay == 0ULL) { 3701 out(O_ALTFP|O_VERB|O_NONL, " default wait "); 3702 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 3703 out(O_ALTFP|O_VERB, NULL); 3704 result = requirements_test(fmep, ep2, Hesitate, 3705 &my_delay); 3706 } else { 3707 result = requirements_test(fmep, ep2, 3708 at_latest_by + ap->arrowp->maxdelay, 3709 &my_delay); 3710 } 3711 if (result == FME_WAIT) { 3712 retval = WAIT_EFFECT; 3713 if (overall_delay > my_delay) 3714 overall_delay = my_delay; 3715 ep2->cached_state |= WAIT_EFFECT; 3716 indent(); 3717 out(O_ALTFP|O_VERB|O_NONL, " EFFECTS WAIT "); 3718 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 3719 out(O_ALTFP|O_VERB, NULL); 3720 indent_push(" E"); 3721 if (mark_arrows(fmep, ep2, PARENT_WAIT, 3722 at_latest_by, &my_delay, 0) == 3723 WAIT_EFFECT) { 3724 retval = WAIT_EFFECT; 3725 if (overall_delay > my_delay) 3726 overall_delay = my_delay; 3727 } 3728 indent_pop(); 3729 } else if (result == FME_DISPROVED) { 3730 indent(); 3731 out(O_ALTFP|O_VERB|O_NONL, 3732 " EFFECTS DISPROVED "); 3733 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 3734 out(O_ALTFP|O_VERB, NULL); 3735 } else { 3736 ep2->cached_state |= mark; 3737 indent(); 3738 if (mark == CREDIBLE_EFFECT) 3739 out(O_ALTFP|O_VERB|O_NONL, 3740 " EFFECTS CREDIBLE "); 3741 else 3742 out(O_ALTFP|O_VERB|O_NONL, 3743 " PARENT EFFECTS WAIT "); 3744 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 3745 out(O_ALTFP|O_VERB, NULL); 3746 indent_push(" E"); 3747 if (mark_arrows(fmep, ep2, mark, at_latest_by, 3748 &my_delay, 0) == WAIT_EFFECT) { 3749 retval = WAIT_EFFECT; 3750 if (overall_delay > my_delay) 3751 overall_delay = my_delay; 3752 } 3753 indent_pop(); 3754 } 3755 } 3756 } 3757 if (retval == WAIT_EFFECT) 3758 *pdelay = overall_delay; 3759 return (retval); 3760 } 3761 3762 static enum fme_state 3763 effects_test(struct fme *fmep, struct event *fault_event, 3764 unsigned long long at_latest_by, unsigned long long *pdelay) 3765 { 3766 struct event *error_event; 3767 enum fme_state return_value = FME_CREDIBLE; 3768 unsigned long long overall_delay = TIMEVAL_EVENTUALLY; 3769 unsigned long long my_delay; 3770 3771 stats_counter_bump(fmep->Ecallcount); 3772 indent_push(" E"); 3773 indent(); 3774 out(O_ALTFP|O_VERB|O_NONL, "->"); 3775 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, fault_event); 3776 out(O_ALTFP|O_VERB, NULL); 3777 3778 if (mark_arrows(fmep, fault_event, CREDIBLE_EFFECT, at_latest_by, 3779 &my_delay, 0) == WAIT_EFFECT) { 3780 return_value = FME_WAIT; 3781 if (overall_delay > my_delay) 3782 overall_delay = my_delay; 3783 } 3784 for (error_event = fmep->observations; 3785 error_event; error_event = error_event->observations) { 3786 indent(); 3787 out(O_ALTFP|O_VERB|O_NONL, " "); 3788 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, error_event); 3789 if (!(error_event->cached_state & CREDIBLE_EFFECT)) { 3790 if (error_event->cached_state & 3791 (PARENT_WAIT|WAIT_EFFECT)) { 3792 out(O_ALTFP|O_VERB, " NOT YET triggered"); 3793 continue; 3794 } 3795 return_value = FME_DISPROVED; 3796 out(O_ALTFP|O_VERB, " NOT triggered"); 3797 break; 3798 } else { 3799 out(O_ALTFP|O_VERB, " triggered"); 3800 } 3801 } 3802 if (return_value == FME_DISPROVED) { 3803 (void) mark_arrows(fmep, fault_event, 0, 0, NULL, 0); 3804 } else { 3805 fault_event->keep_in_tree = 1; 3806 (void) mark_arrows(fmep, fault_event, 0, 0, NULL, 1); 3807 } 3808 3809 indent(); 3810 out(O_ALTFP|O_VERB|O_NONL, "<-EFFECTS %s ", 3811 fme_state2str(return_value)); 3812 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, fault_event); 3813 out(O_ALTFP|O_VERB, NULL); 3814 indent_pop(); 3815 if (return_value == FME_WAIT) 3816 *pdelay = overall_delay; 3817 return (return_value); 3818 } 3819 3820 static enum fme_state 3821 requirements_test(struct fme *fmep, struct event *ep, 3822 unsigned long long at_latest_by, unsigned long long *pdelay) 3823 { 3824 int waiting_events; 3825 int credible_events; 3826 int deferred_events; 3827 enum fme_state return_value = FME_CREDIBLE; 3828 unsigned long long overall_delay = TIMEVAL_EVENTUALLY; 3829 unsigned long long arrow_delay; 3830 unsigned long long my_delay; 3831 struct event *ep2; 3832 struct bubble *bp; 3833 struct arrowlist *ap; 3834 3835 if (ep->cached_state & REQMNTS_CREDIBLE) { 3836 indent(); 3837 out(O_ALTFP|O_VERB|O_NONL, " REQMNTS ALREADY CREDIBLE "); 3838 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3839 out(O_ALTFP|O_VERB, NULL); 3840 return (FME_CREDIBLE); 3841 } 3842 if (ep->cached_state & REQMNTS_DISPROVED) { 3843 indent(); 3844 out(O_ALTFP|O_VERB|O_NONL, " REQMNTS ALREADY DISPROVED "); 3845 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3846 out(O_ALTFP|O_VERB, NULL); 3847 return (FME_DISPROVED); 3848 } 3849 if (ep->cached_state & REQMNTS_WAIT) { 3850 indent(); 3851 *pdelay = ep->cached_delay; 3852 out(O_ALTFP|O_VERB|O_NONL, " REQMNTS ALREADY WAIT "); 3853 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3854 out(O_ALTFP|O_VERB|O_NONL, ", wait for: "); 3855 ptree_timeval(O_ALTFP|O_VERB|O_NONL, &at_latest_by); 3856 out(O_ALTFP|O_VERB, NULL); 3857 return (FME_WAIT); 3858 } 3859 stats_counter_bump(fmep->Rcallcount); 3860 indent_push(" R"); 3861 indent(); 3862 out(O_ALTFP|O_VERB|O_NONL, "->"); 3863 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3864 out(O_ALTFP|O_VERB|O_NONL, ", at latest by: "); 3865 ptree_timeval(O_ALTFP|O_VERB|O_NONL, &at_latest_by); 3866 out(O_ALTFP|O_VERB, NULL); 3867 3868 if (ep->t == N_EREPORT) { 3869 if (ep->count == 0) { 3870 if (fmep->pull >= at_latest_by) { 3871 return_value = FME_DISPROVED; 3872 } else { 3873 ep->cached_delay = *pdelay = at_latest_by; 3874 return_value = FME_WAIT; 3875 } 3876 } 3877 3878 indent(); 3879 switch (return_value) { 3880 case FME_CREDIBLE: 3881 ep->cached_state |= REQMNTS_CREDIBLE; 3882 out(O_ALTFP|O_VERB|O_NONL, "<-REQMNTS CREDIBLE "); 3883 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3884 break; 3885 case FME_DISPROVED: 3886 ep->cached_state |= REQMNTS_DISPROVED; 3887 out(O_ALTFP|O_VERB|O_NONL, "<-REQMNTS DISPROVED "); 3888 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3889 break; 3890 case FME_WAIT: 3891 ep->cached_state |= REQMNTS_WAIT; 3892 out(O_ALTFP|O_VERB|O_NONL, "<-REQMNTS WAIT "); 3893 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3894 out(O_ALTFP|O_VERB|O_NONL, " to "); 3895 ptree_timeval(O_ALTFP|O_VERB|O_NONL, &at_latest_by); 3896 break; 3897 default: 3898 out(O_DIE, "requirements_test: unexpected fme_state"); 3899 break; 3900 } 3901 out(O_ALTFP|O_VERB, NULL); 3902 indent_pop(); 3903 3904 return (return_value); 3905 } 3906 3907 /* this event is not a report, descend the tree */ 3908 for (bp = itree_next_bubble(ep, NULL); bp; 3909 bp = itree_next_bubble(ep, bp)) { 3910 int n; 3911 3912 if (bp->t != B_FROM) 3913 continue; 3914 3915 n = bp->nork; 3916 3917 credible_events = 0; 3918 waiting_events = 0; 3919 deferred_events = 0; 3920 arrow_delay = TIMEVAL_EVENTUALLY; 3921 /* 3922 * n is -1 for 'A' so adjust it. 3923 * XXX just count up the arrows for now. 3924 */ 3925 if (n < 0) { 3926 n = 0; 3927 for (ap = itree_next_arrow(bp, NULL); ap; 3928 ap = itree_next_arrow(bp, ap)) 3929 n++; 3930 indent(); 3931 out(O_ALTFP|O_VERB, " Bubble Counted N=%d", n); 3932 } else { 3933 indent(); 3934 out(O_ALTFP|O_VERB, " Bubble N=%d", n); 3935 } 3936 3937 if (n == 0) 3938 continue; 3939 if (!(bp->mark & (BUBBLE_ELIDED|BUBBLE_OK))) { 3940 for (ap = itree_next_arrow(bp, NULL); ap; 3941 ap = itree_next_arrow(bp, ap)) { 3942 ep2 = ap->arrowp->head->myevent; 3943 platform_set_payloadnvp(ep2->nvp); 3944 (void) checkconstraints(fmep, ap->arrowp); 3945 if (!ap->arrowp->forever_false) { 3946 /* 3947 * if all arrows are invalidated by the 3948 * constraints, then we should elide the 3949 * whole bubble to be consistant with 3950 * the tree creation time behaviour 3951 */ 3952 bp->mark |= BUBBLE_OK; 3953 platform_set_payloadnvp(NULL); 3954 break; 3955 } 3956 platform_set_payloadnvp(NULL); 3957 } 3958 } 3959 for (ap = itree_next_arrow(bp, NULL); ap; 3960 ap = itree_next_arrow(bp, ap)) { 3961 ep2 = ap->arrowp->head->myevent; 3962 if (n <= credible_events) 3963 break; 3964 3965 ap->arrowp->mark |= REQMNTS_COUNTER; 3966 if (triggered(fmep, ep2, REQMNTS_COUNTER)) 3967 /* XXX adding max timevals! */ 3968 switch (requirements_test(fmep, ep2, 3969 at_latest_by + ap->arrowp->maxdelay, 3970 &my_delay)) { 3971 case FME_DEFERRED: 3972 deferred_events++; 3973 break; 3974 case FME_CREDIBLE: 3975 credible_events++; 3976 break; 3977 case FME_DISPROVED: 3978 break; 3979 case FME_WAIT: 3980 if (my_delay < arrow_delay) 3981 arrow_delay = my_delay; 3982 waiting_events++; 3983 break; 3984 default: 3985 out(O_DIE, 3986 "Bug in requirements_test."); 3987 } 3988 else 3989 deferred_events++; 3990 } 3991 if (!(bp->mark & BUBBLE_OK) && waiting_events == 0) { 3992 bp->mark |= BUBBLE_ELIDED; 3993 continue; 3994 } 3995 indent(); 3996 out(O_ALTFP|O_VERB, " Credible: %d Waiting %d", 3997 credible_events + deferred_events, waiting_events); 3998 if (credible_events + deferred_events + waiting_events < n) { 3999 /* Can never meet requirements */ 4000 ep->cached_state |= REQMNTS_DISPROVED; 4001 indent(); 4002 out(O_ALTFP|O_VERB|O_NONL, "<-REQMNTS DISPROVED "); 4003 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 4004 out(O_ALTFP|O_VERB, NULL); 4005 indent_pop(); 4006 return (FME_DISPROVED); 4007 } 4008 if (credible_events + deferred_events < n) { 4009 /* will have to wait */ 4010 /* wait time is shortest known */ 4011 if (arrow_delay < overall_delay) 4012 overall_delay = arrow_delay; 4013 return_value = FME_WAIT; 4014 } else if (credible_events < n) { 4015 if (return_value != FME_WAIT) 4016 return_value = FME_DEFERRED; 4017 } 4018 } 4019 4020 /* 4021 * don't mark as FME_DEFERRED. If this event isn't reached by another 4022 * path, then this will be considered FME_CREDIBLE. But if it is 4023 * reached by a different path so the K-count is met, then might 4024 * get overridden by FME_WAIT or FME_DISPROVED. 4025 */ 4026 if (return_value == FME_WAIT) { 4027 ep->cached_state |= REQMNTS_WAIT; 4028 ep->cached_delay = *pdelay = overall_delay; 4029 } else if (return_value == FME_CREDIBLE) { 4030 ep->cached_state |= REQMNTS_CREDIBLE; 4031 } 4032 indent(); 4033 out(O_ALTFP|O_VERB|O_NONL, "<-REQMNTS %s ", 4034 fme_state2str(return_value)); 4035 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 4036 out(O_ALTFP|O_VERB, NULL); 4037 indent_pop(); 4038 return (return_value); 4039 } 4040 4041 static enum fme_state 4042 causes_test(struct fme *fmep, struct event *ep, 4043 unsigned long long at_latest_by, unsigned long long *pdelay) 4044 { 4045 unsigned long long overall_delay = TIMEVAL_EVENTUALLY; 4046 unsigned long long my_delay; 4047 int credible_results = 0; 4048 int waiting_results = 0; 4049 enum fme_state fstate; 4050 struct event *tail_event; 4051 struct bubble *bp; 4052 struct arrowlist *ap; 4053 int k = 1; 4054 4055 stats_counter_bump(fmep->Ccallcount); 4056 indent_push(" C"); 4057 indent(); 4058 out(O_ALTFP|O_VERB|O_NONL, "->"); 4059 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 4060 out(O_ALTFP|O_VERB, NULL); 4061 4062 for (bp = itree_next_bubble(ep, NULL); bp; 4063 bp = itree_next_bubble(ep, bp)) { 4064 if (bp->t != B_TO) 4065 continue; 4066 k = bp->nork; /* remember the K value */ 4067 for (ap = itree_next_arrow(bp, NULL); ap; 4068 ap = itree_next_arrow(bp, ap)) { 4069 int do_not_follow = 0; 4070 4071 /* 4072 * if we get to the same event multiple times 4073 * only worry about the first one. 4074 */ 4075 if (ap->arrowp->tail->myevent->cached_state & 4076 CAUSES_TESTED) { 4077 indent(); 4078 out(O_ALTFP|O_VERB|O_NONL, 4079 " causes test already run for "); 4080 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, 4081 ap->arrowp->tail->myevent); 4082 out(O_ALTFP|O_VERB, NULL); 4083 continue; 4084 } 4085 4086 /* 4087 * see if false constraint prevents us 4088 * from traversing this arrow 4089 */ 4090 platform_set_payloadnvp(ep->nvp); 4091 if (checkconstraints(fmep, ap->arrowp) == 0) 4092 do_not_follow = 1; 4093 platform_set_payloadnvp(NULL); 4094 if (do_not_follow) { 4095 indent(); 4096 out(O_ALTFP|O_VERB|O_NONL, 4097 " False arrow from "); 4098 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, 4099 ap->arrowp->tail->myevent); 4100 out(O_ALTFP|O_VERB, NULL); 4101 continue; 4102 } 4103 4104 ap->arrowp->tail->myevent->cached_state |= 4105 CAUSES_TESTED; 4106 tail_event = ap->arrowp->tail->myevent; 4107 fstate = hypothesise(fmep, tail_event, at_latest_by, 4108 &my_delay); 4109 4110 switch (fstate) { 4111 case FME_WAIT: 4112 if (my_delay < overall_delay) 4113 overall_delay = my_delay; 4114 waiting_results++; 4115 break; 4116 case FME_CREDIBLE: 4117 credible_results++; 4118 break; 4119 case FME_DISPROVED: 4120 break; 4121 default: 4122 out(O_DIE, "Bug in causes_test"); 4123 } 4124 } 4125 } 4126 /* compare against K */ 4127 if (credible_results + waiting_results < k) { 4128 indent(); 4129 out(O_ALTFP|O_VERB|O_NONL, "<-CAUSES DISPROVED "); 4130 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 4131 out(O_ALTFP|O_VERB, NULL); 4132 indent_pop(); 4133 return (FME_DISPROVED); 4134 } 4135 if (waiting_results != 0) { 4136 *pdelay = overall_delay; 4137 indent(); 4138 out(O_ALTFP|O_VERB|O_NONL, "<-CAUSES WAIT "); 4139 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 4140 out(O_ALTFP|O_VERB|O_NONL, " to "); 4141 ptree_timeval(O_ALTFP|O_VERB|O_NONL, &at_latest_by); 4142 out(O_ALTFP|O_VERB, NULL); 4143 indent_pop(); 4144 return (FME_WAIT); 4145 } 4146 indent(); 4147 out(O_ALTFP|O_VERB|O_NONL, "<-CAUSES CREDIBLE "); 4148 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 4149 out(O_ALTFP|O_VERB, NULL); 4150 indent_pop(); 4151 return (FME_CREDIBLE); 4152 } 4153 4154 static enum fme_state 4155 hypothesise(struct fme *fmep, struct event *ep, 4156 unsigned long long at_latest_by, unsigned long long *pdelay) 4157 { 4158 enum fme_state rtr, otr; 4159 unsigned long long my_delay; 4160 unsigned long long overall_delay = TIMEVAL_EVENTUALLY; 4161 4162 stats_counter_bump(fmep->Hcallcount); 4163 indent_push(" H"); 4164 indent(); 4165 out(O_ALTFP|O_VERB|O_NONL, "->"); 4166 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 4167 out(O_ALTFP|O_VERB|O_NONL, ", at latest by: "); 4168 ptree_timeval(O_ALTFP|O_VERB|O_NONL, &at_latest_by); 4169 out(O_ALTFP|O_VERB, NULL); 4170 4171 rtr = requirements_test(fmep, ep, at_latest_by, &my_delay); 4172 if ((rtr == FME_WAIT) && (my_delay < overall_delay)) 4173 overall_delay = my_delay; 4174 if (rtr != FME_DISPROVED) { 4175 if (is_problem(ep->t)) { 4176 otr = effects_test(fmep, ep, at_latest_by, &my_delay); 4177 if (otr != FME_DISPROVED) { 4178 if (fmep->peek == 0 && ep->is_suspect == 0) { 4179 ep->suspects = fmep->suspects; 4180 ep->is_suspect = 1; 4181 fmep->suspects = ep; 4182 fmep->nsuspects++; 4183 } 4184 } 4185 } else 4186 otr = causes_test(fmep, ep, at_latest_by, &my_delay); 4187 if ((otr == FME_WAIT) && (my_delay < overall_delay)) 4188 overall_delay = my_delay; 4189 if ((otr != FME_DISPROVED) && 4190 ((rtr == FME_WAIT) || (otr == FME_WAIT))) 4191 *pdelay = overall_delay; 4192 } 4193 if (rtr == FME_DISPROVED) { 4194 indent(); 4195 out(O_ALTFP|O_VERB|O_NONL, "<-DISPROVED "); 4196 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 4197 out(O_ALTFP|O_VERB, " (doesn't meet requirements)"); 4198 indent_pop(); 4199 return (FME_DISPROVED); 4200 } 4201 if ((otr == FME_DISPROVED) && is_problem(ep->t)) { 4202 indent(); 4203 out(O_ALTFP|O_VERB|O_NONL, "<-DISPROVED "); 4204 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 4205 out(O_ALTFP|O_VERB, " (doesn't explain all reports)"); 4206 indent_pop(); 4207 return (FME_DISPROVED); 4208 } 4209 if (otr == FME_DISPROVED) { 4210 indent(); 4211 out(O_ALTFP|O_VERB|O_NONL, "<-DISPROVED "); 4212 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 4213 out(O_ALTFP|O_VERB, " (causes are not credible)"); 4214 indent_pop(); 4215 return (FME_DISPROVED); 4216 } 4217 if ((rtr == FME_WAIT) || (otr == FME_WAIT)) { 4218 indent(); 4219 out(O_ALTFP|O_VERB|O_NONL, "<-WAIT "); 4220 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 4221 out(O_ALTFP|O_VERB|O_NONL, " to "); 4222 ptree_timeval(O_ALTFP|O_VERB|O_NONL, &overall_delay); 4223 out(O_ALTFP|O_VERB, NULL); 4224 indent_pop(); 4225 return (FME_WAIT); 4226 } 4227 indent(); 4228 out(O_ALTFP|O_VERB|O_NONL, "<-CREDIBLE "); 4229 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 4230 out(O_ALTFP|O_VERB, NULL); 4231 indent_pop(); 4232 return (FME_CREDIBLE); 4233 } 4234 4235 /* 4236 * fme_istat_load -- reconstitute any persistent istats 4237 */ 4238 void 4239 fme_istat_load(fmd_hdl_t *hdl) 4240 { 4241 int sz; 4242 char *sbuf; 4243 char *ptr; 4244 4245 if ((sz = fmd_buf_size(hdl, NULL, WOBUF_ISTATS)) == 0) { 4246 out(O_ALTFP, "fme_istat_load: No stats"); 4247 return; 4248 } 4249 4250 sbuf = alloca(sz); 4251 4252 fmd_buf_read(hdl, NULL, WOBUF_ISTATS, sbuf, sz); 4253 4254 /* 4255 * pick apart the serialized stats 4256 * 4257 * format is: 4258 * <class-name>, '@', <path>, '\0', <value>, '\0' 4259 * for example: 4260 * "stat.first@stat0/path0\02\0stat.second@stat0/path1\023\0" 4261 * 4262 * since this is parsing our own serialized data, any parsing issues 4263 * are fatal, so we check for them all with ASSERT() below. 4264 */ 4265 ptr = sbuf; 4266 while (ptr < &sbuf[sz]) { 4267 char *sepptr; 4268 struct node *np; 4269 int val; 4270 4271 sepptr = strchr(ptr, '@'); 4272 ASSERT(sepptr != NULL); 4273 *sepptr = '\0'; 4274 4275 /* construct the event */ 4276 np = newnode(T_EVENT, NULL, 0); 4277 np->u.event.ename = newnode(T_NAME, NULL, 0); 4278 np->u.event.ename->u.name.t = N_STAT; 4279 np->u.event.ename->u.name.s = stable(ptr); 4280 np->u.event.ename->u.name.it = IT_ENAME; 4281 np->u.event.ename->u.name.last = np->u.event.ename; 4282 4283 ptr = sepptr + 1; 4284 ASSERT(ptr < &sbuf[sz]); 4285 ptr += strlen(ptr); 4286 ptr++; /* move past the '\0' separating path from value */ 4287 ASSERT(ptr < &sbuf[sz]); 4288 ASSERT(isdigit(*ptr)); 4289 val = atoi(ptr); 4290 ASSERT(val > 0); 4291 ptr += strlen(ptr); 4292 ptr++; /* move past the final '\0' for this entry */ 4293 4294 np->u.event.epname = pathstring2epnamenp(sepptr + 1); 4295 ASSERT(np->u.event.epname != NULL); 4296 4297 istat_bump(np, val); 4298 tree_free(np); 4299 } 4300 4301 istat_save(); 4302 } 4303