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