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