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, Version 1.0 only 6 * (the "License"). You may not use this file except in compliance 7 * with the License. 8 * 9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 10 * or http://www.opensolaris.org/os/licensing. 11 * See the License for the specific language governing permissions 12 * and limitations under the License. 13 * 14 * When distributing Covered Code, include this CDDL HEADER in each 15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 16 * If applicable, add the following below this CDDL HEADER, with the 17 * fields enclosed by brackets "[]" replaced with your own identifying 18 * information: Portions Copyright [yyyy] [name of copyright owner] 19 * 20 * CDDL HEADER END 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 *)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 *)gval->v); 827 suffix = (char *)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 } 1403 1404 static uint_t 1405 avg(uint_t sum, uint_t cnt) 1406 { 1407 unsigned long long s = sum * 10; 1408 1409 return ((s / cnt / 10) + (((s / cnt % 10) >= 5) ? 1 : 0)); 1410 } 1411 1412 static uint8_t 1413 percentof(uint_t part, uint_t whole) 1414 { 1415 unsigned long long p = part * 1000; 1416 1417 return ((p / whole / 10) + (((p / whole % 10) >= 5) ? 1 : 0)); 1418 } 1419 1420 static struct rsl { 1421 struct event *suspect; 1422 nvlist_t *asru; 1423 nvlist_t *fru; 1424 nvlist_t *rsrc; 1425 }; 1426 1427 /* 1428 * rslfree -- free internal members of struct rsl not expected to be 1429 * freed elsewhere. 1430 */ 1431 static void 1432 rslfree(struct rsl *freeme) 1433 { 1434 if (freeme->asru != NULL) 1435 nvlist_free(freeme->asru); 1436 if (freeme->fru != NULL) 1437 nvlist_free(freeme->fru); 1438 if (freeme->rsrc != NULL && freeme->rsrc != freeme->asru) 1439 nvlist_free(freeme->rsrc); 1440 } 1441 1442 /* 1443 * rslcmp -- compare two rsl structures. Use the following 1444 * comparisons to establish cardinality: 1445 * 1446 * 1. Name of the suspect's class. (simple strcmp) 1447 * 2. Name of the suspect's ASRU. (trickier, since nvlist) 1448 * 1449 */ 1450 static int 1451 rslcmp(const void *a, const void *b) 1452 { 1453 struct rsl *r1 = (struct rsl *)a; 1454 struct rsl *r2 = (struct rsl *)b; 1455 int rv; 1456 1457 rv = strcmp(r1->suspect->enode->u.event.ename->u.name.s, 1458 r2->suspect->enode->u.event.ename->u.name.s); 1459 if (rv != 0) 1460 return (rv); 1461 1462 if (r1->asru == NULL && r2->asru == NULL) 1463 return (0); 1464 if (r1->asru == NULL) 1465 return (-1); 1466 if (r2->asru == NULL) 1467 return (1); 1468 return (evnv_cmpnvl(r1->asru, r2->asru, 0)); 1469 } 1470 1471 /* 1472 * rsluniq -- given an array of rsl structures, seek out and "remove" 1473 * any duplicates. Dups are "remove"d by NULLing the suspect pointer 1474 * of the array element. Removal also means updating the number of 1475 * problems and the number of problems which are not faults. User 1476 * provides the first and last element pointers. 1477 */ 1478 static void 1479 rsluniq(struct rsl *first, struct rsl *last, int *nprobs, int *nnonf) 1480 { 1481 struct rsl *cr; 1482 1483 if (*nprobs == 1) 1484 return; 1485 1486 /* 1487 * At this point, we only expect duplicate defects. 1488 * Eversholt's diagnosis algorithm prevents duplicate 1489 * suspects, but we rewrite defects in the platform code after 1490 * the diagnosis is made, and that can introduce new 1491 * duplicates. 1492 */ 1493 while (first <= last) { 1494 if (first->suspect == NULL || !is_defect(first->suspect->t)) { 1495 first++; 1496 continue; 1497 } 1498 cr = first + 1; 1499 while (cr <= last) { 1500 if (is_defect(first->suspect->t)) { 1501 if (rslcmp(first, cr) == 0) { 1502 cr->suspect = NULL; 1503 rslfree(cr); 1504 (*nprobs)--; 1505 (*nnonf)--; 1506 } 1507 } 1508 /* 1509 * assume all defects are in order after our 1510 * sort and short circuit here with "else break" ? 1511 */ 1512 cr++; 1513 } 1514 first++; 1515 } 1516 } 1517 1518 /* 1519 * get_resources -- for a given suspect, determine what ASRU, FRU and 1520 * RSRC nvlists should be advertised in the final suspect list. 1521 */ 1522 void 1523 get_resources(struct event *sp, struct rsl *rsrcs, struct config *croot) 1524 { 1525 struct node *asrudef, *frudef; 1526 nvlist_t *asru, *fru; 1527 nvlist_t *rsrc = NULL; 1528 char *pathstr; 1529 1530 /* 1531 * First find any ASRU and/or FRU defined in the 1532 * initial fault tree. 1533 */ 1534 asrudef = eventprop_lookup(sp, L_ASRU); 1535 frudef = eventprop_lookup(sp, L_FRU); 1536 1537 /* 1538 * Create FMRIs based on those definitions 1539 */ 1540 asru = node2fmri(asrudef); 1541 fru = node2fmri(frudef); 1542 pathstr = ipath2str(NULL, sp->ipp); 1543 1544 /* 1545 * Allow for platform translations of the FMRIs 1546 */ 1547 platform_units_translate(is_defect(sp->t), croot, &asru, &fru, &rsrc, 1548 pathstr); 1549 1550 FREE(pathstr); 1551 rsrcs->suspect = sp; 1552 rsrcs->asru = asru; 1553 rsrcs->fru = fru; 1554 rsrcs->rsrc = rsrc; 1555 } 1556 1557 /* 1558 * trim_suspects -- prior to publishing, we may need to remove some 1559 * suspects from the list. If we're auto-closing upsets, we don't 1560 * want any of those in the published list. If the ASRUs for multiple 1561 * defects resolve to the same ASRU (driver) we only want to publish 1562 * that as a single suspect. 1563 */ 1564 static void 1565 trim_suspects(struct fme *fmep, boolean_t no_upsets, struct rsl **begin, 1566 struct rsl **end) 1567 { 1568 struct event *ep; 1569 struct rsl *rp; 1570 int rpcnt; 1571 1572 /* 1573 * First save the suspects in the psuspects, then copy back 1574 * only the ones we wish to retain. This resets nsuspects to 1575 * zero. 1576 */ 1577 rpcnt = fmep->nsuspects; 1578 save_suspects(fmep); 1579 1580 /* 1581 * allocate an array of resource pointers for the suspects. 1582 * We may end up using less than the full allocation, but this 1583 * is a very short-lived array. publish_suspects() will free 1584 * this array when it's done using it. 1585 */ 1586 rp = *begin = MALLOC(rpcnt * sizeof (struct rsl)); 1587 bzero(rp, rpcnt * sizeof (struct rsl)); 1588 1589 /* first pass, remove any unwanted upsets and populate our array */ 1590 for (ep = fmep->psuspects; ep; ep = ep->psuspects) { 1591 if (no_upsets && is_upset(ep->t)) 1592 continue; 1593 get_resources(ep, rp, fmep->cfgdata->cooked); 1594 rp++; 1595 fmep->nsuspects++; 1596 if (!is_fault(ep->t)) 1597 fmep->nonfault++; 1598 } 1599 1600 /* if all we had was unwanted upsets, we're done */ 1601 if (fmep->nsuspects == 0) 1602 return; 1603 1604 *end = rp - 1; 1605 1606 /* sort the array */ 1607 qsort(*begin, fmep->nsuspects, sizeof (struct rsl), rslcmp); 1608 rsluniq(*begin, *end, &fmep->nsuspects, &fmep->nonfault); 1609 } 1610 1611 /* 1612 * addpayloadprop -- add a payload prop to a problem 1613 */ 1614 static void 1615 addpayloadprop(const char *lhs, struct evalue *rhs, nvlist_t *fault) 1616 { 1617 ASSERT(fault != NULL); 1618 ASSERT(lhs != NULL); 1619 ASSERT(rhs != NULL); 1620 1621 if (rhs->t == UINT64) { 1622 out(O_ALTFP|O_VERB2, "addpayloadprop: %s=%llu", lhs, rhs->v); 1623 1624 if (nvlist_add_uint64(fault, lhs, rhs->v) != 0) 1625 out(O_DIE, 1626 "cannot add payloadprop \"%s\" to fault", lhs); 1627 } else { 1628 out(O_ALTFP|O_VERB2, "addpayloadprop: %s=\"%s\"", 1629 lhs, (char *)rhs->v); 1630 1631 if (nvlist_add_string(fault, lhs, (char *)rhs->v) != 0) 1632 out(O_DIE, 1633 "cannot add payloadprop \"%s\" to fault", lhs); 1634 } 1635 } 1636 1637 static char *Istatbuf; 1638 static char *Istatbufptr; 1639 static int Istatsz; 1640 1641 /* 1642 * istataddsize -- calculate size of istat and add it to Istatsz 1643 */ 1644 /*ARGSUSED2*/ 1645 static void 1646 istataddsize(const struct istat_entry *lhs, struct stats *rhs, void *arg) 1647 { 1648 int val; 1649 1650 ASSERT(lhs != NULL); 1651 ASSERT(rhs != NULL); 1652 1653 if ((val = stats_counter_value(rhs)) == 0) 1654 return; /* skip zero-valued stats */ 1655 1656 /* count up the size of the stat name */ 1657 Istatsz += ipath2strlen(lhs->ename, lhs->ipath); 1658 Istatsz++; /* for the trailing NULL byte */ 1659 1660 /* count up the size of the stat value */ 1661 Istatsz += snprintf(NULL, 0, "%d", val); 1662 Istatsz++; /* for the trailing NULL byte */ 1663 } 1664 1665 /* 1666 * istat2str -- serialize an istat, writing result to *Istatbufptr 1667 */ 1668 /*ARGSUSED2*/ 1669 static void 1670 istat2str(const struct istat_entry *lhs, struct stats *rhs, void *arg) 1671 { 1672 char *str; 1673 int len; 1674 int val; 1675 1676 ASSERT(lhs != NULL); 1677 ASSERT(rhs != NULL); 1678 1679 if ((val = stats_counter_value(rhs)) == 0) 1680 return; /* skip zero-valued stats */ 1681 1682 /* serialize the stat name */ 1683 str = ipath2str(lhs->ename, lhs->ipath); 1684 len = strlen(str); 1685 1686 ASSERT(Istatbufptr + len + 1 < &Istatbuf[Istatsz]); 1687 (void) strlcpy(Istatbufptr, str, &Istatbuf[Istatsz] - Istatbufptr); 1688 Istatbufptr += len; 1689 FREE(str); 1690 *Istatbufptr++ = '\0'; 1691 1692 /* serialize the stat value */ 1693 Istatbufptr += snprintf(Istatbufptr, &Istatbuf[Istatsz] - Istatbufptr, 1694 "%d", val); 1695 *Istatbufptr++ = '\0'; 1696 1697 ASSERT(Istatbufptr <= &Istatbuf[Istatsz]); 1698 } 1699 1700 void 1701 istat_save() 1702 { 1703 if (Istat_need_save == 0) 1704 return; 1705 1706 /* figure out how big the serialzed info is */ 1707 Istatsz = 0; 1708 lut_walk(Istats, (lut_cb)istataddsize, NULL); 1709 1710 if (Istatsz == 0) { 1711 /* no stats to save */ 1712 fmd_buf_destroy(Hdl, NULL, WOBUF_ISTATS); 1713 return; 1714 } 1715 1716 /* create the serialized buffer */ 1717 Istatbufptr = Istatbuf = MALLOC(Istatsz); 1718 lut_walk(Istats, (lut_cb)istat2str, NULL); 1719 1720 /* clear out current saved stats */ 1721 fmd_buf_destroy(Hdl, NULL, WOBUF_ISTATS); 1722 1723 /* write out the new version */ 1724 fmd_buf_write(Hdl, NULL, WOBUF_ISTATS, Istatbuf, Istatsz); 1725 FREE(Istatbuf); 1726 1727 Istat_need_save = 0; 1728 } 1729 1730 int 1731 istat_cmp(struct istat_entry *ent1, struct istat_entry *ent2) 1732 { 1733 if (ent1->ename != ent2->ename) 1734 return (ent2->ename - ent1->ename); 1735 if (ent1->ipath != ent2->ipath) 1736 return ((char *)ent2->ipath - (char *)ent1->ipath); 1737 1738 return (0); 1739 } 1740 1741 /* 1742 * istat-verify -- verify the component associated with a stat still exists 1743 * 1744 * if the component no longer exists, this routine resets the stat and 1745 * returns 0. if the component still exists, it returns 1. 1746 */ 1747 static int 1748 istat_verify(struct node *snp, struct istat_entry *entp) 1749 { 1750 struct stats *statp; 1751 nvlist_t *fmri; 1752 1753 fmri = node2fmri(snp->u.event.epname); 1754 if (platform_path_exists(fmri)) { 1755 nvlist_free(fmri); 1756 return (1); 1757 } 1758 nvlist_free(fmri); 1759 1760 /* component no longer in system. zero out the associated stats */ 1761 if ((statp = (struct stats *) 1762 lut_lookup(Istats, entp, (lut_cmp)istat_cmp)) == NULL || 1763 stats_counter_value(statp) == 0) 1764 return (0); /* stat is already reset */ 1765 1766 Istat_need_save = 1; 1767 stats_counter_reset(statp); 1768 return (0); 1769 } 1770 1771 static void 1772 istat_bump(struct node *snp, int n) 1773 { 1774 struct stats *statp; 1775 struct istat_entry ent; 1776 1777 ASSERT(snp != NULL); 1778 ASSERTinfo(snp->t == T_EVENT, ptree_nodetype2str(snp->t)); 1779 ASSERT(snp->u.event.epname != NULL); 1780 1781 /* class name should be hoisted into a single stable entry */ 1782 ASSERT(snp->u.event.ename->u.name.next == NULL); 1783 ent.ename = snp->u.event.ename->u.name.s; 1784 ent.ipath = ipath(snp->u.event.epname); 1785 1786 if (!istat_verify(snp, &ent)) { 1787 /* component no longer exists in system, nothing to do */ 1788 return; 1789 } 1790 1791 if ((statp = (struct stats *) 1792 lut_lookup(Istats, &ent, (lut_cmp)istat_cmp)) == NULL) { 1793 /* need to create the counter */ 1794 int cnt = 0; 1795 struct node *np; 1796 char *sname; 1797 char *snamep; 1798 struct istat_entry *newentp; 1799 1800 /* count up the size of the stat name */ 1801 np = snp->u.event.ename; 1802 while (np != NULL) { 1803 cnt += strlen(np->u.name.s); 1804 cnt++; /* for the '.' or '@' */ 1805 np = np->u.name.next; 1806 } 1807 np = snp->u.event.epname; 1808 while (np != NULL) { 1809 cnt += snprintf(NULL, 0, "%s%llu", 1810 np->u.name.s, np->u.name.child->u.ull); 1811 cnt++; /* for the '/' or trailing NULL byte */ 1812 np = np->u.name.next; 1813 } 1814 1815 /* build the stat name */ 1816 snamep = sname = alloca(cnt); 1817 np = snp->u.event.ename; 1818 while (np != NULL) { 1819 snamep += snprintf(snamep, &sname[cnt] - snamep, 1820 "%s", np->u.name.s); 1821 np = np->u.name.next; 1822 if (np) 1823 *snamep++ = '.'; 1824 } 1825 *snamep++ = '@'; 1826 np = snp->u.event.epname; 1827 while (np != NULL) { 1828 snamep += snprintf(snamep, &sname[cnt] - snamep, 1829 "%s%llu", np->u.name.s, np->u.name.child->u.ull); 1830 np = np->u.name.next; 1831 if (np) 1832 *snamep++ = '/'; 1833 } 1834 *snamep++ = '\0'; 1835 1836 /* create the new stat & add it to our list */ 1837 newentp = MALLOC(sizeof (*newentp)); 1838 *newentp = ent; 1839 statp = stats_new_counter(NULL, sname, 0); 1840 Istats = lut_add(Istats, (void *)newentp, (void *)statp, 1841 (lut_cmp)istat_cmp); 1842 } 1843 1844 /* if n is non-zero, set that value instead of bumping */ 1845 if (n) { 1846 stats_counter_reset(statp); 1847 stats_counter_add(statp, n); 1848 } else 1849 stats_counter_bump(statp); 1850 Istat_need_save = 1; 1851 } 1852 1853 /*ARGSUSED*/ 1854 static void 1855 istat_destructor(void *left, void *right, void *arg) 1856 { 1857 struct istat_entry *entp = (struct istat_entry *)left; 1858 struct stats *statp = (struct stats *)right; 1859 FREE(entp); 1860 stats_delete(statp); 1861 } 1862 1863 void 1864 istat_fini(void) 1865 { 1866 lut_free(Istats, istat_destructor, NULL); 1867 } 1868 1869 static void 1870 publish_suspects(struct fme *fmep) 1871 { 1872 struct event *ep; 1873 struct rsl *srl = NULL; 1874 struct rsl *erl; 1875 struct rsl *rp; 1876 nvlist_t *fault; 1877 uint8_t cert; 1878 uint_t *frs; 1879 uint_t fravg, frsum, fr; 1880 uint_t messval; 1881 struct node *snp; 1882 int frcnt, fridx; 1883 boolean_t no_upsets = B_FALSE; 1884 boolean_t allfaulty = B_TRUE; 1885 1886 stats_counter_bump(fmep->diags); 1887 1888 /* 1889 * The current fmd interfaces don't allow us to solve a case 1890 * that's already solved. If we make a new case, what of the 1891 * ereports? We don't appear to have an interface that allows 1892 * us to access the ereports attached to a case (if we wanted 1893 * to copy the original case's ereport attachments to the new 1894 * case) and it's also a bit unclear if there would be any 1895 * problems with having ereports attached to multiple cases 1896 * and/or attaching DIAGNOSED ereports to a case. For now, 1897 * we'll just output a message. 1898 */ 1899 if (fmep->posted_suspects || 1900 fmd_case_solved(fmep->hdl, fmep->fmcase)) { 1901 out(O_ALTFP|O_NONL, "Revised diagnosis for case %s: ", 1902 fmd_case_uuid(fmep->hdl, fmep->fmcase)); 1903 for (ep = fmep->suspects; ep; ep = ep->suspects) { 1904 out(O_ALTFP|O_NONL, " "); 1905 itree_pevent_brief(O_ALTFP|O_NONL, ep); 1906 } 1907 out(O_ALTFP, NULL); 1908 return; 1909 } 1910 1911 /* 1912 * If we're auto-closing upsets, we don't want to include them 1913 * in any produced suspect lists or certainty accounting. 1914 */ 1915 if (Autoclose != NULL) 1916 if (strcmp(Autoclose, "true") == 0 || 1917 strcmp(Autoclose, "all") == 0 || 1918 strcmp(Autoclose, "upsets") == 0) 1919 no_upsets = B_TRUE; 1920 1921 trim_suspects(fmep, no_upsets, &srl, &erl); 1922 1923 /* 1924 * If the resulting suspect list has no members, we're 1925 * done. Returning here will simply close the case. 1926 */ 1927 if (fmep->nsuspects == 0) { 1928 out(O_ALTFP, 1929 "[FME%d, case %s (all suspects are upsets)]", 1930 fmep->id, fmd_case_uuid(fmep->hdl, fmep->fmcase)); 1931 FREE(srl); 1932 restore_suspects(fmep); 1933 return; 1934 } 1935 1936 /* 1937 * If the suspect list is all faults, then for a given fault, 1938 * say X of N, X's certainty is computed via: 1939 * 1940 * fitrate(X) / (fitrate(1) + ... + fitrate(N)) * 100 1941 * 1942 * If none of the suspects are faults, and there are N suspects, 1943 * the certainty of a given suspect is 100/N. 1944 * 1945 * If there are are a mixture of faults and other problems in 1946 * the suspect list, we take an average of the faults' 1947 * FITrates and treat this average as the FITrate for any 1948 * non-faults. The fitrate of any given suspect is then 1949 * computed per the first formula above. 1950 */ 1951 if (fmep->nonfault == fmep->nsuspects) { 1952 /* NO faults in the suspect list */ 1953 cert = percentof(1, fmep->nsuspects); 1954 } else { 1955 /* sum the fitrates */ 1956 frs = alloca(fmep->nsuspects * sizeof (uint_t)); 1957 fridx = frcnt = frsum = 0; 1958 1959 for (rp = srl; rp <= erl; rp++) { 1960 struct node *n; 1961 1962 if (rp->suspect == NULL) 1963 continue; 1964 if (!is_fault(rp->suspect->t)) { 1965 frs[fridx++] = 0; 1966 continue; 1967 } 1968 n = eventprop_lookup(rp->suspect, L_FITrate); 1969 if (node2uint(n, &fr) != 0) { 1970 out(O_DEBUG|O_NONL, "event "); 1971 ipath_print(O_DEBUG|O_NONL, 1972 ep->enode->u.event.ename->u.name.s, 1973 ep->ipp); 1974 out(O_DEBUG, " has no FITrate (using 1)"); 1975 fr = 1; 1976 } else if (fr == 0) { 1977 out(O_DEBUG|O_NONL, "event "); 1978 ipath_print(O_DEBUG|O_NONL, 1979 ep->enode->u.event.ename->u.name.s, 1980 ep->ipp); 1981 out(O_DEBUG, " has zero FITrate (using 1)"); 1982 fr = 1; 1983 } 1984 1985 frs[fridx++] = fr; 1986 frsum += fr; 1987 frcnt++; 1988 } 1989 fravg = avg(frsum, frcnt); 1990 for (fridx = 0; fridx < fmep->nsuspects; fridx++) 1991 if (frs[fridx] == 0) { 1992 frs[fridx] = fravg; 1993 frsum += fravg; 1994 } 1995 } 1996 1997 /* Add them in reverse order of our sort, as fmd reverses order */ 1998 for (rp = erl; rp >= srl; rp--) { 1999 if (rp->suspect == NULL) 2000 continue; 2001 if (!is_fault(rp->suspect->t)) 2002 allfaulty = B_FALSE; 2003 if (fmep->nonfault != fmep->nsuspects) 2004 cert = percentof(frs[--fridx], frsum); 2005 fault = fmd_nvl_create_fault(fmep->hdl, 2006 rp->suspect->enode->u.event.ename->u.name.s, 2007 cert, 2008 rp->asru, 2009 rp->fru, 2010 rp->rsrc); 2011 if (fault == NULL) 2012 out(O_DIE, "fault creation failed"); 2013 /* if "message" property exists, add it to the fault */ 2014 if (node2uint(eventprop_lookup(rp->suspect, L_message), 2015 &messval) == 0) { 2016 2017 out(O_ALTFP, 2018 "[FME%d, %s adds message=%d to suspect list]", 2019 fmep->id, 2020 rp->suspect->enode->u.event.ename->u.name.s, 2021 messval); 2022 if (nvlist_add_boolean_value(fault, 2023 FM_SUSPECT_MESSAGE, 2024 (messval) ? B_TRUE : B_FALSE) != 0) { 2025 out(O_DIE, "cannot add no-message to fault"); 2026 } 2027 } 2028 /* add any payload properties */ 2029 lut_walk(rp->suspect->payloadprops, 2030 (lut_cb)addpayloadprop, (void *)fault); 2031 fmd_case_add_suspect(fmep->hdl, fmep->fmcase, fault); 2032 rp->suspect->fault = fault; 2033 rslfree(rp); 2034 /* if "count" property exists, increment the appropriate stat */ 2035 if ((snp = eventprop_lookup(rp->suspect, L_count)) != NULL) { 2036 out(O_ALTFP|O_NONL, 2037 "[FME%d, %s count ", fmep->id, 2038 rp->suspect->enode->u.event.ename->u.name.s); 2039 ptree_name_iter(O_ALTFP|O_NONL, snp); 2040 out(O_ALTFP, "]"); 2041 istat_bump(snp, 0); 2042 } 2043 /* if "action" property exists, evaluate it */ 2044 if ((snp = eventprop_lookup(rp->suspect, L_action)) != NULL) { 2045 struct evalue evalue; 2046 2047 out(O_ALTFP|O_NONL, 2048 "[FME%d, %s action ", fmep->id, 2049 rp->suspect->enode->u.event.ename->u.name.s); 2050 ptree_name_iter(O_ALTFP|O_NONL, snp); 2051 out(O_ALTFP, "]"); 2052 Action_nvl = fault; 2053 (void) eval_expr(snp, NULL, NULL, NULL, NULL, 2054 NULL, 0, &evalue); 2055 } 2056 /* 2057 * if "dupclose" tunable is set, check if the asru is 2058 * already marked as "faulty". 2059 */ 2060 if (Dupclose && allfaulty) { 2061 nvlist_t *asru; 2062 2063 out(O_ALTFP|O_VERB, "FMD%d dupclose check ", fmep->id); 2064 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, rp->suspect); 2065 out(O_ALTFP|O_VERB|O_NONL, " "); 2066 if (nvlist_lookup_nvlist(fault, 2067 FM_FAULT_ASRU, &asru) != 0) { 2068 out(O_ALTFP|O_VERB, "NULL asru"); 2069 allfaulty = B_FALSE; 2070 } else if (fmd_nvl_fmri_faulty(fmep->hdl, asru)) { 2071 out(O_ALTFP|O_VERB, "faulty"); 2072 } else { 2073 out(O_ALTFP|O_VERB, "not faulty"); 2074 allfaulty = B_FALSE; 2075 } 2076 } 2077 2078 } 2079 if (Dupclose && allfaulty) { 2080 out(O_ALTFP, "[dupclose FME%d, case %s]", fmep->id, 2081 fmd_case_uuid(fmep->hdl, fmep->fmcase)); 2082 fmd_case_close(fmep->hdl, fmep->fmcase); 2083 } else { 2084 out(O_ALTFP, "[solving FME%d, case %s]", fmep->id, 2085 fmd_case_uuid(fmep->hdl, fmep->fmcase)); 2086 fmd_case_solve(fmep->hdl, fmep->fmcase); 2087 } 2088 2089 istat_save(); /* write out any istat changes */ 2090 2091 /* 2092 * revert to the original suspect list 2093 */ 2094 FREE(srl); 2095 restore_suspects(fmep); 2096 } 2097 2098 static void 2099 publish_undiagnosable(fmd_hdl_t *hdl, fmd_event_t *ffep) 2100 { 2101 struct case_list *newcase; 2102 nvlist_t *defect; 2103 2104 out(O_ALTFP, 2105 "[undiagnosable ereport received, " 2106 "creating and closing a new case (%s)]", 2107 Undiag_reason ? Undiag_reason : "reason not provided"); 2108 2109 newcase = MALLOC(sizeof (struct case_list)); 2110 newcase->next = NULL; 2111 2112 newcase->fmcase = fmd_case_open(hdl, NULL); 2113 if (Undiagablecaselist != NULL) 2114 newcase->next = Undiagablecaselist; 2115 Undiagablecaselist = newcase; 2116 2117 if (ffep != NULL) 2118 fmd_case_add_ereport(hdl, newcase->fmcase, ffep); 2119 2120 defect = fmd_nvl_create_fault(hdl, UNDIAGNOSABLE_DEFECT, 100, 2121 NULL, NULL, NULL); 2122 if (Undiag_reason != NULL) 2123 (void) nvlist_add_string(defect, UNDIAG_REASON, Undiag_reason); 2124 fmd_case_add_suspect(hdl, newcase->fmcase, defect); 2125 2126 fmd_case_solve(hdl, newcase->fmcase); 2127 fmd_case_close(hdl, newcase->fmcase); 2128 } 2129 2130 static void 2131 fme_undiagnosable(struct fme *f) 2132 { 2133 nvlist_t *defect; 2134 2135 out(O_ALTFP, "[solving/closing FME%d, case %s (%s)]", 2136 f->id, fmd_case_uuid(f->hdl, f->fmcase), 2137 Undiag_reason ? Undiag_reason : "undiagnosable"); 2138 2139 defect = fmd_nvl_create_fault(f->hdl, UNDIAGNOSABLE_DEFECT, 100, 2140 NULL, NULL, NULL); 2141 if (Undiag_reason != NULL) 2142 (void) nvlist_add_string(defect, UNDIAG_REASON, Undiag_reason); 2143 fmd_case_add_suspect(f->hdl, f->fmcase, defect); 2144 fmd_case_solve(f->hdl, f->fmcase); 2145 destroy_fme_bufs(f); 2146 fmd_case_close(f->hdl, f->fmcase); 2147 } 2148 2149 /* 2150 * fme_close_case 2151 * 2152 * Find the requested case amongst our fmes and close it. Free up 2153 * the related fme. 2154 */ 2155 void 2156 fme_close_case(fmd_hdl_t *hdl, fmd_case_t *fmcase) 2157 { 2158 struct case_list *ucasep, *prevcasep = NULL; 2159 struct fme *prev = NULL; 2160 struct fme *fmep; 2161 2162 for (ucasep = Undiagablecaselist; ucasep; ucasep = ucasep->next) { 2163 if (fmcase != ucasep->fmcase) { 2164 prevcasep = ucasep; 2165 continue; 2166 } 2167 2168 if (prevcasep == NULL) 2169 Undiagablecaselist = Undiagablecaselist->next; 2170 else 2171 prevcasep->next = ucasep->next; 2172 2173 FREE(ucasep); 2174 return; 2175 } 2176 2177 for (fmep = FMElist; fmep; fmep = fmep->next) { 2178 if (fmep->hdl == hdl && fmep->fmcase == fmcase) 2179 break; 2180 prev = fmep; 2181 } 2182 2183 if (fmep == NULL) { 2184 out(O_WARN, "Eft asked to close unrecognized case [%s].", 2185 fmd_case_uuid(hdl, fmcase)); 2186 return; 2187 } 2188 2189 if (EFMElist == fmep) 2190 EFMElist = prev; 2191 2192 if (prev == NULL) 2193 FMElist = FMElist->next; 2194 else 2195 prev->next = fmep->next; 2196 2197 fmep->next = NULL; 2198 2199 /* Get rid of any timer this fme has set */ 2200 if (fmep->wull != 0) 2201 fmd_timer_remove(fmep->hdl, fmep->timer); 2202 2203 if (ClosedFMEs == NULL) { 2204 ClosedFMEs = fmep; 2205 } else { 2206 fmep->next = ClosedFMEs; 2207 ClosedFMEs = fmep; 2208 } 2209 2210 Open_fme_count--; 2211 2212 /* See if we can close the overflow FME */ 2213 if (Open_fme_count <= Max_fme) { 2214 for (fmep = FMElist; fmep; fmep = fmep->next) { 2215 if (fmep->overflow && !(fmd_case_closed(fmep->hdl, 2216 fmep->fmcase))) 2217 break; 2218 } 2219 2220 if (fmep != NULL) 2221 fmd_case_close(fmep->hdl, fmep->fmcase); 2222 } 2223 } 2224 2225 /* 2226 * fme_set_timer() 2227 * If the time we need to wait for the given FME is less than the 2228 * current timer, kick that old timer out and establish a new one. 2229 */ 2230 static int 2231 fme_set_timer(struct fme *fmep, unsigned long long wull) 2232 { 2233 out(O_ALTFP|O_VERB|O_NONL, " fme_set_timer: request to wait "); 2234 ptree_timeval(O_ALTFP|O_VERB, &wull); 2235 2236 if (wull <= fmep->pull) { 2237 out(O_ALTFP|O_VERB|O_NONL, "already have waited at least "); 2238 ptree_timeval(O_ALTFP|O_VERB, &fmep->pull); 2239 out(O_ALTFP|O_VERB, NULL); 2240 /* we've waited at least wull already, don't need timer */ 2241 return (0); 2242 } 2243 2244 out(O_ALTFP|O_VERB|O_NONL, " currently "); 2245 if (fmep->wull != 0) { 2246 out(O_ALTFP|O_VERB|O_NONL, "waiting "); 2247 ptree_timeval(O_ALTFP|O_VERB, &fmep->wull); 2248 out(O_ALTFP|O_VERB, NULL); 2249 } else { 2250 out(O_ALTFP|O_VERB|O_NONL, "not waiting"); 2251 out(O_ALTFP|O_VERB, NULL); 2252 } 2253 2254 if (fmep->wull != 0) 2255 if (wull >= fmep->wull) 2256 /* New timer would fire later than established timer */ 2257 return (0); 2258 2259 if (fmep->wull != 0) { 2260 fmd_timer_remove(fmep->hdl, fmep->timer); 2261 if (fmep->timer == fmep->htid) { 2262 out(O_ALTFP, 2263 "[stopped hesitating FME%d, case %s]", 2264 fmep->id, 2265 fmd_case_uuid(fmep->hdl, 2266 fmep->fmcase)); 2267 fmep->htid = 0; 2268 } 2269 } 2270 2271 fmep->timer = fmd_timer_install(fmep->hdl, (void *)fmep, 2272 fmep->e0r, wull); 2273 out(O_ALTFP|O_VERB, "timer set, id is %ld", fmep->timer); 2274 fmep->wull = wull; 2275 return (1); 2276 } 2277 2278 void 2279 fme_timer_fired(struct fme *fmep, id_t tid) 2280 { 2281 struct fme *ffmep = NULL; 2282 2283 for (ffmep = FMElist; ffmep; ffmep = ffmep->next) 2284 if (ffmep == fmep) 2285 break; 2286 2287 if (ffmep == NULL) { 2288 out(O_WARN, "Timer fired for an FME (%p) not in FMEs list.", 2289 (void *)fmep); 2290 return; 2291 } 2292 2293 out(O_ALTFP, "Timer fired %lx %lx", tid, fmep->htid); 2294 if (tid != fmep->htid) { 2295 /* 2296 * normal timer (not the hesitation timer) 2297 */ 2298 fmep->pull = fmep->wull; 2299 fmep->wull = 0; 2300 fmd_buf_write(fmep->hdl, fmep->fmcase, 2301 WOBUF_PULL, (void *)&fmep->pull, sizeof (fmep->pull)); 2302 /* 2303 * no point in heistating if we've already waited. 2304 */ 2305 fmep->hesitated = 1; 2306 } else { 2307 fmep->hesitated = 1; 2308 } 2309 fme_eval(fmep, NULL); 2310 } 2311 2312 /* 2313 * Preserve the fme's suspect list in its psuspects list, NULLing the 2314 * suspects list in the meantime. 2315 */ 2316 static void 2317 save_suspects(struct fme *fmep) 2318 { 2319 struct event *ep; 2320 struct event *nextep; 2321 2322 /* zero out the previous suspect list */ 2323 for (ep = fmep->psuspects; ep; ep = nextep) { 2324 nextep = ep->psuspects; 2325 ep->psuspects = NULL; 2326 } 2327 fmep->psuspects = NULL; 2328 2329 /* zero out the suspect list, copying it to previous suspect list */ 2330 fmep->psuspects = fmep->suspects; 2331 for (ep = fmep->suspects; ep; ep = nextep) { 2332 nextep = ep->suspects; 2333 ep->psuspects = ep->suspects; 2334 ep->suspects = NULL; 2335 ep->is_suspect = 0; 2336 } 2337 fmep->suspects = NULL; 2338 fmep->nsuspects = 0; 2339 fmep->nonfault = 0; 2340 } 2341 2342 /* 2343 * Retrieve the fme's suspect list from its psuspects list. 2344 */ 2345 static void 2346 restore_suspects(struct fme *fmep) 2347 { 2348 struct event *ep; 2349 struct event *nextep; 2350 2351 fmep->nsuspects = fmep->nonfault = 0; 2352 fmep->suspects = fmep->psuspects; 2353 for (ep = fmep->psuspects; ep; ep = nextep) { 2354 fmep->nsuspects++; 2355 if (!is_fault(ep->t)) 2356 fmep->nonfault++; 2357 nextep = ep->psuspects; 2358 ep->suspects = ep->psuspects; 2359 } 2360 } 2361 2362 /* 2363 * this is what we use to call the Emrys prototype code instead of main() 2364 */ 2365 static void 2366 fme_eval(struct fme *fmep, fmd_event_t *ffep) 2367 { 2368 struct event *ep; 2369 unsigned long long my_delay = TIMEVAL_EVENTUALLY; 2370 2371 save_suspects(fmep); 2372 2373 out(O_ALTFP|O_VERB, "Evaluate FME %d", fmep->id); 2374 indent_set(" "); 2375 2376 lut_walk(fmep->eventtree, (lut_cb)clear_arrows, (void *)fmep); 2377 fmep->state = hypothesise(fmep, fmep->e0, fmep->ull, &my_delay); 2378 2379 out(O_ALTFP|O_VERB|O_NONL, "FME%d state: %s, suspect list:", fmep->id, 2380 fme_state2str(fmep->state)); 2381 for (ep = fmep->suspects; ep; ep = ep->suspects) { 2382 out(O_ALTFP|O_VERB|O_NONL, " "); 2383 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 2384 } 2385 out(O_ALTFP|O_VERB, NULL); 2386 2387 if (fmep->posted_suspects) { 2388 /* 2389 * this FME has already posted a diagnosis, so see if 2390 * the event changed the diagnosis and print a warning 2391 * if it did. 2392 * 2393 */ 2394 if (suspects_changed(fmep)) { 2395 print_suspects(SLCHANGED, fmep); 2396 publish_suspects(fmep); 2397 } 2398 } else { 2399 switch (fmep->state) { 2400 case FME_CREDIBLE: 2401 /* 2402 * if the suspect list contains any upsets, we 2403 * turn off the hesitation logic (by setting 2404 * the hesitate flag which normally indicates 2405 * we've already done the hesitate logic). 2406 * this is done because hesitating with upsets 2407 * causes us to explain away additional soft errors 2408 * while the upset FME stays open. 2409 */ 2410 if (fmep->hesitated == 0) { 2411 struct event *s; 2412 2413 for (s = fmep->suspects; s; s = s->suspects) { 2414 if (s->t == N_UPSET) { 2415 fmep->hesitated = 1; 2416 break; 2417 } 2418 } 2419 } 2420 2421 if (Hesitate && 2422 fmep->suspects != NULL && 2423 fmep->suspects->suspects != NULL && 2424 fmep->hesitated == 0) { 2425 /* 2426 * about to publish multi-entry suspect list, 2427 * set the hesitation timer if not already set. 2428 */ 2429 if (fmep->htid == 0) { 2430 out(O_ALTFP|O_NONL, 2431 "[hesitate FME%d, case %s ", 2432 fmep->id, 2433 fmd_case_uuid(fmep->hdl, 2434 fmep->fmcase)); 2435 ptree_timeval(O_ALTFP|O_NONL, 2436 (unsigned long long *)&Hesitate); 2437 out(O_ALTFP, "]"); 2438 if (fme_set_timer(fmep, Hesitate)) 2439 fmep->htid = fmep->timer; 2440 } else { 2441 out(O_ALTFP, 2442 "[still hesitating FME%d, case %s]", 2443 fmep->id, 2444 fmd_case_uuid(fmep->hdl, 2445 fmep->fmcase)); 2446 } 2447 } else { 2448 print_suspects(SLNEW, fmep); 2449 (void) upsets_eval(fmep, ffep); 2450 publish_suspects(fmep); 2451 fmep->posted_suspects = 1; 2452 fmd_buf_write(fmep->hdl, fmep->fmcase, 2453 WOBUF_POSTD, 2454 (void *)&fmep->posted_suspects, 2455 sizeof (fmep->posted_suspects)); 2456 } 2457 break; 2458 2459 case FME_WAIT: 2460 /* 2461 * singleton suspect list implies 2462 * no point in waiting 2463 */ 2464 if (fmep->suspects && 2465 fmep->suspects->suspects == NULL) { 2466 print_suspects(SLNEW, fmep); 2467 (void) upsets_eval(fmep, ffep); 2468 publish_suspects(fmep); 2469 fmep->posted_suspects = 1; 2470 fmd_buf_write(fmep->hdl, fmep->fmcase, 2471 WOBUF_POSTD, 2472 (void *)&fmep->posted_suspects, 2473 sizeof (fmep->posted_suspects)); 2474 fmep->state = FME_CREDIBLE; 2475 } else { 2476 ASSERT(my_delay > fmep->ull); 2477 (void) fme_set_timer(fmep, my_delay); 2478 print_suspects(SLWAIT, fmep); 2479 } 2480 break; 2481 2482 case FME_DISPROVED: 2483 print_suspects(SLDISPROVED, fmep); 2484 Undiag_reason = UD_UNSOLVD; 2485 fme_undiagnosable(fmep); 2486 break; 2487 } 2488 } 2489 2490 if (fmep->posted_suspects == 1 && Autoclose != NULL) { 2491 int doclose = 0; 2492 2493 if (strcmp(Autoclose, "true") == 0 || 2494 strcmp(Autoclose, "all") == 0) 2495 doclose = 1; 2496 2497 if (strcmp(Autoclose, "upsets") == 0) { 2498 doclose = 1; 2499 for (ep = fmep->suspects; ep; ep = ep->suspects) { 2500 if (ep->t != N_UPSET) { 2501 doclose = 0; 2502 break; 2503 } 2504 } 2505 } 2506 2507 if (doclose) { 2508 out(O_ALTFP, "[closing FME%d, case %s (autoclose)]", 2509 fmep->id, fmd_case_uuid(fmep->hdl, fmep->fmcase)); 2510 2511 destroy_fme_bufs(fmep); 2512 fmd_case_close(fmep->hdl, fmep->fmcase); 2513 } 2514 } 2515 } 2516 2517 static void indent(void); 2518 static int triggered(struct fme *fmep, struct event *ep, int mark); 2519 static enum fme_state effects_test(struct fme *fmep, 2520 struct event *fault_event, unsigned long long at_latest_by, 2521 unsigned long long *pdelay); 2522 static enum fme_state requirements_test(struct fme *fmep, struct event *ep, 2523 unsigned long long at_latest_by, unsigned long long *pdelay); 2524 static enum fme_state causes_test(struct fme *fmep, struct event *ep, 2525 unsigned long long at_latest_by, unsigned long long *pdelay); 2526 2527 static int 2528 checkconstraints(struct fme *fmep, struct arrow *arrowp) 2529 { 2530 struct constraintlist *ctp; 2531 struct evalue value; 2532 2533 if (arrowp->forever_false) { 2534 char *sep = ""; 2535 indent(); 2536 out(O_ALTFP|O_VERB|O_NONL, " Forever false constraint: "); 2537 for (ctp = arrowp->constraints; ctp != NULL; ctp = ctp->next) { 2538 out(O_ALTFP|O_VERB|O_NONL, sep); 2539 ptree(O_ALTFP|O_VERB|O_NONL, ctp->cnode, 1, 0); 2540 sep = ", "; 2541 } 2542 out(O_ALTFP|O_VERB, NULL); 2543 return (0); 2544 } 2545 2546 for (ctp = arrowp->constraints; ctp != NULL; ctp = ctp->next) { 2547 if (eval_expr(ctp->cnode, NULL, NULL, 2548 &fmep->globals, fmep->cfgdata->cooked, 2549 arrowp, 0, &value)) { 2550 /* evaluation successful */ 2551 if (value.t == UNDEFINED || value.v == 0) { 2552 /* known false */ 2553 arrowp->forever_false = 1; 2554 indent(); 2555 out(O_ALTFP|O_VERB|O_NONL, 2556 " False constraint: "); 2557 ptree(O_ALTFP|O_VERB|O_NONL, ctp->cnode, 1, 0); 2558 out(O_ALTFP|O_VERB, NULL); 2559 return (0); 2560 } 2561 } else { 2562 /* evaluation unsuccessful -- unknown value */ 2563 indent(); 2564 out(O_ALTFP|O_VERB|O_NONL, 2565 " Deferred constraint: "); 2566 ptree(O_ALTFP|O_VERB|O_NONL, ctp->cnode, 1, 0); 2567 out(O_ALTFP|O_VERB, NULL); 2568 return (2); 2569 } 2570 } 2571 /* known true */ 2572 return (1); 2573 } 2574 2575 static int 2576 triggered(struct fme *fmep, struct event *ep, int mark) 2577 { 2578 struct bubble *bp; 2579 struct arrowlist *ap; 2580 int count = 0; 2581 2582 stats_counter_bump(fmep->Tcallcount); 2583 for (bp = itree_next_bubble(ep, NULL); bp; 2584 bp = itree_next_bubble(ep, bp)) { 2585 if (bp->t != B_TO) 2586 continue; 2587 for (ap = itree_next_arrow(bp, NULL); ap; 2588 ap = itree_next_arrow(bp, ap)) { 2589 /* check count of marks against K in the bubble */ 2590 if ((ap->arrowp->mark & mark) && 2591 ++count >= bp->nork) 2592 return (1); 2593 } 2594 } 2595 return (0); 2596 } 2597 2598 static int 2599 mark_arrows(struct fme *fmep, struct event *ep, int mark, 2600 unsigned long long at_latest_by, unsigned long long *pdelay) 2601 { 2602 struct bubble *bp; 2603 struct arrowlist *ap; 2604 unsigned long long overall_delay = TIMEVAL_EVENTUALLY; 2605 unsigned long long my_delay; 2606 enum fme_state result; 2607 int retval = 0; 2608 2609 for (bp = itree_next_bubble(ep, NULL); bp; 2610 bp = itree_next_bubble(ep, bp)) { 2611 if (bp->t != B_FROM) 2612 continue; 2613 stats_counter_bump(fmep->Marrowcount); 2614 for (ap = itree_next_arrow(bp, NULL); ap; 2615 ap = itree_next_arrow(bp, ap)) { 2616 struct event *ep2 = ap->arrowp->head->myevent; 2617 /* 2618 * if we're clearing marks, we can avoid doing 2619 * all that work evaluating constraints. 2620 */ 2621 if (mark == 0) { 2622 ap->arrowp->mark &= ~EFFECTS_COUNTER; 2623 ep2->cached_state &= 2624 ~(WAIT_EFFECT|CREDIBLE_EFFECT|PARENT_WAIT); 2625 (void) mark_arrows(fmep, ep2, mark, 0, NULL); 2626 continue; 2627 } 2628 if (ep2->cached_state & REQMNTS_DISPROVED) { 2629 indent(); 2630 out(O_ALTFP|O_VERB|O_NONL, 2631 " ALREADY DISPROVED "); 2632 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 2633 out(O_ALTFP|O_VERB, NULL); 2634 continue; 2635 } 2636 if (ep2->cached_state & WAIT_EFFECT) { 2637 indent(); 2638 out(O_ALTFP|O_VERB|O_NONL, 2639 " ALREADY EFFECTS WAIT "); 2640 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 2641 out(O_ALTFP|O_VERB, NULL); 2642 continue; 2643 } 2644 if (ep2->cached_state & CREDIBLE_EFFECT) { 2645 indent(); 2646 out(O_ALTFP|O_VERB|O_NONL, 2647 " ALREADY EFFECTS CREDIBLE "); 2648 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 2649 out(O_ALTFP|O_VERB, NULL); 2650 continue; 2651 } 2652 if ((ep2->cached_state & PARENT_WAIT) && 2653 (mark & PARENT_WAIT)) { 2654 indent(); 2655 out(O_ALTFP|O_VERB|O_NONL, 2656 " ALREADY PARENT EFFECTS WAIT "); 2657 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 2658 out(O_ALTFP|O_VERB, NULL); 2659 continue; 2660 } 2661 platform_set_payloadnvp(ep2->nvp); 2662 if (checkconstraints(fmep, ap->arrowp) != 1) { 2663 platform_set_payloadnvp(NULL); 2664 indent(); 2665 out(O_ALTFP|O_VERB|O_NONL, 2666 " CONSTRAINTS FAIL "); 2667 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 2668 out(O_ALTFP|O_VERB, NULL); 2669 continue; 2670 } 2671 platform_set_payloadnvp(NULL); 2672 ap->arrowp->mark |= EFFECTS_COUNTER; 2673 if (!triggered(fmep, ep2, EFFECTS_COUNTER)) { 2674 indent(); 2675 out(O_ALTFP|O_VERB|O_NONL, 2676 " K-COUNT NOT YET MET "); 2677 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 2678 out(O_ALTFP|O_VERB, NULL); 2679 continue; 2680 } 2681 ep2->cached_state &= ~PARENT_WAIT; 2682 result = requirements_test(fmep, ep2, at_latest_by + 2683 ap->arrowp->maxdelay, 2684 &my_delay); 2685 if (result == FME_WAIT) { 2686 retval = WAIT_EFFECT; 2687 if (overall_delay > my_delay) 2688 overall_delay = my_delay; 2689 ep2->cached_state |= WAIT_EFFECT; 2690 indent(); 2691 out(O_ALTFP|O_VERB|O_NONL, " EFFECTS WAIT "); 2692 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 2693 out(O_ALTFP|O_VERB, NULL); 2694 indent_push(" E"); 2695 if (mark_arrows(fmep, ep2, PARENT_WAIT, 2696 at_latest_by, &my_delay) == WAIT_EFFECT) { 2697 retval = WAIT_EFFECT; 2698 if (overall_delay > my_delay) 2699 overall_delay = my_delay; 2700 } 2701 indent_pop(); 2702 } else if (result == FME_DISPROVED) { 2703 indent(); 2704 out(O_ALTFP|O_VERB|O_NONL, 2705 " EFFECTS DISPROVED "); 2706 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 2707 out(O_ALTFP|O_VERB, NULL); 2708 } else { 2709 ep2->cached_state |= mark; 2710 indent(); 2711 if (mark == CREDIBLE_EFFECT) 2712 out(O_ALTFP|O_VERB|O_NONL, 2713 " EFFECTS CREDIBLE "); 2714 else 2715 out(O_ALTFP|O_VERB|O_NONL, 2716 " PARENT EFFECTS WAIT "); 2717 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep2); 2718 out(O_ALTFP|O_VERB, NULL); 2719 indent_push(" E"); 2720 if (mark_arrows(fmep, ep2, mark, at_latest_by, 2721 &my_delay) == WAIT_EFFECT) { 2722 retval = WAIT_EFFECT; 2723 if (overall_delay > my_delay) 2724 overall_delay = my_delay; 2725 } 2726 indent_pop(); 2727 } 2728 } 2729 } 2730 if (retval == WAIT_EFFECT) 2731 *pdelay = overall_delay; 2732 return (retval); 2733 } 2734 2735 static enum fme_state 2736 effects_test(struct fme *fmep, struct event *fault_event, 2737 unsigned long long at_latest_by, unsigned long long *pdelay) 2738 { 2739 struct event *error_event; 2740 enum fme_state return_value = FME_CREDIBLE; 2741 unsigned long long overall_delay = TIMEVAL_EVENTUALLY; 2742 unsigned long long my_delay; 2743 2744 stats_counter_bump(fmep->Ecallcount); 2745 indent_push(" E"); 2746 indent(); 2747 out(O_ALTFP|O_VERB|O_NONL, "->"); 2748 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, fault_event); 2749 out(O_ALTFP|O_VERB, NULL); 2750 2751 (void) mark_arrows(fmep, fault_event, CREDIBLE_EFFECT, at_latest_by, 2752 &my_delay); 2753 for (error_event = fmep->observations; 2754 error_event; error_event = error_event->observations) { 2755 indent(); 2756 out(O_ALTFP|O_VERB|O_NONL, " "); 2757 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, error_event); 2758 if (!(error_event->cached_state & CREDIBLE_EFFECT)) { 2759 if (error_event->cached_state & 2760 (PARENT_WAIT|WAIT_EFFECT)) { 2761 return_value = FME_WAIT; 2762 if (overall_delay > my_delay) 2763 overall_delay = my_delay; 2764 out(O_ALTFP|O_VERB, " NOT YET triggered"); 2765 continue; 2766 } 2767 return_value = FME_DISPROVED; 2768 out(O_ALTFP|O_VERB, " NOT triggered"); 2769 break; 2770 } else { 2771 out(O_ALTFP|O_VERB, " triggered"); 2772 } 2773 } 2774 (void) mark_arrows(fmep, fault_event, 0, 0, NULL); 2775 2776 indent(); 2777 out(O_ALTFP|O_VERB|O_NONL, "<-EFFECTS %s ", 2778 fme_state2str(return_value)); 2779 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, fault_event); 2780 out(O_ALTFP|O_VERB, NULL); 2781 indent_pop(); 2782 if (return_value == FME_WAIT) 2783 *pdelay = overall_delay; 2784 return (return_value); 2785 } 2786 2787 static enum fme_state 2788 requirements_test(struct fme *fmep, struct event *ep, 2789 unsigned long long at_latest_by, unsigned long long *pdelay) 2790 { 2791 int waiting_events; 2792 int credible_events; 2793 int deferred_events; 2794 enum fme_state return_value = FME_CREDIBLE; 2795 unsigned long long overall_delay = TIMEVAL_EVENTUALLY; 2796 unsigned long long arrow_delay; 2797 unsigned long long my_delay; 2798 struct event *ep2; 2799 struct bubble *bp; 2800 struct arrowlist *ap; 2801 2802 if (ep->cached_state & REQMNTS_CREDIBLE) { 2803 indent(); 2804 out(O_ALTFP|O_VERB|O_NONL, " REQMNTS ALREADY CREDIBLE "); 2805 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 2806 out(O_ALTFP|O_VERB, NULL); 2807 return (FME_CREDIBLE); 2808 } 2809 if (ep->cached_state & REQMNTS_DISPROVED) { 2810 indent(); 2811 out(O_ALTFP|O_VERB|O_NONL, " REQMNTS ALREADY DISPROVED "); 2812 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 2813 out(O_ALTFP|O_VERB, NULL); 2814 return (FME_DISPROVED); 2815 } 2816 if (ep->cached_state & REQMNTS_WAIT) { 2817 indent(); 2818 *pdelay = ep->cached_delay; 2819 out(O_ALTFP|O_VERB|O_NONL, " REQMNTS ALREADY WAIT "); 2820 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 2821 out(O_ALTFP|O_VERB|O_NONL, ", wait for: "); 2822 ptree_timeval(O_ALTFP|O_VERB|O_NONL, &at_latest_by); 2823 out(O_ALTFP|O_VERB, NULL); 2824 return (FME_WAIT); 2825 } 2826 stats_counter_bump(fmep->Rcallcount); 2827 indent_push(" R"); 2828 indent(); 2829 out(O_ALTFP|O_VERB|O_NONL, "->"); 2830 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 2831 out(O_ALTFP|O_VERB|O_NONL, ", at latest by: "); 2832 ptree_timeval(O_ALTFP|O_VERB|O_NONL, &at_latest_by); 2833 out(O_ALTFP|O_VERB, NULL); 2834 2835 if (ep->t == N_EREPORT) { 2836 if (ep->count == 0) { 2837 if (fmep->pull >= at_latest_by) { 2838 return_value = FME_DISPROVED; 2839 } else { 2840 ep->cached_delay = *pdelay = at_latest_by; 2841 return_value = FME_WAIT; 2842 } 2843 } 2844 2845 indent(); 2846 switch (return_value) { 2847 case FME_CREDIBLE: 2848 ep->cached_state |= REQMNTS_CREDIBLE; 2849 out(O_ALTFP|O_VERB|O_NONL, "<-REQMNTS CREDIBLE "); 2850 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 2851 break; 2852 case FME_DISPROVED: 2853 ep->cached_state |= REQMNTS_DISPROVED; 2854 out(O_ALTFP|O_VERB|O_NONL, "<-REQMNTS DISPROVED "); 2855 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 2856 break; 2857 case FME_WAIT: 2858 ep->cached_state |= REQMNTS_WAIT; 2859 out(O_ALTFP|O_VERB|O_NONL, "<-REQMNTS WAIT "); 2860 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 2861 out(O_ALTFP|O_VERB|O_NONL, " to "); 2862 ptree_timeval(O_ALTFP|O_VERB|O_NONL, &at_latest_by); 2863 break; 2864 default: 2865 out(O_DIE, "requirements_test: unexpected fme_state"); 2866 break; 2867 } 2868 out(O_ALTFP|O_VERB, NULL); 2869 indent_pop(); 2870 2871 return (return_value); 2872 } 2873 2874 /* this event is not a report, descend the tree */ 2875 for (bp = itree_next_bubble(ep, NULL); bp; 2876 bp = itree_next_bubble(ep, bp)) { 2877 int n; 2878 2879 if (bp->t != B_FROM) 2880 continue; 2881 2882 n = bp->nork; 2883 2884 credible_events = 0; 2885 waiting_events = 0; 2886 deferred_events = 0; 2887 arrow_delay = TIMEVAL_EVENTUALLY; 2888 /* 2889 * n is -1 for 'A' so adjust it. 2890 * XXX just count up the arrows for now. 2891 */ 2892 if (n < 0) { 2893 n = 0; 2894 for (ap = itree_next_arrow(bp, NULL); ap; 2895 ap = itree_next_arrow(bp, ap)) 2896 n++; 2897 indent(); 2898 out(O_ALTFP|O_VERB, " Bubble Counted N=%d", n); 2899 } else { 2900 indent(); 2901 out(O_ALTFP|O_VERB, " Bubble N=%d", n); 2902 } 2903 2904 if (n == 0) 2905 continue; 2906 if (!(bp->mark & (BUBBLE_ELIDED|BUBBLE_OK))) { 2907 for (ap = itree_next_arrow(bp, NULL); ap; 2908 ap = itree_next_arrow(bp, ap)) { 2909 ep2 = ap->arrowp->head->myevent; 2910 platform_set_payloadnvp(ep2->nvp); 2911 if (checkconstraints(fmep, ap->arrowp) == 0) { 2912 /* 2913 * if any arrow is invalidated by the 2914 * constraints, then we should elide the 2915 * whole bubble to be consistant with 2916 * the tree creation time behaviour 2917 */ 2918 bp->mark |= BUBBLE_ELIDED; 2919 platform_set_payloadnvp(NULL); 2920 break; 2921 } 2922 platform_set_payloadnvp(NULL); 2923 } 2924 } 2925 if (bp->mark & BUBBLE_ELIDED) 2926 continue; 2927 bp->mark |= BUBBLE_OK; 2928 for (ap = itree_next_arrow(bp, NULL); ap; 2929 ap = itree_next_arrow(bp, ap)) { 2930 ep2 = ap->arrowp->head->myevent; 2931 if (n <= credible_events) 2932 break; 2933 2934 ap->arrowp->mark |= REQMNTS_COUNTER; 2935 if (triggered(fmep, ep2, REQMNTS_COUNTER)) 2936 /* XXX adding max timevals! */ 2937 switch (requirements_test(fmep, ep2, 2938 at_latest_by + ap->arrowp->maxdelay, 2939 &my_delay)) { 2940 case FME_DEFERRED: 2941 deferred_events++; 2942 break; 2943 case FME_CREDIBLE: 2944 credible_events++; 2945 break; 2946 case FME_DISPROVED: 2947 break; 2948 case FME_WAIT: 2949 if (my_delay < arrow_delay) 2950 arrow_delay = my_delay; 2951 waiting_events++; 2952 break; 2953 default: 2954 out(O_DIE, 2955 "Bug in requirements_test."); 2956 } 2957 else 2958 deferred_events++; 2959 } 2960 indent(); 2961 out(O_ALTFP|O_VERB, " Credible: %d Waiting %d", 2962 credible_events + deferred_events, waiting_events); 2963 if (credible_events + deferred_events + waiting_events < n) { 2964 /* Can never meet requirements */ 2965 ep->cached_state |= REQMNTS_DISPROVED; 2966 indent(); 2967 out(O_ALTFP|O_VERB|O_NONL, "<-REQMNTS DISPROVED "); 2968 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 2969 out(O_ALTFP|O_VERB, NULL); 2970 indent_pop(); 2971 return (FME_DISPROVED); 2972 } 2973 if (credible_events + deferred_events < n) { 2974 /* will have to wait */ 2975 /* wait time is shortest known */ 2976 if (arrow_delay < overall_delay) 2977 overall_delay = arrow_delay; 2978 return_value = FME_WAIT; 2979 } else if (credible_events < n) { 2980 if (return_value != FME_WAIT) 2981 return_value = FME_DEFERRED; 2982 } 2983 } 2984 2985 /* 2986 * don't mark as FME_DEFERRED. If this event isn't reached by another 2987 * path, then this will be considered FME_CREDIBLE. But if it is 2988 * reached by a different path so the K-count is met, then might 2989 * get overridden by FME_WAIT or FME_DISPROVED. 2990 */ 2991 if (return_value == FME_WAIT) { 2992 ep->cached_state |= REQMNTS_WAIT; 2993 ep->cached_delay = *pdelay = overall_delay; 2994 } else if (return_value == FME_CREDIBLE) { 2995 ep->cached_state |= REQMNTS_CREDIBLE; 2996 } 2997 indent(); 2998 out(O_ALTFP|O_VERB|O_NONL, "<-REQMNTS %s ", 2999 fme_state2str(return_value)); 3000 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3001 out(O_ALTFP|O_VERB, NULL); 3002 indent_pop(); 3003 return (return_value); 3004 } 3005 3006 static enum fme_state 3007 causes_test(struct fme *fmep, struct event *ep, 3008 unsigned long long at_latest_by, unsigned long long *pdelay) 3009 { 3010 unsigned long long overall_delay = TIMEVAL_EVENTUALLY; 3011 unsigned long long my_delay; 3012 int credible_results = 0; 3013 int waiting_results = 0; 3014 enum fme_state fstate; 3015 struct event *tail_event; 3016 struct bubble *bp; 3017 struct arrowlist *ap; 3018 int k = 1; 3019 3020 stats_counter_bump(fmep->Ccallcount); 3021 indent_push(" C"); 3022 indent(); 3023 out(O_ALTFP|O_VERB|O_NONL, "->"); 3024 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3025 out(O_ALTFP|O_VERB, NULL); 3026 3027 for (bp = itree_next_bubble(ep, NULL); bp; 3028 bp = itree_next_bubble(ep, bp)) { 3029 if (bp->t != B_TO) 3030 continue; 3031 k = bp->nork; /* remember the K value */ 3032 for (ap = itree_next_arrow(bp, NULL); ap; 3033 ap = itree_next_arrow(bp, ap)) { 3034 int do_not_follow = 0; 3035 3036 /* 3037 * if we get to the same event multiple times 3038 * only worry about the first one. 3039 */ 3040 if (ap->arrowp->tail->myevent->cached_state & 3041 CAUSES_TESTED) { 3042 indent(); 3043 out(O_ALTFP|O_VERB|O_NONL, 3044 " causes test already run for "); 3045 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, 3046 ap->arrowp->tail->myevent); 3047 out(O_ALTFP|O_VERB, NULL); 3048 continue; 3049 } 3050 3051 /* 3052 * see if false constraint prevents us 3053 * from traversing this arrow 3054 */ 3055 platform_set_payloadnvp(ep->nvp); 3056 if (checkconstraints(fmep, ap->arrowp) != 1) 3057 do_not_follow = 1; 3058 platform_set_payloadnvp(NULL); 3059 if (do_not_follow) { 3060 indent(); 3061 out(O_ALTFP|O_VERB|O_NONL, 3062 " False arrow from "); 3063 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, 3064 ap->arrowp->tail->myevent); 3065 out(O_ALTFP|O_VERB, NULL); 3066 continue; 3067 } 3068 3069 ap->arrowp->tail->myevent->cached_state |= 3070 CAUSES_TESTED; 3071 tail_event = ap->arrowp->tail->myevent; 3072 fstate = hypothesise(fmep, tail_event, at_latest_by, 3073 &my_delay); 3074 3075 switch (fstate) { 3076 case FME_WAIT: 3077 if (my_delay < overall_delay) 3078 overall_delay = my_delay; 3079 waiting_results++; 3080 break; 3081 case FME_CREDIBLE: 3082 credible_results++; 3083 break; 3084 case FME_DISPROVED: 3085 break; 3086 default: 3087 out(O_DIE, "Bug in causes_test"); 3088 } 3089 } 3090 } 3091 /* compare against K */ 3092 if (credible_results + waiting_results < k) { 3093 indent(); 3094 out(O_ALTFP|O_VERB|O_NONL, "<-CAUSES DISPROVED "); 3095 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3096 out(O_ALTFP|O_VERB, NULL); 3097 indent_pop(); 3098 return (FME_DISPROVED); 3099 } 3100 if (waiting_results != 0) { 3101 *pdelay = overall_delay; 3102 indent(); 3103 out(O_ALTFP|O_VERB|O_NONL, "<-CAUSES WAIT "); 3104 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3105 out(O_ALTFP|O_VERB|O_NONL, " to "); 3106 ptree_timeval(O_ALTFP|O_VERB|O_NONL, &at_latest_by); 3107 out(O_ALTFP|O_VERB, NULL); 3108 indent_pop(); 3109 return (FME_WAIT); 3110 } 3111 indent(); 3112 out(O_ALTFP|O_VERB|O_NONL, "<-CAUSES CREDIBLE "); 3113 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3114 out(O_ALTFP|O_VERB, NULL); 3115 indent_pop(); 3116 return (FME_CREDIBLE); 3117 } 3118 3119 static enum fme_state 3120 hypothesise(struct fme *fmep, struct event *ep, 3121 unsigned long long at_latest_by, unsigned long long *pdelay) 3122 { 3123 enum fme_state rtr, otr; 3124 unsigned long long my_delay; 3125 unsigned long long overall_delay = TIMEVAL_EVENTUALLY; 3126 3127 stats_counter_bump(fmep->Hcallcount); 3128 indent_push(" H"); 3129 indent(); 3130 out(O_ALTFP|O_VERB|O_NONL, "->"); 3131 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3132 out(O_ALTFP|O_VERB|O_NONL, ", at latest by: "); 3133 ptree_timeval(O_ALTFP|O_VERB|O_NONL, &at_latest_by); 3134 out(O_ALTFP|O_VERB, NULL); 3135 3136 rtr = requirements_test(fmep, ep, at_latest_by, &my_delay); 3137 if ((rtr == FME_WAIT) && (my_delay < overall_delay)) 3138 overall_delay = my_delay; 3139 if (rtr != FME_DISPROVED) { 3140 if (is_problem(ep->t)) { 3141 otr = effects_test(fmep, ep, at_latest_by, &my_delay); 3142 if (otr != FME_DISPROVED) { 3143 if (fmep->peek == 0 && ep->is_suspect++ == 0) { 3144 ep->suspects = fmep->suspects; 3145 fmep->suspects = ep; 3146 fmep->nsuspects++; 3147 if (!is_fault(ep->t)) 3148 fmep->nonfault++; 3149 } 3150 } 3151 } else 3152 otr = causes_test(fmep, ep, at_latest_by, &my_delay); 3153 if ((otr == FME_WAIT) && (my_delay < overall_delay)) 3154 overall_delay = my_delay; 3155 if ((otr != FME_DISPROVED) && 3156 ((rtr == FME_WAIT) || (otr == FME_WAIT))) 3157 *pdelay = overall_delay; 3158 } 3159 if (rtr == FME_DISPROVED) { 3160 indent(); 3161 out(O_ALTFP|O_VERB|O_NONL, "<-DISPROVED "); 3162 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3163 out(O_ALTFP|O_VERB, " (doesn't meet requirements)"); 3164 indent_pop(); 3165 return (FME_DISPROVED); 3166 } 3167 if ((otr == FME_DISPROVED) && is_problem(ep->t)) { 3168 indent(); 3169 out(O_ALTFP|O_VERB|O_NONL, "<-DISPROVED "); 3170 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3171 out(O_ALTFP|O_VERB, " (doesn't explain all reports)"); 3172 indent_pop(); 3173 return (FME_DISPROVED); 3174 } 3175 if (otr == FME_DISPROVED) { 3176 indent(); 3177 out(O_ALTFP|O_VERB|O_NONL, "<-DISPROVED "); 3178 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3179 out(O_ALTFP|O_VERB, " (causes are not credible)"); 3180 indent_pop(); 3181 return (FME_DISPROVED); 3182 } 3183 if ((rtr == FME_WAIT) || (otr == FME_WAIT)) { 3184 indent(); 3185 out(O_ALTFP|O_VERB|O_NONL, "<-WAIT "); 3186 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3187 out(O_ALTFP|O_VERB|O_NONL, " to "); 3188 ptree_timeval(O_ALTFP|O_VERB|O_NONL, &overall_delay); 3189 out(O_ALTFP|O_VERB, NULL); 3190 indent_pop(); 3191 return (FME_WAIT); 3192 } 3193 indent(); 3194 out(O_ALTFP|O_VERB|O_NONL, "<-CREDIBLE "); 3195 itree_pevent_brief(O_ALTFP|O_VERB|O_NONL, ep); 3196 out(O_ALTFP|O_VERB, NULL); 3197 indent_pop(); 3198 return (FME_CREDIBLE); 3199 } 3200 3201 /* 3202 * fme_istat_load -- reconstitute any persistent istats 3203 */ 3204 void 3205 fme_istat_load(fmd_hdl_t *hdl) 3206 { 3207 int sz; 3208 char *sbuf; 3209 char *ptr; 3210 3211 if ((sz = fmd_buf_size(hdl, NULL, WOBUF_ISTATS)) == 0) { 3212 out(O_ALTFP, "fme_istat_load: No stats"); 3213 return; 3214 } 3215 3216 sbuf = alloca(sz); 3217 3218 fmd_buf_read(hdl, NULL, WOBUF_ISTATS, sbuf, sz); 3219 3220 /* 3221 * pick apart the serialized stats 3222 * 3223 * format is: 3224 * <class-name>, '@', <path>, '\0', <value>, '\0' 3225 * for example: 3226 * "stat.first@stat0/path0\02\0stat.second@stat0/path1\023\0" 3227 * 3228 * since this is parsing our own serialized data, any parsing issues 3229 * are fatal, so we check for them all with ASSERT() below. 3230 */ 3231 ptr = sbuf; 3232 while (ptr < &sbuf[sz]) { 3233 char *sepptr; 3234 struct node *np; 3235 int val; 3236 3237 sepptr = strchr(ptr, '@'); 3238 ASSERT(sepptr != NULL); 3239 *sepptr = '\0'; 3240 3241 /* construct the event */ 3242 np = newnode(T_EVENT, NULL, 0); 3243 np->u.event.ename = newnode(T_NAME, NULL, 0); 3244 np->u.event.ename->u.name.t = N_STAT; 3245 np->u.event.ename->u.name.s = stable(ptr); 3246 np->u.event.ename->u.name.it = IT_ENAME; 3247 np->u.event.ename->u.name.last = np->u.event.ename; 3248 3249 ptr = sepptr + 1; 3250 ASSERT(ptr < &sbuf[sz]); 3251 ptr += strlen(ptr); 3252 ptr++; /* move past the '\0' separating path from value */ 3253 ASSERT(ptr < &sbuf[sz]); 3254 ASSERT(isdigit(*ptr)); 3255 val = atoi(ptr); 3256 ASSERT(val > 0); 3257 ptr += strlen(ptr); 3258 ptr++; /* move past the final '\0' for this entry */ 3259 3260 np->u.event.epname = pathstring2epnamenp(sepptr + 1); 3261 ASSERT(np->u.event.epname != NULL); 3262 3263 istat_bump(np, val); 3264 tree_free(np); 3265 } 3266 3267 istat_save(); 3268 } 3269