Print this page
7711 SMF: Finish implementing support for degraded state
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/cmd/svc/startd/method.c
+++ new/usr/src/cmd/svc/startd/method.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21
22 22 /*
23 23 * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
24 24 * Copyright 2011 Joyent Inc.
25 25 */
26 26
27 27 /*
28 28 * method.c - method execution functions
29 29 *
30 30 * This file contains the routines needed to run a method: a fork(2)-exec(2)
31 31 * invocation monitored using either the contract filesystem or waitpid(2).
32 32 * (Plain fork1(2) support is provided in fork.c.)
33 33 *
34 34 * Contract Transfer
35 35 * When we restart a service, we want to transfer any contracts that the old
36 36 * service's contract inherited. This means that (a) we must not abandon the
37 37 * old contract when the service dies and (b) we must write the id of the old
38 38 * contract into the terms of the new contract. There should be limits to
39 39 * (a), though, since we don't want to keep the contract around forever. To
40 40 * this end we'll say that services in the offline state may have a contract
41 41 * to be transfered and services in the disabled or maintenance states cannot.
42 42 * This means that when a service transitions from online (or degraded) to
43 43 * offline, the contract should be preserved, and when the service transitions
44 44 * from offline to online (i.e., the start method), we'll transfer inherited
45 45 * contracts.
46 46 */
47 47
48 48 #include <sys/contract/process.h>
49 49 #include <sys/ctfs.h>
50 50 #include <sys/stat.h>
51 51 #include <sys/time.h>
52 52 #include <sys/types.h>
53 53 #include <sys/uio.h>
54 54 #include <sys/wait.h>
55 55 #include <alloca.h>
56 56 #include <assert.h>
57 57 #include <errno.h>
58 58 #include <fcntl.h>
59 59 #include <libcontract.h>
60 60 #include <libcontract_priv.h>
61 61 #include <libgen.h>
62 62 #include <librestart.h>
63 63 #include <libscf.h>
64 64 #include <limits.h>
65 65 #include <port.h>
66 66 #include <sac.h>
67 67 #include <signal.h>
68 68 #include <stdlib.h>
69 69 #include <string.h>
70 70 #include <strings.h>
71 71 #include <unistd.h>
72 72 #include <atomic.h>
73 73 #include <poll.h>
74 74 #include <libscf_priv.h>
75 75
76 76 #include "startd.h"
77 77
78 78 #define SBIN_SH "/sbin/sh"
79 79
80 80 /*
81 81 * Used to tell if contracts are in the process of being
82 82 * stored into the svc.startd internal hash table.
83 83 */
84 84 volatile uint16_t storing_contract = 0;
85 85
86 86 /*
87 87 * Mapping from restart_on method-type to contract events. Must correspond to
88 88 * enum method_restart_t.
89 89 */
90 90 static uint_t method_events[] = {
91 91 /* METHOD_RESTART_ALL */
92 92 CT_PR_EV_HWERR | CT_PR_EV_SIGNAL | CT_PR_EV_CORE | CT_PR_EV_EMPTY,
93 93 /* METHOD_RESTART_EXTERNAL_FAULT */
94 94 CT_PR_EV_HWERR | CT_PR_EV_SIGNAL,
95 95 /* METHOD_RESTART_ANY_FAULT */
96 96 CT_PR_EV_HWERR | CT_PR_EV_SIGNAL | CT_PR_EV_CORE
97 97 };
98 98
99 99 /*
100 100 * method_record_start(restarter_inst_t *)
101 101 * Record a service start for rate limiting. Place the current time
102 102 * in the circular array of instance starts.
103 103 */
104 104 static void
105 105 method_record_start(restarter_inst_t *inst)
106 106 {
107 107 int index = inst->ri_start_index++ % RINST_START_TIMES;
108 108
109 109 inst->ri_start_time[index] = gethrtime();
110 110 }
111 111
112 112 /*
113 113 * method_rate_critical(restarter_inst_t *)
114 114 * Return true if the average start interval is less than the permitted
115 115 * interval. The implicit interval defaults to RINST_FAILURE_RATE_NS and
116 116 * RINST_START_TIMES but may be overridden with the svc properties
117 117 * startd/critical_failure_count and startd/critical_failure_period
118 118 * which represent the number of failures to consider and the amount of
119 119 * time in seconds in which that number may occur, respectively. Note that
120 120 * this time is measured as of the transition to 'enabled' rather than wall
121 121 * clock time.
122 122 * Implicit success if insufficient measurements for an average exist.
123 123 */
124 124 int
125 125 method_rate_critical(restarter_inst_t *inst)
126 126 {
127 127 hrtime_t critical_failure_period;
128 128 uint_t critical_failure_count = RINST_START_TIMES;
129 129 uint_t n = inst->ri_start_index;
130 130 hrtime_t avg_ns = 0;
131 131 uint64_t scf_fr, scf_st;
132 132 scf_propvec_t *prop = NULL;
133 133 scf_propvec_t restart_critical[] = {
134 134 { "critical_failure_period", NULL, SCF_TYPE_INTEGER, NULL, 0 },
135 135 { "critical_failure_count", NULL, SCF_TYPE_INTEGER, NULL, 0 },
136 136 { NULL }
137 137 };
138 138
139 139 if (instance_is_wait_style(inst))
140 140 critical_failure_period = RINST_WT_SVC_FAILURE_RATE_NS;
141 141 else
142 142 critical_failure_period = RINST_FAILURE_RATE_NS;
143 143
144 144 restart_critical[0].pv_ptr = &scf_fr;
145 145 restart_critical[1].pv_ptr = &scf_st;
146 146
147 147 if (scf_read_propvec(inst->ri_i.i_fmri, "startd",
148 148 B_TRUE, restart_critical, &prop) != SCF_FAILED) {
149 149 /*
150 150 * critical_failure_period is expressed
151 151 * in seconds but tracked in ns
152 152 */
153 153 critical_failure_period = (hrtime_t)scf_fr * NANOSEC;
154 154 critical_failure_count = (uint_t)scf_st;
155 155 }
156 156 if (inst->ri_start_index < critical_failure_count)
157 157 return (0);
158 158
159 159 avg_ns =
160 160 (inst->ri_start_time[(n - 1) % critical_failure_count] -
161 161 inst->ri_start_time[n % critical_failure_count]) /
162 162 (critical_failure_count - 1);
163 163
164 164 return (avg_ns < critical_failure_period);
165 165 }
166 166
167 167 /*
168 168 * int method_is_transient()
169 169 * Determine if the method for the given instance is transient,
170 170 * from a contract perspective. Return 1 if it is, and 0 if it isn't.
171 171 */
172 172 static int
173 173 method_is_transient(restarter_inst_t *inst, int type)
174 174 {
175 175 if (instance_is_transient_style(inst) || type != METHOD_START)
176 176 return (1);
177 177 else
178 178 return (0);
179 179 }
180 180
181 181 /*
182 182 * void method_store_contract()
183 183 * Store the newly created contract id into local structures and
184 184 * the repository. If the repository connection is broken it is rebound.
185 185 */
186 186 static void
187 187 method_store_contract(restarter_inst_t *inst, int type, ctid_t *cid)
188 188 {
189 189 int r;
190 190 boolean_t primary;
191 191
192 192 if (errno = contract_latest(cid))
193 193 uu_die("%s: Couldn't get new contract's id", inst->ri_i.i_fmri);
194 194
195 195 primary = !method_is_transient(inst, type);
196 196
197 197 if (!primary) {
198 198 if (inst->ri_i.i_transient_ctid != 0) {
199 199 log_framework(LOG_INFO,
200 200 "%s: transient ctid expected to be 0 but "
201 201 "was set to %ld\n", inst->ri_i.i_fmri,
202 202 inst->ri_i.i_transient_ctid);
203 203 }
204 204
205 205 inst->ri_i.i_transient_ctid = *cid;
206 206 } else {
207 207 if (inst->ri_i.i_primary_ctid != 0) {
208 208 /*
209 209 * There was an old contract that we transferred.
210 210 * Remove it.
211 211 */
212 212 method_remove_contract(inst, B_TRUE, B_FALSE);
213 213 }
214 214
215 215 if (inst->ri_i.i_primary_ctid != 0) {
216 216 log_framework(LOG_INFO,
217 217 "%s: primary ctid expected to be 0 but "
218 218 "was set to %ld\n", inst->ri_i.i_fmri,
219 219 inst->ri_i.i_primary_ctid);
220 220 }
221 221
222 222 inst->ri_i.i_primary_ctid = *cid;
223 223 inst->ri_i.i_primary_ctid_stopped = 0;
224 224
225 225 log_framework(LOG_DEBUG, "Storing primary contract %ld for "
226 226 "%s.\n", *cid, inst->ri_i.i_fmri);
227 227
228 228 contract_hash_store(*cid, inst->ri_id);
229 229 }
230 230
231 231 again:
232 232 if (inst->ri_mi_deleted)
233 233 return;
234 234
235 235 r = restarter_store_contract(inst->ri_m_inst, *cid, primary ?
236 236 RESTARTER_CONTRACT_PRIMARY : RESTARTER_CONTRACT_TRANSIENT);
237 237 switch (r) {
238 238 case 0:
239 239 break;
240 240
241 241 case ECANCELED:
242 242 inst->ri_mi_deleted = B_TRUE;
243 243 break;
244 244
245 245 case ECONNABORTED:
246 246 libscf_handle_rebind(scf_instance_handle(inst->ri_m_inst));
247 247 /* FALLTHROUGH */
248 248
249 249 case EBADF:
250 250 libscf_reget_instance(inst);
251 251 goto again;
252 252
253 253 case ENOMEM:
254 254 case EPERM:
255 255 case EACCES:
256 256 case EROFS:
257 257 uu_die("%s: Couldn't store contract id %ld",
258 258 inst->ri_i.i_fmri, *cid);
259 259 /* NOTREACHED */
260 260
261 261 case EINVAL:
262 262 default:
263 263 bad_error("restarter_store_contract", r);
264 264 }
265 265 }
266 266
267 267 /*
268 268 * void method_remove_contract()
269 269 * Remove any non-permanent contracts from internal structures and
270 270 * the repository, then abandon them.
271 271 * Returns
272 272 * 0 - success
273 273 * ECANCELED - inst was deleted from the repository
274 274 *
275 275 * If the repository connection was broken, it is rebound.
276 276 */
277 277 void
278 278 method_remove_contract(restarter_inst_t *inst, boolean_t primary,
279 279 boolean_t abandon)
280 280 {
281 281 ctid_t * const ctidp = primary ? &inst->ri_i.i_primary_ctid :
282 282 &inst->ri_i.i_transient_ctid;
283 283
284 284 int r;
285 285
286 286 assert(*ctidp != 0);
287 287
288 288 log_framework(LOG_DEBUG, "Removing %s contract %lu for %s.\n",
289 289 primary ? "primary" : "transient", *ctidp, inst->ri_i.i_fmri);
290 290
291 291 if (abandon)
292 292 contract_abandon(*ctidp);
293 293
294 294 again:
295 295 if (inst->ri_mi_deleted) {
296 296 r = ECANCELED;
297 297 goto out;
298 298 }
299 299
300 300 r = restarter_remove_contract(inst->ri_m_inst, *ctidp, primary ?
301 301 RESTARTER_CONTRACT_PRIMARY : RESTARTER_CONTRACT_TRANSIENT);
302 302 switch (r) {
303 303 case 0:
304 304 break;
305 305
306 306 case ECANCELED:
307 307 inst->ri_mi_deleted = B_TRUE;
308 308 break;
309 309
310 310 case ECONNABORTED:
311 311 libscf_handle_rebind(scf_instance_handle(inst->ri_m_inst));
312 312 /* FALLTHROUGH */
313 313
314 314 case EBADF:
315 315 libscf_reget_instance(inst);
316 316 goto again;
317 317
318 318 case ENOMEM:
319 319 case EPERM:
320 320 case EACCES:
321 321 case EROFS:
322 322 log_error(LOG_INFO, "%s: Couldn't remove contract id %ld: "
323 323 "%s.\n", inst->ri_i.i_fmri, *ctidp, strerror(r));
324 324 break;
325 325
326 326 case EINVAL:
327 327 default:
328 328 bad_error("restarter_remove_contract", r);
329 329 }
330 330
331 331 out:
332 332 if (primary)
333 333 contract_hash_remove(*ctidp);
334 334
335 335 *ctidp = 0;
336 336 }
337 337
338 338 static const char *method_names[] = { "start", "stop", "refresh" };
339 339
340 340 /*
341 341 * int method_ready_contract(restarter_inst_t *, int, method_restart_t, int)
342 342 *
343 343 * Activate a contract template for the type method of inst. type,
344 344 * restart_on, and cte_mask dictate the critical events term of the contract.
345 345 * Returns
346 346 * 0 - success
347 347 * ECANCELED - inst has been deleted from the repository
348 348 */
349 349 static int
350 350 method_ready_contract(restarter_inst_t *inst, int type,
351 351 method_restart_t restart_on, uint_t cte_mask)
352 352 {
353 353 int tmpl, err, istrans, iswait, ret;
354 354 uint_t cevents, fevents;
355 355
356 356 /*
357 357 * Correctly supporting wait-style services is tricky without
358 358 * rearchitecting startd to cope with multiple event sources
359 359 * simultaneously trying to stop an instance. Until a better
360 360 * solution is implemented, we avoid this problem for
361 361 * wait-style services by making contract events fatal and
362 362 * letting the wait code alone handle stopping the service.
363 363 */
364 364 iswait = instance_is_wait_style(inst);
365 365 istrans = method_is_transient(inst, type);
366 366
367 367 tmpl = open64(CTFS_ROOT "/process/template", O_RDWR);
368 368 if (tmpl == -1)
369 369 uu_die("Could not create contract template");
370 370
371 371 /*
372 372 * We assume non-login processes are unlikely to create
373 373 * multiple process groups, and set CT_PR_PGRPONLY for all
374 374 * wait-style services' contracts.
375 375 */
376 376 err = ct_pr_tmpl_set_param(tmpl, CT_PR_INHERIT | CT_PR_REGENT |
377 377 (iswait ? CT_PR_PGRPONLY : 0));
378 378 assert(err == 0);
379 379
380 380 if (istrans) {
381 381 cevents = 0;
382 382 fevents = 0;
383 383 } else {
384 384 assert(restart_on >= 0);
385 385 assert(restart_on <= METHOD_RESTART_ANY_FAULT);
386 386 cevents = method_events[restart_on] & ~cte_mask;
387 387 fevents = iswait ?
388 388 (method_events[restart_on] & ~cte_mask & CT_PR_ALLFATAL) :
389 389 0;
390 390 }
391 391
392 392 err = ct_tmpl_set_critical(tmpl, cevents);
393 393 assert(err == 0);
394 394
395 395 err = ct_tmpl_set_informative(tmpl, 0);
396 396 assert(err == 0);
397 397 err = ct_pr_tmpl_set_fatal(tmpl, fevents);
398 398 assert(err == 0);
399 399
400 400 err = ct_tmpl_set_cookie(tmpl, istrans ? METHOD_OTHER_COOKIE :
401 401 METHOD_START_COOKIE);
402 402 assert(err == 0);
403 403
404 404 if (type == METHOD_START && inst->ri_i.i_primary_ctid != 0) {
405 405 ret = ct_pr_tmpl_set_transfer(tmpl, inst->ri_i.i_primary_ctid);
406 406 switch (ret) {
407 407 case 0:
408 408 break;
409 409
410 410 case ENOTEMPTY:
411 411 /* No contracts for you! */
412 412 method_remove_contract(inst, B_TRUE, B_TRUE);
413 413 if (inst->ri_mi_deleted) {
414 414 ret = ECANCELED;
415 415 goto out;
416 416 }
417 417 break;
418 418
419 419 case EINVAL:
420 420 case ESRCH:
421 421 case EACCES:
422 422 default:
423 423 bad_error("ct_pr_tmpl_set_transfer", ret);
424 424 }
425 425 }
426 426
427 427 err = ct_pr_tmpl_set_svc_fmri(tmpl, inst->ri_i.i_fmri);
428 428 assert(err == 0);
429 429 err = ct_pr_tmpl_set_svc_aux(tmpl, method_names[type]);
430 430 assert(err == 0);
431 431
432 432 err = ct_tmpl_activate(tmpl);
433 433 assert(err == 0);
434 434
435 435 ret = 0;
436 436
437 437 out:
438 438 err = close(tmpl);
439 439 assert(err == 0);
440 440
441 441 return (ret);
442 442 }
443 443
444 444 static void
445 445 exec_method(const restarter_inst_t *inst, int type, const char *method,
446 446 struct method_context *mcp, uint8_t need_session)
447 447 {
448 448 char *cmd;
449 449 const char *errf;
450 450 char **nenv;
451 451 int rsmc_errno = 0;
452 452
453 453 cmd = uu_msprintf("exec %s", method);
454 454
455 455 if (inst->ri_utmpx_prefix[0] != '\0' && inst->ri_utmpx_prefix != NULL)
456 456 (void) utmpx_mark_init(getpid(), inst->ri_utmpx_prefix);
457 457
458 458 setlog(inst->ri_logstem);
459 459 log_instance(inst, B_FALSE, "Executing %s method (\"%s\").",
460 460 method_names[type], method);
461 461
462 462 if (need_session)
463 463 (void) setpgrp();
464 464
465 465 /* Set credentials. */
466 466 rsmc_errno = restarter_set_method_context(mcp, &errf);
467 467 if (rsmc_errno != 0) {
468 468 log_instance(inst, B_FALSE,
469 469 "svc.startd could not set context for method: ");
470 470
471 471 if (rsmc_errno == -1) {
472 472 if (strcmp(errf, "core_set_process_path") == 0) {
473 473 log_instance(inst, B_FALSE,
474 474 "Could not set corefile path.");
475 475 } else if (strcmp(errf, "setproject") == 0) {
476 476 log_instance(inst, B_FALSE, "%s: a resource "
477 477 "control assignment failed", errf);
478 478 } else if (strcmp(errf, "pool_set_binding") == 0) {
479 479 log_instance(inst, B_FALSE, "%s: a system "
480 480 "error occurred", errf);
481 481 } else {
482 482 #ifndef NDEBUG
483 483 uu_warn("%s:%d: Bad function name \"%s\" for "
484 484 "error %d from "
485 485 "restarter_set_method_context().\n",
486 486 __FILE__, __LINE__, errf, rsmc_errno);
487 487 #endif
488 488 abort();
489 489 }
490 490
491 491 exit(1);
492 492 }
493 493
494 494 if (errf != NULL && strcmp(errf, "pool_set_binding") == 0) {
495 495 switch (rsmc_errno) {
496 496 case ENOENT:
497 497 log_instance(inst, B_FALSE, "%s: the pool "
498 498 "could not be found", errf);
499 499 break;
500 500
501 501 case EBADF:
502 502 log_instance(inst, B_FALSE, "%s: the "
503 503 "configuration is invalid", errf);
504 504 break;
505 505
506 506 case EINVAL:
507 507 log_instance(inst, B_FALSE, "%s: pool name "
508 508 "\"%s\" is invalid", errf,
509 509 mcp->resource_pool);
510 510 break;
511 511
512 512 default:
513 513 #ifndef NDEBUG
514 514 uu_warn("%s:%d: Bad error %d for function %s "
515 515 "in restarter_set_method_context().\n",
516 516 __FILE__, __LINE__, rsmc_errno, errf);
517 517 #endif
518 518 abort();
519 519 }
520 520
521 521 exit(SMF_EXIT_ERR_CONFIG);
522 522 }
523 523
524 524 if (errf != NULL && strcmp(errf, "chdir") == 0) {
525 525 switch (rsmc_errno) {
526 526 case EACCES:
527 527 case EFAULT:
528 528 case EIO:
529 529 case ELOOP:
530 530 case ENAMETOOLONG:
531 531 case ENOENT:
532 532 case ENOLINK:
533 533 case ENOTDIR:
534 534 log_instance(inst, B_FALSE, "%s: %s (\"%s\")",
535 535 errf,
536 536 strerror(rsmc_errno), mcp->working_dir);
537 537 break;
538 538
539 539 default:
540 540 #ifndef NDEBUG
541 541 uu_warn("%s:%d: Bad error %d for function %s "
542 542 "in restarter_set_method_context().\n",
543 543 __FILE__, __LINE__, rsmc_errno, errf);
544 544 #endif
545 545 abort();
546 546 }
547 547
548 548 exit(SMF_EXIT_ERR_CONFIG);
549 549 }
550 550
551 551 if (errf != NULL) {
552 552 errno = rsmc_errno;
553 553 perror(errf);
554 554
555 555 switch (rsmc_errno) {
556 556 case EINVAL:
557 557 case EPERM:
558 558 case ENOENT:
559 559 case ENAMETOOLONG:
560 560 case ERANGE:
561 561 case ESRCH:
562 562 exit(SMF_EXIT_ERR_CONFIG);
563 563 /* NOTREACHED */
564 564
565 565 default:
566 566 exit(1);
567 567 }
568 568 }
569 569
570 570 switch (rsmc_errno) {
571 571 case ENOMEM:
572 572 log_instance(inst, B_FALSE, "Out of memory.");
573 573 exit(1);
574 574 /* NOTREACHED */
575 575
576 576 case ENOENT:
577 577 log_instance(inst, B_FALSE, "Missing passwd entry for "
578 578 "user.");
579 579 exit(SMF_EXIT_ERR_CONFIG);
580 580 /* NOTREACHED */
581 581
582 582 default:
583 583 #ifndef NDEBUG
584 584 uu_warn("%s:%d: Bad miscellaneous error %d from "
585 585 "restarter_set_method_context().\n", __FILE__,
586 586 __LINE__, rsmc_errno);
587 587 #endif
588 588 abort();
589 589 }
590 590 }
591 591
592 592 nenv = set_smf_env(mcp->env, mcp->env_sz, NULL, inst,
593 593 method_names[type]);
594 594
595 595 log_preexec();
596 596
597 597 (void) execle(SBIN_SH, SBIN_SH, "-c", cmd, NULL, nenv);
598 598
599 599 exit(10);
600 600 }
601 601
602 602 static void
603 603 write_status(restarter_inst_t *inst, const char *mname, int stat)
604 604 {
605 605 int r;
606 606
607 607 again:
608 608 if (inst->ri_mi_deleted)
609 609 return;
610 610
611 611 r = libscf_write_method_status(inst->ri_m_inst, mname, stat);
612 612 switch (r) {
613 613 case 0:
614 614 break;
615 615
616 616 case ECONNABORTED:
617 617 libscf_reget_instance(inst);
618 618 goto again;
619 619
620 620 case ECANCELED:
621 621 inst->ri_mi_deleted = 1;
622 622 break;
623 623
624 624 case EPERM:
625 625 case EACCES:
626 626 case EROFS:
627 627 log_framework(LOG_INFO, "Could not write exit status "
628 628 "for %s method of %s: %s.\n", mname,
629 629 inst->ri_i.i_fmri, strerror(r));
630 630 break;
631 631
632 632 case ENAMETOOLONG:
633 633 default:
634 634 bad_error("libscf_write_method_status", r);
635 635 }
636 636 }
637 637
638 638 /*
639 639 * int method_run()
640 640 * Execute the type method of instp. If it requires a fork(), wait for it
641 641 * to return and return its exit code in *exit_code. Otherwise set
642 642 * *exit_code to 0 if the method succeeds & -1 if it fails. If the
643 643 * repository connection is broken, it is rebound, but inst may not be
644 644 * reset.
645 645 * Returns
646 646 * 0 - success
647 647 * EINVAL - A correct method or method context couldn't be retrieved.
648 648 * EIO - Contract kill failed.
649 649 * EFAULT - Method couldn't be executed successfully.
650 650 * ELOOP - Retry threshold exceeded.
651 651 * ECANCELED - inst was deleted from the repository before method was run
652 652 * ERANGE - Timeout retry threshold exceeded.
653 653 * EAGAIN - Failed due to external cause, retry.
654 654 */
655 655 int
656 656 method_run(restarter_inst_t **instp, int type, int *exit_code)
657 657 {
658 658 char *method;
659 659 int ret_status;
660 660 pid_t pid;
661 661 method_restart_t restart_on;
662 662 uint_t cte_mask;
663 663 uint8_t need_session;
664 664 scf_handle_t *h;
665 665 scf_snapshot_t *snap;
666 666 const char *mname;
667 667 mc_error_t *m_error;
668 668 struct method_context *mcp;
669 669 int result = 0, timeout_fired = 0;
670 670 int sig, r;
671 671 boolean_t transient;
672 672 uint64_t timeout;
673 673 uint8_t timeout_retry;
674 674 ctid_t ctid;
675 675 int ctfd = -1;
676 676 restarter_inst_t *inst = *instp;
677 677 int id = inst->ri_id;
678 678 int forkerr;
679 679
680 680 assert(MUTEX_HELD(&inst->ri_lock));
681 681 assert(instance_in_transition(inst));
682 682
683 683 if (inst->ri_mi_deleted)
684 684 return (ECANCELED);
685 685
686 686 *exit_code = 0;
687 687
688 688 assert(0 <= type && type <= 2);
689 689 mname = method_names[type];
690 690
691 691 if (type == METHOD_START)
692 692 inst->ri_pre_online_hook();
693 693
694 694 h = scf_instance_handle(inst->ri_m_inst);
695 695
696 696 snap = scf_snapshot_create(h);
697 697 if (snap == NULL ||
698 698 scf_instance_get_snapshot(inst->ri_m_inst, "running", snap) != 0) {
699 699 log_framework(LOG_DEBUG,
700 700 "Could not get running snapshot for %s. "
701 701 "Using editing version to run method %s.\n",
702 702 inst->ri_i.i_fmri, mname);
703 703 scf_snapshot_destroy(snap);
704 704 snap = NULL;
705 705 }
706 706
707 707 /*
708 708 * After this point, we may be logging to the instance log.
709 709 * Make sure we've noted where that log is as a property of
710 710 * the instance.
711 711 */
712 712 r = libscf_note_method_log(inst->ri_m_inst, st->st_log_prefix,
713 713 inst->ri_logstem);
714 714 if (r != 0) {
715 715 log_framework(LOG_WARNING,
716 716 "%s: couldn't note log location: %s\n",
717 717 inst->ri_i.i_fmri, strerror(r));
718 718 }
719 719
720 720 if ((method = libscf_get_method(h, type, inst, snap, &restart_on,
721 721 &cte_mask, &need_session, &timeout, &timeout_retry)) == NULL) {
722 722 if (errno == LIBSCF_PGROUP_ABSENT) {
723 723 log_framework(LOG_DEBUG,
724 724 "%s: instance has no method property group '%s'.\n",
725 725 inst->ri_i.i_fmri, mname);
726 726 if (type == METHOD_REFRESH)
727 727 log_instance(inst, B_TRUE, "No '%s' method "
728 728 "defined. Treating as :true.", mname);
729 729 else
730 730 log_instance(inst, B_TRUE, "Method property "
731 731 "group '%s' is not present.", mname);
732 732 scf_snapshot_destroy(snap);
733 733 return (0);
734 734 } else if (errno == LIBSCF_PROPERTY_ABSENT) {
735 735 log_framework(LOG_DEBUG,
736 736 "%s: instance has no '%s/exec' method property.\n",
737 737 inst->ri_i.i_fmri, mname);
738 738 log_instance(inst, B_TRUE, "Method property '%s/exec "
739 739 "is not present.", mname);
740 740 scf_snapshot_destroy(snap);
741 741 return (0);
742 742 } else {
743 743 log_error(LOG_WARNING,
744 744 "%s: instance libscf_get_method failed\n",
745 745 inst->ri_i.i_fmri);
746 746 scf_snapshot_destroy(snap);
747 747 return (EINVAL);
748 748 }
749 749 }
750 750
751 751 /* open service contract if stopping a non-transient service */
752 752 if (type == METHOD_STOP && (!instance_is_transient_style(inst))) {
753 753 if (inst->ri_i.i_primary_ctid == 0) {
754 754 /* service is not running, nothing to stop */
755 755 log_framework(LOG_DEBUG, "%s: instance has no primary "
756 756 "contract, no service to stop.\n",
757 757 inst->ri_i.i_fmri);
758 758 scf_snapshot_destroy(snap);
759 759 return (0);
760 760 }
761 761 if ((ctfd = contract_open(inst->ri_i.i_primary_ctid, "process",
762 762 "events", O_RDONLY)) < 0) {
763 763 result = EFAULT;
764 764 log_instance(inst, B_TRUE, "Could not open service "
765 765 "contract %ld. Stop method not run.",
766 766 inst->ri_i.i_primary_ctid);
767 767 goto out;
768 768 }
769 769 }
770 770
771 771 if (restarter_is_null_method(method)) {
772 772 log_framework(LOG_DEBUG, "%s: null method succeeds\n",
773 773 inst->ri_i.i_fmri);
774 774
775 775 log_instance(inst, B_TRUE, "Executing %s method (null).",
776 776 mname);
777 777
778 778 if (type == METHOD_START)
779 779 write_status(inst, mname, 0);
780 780 goto out;
781 781 }
782 782
783 783 sig = restarter_is_kill_method(method);
784 784 if (sig >= 0) {
785 785
786 786 if (inst->ri_i.i_primary_ctid == 0) {
787 787 log_error(LOG_ERR, "%s: :kill with no contract\n",
788 788 inst->ri_i.i_fmri);
789 789 log_instance(inst, B_TRUE, "Invalid use of \":kill\" "
790 790 "as stop method for transient service.");
791 791 result = EINVAL;
792 792 goto out;
793 793 }
794 794
795 795 log_framework(LOG_DEBUG,
796 796 "%s: :killing contract with signal %d\n",
797 797 inst->ri_i.i_fmri, sig);
798 798
799 799 log_instance(inst, B_TRUE, "Executing %s method (:kill).",
800 800 mname);
801 801
802 802 if (contract_kill(inst->ri_i.i_primary_ctid, sig,
803 803 inst->ri_i.i_fmri) != 0) {
804 804 result = EIO;
805 805 goto out;
806 806 } else
807 807 goto assured_kill;
808 808 }
809 809
810 810 log_framework(LOG_DEBUG, "%s: forking to run method %s\n",
811 811 inst->ri_i.i_fmri, method);
812 812
813 813 m_error = restarter_get_method_context(RESTARTER_METHOD_CONTEXT_VERSION,
814 814 inst->ri_m_inst, snap, mname, method, &mcp);
815 815
816 816 if (m_error != NULL) {
817 817 log_instance(inst, B_TRUE, "%s", m_error->msg);
818 818 restarter_mc_error_destroy(m_error);
819 819 result = EINVAL;
820 820 goto out;
821 821 }
822 822
823 823 r = method_ready_contract(inst, type, restart_on, cte_mask);
824 824 if (r != 0) {
825 825 assert(r == ECANCELED);
826 826 assert(inst->ri_mi_deleted);
827 827 restarter_free_method_context(mcp);
828 828 result = ECANCELED;
829 829 goto out;
830 830 }
831 831
832 832 /*
833 833 * Validate safety of method contexts, to save children work.
834 834 */
835 835 if (!restarter_rm_libs_loadable())
836 836 log_framework(LOG_DEBUG, "%s: method contexts limited "
837 837 "to root-accessible libraries\n", inst->ri_i.i_fmri);
838 838
839 839 /*
840 840 * For wait-style svc, sanity check that method exists to prevent an
841 841 * infinite loop.
842 842 */
843 843 if (instance_is_wait_style(inst) && type == METHOD_START) {
844 844 char *pend;
845 845 struct stat64 sbuf;
846 846
847 847 /*
848 848 * We need to handle start method strings that have arguments,
849 849 * such as '/lib/svc/method/console-login %i'.
850 850 */
851 851 if ((pend = strchr(method, ' ')) != NULL)
852 852 *pend = '\0';
853 853
854 854 if (*method == '/' && stat64(method, &sbuf) == -1 &&
855 855 errno == ENOENT) {
856 856 log_instance(inst, B_TRUE, "Missing start method (%s), "
857 857 "changing state to maintenance.", method);
858 858 restarter_free_method_context(mcp);
859 859 result = ENOENT;
860 860 goto out;
861 861 }
862 862 if (pend != NULL)
863 863 *pend = ' ';
864 864 }
865 865
866 866 /*
867 867 * If the service is restarting too quickly, send it to
868 868 * maintenance.
869 869 */
870 870 if (type == METHOD_START) {
871 871 method_record_start(inst);
872 872 if (method_rate_critical(inst) &&
873 873 !instance_is_wait_style(inst)) {
874 874 log_instance(inst, B_TRUE, "Restarting too quickly, "
875 875 "changing state to maintenance.");
876 876 result = ELOOP;
877 877 restarter_free_method_context(mcp);
878 878 goto out;
879 879 }
880 880 }
881 881
882 882 atomic_add_16(&storing_contract, 1);
883 883 pid = startd_fork1(&forkerr);
884 884 if (pid == 0)
885 885 exec_method(inst, type, method, mcp, need_session);
886 886
887 887 if (pid == -1) {
888 888 atomic_add_16(&storing_contract, -1);
889 889 if (forkerr == EAGAIN)
890 890 result = EAGAIN;
891 891 else
892 892 result = EFAULT;
893 893
894 894 log_error(LOG_WARNING,
895 895 "%s: Couldn't fork to execute method %s: %s\n",
896 896 inst->ri_i.i_fmri, method, strerror(forkerr));
897 897
898 898 restarter_free_method_context(mcp);
899 899 goto out;
900 900 }
901 901
902 902
903 903 /*
904 904 * Get the contract id, decide whether it is primary or transient, and
905 905 * stash it in inst & the repository.
906 906 */
907 907 method_store_contract(inst, type, &ctid);
908 908 atomic_add_16(&storing_contract, -1);
909 909
910 910 restarter_free_method_context(mcp);
911 911
912 912 /*
913 913 * Similarly for the start method PID.
914 914 */
915 915 if (type == METHOD_START && !inst->ri_mi_deleted)
916 916 (void) libscf_write_start_pid(inst->ri_m_inst, pid);
917 917
918 918 if (instance_is_wait_style(inst) && type == METHOD_START) {
919 919 /* Wait style instances don't get timeouts on start methods. */
920 920 if (wait_register(pid, inst->ri_i.i_fmri, 1, 0)) {
921 921 log_error(LOG_WARNING,
922 922 "%s: couldn't register %ld for wait\n",
923 923 inst->ri_i.i_fmri, pid);
924 924 result = EFAULT;
925 925 goto contract_out;
926 926 }
927 927 write_status(inst, mname, 0);
928 928
929 929 } else {
930 930 int r, err;
931 931 time_t start_time;
932 932 time_t end_time;
933 933
934 934 /*
935 935 * Because on upgrade/live-upgrade we may have no chance
936 936 * to override faulty timeout values on the way to
937 937 * manifest import, all services on the path to manifest
938 938 * import are treated the same as INFINITE timeout services.
939 939 */
940 940
941 941 start_time = time(NULL);
942 942 if (timeout != METHOD_TIMEOUT_INFINITE && !is_timeout_ovr(inst))
943 943 timeout_insert(inst, ctid, timeout);
944 944 else
945 945 timeout = METHOD_TIMEOUT_INFINITE;
946 946
947 947 /* Unlock the instance while waiting for the method. */
948 948 MUTEX_UNLOCK(&inst->ri_lock);
949 949
950 950 do {
951 951 r = waitpid(pid, &ret_status, NULL);
952 952 } while (r == -1 && errno == EINTR);
953 953 if (r == -1)
954 954 err = errno;
955 955
956 956 /* Re-grab the lock. */
957 957 inst = inst_lookup_by_id(id);
958 958
959 959 /*
960 960 * inst can't be removed, as the removal thread waits
961 961 * for completion of this one.
962 962 */
963 963 assert(inst != NULL);
964 964 *instp = inst;
965 965
966 966 if (inst->ri_timeout != NULL && inst->ri_timeout->te_fired)
967 967 timeout_fired = 1;
968 968
969 969 timeout_remove(inst, ctid);
970 970
971 971 log_framework(LOG_DEBUG,
972 972 "%s method for %s exited with status %d.\n", mname,
973 973 inst->ri_i.i_fmri, WEXITSTATUS(ret_status));
974 974
975 975 if (r == -1) {
976 976 log_error(LOG_WARNING,
977 977 "Couldn't waitpid() for %s method of %s (%s).\n",
978 978 mname, inst->ri_i.i_fmri, strerror(err));
979 979 result = EFAULT;
980 980 goto contract_out;
981 981 }
982 982
983 983 if (type == METHOD_START)
984 984 write_status(inst, mname, ret_status);
985 985
986 986 /* return ERANGE if this service doesn't retry on timeout */
987 987 if (timeout_fired == 1 && timeout_retry == 0) {
988 988 result = ERANGE;
989 989 goto contract_out;
990 990 }
991 991
992 992 if (!WIFEXITED(ret_status)) {
993 993 /*
994 994 * If method didn't exit itself (it was killed by an
995 995 * external entity, etc.), consider the entire
996 996 * method_run as failed.
997 997 */
998 998 if (WIFSIGNALED(ret_status)) {
999 999 char buf[SIG2STR_MAX];
1000 1000 (void) sig2str(WTERMSIG(ret_status), buf);
1001 1001
1002 1002 log_error(LOG_WARNING, "%s: Method \"%s\" "
1003 1003 "failed due to signal %s.\n",
1004 1004 inst->ri_i.i_fmri, method, buf);
1005 1005 log_instance(inst, B_TRUE, "Method \"%s\" "
1006 1006 "failed due to signal %s.", mname, buf);
1007 1007 } else {
1008 1008 log_error(LOG_WARNING, "%s: Method \"%s\" "
1009 1009 "failed with exit status %d.\n",
1010 1010 inst->ri_i.i_fmri, method,
1011 1011 WEXITSTATUS(ret_status));
1012 1012 log_instance(inst, B_TRUE, "Method \"%s\" "
1013 1013 "failed with exit status %d.", mname,
1014 1014 WEXITSTATUS(ret_status));
1015 1015 }
1016 1016 result = EAGAIN;
1017 1017 goto contract_out;
1018 1018 }
1019 1019
1020 1020 *exit_code = WEXITSTATUS(ret_status);
1021 1021 if (*exit_code != 0) {
1022 1022 log_error(LOG_WARNING,
1023 1023 "%s: Method \"%s\" failed with exit status %d.\n",
1024 1024 inst->ri_i.i_fmri, method, WEXITSTATUS(ret_status));
1025 1025 }
1026 1026
1027 1027 log_instance(inst, B_TRUE, "Method \"%s\" exited with status "
1028 1028 "%d.", mname, *exit_code);
1029 1029
1030 1030 if (*exit_code != 0)
1031 1031 goto contract_out;
1032 1032
1033 1033 end_time = time(NULL);
1034 1034
1035 1035 /* Give service contract remaining seconds to empty */
1036 1036 if (timeout != METHOD_TIMEOUT_INFINITE)
1037 1037 timeout -= (end_time - start_time);
1038 1038 }
1039 1039
1040 1040 assured_kill:
1041 1041 /*
1042 1042 * For stop methods, assure that the service contract has emptied
1043 1043 * before returning.
1044 1044 */
1045 1045 if (type == METHOD_STOP && (!instance_is_transient_style(inst)) &&
1046 1046 !(contract_is_empty(inst->ri_i.i_primary_ctid))) {
1047 1047 int times = 0;
1048 1048
1049 1049 if (timeout != METHOD_TIMEOUT_INFINITE)
1050 1050 timeout_insert(inst, inst->ri_i.i_primary_ctid,
1051 1051 timeout);
1052 1052
1053 1053 for (;;) {
1054 1054 /*
1055 1055 * Check frequently at first, then back off. This
1056 1056 * keeps startd from idling while shutting down.
1057 1057 */
1058 1058 if (times < 20) {
1059 1059 (void) poll(NULL, 0, 5);
1060 1060 times++;
1061 1061 } else {
1062 1062 (void) poll(NULL, 0, 100);
1063 1063 }
1064 1064 if (contract_is_empty(inst->ri_i.i_primary_ctid))
1065 1065 break;
1066 1066 }
1067 1067
1068 1068 if (timeout != METHOD_TIMEOUT_INFINITE)
1069 1069 if (inst->ri_timeout->te_fired)
1070 1070 result = EFAULT;
1071 1071
1072 1072 timeout_remove(inst, inst->ri_i.i_primary_ctid);
1073 1073 }
1074 1074
1075 1075 contract_out:
1076 1076 /* Abandon contracts for transient methods & methods that fail. */
1077 1077 transient = method_is_transient(inst, type);
1078 1078 if ((transient || *exit_code != 0 || result != 0) &&
1079 1079 (restarter_is_kill_method(method) < 0))
1080 1080 method_remove_contract(inst, !transient, B_TRUE);
1081 1081
1082 1082 out:
1083 1083 if (ctfd >= 0)
1084 1084 (void) close(ctfd);
1085 1085 scf_snapshot_destroy(snap);
1086 1086 free(method);
1087 1087 return (result);
1088 1088 }
1089 1089
1090 1090 /*
1091 1091 * The method thread executes a service method to effect a state transition.
1092 1092 * The next_state of info->sf_id should be non-_NONE on entrance, and it will
1093 1093 * be _NONE on exit (state will either be what next_state was (on success), or
1094 1094 * it will be _MAINT (on error)).
1095 1095 *
1096 1096 * There are six classes of methods to consider: start & other (stop, refresh)
1097 1097 * for each of "normal" services, wait services, and transient services. For
1098 1098 * each, the method must be fetched from the repository & executed. fork()ed
1099 1099 * methods must be waited on, except for the start method of wait services
1100 1100 * (which must be registered with the wait subsystem via wait_register()). If
1101 1101 * the method succeeded (returned 0), then for start methods its contract
1102 1102 * should be recorded as the primary contract for the service. For other
1103 1103 * methods, it should be abandoned. If the method fails, then depending on
1104 1104 * the failure, either the method should be reexecuted or the service should
1105 1105 * be put into maintenance. Either way the contract should be abandoned.
1106 1106 */
1107 1107 void *
1108 1108 method_thread(void *arg)
1109 1109 {
1110 1110 fork_info_t *info = arg;
1111 1111 restarter_inst_t *inst;
1112 1112 scf_handle_t *local_handle;
1113 1113 scf_instance_t *s_inst = NULL;
1114 1114 int r, exit_code;
1115 1115 boolean_t retryable;
1116 1116 restarter_str_t reason;
1117 1117
1118 1118 assert(0 <= info->sf_method_type && info->sf_method_type <= 2);
1119 1119
1120 1120 /* Get (and lock) the restarter_inst_t. */
1121 1121 inst = inst_lookup_by_id(info->sf_id);
1122 1122
1123 1123 assert(inst->ri_method_thread != 0);
1124 1124 assert(instance_in_transition(inst) == 1);
1125 1125
1126 1126 /*
1127 1127 * We cannot leave this function with inst in transition, because
1128 1128 * protocol.c withholds messages for inst otherwise.
1129 1129 */
1130 1130
1131 1131 log_framework(LOG_DEBUG, "method_thread() running %s method for %s.\n",
1132 1132 method_names[info->sf_method_type], inst->ri_i.i_fmri);
1133 1133
1134 1134 local_handle = libscf_handle_create_bound_loop();
1135 1135
1136 1136 rebind_retry:
1137 1137 /* get scf_instance_t */
1138 1138 switch (r = libscf_fmri_get_instance(local_handle, inst->ri_i.i_fmri,
1139 1139 &s_inst)) {
1140 1140 case 0:
1141 1141 break;
1142 1142
1143 1143 case ECONNABORTED:
1144 1144 libscf_handle_rebind(local_handle);
1145 1145 goto rebind_retry;
1146 1146
1147 1147 case ENOENT:
1148 1148 /*
1149 1149 * It's not there, but we need to call this so protocol.c
1150 1150 * doesn't think it's in transition anymore.
1151 1151 */
1152 1152 (void) restarter_instance_update_states(local_handle, inst,
1153 1153 inst->ri_i.i_state, RESTARTER_STATE_NONE, RERR_NONE,
1154 1154 restarter_str_none);
1155 1155 goto out;
1156 1156
1157 1157 case EINVAL:
1158 1158 case ENOTSUP:
1159 1159 default:
1160 1160 bad_error("libscf_fmri_get_instance", r);
1161 1161 }
|
↓ open down ↓ |
1161 lines elided |
↑ open up ↑ |
1162 1162
1163 1163 inst->ri_m_inst = s_inst;
1164 1164 inst->ri_mi_deleted = B_FALSE;
1165 1165
1166 1166 retry:
1167 1167 if (info->sf_method_type == METHOD_START)
1168 1168 log_transition(inst, START_REQUESTED);
1169 1169
1170 1170 r = method_run(&inst, info->sf_method_type, &exit_code);
1171 1171
1172 - if (r == 0 && exit_code == 0) {
1172 + if (r == 0 && (exit_code == 0 || exit_code == SMF_EXIT_MON_DEGRADE)) {
1173 1173 /* Success! */
1174 1174 assert(inst->ri_i.i_next_state != RESTARTER_STATE_NONE);
1175 1175
1176 1176 /*
1177 1177 * When a stop method succeeds, remove the primary contract of
1178 1178 * the service, unless we're going to offline, in which case
1179 1179 * retain the contract so we can transfer inherited contracts to
1180 1180 * the replacement service.
1181 1181 */
1182 1182
1183 1183 if (info->sf_method_type == METHOD_STOP &&
1184 1184 inst->ri_i.i_primary_ctid != 0) {
1185 1185 if (inst->ri_i.i_next_state == RESTARTER_STATE_OFFLINE)
1186 1186 inst->ri_i.i_primary_ctid_stopped = 1;
1187 1187 else
1188 1188 method_remove_contract(inst, B_TRUE, B_TRUE);
1189 1189 }
1190 +
1190 1191 /*
1192 + * When a start method returns with SMF_EXIT_MON_DEGRADE we
1193 + * transition the service into degraded.
1194 + */
1195 + if (info->sf_method_type == METHOD_START &&
1196 + exit_code == SMF_EXIT_MON_DEGRADE) {
1197 + inst->ri_i.i_next_state = RESTARTER_STATE_DEGRADED;
1198 + info->sf_reason = restarter_str_method_failed;
1199 + }
1200 +
1201 + /*
1191 1202 * We don't care whether the handle was rebound because this is
1192 1203 * the last thing we do with it.
1193 1204 */
1194 1205 (void) restarter_instance_update_states(local_handle, inst,
1195 1206 inst->ri_i.i_next_state, RESTARTER_STATE_NONE,
1196 1207 info->sf_event_type, info->sf_reason);
1197 1208
1198 1209 (void) update_fault_count(inst, FAULT_COUNT_RESET);
1199 1210
1200 1211 goto out;
1201 1212 }
1202 1213
1203 1214 /* Failure. Retry or go to maintenance. */
1204 1215
1205 1216 if (r != 0 && r != EAGAIN) {
1206 1217 retryable = B_FALSE;
1207 1218 } else {
1208 1219 switch (exit_code) {
1209 1220 case SMF_EXIT_ERR_CONFIG:
1210 1221 case SMF_EXIT_ERR_NOSMF:
1211 1222 case SMF_EXIT_ERR_PERM:
1212 1223 case SMF_EXIT_ERR_FATAL:
1213 1224 retryable = B_FALSE;
1214 1225 break;
1215 1226
1216 1227 default:
1217 1228 retryable = B_TRUE;
1218 1229 }
1219 1230 }
1220 1231
1221 1232 if (retryable && update_fault_count(inst, FAULT_COUNT_INCR) != 1)
1222 1233 goto retry;
1223 1234
1224 1235 /* maintenance */
1225 1236 if (r == ELOOP)
1226 1237 log_transition(inst, START_FAILED_REPEATEDLY);
1227 1238 else if (r == ERANGE)
1228 1239 log_transition(inst, START_FAILED_TIMEOUT_FATAL);
1229 1240 else if (exit_code == SMF_EXIT_ERR_CONFIG)
1230 1241 log_transition(inst, START_FAILED_CONFIGURATION);
1231 1242 else if (exit_code == SMF_EXIT_ERR_FATAL)
1232 1243 log_transition(inst, START_FAILED_FATAL);
1233 1244 else
1234 1245 log_transition(inst, START_FAILED_OTHER);
1235 1246
1236 1247 if (r == ELOOP) {
1237 1248 reason = restarter_str_restarting_too_quickly;
1238 1249 } else if (retryable) {
1239 1250 reason = restarter_str_fault_threshold_reached;
1240 1251 } else {
1241 1252 reason = restarter_str_method_failed;
1242 1253 }
1243 1254
1244 1255 (void) restarter_instance_update_states(local_handle, inst,
1245 1256 RESTARTER_STATE_MAINT, RESTARTER_STATE_NONE, RERR_FAULT,
1246 1257 reason);
1247 1258
1248 1259 if (!method_is_transient(inst, info->sf_method_type) &&
1249 1260 inst->ri_i.i_primary_ctid != 0)
1250 1261 method_remove_contract(inst, B_TRUE, B_TRUE);
1251 1262
1252 1263 out:
1253 1264 inst->ri_method_thread = 0;
1254 1265
1255 1266 /*
1256 1267 * Unlock the mutex after broadcasting to avoid a race condition
1257 1268 * with restarter_delete_inst() when the 'inst' structure is freed.
1258 1269 */
1259 1270 (void) pthread_cond_broadcast(&inst->ri_method_cv);
1260 1271 MUTEX_UNLOCK(&inst->ri_lock);
1261 1272
1262 1273 scf_instance_destroy(s_inst);
1263 1274 scf_handle_destroy(local_handle);
1264 1275 startd_free(info, sizeof (fork_info_t));
1265 1276 return (NULL);
1266 1277 }
|
↓ open down ↓ |
66 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX