xref: /titanic_50/usr/src/uts/common/contract/process.c (revision 2654012f83cec5dc15b61dfe3e4a4915f186e7a6)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/mutex.h>
29 #include <sys/debug.h>
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/kmem.h>
33 #include <sys/thread.h>
34 #include <sys/id_space.h>
35 #include <sys/avl.h>
36 #include <sys/list.h>
37 #include <sys/sysmacros.h>
38 #include <sys/proc.h>
39 #include <sys/contract.h>
40 #include <sys/contract_impl.h>
41 #include <sys/contract/process.h>
42 #include <sys/contract/process_impl.h>
43 #include <sys/cmn_err.h>
44 #include <sys/nvpair.h>
45 #include <sys/policy.h>
46 #include <sys/refstr.h>
47 #include <sys/sunddi.h>
48 
49 /*
50  * Process Contracts
51  * -----------------
52  *
53  * Generally speaking, a process contract is a contract between a
54  * process and a set of its descendent processes.  In some cases, when
55  * the child processes outlive the author of the contract, the contract
56  * may be held by (and therefore be between the child processes and) a
57  * successor process which adopts the contract after the death of the
58  * original author.
59  *
60  * The process contract adds two new concepts to the Solaris process
61  * model.  The first is that a process contract forms a rigid fault
62  * boundary around a set of processes.  Hardware, software, and even
63  * administrator errors impacting a process in a process contract
64  * generate specific events and can be requested to atomically shutdown
65  * all processes in the contract.  The second is that a process
66  * contract is a process collective whose leader is not a member of the
67  * collective.  This means that the leader can reliably react to events
68  * in the collective, and may also act upon the collective without
69  * special casing itself.
70  *
71  * A composite outcome of these two concepts is that we can now create
72  * a tree of process contracts, rooted at init(1M), which represent
73  * services and subservices that are reliably observed and can be
74  * restarted when fatal errors occur.  The service management framework
75  * (SMF) realizes this structure.
76  *
77  * For more details, see the "restart agreements" case, PSARC 2003/193.
78  *
79  * There are four sets of routines in this file: the process contract
80  * standard template operations, the process contract standard contract
81  * operations, a couple routines used only by the contract subsystem to
82  * handle process contracts' unique role as a temporary holder of
83  * abandoned contracts, and the interfaces which allow the system to
84  * create and act upon process contracts.  The first two are defined by
85  * the contracts framework and won't be discussed further.  As for the
86  * remaining two:
87  *
88  * Special framework interfaces
89  * ----------------------------
90  *
91  * contract_process_accept - determines if a process contract is a
92  *   regent, i.e. if it can inherit other contracts.
93  *
94  * contract_process_take - tells a regent process contract to inherit
95  *   an abandoned contract
96  *
97  * contract_process_adopt - tells a regent process contract that a
98  *   contract it has inherited is being adopted by a process.
99  *
100  * Process contract interfaces
101  * ---------------------------
102  *
103  * contract_process_fork - called when a process is created; adds the
104  *   new process to an existing contract or to a newly created one.
105  *
106  * contract_process_exit - called when a process exits
107  *
108  * contract_process_core - called when a process would have dumped core
109  *   (even if a core file wasn't generated)
110  *
111  * contract_process_hwerr - called when a process was killed because of
112  *   an uncorrectable hardware error
113  *
114  * contract_process_sig - called when a process was killed by a fatal
115  *   signal sent by a process in another process contract
116  *
117  */
118 
119 ct_type_t *process_type;
120 ctmpl_process_t *sys_process_tmpl;
121 refstr_t *conp_svc_aux_default;
122 
123 /*
124  * Macro predicates for determining when events should be sent and how.
125  */
126 #define	EVSENDP(ctp, flag) \
127 	((ctp->conp_contract.ct_ev_info | ctp->conp_contract.ct_ev_crit) & flag)
128 
129 #define	EVINFOP(ctp, flag) \
130 	((ctp->conp_contract.ct_ev_crit & flag) == 0)
131 
132 #define	EVFATALP(ctp, flag) \
133 	(ctp->conp_ev_fatal & flag)
134 
135 
136 /*
137  * Process contract template implementation
138  */
139 
140 /*
141  * ctmpl_process_dup
142  *
143  * The process contract template dup entry point.  Other than the
144  * to-be-subsumed contract, which must be held, this simply copies all
145  * the fields of the original.
146  */
147 static struct ct_template *
148 ctmpl_process_dup(struct ct_template *template)
149 {
150 	ctmpl_process_t *new;
151 	ctmpl_process_t *old = template->ctmpl_data;
152 
153 	new = kmem_alloc(sizeof (ctmpl_process_t), KM_SLEEP);
154 
155 	ctmpl_copy(&new->ctp_ctmpl, template);
156 	new->ctp_ctmpl.ctmpl_data = new;
157 
158 	new->ctp_subsume = old->ctp_subsume;
159 	if (new->ctp_subsume)
160 		contract_hold(new->ctp_subsume);
161 	new->ctp_params = old->ctp_params;
162 	new->ctp_ev_fatal = old->ctp_ev_fatal;
163 	new->ctp_svc_fmri = old->ctp_svc_fmri;
164 	if (new->ctp_svc_fmri != NULL) {
165 		refstr_hold(new->ctp_svc_fmri);
166 	}
167 	new->ctp_svc_aux = old->ctp_svc_aux;
168 	if (new->ctp_svc_aux != NULL) {
169 		refstr_hold(new->ctp_svc_aux);
170 	}
171 
172 	return (&new->ctp_ctmpl);
173 }
174 
175 /*
176  * ctmpl_process_free
177  *
178  * The process contract template free entry point.  Just releases a
179  * to-be-subsumed contract and frees the template.
180  */
181 static void
182 ctmpl_process_free(struct ct_template *template)
183 {
184 	ctmpl_process_t *ctp = template->ctmpl_data;
185 
186 	if (ctp->ctp_subsume)
187 		contract_rele(ctp->ctp_subsume);
188 	if (ctp->ctp_svc_fmri != NULL) {
189 		refstr_rele(ctp->ctp_svc_fmri);
190 	}
191 	if (ctp->ctp_svc_aux != NULL) {
192 		refstr_rele(ctp->ctp_svc_aux);
193 	}
194 	kmem_free(template, sizeof (ctmpl_process_t));
195 }
196 
197 /*
198  * SAFE_EV is the set of events which a non-privileged process is
199  * allowed to make critical but not fatal or if the PGRPONLY parameter
200  * is set.  EXCESS tells us if "value", a critical event set, requires
201  * additional privilege given the template "ctp".
202  */
203 #define	SAFE_EV			(CT_PR_EV_EMPTY)
204 #define	EXCESS(ctp, value)	\
205 	(((value) & ~((ctp)->ctp_ev_fatal | SAFE_EV)) || \
206 	(((value) & ~SAFE_EV) && (ctp->ctp_params & CT_PR_PGRPONLY)))
207 
208 /*
209  * ctmpl_process_set
210  *
211  * The process contract template set entry point.  None of the terms
212  * may be unconditionally set, and setting the parameters or fatal
213  * event set may result in events being implicitly removed from to the
214  * critical event set and added to the informative event set.  The
215  * (admittedly subtle) reason we implicitly change the critical event
216  * set when the parameter or fatal event set is modified but not the
217  * other way around is because a change to the critical event set only
218  * affects the contract's owner, whereas a change to the parameter set
219  * and fatal set can affect the execution of the application running in
220  * the contract (and should therefore be only made explicitly).  We
221  * allow implicit changes at all so that setting contract terms doesn't
222  * become a complex dance dependent on the template's initial state and
223  * the desired terms.
224  */
225 static int
226 ctmpl_process_set(struct ct_template *tmpl, ct_param_t *param, const cred_t *cr)
227 {
228 	ctmpl_process_t *ctp = tmpl->ctmpl_data;
229 	contract_t *ct;
230 	int error;
231 	uint64_t param_value;
232 	char *str_value;
233 
234 	if ((param->ctpm_id == CTPP_SVC_FMRI) ||
235 	    (param->ctpm_id == CTPP_CREATOR_AUX)) {
236 		str_value = (char *)param->ctpm_value;
237 		str_value[param->ctpm_size - 1] = '\0';
238 	} else {
239 		if (param->ctpm_size < sizeof (uint64_t))
240 			return (EINVAL);
241 		param_value = *(uint64_t *)param->ctpm_value;
242 		/*
243 		 * No process contract parameters are > 32 bits.
244 		 * Unless it is a string.
245 		 */
246 		if (param_value & ~UINT32_MAX)
247 			return (EINVAL);
248 	}
249 
250 	switch (param->ctpm_id) {
251 	case CTPP_SUBSUME:
252 		if (param_value != 0) {
253 			/*
254 			 * Ensure that the contract exists, that we
255 			 * hold the contract, and that the contract is
256 			 * empty.
257 			 */
258 			ct = contract_type_ptr(process_type, param_value,
259 			    curproc->p_zone->zone_uniqid);
260 			if (ct == NULL)
261 				return (ESRCH);
262 			if (ct->ct_owner != curproc) {
263 				contract_rele(ct);
264 				return (EACCES);
265 			}
266 			if (((cont_process_t *)ct->ct_data)->conp_nmembers) {
267 				contract_rele(ct);
268 				return (ENOTEMPTY);
269 			}
270 		} else {
271 			ct = NULL;
272 		}
273 		if (ctp->ctp_subsume)
274 			contract_rele(ctp->ctp_subsume);
275 		ctp->ctp_subsume = ct;
276 		break;
277 	case CTPP_PARAMS:
278 		if (param_value & ~CT_PR_ALLPARAM)
279 			return (EINVAL);
280 		ctp->ctp_params = param_value;
281 		/*
282 		 * If an unprivileged process requests that
283 		 * CT_PR_PGRPONLY be set, remove any unsafe events from
284 		 * the critical event set and add them to the
285 		 * informative event set.
286 		 */
287 		if ((ctp->ctp_params & CT_PR_PGRPONLY) &&
288 		    EXCESS(ctp, tmpl->ctmpl_ev_crit) &&
289 		    !secpolicy_contract_event_choice(cr)) {
290 			tmpl->ctmpl_ev_info |= (tmpl->ctmpl_ev_crit & ~SAFE_EV);
291 			tmpl->ctmpl_ev_crit &= SAFE_EV;
292 		}
293 
294 		break;
295 	case CTPP_SVC_FMRI:
296 		if (error = secpolicy_contract_identity(cr))
297 			return (error);
298 		if (ctp->ctp_svc_fmri != NULL)
299 			refstr_rele(ctp->ctp_svc_fmri);
300 		if (strcmp(CT_PR_SVC_DEFAULT, str_value) == 0)
301 			ctp->ctp_svc_fmri = NULL;
302 		else
303 			ctp->ctp_svc_fmri =
304 			    refstr_alloc(str_value);
305 		break;
306 	case CTPP_CREATOR_AUX:
307 		if (ctp->ctp_svc_aux != NULL)
308 			refstr_rele(ctp->ctp_svc_aux);
309 		if (param->ctpm_size == 1) /* empty string */
310 			ctp->ctp_svc_aux = NULL;
311 		else
312 			ctp->ctp_svc_aux =
313 			    refstr_alloc(str_value);
314 		break;
315 	case CTP_EV_CRITICAL:
316 		/*
317 		 * We simply don't allow adding events to the critical
318 		 * event set which aren't permitted by our policy or by
319 		 * privilege.
320 		 */
321 		if (EXCESS(ctp, param_value) &&
322 		    (error = secpolicy_contract_event(cr)) != 0)
323 			return (error);
324 		tmpl->ctmpl_ev_crit = param_value;
325 		break;
326 	case CTPP_EV_FATAL:
327 		if (param_value & ~CT_PR_ALLFATAL)
328 			return (EINVAL);
329 		ctp->ctp_ev_fatal = param_value;
330 		/*
331 		 * Check to see if an unprivileged process is
332 		 * requesting that events be removed from the fatal
333 		 * event set which are still in the critical event set.
334 		 */
335 		if (EXCESS(ctp, tmpl->ctmpl_ev_crit) &&
336 		    !secpolicy_contract_event_choice(cr)) {
337 			int allowed =
338 			    SAFE_EV | (ctp->ctp_params & CT_PR_PGRPONLY) ?
339 			    0 : ctp->ctp_ev_fatal;
340 			tmpl->ctmpl_ev_info |= (tmpl->ctmpl_ev_crit & ~allowed);
341 			tmpl->ctmpl_ev_crit &= allowed;
342 		}
343 		break;
344 	default:
345 		return (EINVAL);
346 	}
347 
348 	return (0);
349 }
350 
351 /*
352  * ctmpl_process_get
353  *
354  * The process contract template get entry point.  Simply fetches and
355  * returns the requested term.
356  */
357 static int
358 ctmpl_process_get(struct ct_template *template, ct_param_t *param)
359 {
360 	ctmpl_process_t *ctp = template->ctmpl_data;
361 	uint64_t *param_value = param->ctpm_value;
362 
363 	if (param->ctpm_id == CTPP_SUBSUME ||
364 	    param->ctpm_id == CTPP_PARAMS ||
365 	    param->ctpm_id == CTPP_EV_FATAL) {
366 		if (param->ctpm_size < sizeof (uint64_t))
367 			return (EINVAL);
368 		param->ctpm_size = sizeof (uint64_t);
369 	}
370 
371 	switch (param->ctpm_id) {
372 	case CTPP_SUBSUME:
373 		*param_value = ctp->ctp_subsume ?
374 		    ctp->ctp_subsume->ct_id : 0;
375 		break;
376 	case CTPP_PARAMS:
377 		*param_value = ctp->ctp_params;
378 		break;
379 	case CTPP_SVC_FMRI:
380 		if (ctp->ctp_svc_fmri == NULL) {
381 			param->ctpm_size =
382 			    strlcpy((char *)param->ctpm_value,
383 			    CT_PR_SVC_DEFAULT, param->ctpm_size);
384 		} else {
385 			param->ctpm_size =
386 			    strlcpy((char *)param->ctpm_value,
387 			    refstr_value(ctp->ctp_svc_fmri), param->ctpm_size);
388 		}
389 		param->ctpm_size++;
390 		break;
391 	case CTPP_CREATOR_AUX:
392 		if (ctp->ctp_svc_aux == NULL) {
393 			param->ctpm_size =
394 			    strlcpy((char *)param->ctpm_value,
395 			    refstr_value(conp_svc_aux_default),
396 			    param->ctpm_size);
397 		} else {
398 			param->ctpm_size =
399 			    strlcpy((char *)param->ctpm_value,
400 			    refstr_value(ctp->ctp_svc_aux), param->ctpm_size);
401 		}
402 		param->ctpm_size++;
403 		break;
404 	case CTPP_EV_FATAL:
405 		*param_value = ctp->ctp_ev_fatal;
406 		break;
407 	default:
408 		return (EINVAL);
409 	}
410 
411 	return (0);
412 }
413 
414 static ctmplops_t ctmpl_process_ops = {
415 	ctmpl_process_dup,		/* ctop_dup */
416 	ctmpl_process_free,		/* ctop_free */
417 	ctmpl_process_set,		/* ctop_set */
418 	ctmpl_process_get,		/* ctop_get */
419 	ctmpl_create_inval,		/* ctop_create */
420 	CT_PR_ALLEVENT
421 };
422 
423 
424 /*
425  * Process contract implementation
426  */
427 
428 /*
429  * ctmpl_process_default
430  *
431  * The process contract default template entry point.  Creates a
432  * process contract template with no parameters set, with informative
433  * core and signal events, critical empty and hwerr events, and fatal
434  * hwerr events.
435  */
436 static ct_template_t *
437 contract_process_default(void)
438 {
439 	ctmpl_process_t *new;
440 
441 	new = kmem_alloc(sizeof (ctmpl_process_t), KM_SLEEP);
442 	ctmpl_init(&new->ctp_ctmpl, &ctmpl_process_ops, process_type, new);
443 
444 	new->ctp_subsume = NULL;
445 	new->ctp_params = 0;
446 	new->ctp_ctmpl.ctmpl_ev_info = CT_PR_EV_CORE | CT_PR_EV_SIGNAL;
447 	new->ctp_ctmpl.ctmpl_ev_crit = CT_PR_EV_EMPTY | CT_PR_EV_HWERR;
448 	new->ctp_ev_fatal = CT_PR_EV_HWERR;
449 	new->ctp_svc_fmri = NULL;
450 	new->ctp_svc_aux = NULL;
451 
452 	return (&new->ctp_ctmpl);
453 }
454 
455 /*
456  * contract_process_free
457  *
458  * The process contract free entry point.
459  */
460 static void
461 contract_process_free(contract_t *ct)
462 {
463 	cont_process_t *ctp = ct->ct_data;
464 	crfree(ctp->conp_cred);
465 	list_destroy(&ctp->conp_members);
466 	list_destroy(&ctp->conp_inherited);
467 	if (ctp->conp_svc_fmri != NULL) {
468 		refstr_rele(ctp->conp_svc_fmri);
469 	}
470 	if (ctp->conp_svc_aux != NULL) {
471 		refstr_rele(ctp->conp_svc_aux);
472 	}
473 	if (ctp->conp_svc_creator != NULL) {
474 		refstr_rele(ctp->conp_svc_creator);
475 	}
476 	kmem_free(ctp, sizeof (cont_process_t));
477 }
478 
479 /*
480  * contract_process_cankill
481  *
482  * Determine if the contract author had or if the process generating
483  * the event, sp, has adequate privileges to kill process tp.
484  */
485 static int
486 contract_process_cankill(proc_t *tp, proc_t *sp, cont_process_t *ctp)
487 {
488 	int cankill;
489 
490 	mutex_enter(&tp->p_crlock);
491 	cankill = hasprocperm(tp->p_cred, ctp->conp_cred);
492 	mutex_exit(&tp->p_crlock);
493 	if (cankill || (sp && prochasprocperm(tp, sp, CRED())))
494 		return (1);
495 
496 	return (0);
497 }
498 
499 /*
500  * contract_process_kill
501  *
502  * Kills all processes in a contract, or all processes in the
503  * intersection of a contract and ex's process group (if ex is non-NULL
504  * and the contract's PGRPONLY parameter is set).  If checkpriv is
505  * true, only those processes which may be signaled by the contract
506  * author or ex are killed.
507  */
508 static void
509 contract_process_kill(contract_t *ct, proc_t *ex, int checkpriv)
510 {
511 	cont_process_t *ctp = ct->ct_data;
512 	proc_t *p;
513 	pid_t pgrp = -1;
514 
515 	ASSERT(MUTEX_HELD(&ct->ct_lock));
516 
517 	if (ex && (ctp->conp_params & CT_PR_PGRPONLY)) {
518 		pgrp = ex->p_pgrp;
519 		mutex_enter(&pidlock);
520 	}
521 
522 	for (p = list_head(&ctp->conp_members); p != NULL;
523 	    p = list_next(&ctp->conp_members, p)) {
524 		if ((p == ex) || (pgrp != -1 && p->p_pgrp != pgrp) ||
525 		    (checkpriv && !contract_process_cankill(p, ex, ctp)))
526 			continue;
527 
528 		psignal(p, SIGKILL);
529 	}
530 
531 	if (pgrp != -1)
532 		mutex_exit(&pidlock);
533 }
534 
535 
536 /*
537  * contract_process_accept
538  *
539  * Tests if the process contract is willing to act as a regent for
540  * inherited contracts.  Though brief and only called from one place,
541  * this functionality is kept here to avoid including knowledge of
542  * process contract implementation in the generic contract code.
543  */
544 int
545 contract_process_accept(contract_t *parent)
546 {
547 	cont_process_t *ctp = parent->ct_data;
548 
549 	ASSERT(parent->ct_type == process_type);
550 
551 	return (ctp->conp_params & CT_PR_REGENT);
552 }
553 
554 /*
555  * contract_process_take
556  *
557  * Executes the process contract side of inheriting a contract.
558  */
559 void
560 contract_process_take(contract_t *parent, contract_t *child)
561 {
562 	cont_process_t *ctp = parent->ct_data;
563 
564 	ASSERT(MUTEX_HELD(&parent->ct_lock));
565 	ASSERT(MUTEX_HELD(&child->ct_lock));
566 	ASSERT(parent->ct_type == process_type);
567 	ASSERT(ctp->conp_params & CT_PR_REGENT);
568 
569 	list_insert_head(&ctp->conp_inherited, child);
570 	ctp->conp_ninherited++;
571 }
572 
573 /*
574  * contract_process_adopt
575  *
576  * Executes the process contract side of adopting a contract.
577  */
578 void
579 contract_process_adopt(contract_t *ct, proc_t *p)
580 {
581 	cont_process_t *parent = p->p_ct_process;
582 
583 	ASSERT(MUTEX_HELD(&parent->conp_contract.ct_lock));
584 	ASSERT(MUTEX_HELD(&ct->ct_lock));
585 
586 	list_remove(&parent->conp_inherited, ct);
587 	parent->conp_ninherited--;
588 
589 	/*
590 	 * We drop the parent lock first because a) we are passing the
591 	 * contract reference to the child, and b) contract_adopt
592 	 * expects us to return with the contract lock held.
593 	 */
594 	mutex_exit(&parent->conp_contract.ct_lock);
595 }
596 
597 /*
598  * contract_process_abandon
599  *
600  * The process contract abandon entry point.
601  */
602 static void
603 contract_process_abandon(contract_t *ct)
604 {
605 	cont_process_t *ctp = ct->ct_data;
606 
607 	ASSERT(MUTEX_HELD(&ct->ct_lock));
608 
609 	/*
610 	 * Shall we stay or shall we go?
611 	 */
612 	if (list_head(&ctp->conp_members) == NULL) {
613 		contract_destroy(ct);
614 	} else {
615 		/*
616 		 * Strictly speaking, we actually do orphan the contract.
617 		 * Assuming our credentials allow us to kill all
618 		 * processes in the contract, this is only temporary.
619 		 */
620 		if (ctp->conp_params & CT_PR_NOORPHAN)
621 			contract_process_kill(ct, NULL, B_TRUE);
622 		contract_orphan(ct);
623 		mutex_exit(&ct->ct_lock);
624 		contract_rele(ct);
625 	}
626 }
627 
628 /*
629  * contract_process_destroy
630  *
631  * The process contract destroy entry point.
632  */
633 static void
634 contract_process_destroy(contract_t *ct)
635 {
636 	cont_process_t *ctp = ct->ct_data;
637 	contract_t *cct;
638 
639 	ASSERT(MUTEX_HELD(&ct->ct_lock));
640 
641 	/*
642 	 * contract_destroy all empty children, kill or orphan the rest
643 	 */
644 	while (cct = list_head(&ctp->conp_inherited)) {
645 		mutex_enter(&cct->ct_lock);
646 
647 		ASSERT(cct->ct_state == CTS_INHERITED);
648 
649 		list_remove(&ctp->conp_inherited, cct);
650 		ctp->conp_ninherited--;
651 		cct->ct_regent = NULL;
652 		cct->ct_type->ct_type_ops->contop_abandon(cct);
653 	}
654 }
655 
656 /*
657  * contract_process_status
658  *
659  * The process contract status entry point.
660  */
661 static void
662 contract_process_status(contract_t *ct, zone_t *zone, int detail, nvlist_t *nvl,
663     void *status, model_t model)
664 {
665 	cont_process_t *ctp = ct->ct_data;
666 	uint32_t *pids, *ctids;
667 	uint_t npids, nctids;
668 	uint_t spids, sctids;
669 	ctid_t local_svc_zone_enter;
670 
671 	if (detail == CTD_FIXED) {
672 		mutex_enter(&ct->ct_lock);
673 		contract_status_common(ct, zone, status, model);
674 		local_svc_zone_enter = ctp->conp_svc_zone_enter;
675 		mutex_exit(&ct->ct_lock);
676 	} else {
677 		contract_t *cnext;
678 		proc_t *pnext;
679 		uint_t loc;
680 
681 		ASSERT(detail == CTD_ALL);
682 		mutex_enter(&ct->ct_lock);
683 		for (;;) {
684 			spids = ctp->conp_nmembers + 5;
685 			sctids = ctp->conp_ninherited + 5;
686 			mutex_exit(&ct->ct_lock);
687 
688 			pids = kmem_alloc(spids * sizeof (uint32_t), KM_SLEEP);
689 			ctids = kmem_alloc(sctids * sizeof (uint32_t),
690 			    KM_SLEEP);
691 
692 			mutex_enter(&ct->ct_lock);
693 			npids = ctp->conp_nmembers;
694 			nctids = ctp->conp_ninherited;
695 			if (spids >= npids && sctids >= nctids)
696 				break;
697 
698 			kmem_free(pids, spids * sizeof (uint32_t));
699 			kmem_free(ctids, sctids * sizeof (uint32_t));
700 		}
701 		contract_status_common(ct, zone, status, model);
702 		for (loc = 0, cnext = list_head(&ctp->conp_inherited); cnext;
703 		    cnext = list_next(&ctp->conp_inherited, cnext))
704 			ctids[loc++] = cnext->ct_id;
705 		ASSERT(loc == nctids);
706 		for (loc = 0, pnext = list_head(&ctp->conp_members); pnext;
707 		    pnext = list_next(&ctp->conp_members, pnext))
708 			pids[loc++] = pnext->p_pid;
709 		ASSERT(loc == npids);
710 		local_svc_zone_enter = ctp->conp_svc_zone_enter;
711 		mutex_exit(&ct->ct_lock);
712 	}
713 
714 	/*
715 	 * Contract terms are static; there's no need to hold the
716 	 * contract lock while accessing them.
717 	 */
718 	VERIFY(nvlist_add_uint32(nvl, CTPS_PARAMS, ctp->conp_params) == 0);
719 	VERIFY(nvlist_add_uint32(nvl, CTPS_EV_FATAL, ctp->conp_ev_fatal) == 0);
720 	if (detail == CTD_ALL) {
721 		VERIFY(nvlist_add_uint32_array(nvl, CTPS_MEMBERS, pids,
722 		    npids) == 0);
723 		VERIFY(nvlist_add_uint32_array(nvl, CTPS_CONTRACTS, ctids,
724 		    nctids) == 0);
725 		VERIFY(nvlist_add_string(nvl, CTPS_CREATOR_AUX,
726 		    refstr_value(ctp->conp_svc_aux)) == 0);
727 		VERIFY(nvlist_add_string(nvl, CTPS_SVC_CREATOR,
728 		    refstr_value(ctp->conp_svc_creator)) == 0);
729 		kmem_free(pids, spids * sizeof (uint32_t));
730 		kmem_free(ctids, sctids * sizeof (uint32_t));
731 	}
732 
733 	/*
734 	 * if we are in a local zone and svc_fmri was inherited from
735 	 * the global zone, we provide fake svc_fmri and svc_ctid
736 	 */
737 	if (local_svc_zone_enter == 0||
738 	    zone->zone_uniqid == GLOBAL_ZONEUNIQID) {
739 		if (detail > CTD_COMMON) {
740 			VERIFY(nvlist_add_int32(nvl, CTPS_SVC_CTID,
741 			    ctp->conp_svc_ctid) == 0);
742 		}
743 		if (detail == CTD_ALL) {
744 			VERIFY(nvlist_add_string(nvl, CTPS_SVC_FMRI,
745 			    refstr_value(ctp->conp_svc_fmri)) == 0);
746 		}
747 	} else {
748 		if (detail > CTD_COMMON) {
749 			VERIFY(nvlist_add_int32(nvl, CTPS_SVC_CTID,
750 			    local_svc_zone_enter) == 0);
751 		}
752 		if (detail == CTD_ALL) {
753 			VERIFY(nvlist_add_string(nvl, CTPS_SVC_FMRI,
754 			    CT_PR_SVC_FMRI_ZONE_ENTER) == 0);
755 		}
756 	}
757 }
758 
759 /*ARGSUSED*/
760 static int
761 contract_process_newct(contract_t *ct)
762 {
763 	return (0);
764 }
765 
766 /* process contracts don't negotiate */
767 static contops_t contract_process_ops = {
768 	contract_process_free,		/* contop_free */
769 	contract_process_abandon,	/* contop_abandon */
770 	contract_process_destroy,	/* contop_destroy */
771 	contract_process_status,	/* contop_status */
772 	contract_ack_inval,		/* contop_ack */
773 	contract_ack_inval,		/* contop_nack */
774 	contract_qack_inval,		/* contop_qack */
775 	contract_process_newct		/* contop_newct */
776 };
777 
778 /*
779  * contract_process_init
780  *
781  * Initializes the process contract type.  Also creates a template for
782  * use by newproc() when it creates user processes.
783  */
784 void
785 contract_process_init(void)
786 {
787 	process_type = contract_type_init(CTT_PROCESS, "process",
788 	    &contract_process_ops, contract_process_default);
789 
790 	/*
791 	 * Create a template for use with init(1M) and other
792 	 * kernel-started processes.
793 	 */
794 	sys_process_tmpl = kmem_alloc(sizeof (ctmpl_process_t), KM_SLEEP);
795 	ctmpl_init(&sys_process_tmpl->ctp_ctmpl, &ctmpl_process_ops,
796 	    process_type, sys_process_tmpl);
797 	sys_process_tmpl->ctp_subsume = NULL;
798 	sys_process_tmpl->ctp_params = CT_PR_NOORPHAN;
799 	sys_process_tmpl->ctp_ev_fatal = CT_PR_EV_HWERR;
800 	sys_process_tmpl->ctp_svc_fmri =
801 	    refstr_alloc("svc:/system/init:default");
802 	sys_process_tmpl->ctp_svc_aux = refstr_alloc("");
803 	conp_svc_aux_default = sys_process_tmpl->ctp_svc_aux;
804 	refstr_hold(conp_svc_aux_default);
805 }
806 
807 /*
808  * contract_process_create
809  *
810  * create a process contract given template "tmpl" and parent process
811  * "parent".  May fail and return NULL if project.max-contracts would
812  * have been exceeded.
813  */
814 static cont_process_t *
815 contract_process_create(ctmpl_process_t *tmpl, proc_t *parent, int canfail)
816 {
817 	cont_process_t *ctp;
818 
819 	ASSERT(tmpl != NULL);
820 
821 	(void) contract_type_pbundle(process_type, parent);
822 
823 	ctp = kmem_zalloc(sizeof (cont_process_t), KM_SLEEP);
824 
825 	list_create(&ctp->conp_members, sizeof (proc_t),
826 	    offsetof(proc_t, p_ct_member));
827 	list_create(&ctp->conp_inherited, sizeof (contract_t),
828 	    offsetof(contract_t, ct_ctlist));
829 	mutex_enter(&tmpl->ctp_ctmpl.ctmpl_lock);
830 	ctp->conp_params = tmpl->ctp_params;
831 	ctp->conp_ev_fatal = tmpl->ctp_ev_fatal;
832 	crhold(ctp->conp_cred = CRED());
833 
834 	if (contract_ctor(&ctp->conp_contract, process_type, &tmpl->ctp_ctmpl,
835 	    ctp, (ctp->conp_params & CT_PR_INHERIT) ? CTF_INHERIT : 0,
836 	    parent, canfail)) {
837 		mutex_exit(&tmpl->ctp_ctmpl.ctmpl_lock);
838 		contract_process_free(&ctp->conp_contract);
839 		return (NULL);
840 	}
841 
842 	/*
843 	 * inherit svc_fmri if not defined by consumer. In this case, inherit
844 	 * also svc_ctid to keep track of the contract id where
845 	 * svc_fmri was set
846 	 */
847 	if (tmpl->ctp_svc_fmri == NULL) {
848 		ctp->conp_svc_fmri = parent->p_ct_process->conp_svc_fmri;
849 		ctp->conp_svc_ctid = parent->p_ct_process->conp_svc_ctid;
850 		ctp->conp_svc_zone_enter =
851 		    parent->p_ct_process->conp_svc_zone_enter;
852 	} else {
853 		ctp->conp_svc_fmri = tmpl->ctp_svc_fmri;
854 		ctp->conp_svc_ctid = ctp->conp_contract.ct_id;
855 		/* make svc_zone_enter flag false when svc_fmri is set */
856 		ctp->conp_svc_zone_enter = 0;
857 	}
858 	refstr_hold(ctp->conp_svc_fmri);
859 	/* set svc_aux to default value if not defined in template */
860 	if (tmpl->ctp_svc_aux == NULL) {
861 		ctp->conp_svc_aux = conp_svc_aux_default;
862 	} else {
863 		ctp->conp_svc_aux = tmpl->ctp_svc_aux;
864 	}
865 	refstr_hold(ctp->conp_svc_aux);
866 	/*
867 	 * set svc_creator to execname
868 	 * We special case pid0 because when newproc() creates
869 	 * the init process, the p_user.u_comm field of sched's proc_t
870 	 * has not been populated yet.
871 	 */
872 	if (parent->p_pidp == &pid0) /* if the kernel is the creator */
873 		ctp->conp_svc_creator = refstr_alloc("sched");
874 	else
875 		ctp->conp_svc_creator = refstr_alloc(parent->p_user.u_comm);
876 
877 	/*
878 	 * Transfer subcontracts only after new contract is visible.
879 	 * Also, only transfer contracts if the parent matches -- we
880 	 * don't want to create a cycle in the tree of contracts.
881 	 */
882 	if (tmpl->ctp_subsume && tmpl->ctp_subsume->ct_owner == parent) {
883 		cont_process_t *sct = tmpl->ctp_subsume->ct_data;
884 		contract_t *ct;
885 
886 		mutex_enter(&tmpl->ctp_subsume->ct_lock);
887 		mutex_enter(&ctp->conp_contract.ct_lock);
888 		while (ct = list_head(&sct->conp_inherited)) {
889 			mutex_enter(&ct->ct_lock);
890 			list_remove(&sct->conp_inherited, ct);
891 			list_insert_tail(&ctp->conp_inherited, ct);
892 			ct->ct_regent = &ctp->conp_contract;
893 			mutex_exit(&ct->ct_lock);
894 		}
895 		ctp->conp_ninherited += sct->conp_ninherited;
896 		sct->conp_ninherited = 0;
897 		mutex_exit(&ctp->conp_contract.ct_lock);
898 		mutex_exit(&tmpl->ctp_subsume->ct_lock);
899 
900 		/*
901 		 * Automatically abandon the contract.
902 		 */
903 		(void) contract_abandon(tmpl->ctp_subsume, parent, 1);
904 	}
905 
906 	mutex_exit(&tmpl->ctp_ctmpl.ctmpl_lock);
907 
908 	return (ctp);
909 }
910 
911 /*
912  * contract_process_exit
913  *
914  * Called on process exit.  Removes process p from process contract
915  * ctp.  Generates an exit event, if requested.  Generates an empty
916  * event, if p is the last member of the the process contract and empty
917  * events were requested.
918  */
919 void
920 contract_process_exit(cont_process_t *ctp, proc_t *p, int exitstatus)
921 {
922 	contract_t *ct = &ctp->conp_contract;
923 	ct_kevent_t *event;
924 	int empty;
925 
926 	/*
927 	 * Remove self from process contract.
928 	 */
929 	mutex_enter(&ct->ct_lock);
930 	list_remove(&ctp->conp_members, p);
931 	ctp->conp_nmembers--;
932 	mutex_enter(&p->p_lock);	/* in case /proc is watching */
933 	p->p_ct_process = NULL;
934 	mutex_exit(&p->p_lock);
935 
936 	/*
937 	 * We check for emptiness before dropping the contract lock to
938 	 * send the exit event, otherwise we could end up with two
939 	 * empty events.
940 	 */
941 	empty = (list_head(&ctp->conp_members) == NULL);
942 	if (EVSENDP(ctp, CT_PR_EV_EXIT)) {
943 		nvlist_t *nvl;
944 
945 		mutex_exit(&ct->ct_lock);
946 		VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
947 		VERIFY(nvlist_add_uint32(nvl, CTPE_PID, p->p_pid) == 0);
948 		VERIFY(nvlist_add_int32(nvl, CTPE_EXITSTATUS, exitstatus) == 0);
949 
950 		event = kmem_zalloc(sizeof (ct_kevent_t), KM_SLEEP);
951 		event->cte_flags = EVINFOP(ctp, CT_PR_EV_EXIT) ? CTE_INFO : 0;
952 		event->cte_type = CT_PR_EV_EXIT;
953 		(void) cte_publish_all(ct, event, nvl, NULL);
954 		mutex_enter(&ct->ct_lock);
955 	}
956 	if (empty) {
957 		/*
958 		 * Send EMPTY message.
959 		 */
960 		if (EVSENDP(ctp, CT_PR_EV_EMPTY)) {
961 			nvlist_t *nvl;
962 
963 			mutex_exit(&ct->ct_lock);
964 			VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
965 			    KM_SLEEP) == 0);
966 			VERIFY(nvlist_add_uint32(nvl, CTPE_PID, p->p_pid) == 0);
967 
968 			event = kmem_zalloc(sizeof (ct_kevent_t), KM_SLEEP);
969 			event->cte_flags = EVINFOP(ctp, CT_PR_EV_EMPTY) ?
970 			    CTE_INFO : 0;
971 			event->cte_type = CT_PR_EV_EMPTY;
972 			(void) cte_publish_all(ct, event, nvl, NULL);
973 			mutex_enter(&ct->ct_lock);
974 		}
975 
976 		/*
977 		 * The last one to leave an orphaned contract turns out
978 		 * the lights.
979 		 */
980 		if (ct->ct_state == CTS_ORPHAN) {
981 			contract_destroy(ct);
982 			return;
983 		}
984 	}
985 	mutex_exit(&ct->ct_lock);
986 	contract_rele(ct);
987 }
988 
989 /*
990  * contract_process_fork
991  *
992  * Called on process fork.  If the current lwp has a active process
993  * contract template, we attempt to create a new process contract.
994  * Failure to create a process contract when required is a failure in
995  * fork so, in such an event, we return NULL.
996  *
997  * Assuming we succeeded or skipped the previous step, we add the child
998  * process to the new contract (success) or to the parent's process
999  * contract (skip).  If requested, we also send a fork event to that
1000  * contract.
1001  *
1002  * Because contract_process_fork() may fail, and because we would
1003  * prefer that process contracts not be created for processes which
1004  * don't complete forking, this should be the last function called
1005  * before the "all clear" point in cfork.
1006  */
1007 cont_process_t *
1008 contract_process_fork(ctmpl_process_t *rtmpl, proc_t *cp, proc_t *pp,
1009     int canfail)
1010 {
1011 	contract_t *ct;
1012 	cont_process_t *ctp;
1013 	ct_kevent_t *event;
1014 	ct_template_t *tmpl;
1015 
1016 	if (rtmpl == NULL && (tmpl = ttolwp(curthread)->lwp_ct_active[
1017 	    process_type->ct_type_index]) != NULL)
1018 		rtmpl = tmpl->ctmpl_data;
1019 
1020 	if (rtmpl == NULL)
1021 		ctp = curproc->p_ct_process;
1022 	else if ((ctp = contract_process_create(rtmpl, pp, canfail)) == NULL)
1023 		return (NULL);
1024 
1025 	ct = &ctp->conp_contract;
1026 	/*
1027 	 * Prevent contract_process_kill() from missing forked children
1028 	 * by failing forks by parents that have just been killed.
1029 	 * It's not worth hoisting the ctp test since contract creation
1030 	 * is by no means the common case.
1031 	 */
1032 	mutex_enter(&ct->ct_lock);
1033 	mutex_enter(&pp->p_lock);
1034 	if (ctp == curproc->p_ct_process && (pp->p_flag & SKILLED) != 0 &&
1035 	    canfail) {
1036 		mutex_exit(&pp->p_lock);
1037 		mutex_exit(&ct->ct_lock);
1038 		return (NULL);
1039 	}
1040 	cp->p_ct_process = ctp;
1041 	mutex_exit(&pp->p_lock);
1042 	contract_hold(ct);
1043 	list_insert_head(&ctp->conp_members, cp);
1044 	ctp->conp_nmembers++;
1045 	mutex_exit(&ct->ct_lock);
1046 	if (EVSENDP(ctp, CT_PR_EV_FORK)) {
1047 		nvlist_t *nvl;
1048 
1049 		VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1050 		VERIFY(nvlist_add_uint32(nvl, CTPE_PID, cp->p_pid) == 0);
1051 		VERIFY(nvlist_add_uint32(nvl, CTPE_PPID, pp->p_pid) == 0);
1052 
1053 		event = kmem_zalloc(sizeof (ct_kevent_t), KM_SLEEP);
1054 		event->cte_flags = EVINFOP(ctp, CT_PR_EV_FORK) ? CTE_INFO : 0;
1055 		event->cte_type = CT_PR_EV_FORK;
1056 		(void) cte_publish_all(ct, event, nvl, NULL);
1057 	}
1058 	return (ctp);
1059 }
1060 
1061 /*
1062  * contract_process_core
1063  *
1064  * Called on core file generation attempts.  Generates a core event, if
1065  * requested, containing the names of the process, global, and
1066  * system-global ("zone") core files.  If dumping core is in the fatal
1067  * event set, calls contract_process_kill().
1068  */
1069 void
1070 contract_process_core(cont_process_t *ctp, proc_t *p, int sig,
1071     const char *process, const char *global, const char *zone)
1072 {
1073 	contract_t *ct = &ctp->conp_contract;
1074 
1075 	if (EVSENDP(ctp, CT_PR_EV_CORE)) {
1076 		ct_kevent_t *event;
1077 		nvlist_t *nvl, *gnvl = NULL;
1078 
1079 		VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1080 		VERIFY(nvlist_add_uint32(nvl, CTPE_PID, p->p_pid) == 0);
1081 		VERIFY(nvlist_add_uint32(nvl, CTPE_SIGNAL, sig) == 0);
1082 		if (process)
1083 			VERIFY(nvlist_add_string(nvl, CTPE_PCOREFILE,
1084 			    (char *)process) == 0);
1085 		if (global)
1086 			VERIFY(nvlist_add_string(nvl, CTPE_GCOREFILE,
1087 			    (char *)global) == 0);
1088 
1089 		if (zone) {
1090 			/*
1091 			 * Only the global zone is informed of the
1092 			 * local-zone generated global-zone core.
1093 			 */
1094 			VERIFY(nvlist_alloc(&gnvl, NV_UNIQUE_NAME,
1095 			    KM_SLEEP) == 0);
1096 			VERIFY(nvlist_add_string(gnvl, CTPE_ZCOREFILE,
1097 			    (char *)zone) == 0);
1098 		}
1099 
1100 		event = kmem_zalloc(sizeof (ct_kevent_t), KM_SLEEP);
1101 		event->cte_flags = EVINFOP(ctp, CT_PR_EV_CORE) ? CTE_INFO : 0;
1102 		event->cte_type = CT_PR_EV_CORE;
1103 		(void) cte_publish_all(ct, event, nvl, gnvl);
1104 	}
1105 
1106 	if (EVFATALP(ctp, CT_PR_EV_CORE)) {
1107 		mutex_enter(&ct->ct_lock);
1108 		contract_process_kill(ct, p, B_TRUE);
1109 		mutex_exit(&ct->ct_lock);
1110 	}
1111 }
1112 
1113 /*
1114  * contract_process_hwerr
1115  *
1116  * Called when a process is killed by an unrecoverable hardware error.
1117  * Generates an hwerr event, if requested.  If hardware errors are in
1118  * the fatal event set, calls contract_process_kill().
1119  */
1120 void
1121 contract_process_hwerr(cont_process_t *ctp, proc_t *p)
1122 {
1123 	contract_t *ct = &ctp->conp_contract;
1124 
1125 	if (EVSENDP(ctp, CT_PR_EV_HWERR)) {
1126 		ct_kevent_t *event;
1127 		nvlist_t *nvl;
1128 
1129 		VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1130 		VERIFY(nvlist_add_uint32(nvl, CTPE_PID, p->p_pid) == 0);
1131 
1132 		event = kmem_zalloc(sizeof (ct_kevent_t), KM_SLEEP);
1133 		event->cte_flags = EVINFOP(ctp, CT_PR_EV_HWERR) ? CTE_INFO : 0;
1134 		event->cte_type = CT_PR_EV_HWERR;
1135 		(void) cte_publish_all(ct, event, nvl, NULL);
1136 	}
1137 
1138 	if (EVFATALP(ctp, CT_PR_EV_HWERR)) {
1139 		mutex_enter(&ct->ct_lock);
1140 		contract_process_kill(ct, p, B_FALSE);
1141 		mutex_exit(&ct->ct_lock);
1142 	}
1143 }
1144 
1145 /*
1146  * contract_process_sig
1147  *
1148  * Called when a process is killed by a signal originating from a
1149  * process outside of its process contract or its process contract's
1150  * holder.  Generates an signal event, if requested, containing the
1151  * signal number, and the sender's pid and contract id (if available).
1152  * If signals are in the fatal event set, calls
1153  * contract_process_kill().
1154  */
1155 void
1156 contract_process_sig(cont_process_t *ctp, proc_t *p, int sig, pid_t pid,
1157     ctid_t ctid, zoneid_t zoneid)
1158 {
1159 	contract_t *ct = &ctp->conp_contract;
1160 
1161 	if (EVSENDP(ctp, CT_PR_EV_SIGNAL)) {
1162 		ct_kevent_t *event;
1163 		nvlist_t *dest, *nvl, *gnvl = NULL;
1164 
1165 		VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1166 		VERIFY(nvlist_add_uint32(nvl, CTPE_PID, p->p_pid) == 0);
1167 		VERIFY(nvlist_add_uint32(nvl, CTPE_SIGNAL, sig) == 0);
1168 
1169 		if (zoneid >= 0 && p->p_zone->zone_id != zoneid) {
1170 			VERIFY(nvlist_alloc(&gnvl, NV_UNIQUE_NAME,
1171 			    KM_SLEEP) == 0);
1172 			dest = gnvl;
1173 		} else {
1174 			dest = nvl;
1175 		}
1176 
1177 		if (pid != -1)
1178 			VERIFY(nvlist_add_uint32(dest, CTPE_SENDER, pid) == 0);
1179 		if (ctid != 0)
1180 			VERIFY(nvlist_add_uint32(dest, CTPE_SENDCT, ctid) == 0);
1181 
1182 		event = kmem_zalloc(sizeof (ct_kevent_t), KM_SLEEP);
1183 		event->cte_flags = EVINFOP(ctp, CT_PR_EV_SIGNAL) ? CTE_INFO : 0;
1184 		event->cte_type = CT_PR_EV_SIGNAL;
1185 		(void) cte_publish_all(ct, event, nvl, gnvl);
1186 	}
1187 
1188 	if (EVFATALP(ctp, CT_PR_EV_SIGNAL)) {
1189 		mutex_enter(&ct->ct_lock);
1190 		contract_process_kill(ct, p, B_TRUE);
1191 		mutex_exit(&ct->ct_lock);
1192 	}
1193 }
1194