xref: /freebsd/sys/security/audit/audit.c (revision 2a01feab)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1999-2005 Apple Inc.
5  * Copyright (c) 2006-2007, 2016-2018 Robert N. M. Watson
6  * All rights reserved.
7  *
8  * Portions of this software were developed by BAE Systems, the University of
9  * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL
10  * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent
11  * Computing (TC) research program.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1.  Redistributions of source code must retain the above copyright
17  *     notice, this list of conditions and the following disclaimer.
18  * 2.  Redistributions in binary form must reproduce the above copyright
19  *     notice, this list of conditions and the following disclaimer in the
20  *     documentation and/or other materials provided with the distribution.
21  * 3.  Neither the name of Apple Inc. ("Apple") nor the names of
22  *     its contributors may be used to endorse or promote products derived
23  *     from this software without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY APPLE AND ITS CONTRIBUTORS "AS IS" AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED. IN NO EVENT SHALL APPLE OR ITS CONTRIBUTORS BE LIABLE FOR
29  * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
33  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
34  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
35  * POSSIBILITY OF SUCH DAMAGE.
36  */
37 
38 #include <sys/cdefs.h>
39 __FBSDID("$FreeBSD$");
40 
41 #include <sys/param.h>
42 #include <sys/condvar.h>
43 #include <sys/conf.h>
44 #include <sys/file.h>
45 #include <sys/filedesc.h>
46 #include <sys/fcntl.h>
47 #include <sys/ipc.h>
48 #include <sys/jail.h>
49 #include <sys/kernel.h>
50 #include <sys/kthread.h>
51 #include <sys/malloc.h>
52 #include <sys/mount.h>
53 #include <sys/namei.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/queue.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/protosw.h>
60 #include <sys/domain.h>
61 #include <sys/sysctl.h>
62 #include <sys/sysproto.h>
63 #include <sys/sysent.h>
64 #include <sys/systm.h>
65 #include <sys/ucred.h>
66 #include <sys/uio.h>
67 #include <sys/un.h>
68 #include <sys/unistd.h>
69 #include <sys/vnode.h>
70 
71 #include <bsm/audit.h>
72 #include <bsm/audit_internal.h>
73 #include <bsm/audit_kevents.h>
74 
75 #include <netinet/in.h>
76 #include <netinet/in_pcb.h>
77 
78 #include <security/audit/audit.h>
79 #include <security/audit/audit_private.h>
80 
81 #include <vm/uma.h>
82 
83 FEATURE(audit, "BSM audit support");
84 
85 static uma_zone_t	audit_record_zone;
86 static MALLOC_DEFINE(M_AUDITCRED, "audit_cred", "Audit cred storage");
87 MALLOC_DEFINE(M_AUDITDATA, "audit_data", "Audit data storage");
88 MALLOC_DEFINE(M_AUDITPATH, "audit_path", "Audit path storage");
89 MALLOC_DEFINE(M_AUDITTEXT, "audit_text", "Audit text storage");
90 MALLOC_DEFINE(M_AUDITGIDSET, "audit_gidset", "Audit GID set storage");
91 
92 static SYSCTL_NODE(_security, OID_AUTO, audit, CTLFLAG_RW, 0,
93     "TrustedBSD audit controls");
94 
95 /*
96  * Audit control settings that are set/read by system calls and are hence
97  * non-static.
98  *
99  * Define the audit control flags.
100  */
101 int			audit_trail_enabled;
102 int			audit_trail_suspended;
103 #ifdef KDTRACE_HOOKS
104 u_int			audit_dtrace_enabled;
105 #endif
106 int __read_frequently	audit_syscalls_enabled;
107 
108 /*
109  * Flags controlling behavior in low storage situations.  Should we panic if
110  * a write fails?  Should we fail stop if we're out of disk space?
111  */
112 int			audit_panic_on_write_fail;
113 int			audit_fail_stop;
114 int			audit_argv;
115 int			audit_arge;
116 
117 /*
118  * Are we currently "failing stop" due to out of disk space?
119  */
120 int			audit_in_failure;
121 
122 /*
123  * Global audit statistics.
124  */
125 struct audit_fstat	audit_fstat;
126 
127 /*
128  * Preselection mask for non-attributable events.
129  */
130 struct au_mask		audit_nae_mask;
131 
132 /*
133  * Mutex to protect global variables shared between various threads and
134  * processes.
135  */
136 struct mtx		audit_mtx;
137 
138 /*
139  * Queue of audit records ready for delivery to disk.  We insert new records
140  * at the tail, and remove records from the head.  Also, a count of the
141  * number of records used for checking queue depth.  In addition, a counter
142  * of records that we have allocated but are not yet in the queue, which is
143  * needed to estimate the total size of the combined set of records
144  * outstanding in the system.
145  */
146 struct kaudit_queue	audit_q;
147 int			audit_q_len;
148 int			audit_pre_q_len;
149 
150 /*
151  * Audit queue control settings (minimum free, low/high water marks, etc.)
152  */
153 struct au_qctrl		audit_qctrl;
154 
155 /*
156  * Condition variable to signal to the worker that it has work to do: either
157  * new records are in the queue, or a log replacement is taking place.
158  */
159 struct cv		audit_worker_cv;
160 
161 /*
162  * Condition variable to flag when crossing the low watermark, meaning that
163  * threads blocked due to hitting the high watermark can wake up and continue
164  * to commit records.
165  */
166 struct cv		audit_watermark_cv;
167 
168 /*
169  * Condition variable for  auditing threads wait on when in fail-stop mode.
170  * Threads wait on this CV forever (and ever), never seeing the light of day
171  * again.
172  */
173 static struct cv	audit_fail_cv;
174 
175 /*
176  * Optional DTrace audit provider support: function pointers for preselection
177  * and commit events.
178  */
179 #ifdef KDTRACE_HOOKS
180 void	*(*dtaudit_hook_preselect)(au_id_t auid, au_event_t event,
181 	    au_class_t class);
182 int	(*dtaudit_hook_commit)(struct kaudit_record *kar, au_id_t auid,
183 	    au_event_t event, au_class_t class, int sorf);
184 void	(*dtaudit_hook_bsm)(struct kaudit_record *kar, au_id_t auid,
185 	    au_event_t event, au_class_t class, int sorf,
186 	    void *bsm_data, size_t bsm_lenlen);
187 #endif
188 
189 /*
190  * Kernel audit information.  This will store the current audit address
191  * or host information that the kernel will use when it's generating
192  * audit records.  This data is modified by the A_GET{SET}KAUDIT auditon(2)
193  * command.
194  */
195 static struct auditinfo_addr	audit_kinfo;
196 static struct rwlock		audit_kinfo_lock;
197 
198 #define	KINFO_LOCK_INIT()	rw_init(&audit_kinfo_lock, \
199 				    "audit_kinfo_lock")
200 #define	KINFO_RLOCK()		rw_rlock(&audit_kinfo_lock)
201 #define	KINFO_WLOCK()		rw_wlock(&audit_kinfo_lock)
202 #define	KINFO_RUNLOCK()		rw_runlock(&audit_kinfo_lock)
203 #define	KINFO_WUNLOCK()		rw_wunlock(&audit_kinfo_lock)
204 
205 /*
206  * Check various policies to see if we should enable system-call audit hooks.
207  * Note that despite the mutex being held, we want to assign a value exactly
208  * once, as checks of the flag are performed lock-free for performance
209  * reasons.  The mutex is used to get a consistent snapshot of policy state --
210  * e.g., safely accessing the two audit_trail flags.
211  */
212 void
213 audit_syscalls_enabled_update(void)
214 {
215 
216 	mtx_lock(&audit_mtx);
217 #ifdef KDTRACE_HOOKS
218 	if (audit_dtrace_enabled)
219 		audit_syscalls_enabled = 1;
220 	else {
221 #endif
222 		if (audit_trail_enabled && !audit_trail_suspended)
223 			audit_syscalls_enabled = 1;
224 		else
225 			audit_syscalls_enabled = 0;
226 #ifdef KDTRACE_HOOKS
227 	}
228 #endif
229 	mtx_unlock(&audit_mtx);
230 }
231 
232 void
233 audit_set_kinfo(struct auditinfo_addr *ak)
234 {
235 
236 	KASSERT(ak->ai_termid.at_type == AU_IPv4 ||
237 	    ak->ai_termid.at_type == AU_IPv6,
238 	    ("audit_set_kinfo: invalid address type"));
239 
240 	KINFO_WLOCK();
241 	audit_kinfo = *ak;
242 	KINFO_WUNLOCK();
243 }
244 
245 void
246 audit_get_kinfo(struct auditinfo_addr *ak)
247 {
248 
249 	KASSERT(audit_kinfo.ai_termid.at_type == AU_IPv4 ||
250 	    audit_kinfo.ai_termid.at_type == AU_IPv6,
251 	    ("audit_set_kinfo: invalid address type"));
252 
253 	KINFO_RLOCK();
254 	*ak = audit_kinfo;
255 	KINFO_RUNLOCK();
256 }
257 
258 /*
259  * Construct an audit record for the passed thread.
260  */
261 static int
262 audit_record_ctor(void *mem, int size, void *arg, int flags)
263 {
264 	struct kaudit_record *ar;
265 	struct thread *td;
266 	struct ucred *cred;
267 	struct prison *pr;
268 
269 	KASSERT(sizeof(*ar) == size, ("audit_record_ctor: wrong size"));
270 
271 	td = arg;
272 	ar = mem;
273 	bzero(ar, sizeof(*ar));
274 	ar->k_ar.ar_magic = AUDIT_RECORD_MAGIC;
275 	nanotime(&ar->k_ar.ar_starttime);
276 
277 	/*
278 	 * Export the subject credential.
279 	 */
280 	cred = td->td_ucred;
281 	cru2x(cred, &ar->k_ar.ar_subj_cred);
282 	ar->k_ar.ar_subj_ruid = cred->cr_ruid;
283 	ar->k_ar.ar_subj_rgid = cred->cr_rgid;
284 	ar->k_ar.ar_subj_egid = cred->cr_groups[0];
285 	ar->k_ar.ar_subj_auid = cred->cr_audit.ai_auid;
286 	ar->k_ar.ar_subj_asid = cred->cr_audit.ai_asid;
287 	ar->k_ar.ar_subj_pid = td->td_proc->p_pid;
288 	ar->k_ar.ar_subj_amask = cred->cr_audit.ai_mask;
289 	ar->k_ar.ar_subj_term_addr = cred->cr_audit.ai_termid;
290 	/*
291 	 * If this process is jailed, make sure we capture the name of the
292 	 * jail so we can use it to generate a zonename token when we covert
293 	 * this record to BSM.
294 	 */
295 	if (jailed(cred)) {
296 		pr = cred->cr_prison;
297 		(void) strlcpy(ar->k_ar.ar_jailname, pr->pr_name,
298 		    sizeof(ar->k_ar.ar_jailname));
299 	} else
300 		ar->k_ar.ar_jailname[0] = '\0';
301 	return (0);
302 }
303 
304 static void
305 audit_record_dtor(void *mem, int size, void *arg)
306 {
307 	struct kaudit_record *ar;
308 
309 	KASSERT(sizeof(*ar) == size, ("audit_record_dtor: wrong size"));
310 
311 	ar = mem;
312 	if (ar->k_ar.ar_arg_upath1 != NULL)
313 		free(ar->k_ar.ar_arg_upath1, M_AUDITPATH);
314 	if (ar->k_ar.ar_arg_upath2 != NULL)
315 		free(ar->k_ar.ar_arg_upath2, M_AUDITPATH);
316 	if (ar->k_ar.ar_arg_text != NULL)
317 		free(ar->k_ar.ar_arg_text, M_AUDITTEXT);
318 	if (ar->k_udata != NULL)
319 		free(ar->k_udata, M_AUDITDATA);
320 	if (ar->k_ar.ar_arg_argv != NULL)
321 		free(ar->k_ar.ar_arg_argv, M_AUDITTEXT);
322 	if (ar->k_ar.ar_arg_envv != NULL)
323 		free(ar->k_ar.ar_arg_envv, M_AUDITTEXT);
324 	if (ar->k_ar.ar_arg_groups.gidset != NULL)
325 		free(ar->k_ar.ar_arg_groups.gidset, M_AUDITGIDSET);
326 }
327 
328 /*
329  * Initialize the Audit subsystem: configuration state, work queue,
330  * synchronization primitives, worker thread, and trigger device node.  Also
331  * call into the BSM assembly code to initialize it.
332  */
333 static void
334 audit_init(void)
335 {
336 
337 	audit_trail_enabled = 0;
338 	audit_trail_suspended = 0;
339 	audit_syscalls_enabled = 0;
340 	audit_panic_on_write_fail = 0;
341 	audit_fail_stop = 0;
342 	audit_in_failure = 0;
343 	audit_argv = 0;
344 	audit_arge = 0;
345 
346 	audit_fstat.af_filesz = 0;	/* '0' means unset, unbounded. */
347 	audit_fstat.af_currsz = 0;
348 	audit_nae_mask.am_success = 0;
349 	audit_nae_mask.am_failure = 0;
350 
351 	TAILQ_INIT(&audit_q);
352 	audit_q_len = 0;
353 	audit_pre_q_len = 0;
354 	audit_qctrl.aq_hiwater = AQ_HIWATER;
355 	audit_qctrl.aq_lowater = AQ_LOWATER;
356 	audit_qctrl.aq_bufsz = AQ_BUFSZ;
357 	audit_qctrl.aq_minfree = AU_FS_MINFREE;
358 
359 	audit_kinfo.ai_termid.at_type = AU_IPv4;
360 	audit_kinfo.ai_termid.at_addr[0] = INADDR_ANY;
361 
362 	mtx_init(&audit_mtx, "audit_mtx", NULL, MTX_DEF);
363 	KINFO_LOCK_INIT();
364 	cv_init(&audit_worker_cv, "audit_worker_cv");
365 	cv_init(&audit_watermark_cv, "audit_watermark_cv");
366 	cv_init(&audit_fail_cv, "audit_fail_cv");
367 
368 	audit_record_zone = uma_zcreate("audit_record",
369 	    sizeof(struct kaudit_record), audit_record_ctor,
370 	    audit_record_dtor, NULL, NULL, UMA_ALIGN_PTR, 0);
371 
372 	/* First initialisation of audit_syscalls_enabled. */
373 	audit_syscalls_enabled_update();
374 
375 	/* Initialize the BSM audit subsystem. */
376 	kau_init();
377 
378 	audit_trigger_init();
379 
380 	/* Register shutdown handler. */
381 	EVENTHANDLER_REGISTER(shutdown_pre_sync, audit_shutdown, NULL,
382 	    SHUTDOWN_PRI_FIRST);
383 
384 	/* Start audit worker thread. */
385 	audit_worker_init();
386 }
387 
388 SYSINIT(audit_init, SI_SUB_AUDIT, SI_ORDER_FIRST, audit_init, NULL);
389 
390 /*
391  * Drain the audit queue and close the log at shutdown.  Note that this can
392  * be called both from the system shutdown path and also from audit
393  * configuration syscalls, so 'arg' and 'howto' are ignored.
394  *
395  * XXXRW: In FreeBSD 7.x and 8.x, this fails to wait for the record queue to
396  * drain before returning, which could lead to lost records on shutdown.
397  */
398 void
399 audit_shutdown(void *arg, int howto)
400 {
401 
402 	audit_rotate_vnode(NULL, NULL);
403 }
404 
405 /*
406  * Return the current thread's audit record, if any.
407  */
408 struct kaudit_record *
409 currecord(void)
410 {
411 
412 	return (curthread->td_ar);
413 }
414 
415 /*
416  * XXXAUDIT: Shouldn't there be logic here to sleep waiting on available
417  * pre_q space, suspending the system call until there is room?
418  */
419 struct kaudit_record *
420 audit_new(int event, struct thread *td)
421 {
422 	struct kaudit_record *ar;
423 
424 	/*
425 	 * Note: the number of outstanding uncommitted audit records is
426 	 * limited to the number of concurrent threads servicing system calls
427 	 * in the kernel.
428 	 */
429 	ar = uma_zalloc_arg(audit_record_zone, td, M_WAITOK);
430 	ar->k_ar.ar_event = event;
431 
432 	mtx_lock(&audit_mtx);
433 	audit_pre_q_len++;
434 	mtx_unlock(&audit_mtx);
435 
436 	return (ar);
437 }
438 
439 void
440 audit_free(struct kaudit_record *ar)
441 {
442 
443 	uma_zfree(audit_record_zone, ar);
444 }
445 
446 void
447 audit_commit(struct kaudit_record *ar, int error, int retval)
448 {
449 	au_event_t event;
450 	au_class_t class;
451 	au_id_t auid;
452 	int sorf;
453 	struct au_mask *aumask;
454 
455 	if (ar == NULL)
456 		return;
457 
458 	ar->k_ar.ar_errno = error;
459 	ar->k_ar.ar_retval = retval;
460 	nanotime(&ar->k_ar.ar_endtime);
461 
462 	/*
463 	 * Decide whether to commit the audit record by checking the error
464 	 * value from the system call and using the appropriate audit mask.
465 	 */
466 	if (ar->k_ar.ar_subj_auid == AU_DEFAUDITID)
467 		aumask = &audit_nae_mask;
468 	else
469 		aumask = &ar->k_ar.ar_subj_amask;
470 
471 	if (error)
472 		sorf = AU_PRS_FAILURE;
473 	else
474 		sorf = AU_PRS_SUCCESS;
475 
476 	/*
477 	 * syscalls.master sometimes contains a prototype event number, which
478 	 * we will transform into a more specific event number now that we
479 	 * have more complete information gathered during the system call.
480 	 */
481 	switch(ar->k_ar.ar_event) {
482 	case AUE_OPEN_RWTC:
483 		ar->k_ar.ar_event = audit_flags_and_error_to_openevent(
484 		    ar->k_ar.ar_arg_fflags, error);
485 		break;
486 
487 	case AUE_OPENAT_RWTC:
488 		ar->k_ar.ar_event = audit_flags_and_error_to_openatevent(
489 		    ar->k_ar.ar_arg_fflags, error);
490 		break;
491 
492 	case AUE_SYSCTL:
493 		ar->k_ar.ar_event = audit_ctlname_to_sysctlevent(
494 		    ar->k_ar.ar_arg_ctlname, ar->k_ar.ar_valid_arg);
495 		break;
496 
497 	case AUE_AUDITON:
498 		/* Convert the auditon() command to an event. */
499 		ar->k_ar.ar_event = auditon_command_event(ar->k_ar.ar_arg_cmd);
500 		break;
501 
502 	case AUE_MSGSYS:
503 		if (ARG_IS_VALID(ar, ARG_SVIPC_WHICH))
504 			ar->k_ar.ar_event =
505 			    audit_msgsys_to_event(ar->k_ar.ar_arg_svipc_which);
506 		break;
507 
508 	case AUE_SEMSYS:
509 		if (ARG_IS_VALID(ar, ARG_SVIPC_WHICH))
510 			ar->k_ar.ar_event =
511 			    audit_semsys_to_event(ar->k_ar.ar_arg_svipc_which);
512 		break;
513 
514 	case AUE_SHMSYS:
515 		if (ARG_IS_VALID(ar, ARG_SVIPC_WHICH))
516 			ar->k_ar.ar_event =
517 			    audit_shmsys_to_event(ar->k_ar.ar_arg_svipc_which);
518 		break;
519 	}
520 
521 	auid = ar->k_ar.ar_subj_auid;
522 	event = ar->k_ar.ar_event;
523 	class = au_event_class(event);
524 
525 	ar->k_ar_commit |= AR_COMMIT_KERNEL;
526 	if (au_preselect(event, class, aumask, sorf) != 0)
527 		ar->k_ar_commit |= AR_PRESELECT_TRAIL;
528 	if (audit_pipe_preselect(auid, event, class, sorf,
529 	    ar->k_ar_commit & AR_PRESELECT_TRAIL) != 0)
530 		ar->k_ar_commit |= AR_PRESELECT_PIPE;
531 #ifdef KDTRACE_HOOKS
532 	/*
533 	 * Expose the audit record to DTrace, both to allow the "commit" probe
534 	 * to fire if it's desirable, and also to allow a decision to be made
535 	 * about later firing with BSM in the audit worker.
536 	 */
537 	if (dtaudit_hook_commit != NULL) {
538 		if (dtaudit_hook_commit(ar, auid, event, class, sorf) != 0)
539 			ar->k_ar_commit |= AR_PRESELECT_DTRACE;
540 	}
541 #endif
542 
543 	if ((ar->k_ar_commit & (AR_PRESELECT_TRAIL | AR_PRESELECT_PIPE |
544 	    AR_PRESELECT_USER_TRAIL | AR_PRESELECT_USER_PIPE |
545 	    AR_PRESELECT_DTRACE)) == 0) {
546 		mtx_lock(&audit_mtx);
547 		audit_pre_q_len--;
548 		mtx_unlock(&audit_mtx);
549 		audit_free(ar);
550 		return;
551 	}
552 
553 	/*
554 	 * Note: it could be that some records initiated while audit was
555 	 * enabled should still be committed?
556 	 *
557 	 * NB: The check here is not for audit_syscalls because any
558 	 * DTrace-related obligations have been fulfilled above -- we're just
559 	 * down to the trail and pipes now.
560 	 */
561 	mtx_lock(&audit_mtx);
562 	if (audit_trail_suspended || !audit_trail_enabled) {
563 		audit_pre_q_len--;
564 		mtx_unlock(&audit_mtx);
565 		audit_free(ar);
566 		return;
567 	}
568 
569 	/*
570 	 * Constrain the number of committed audit records based on the
571 	 * configurable parameter.
572 	 */
573 	while (audit_q_len >= audit_qctrl.aq_hiwater)
574 		cv_wait(&audit_watermark_cv, &audit_mtx);
575 
576 	TAILQ_INSERT_TAIL(&audit_q, ar, k_q);
577 	audit_q_len++;
578 	audit_pre_q_len--;
579 	cv_signal(&audit_worker_cv);
580 	mtx_unlock(&audit_mtx);
581 }
582 
583 /*
584  * audit_syscall_enter() is called on entry to each system call.  It is
585  * responsible for deciding whether or not to audit the call (preselection),
586  * and if so, allocating a per-thread audit record.  audit_new() will fill in
587  * basic thread/credential properties.
588  *
589  * This function will be entered only if audit_syscalls_enabled was set in the
590  * macro wrapper for this function.  It could be cleared by the time this
591  * function runs, but that is an acceptable race.
592  */
593 void
594 audit_syscall_enter(unsigned short code, struct thread *td)
595 {
596 	struct au_mask *aumask;
597 #ifdef KDTRACE_HOOKS
598 	void *dtaudit_state;
599 #endif
600 	au_class_t class;
601 	au_event_t event;
602 	au_id_t auid;
603 	int record_needed;
604 
605 	KASSERT(td->td_ar == NULL, ("audit_syscall_enter: td->td_ar != NULL"));
606 	KASSERT((td->td_pflags & TDP_AUDITREC) == 0,
607 	    ("audit_syscall_enter: TDP_AUDITREC set"));
608 
609 	/*
610 	 * In FreeBSD, each ABI has its own system call table, and hence
611 	 * mapping of system call codes to audit events.  Convert the code to
612 	 * an audit event identifier using the process system call table
613 	 * reference.  In Darwin, there's only one, so we use the global
614 	 * symbol for the system call table.  No audit record is generated
615 	 * for bad system calls, as no operation has been performed.
616 	 */
617 	if (code >= td->td_proc->p_sysent->sv_size)
618 		return;
619 
620 	event = td->td_proc->p_sysent->sv_table[code].sy_auevent;
621 	if (event == AUE_NULL)
622 		return;
623 
624 	/*
625 	 * Check which audit mask to use; either the kernel non-attributable
626 	 * event mask or the process audit mask.
627 	 */
628 	auid = td->td_ucred->cr_audit.ai_auid;
629 	if (auid == AU_DEFAUDITID)
630 		aumask = &audit_nae_mask;
631 	else
632 		aumask = &td->td_ucred->cr_audit.ai_mask;
633 
634 	/*
635 	 * Determine whether trail or pipe preselection would like an audit
636 	 * record allocated for this system call.
637 	 */
638 	class = au_event_class(event);
639 	if (au_preselect(event, class, aumask, AU_PRS_BOTH)) {
640 		/*
641 		 * If we're out of space and need to suspend unprivileged
642 		 * processes, do that here rather than trying to allocate
643 		 * another audit record.
644 		 *
645 		 * Note: we might wish to be able to continue here in the
646 		 * future, if the system recovers.  That should be possible
647 		 * by means of checking the condition in a loop around
648 		 * cv_wait().  It might be desirable to reevaluate whether an
649 		 * audit record is still required for this event by
650 		 * re-calling au_preselect().
651 		 */
652 		if (audit_in_failure &&
653 		    priv_check(td, PRIV_AUDIT_FAILSTOP) != 0) {
654 			cv_wait(&audit_fail_cv, &audit_mtx);
655 			panic("audit_failing_stop: thread continued");
656 		}
657 		record_needed = 1;
658 	} else if (audit_pipe_preselect(auid, event, class, AU_PRS_BOTH, 0)) {
659 		record_needed = 1;
660 	} else {
661 		record_needed = 0;
662 	}
663 
664 	/*
665 	 * After audit trails and pipes have made their policy choices, DTrace
666 	 * may request that records be generated as well.  This is a slightly
667 	 * complex affair, as the DTrace audit provider needs the audit
668 	 * framework to maintain some state on the audit record, which has not
669 	 * been allocated at the point where the decision has to be made.
670 	 * This hook must run even if we are not changing the decision, as
671 	 * DTrace may want to stick event state onto a record we were going to
672 	 * produce due to the trail or pipes.  The event state returned by the
673 	 * DTrace provider must be safe without locks held between here and
674 	 * below -- i.e., dtaudit_state must must refer to stable memory.
675 	 */
676 #ifdef KDTRACE_HOOKS
677 	dtaudit_state = NULL;
678         if (dtaudit_hook_preselect != NULL) {
679 		dtaudit_state = dtaudit_hook_preselect(auid, event, class);
680 		if (dtaudit_state != NULL)
681 			record_needed = 1;
682 	}
683 #endif
684 
685 	/*
686 	 * If a record is required, allocate it and attach it to the thread
687 	 * for use throughout the system call.  Also attach DTrace state if
688 	 * required.
689 	 *
690 	 * XXXRW: If we decide to reference count the evname_elem underlying
691 	 * dtaudit_state, we will need to free here if no record is allocated
692 	 * or allocatable.
693 	 */
694 	if (record_needed) {
695 		td->td_ar = audit_new(event, td);
696 		if (td->td_ar != NULL) {
697 			td->td_pflags |= TDP_AUDITREC;
698 #ifdef KDTRACE_HOOKS
699 			td->td_ar->k_dtaudit_state = dtaudit_state;
700 #endif
701 		}
702 	} else
703 		td->td_ar = NULL;
704 }
705 
706 /*
707  * audit_syscall_exit() is called from the return of every system call, or in
708  * the event of exit1(), during the execution of exit1().  It is responsible
709  * for committing the audit record, if any, along with return condition.
710  */
711 void
712 audit_syscall_exit(int error, struct thread *td)
713 {
714 	int retval;
715 
716 	/*
717 	 * Commit the audit record as desired; once we pass the record into
718 	 * audit_commit(), the memory is owned by the audit subsystem.  The
719 	 * return value from the system call is stored on the user thread.
720 	 * If there was an error, the return value is set to -1, imitating
721 	 * the behavior of the cerror routine.
722 	 */
723 	if (error)
724 		retval = -1;
725 	else
726 		retval = td->td_retval[0];
727 
728 	audit_commit(td->td_ar, error, retval);
729 	td->td_ar = NULL;
730 	td->td_pflags &= ~TDP_AUDITREC;
731 }
732 
733 void
734 audit_cred_copy(struct ucred *src, struct ucred *dest)
735 {
736 
737 	bcopy(&src->cr_audit, &dest->cr_audit, sizeof(dest->cr_audit));
738 }
739 
740 void
741 audit_cred_destroy(struct ucred *cred)
742 {
743 
744 }
745 
746 void
747 audit_cred_init(struct ucred *cred)
748 {
749 
750 	bzero(&cred->cr_audit, sizeof(cred->cr_audit));
751 }
752 
753 /*
754  * Initialize audit information for the first kernel process (proc 0) and for
755  * the first user process (init).
756  */
757 void
758 audit_cred_kproc0(struct ucred *cred)
759 {
760 
761 	cred->cr_audit.ai_auid = AU_DEFAUDITID;
762 	cred->cr_audit.ai_termid.at_type = AU_IPv4;
763 }
764 
765 void
766 audit_cred_proc1(struct ucred *cred)
767 {
768 
769 	cred->cr_audit.ai_auid = AU_DEFAUDITID;
770 	cred->cr_audit.ai_termid.at_type = AU_IPv4;
771 }
772 
773 void
774 audit_thread_alloc(struct thread *td)
775 {
776 
777 	td->td_ar = NULL;
778 }
779 
780 void
781 audit_thread_free(struct thread *td)
782 {
783 
784 	KASSERT(td->td_ar == NULL, ("audit_thread_free: td_ar != NULL"));
785 	KASSERT((td->td_pflags & TDP_AUDITREC) == 0,
786 	    ("audit_thread_free: TDP_AUDITREC set"));
787 }
788 
789 void
790 audit_proc_coredump(struct thread *td, char *path, int errcode)
791 {
792 	struct kaudit_record *ar;
793 	struct au_mask *aumask;
794 	struct ucred *cred;
795 	au_class_t class;
796 	int ret, sorf;
797 	char **pathp;
798 	au_id_t auid;
799 
800 	ret = 0;
801 
802 	/*
803 	 * Make sure we are using the correct preselection mask.
804 	 */
805 	cred = td->td_ucred;
806 	auid = cred->cr_audit.ai_auid;
807 	if (auid == AU_DEFAUDITID)
808 		aumask = &audit_nae_mask;
809 	else
810 		aumask = &cred->cr_audit.ai_mask;
811 	/*
812 	 * It's possible for coredump(9) generation to fail.  Make sure that
813 	 * we handle this case correctly for preselection.
814 	 */
815 	if (errcode != 0)
816 		sorf = AU_PRS_FAILURE;
817 	else
818 		sorf = AU_PRS_SUCCESS;
819 	class = au_event_class(AUE_CORE);
820 	if (au_preselect(AUE_CORE, class, aumask, sorf) == 0 &&
821 	    audit_pipe_preselect(auid, AUE_CORE, class, sorf, 0) == 0)
822 		return;
823 
824 	/*
825 	 * If we are interested in seeing this audit record, allocate it.
826 	 * Where possible coredump records should contain a pathname and arg32
827 	 * (signal) tokens.
828 	 */
829 	ar = audit_new(AUE_CORE, td);
830 	if (ar == NULL)
831 		return;
832 	if (path != NULL) {
833 		pathp = &ar->k_ar.ar_arg_upath1;
834 		*pathp = malloc(MAXPATHLEN, M_AUDITPATH, M_WAITOK);
835 		audit_canon_path(td, AT_FDCWD, path, *pathp);
836 		ARG_SET_VALID(ar, ARG_UPATH1);
837 	}
838 	ar->k_ar.ar_arg_signum = td->td_proc->p_sig;
839 	ARG_SET_VALID(ar, ARG_SIGNUM);
840 	if (errcode != 0)
841 		ret = 1;
842 	audit_commit(ar, errcode, ret);
843 }
844