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/eventhandler.h>
45#include <sys/file.h>
46#include <sys/filedesc.h>
47#include <sys/fcntl.h>
48#include <sys/ipc.h>
49#include <sys/jail.h>
50#include <sys/kernel.h>
51#include <sys/kthread.h>
52#include <sys/malloc.h>
53#include <sys/mount.h>
54#include <sys/namei.h>
55#include <sys/priv.h>
56#include <sys/proc.h>
57#include <sys/queue.h>
58#include <sys/socket.h>
59#include <sys/socketvar.h>
60#include <sys/protosw.h>
61#include <sys/domain.h>
62#include <sys/sysctl.h>
63#include <sys/sysproto.h>
64#include <sys/sysent.h>
65#include <sys/systm.h>
66#include <sys/ucred.h>
67#include <sys/uio.h>
68#include <sys/un.h>
69#include <sys/unistd.h>
70#include <sys/vnode.h>
71
72#include <bsm/audit.h>
73#include <bsm/audit_internal.h>
74#include <bsm/audit_kevents.h>
75
76#include <netinet/in.h>
77#include <netinet/in_pcb.h>
78
79#include <security/audit/audit.h>
80#include <security/audit/audit_private.h>
81
82#include <vm/uma.h>
83
84FEATURE(audit, "BSM audit support");
85
86static uma_zone_t	audit_record_zone;
87static MALLOC_DEFINE(M_AUDITCRED, "audit_cred", "Audit cred storage");
88MALLOC_DEFINE(M_AUDITDATA, "audit_data", "Audit data storage");
89MALLOC_DEFINE(M_AUDITPATH, "audit_path", "Audit path storage");
90MALLOC_DEFINE(M_AUDITTEXT, "audit_text", "Audit text storage");
91MALLOC_DEFINE(M_AUDITGIDSET, "audit_gidset", "Audit GID set storage");
92
93static SYSCTL_NODE(_security, OID_AUTO, audit, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
94    "TrustedBSD audit controls");
95
96/*
97 * Audit control settings that are set/read by system calls and are hence
98 * non-static.
99 *
100 * Define the audit control flags.
101 */
102int			audit_trail_enabled;
103int			audit_trail_suspended;
104#ifdef KDTRACE_HOOKS
105u_int			audit_dtrace_enabled;
106#endif
107bool __read_frequently	audit_syscalls_enabled;
108
109/*
110 * Flags controlling behavior in low storage situations.  Should we panic if
111 * a write fails?  Should we fail stop if we're out of disk space?
112 */
113int			audit_panic_on_write_fail;
114int			audit_fail_stop;
115int			audit_argv;
116int			audit_arge;
117
118/*
119 * Are we currently "failing stop" due to out of disk space?
120 */
121int			audit_in_failure;
122
123/*
124 * Global audit statistics.
125 */
126struct audit_fstat	audit_fstat;
127
128/*
129 * Preselection mask for non-attributable events.
130 */
131struct au_mask		audit_nae_mask;
132
133/*
134 * Mutex to protect global variables shared between various threads and
135 * processes.
136 */
137struct mtx		audit_mtx;
138
139/*
140 * Queue of audit records ready for delivery to disk.  We insert new records
141 * at the tail, and remove records from the head.  Also, a count of the
142 * number of records used for checking queue depth.  In addition, a counter
143 * of records that we have allocated but are not yet in the queue, which is
144 * needed to estimate the total size of the combined set of records
145 * outstanding in the system.
146 */
147struct kaudit_queue	audit_q;
148int			audit_q_len;
149int			audit_pre_q_len;
150
151/*
152 * Audit queue control settings (minimum free, low/high water marks, etc.)
153 */
154struct au_qctrl		audit_qctrl;
155
156/*
157 * Condition variable to signal to the worker that it has work to do: either
158 * new records are in the queue, or a log replacement is taking place.
159 */
160struct cv		audit_worker_cv;
161
162/*
163 * Condition variable to flag when crossing the low watermark, meaning that
164 * threads blocked due to hitting the high watermark can wake up and continue
165 * to commit records.
166 */
167struct cv		audit_watermark_cv;
168
169/*
170 * Condition variable for  auditing threads wait on when in fail-stop mode.
171 * Threads wait on this CV forever (and ever), never seeing the light of day
172 * again.
173 */
174static struct cv	audit_fail_cv;
175
176/*
177 * Optional DTrace audit provider support: function pointers for preselection
178 * and commit events.
179 */
180#ifdef KDTRACE_HOOKS
181void	*(*dtaudit_hook_preselect)(au_id_t auid, au_event_t event,
182	    au_class_t class);
183int	(*dtaudit_hook_commit)(struct kaudit_record *kar, au_id_t auid,
184	    au_event_t event, au_class_t class, int sorf);
185void	(*dtaudit_hook_bsm)(struct kaudit_record *kar, au_id_t auid,
186	    au_event_t event, au_class_t class, int sorf,
187	    void *bsm_data, size_t bsm_lenlen);
188#endif
189
190/*
191 * Kernel audit information.  This will store the current audit address
192 * or host information that the kernel will use when it's generating
193 * audit records.  This data is modified by the A_GET{SET}KAUDIT auditon(2)
194 * command.
195 */
196static struct auditinfo_addr	audit_kinfo;
197static struct rwlock		audit_kinfo_lock;
198
199#define	KINFO_LOCK_INIT()	rw_init(&audit_kinfo_lock, \
200				    "audit_kinfo_lock")
201#define	KINFO_RLOCK()		rw_rlock(&audit_kinfo_lock)
202#define	KINFO_WLOCK()		rw_wlock(&audit_kinfo_lock)
203#define	KINFO_RUNLOCK()		rw_runlock(&audit_kinfo_lock)
204#define	KINFO_WUNLOCK()		rw_wunlock(&audit_kinfo_lock)
205
206/*
207 * Check various policies to see if we should enable system-call audit hooks.
208 * Note that despite the mutex being held, we want to assign a value exactly
209 * once, as checks of the flag are performed lock-free for performance
210 * reasons.  The mutex is used to get a consistent snapshot of policy state --
211 * e.g., safely accessing the two audit_trail flags.
212 */
213void
214audit_syscalls_enabled_update(void)
215{
216
217	mtx_lock(&audit_mtx);
218#ifdef KDTRACE_HOOKS
219	if (audit_dtrace_enabled)
220		audit_syscalls_enabled = true;
221	else {
222#endif
223		if (audit_trail_enabled && !audit_trail_suspended)
224			audit_syscalls_enabled = true;
225		else
226			audit_syscalls_enabled = false;
227#ifdef KDTRACE_HOOKS
228	}
229#endif
230	mtx_unlock(&audit_mtx);
231}
232
233void
234audit_set_kinfo(struct auditinfo_addr *ak)
235{
236
237	KASSERT(ak->ai_termid.at_type == AU_IPv4 ||
238	    ak->ai_termid.at_type == AU_IPv6,
239	    ("audit_set_kinfo: invalid address type"));
240
241	KINFO_WLOCK();
242	audit_kinfo = *ak;
243	KINFO_WUNLOCK();
244}
245
246void
247audit_get_kinfo(struct auditinfo_addr *ak)
248{
249
250	KASSERT(audit_kinfo.ai_termid.at_type == AU_IPv4 ||
251	    audit_kinfo.ai_termid.at_type == AU_IPv6,
252	    ("audit_set_kinfo: invalid address type"));
253
254	KINFO_RLOCK();
255	*ak = audit_kinfo;
256	KINFO_RUNLOCK();
257}
258
259/*
260 * Construct an audit record for the passed thread.
261 */
262static int
263audit_record_ctor(void *mem, int size, void *arg, int flags)
264{
265	struct kaudit_record *ar;
266	struct thread *td;
267	struct ucred *cred;
268	struct prison *pr;
269
270	KASSERT(sizeof(*ar) == size, ("audit_record_ctor: wrong size"));
271
272	td = arg;
273	ar = mem;
274	bzero(ar, sizeof(*ar));
275	ar->k_ar.ar_magic = AUDIT_RECORD_MAGIC;
276	nanotime(&ar->k_ar.ar_starttime);
277
278	/*
279	 * Export the subject credential.
280	 */
281	cred = td->td_ucred;
282	cru2x(cred, &ar->k_ar.ar_subj_cred);
283	ar->k_ar.ar_subj_ruid = cred->cr_ruid;
284	ar->k_ar.ar_subj_rgid = cred->cr_rgid;
285	ar->k_ar.ar_subj_egid = cred->cr_groups[0];
286	ar->k_ar.ar_subj_auid = cred->cr_audit.ai_auid;
287	ar->k_ar.ar_subj_asid = cred->cr_audit.ai_asid;
288	ar->k_ar.ar_subj_pid = td->td_proc->p_pid;
289	ar->k_ar.ar_subj_amask = cred->cr_audit.ai_mask;
290	ar->k_ar.ar_subj_term_addr = cred->cr_audit.ai_termid;
291	/*
292	 * If this process is jailed, make sure we capture the name of the
293	 * jail so we can use it to generate a zonename token when we covert
294	 * this record to BSM.
295	 */
296	if (jailed(cred)) {
297		pr = cred->cr_prison;
298		(void) strlcpy(ar->k_ar.ar_jailname, pr->pr_name,
299		    sizeof(ar->k_ar.ar_jailname));
300	} else
301		ar->k_ar.ar_jailname[0] = '\0';
302	return (0);
303}
304
305static void
306audit_record_dtor(void *mem, int size, void *arg)
307{
308	struct kaudit_record *ar;
309
310	KASSERT(sizeof(*ar) == size, ("audit_record_dtor: wrong size"));
311
312	ar = mem;
313	if (ar->k_ar.ar_arg_upath1 != NULL)
314		free(ar->k_ar.ar_arg_upath1, M_AUDITPATH);
315	if (ar->k_ar.ar_arg_upath2 != NULL)
316		free(ar->k_ar.ar_arg_upath2, M_AUDITPATH);
317	if (ar->k_ar.ar_arg_text != NULL)
318		free(ar->k_ar.ar_arg_text, M_AUDITTEXT);
319	if (ar->k_udata != NULL)
320		free(ar->k_udata, M_AUDITDATA);
321	if (ar->k_ar.ar_arg_argv != NULL)
322		free(ar->k_ar.ar_arg_argv, M_AUDITTEXT);
323	if (ar->k_ar.ar_arg_envv != NULL)
324		free(ar->k_ar.ar_arg_envv, M_AUDITTEXT);
325	if (ar->k_ar.ar_arg_groups.gidset != NULL)
326		free(ar->k_ar.ar_arg_groups.gidset, M_AUDITGIDSET);
327}
328
329/*
330 * Initialize the Audit subsystem: configuration state, work queue,
331 * synchronization primitives, worker thread, and trigger device node.  Also
332 * call into the BSM assembly code to initialize it.
333 */
334static void
335audit_init(void)
336{
337
338	audit_trail_enabled = 0;
339	audit_trail_suspended = 0;
340	audit_syscalls_enabled = false;
341	audit_panic_on_write_fail = 0;
342	audit_fail_stop = 0;
343	audit_in_failure = 0;
344	audit_argv = 0;
345	audit_arge = 0;
346
347	audit_fstat.af_filesz = 0;	/* '0' means unset, unbounded. */
348	audit_fstat.af_currsz = 0;
349	audit_nae_mask.am_success = 0;
350	audit_nae_mask.am_failure = 0;
351
352	TAILQ_INIT(&audit_q);
353	audit_q_len = 0;
354	audit_pre_q_len = 0;
355	audit_qctrl.aq_hiwater = AQ_HIWATER;
356	audit_qctrl.aq_lowater = AQ_LOWATER;
357	audit_qctrl.aq_bufsz = AQ_BUFSZ;
358	audit_qctrl.aq_minfree = AU_FS_MINFREE;
359
360	audit_kinfo.ai_termid.at_type = AU_IPv4;
361	audit_kinfo.ai_termid.at_addr[0] = INADDR_ANY;
362
363	mtx_init(&audit_mtx, "audit_mtx", NULL, MTX_DEF);
364	KINFO_LOCK_INIT();
365	cv_init(&audit_worker_cv, "audit_worker_cv");
366	cv_init(&audit_watermark_cv, "audit_watermark_cv");
367	cv_init(&audit_fail_cv, "audit_fail_cv");
368
369	audit_record_zone = uma_zcreate("audit_record",
370	    sizeof(struct kaudit_record), audit_record_ctor,
371	    audit_record_dtor, NULL, NULL, UMA_ALIGN_PTR, 0);
372
373	/* First initialisation of audit_syscalls_enabled. */
374	audit_syscalls_enabled_update();
375
376	/* Initialize the BSM audit subsystem. */
377	kau_init();
378
379	audit_trigger_init();
380
381	/* Register shutdown handler. */
382	EVENTHANDLER_REGISTER(shutdown_pre_sync, audit_shutdown, NULL,
383	    SHUTDOWN_PRI_FIRST);
384
385	/* Start audit worker thread. */
386	audit_worker_init();
387}
388
389SYSINIT(audit_init, SI_SUB_AUDIT, SI_ORDER_FIRST, audit_init, NULL);
390
391/*
392 * Drain the audit queue and close the log at shutdown.  Note that this can
393 * be called both from the system shutdown path and also from audit
394 * configuration syscalls, so 'arg' and 'howto' are ignored.
395 *
396 * XXXRW: In FreeBSD 7.x and 8.x, this fails to wait for the record queue to
397 * drain before returning, which could lead to lost records on shutdown.
398 */
399void
400audit_shutdown(void *arg, int howto)
401{
402
403	audit_rotate_vnode(NULL, NULL);
404}
405
406/*
407 * Return the current thread's audit record, if any.
408 */
409struct kaudit_record *
410currecord(void)
411{
412
413	return (curthread->td_ar);
414}
415
416/*
417 * XXXAUDIT: Shouldn't there be logic here to sleep waiting on available
418 * pre_q space, suspending the system call until there is room?
419 */
420struct kaudit_record *
421audit_new(int event, struct thread *td)
422{
423	struct kaudit_record *ar;
424
425	/*
426	 * Note: the number of outstanding uncommitted audit records is
427	 * limited to the number of concurrent threads servicing system calls
428	 * in the kernel.
429	 */
430	ar = uma_zalloc_arg(audit_record_zone, td, M_WAITOK);
431	ar->k_ar.ar_event = event;
432
433	mtx_lock(&audit_mtx);
434	audit_pre_q_len++;
435	mtx_unlock(&audit_mtx);
436
437	return (ar);
438}
439
440void
441audit_free(struct kaudit_record *ar)
442{
443
444	uma_zfree(audit_record_zone, ar);
445}
446
447void
448audit_commit(struct kaudit_record *ar, int error, int retval)
449{
450	au_event_t event;
451	au_class_t class;
452	au_id_t auid;
453	int sorf;
454	struct au_mask *aumask;
455
456	if (ar == NULL)
457		return;
458
459	ar->k_ar.ar_errno = error;
460	ar->k_ar.ar_retval = retval;
461	nanotime(&ar->k_ar.ar_endtime);
462
463	/*
464	 * Decide whether to commit the audit record by checking the error
465	 * value from the system call and using the appropriate audit mask.
466	 */
467	if (ar->k_ar.ar_subj_auid == AU_DEFAUDITID)
468		aumask = &audit_nae_mask;
469	else
470		aumask = &ar->k_ar.ar_subj_amask;
471
472	if (error)
473		sorf = AU_PRS_FAILURE;
474	else
475		sorf = AU_PRS_SUCCESS;
476
477	/*
478	 * syscalls.master sometimes contains a prototype event number, which
479	 * we will transform into a more specific event number now that we
480	 * have more complete information gathered during the system call.
481	 */
482	switch(ar->k_ar.ar_event) {
483	case AUE_OPEN_RWTC:
484		ar->k_ar.ar_event = audit_flags_and_error_to_openevent(
485		    ar->k_ar.ar_arg_fflags, error);
486		break;
487
488	case AUE_OPENAT_RWTC:
489		ar->k_ar.ar_event = audit_flags_and_error_to_openatevent(
490		    ar->k_ar.ar_arg_fflags, error);
491		break;
492
493	case AUE_SYSCTL:
494		ar->k_ar.ar_event = audit_ctlname_to_sysctlevent(
495		    ar->k_ar.ar_arg_ctlname, ar->k_ar.ar_valid_arg);
496		break;
497
498	case AUE_AUDITON:
499		/* Convert the auditon() command to an event. */
500		ar->k_ar.ar_event = auditon_command_event(ar->k_ar.ar_arg_cmd);
501		break;
502
503	case AUE_MSGSYS:
504		if (ARG_IS_VALID(ar, ARG_SVIPC_WHICH))
505			ar->k_ar.ar_event =
506			    audit_msgsys_to_event(ar->k_ar.ar_arg_svipc_which);
507		break;
508
509	case AUE_SEMSYS:
510		if (ARG_IS_VALID(ar, ARG_SVIPC_WHICH))
511			ar->k_ar.ar_event =
512			    audit_semsys_to_event(ar->k_ar.ar_arg_svipc_which);
513		break;
514
515	case AUE_SHMSYS:
516		if (ARG_IS_VALID(ar, ARG_SVIPC_WHICH))
517			ar->k_ar.ar_event =
518			    audit_shmsys_to_event(ar->k_ar.ar_arg_svipc_which);
519		break;
520	}
521
522	auid = ar->k_ar.ar_subj_auid;
523	event = ar->k_ar.ar_event;
524	class = au_event_class(event);
525
526	ar->k_ar_commit |= AR_COMMIT_KERNEL;
527	if (au_preselect(event, class, aumask, sorf) != 0)
528		ar->k_ar_commit |= AR_PRESELECT_TRAIL;
529	if (audit_pipe_preselect(auid, event, class, sorf,
530	    ar->k_ar_commit & AR_PRESELECT_TRAIL) != 0)
531		ar->k_ar_commit |= AR_PRESELECT_PIPE;
532#ifdef KDTRACE_HOOKS
533	/*
534	 * Expose the audit record to DTrace, both to allow the "commit" probe
535	 * to fire if it's desirable, and also to allow a decision to be made
536	 * about later firing with BSM in the audit worker.
537	 */
538	if (dtaudit_hook_commit != NULL) {
539		if (dtaudit_hook_commit(ar, auid, event, class, sorf) != 0)
540			ar->k_ar_commit |= AR_PRESELECT_DTRACE;
541	}
542#endif
543
544	if ((ar->k_ar_commit & (AR_PRESELECT_TRAIL | AR_PRESELECT_PIPE |
545	    AR_PRESELECT_USER_TRAIL | AR_PRESELECT_USER_PIPE |
546	    AR_PRESELECT_DTRACE)) == 0) {
547		mtx_lock(&audit_mtx);
548		audit_pre_q_len--;
549		mtx_unlock(&audit_mtx);
550		audit_free(ar);
551		return;
552	}
553
554	/*
555	 * Note: it could be that some records initiated while audit was
556	 * enabled should still be committed?
557	 *
558	 * NB: The check here is not for audit_syscalls because any
559	 * DTrace-related obligations have been fulfilled above -- we're just
560	 * down to the trail and pipes now.
561	 */
562	mtx_lock(&audit_mtx);
563	if (audit_trail_suspended || !audit_trail_enabled) {
564		audit_pre_q_len--;
565		mtx_unlock(&audit_mtx);
566		audit_free(ar);
567		return;
568	}
569
570	/*
571	 * Constrain the number of committed audit records based on the
572	 * configurable parameter.
573	 */
574	while (audit_q_len >= audit_qctrl.aq_hiwater)
575		cv_wait(&audit_watermark_cv, &audit_mtx);
576
577	TAILQ_INSERT_TAIL(&audit_q, ar, k_q);
578	audit_q_len++;
579	audit_pre_q_len--;
580	cv_signal(&audit_worker_cv);
581	mtx_unlock(&audit_mtx);
582}
583
584/*
585 * audit_syscall_enter() is called on entry to each system call.  It is
586 * responsible for deciding whether or not to audit the call (preselection),
587 * and if so, allocating a per-thread audit record.  audit_new() will fill in
588 * basic thread/credential properties.
589 *
590 * This function will be entered only if audit_syscalls_enabled was set in the
591 * macro wrapper for this function.  It could be cleared by the time this
592 * function runs, but that is an acceptable race.
593 */
594void
595audit_syscall_enter(unsigned short code, struct thread *td)
596{
597	struct au_mask *aumask;
598#ifdef KDTRACE_HOOKS
599	void *dtaudit_state;
600#endif
601	au_class_t class;
602	au_event_t event;
603	au_id_t auid;
604	int record_needed;
605
606	KASSERT(td->td_ar == NULL, ("audit_syscall_enter: td->td_ar != NULL"));
607	KASSERT((td->td_pflags & TDP_AUDITREC) == 0,
608	    ("audit_syscall_enter: TDP_AUDITREC set"));
609
610	/*
611	 * In FreeBSD, each ABI has its own system call table, and hence
612	 * mapping of system call codes to audit events.  Convert the code to
613	 * an audit event identifier using the process system call table
614	 * reference.  In Darwin, there's only one, so we use the global
615	 * symbol for the system call table.  No audit record is generated
616	 * for bad system calls, as no operation has been performed.
617	 */
618	if (code >= td->td_proc->p_sysent->sv_size)
619		return;
620
621	event = td->td_proc->p_sysent->sv_table[code].sy_auevent;
622	if (event == AUE_NULL)
623		return;
624
625	/*
626	 * Check which audit mask to use; either the kernel non-attributable
627	 * event mask or the process audit mask.
628	 */
629	auid = td->td_ucred->cr_audit.ai_auid;
630	if (auid == AU_DEFAUDITID)
631		aumask = &audit_nae_mask;
632	else
633		aumask = &td->td_ucred->cr_audit.ai_mask;
634
635	/*
636	 * Determine whether trail or pipe preselection would like an audit
637	 * record allocated for this system call.
638	 */
639	class = au_event_class(event);
640	if (au_preselect(event, class, aumask, AU_PRS_BOTH)) {
641		/*
642		 * If we're out of space and need to suspend unprivileged
643		 * processes, do that here rather than trying to allocate
644		 * another audit record.
645		 *
646		 * Note: we might wish to be able to continue here in the
647		 * future, if the system recovers.  That should be possible
648		 * by means of checking the condition in a loop around
649		 * cv_wait().  It might be desirable to reevaluate whether an
650		 * audit record is still required for this event by
651		 * re-calling au_preselect().
652		 */
653		if (audit_in_failure &&
654		    priv_check(td, PRIV_AUDIT_FAILSTOP) != 0) {
655			cv_wait(&audit_fail_cv, &audit_mtx);
656			panic("audit_failing_stop: thread continued");
657		}
658		record_needed = 1;
659	} else if (audit_pipe_preselect(auid, event, class, AU_PRS_BOTH, 0)) {
660		record_needed = 1;
661	} else {
662		record_needed = 0;
663	}
664
665	/*
666	 * After audit trails and pipes have made their policy choices, DTrace
667	 * may request that records be generated as well.  This is a slightly
668	 * complex affair, as the DTrace audit provider needs the audit
669	 * framework to maintain some state on the audit record, which has not
670	 * been allocated at the point where the decision has to be made.
671	 * This hook must run even if we are not changing the decision, as
672	 * DTrace may want to stick event state onto a record we were going to
673	 * produce due to the trail or pipes.  The event state returned by the
674	 * DTrace provider must be safe without locks held between here and
675	 * below -- i.e., dtaudit_state must must refer to stable memory.
676	 */
677#ifdef KDTRACE_HOOKS
678	dtaudit_state = NULL;
679        if (dtaudit_hook_preselect != NULL) {
680		dtaudit_state = dtaudit_hook_preselect(auid, event, class);
681		if (dtaudit_state != NULL)
682			record_needed = 1;
683	}
684#endif
685
686	/*
687	 * If a record is required, allocate it and attach it to the thread
688	 * for use throughout the system call.  Also attach DTrace state if
689	 * required.
690	 *
691	 * XXXRW: If we decide to reference count the evname_elem underlying
692	 * dtaudit_state, we will need to free here if no record is allocated
693	 * or allocatable.
694	 */
695	if (record_needed) {
696		td->td_ar = audit_new(event, td);
697		if (td->td_ar != NULL) {
698			td->td_pflags |= TDP_AUDITREC;
699#ifdef KDTRACE_HOOKS
700			td->td_ar->k_dtaudit_state = dtaudit_state;
701#endif
702		}
703	} else
704		td->td_ar = NULL;
705}
706
707/*
708 * audit_syscall_exit() is called from the return of every system call, or in
709 * the event of exit1(), during the execution of exit1().  It is responsible
710 * for committing the audit record, if any, along with return condition.
711 */
712void
713audit_syscall_exit(int error, struct thread *td)
714{
715	int retval;
716
717	/*
718	 * Commit the audit record as desired; once we pass the record into
719	 * audit_commit(), the memory is owned by the audit subsystem.  The
720	 * return value from the system call is stored on the user thread.
721	 * If there was an error, the return value is set to -1, imitating
722	 * the behavior of the cerror routine.
723	 */
724	if (error)
725		retval = -1;
726	else
727		retval = td->td_retval[0];
728
729	audit_commit(td->td_ar, error, retval);
730	td->td_ar = NULL;
731	td->td_pflags &= ~TDP_AUDITREC;
732}
733
734void
735audit_cred_copy(struct ucred *src, struct ucred *dest)
736{
737
738	bcopy(&src->cr_audit, &dest->cr_audit, sizeof(dest->cr_audit));
739}
740
741void
742audit_cred_destroy(struct ucred *cred)
743{
744
745}
746
747void
748audit_cred_init(struct ucred *cred)
749{
750
751	bzero(&cred->cr_audit, sizeof(cred->cr_audit));
752}
753
754/*
755 * Initialize audit information for the first kernel process (proc 0) and for
756 * the first user process (init).
757 */
758void
759audit_cred_kproc0(struct ucred *cred)
760{
761
762	cred->cr_audit.ai_auid = AU_DEFAUDITID;
763	cred->cr_audit.ai_termid.at_type = AU_IPv4;
764}
765
766void
767audit_cred_proc1(struct ucred *cred)
768{
769
770	cred->cr_audit.ai_auid = AU_DEFAUDITID;
771	cred->cr_audit.ai_termid.at_type = AU_IPv4;
772}
773
774void
775audit_thread_alloc(struct thread *td)
776{
777
778	td->td_ar = NULL;
779}
780
781void
782audit_thread_free(struct thread *td)
783{
784
785	KASSERT(td->td_ar == NULL, ("audit_thread_free: td_ar != NULL"));
786	KASSERT((td->td_pflags & TDP_AUDITREC) == 0,
787	    ("audit_thread_free: TDP_AUDITREC set"));
788}
789
790void
791audit_proc_coredump(struct thread *td, char *path, int errcode)
792{
793	struct kaudit_record *ar;
794	struct au_mask *aumask;
795	struct ucred *cred;
796	au_class_t class;
797	int ret, sorf;
798	char **pathp;
799	au_id_t auid;
800
801	ret = 0;
802
803	/*
804	 * Make sure we are using the correct preselection mask.
805	 */
806	cred = td->td_ucred;
807	auid = cred->cr_audit.ai_auid;
808	if (auid == AU_DEFAUDITID)
809		aumask = &audit_nae_mask;
810	else
811		aumask = &cred->cr_audit.ai_mask;
812	/*
813	 * It's possible for coredump(9) generation to fail.  Make sure that
814	 * we handle this case correctly for preselection.
815	 */
816	if (errcode != 0)
817		sorf = AU_PRS_FAILURE;
818	else
819		sorf = AU_PRS_SUCCESS;
820	class = au_event_class(AUE_CORE);
821	if (au_preselect(AUE_CORE, class, aumask, sorf) == 0 &&
822	    audit_pipe_preselect(auid, AUE_CORE, class, sorf, 0) == 0)
823		return;
824
825	/*
826	 * If we are interested in seeing this audit record, allocate it.
827	 * Where possible coredump records should contain a pathname and arg32
828	 * (signal) tokens.
829	 */
830	ar = audit_new(AUE_CORE, td);
831	if (ar == NULL)
832		return;
833	if (path != NULL) {
834		pathp = &ar->k_ar.ar_arg_upath1;
835		*pathp = malloc(MAXPATHLEN, M_AUDITPATH, M_WAITOK);
836		audit_canon_path(td, AT_FDCWD, path, *pathp);
837		ARG_SET_VALID(ar, ARG_UPATH1);
838	}
839	ar->k_ar.ar_arg_signum = td->td_proc->p_sig;
840	ARG_SET_VALID(ar, ARG_SIGNUM);
841	if (errcode != 0)
842		ret = 1;
843	audit_commit(ar, errcode, ret);
844}
845