audit.c revision 159275 
1/*
2 * Copyright (c) 1999-2005 Apple Computer, Inc.
3 * Copyright (c) 2006 Robert N. M. Watson
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1.  Redistributions of source code must retain the above copyright
10 *     notice, this list of conditions and the following disclaimer.
11 * 2.  Redistributions in binary form must reproduce the above copyright
12 *     notice, this list of conditions and the following disclaimer in the
13 *     documentation and/or other materials provided with the distribution.
14 * 3.  Neither the name of Apple Computer, Inc. ("Apple") nor the names of
15 *     its contributors may be used to endorse or promote products derived
16 *     from this software without specific prior written permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY APPLE AND ITS CONTRIBUTORS "AS IS" AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL APPLE OR ITS CONTRIBUTORS BE LIABLE FOR
22 * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
26 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
27 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
28 * POSSIBILITY OF SUCH DAMAGE.
29 *
30 * $FreeBSD: head/sys/security/audit/audit.c 159275 2006-06-05 15:38:12Z rwatson $
31 */
32
33#include <sys/param.h>
34#include <sys/condvar.h>
35#include <sys/conf.h>
36#include <sys/file.h>
37#include <sys/filedesc.h>
38#include <sys/fcntl.h>
39#include <sys/ipc.h>
40#include <sys/kernel.h>
41#include <sys/kthread.h>
42#include <sys/malloc.h>
43#include <sys/mount.h>
44#include <sys/namei.h>
45#include <sys/proc.h>
46#include <sys/queue.h>
47#include <sys/socket.h>
48#include <sys/socketvar.h>
49#include <sys/protosw.h>
50#include <sys/domain.h>
51#include <sys/sysproto.h>
52#include <sys/sysent.h>
53#include <sys/systm.h>
54#include <sys/ucred.h>
55#include <sys/uio.h>
56#include <sys/un.h>
57#include <sys/unistd.h>
58#include <sys/vnode.h>
59
60#include <bsm/audit.h>
61#include <bsm/audit_internal.h>
62#include <bsm/audit_kevents.h>
63
64#include <netinet/in.h>
65#include <netinet/in_pcb.h>
66
67#include <security/audit/audit.h>
68#include <security/audit/audit_private.h>
69
70#include <vm/uma.h>
71
72static uma_zone_t	audit_record_zone;
73static MALLOC_DEFINE(M_AUDITPROC, "audit_proc", "Audit process storage");
74MALLOC_DEFINE(M_AUDITDATA, "audit_data", "Audit data storage");
75MALLOC_DEFINE(M_AUDITPATH, "audit_path", "Audit path storage");
76MALLOC_DEFINE(M_AUDITTEXT, "audit_text", "Audit text storage");
77
78/*
79 * Audit control settings that are set/read by system calls and are
80 * hence non-static.
81 */
82/*
83 * Define the audit control flags.
84 */
85int			audit_enabled;
86int			audit_suspended;
87
88/*
89 * Flags controlling behavior in low storage situations.
90 * Should we panic if a write fails?  Should we fail stop
91 * if we're out of disk space?
92 */
93int			audit_panic_on_write_fail;
94int			audit_fail_stop;
95
96/*
97 * Are we currently "failing stop" due to out of disk space?
98 */
99int			audit_in_failure;
100
101/*
102 * Global audit statistiscs.
103 */
104struct audit_fstat	audit_fstat;
105
106/*
107 * Preselection mask for non-attributable events.
108 */
109struct au_mask		audit_nae_mask;
110
111/*
112 * Mutex to protect global variables shared between various threads and
113 * processes.
114 */
115struct mtx		audit_mtx;
116
117/*
118 * Queue of audit records ready for delivery to disk.  We insert new
119 * records at the tail, and remove records from the head.  Also,
120 * a count of the number of records used for checking queue depth.
121 * In addition, a counter of records that we have allocated but are
122 * not yet in the queue, which is needed to estimate the total
123 * size of the combined set of records outstanding in the system.
124 */
125struct kaudit_queue	audit_q;
126int			audit_q_len;
127int			audit_pre_q_len;
128
129/*
130 * Audit queue control settings (minimum free, low/high water marks, etc.)
131 */
132struct au_qctrl		audit_qctrl;
133
134/*
135 * Condition variable to signal to the worker that it has work to do:
136 * either new records are in the queue, or a log replacement is taking
137 * place.
138 */
139struct cv		audit_worker_cv;
140
141/*
142 * Condition variable to flag when crossing the low watermark, meaning that
143 * threads blocked due to hitting the high watermark can wake up and continue
144 * to commit records.
145 */
146struct cv		audit_watermark_cv;
147
148/*
149 * Condition variable for  auditing threads wait on when in fail-stop mode.
150 * Threads wait on this CV forever (and ever), never seeing the light of
151 * day again.
152 */
153static struct cv	audit_fail_cv;
154
155/*
156 * Construct an audit record for the passed thread.
157 */
158static int
159audit_record_ctor(void *mem, int size, void *arg, int flags)
160{
161	struct kaudit_record *ar;
162	struct thread *td;
163
164	KASSERT(sizeof(*ar) == size, ("audit_record_ctor: wrong size"));
165
166	td = arg;
167	ar = mem;
168	bzero(ar, sizeof(*ar));
169	ar->k_ar.ar_magic = AUDIT_RECORD_MAGIC;
170	nanotime(&ar->k_ar.ar_starttime);
171
172	/*
173	 * Export the subject credential.
174	 *
175	 * XXXAUDIT: td_ucred access is OK without proc lock, but some other
176	 * fields here may require the proc lock.
177	 */
178	cru2x(td->td_ucred, &ar->k_ar.ar_subj_cred);
179	ar->k_ar.ar_subj_ruid = td->td_ucred->cr_ruid;
180	ar->k_ar.ar_subj_rgid = td->td_ucred->cr_rgid;
181	ar->k_ar.ar_subj_egid = td->td_ucred->cr_groups[0];
182	ar->k_ar.ar_subj_auid = td->td_proc->p_au->ai_auid;
183	ar->k_ar.ar_subj_asid = td->td_proc->p_au->ai_asid;
184	ar->k_ar.ar_subj_pid = td->td_proc->p_pid;
185	ar->k_ar.ar_subj_amask = td->td_proc->p_au->ai_mask;
186	ar->k_ar.ar_subj_term = td->td_proc->p_au->ai_termid;
187	bcopy(td->td_proc->p_comm, ar->k_ar.ar_subj_comm, MAXCOMLEN);
188
189	return (0);
190}
191
192static void
193audit_record_dtor(void *mem, int size, void *arg)
194{
195	struct kaudit_record *ar;
196
197	KASSERT(sizeof(*ar) == size, ("audit_record_dtor: wrong size"));
198
199	ar = mem;
200	if (ar->k_ar.ar_arg_upath1 != NULL)
201		free(ar->k_ar.ar_arg_upath1, M_AUDITPATH);
202	if (ar->k_ar.ar_arg_upath2 != NULL)
203		free(ar->k_ar.ar_arg_upath2, M_AUDITPATH);
204	if (ar->k_ar.ar_arg_text != NULL)
205		free(ar->k_ar.ar_arg_text, M_AUDITTEXT);
206	if (ar->k_udata != NULL)
207		free(ar->k_udata, M_AUDITDATA);
208}
209
210/*
211 * Initialize the Audit subsystem: configuration state, work queue,
212 * synchronization primitives, worker thread, and trigger device node.  Also
213 * call into the BSM assembly code to initialize it.
214 */
215static void
216audit_init(void)
217{
218
219	printf("Security auditing service present\n");
220	audit_enabled = 0;
221	audit_suspended = 0;
222	audit_panic_on_write_fail = 0;
223	audit_fail_stop = 0;
224	audit_in_failure = 0;
225
226	audit_fstat.af_filesz = 0;	/* '0' means unset, unbounded */
227	audit_fstat.af_currsz = 0;
228	audit_nae_mask.am_success = AU_NULL;
229	audit_nae_mask.am_failure = AU_NULL;
230
231	TAILQ_INIT(&audit_q);
232	audit_q_len = 0;
233	audit_pre_q_len = 0;
234	audit_qctrl.aq_hiwater = AQ_HIWATER;
235	audit_qctrl.aq_lowater = AQ_LOWATER;
236	audit_qctrl.aq_bufsz = AQ_BUFSZ;
237	audit_qctrl.aq_minfree = AU_FS_MINFREE;
238
239	mtx_init(&audit_mtx, "audit_mtx", NULL, MTX_DEF);
240	cv_init(&audit_worker_cv, "audit_worker_cv");
241	cv_init(&audit_watermark_cv, "audit_watermark_cv");
242	cv_init(&audit_fail_cv, "audit_fail_cv");
243
244	audit_record_zone = uma_zcreate("audit_record",
245	    sizeof(struct kaudit_record), audit_record_ctor,
246	    audit_record_dtor, NULL, NULL, UMA_ALIGN_PTR, 0);
247
248	/* Initialize the BSM audit subsystem. */
249	kau_init();
250
251	audit_trigger_init();
252
253	/* Register shutdown handler. */
254	EVENTHANDLER_REGISTER(shutdown_pre_sync, audit_shutdown, NULL,
255	    SHUTDOWN_PRI_FIRST);
256
257	/* Start audit worker thread. */
258	audit_worker_init();
259}
260
261SYSINIT(audit_init, SI_SUB_AUDIT, SI_ORDER_FIRST, audit_init, NULL)
262
263/*
264 * Drain the audit queue and close the log at shutdown.  Note that this can
265 * be called both from the system shutdown path and also from audit
266 * configuration syscalls, so 'arg' and 'howto' are ignored.
267 */
268void
269audit_shutdown(void *arg, int howto)
270{
271
272	audit_rotate_vnode(NULL, NULL);
273}
274
275/*
276 * Return the current thread's audit record, if any.
277 */
278__inline__ struct kaudit_record *
279currecord(void)
280{
281
282	return (curthread->td_ar);
283}
284
285/*
286 * MPSAFE
287 *
288 * XXXAUDIT: There are a number of races present in the code below due to
289 * release and re-grab of the mutex.  The code should be revised to become
290 * slightly less racy.
291 *
292 * XXXAUDIT: Shouldn't there be logic here to sleep waiting on available
293 * pre_q space, suspending the system call until there is room?
294 */
295struct kaudit_record *
296audit_new(int event, struct thread *td)
297{
298	struct kaudit_record *ar;
299	int no_record;
300
301	mtx_lock(&audit_mtx);
302	no_record = (audit_suspended || !audit_enabled);
303	mtx_unlock(&audit_mtx);
304	if (no_record)
305		return (NULL);
306
307	/*
308	 * XXX: The number of outstanding uncommitted audit records is
309	 * limited to the number of concurrent threads servicing system
310	 * calls in the kernel.
311	 */
312	ar = uma_zalloc_arg(audit_record_zone, td, M_WAITOK);
313	ar->k_ar.ar_event = event;
314
315	mtx_lock(&audit_mtx);
316	audit_pre_q_len++;
317	mtx_unlock(&audit_mtx);
318
319	return (ar);
320}
321
322void
323audit_free(struct kaudit_record *ar)
324{
325
326	uma_zfree(audit_record_zone, ar);
327}
328
329/*
330 * MPSAFE
331 */
332void
333audit_commit(struct kaudit_record *ar, int error, int retval)
334{
335	au_event_t event;
336	au_class_t class;
337	au_id_t auid;
338	int sorf;
339	struct au_mask *aumask;
340
341	if (ar == NULL)
342		return;
343
344	/*
345	 * Decide whether to commit the audit record by checking the
346	 * error value from the system call and using the appropriate
347	 * audit mask.
348	 *
349	 * XXXAUDIT: Synchronize access to audit_nae_mask?
350	 */
351	if (ar->k_ar.ar_subj_auid == AU_DEFAUDITID)
352		aumask = &audit_nae_mask;
353	else
354		aumask = &ar->k_ar.ar_subj_amask;
355
356	if (error)
357		sorf = AU_PRS_FAILURE;
358	else
359		sorf = AU_PRS_SUCCESS;
360
361	switch(ar->k_ar.ar_event) {
362
363	case AUE_OPEN_RWTC:
364		/* The open syscall always writes a AUE_OPEN_RWTC event; change
365		 * it to the proper type of event based on the flags and the
366		 * error value.
367		 */
368		ar->k_ar.ar_event = flags_and_error_to_openevent(
369		    ar->k_ar.ar_arg_fflags, error);
370		break;
371
372	case AUE_SYSCTL:
373		ar->k_ar.ar_event = ctlname_to_sysctlevent(
374		    ar->k_ar.ar_arg_ctlname, ar->k_ar.ar_valid_arg);
375		break;
376
377	case AUE_AUDITON:
378		/* Convert the auditon() command to an event */
379		ar->k_ar.ar_event = auditon_command_event(ar->k_ar.ar_arg_cmd);
380		break;
381	}
382
383	auid = ar->k_ar.ar_subj_auid;
384	event = ar->k_ar.ar_event;
385	class = au_event_class(event);
386
387	ar->k_ar_commit |= AR_COMMIT_KERNEL;
388	if (au_preselect(event, class, aumask, sorf) != 0)
389		ar->k_ar_commit |= AR_PRESELECT_TRAIL;
390	if (audit_pipe_preselect(auid, event, class, sorf,
391	    ar->k_ar_commit & AR_PRESELECT_TRAIL) != 0)
392		ar->k_ar_commit |= AR_PRESELECT_PIPE;
393	if ((ar->k_ar_commit & (AR_PRESELECT_TRAIL | AR_PRESELECT_PIPE)) ==
394	    0) {
395		mtx_lock(&audit_mtx);
396		audit_pre_q_len--;
397		mtx_unlock(&audit_mtx);
398		audit_free(ar);
399		return;
400	}
401
402	ar->k_ar.ar_errno = error;
403	ar->k_ar.ar_retval = retval;
404
405	/*
406	 * We might want to do some system-wide post-filtering
407	 * here at some point.
408	 */
409
410	/*
411	 * Timestamp system call end.
412	 */
413	nanotime(&ar->k_ar.ar_endtime);
414
415	mtx_lock(&audit_mtx);
416
417	/*
418	 * Note: it could be that some records initiated while audit was
419	 * enabled should still be committed?
420	 */
421	if (audit_suspended || !audit_enabled) {
422		audit_pre_q_len--;
423		mtx_unlock(&audit_mtx);
424		audit_free(ar);
425		return;
426	}
427
428	/*
429	 * Constrain the number of committed audit records based on
430	 * the configurable parameter.
431	 */
432	while (audit_q_len >= audit_qctrl.aq_hiwater) {
433		AUDIT_PRINTF(("audit_commit: sleeping to wait for "
434		   "audit queue to drain below high water mark\n"));
435		cv_wait(&audit_watermark_cv, &audit_mtx);
436		AUDIT_PRINTF(("audit_commit: woke up waiting for "
437		   "audit queue draining\n"));
438	}
439
440	TAILQ_INSERT_TAIL(&audit_q, ar, k_q);
441	audit_q_len++;
442	audit_pre_q_len--;
443	cv_signal(&audit_worker_cv);
444	mtx_unlock(&audit_mtx);
445}
446
447/*
448 * audit_syscall_enter() is called on entry to each system call.  It is
449 * responsible for deciding whether or not to audit the call (preselection),
450 * and if so, allocating a per-thread audit record.  audit_new() will fill in
451 * basic thread/credential properties.
452 */
453void
454audit_syscall_enter(unsigned short code, struct thread *td)
455{
456	struct au_mask *aumask;
457	au_class_t class;
458	au_event_t event;
459	au_id_t auid;
460
461	KASSERT(td->td_ar == NULL, ("audit_syscall_enter: td->td_ar != NULL"));
462
463	/*
464	 * In FreeBSD, each ABI has its own system call table, and hence
465	 * mapping of system call codes to audit events.  Convert the code to
466	 * an audit event identifier using the process system call table
467	 * reference.  In Darwin, there's only one, so we use the global
468	 * symbol for the system call table.
469	 *
470	 * XXXAUDIT: Should we audit that a bad system call was made, and if
471	 * so, how?
472	 */
473	if (code >= td->td_proc->p_sysent->sv_size)
474		return;
475
476	event = td->td_proc->p_sysent->sv_table[code].sy_auevent;
477	if (event == AUE_NULL)
478		return;
479
480	/*
481	 * Check which audit mask to use; either the kernel non-attributable
482	 * event mask or the process audit mask.
483	 */
484	auid = td->td_proc->p_au->ai_auid;
485	if (auid == AU_DEFAUDITID)
486		aumask = &audit_nae_mask;
487	else
488		aumask = &td->td_proc->p_au->ai_mask;
489
490	/*
491	 * Allocate an audit record, if preselection allows it, and store
492	 * in the thread for later use.
493	 */
494	class = au_event_class(event);
495	if (au_preselect(event, class, aumask, AU_PRS_BOTH)) {
496		/*
497		 * If we're out of space and need to suspend unprivileged
498		 * processes, do that here rather than trying to allocate
499		 * another audit record.
500		 *
501		 * XXXRW: We might wish to be able to continue here in the
502		 * future, if the system recovers.  That should be possible
503		 * by means of checking the condition in a loop around
504		 * cv_wait().  It might be desirable to reevaluate whether an
505		 * audit record is still required for this event by
506		 * re-calling au_preselect().
507		 */
508		if (audit_in_failure && suser(td) != 0) {
509			cv_wait(&audit_fail_cv, &audit_mtx);
510			panic("audit_failing_stop: thread continued");
511		}
512		td->td_ar = audit_new(event, td);
513	} else if (audit_pipe_preselect(auid, event, class, AU_PRS_BOTH, 0))
514		td->td_ar = audit_new(event, td);
515	else
516		td->td_ar = NULL;
517}
518
519/*
520 * audit_syscall_exit() is called from the return of every system call, or in
521 * the event of exit1(), during the execution of exit1().  It is responsible
522 * for committing the audit record, if any, along with return condition.
523 */
524void
525audit_syscall_exit(int error, struct thread *td)
526{
527	int retval;
528
529	/*
530	 * Commit the audit record as desired; once we pass the record
531	 * into audit_commit(), the memory is owned by the audit
532	 * subsystem.
533	 * The return value from the system call is stored on the user
534	 * thread. If there was an error, the return value is set to -1,
535	 * imitating the behavior of the cerror routine.
536	 */
537	if (error)
538		retval = -1;
539	else
540		retval = td->td_retval[0];
541
542	audit_commit(td->td_ar, error, retval);
543	if (td->td_ar != NULL)
544		AUDIT_PRINTF(("audit record committed by pid %d\n",
545			td->td_proc->p_pid));
546	td->td_ar = NULL;
547
548}
549
550/*
551 * Allocate storage for a new process (init, or otherwise).
552 */
553void
554audit_proc_alloc(struct proc *p)
555{
556
557	KASSERT(p->p_au == NULL, ("audit_proc_alloc: p->p_au != NULL (%d)",
558	    p->p_pid));
559	p->p_au = malloc(sizeof(*(p->p_au)), M_AUDITPROC, M_WAITOK);
560	/* XXXAUDIT: Zero?  Slab allocate? */
561	//printf("audit_proc_alloc: pid %d p_au %p\n", p->p_pid, p->p_au);
562}
563
564/*
565 * Allocate storage for a new thread.
566 */
567void
568audit_thread_alloc(struct thread *td)
569{
570
571	td->td_ar = NULL;
572}
573
574/*
575 * Thread destruction.
576 */
577void
578audit_thread_free(struct thread *td)
579{
580
581	KASSERT(td->td_ar == NULL, ("audit_thread_free: td_ar != NULL"));
582}
583
584/*
585 * Initialize the audit information for the a process, presumably the first
586 * process in the system.
587 * XXX It is not clear what the initial values should be for audit ID,
588 * session ID, etc.
589 */
590void
591audit_proc_kproc0(struct proc *p)
592{
593
594	KASSERT(p->p_au != NULL, ("audit_proc_kproc0: p->p_au == NULL (%d)",
595	    p->p_pid));
596	//printf("audit_proc_kproc0: pid %d p_au %p\n", p->p_pid, p->p_au);
597	bzero(p->p_au, sizeof(*(p)->p_au));
598}
599
600void
601audit_proc_init(struct proc *p)
602{
603
604	KASSERT(p->p_au != NULL, ("audit_proc_init: p->p_au == NULL (%d)",
605	    p->p_pid));
606	//printf("audit_proc_init: pid %d p_au %p\n", p->p_pid, p->p_au);
607	bzero(p->p_au, sizeof(*(p)->p_au));
608	p->p_au->ai_auid = AU_DEFAUDITID;
609}
610
611/*
612 * Copy the audit info from the parent process to the child process when
613 * a fork takes place.
614 */
615void
616audit_proc_fork(struct proc *parent, struct proc *child)
617{
618
619	PROC_LOCK_ASSERT(parent, MA_OWNED);
620	PROC_LOCK_ASSERT(child, MA_OWNED);
621	KASSERT(parent->p_au != NULL,
622	    ("audit_proc_fork: parent->p_au == NULL (%d)", parent->p_pid));
623	KASSERT(child->p_au != NULL,
624	    ("audit_proc_fork: child->p_au == NULL (%d)", child->p_pid));
625	//printf("audit_proc_fork: parent pid %d p_au %p\n", parent->p_pid,
626	//    parent->p_au);
627	//printf("audit_proc_fork: child pid %d p_au %p\n", child->p_pid,
628	//    child->p_au);
629	bcopy(parent->p_au, child->p_au, sizeof(*child->p_au));
630	/*
631	 * XXXAUDIT: Zero pointers to external memory, or assert they are
632	 * zero?
633	 */
634}
635
636/*
637 * Free the auditing structure for the process.
638 */
639void
640audit_proc_free(struct proc *p)
641{
642
643	KASSERT(p->p_au != NULL, ("p->p_au == NULL (%d)", p->p_pid));
644	//printf("audit_proc_free: pid %d p_au %p\n", p->p_pid, p->p_au);
645	/*
646	 * XXXAUDIT: Assert that external memory pointers are NULL?
647	 */
648	free(p->p_au, M_AUDITPROC);
649	p->p_au = NULL;
650}
651