1/*-
2 * Copyright (c) 2006 Robert N. M. Watson
3 * Copyright (c) 2008-2009 Apple, Inc.
4 * All rights reserved.
5 *
6 * This software was developed by Robert Watson for the TrustedBSD Project.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 *    notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30#include <stdarg.h>
31#include <sys/param.h>
32#include <sys/systm.h>
33#include <sys/kernel.h>
34#include <sys/fcntl.h>
35#include <sys/conf.h>
36#include <sys/poll.h>
37#include <sys/user.h>
38#include <sys/signalvar.h>
39#include <miscfs/devfs/devfs.h>
40
41#include <bsm/audit.h>
42#include <security/audit/audit.h>
43#include <security/audit/audit_ioctl.h>
44#include <security/audit/audit_bsd.h>
45#include <security/audit/audit_private.h>
46
47#if CONFIG_AUDIT
48/*
49 * Implementation of a clonable special device providing a live stream of BSM
50 * audit data.  Consumers receive a "tee" of the system audit trail by
51 * default, but may also define alternative event selections using ioctls.
52 * This interface provides unreliable but timely access to audit events.
53 * Consumers should be very careful to avoid introducing event cycles.
54 */
55
56/*
57 * Memory types.
58 */
59static MALLOC_DEFINE(M_AUDIT_PIPE, "audit_pipe", "Audit pipes");
60static MALLOC_DEFINE(M_AUDIT_PIPE_ENTRY, "audit_pipeent",
61    "Audit pipe entries and buffers");
62static MALLOC_DEFINE(M_AUDIT_PIPE_PRESELECT, "audit_pipe_presel",
63    "Audit pipe preselection structure");
64
65/*
66 * Audit pipe buffer parameters.
67 */
68#define	AUDIT_PIPE_QLIMIT_DEFAULT	(128)
69#define	AUDIT_PIPE_QLIMIT_MIN		(1)
70#define	AUDIT_PIPE_QLIMIT_MAX		(1024)
71
72/*
73 * Description of an entry in an audit_pipe.
74 */
75struct audit_pipe_entry {
76	void				*ape_record;
77	u_int				 ape_record_len;
78	TAILQ_ENTRY(audit_pipe_entry)	 ape_queue;
79};
80
81/*
82 * Audit pipes allow processes to express "interest" in the set of records
83 * that are delivered via the pipe.  They do this in a similar manner to the
84 * mechanism for audit trail configuration, by expressing two global masks,
85 * and optionally expressing per-auid masks.  The following data structure is
86 * the per-auid mask description.  The global state is stored in the audit
87 * pipe data structure.
88 *
89 * We may want to consider a more space/time-efficient data structure once
90 * usage patterns for per-auid specifications are clear.
91 */
92struct audit_pipe_preselect {
93	au_id_t					 app_auid;
94	au_mask_t				 app_mask;
95	TAILQ_ENTRY(audit_pipe_preselect)	 app_list;
96};
97
98/*
99 * Description of an individual audit_pipe.  Consists largely of a bounded
100 * length queue.
101 */
102#define	AUDIT_PIPE_ASYNC	0x00000001
103#define	AUDIT_PIPE_NBIO		0x00000002
104struct audit_pipe {
105	int				 ap_open;	/* Device open? */
106	u_int				 ap_flags;
107
108	struct selinfo			 ap_selinfo;
109	pid_t				 ap_sigio;
110
111	/*
112	 * Per-pipe mutex protecting most fields in this data structure.
113	 */
114	struct mtx			 ap_mtx;
115
116	/*
117	 * Per-pipe sleep lock serializing user-generated reads and flushes.
118	 * uiomove() is called to copy out the current head record's data
119	 * while the record remains in the queue, so we prevent other threads
120	 * from removing it using this lock.
121	 */
122	struct slck			 ap_sx;
123
124	/*
125	 * Condition variable to signal when data has been delivered to a
126	 * pipe.
127	 */
128	struct cv			 ap_cv;
129
130	/*
131	 * Various queue-related variables:  qlen and qlimit are a count of
132	 * records in the queue; qbyteslen is the number of bytes of data
133	 * across all records, and qoffset is the amount read so far of the
134	 * first record in the queue.  The number of bytes available for
135	 * reading in the queue is qbyteslen - qoffset.
136	 */
137	u_int				 ap_qlen;
138	u_int				 ap_qlimit;
139	u_int				 ap_qbyteslen;
140	u_int				 ap_qoffset;
141
142	/*
143	 * Per-pipe operation statistics.
144	 */
145	u_int64_t			 ap_inserts;	/* Records added. */
146	u_int64_t			 ap_reads;	/* Records read. */
147	u_int64_t			 ap_drops;	/* Records dropped. */
148
149	/*
150	 * Fields relating to pipe interest: global masks for unmatched
151	 * processes (attributable, non-attributable), and a list of specific
152	 * interest specifications by auid.
153	 */
154	int				 ap_preselect_mode;
155	au_mask_t			 ap_preselect_flags;
156	au_mask_t			 ap_preselect_naflags;
157	TAILQ_HEAD(, audit_pipe_preselect)	ap_preselect_list;
158
159	/*
160	 * Current pending record list.  Protected by a combination of ap_mtx
161	 * and ap_sx.  Note particularly that *both* locks are required to
162	 * remove a record from the head of the queue, as an in-progress read
163	 * may sleep while copying and therefore cannot hold ap_mtx.
164	 */
165	TAILQ_HEAD(, audit_pipe_entry)	 ap_queue;
166
167	/*
168	 * Global pipe list.
169	 */
170	TAILQ_ENTRY(audit_pipe)		 ap_list;
171};
172
173#define	AUDIT_PIPE_LOCK(ap)		mtx_lock(&(ap)->ap_mtx)
174#define	AUDIT_PIPE_LOCK_ASSERT(ap)	mtx_assert(&(ap)->ap_mtx, MA_OWNED)
175#define	AUDIT_PIPE_LOCK_DESTROY(ap)	mtx_destroy(&(ap)->ap_mtx)
176#define	AUDIT_PIPE_LOCK_INIT(ap)	mtx_init(&(ap)->ap_mtx, \
177					    "audit_pipe_mtx", NULL, MTX_DEF)
178#define	AUDIT_PIPE_UNLOCK(ap)		mtx_unlock(&(ap)->ap_mtx)
179#define	AUDIT_PIPE_MTX(ap)		(&(ap)->ap_mtx)
180
181#define	AUDIT_PIPE_SX_LOCK_DESTROY(ap)	slck_destroy(&(ap)->ap_sx)
182#define	AUDIT_PIPE_SX_LOCK_INIT(ap)	slck_init(&(ap)->ap_sx, "audit_pipe_sx")
183#define	AUDIT_PIPE_SX_XLOCK_ASSERT(ap)	slck_assert(&(ap)->ap_sx, SA_XLOCKED)
184#define	AUDIT_PIPE_SX_XLOCK_SIG(ap)	slck_lock_sig(&(ap)->ap_sx)
185#define	AUDIT_PIPE_SX_XUNLOCK(ap)	slck_unlock(&(ap)->ap_sx)
186
187
188/*
189 * Global list of audit pipes, rwlock to protect it.  Individual record
190 * queues on pipes are protected by per-pipe locks; these locks synchronize
191 * between threads walking the list to deliver to individual pipes and add/
192 * remove of pipes, and are mostly acquired for read.
193 */
194static TAILQ_HEAD(, audit_pipe)	 audit_pipe_list;
195static struct rwlock		 audit_pipe_lock;
196
197#define	AUDIT_PIPE_LIST_LOCK_INIT()	rw_init(&audit_pipe_lock, \
198					    "audit_pipe_list_lock")
199#define	AUDIT_PIPE_LIST_RLOCK()  	rw_rlock(&audit_pipe_lock)
200#define	AUDIT_PIPE_LIST_RUNLOCK()	rw_runlock(&audit_pipe_lock)
201#define	AUDIT_PIPE_LIST_WLOCK()		rw_wlock(&audit_pipe_lock)
202#define	AUDIT_PIPE_LIST_WLOCK_ASSERT()	rw_assert(&audit_pipe_lock, \
203					    RA_WLOCKED)
204#define	AUDIT_PIPE_LIST_WUNLOCK()	rw_wunlock(&audit_pipe_lock)
205
206/*
207 * Cloning related variables and constants.
208 */
209#define	AUDIT_PIPE_NAME		"auditpipe"
210#define	MAX_AUDIT_PIPES		32
211static int audit_pipe_major;
212
213/*
214 * dev_t doesn't have a pointer for "softc" data.  So we have to keep track of
215 * it with the following global array (indexed by the minor number).
216 *
217 * XXX We may want to dynamically grow this as needed.
218 */
219static struct audit_pipe	*audit_pipe_dtab[MAX_AUDIT_PIPES];
220
221
222/*
223 * Special device methods and definition.
224 */
225static open_close_fcn_t		audit_pipe_open;
226static open_close_fcn_t		audit_pipe_close;
227static read_write_fcn_t		audit_pipe_read;
228static ioctl_fcn_t		audit_pipe_ioctl;
229static select_fcn_t		audit_pipe_poll;
230
231static struct cdevsw audit_pipe_cdevsw = {
232	.d_open      =          audit_pipe_open,
233	.d_close     =          audit_pipe_close,
234	.d_read      =          audit_pipe_read,
235	.d_write     =          eno_rdwrt,
236	.d_ioctl     =          audit_pipe_ioctl,
237	.d_stop      =          eno_stop,
238	.d_reset     =          eno_reset,
239	.d_ttys      =          NULL,
240	.d_select    =          audit_pipe_poll,
241	.d_mmap      =          eno_mmap,
242	.d_strategy  =          eno_strat,
243	.d_type      =          0
244};
245
246/*
247 * Some global statistics on audit pipes.
248 */
249static int		audit_pipe_count;	/* Current number of pipes. */
250static u_int64_t	audit_pipe_ever;	/* Pipes ever allocated. */
251static u_int64_t	audit_pipe_records;	/* Records seen. */
252static u_int64_t	audit_pipe_drops;	/* Global record drop count. */
253
254/*
255 * Free an audit pipe entry.
256 */
257static void
258audit_pipe_entry_free(struct audit_pipe_entry *ape)
259{
260
261	free(ape->ape_record, M_AUDIT_PIPE_ENTRY);
262	free(ape, M_AUDIT_PIPE_ENTRY);
263}
264
265/*
266 * Find an audit pipe preselection specification for an auid, if any.
267 */
268static struct audit_pipe_preselect *
269audit_pipe_preselect_find(struct audit_pipe *ap, au_id_t auid)
270{
271	struct audit_pipe_preselect *app;
272
273	AUDIT_PIPE_LOCK_ASSERT(ap);
274
275	TAILQ_FOREACH(app, &ap->ap_preselect_list, app_list) {
276		if (app->app_auid == auid)
277			return (app);
278	}
279	return (NULL);
280}
281
282/*
283 * Query the per-pipe mask for a specific auid.
284 */
285static int
286audit_pipe_preselect_get(struct audit_pipe *ap, au_id_t auid,
287    au_mask_t *maskp)
288{
289	struct audit_pipe_preselect *app;
290	int error;
291
292	AUDIT_PIPE_LOCK(ap);
293	app = audit_pipe_preselect_find(ap, auid);
294	if (app != NULL) {
295		*maskp = app->app_mask;
296		error = 0;
297	} else
298		error = ENOENT;
299	AUDIT_PIPE_UNLOCK(ap);
300	return (error);
301}
302
303/*
304 * Set the per-pipe mask for a specific auid.  Add a new entry if needed;
305 * otherwise, update the current entry.
306 */
307static void
308audit_pipe_preselect_set(struct audit_pipe *ap, au_id_t auid, au_mask_t mask)
309{
310	struct audit_pipe_preselect *app, *app_new;
311
312	/*
313	 * Pessimistically assume that the auid doesn't already have a mask
314	 * set, and allocate.  We will free it if it is unneeded.
315	 */
316	app_new = malloc(sizeof(*app_new), M_AUDIT_PIPE_PRESELECT, M_WAITOK);
317	AUDIT_PIPE_LOCK(ap);
318	app = audit_pipe_preselect_find(ap, auid);
319	if (app == NULL) {
320		app = app_new;
321		app_new = NULL;
322		app->app_auid = auid;
323		TAILQ_INSERT_TAIL(&ap->ap_preselect_list, app, app_list);
324	}
325	app->app_mask = mask;
326	AUDIT_PIPE_UNLOCK(ap);
327	if (app_new != NULL)
328		free(app_new, M_AUDIT_PIPE_PRESELECT);
329}
330
331/*
332 * Delete a per-auid mask on an audit pipe.
333 */
334static int
335audit_pipe_preselect_delete(struct audit_pipe *ap, au_id_t auid)
336{
337	struct audit_pipe_preselect *app;
338	int error;
339
340	AUDIT_PIPE_LOCK(ap);
341	app = audit_pipe_preselect_find(ap, auid);
342	if (app != NULL) {
343		TAILQ_REMOVE(&ap->ap_preselect_list, app, app_list);
344		error = 0;
345	} else
346		error = ENOENT;
347	AUDIT_PIPE_UNLOCK(ap);
348	if (app != NULL)
349		free(app, M_AUDIT_PIPE_PRESELECT);
350	return (error);
351}
352
353/*
354 * Delete all per-auid masks on an audit pipe.
355 */
356static void
357audit_pipe_preselect_flush_locked(struct audit_pipe *ap)
358{
359	struct audit_pipe_preselect *app;
360
361	AUDIT_PIPE_LOCK_ASSERT(ap);
362
363	while ((app = TAILQ_FIRST(&ap->ap_preselect_list)) != NULL) {
364		TAILQ_REMOVE(&ap->ap_preselect_list, app, app_list);
365		free(app, M_AUDIT_PIPE_PRESELECT);
366	}
367}
368
369static void
370audit_pipe_preselect_flush(struct audit_pipe *ap)
371{
372
373	AUDIT_PIPE_LOCK(ap);
374	audit_pipe_preselect_flush_locked(ap);
375	AUDIT_PIPE_UNLOCK(ap);
376}
377
378/*-
379 * Determine whether a specific audit pipe matches a record with these
380 * properties.  Algorithm is as follows:
381 *
382 * - If the pipe is configured to track the default trail configuration, then
383 *   use the results of global preselection matching.
384 * - If not, search for a specifically configured auid entry matching the
385 *   event.  If an entry is found, use that.
386 * - Otherwise, use the default flags or naflags configured for the pipe.
387 */
388static int
389audit_pipe_preselect_check(struct audit_pipe *ap, au_id_t auid,
390    au_event_t event, au_class_t class, int sorf, int trail_preselect)
391{
392	struct audit_pipe_preselect *app;
393
394	AUDIT_PIPE_LOCK_ASSERT(ap);
395
396	switch (ap->ap_preselect_mode) {
397	case AUDITPIPE_PRESELECT_MODE_TRAIL:
398		return (trail_preselect);
399
400	case AUDITPIPE_PRESELECT_MODE_LOCAL:
401		app = audit_pipe_preselect_find(ap, auid);
402		if (app == NULL) {
403			if (auid == (uid_t)AU_DEFAUDITID)
404				return (au_preselect(event, class,
405				    &ap->ap_preselect_naflags, sorf));
406			else
407				return (au_preselect(event, class,
408				    &ap->ap_preselect_flags, sorf));
409		} else
410			return (au_preselect(event, class, &app->app_mask,
411			    sorf));
412
413	default:
414		panic("audit_pipe_preselect_check: mode %d",
415		    ap->ap_preselect_mode);
416	}
417
418	return (0);
419}
420
421/*
422 * Determine whether there exists a pipe interested in a record with specific
423 * properties.
424 */
425int
426audit_pipe_preselect(au_id_t auid, au_event_t event, au_class_t class,
427    int sorf, int trail_preselect)
428{
429	struct audit_pipe *ap;
430
431	/* Lockless read to avoid acquiring the global lock if not needed. */
432	if (TAILQ_EMPTY(&audit_pipe_list))
433		return (0);
434
435	AUDIT_PIPE_LIST_RLOCK();
436	TAILQ_FOREACH(ap, &audit_pipe_list, ap_list) {
437		 AUDIT_PIPE_LOCK(ap);
438		if (audit_pipe_preselect_check(ap, auid, event, class, sorf,
439		    trail_preselect)) {
440			AUDIT_PIPE_UNLOCK(ap);
441			AUDIT_PIPE_LIST_RUNLOCK();
442			return (1);
443		}
444		AUDIT_PIPE_UNLOCK(ap);
445	}
446	AUDIT_PIPE_LIST_RUNLOCK();
447	return (0);
448}
449
450/*
451 * Append individual record to a queue -- allocate queue-local buffer, and
452 * add to the queue.  If the queue is full or we can't allocate memory, drop
453 * the newest record.
454 */
455static void
456audit_pipe_append(struct audit_pipe *ap, void *record, u_int record_len)
457{
458	struct audit_pipe_entry *ape;
459
460	AUDIT_PIPE_LOCK_ASSERT(ap);
461
462	if (ap->ap_qlen >= ap->ap_qlimit) {
463		ap->ap_drops++;
464		audit_pipe_drops++;
465		return;
466	}
467
468	ape = malloc(sizeof(*ape), M_AUDIT_PIPE_ENTRY, M_NOWAIT | M_ZERO);
469	if (ape == NULL) {
470		ap->ap_drops++;
471		audit_pipe_drops++;
472		return;
473	}
474
475	ape->ape_record = malloc(record_len, M_AUDIT_PIPE_ENTRY, M_NOWAIT);
476	if (ape->ape_record == NULL) {
477		free(ape, M_AUDIT_PIPE_ENTRY);
478		ap->ap_drops++;
479		audit_pipe_drops++;
480		return;
481	}
482
483	bcopy(record, ape->ape_record, record_len);
484	ape->ape_record_len = record_len;
485
486	TAILQ_INSERT_TAIL(&ap->ap_queue, ape, ape_queue);
487	ap->ap_inserts++;
488	ap->ap_qlen++;
489	ap->ap_qbyteslen += ape->ape_record_len;
490	selwakeup(&ap->ap_selinfo);
491	if (ap->ap_flags & AUDIT_PIPE_ASYNC)
492		pgsigio(ap->ap_sigio, SIGIO);
493#if 0  /* XXX - fix select */
494	selwakeuppri(&ap->ap_selinfo, PSOCK);
495	KNOTE_LOCKED(&ap->ap_selinfo.si_note, 0);
496	if (ap->ap_flags & AUDIT_PIPE_ASYNC)
497		pgsigio(&ap->ap_sigio, SIGIO, 0);
498#endif
499	cv_broadcast(&ap->ap_cv);
500}
501
502/*
503 * audit_pipe_submit(): audit_worker submits audit records via this
504 * interface, which arranges for them to be delivered to pipe queues.
505 */
506void
507audit_pipe_submit(au_id_t auid, au_event_t event, au_class_t class, int sorf,
508    int trail_select, void *record, u_int record_len)
509{
510	struct audit_pipe *ap;
511
512	/*
513	 * Lockless read to avoid lock overhead if pipes are not in use.
514	 */
515	if (TAILQ_FIRST(&audit_pipe_list) == NULL)
516		return;
517
518	AUDIT_PIPE_LIST_RLOCK();
519	TAILQ_FOREACH(ap, &audit_pipe_list, ap_list) {
520		AUDIT_PIPE_LOCK(ap);
521		if (audit_pipe_preselect_check(ap, auid, event, class, sorf,
522		    trail_select))
523			audit_pipe_append(ap, record, record_len);
524		AUDIT_PIPE_UNLOCK(ap);
525	}
526	AUDIT_PIPE_LIST_RUNLOCK();
527
528	/* Unlocked increment. */
529	audit_pipe_records++;
530}
531
532/*
533 * audit_pipe_submit_user(): the same as audit_pipe_submit(), except that
534 * since we don't currently have selection information available, it is
535 * delivered to the pipe unconditionally.
536 *
537 * XXXRW: This is a bug.  The BSM check routine for submitting a user record
538 * should parse that information and return it.
539 */
540void
541audit_pipe_submit_user(void *record, u_int record_len)
542{
543	struct audit_pipe *ap;
544
545	/*
546	 * Lockless read to avoid lock overhead if pipes are not in use.
547	 */
548	if (TAILQ_FIRST(&audit_pipe_list) == NULL)
549		return;
550
551	AUDIT_PIPE_LIST_RLOCK();
552	TAILQ_FOREACH(ap, &audit_pipe_list, ap_list) {
553		AUDIT_PIPE_LOCK(ap);
554		audit_pipe_append(ap, record, record_len);
555		AUDIT_PIPE_UNLOCK(ap);
556	}
557	AUDIT_PIPE_LIST_RUNLOCK();
558
559	/* Unlocked increment. */
560	audit_pipe_records++;
561}
562
563/*
564 * Allocate a new audit pipe.  Connects the pipe, on success, to the global
565 * list and updates statistics.
566 */
567static struct audit_pipe *
568audit_pipe_alloc(void)
569{
570	struct audit_pipe *ap;
571
572	AUDIT_PIPE_LIST_WLOCK_ASSERT();
573
574	ap = malloc(sizeof(*ap), M_AUDIT_PIPE, M_WAITOK | M_ZERO);
575	if (ap == NULL)
576		return (NULL);
577
578	ap->ap_qlimit = AUDIT_PIPE_QLIMIT_DEFAULT;
579	TAILQ_INIT(&ap->ap_queue);
580#ifndef  __APPLE__
581	knlist_init(&ap->ap_selinfo.si_note, AUDIT_PIPE_MTX(ap), NULL, NULL,
582	    NULL);
583#endif
584	AUDIT_PIPE_LOCK_INIT(ap);
585	AUDIT_PIPE_SX_LOCK_INIT(ap);
586	cv_init(&ap->ap_cv, "audit_pipe");
587
588	/*
589	 * Default flags, naflags, and auid-specific preselection settings to
590	 * 0.  Initialize the mode to the global trail so that if praudit(1)
591	 * is run on /dev/auditpipe, it sees events associated with the
592	 * default trail.  Pipe-aware application can clear the flag, set
593	 * custom masks, and flush the pipe as needed.
594	 */
595	bzero(&ap->ap_preselect_flags, sizeof(ap->ap_preselect_flags));
596	bzero(&ap->ap_preselect_naflags, sizeof(ap->ap_preselect_naflags));
597	TAILQ_INIT(&ap->ap_preselect_list);
598	ap->ap_preselect_mode = AUDITPIPE_PRESELECT_MODE_TRAIL;
599
600	/*
601	 * Add to global list and update global statistics.
602	 */
603	TAILQ_INSERT_HEAD(&audit_pipe_list, ap, ap_list);
604	audit_pipe_count++;
605	audit_pipe_ever++;
606
607	return (ap);
608}
609
610/*
611 * Flush all records currently present in an audit pipe; assume mutex is held.
612 */
613static void
614audit_pipe_flush(struct audit_pipe *ap)
615{
616	struct audit_pipe_entry *ape;
617
618	AUDIT_PIPE_LOCK_ASSERT(ap);
619
620	while ((ape = TAILQ_FIRST(&ap->ap_queue)) != NULL) {
621		TAILQ_REMOVE(&ap->ap_queue, ape, ape_queue);
622		ap->ap_qbyteslen -= ape->ape_record_len;
623		audit_pipe_entry_free(ape);
624		ap->ap_qlen--;
625	}
626	ap->ap_qoffset = 0;
627
628	KASSERT(ap->ap_qlen == 0, ("audit_pipe_free: ap_qbyteslen"));
629	KASSERT(ap->ap_qbyteslen == 0, ("audit_pipe_flush: ap_qbyteslen"));
630}
631
632/*
633 * Free an audit pipe; this means freeing all preselection state and all
634 * records in the pipe.  Assumes global write lock and pipe mutex are held to
635 * revent any new records from being inserted during the free, and that the
636 * audit pipe is still on the global list.
637 */
638static void
639audit_pipe_free(struct audit_pipe *ap)
640{
641
642	AUDIT_PIPE_LIST_WLOCK_ASSERT();
643	AUDIT_PIPE_LOCK_ASSERT(ap);
644
645	audit_pipe_preselect_flush_locked(ap);
646	audit_pipe_flush(ap);
647	cv_destroy(&ap->ap_cv);
648	AUDIT_PIPE_SX_LOCK_DESTROY(ap);
649	AUDIT_PIPE_LOCK_DESTROY(ap);
650#ifndef  __APPLE__
651	knlist_destroy(&ap->ap_selinfo.si_note);
652#endif
653	TAILQ_REMOVE(&audit_pipe_list, ap, ap_list);
654	free(ap, M_AUDIT_PIPE);
655	audit_pipe_count--;
656}
657
658/*
659 * Audit pipe clone routine -- provides a new minor number, or to return (-1),
660 * if one can't be provided.  Called with DEVFS_LOCK held.
661 */
662static int
663audit_pipe_clone(__unused dev_t dev, int action)
664{
665	int i;
666
667	if (action == DEVFS_CLONE_ALLOC) {
668		for(i = 0; i < MAX_AUDIT_PIPES; i++)
669			if (audit_pipe_dtab[i] == NULL)
670				return (i);
671
672		/*
673		 * XXX Should really return -1 here but that seems to hang
674		 * things in devfs.  Instead return 0 and let _open() tell
675		 * userland the bad news.
676		 */
677		return (0);
678	}
679
680	return (-1);
681}
682
683/*
684 * Audit pipe open method.  Explicit privilege check isn't used as this
685 * allows file permissions on the special device to be used to grant audit
686 * review access.  Those file permissions should be managed carefully.
687 */
688static int
689audit_pipe_open(dev_t dev, __unused int flags,  __unused int devtype,
690    __unused proc_t p)
691{
692	struct audit_pipe *ap;
693	int u;
694
695	u = minor(dev);
696	if (u < 0 || u > MAX_AUDIT_PIPES)
697		return (ENXIO);
698
699	AUDIT_PIPE_LIST_WLOCK();
700	ap = audit_pipe_dtab[u];
701	if (ap == NULL) {
702		ap = audit_pipe_alloc();
703		if (ap == NULL) {
704			AUDIT_PIPE_LIST_WUNLOCK();
705			return (ENOMEM);
706		}
707		audit_pipe_dtab[u] = ap;
708	} else {
709		KASSERT(ap->ap_open, ("audit_pipe_open: ap && !ap_open"));
710		AUDIT_PIPE_LIST_WUNLOCK();
711		return (EBUSY);
712	}
713	ap->ap_open = 1;
714	AUDIT_PIPE_LIST_WUNLOCK();
715#ifndef __APPLE__
716	fsetown(td->td_proc->p_pid, &ap->ap_sigio);
717#endif
718	return (0);
719}
720
721/*
722 * Close audit pipe, tear down all records, etc.
723 */
724static int
725audit_pipe_close(dev_t dev, __unused int flags, __unused int devtype,
726    __unused proc_t p)
727{
728	struct audit_pipe *ap;
729	int u;
730
731	u = minor(dev);
732	ap = audit_pipe_dtab[u];
733	KASSERT(ap != NULL, ("audit_pipe_close: ap == NULL"));
734	KASSERT(ap->ap_open, ("audit_pipe_close: !ap_open"));
735
736#ifndef __APPLE__
737	funsetown(&ap->ap_sigio);
738#endif
739	AUDIT_PIPE_LIST_WLOCK();
740	AUDIT_PIPE_LOCK(ap);
741	ap->ap_open = 0;
742	audit_pipe_free(ap);
743	audit_pipe_dtab[u] = NULL;
744	AUDIT_PIPE_LIST_WUNLOCK();
745	return (0);
746}
747
748/*
749 * Audit pipe ioctl() routine.  Handle file descriptor and audit pipe layer
750 * commands.
751 */
752static int
753audit_pipe_ioctl(dev_t dev, u_long cmd, caddr_t data,
754    __unused int flag, __unused proc_t p)
755{
756	struct auditpipe_ioctl_preselect *aip;
757	struct audit_pipe *ap;
758	au_mask_t *maskp;
759	int error, mode;
760	au_id_t auid;
761
762	ap = audit_pipe_dtab[minor(dev)];
763	KASSERT(ap != NULL, ("audit_pipe_ioctl: ap == NULL"));
764
765	/*
766	 * Audit pipe ioctls: first come standard device node ioctls, then
767	 * manipulation of pipe settings, and finally, statistics query
768	 * ioctls.
769	 */
770	switch (cmd) {
771	case FIONBIO:
772		AUDIT_PIPE_LOCK(ap);
773		if (*(int *)data)
774			ap->ap_flags |= AUDIT_PIPE_NBIO;
775		else
776			ap->ap_flags &= ~AUDIT_PIPE_NBIO;
777		AUDIT_PIPE_UNLOCK(ap);
778		error = 0;
779		break;
780
781	case FIONREAD:
782		AUDIT_PIPE_LOCK(ap);
783		*(int *)data = ap->ap_qbyteslen - ap->ap_qoffset;
784		AUDIT_PIPE_UNLOCK(ap);
785		error = 0;
786		break;
787
788	case FIOASYNC:
789		AUDIT_PIPE_LOCK(ap);
790		if (*(int *)data)
791			ap->ap_flags |= AUDIT_PIPE_ASYNC;
792		else
793			ap->ap_flags &= ~AUDIT_PIPE_ASYNC;
794		AUDIT_PIPE_UNLOCK(ap);
795		error = 0;
796		break;
797
798#ifndef __APPLE__
799	case FIOSETOWN:
800		error = fsetown(*(int *)data, &ap->ap_sigio);
801		break;
802
803	case FIOGETOWN:
804		*(int *)data = fgetown(&ap->ap_sigio);
805		error = 0;
806		break;
807#endif /* !__APPLE__ */
808
809	case AUDITPIPE_GET_QLEN:
810		*(u_int *)data = ap->ap_qlen;
811		error = 0;
812		break;
813
814	case AUDITPIPE_GET_QLIMIT:
815		*(u_int *)data = ap->ap_qlimit;
816		error = 0;
817		break;
818
819	case AUDITPIPE_SET_QLIMIT:
820		/* Lockless integer write. */
821		if (*(u_int *)data >= AUDIT_PIPE_QLIMIT_MIN ||
822		    *(u_int *)data <= AUDIT_PIPE_QLIMIT_MAX) {
823			ap->ap_qlimit = *(u_int *)data;
824			error = 0;
825		} else
826			error = EINVAL;
827		break;
828
829	case AUDITPIPE_GET_QLIMIT_MIN:
830		*(u_int *)data = AUDIT_PIPE_QLIMIT_MIN;
831		error = 0;
832		break;
833
834	case AUDITPIPE_GET_QLIMIT_MAX:
835		*(u_int *)data = AUDIT_PIPE_QLIMIT_MAX;
836		error = 0;
837		break;
838
839	case AUDITPIPE_GET_PRESELECT_FLAGS:
840		AUDIT_PIPE_LOCK(ap);
841		maskp = (au_mask_t *)data;
842		*maskp = ap->ap_preselect_flags;
843		AUDIT_PIPE_UNLOCK(ap);
844		error = 0;
845		break;
846
847	case AUDITPIPE_SET_PRESELECT_FLAGS:
848		AUDIT_PIPE_LOCK(ap);
849		maskp = (au_mask_t *)data;
850		ap->ap_preselect_flags = *maskp;
851		AUDIT_CHECK_IF_KEVENTS_MASK(ap->ap_preselect_flags);
852		AUDIT_PIPE_UNLOCK(ap);
853		error = 0;
854		break;
855
856	case AUDITPIPE_GET_PRESELECT_NAFLAGS:
857		AUDIT_PIPE_LOCK(ap);
858		maskp = (au_mask_t *)data;
859		*maskp = ap->ap_preselect_naflags;
860		AUDIT_PIPE_UNLOCK(ap);
861		error = 0;
862		break;
863
864	case AUDITPIPE_SET_PRESELECT_NAFLAGS:
865		AUDIT_PIPE_LOCK(ap);
866		maskp = (au_mask_t *)data;
867		ap->ap_preselect_naflags = *maskp;
868		AUDIT_CHECK_IF_KEVENTS_MASK(ap->ap_preselect_naflags);
869		AUDIT_PIPE_UNLOCK(ap);
870		error = 0;
871		break;
872
873	case AUDITPIPE_GET_PRESELECT_AUID:
874		aip = (struct auditpipe_ioctl_preselect *)data;
875		error = audit_pipe_preselect_get(ap, aip->aip_auid,
876		    &aip->aip_mask);
877		break;
878
879	case AUDITPIPE_SET_PRESELECT_AUID:
880		aip = (struct auditpipe_ioctl_preselect *)data;
881		audit_pipe_preselect_set(ap, aip->aip_auid, aip->aip_mask);
882		error = 0;
883		break;
884
885	case AUDITPIPE_DELETE_PRESELECT_AUID:
886		auid = *(au_id_t *)data;
887		error = audit_pipe_preselect_delete(ap, auid);
888		break;
889
890	case AUDITPIPE_FLUSH_PRESELECT_AUID:
891		audit_pipe_preselect_flush(ap);
892		error = 0;
893		break;
894
895	case AUDITPIPE_GET_PRESELECT_MODE:
896		AUDIT_PIPE_LOCK(ap);
897		*(int *)data = ap->ap_preselect_mode;
898		AUDIT_PIPE_UNLOCK(ap);
899		error = 0;
900		break;
901
902	case AUDITPIPE_SET_PRESELECT_MODE:
903		mode = *(int *)data;
904		switch (mode) {
905		case AUDITPIPE_PRESELECT_MODE_TRAIL:
906		case AUDITPIPE_PRESELECT_MODE_LOCAL:
907			AUDIT_PIPE_LOCK(ap);
908			ap->ap_preselect_mode = mode;
909			AUDIT_PIPE_UNLOCK(ap);
910			error = 0;
911			break;
912
913		default:
914			error = EINVAL;
915		}
916		break;
917
918	case AUDITPIPE_FLUSH:
919		if (AUDIT_PIPE_SX_XLOCK_SIG(ap) != 0)
920			return (EINTR);
921		AUDIT_PIPE_LOCK(ap);
922		audit_pipe_flush(ap);
923		AUDIT_PIPE_UNLOCK(ap);
924		AUDIT_PIPE_SX_XUNLOCK(ap);
925		error = 0;
926		break;
927
928	case AUDITPIPE_GET_MAXAUDITDATA:
929		*(u_int *)data = MAXAUDITDATA;
930		error = 0;
931		break;
932
933	case AUDITPIPE_GET_INSERTS:
934		*(u_int *)data = ap->ap_inserts;
935		error = 0;
936		break;
937
938	case AUDITPIPE_GET_READS:
939		*(u_int *)data = ap->ap_reads;
940		error = 0;
941		break;
942
943	case AUDITPIPE_GET_DROPS:
944		*(u_int *)data = ap->ap_drops;
945		error = 0;
946		break;
947
948	case AUDITPIPE_GET_TRUNCATES:
949		*(u_int *)data = 0;
950		error = 0;
951		break;
952
953	default:
954		error = ENOTTY;
955	}
956	return (error);
957}
958
959/*
960 * Audit pipe read.  Read one or more partial or complete records to user
961 * memory.
962 */
963static int
964audit_pipe_read(dev_t dev, struct uio *uio, __unused int flag)
965{
966	struct audit_pipe_entry *ape;
967	struct audit_pipe *ap;
968	u_int toread;
969	int error;
970
971	ap = audit_pipe_dtab[minor(dev)];
972	KASSERT(ap != NULL, ("audit_pipe_read: ap == NULL"));
973
974	/*
975	 * We hold an sleep lock over read and flush because we rely on the
976	 * stability of a record in the queue during uiomove(9).
977	 */
978	if (AUDIT_PIPE_SX_XLOCK_SIG(ap) != 0)
979		return (EINTR);
980	AUDIT_PIPE_LOCK(ap);
981	while (TAILQ_EMPTY(&ap->ap_queue)) {
982		if (ap->ap_flags & AUDIT_PIPE_NBIO) {
983			AUDIT_PIPE_UNLOCK(ap);
984			AUDIT_PIPE_SX_XUNLOCK(ap);
985			return (EAGAIN);
986		}
987		error = cv_wait_sig(&ap->ap_cv, AUDIT_PIPE_MTX(ap));
988		if (error) {
989			AUDIT_PIPE_UNLOCK(ap);
990			AUDIT_PIPE_SX_XUNLOCK(ap);
991			return (error);
992		}
993	}
994
995	/*
996	 * Copy as many remaining bytes from the current record to userspace
997	 * as we can.  Keep processing records until we run out of records in
998	 * the queue, or until the user buffer runs out of space.
999	 *
1000	 * Note: we rely on the sleep lock to maintain ape's stability here.
1001	 */
1002	ap->ap_reads++;
1003	while ((ape = TAILQ_FIRST(&ap->ap_queue)) != NULL &&
1004	    uio_resid(uio) > 0) {
1005		AUDIT_PIPE_LOCK_ASSERT(ap);
1006
1007		KASSERT(ape->ape_record_len > ap->ap_qoffset,
1008		    ("audit_pipe_read: record_len > qoffset (1)"));
1009		toread = MIN(ape->ape_record_len - ap->ap_qoffset,
1010		    uio_resid(uio));
1011		AUDIT_PIPE_UNLOCK(ap);
1012		error = uiomove((char *)ape->ape_record + ap->ap_qoffset,
1013		    toread, uio);
1014		if (error) {
1015			AUDIT_PIPE_SX_XUNLOCK(ap);
1016			return (error);
1017		}
1018
1019		/*
1020		 * If the copy succeeded, update book-keeping, and if no
1021		 * bytes remain in the current record, free it.
1022		 */
1023		AUDIT_PIPE_LOCK(ap);
1024		KASSERT(TAILQ_FIRST(&ap->ap_queue) == ape,
1025		    ("audit_pipe_read: queue out of sync after uiomove"));
1026		ap->ap_qoffset += toread;
1027		KASSERT(ape->ape_record_len >= ap->ap_qoffset,
1028		    ("audit_pipe_read: record_len >= qoffset (2)"));
1029		if (ap->ap_qoffset == ape->ape_record_len) {
1030			TAILQ_REMOVE(&ap->ap_queue, ape, ape_queue);
1031			ap->ap_qbyteslen -= ape->ape_record_len;
1032			audit_pipe_entry_free(ape);
1033			ap->ap_qlen--;
1034			ap->ap_qoffset = 0;
1035		}
1036	}
1037	AUDIT_PIPE_UNLOCK(ap);
1038	AUDIT_PIPE_SX_XUNLOCK(ap);
1039	return (0);
1040}
1041
1042/*
1043 * Audit pipe poll.
1044 */
1045static int
1046audit_pipe_poll(dev_t dev, int events, void *wql, struct proc *p)
1047{
1048	struct audit_pipe *ap;
1049	int revents;
1050
1051	revents = 0;
1052	ap = audit_pipe_dtab[minor(dev)];
1053	KASSERT(ap != NULL, ("audit_pipe_poll: ap == NULL"));
1054
1055	if (events & (POLLIN | POLLRDNORM)) {
1056		AUDIT_PIPE_LOCK(ap);
1057		if (TAILQ_FIRST(&ap->ap_queue) != NULL)
1058			revents |= events & (POLLIN | POLLRDNORM);
1059		else
1060			selrecord(p, &ap->ap_selinfo, wql);
1061		AUDIT_PIPE_UNLOCK(ap);
1062	}
1063	return (revents);
1064}
1065
1066#ifndef __APPLE__
1067/*
1068 * Return true if there are records available for reading on the pipe.
1069 */
1070static int
1071audit_pipe_kqread(struct knote *kn, long hint)
1072{
1073	struct audit_pipe *ap;
1074
1075	ap = (struct audit_pipe *)kn->kn_hook;
1076	KASSERT(ap != NULL, ("audit_pipe_kqread: ap == NULL"));
1077	AUDIT_PIPE_LOCK_ASSERT(ap);
1078
1079	if (ap->ap_qlen != 0) {
1080		kn->kn_data = ap->ap_qbyteslen - ap->ap_qoffset;
1081		return (1);
1082	} else {
1083		kn->kn_data = 0;
1084		return (0);
1085	}
1086}
1087
1088/*
1089 * Detach kqueue state from audit pipe.
1090 */
1091static void
1092audit_pipe_kqdetach(struct knote *kn)
1093{
1094	struct audit_pipe *ap;
1095
1096	ap = (struct audit_pipe *)kn->kn_hook;
1097	KASSERT(ap != NULL, ("audit_pipe_kqdetach: ap == NULL"));
1098
1099	AUDIT_PIPE_LOCK(ap);
1100	knlist_remove(&ap->ap_selinfo.si_note, kn, 1);
1101	AUDIT_PIPE_UNLOCK(ap);
1102}
1103#endif /* !__APPLE__ */
1104
1105static void *devnode;
1106
1107int
1108audit_pipe_init(void)
1109{
1110	dev_t dev;
1111
1112	TAILQ_INIT(&audit_pipe_list);
1113	AUDIT_PIPE_LIST_LOCK_INIT();
1114
1115	audit_pipe_major = cdevsw_add(-1, &audit_pipe_cdevsw);
1116	if (audit_pipe_major < 0)
1117	  return (KERN_FAILURE);
1118
1119	dev = makedev(audit_pipe_major, 0);
1120	devnode = devfs_make_node_clone(dev, DEVFS_CHAR, UID_ROOT, GID_WHEEL,
1121	    0600, audit_pipe_clone, "auditpipe", 0);
1122
1123	if (devnode == NULL)
1124		return (KERN_FAILURE);
1125
1126	return (KERN_SUCCESS);
1127}
1128
1129int
1130audit_pipe_shutdown(void)
1131{
1132
1133	/* unwind everything */
1134	devfs_remove(devnode);
1135	(void) cdevsw_remove(audit_pipe_major, &audit_pipe_cdevsw);
1136
1137	return (KERN_SUCCESS);
1138}
1139
1140#endif /* CONFIG_AUDIT */
1141