xref: /openbmc/linux/kernel/audit.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
1 /* audit.c -- Auditing support
2  * Gateway between the kernel (e.g., selinux) and the user-space audit daemon.
3  * System-call specific features have moved to auditsc.c
4  *
5  * Copyright 2003-2004 Red Hat Inc., Durham, North Carolina.
6  * All Rights Reserved.
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; either version 2 of the License, or
11  * (at your option) any later version.
12  *
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17  *
18  * You should have received a copy of the GNU General Public License
19  * along with this program; if not, write to the Free Software
20  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
21  *
22  * Written by Rickard E. (Rik) Faith <faith@redhat.com>
23  *
24  * Goals: 1) Integrate fully with SELinux.
25  *	  2) Minimal run-time overhead:
26  *	     a) Minimal when syscall auditing is disabled (audit_enable=0).
27  *	     b) Small when syscall auditing is enabled and no audit record
28  *		is generated (defer as much work as possible to record
29  *		generation time):
30  *		i) context is allocated,
31  *		ii) names from getname are stored without a copy, and
32  *		iii) inode information stored from path_lookup.
33  *	  3) Ability to disable syscall auditing at boot time (audit=0).
34  *	  4) Usable by other parts of the kernel (if audit_log* is called,
35  *	     then a syscall record will be generated automatically for the
36  *	     current syscall).
37  *	  5) Netlink interface to user-space.
38  *	  6) Support low-overhead kernel-based filtering to minimize the
39  *	     information that must be passed to user-space.
40  *
41  * Example user-space utilities: http://people.redhat.com/sgrubb/audit/
42  */
43 
44 #include <linux/init.h>
45 #include <asm/atomic.h>
46 #include <asm/types.h>
47 #include <linux/mm.h>
48 #include <linux/module.h>
49 #include <linux/err.h>
50 #include <linux/kthread.h>
51 
52 #include <linux/audit.h>
53 
54 #include <net/sock.h>
55 #include <linux/skbuff.h>
56 #include <linux/netlink.h>
57 
58 /* No auditing will take place until audit_initialized != 0.
59  * (Initialization happens after skb_init is called.) */
60 static int	audit_initialized;
61 
62 /* No syscall auditing will take place unless audit_enabled != 0. */
63 int		audit_enabled;
64 
65 /* Default state when kernel boots without any parameters. */
66 static int	audit_default;
67 
68 /* If auditing cannot proceed, audit_failure selects what happens. */
69 static int	audit_failure = AUDIT_FAIL_PRINTK;
70 
71 /* If audit records are to be written to the netlink socket, audit_pid
72  * contains the (non-zero) pid. */
73 int		audit_pid;
74 
75 /* If audit_limit is non-zero, limit the rate of sending audit records
76  * to that number per second.  This prevents DoS attacks, but results in
77  * audit records being dropped. */
78 static int	audit_rate_limit;
79 
80 /* Number of outstanding audit_buffers allowed. */
81 static int	audit_backlog_limit = 64;
82 static int	audit_backlog_wait_time = 60 * HZ;
83 static int	audit_backlog_wait_overflow = 0;
84 
85 /* The identity of the user shutting down the audit system. */
86 uid_t		audit_sig_uid = -1;
87 pid_t		audit_sig_pid = -1;
88 
89 /* Records can be lost in several ways:
90    0) [suppressed in audit_alloc]
91    1) out of memory in audit_log_start [kmalloc of struct audit_buffer]
92    2) out of memory in audit_log_move [alloc_skb]
93    3) suppressed due to audit_rate_limit
94    4) suppressed due to audit_backlog_limit
95 */
96 static atomic_t    audit_lost = ATOMIC_INIT(0);
97 
98 /* The netlink socket. */
99 static struct sock *audit_sock;
100 
101 /* The audit_freelist is a list of pre-allocated audit buffers (if more
102  * than AUDIT_MAXFREE are in use, the audit buffer is freed instead of
103  * being placed on the freelist). */
104 static DEFINE_SPINLOCK(audit_freelist_lock);
105 static int	   audit_freelist_count = 0;
106 static LIST_HEAD(audit_freelist);
107 
108 static struct sk_buff_head audit_skb_queue;
109 static struct task_struct *kauditd_task;
110 static DECLARE_WAIT_QUEUE_HEAD(kauditd_wait);
111 static DECLARE_WAIT_QUEUE_HEAD(audit_backlog_wait);
112 
113 /* The netlink socket is only to be read by 1 CPU, which lets us assume
114  * that list additions and deletions never happen simultaneously in
115  * auditsc.c */
116 DECLARE_MUTEX(audit_netlink_sem);
117 
118 /* AUDIT_BUFSIZ is the size of the temporary buffer used for formatting
119  * audit records.  Since printk uses a 1024 byte buffer, this buffer
120  * should be at least that large. */
121 #define AUDIT_BUFSIZ 1024
122 
123 /* AUDIT_MAXFREE is the number of empty audit_buffers we keep on the
124  * audit_freelist.  Doing so eliminates many kmalloc/kfree calls. */
125 #define AUDIT_MAXFREE  (2*NR_CPUS)
126 
127 /* The audit_buffer is used when formatting an audit record.  The caller
128  * locks briefly to get the record off the freelist or to allocate the
129  * buffer, and locks briefly to send the buffer to the netlink layer or
130  * to place it on a transmit queue.  Multiple audit_buffers can be in
131  * use simultaneously. */
132 struct audit_buffer {
133 	struct list_head     list;
134 	struct sk_buff       *skb;	/* formatted skb ready to send */
135 	struct audit_context *ctx;	/* NULL or associated context */
136 	gfp_t		     gfp_mask;
137 };
138 
139 static void audit_set_pid(struct audit_buffer *ab, pid_t pid)
140 {
141 	struct nlmsghdr *nlh = (struct nlmsghdr *)ab->skb->data;
142 	nlh->nlmsg_pid = pid;
143 }
144 
145 static void audit_panic(const char *message)
146 {
147 	switch (audit_failure)
148 	{
149 	case AUDIT_FAIL_SILENT:
150 		break;
151 	case AUDIT_FAIL_PRINTK:
152 		printk(KERN_ERR "audit: %s\n", message);
153 		break;
154 	case AUDIT_FAIL_PANIC:
155 		panic("audit: %s\n", message);
156 		break;
157 	}
158 }
159 
160 static inline int audit_rate_check(void)
161 {
162 	static unsigned long	last_check = 0;
163 	static int		messages   = 0;
164 	static DEFINE_SPINLOCK(lock);
165 	unsigned long		flags;
166 	unsigned long		now;
167 	unsigned long		elapsed;
168 	int			retval	   = 0;
169 
170 	if (!audit_rate_limit) return 1;
171 
172 	spin_lock_irqsave(&lock, flags);
173 	if (++messages < audit_rate_limit) {
174 		retval = 1;
175 	} else {
176 		now     = jiffies;
177 		elapsed = now - last_check;
178 		if (elapsed > HZ) {
179 			last_check = now;
180 			messages   = 0;
181 			retval     = 1;
182 		}
183 	}
184 	spin_unlock_irqrestore(&lock, flags);
185 
186 	return retval;
187 }
188 
189 /* Emit at least 1 message per second, even if audit_rate_check is
190  * throttling. */
191 void audit_log_lost(const char *message)
192 {
193 	static unsigned long	last_msg = 0;
194 	static DEFINE_SPINLOCK(lock);
195 	unsigned long		flags;
196 	unsigned long		now;
197 	int			print;
198 
199 	atomic_inc(&audit_lost);
200 
201 	print = (audit_failure == AUDIT_FAIL_PANIC || !audit_rate_limit);
202 
203 	if (!print) {
204 		spin_lock_irqsave(&lock, flags);
205 		now = jiffies;
206 		if (now - last_msg > HZ) {
207 			print = 1;
208 			last_msg = now;
209 		}
210 		spin_unlock_irqrestore(&lock, flags);
211 	}
212 
213 	if (print) {
214 		printk(KERN_WARNING
215 		       "audit: audit_lost=%d audit_rate_limit=%d audit_backlog_limit=%d\n",
216 		       atomic_read(&audit_lost),
217 		       audit_rate_limit,
218 		       audit_backlog_limit);
219 		audit_panic(message);
220 	}
221 
222 }
223 
224 static int audit_set_rate_limit(int limit, uid_t loginuid)
225 {
226 	int old		 = audit_rate_limit;
227 	audit_rate_limit = limit;
228 	audit_log(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE,
229 			"audit_rate_limit=%d old=%d by auid=%u",
230 			audit_rate_limit, old, loginuid);
231 	return old;
232 }
233 
234 static int audit_set_backlog_limit(int limit, uid_t loginuid)
235 {
236 	int old		 = audit_backlog_limit;
237 	audit_backlog_limit = limit;
238 	audit_log(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE,
239 			"audit_backlog_limit=%d old=%d by auid=%u",
240 			audit_backlog_limit, old, loginuid);
241 	return old;
242 }
243 
244 static int audit_set_enabled(int state, uid_t loginuid)
245 {
246 	int old		 = audit_enabled;
247 	if (state != 0 && state != 1)
248 		return -EINVAL;
249 	audit_enabled = state;
250 	audit_log(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE,
251 			"audit_enabled=%d old=%d by auid=%u",
252 			audit_enabled, old, loginuid);
253 	return old;
254 }
255 
256 static int audit_set_failure(int state, uid_t loginuid)
257 {
258 	int old		 = audit_failure;
259 	if (state != AUDIT_FAIL_SILENT
260 	    && state != AUDIT_FAIL_PRINTK
261 	    && state != AUDIT_FAIL_PANIC)
262 		return -EINVAL;
263 	audit_failure = state;
264 	audit_log(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE,
265 			"audit_failure=%d old=%d by auid=%u",
266 			audit_failure, old, loginuid);
267 	return old;
268 }
269 
270 int kauditd_thread(void *dummy)
271 {
272 	struct sk_buff *skb;
273 
274 	while (1) {
275 		skb = skb_dequeue(&audit_skb_queue);
276 		wake_up(&audit_backlog_wait);
277 		if (skb) {
278 			if (audit_pid) {
279 				int err = netlink_unicast(audit_sock, skb, audit_pid, 0);
280 				if (err < 0) {
281 					BUG_ON(err != -ECONNREFUSED); /* Shoudn't happen */
282 					printk(KERN_ERR "audit: *NO* daemon at audit_pid=%d\n", audit_pid);
283 					audit_pid = 0;
284 				}
285 			} else {
286 				printk(KERN_NOTICE "%s\n", skb->data + NLMSG_SPACE(0));
287 				kfree_skb(skb);
288 			}
289 		} else {
290 			DECLARE_WAITQUEUE(wait, current);
291 			set_current_state(TASK_INTERRUPTIBLE);
292 			add_wait_queue(&kauditd_wait, &wait);
293 
294 			if (!skb_queue_len(&audit_skb_queue))
295 				schedule();
296 
297 			__set_current_state(TASK_RUNNING);
298 			remove_wait_queue(&kauditd_wait, &wait);
299 		}
300 	}
301 }
302 
303 void audit_send_reply(int pid, int seq, int type, int done, int multi,
304 		      void *payload, int size)
305 {
306 	struct sk_buff	*skb;
307 	struct nlmsghdr	*nlh;
308 	int		len = NLMSG_SPACE(size);
309 	void		*data;
310 	int		flags = multi ? NLM_F_MULTI : 0;
311 	int		t     = done  ? NLMSG_DONE  : type;
312 
313 	skb = alloc_skb(len, GFP_KERNEL);
314 	if (!skb)
315 		return;
316 
317 	nlh		 = NLMSG_PUT(skb, pid, seq, t, size);
318 	nlh->nlmsg_flags = flags;
319 	data		 = NLMSG_DATA(nlh);
320 	memcpy(data, payload, size);
321 
322 	/* Ignore failure. It'll only happen if the sender goes away,
323 	   because our timeout is set to infinite. */
324 	netlink_unicast(audit_sock, skb, pid, 0);
325 	return;
326 
327 nlmsg_failure:			/* Used by NLMSG_PUT */
328 	if (skb)
329 		kfree_skb(skb);
330 }
331 
332 /*
333  * Check for appropriate CAP_AUDIT_ capabilities on incoming audit
334  * control messages.
335  */
336 static int audit_netlink_ok(kernel_cap_t eff_cap, u16 msg_type)
337 {
338 	int err = 0;
339 
340 	switch (msg_type) {
341 	case AUDIT_GET:
342 	case AUDIT_LIST:
343 	case AUDIT_SET:
344 	case AUDIT_ADD:
345 	case AUDIT_DEL:
346 	case AUDIT_SIGNAL_INFO:
347 		if (!cap_raised(eff_cap, CAP_AUDIT_CONTROL))
348 			err = -EPERM;
349 		break;
350 	case AUDIT_USER:
351 	case AUDIT_FIRST_USER_MSG...AUDIT_LAST_USER_MSG:
352 		if (!cap_raised(eff_cap, CAP_AUDIT_WRITE))
353 			err = -EPERM;
354 		break;
355 	default:  /* bad msg */
356 		err = -EINVAL;
357 	}
358 
359 	return err;
360 }
361 
362 static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
363 {
364 	u32			uid, pid, seq;
365 	void			*data;
366 	struct audit_status	*status_get, status_set;
367 	int			err;
368 	struct audit_buffer	*ab;
369 	u16			msg_type = nlh->nlmsg_type;
370 	uid_t			loginuid; /* loginuid of sender */
371 	struct audit_sig_info   sig_data;
372 
373 	err = audit_netlink_ok(NETLINK_CB(skb).eff_cap, msg_type);
374 	if (err)
375 		return err;
376 
377 	/* As soon as there's any sign of userspace auditd, start kauditd to talk to it */
378 	if (!kauditd_task)
379 		kauditd_task = kthread_run(kauditd_thread, NULL, "kauditd");
380 	if (IS_ERR(kauditd_task)) {
381 		err = PTR_ERR(kauditd_task);
382 		kauditd_task = NULL;
383 		return err;
384 	}
385 
386 	pid  = NETLINK_CREDS(skb)->pid;
387 	uid  = NETLINK_CREDS(skb)->uid;
388 	loginuid = NETLINK_CB(skb).loginuid;
389 	seq  = nlh->nlmsg_seq;
390 	data = NLMSG_DATA(nlh);
391 
392 	switch (msg_type) {
393 	case AUDIT_GET:
394 		status_set.enabled	 = audit_enabled;
395 		status_set.failure	 = audit_failure;
396 		status_set.pid		 = audit_pid;
397 		status_set.rate_limit	 = audit_rate_limit;
398 		status_set.backlog_limit = audit_backlog_limit;
399 		status_set.lost		 = atomic_read(&audit_lost);
400 		status_set.backlog	 = skb_queue_len(&audit_skb_queue);
401 		audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_GET, 0, 0,
402 				 &status_set, sizeof(status_set));
403 		break;
404 	case AUDIT_SET:
405 		if (nlh->nlmsg_len < sizeof(struct audit_status))
406 			return -EINVAL;
407 		status_get   = (struct audit_status *)data;
408 		if (status_get->mask & AUDIT_STATUS_ENABLED) {
409 			err = audit_set_enabled(status_get->enabled, loginuid);
410 			if (err < 0) return err;
411 		}
412 		if (status_get->mask & AUDIT_STATUS_FAILURE) {
413 			err = audit_set_failure(status_get->failure, loginuid);
414 			if (err < 0) return err;
415 		}
416 		if (status_get->mask & AUDIT_STATUS_PID) {
417 			int old   = audit_pid;
418 			audit_pid = status_get->pid;
419 			audit_log(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE,
420 				"audit_pid=%d old=%d by auid=%u",
421 				  audit_pid, old, loginuid);
422 		}
423 		if (status_get->mask & AUDIT_STATUS_RATE_LIMIT)
424 			audit_set_rate_limit(status_get->rate_limit, loginuid);
425 		if (status_get->mask & AUDIT_STATUS_BACKLOG_LIMIT)
426 			audit_set_backlog_limit(status_get->backlog_limit,
427 							loginuid);
428 		break;
429 	case AUDIT_USER:
430 	case AUDIT_FIRST_USER_MSG...AUDIT_LAST_USER_MSG:
431 		if (!audit_enabled && msg_type != AUDIT_USER_AVC)
432 			return 0;
433 
434 		err = audit_filter_user(&NETLINK_CB(skb), msg_type);
435 		if (err == 1) {
436 			err = 0;
437 			ab = audit_log_start(NULL, GFP_KERNEL, msg_type);
438 			if (ab) {
439 				audit_log_format(ab,
440 						 "user pid=%d uid=%u auid=%u msg='%.1024s'",
441 						 pid, uid, loginuid, (char *)data);
442 				audit_set_pid(ab, pid);
443 				audit_log_end(ab);
444 			}
445 		}
446 		break;
447 	case AUDIT_ADD:
448 	case AUDIT_DEL:
449 		if (nlh->nlmsg_len < sizeof(struct audit_rule))
450 			return -EINVAL;
451 		/* fallthrough */
452 	case AUDIT_LIST:
453 		err = audit_receive_filter(nlh->nlmsg_type, NETLINK_CB(skb).pid,
454 					   uid, seq, data, loginuid);
455 		break;
456 	case AUDIT_SIGNAL_INFO:
457 		sig_data.uid = audit_sig_uid;
458 		sig_data.pid = audit_sig_pid;
459 		audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_SIGNAL_INFO,
460 				0, 0, &sig_data, sizeof(sig_data));
461 		break;
462 	default:
463 		err = -EINVAL;
464 		break;
465 	}
466 
467 	return err < 0 ? err : 0;
468 }
469 
470 /* Get message from skb (based on rtnetlink_rcv_skb).  Each message is
471  * processed by audit_receive_msg.  Malformed skbs with wrong length are
472  * discarded silently.  */
473 static void audit_receive_skb(struct sk_buff *skb)
474 {
475 	int		err;
476 	struct nlmsghdr	*nlh;
477 	u32		rlen;
478 
479 	while (skb->len >= NLMSG_SPACE(0)) {
480 		nlh = (struct nlmsghdr *)skb->data;
481 		if (nlh->nlmsg_len < sizeof(*nlh) || skb->len < nlh->nlmsg_len)
482 			return;
483 		rlen = NLMSG_ALIGN(nlh->nlmsg_len);
484 		if (rlen > skb->len)
485 			rlen = skb->len;
486 		if ((err = audit_receive_msg(skb, nlh))) {
487 			netlink_ack(skb, nlh, err);
488 		} else if (nlh->nlmsg_flags & NLM_F_ACK)
489 			netlink_ack(skb, nlh, 0);
490 		skb_pull(skb, rlen);
491 	}
492 }
493 
494 /* Receive messages from netlink socket. */
495 static void audit_receive(struct sock *sk, int length)
496 {
497 	struct sk_buff  *skb;
498 	unsigned int qlen;
499 
500 	down(&audit_netlink_sem);
501 
502 	for (qlen = skb_queue_len(&sk->sk_receive_queue); qlen; qlen--) {
503 		skb = skb_dequeue(&sk->sk_receive_queue);
504 		audit_receive_skb(skb);
505 		kfree_skb(skb);
506 	}
507 	up(&audit_netlink_sem);
508 }
509 
510 
511 /* Initialize audit support at boot time. */
512 static int __init audit_init(void)
513 {
514 	printk(KERN_INFO "audit: initializing netlink socket (%s)\n",
515 	       audit_default ? "enabled" : "disabled");
516 	audit_sock = netlink_kernel_create(NETLINK_AUDIT, 0, audit_receive,
517 					   THIS_MODULE);
518 	if (!audit_sock)
519 		audit_panic("cannot initialize netlink socket");
520 
521 	audit_sock->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
522 	skb_queue_head_init(&audit_skb_queue);
523 	audit_initialized = 1;
524 	audit_enabled = audit_default;
525 	audit_log(NULL, GFP_KERNEL, AUDIT_KERNEL, "initialized");
526 	return 0;
527 }
528 __initcall(audit_init);
529 
530 /* Process kernel command-line parameter at boot time.  audit=0 or audit=1. */
531 static int __init audit_enable(char *str)
532 {
533 	audit_default = !!simple_strtol(str, NULL, 0);
534 	printk(KERN_INFO "audit: %s%s\n",
535 	       audit_default ? "enabled" : "disabled",
536 	       audit_initialized ? "" : " (after initialization)");
537 	if (audit_initialized)
538 		audit_enabled = audit_default;
539 	return 0;
540 }
541 
542 __setup("audit=", audit_enable);
543 
544 static void audit_buffer_free(struct audit_buffer *ab)
545 {
546 	unsigned long flags;
547 
548 	if (!ab)
549 		return;
550 
551 	if (ab->skb)
552 		kfree_skb(ab->skb);
553 
554 	spin_lock_irqsave(&audit_freelist_lock, flags);
555 	if (++audit_freelist_count > AUDIT_MAXFREE)
556 		kfree(ab);
557 	else
558 		list_add(&ab->list, &audit_freelist);
559 	spin_unlock_irqrestore(&audit_freelist_lock, flags);
560 }
561 
562 static struct audit_buffer * audit_buffer_alloc(struct audit_context *ctx,
563 						gfp_t gfp_mask, int type)
564 {
565 	unsigned long flags;
566 	struct audit_buffer *ab = NULL;
567 	struct nlmsghdr *nlh;
568 
569 	spin_lock_irqsave(&audit_freelist_lock, flags);
570 	if (!list_empty(&audit_freelist)) {
571 		ab = list_entry(audit_freelist.next,
572 				struct audit_buffer, list);
573 		list_del(&ab->list);
574 		--audit_freelist_count;
575 	}
576 	spin_unlock_irqrestore(&audit_freelist_lock, flags);
577 
578 	if (!ab) {
579 		ab = kmalloc(sizeof(*ab), gfp_mask);
580 		if (!ab)
581 			goto err;
582 	}
583 
584 	ab->skb = alloc_skb(AUDIT_BUFSIZ, gfp_mask);
585 	if (!ab->skb)
586 		goto err;
587 
588 	ab->ctx = ctx;
589 	ab->gfp_mask = gfp_mask;
590 	nlh = (struct nlmsghdr *)skb_put(ab->skb, NLMSG_SPACE(0));
591 	nlh->nlmsg_type = type;
592 	nlh->nlmsg_flags = 0;
593 	nlh->nlmsg_pid = 0;
594 	nlh->nlmsg_seq = 0;
595 	return ab;
596 err:
597 	audit_buffer_free(ab);
598 	return NULL;
599 }
600 
601 /* Compute a serial number for the audit record.  Audit records are
602  * written to user-space as soon as they are generated, so a complete
603  * audit record may be written in several pieces.  The timestamp of the
604  * record and this serial number are used by the user-space tools to
605  * determine which pieces belong to the same audit record.  The
606  * (timestamp,serial) tuple is unique for each syscall and is live from
607  * syscall entry to syscall exit.
608  *
609  * NOTE: Another possibility is to store the formatted records off the
610  * audit context (for those records that have a context), and emit them
611  * all at syscall exit.  However, this could delay the reporting of
612  * significant errors until syscall exit (or never, if the system
613  * halts). */
614 
615 unsigned int audit_serial(void)
616 {
617 	static spinlock_t serial_lock = SPIN_LOCK_UNLOCKED;
618 	static unsigned int serial = 0;
619 
620 	unsigned long flags;
621 	unsigned int ret;
622 
623 	spin_lock_irqsave(&serial_lock, flags);
624 	do {
625 		ret = ++serial;
626 	} while (unlikely(!ret));
627 	spin_unlock_irqrestore(&serial_lock, flags);
628 
629 	return ret;
630 }
631 
632 static inline void audit_get_stamp(struct audit_context *ctx,
633 				   struct timespec *t, unsigned int *serial)
634 {
635 	if (ctx)
636 		auditsc_get_stamp(ctx, t, serial);
637 	else {
638 		*t = CURRENT_TIME;
639 		*serial = audit_serial();
640 	}
641 }
642 
643 /* Obtain an audit buffer.  This routine does locking to obtain the
644  * audit buffer, but then no locking is required for calls to
645  * audit_log_*format.  If the tsk is a task that is currently in a
646  * syscall, then the syscall is marked as auditable and an audit record
647  * will be written at syscall exit.  If there is no associated task, tsk
648  * should be NULL. */
649 
650 struct audit_buffer *audit_log_start(struct audit_context *ctx, gfp_t gfp_mask,
651 				     int type)
652 {
653 	struct audit_buffer	*ab	= NULL;
654 	struct timespec		t;
655 	unsigned int		serial;
656 	int reserve;
657 	unsigned long timeout_start = jiffies;
658 
659 	if (!audit_initialized)
660 		return NULL;
661 
662 	if (gfp_mask & __GFP_WAIT)
663 		reserve = 0;
664 	else
665 		reserve = 5; /* Allow atomic callers to go up to five
666 				entries over the normal backlog limit */
667 
668 	while (audit_backlog_limit
669 	       && skb_queue_len(&audit_skb_queue) > audit_backlog_limit + reserve) {
670 		if (gfp_mask & __GFP_WAIT && audit_backlog_wait_time
671 		    && time_before(jiffies, timeout_start + audit_backlog_wait_time)) {
672 
673 			/* Wait for auditd to drain the queue a little */
674 			DECLARE_WAITQUEUE(wait, current);
675 			set_current_state(TASK_INTERRUPTIBLE);
676 			add_wait_queue(&audit_backlog_wait, &wait);
677 
678 			if (audit_backlog_limit &&
679 			    skb_queue_len(&audit_skb_queue) > audit_backlog_limit)
680 				schedule_timeout(timeout_start + audit_backlog_wait_time - jiffies);
681 
682 			__set_current_state(TASK_RUNNING);
683 			remove_wait_queue(&audit_backlog_wait, &wait);
684 			continue;
685 		}
686 		if (audit_rate_check())
687 			printk(KERN_WARNING
688 			       "audit: audit_backlog=%d > "
689 			       "audit_backlog_limit=%d\n",
690 			       skb_queue_len(&audit_skb_queue),
691 			       audit_backlog_limit);
692 		audit_log_lost("backlog limit exceeded");
693 		audit_backlog_wait_time = audit_backlog_wait_overflow;
694 		wake_up(&audit_backlog_wait);
695 		return NULL;
696 	}
697 
698 	ab = audit_buffer_alloc(ctx, gfp_mask, type);
699 	if (!ab) {
700 		audit_log_lost("out of memory in audit_log_start");
701 		return NULL;
702 	}
703 
704 	audit_get_stamp(ab->ctx, &t, &serial);
705 
706 	audit_log_format(ab, "audit(%lu.%03lu:%u): ",
707 			 t.tv_sec, t.tv_nsec/1000000, serial);
708 	return ab;
709 }
710 
711 /**
712  * audit_expand - expand skb in the audit buffer
713  * @ab: audit_buffer
714  *
715  * Returns 0 (no space) on failed expansion, or available space if
716  * successful.
717  */
718 static inline int audit_expand(struct audit_buffer *ab, int extra)
719 {
720 	struct sk_buff *skb = ab->skb;
721 	int ret = pskb_expand_head(skb, skb_headroom(skb), extra,
722 				   ab->gfp_mask);
723 	if (ret < 0) {
724 		audit_log_lost("out of memory in audit_expand");
725 		return 0;
726 	}
727 	return skb_tailroom(skb);
728 }
729 
730 /* Format an audit message into the audit buffer.  If there isn't enough
731  * room in the audit buffer, more room will be allocated and vsnprint
732  * will be called a second time.  Currently, we assume that a printk
733  * can't format message larger than 1024 bytes, so we don't either. */
734 static void audit_log_vformat(struct audit_buffer *ab, const char *fmt,
735 			      va_list args)
736 {
737 	int len, avail;
738 	struct sk_buff *skb;
739 	va_list args2;
740 
741 	if (!ab)
742 		return;
743 
744 	BUG_ON(!ab->skb);
745 	skb = ab->skb;
746 	avail = skb_tailroom(skb);
747 	if (avail == 0) {
748 		avail = audit_expand(ab, AUDIT_BUFSIZ);
749 		if (!avail)
750 			goto out;
751 	}
752 	va_copy(args2, args);
753 	len = vsnprintf(skb->tail, avail, fmt, args);
754 	if (len >= avail) {
755 		/* The printk buffer is 1024 bytes long, so if we get
756 		 * here and AUDIT_BUFSIZ is at least 1024, then we can
757 		 * log everything that printk could have logged. */
758 		avail = audit_expand(ab, max_t(unsigned, AUDIT_BUFSIZ, 1+len-avail));
759 		if (!avail)
760 			goto out;
761 		len = vsnprintf(skb->tail, avail, fmt, args2);
762 	}
763 	if (len > 0)
764 		skb_put(skb, len);
765 out:
766 	return;
767 }
768 
769 /* Format a message into the audit buffer.  All the work is done in
770  * audit_log_vformat. */
771 void audit_log_format(struct audit_buffer *ab, const char *fmt, ...)
772 {
773 	va_list args;
774 
775 	if (!ab)
776 		return;
777 	va_start(args, fmt);
778 	audit_log_vformat(ab, fmt, args);
779 	va_end(args);
780 }
781 
782 /* This function will take the passed buf and convert it into a string of
783  * ascii hex digits. The new string is placed onto the skb. */
784 void audit_log_hex(struct audit_buffer *ab, const unsigned char *buf,
785 		size_t len)
786 {
787 	int i, avail, new_len;
788 	unsigned char *ptr;
789 	struct sk_buff *skb;
790 	static const unsigned char *hex = "0123456789ABCDEF";
791 
792 	BUG_ON(!ab->skb);
793 	skb = ab->skb;
794 	avail = skb_tailroom(skb);
795 	new_len = len<<1;
796 	if (new_len >= avail) {
797 		/* Round the buffer request up to the next multiple */
798 		new_len = AUDIT_BUFSIZ*(((new_len-avail)/AUDIT_BUFSIZ) + 1);
799 		avail = audit_expand(ab, new_len);
800 		if (!avail)
801 			return;
802 	}
803 
804 	ptr = skb->tail;
805 	for (i=0; i<len; i++) {
806 		*ptr++ = hex[(buf[i] & 0xF0)>>4]; /* Upper nibble */
807 		*ptr++ = hex[buf[i] & 0x0F];	  /* Lower nibble */
808 	}
809 	*ptr = 0;
810 	skb_put(skb, len << 1); /* new string is twice the old string */
811 }
812 
813 /* This code will escape a string that is passed to it if the string
814  * contains a control character, unprintable character, double quote mark,
815  * or a space. Unescaped strings will start and end with a double quote mark.
816  * Strings that are escaped are printed in hex (2 digits per char). */
817 void audit_log_untrustedstring(struct audit_buffer *ab, const char *string)
818 {
819 	const unsigned char *p = string;
820 
821 	while (*p) {
822 		if (*p == '"' || *p < 0x21 || *p > 0x7f) {
823 			audit_log_hex(ab, string, strlen(string));
824 			return;
825 		}
826 		p++;
827 	}
828 	audit_log_format(ab, "\"%s\"", string);
829 }
830 
831 /* This is a helper-function to print the escaped d_path */
832 void audit_log_d_path(struct audit_buffer *ab, const char *prefix,
833 		      struct dentry *dentry, struct vfsmount *vfsmnt)
834 {
835 	char *p, *path;
836 
837 	if (prefix)
838 		audit_log_format(ab, " %s", prefix);
839 
840 	/* We will allow 11 spaces for ' (deleted)' to be appended */
841 	path = kmalloc(PATH_MAX+11, ab->gfp_mask);
842 	if (!path) {
843 		audit_log_format(ab, "<no memory>");
844 		return;
845 	}
846 	p = d_path(dentry, vfsmnt, path, PATH_MAX+11);
847 	if (IS_ERR(p)) { /* Should never happen since we send PATH_MAX */
848 		/* FIXME: can we save some information here? */
849 		audit_log_format(ab, "<too long>");
850 	} else
851 		audit_log_untrustedstring(ab, p);
852 	kfree(path);
853 }
854 
855 /* The netlink_* functions cannot be called inside an irq context, so
856  * the audit buffer is places on a queue and a tasklet is scheduled to
857  * remove them from the queue outside the irq context.  May be called in
858  * any context. */
859 void audit_log_end(struct audit_buffer *ab)
860 {
861 	if (!ab)
862 		return;
863 	if (!audit_rate_check()) {
864 		audit_log_lost("rate limit exceeded");
865 	} else {
866 		if (audit_pid) {
867 			struct nlmsghdr *nlh = (struct nlmsghdr *)ab->skb->data;
868 			nlh->nlmsg_len = ab->skb->len - NLMSG_SPACE(0);
869 			skb_queue_tail(&audit_skb_queue, ab->skb);
870 			ab->skb = NULL;
871 			wake_up_interruptible(&kauditd_wait);
872 		} else {
873 			printk(KERN_NOTICE "%s\n", ab->skb->data + NLMSG_SPACE(0));
874 		}
875 	}
876 	audit_buffer_free(ab);
877 }
878 
879 /* Log an audit record.  This is a convenience function that calls
880  * audit_log_start, audit_log_vformat, and audit_log_end.  It may be
881  * called in any context. */
882 void audit_log(struct audit_context *ctx, gfp_t gfp_mask, int type,
883 	       const char *fmt, ...)
884 {
885 	struct audit_buffer *ab;
886 	va_list args;
887 
888 	ab = audit_log_start(ctx, gfp_mask, type);
889 	if (ab) {
890 		va_start(args, fmt);
891 		audit_log_vformat(ab, fmt, args);
892 		va_end(args);
893 		audit_log_end(ab);
894 	}
895 }
896