1 /* audit.c -- Auditing support -*- linux-c -*-
2 * Gateway between the kernel (e.g., selinux) and the user-space audit daemon.
3 * System-call specific features have moved to auditsc.c
5 * Copyright 2003-2004 Red Hat Inc., Durham, North Carolina.
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.
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.
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
22 * Written by Rickard E. (Rik) Faith <faith@redhat.com>
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
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
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.
41 * Example user-space utilities: http://people.redhat.com/faith/audit/
44 #include <linux/init.h>
45 #include <asm/atomic.h>
46 #include <asm/types.h>
48 #include <linux/module.h>
50 #include <linux/audit.h>
53 #include <linux/skbuff.h>
54 #include <linux/netlink.h>
56 /* No auditing will take place until audit_initialized != 0.
57 * (Initialization happens after skb_init is called.) */
58 static int audit_initialized
;
60 /* No syscall auditing will take place unless audit_enabled != 0. */
63 /* Default state when kernel boots without any parameters. */
64 static int audit_default
;
66 /* If auditing cannot proceed, audit_failure selects what happens. */
67 static int audit_failure
= AUDIT_FAIL_PRINTK
;
69 /* If audit records are to be written to the netlink socket, audit_pid
70 * contains the (non-zero) pid. */
73 /* If audit_limit is non-zero, limit the rate of sending audit records
74 * to that number per second. This prevents DoS attacks, but results in
75 * audit records being dropped. */
76 static int audit_rate_limit
;
78 /* Number of outstanding audit_buffers allowed. */
79 static int audit_backlog_limit
= 64;
80 static atomic_t audit_backlog
= ATOMIC_INIT(0);
82 /* Records can be lost in several ways:
83 0) [suppressed in audit_alloc]
84 1) out of memory in audit_log_start [kmalloc of struct audit_buffer]
85 2) out of memory in audit_log_move [alloc_skb]
86 3) suppressed due to audit_rate_limit
87 4) suppressed due to audit_backlog_limit
89 static atomic_t audit_lost
= ATOMIC_INIT(0);
91 /* The netlink socket. */
92 static struct sock
*audit_sock
;
94 /* There are two lists of audit buffers. The txlist contains audit
95 * buffers that cannot be sent immediately to the netlink device because
96 * we are in an irq context (these are sent later in a tasklet).
98 * The second list is a list of pre-allocated audit buffers (if more
99 * than AUDIT_MAXFREE are in use, the audit buffer is freed instead of
100 * being placed on the freelist). */
101 static spinlock_t audit_txlist_lock
= SPIN_LOCK_UNLOCKED
;
102 static spinlock_t audit_freelist_lock
= SPIN_LOCK_UNLOCKED
;
103 static int audit_freelist_count
= 0;
104 static LIST_HEAD(audit_txlist
);
105 static LIST_HEAD(audit_freelist
);
107 /* There are three lists of rules -- one to search at task creation
108 * time, one to search at syscall entry time, and another to search at
109 * syscall exit time. */
110 static LIST_HEAD(audit_tsklist
);
111 static LIST_HEAD(audit_entlist
);
112 static LIST_HEAD(audit_extlist
);
114 /* The netlink socket is only to be read by 1 CPU, which lets us assume
115 * that list additions and deletions never happen simultaneiously in
117 static DECLARE_MUTEX(audit_netlink_sem
);
119 /* AUDIT_BUFSIZ is the size of the temporary buffer used for formatting
120 * audit records. Since printk uses a 1024 byte buffer, this buffer
121 * should be at least that large. */
122 #define AUDIT_BUFSIZ 1024
124 /* AUDIT_MAXFREE is the number of empty audit_buffers we keep on the
125 * audit_freelist. Doing so eliminates many kmalloc/kfree calls. */
126 #define AUDIT_MAXFREE (2*NR_CPUS)
128 /* The audit_buffer is used when formatting an audit record. The caller
129 * locks briefly to get the record off the freelist or to allocate the
130 * buffer, and locks briefly to send the buffer to the netlink layer or
131 * to place it on a transmit queue. Multiple audit_buffers can be in
132 * use simultaneously. */
133 struct audit_buffer
{
134 struct list_head list
;
135 struct sk_buff_head sklist
; /* formatted skbs ready to send */
136 struct audit_context
*ctx
; /* NULL or associated context */
137 int len
; /* used area of tmp */
138 char tmp
[AUDIT_BUFSIZ
];
140 /* Pointer to header and contents */
141 struct nlmsghdr
*nlh
;
145 int count
; /* Times requeued */
149 struct list_head list
;
150 struct audit_rule rule
;
153 static void audit_panic(const char *message
)
155 switch (audit_failure
)
157 case AUDIT_FAIL_SILENT
:
159 case AUDIT_FAIL_PRINTK
:
160 printk(KERN_ERR
"audit: %s\n", message
);
162 case AUDIT_FAIL_PANIC
:
168 static inline int audit_rate_check(void)
170 static unsigned long last_check
= 0;
171 static int messages
= 0;
172 static spinlock_t lock
= SPIN_LOCK_UNLOCKED
;
175 unsigned long elapsed
;
178 if (!audit_rate_limit
) return 1;
180 spin_lock_irqsave(&lock
, flags
);
181 if (++messages
< audit_rate_limit
) {
185 elapsed
= now
- last_check
;
192 spin_unlock_irqrestore(&lock
, flags
);
197 /* Emit at least 1 message per second, even if audit_rate_check is
199 void audit_log_lost(const char *message
)
201 static unsigned long last_msg
= 0;
202 static spinlock_t lock
= SPIN_LOCK_UNLOCKED
;
207 atomic_inc(&audit_lost
);
209 print
= (audit_failure
== AUDIT_FAIL_PANIC
|| !audit_rate_limit
);
212 spin_lock_irqsave(&lock
, flags
);
214 if (now
- last_msg
> HZ
) {
218 spin_unlock_irqrestore(&lock
, flags
);
223 "audit: audit_lost=%d audit_backlog=%d"
224 " audit_rate_limit=%d audit_backlog_limit=%d\n",
225 atomic_read(&audit_lost
),
226 atomic_read(&audit_backlog
),
228 audit_backlog_limit
);
229 audit_panic(message
);
234 int audit_set_rate_limit(int limit
)
236 int old
= audit_rate_limit
;
237 audit_rate_limit
= limit
;
238 audit_log(current
->audit_context
, "audit_rate_limit=%d old=%d",
239 audit_rate_limit
, old
);
243 int audit_set_backlog_limit(int limit
)
245 int old
= audit_backlog_limit
;
246 audit_backlog_limit
= limit
;
247 audit_log(current
->audit_context
, "audit_backlog_limit=%d old=%d",
248 audit_backlog_limit
, old
);
252 int audit_set_enabled(int state
)
254 int old
= audit_enabled
;
255 if (state
!= 0 && state
!= 1)
257 audit_enabled
= state
;
258 audit_log(current
->audit_context
, "audit_enabled=%d old=%d",
263 int audit_set_failure(int state
)
265 int old
= audit_failure
;
266 if (state
!= AUDIT_FAIL_SILENT
267 && state
!= AUDIT_FAIL_PRINTK
268 && state
!= AUDIT_FAIL_PANIC
)
270 audit_failure
= state
;
271 audit_log(current
->audit_context
, "audit_failure=%d old=%d",
277 void audit_send_reply(int pid
, int seq
, int type
, int done
, int multi
,
278 void *payload
, int size
)
281 struct nlmsghdr
*nlh
;
282 int len
= NLMSG_SPACE(size
);
284 int flags
= multi
? NLM_F_MULTI
: 0;
285 int t
= done
? NLMSG_DONE
: type
;
287 skb
= alloc_skb(len
, GFP_KERNEL
);
291 nlh
= NLMSG_PUT(skb
, pid
, seq
, t
, len
- sizeof(*nlh
));
292 nlh
->nlmsg_flags
= flags
;
293 data
= NLMSG_DATA(nlh
);
294 memcpy(data
, payload
, size
);
295 netlink_unicast(audit_sock
, skb
, pid
, MSG_DONTWAIT
);
298 nlmsg_failure
: /* Used by NLMSG_PUT */
303 static int audit_receive_msg(struct sk_buff
*skb
, struct nlmsghdr
*nlh
)
307 struct audit_status
*status_get
, status_set
;
308 struct audit_login
*login
;
310 struct audit_buffer
*ab
;
312 pid
= NETLINK_CREDS(skb
)->pid
;
313 uid
= NETLINK_CREDS(skb
)->uid
;
314 seq
= nlh
->nlmsg_seq
;
315 data
= NLMSG_DATA(nlh
);
317 switch (nlh
->nlmsg_type
) {
319 status_set
.enabled
= audit_enabled
;
320 status_set
.failure
= audit_failure
;
321 status_set
.pid
= audit_pid
;
322 status_set
.rate_limit
= audit_rate_limit
;
323 status_set
.backlog_limit
= audit_backlog_limit
;
324 status_set
.lost
= atomic_read(&audit_lost
);
325 status_set
.backlog
= atomic_read(&audit_backlog
);
326 audit_send_reply(pid
, seq
, AUDIT_GET
, 0, 0,
327 &status_set
, sizeof(status_set
));
330 if (!capable(CAP_SYS_ADMIN
))
332 status_get
= (struct audit_status
*)data
;
333 if (status_get
->mask
& AUDIT_STATUS_ENABLED
) {
334 err
= audit_set_enabled(status_get
->enabled
);
335 if (err
< 0) return err
;
337 if (status_get
->mask
& AUDIT_STATUS_FAILURE
) {
338 err
= audit_set_failure(status_get
->failure
);
339 if (err
< 0) return err
;
341 if (status_get
->mask
& AUDIT_STATUS_PID
) {
343 audit_pid
= status_get
->pid
;
344 audit_log(current
->audit_context
,
345 "audit_pid=%d old=%d", audit_pid
, old
);
347 if (status_get
->mask
& AUDIT_STATUS_RATE_LIMIT
)
348 audit_set_rate_limit(status_get
->rate_limit
);
349 if (status_get
->mask
& AUDIT_STATUS_BACKLOG_LIMIT
)
350 audit_set_backlog_limit(status_get
->backlog_limit
);
353 ab
= audit_log_start(NULL
);
355 break; /* audit_panic has been called */
357 "user pid=%d uid=%d length=%d msg='%.1024s'",
360 - ((char *)data
- (char *)nlh
)),
362 ab
->type
= AUDIT_USER
;
367 if (!capable(CAP_SYS_ADMIN
))
369 login
= (struct audit_login
*)data
;
370 ab
= audit_log_start(NULL
);
372 audit_log_format(ab
, "login pid=%d uid=%d loginuid=%d"
373 " length=%d msg='%.1024s'",
378 ab
->type
= AUDIT_LOGIN
;
382 #ifdef CONFIG_AUDITSYSCALL
383 err
= audit_set_loginuid(current
->audit_context
,
390 #ifdef CONFIG_AUDITSYSCALL
391 err
= audit_receive_filter(nlh
->nlmsg_type
, pid
, uid
, seq
,
402 return err
< 0 ? err
: 0;
405 /* Get message from skb (based on rtnetlink_rcv_skb). Each message is
406 * processed by audit_receive_msg. Malformed skbs with wrong length are
407 * discarded silently. */
408 static int audit_receive_skb(struct sk_buff
*skb
)
411 struct nlmsghdr
*nlh
;
414 while (skb
->len
>= NLMSG_SPACE(0)) {
415 nlh
= (struct nlmsghdr
*)skb
->data
;
416 if (nlh
->nlmsg_len
< sizeof(*nlh
) || skb
->len
< nlh
->nlmsg_len
)
418 rlen
= NLMSG_ALIGN(nlh
->nlmsg_len
);
421 if ((err
= audit_receive_msg(skb
, nlh
))) {
422 netlink_ack(skb
, nlh
, -err
);
423 } else if (nlh
->nlmsg_flags
& NLM_F_ACK
)
424 netlink_ack(skb
, nlh
, 0);
430 /* Receive messages from netlink socket. */
431 static void audit_receive(struct sock
*sk
, int length
)
435 if (down_trylock(&audit_netlink_sem
))
438 /* FIXME: this must not cause starvation */
439 while ((skb
= skb_dequeue(&sk
->sk_receive_queue
))) {
440 if (audit_receive_skb(skb
) && skb
->len
)
441 skb_queue_head(&sk
->sk_receive_queue
, skb
);
445 up(&audit_netlink_sem
);
448 /* Move data from tmp buffer into an skb. This is an extra copy, and
449 * that is unfortunate. However, the copy will only occur when a record
450 * is being written to user space, which is already a high-overhead
451 * operation. (Elimination of the copy is possible, for example, by
452 * writing directly into a pre-allocated skb, at the cost of wasting
454 static void audit_log_move(struct audit_buffer
*ab
)
458 int extra
= ab
->nlh
? 0 : NLMSG_SPACE(0);
460 skb
= skb_peek(&ab
->sklist
);
461 if (!skb
|| skb_tailroom(skb
) <= ab
->len
+ extra
) {
462 skb
= alloc_skb(2 * ab
->len
+ extra
, GFP_ATOMIC
);
464 ab
->len
= 0; /* Lose information in ab->tmp */
465 audit_log_lost("out of memory in audit_log_move");
468 __skb_queue_tail(&ab
->sklist
, skb
);
470 ab
->nlh
= (struct nlmsghdr
*)skb_put(skb
,
473 start
= skb_put(skb
, ab
->len
);
474 memcpy(start
, ab
->tmp
, ab
->len
);
478 /* Iterate over the skbuff in the audit_buffer, sending their contents
480 static inline int audit_log_drain(struct audit_buffer
*ab
)
484 while ((skb
= skb_dequeue(&ab
->sklist
))) {
489 ab
->nlh
->nlmsg_len
= ab
->total
;
490 ab
->nlh
->nlmsg_type
= ab
->type
;
491 ab
->nlh
->nlmsg_flags
= 0;
492 ab
->nlh
->nlmsg_seq
= 0;
493 ab
->nlh
->nlmsg_pid
= ab
->pid
;
495 skb_get(skb
); /* because netlink_* frees */
496 retval
= netlink_unicast(audit_sock
, skb
, audit_pid
,
499 if (retval
== -EAGAIN
&& ab
->count
< 5) {
501 audit_log_end_irq(ab
);
505 if (retval
== -ECONNREFUSED
) {
507 "audit: *NO* daemon at audit_pid=%d\n",
511 audit_log_lost("netlink socket too busy");
513 if (!audit_pid
) { /* No daemon */
514 int offset
= ab
->nlh
? NLMSG_SPACE(0) : 0;
515 int len
= skb
->len
- offset
;
516 printk(KERN_ERR
"%*.*s\n",
517 len
, len
, skb
->data
+ offset
);
525 /* Initialize audit support at boot time. */
526 int __init
audit_init(void)
528 printk(KERN_INFO
"audit: initializing netlink socket (%s)\n",
529 audit_default
? "enabled" : "disabled");
530 audit_sock
= netlink_kernel_create(NETLINK_AUDIT
, audit_receive
);
532 audit_panic("cannot initialize netlink socket");
534 audit_initialized
= 1;
535 audit_enabled
= audit_default
;
536 audit_log(NULL
, "initialized");
541 /* Without CONFIG_NET, we have no skbuffs. For now, print what we have
543 static void audit_log_move(struct audit_buffer
*ab
)
545 printk(KERN_ERR
"%*.*s\n", ab
->len
, ab
->len
, ab
->tmp
);
549 static inline int audit_log_drain(struct audit_buffer
*ab
)
554 /* Initialize audit support at boot time. */
555 int __init
audit_init(void)
557 printk(KERN_INFO
"audit: initializing WITHOUT netlink support\n");
561 audit_initialized
= 1;
562 audit_enabled
= audit_default
;
563 audit_log(NULL
, "initialized");
568 __initcall(audit_init
);
570 /* Process kernel command-line parameter at boot time. audit=0 or audit=1. */
571 static int __init
audit_enable(char *str
)
573 audit_default
= !!simple_strtol(str
, NULL
, 0);
574 printk(KERN_INFO
"audit: %s%s\n",
575 audit_default
? "enabled" : "disabled",
576 audit_initialized
? "" : " (after initialization)");
577 if (audit_initialized
)
578 audit_enabled
= audit_default
;
582 __setup("audit=", audit_enable
);
585 /* Obtain an audit buffer. This routine does locking to obtain the
586 * audit buffer, but then no locking is required for calls to
587 * audit_log_*format. If the tsk is a task that is currently in a
588 * syscall, then the syscall is marked as auditable and an audit record
589 * will be written at syscall exit. If there is no associated task, tsk
591 struct audit_buffer
*audit_log_start(struct audit_context
*ctx
)
593 struct audit_buffer
*ab
= NULL
;
598 if (!audit_initialized
)
601 if (audit_backlog_limit
602 && atomic_read(&audit_backlog
) > audit_backlog_limit
) {
603 if (audit_rate_check())
605 "audit: audit_backlog=%d > "
606 "audit_backlog_limit=%d\n",
607 atomic_read(&audit_backlog
),
608 audit_backlog_limit
);
609 audit_log_lost("backlog limit exceeded");
613 spin_lock_irqsave(&audit_freelist_lock
, flags
);
614 if (!list_empty(&audit_freelist
)) {
615 ab
= list_entry(audit_freelist
.next
,
616 struct audit_buffer
, list
);
618 --audit_freelist_count
;
620 spin_unlock_irqrestore(&audit_freelist_lock
, flags
);
623 ab
= kmalloc(sizeof(*ab
), GFP_ATOMIC
);
625 audit_log_lost("audit: out of memory in audit_log_start");
629 atomic_inc(&audit_backlog
);
630 skb_queue_head_init(&ab
->sklist
);
636 ab
->type
= AUDIT_KERNEL
;
640 #ifdef CONFIG_AUDITSYSCALL
642 audit_get_stamp(ab
->ctx
, &t
, &serial
);
647 audit_log_format(ab
, "audit(%lu.%03lu:%u): ",
648 t
.tv_sec
, t
.tv_nsec
/1000000, serial
);
653 /* Format an audit message into the audit buffer. If there isn't enough
654 * room in the audit buffer, more room will be allocated and vsnprint
655 * will be called a second time. Currently, we assume that a printk
656 * can't format message larger than 1024 bytes, so we don't either. */
657 static void audit_log_vformat(struct audit_buffer
*ab
, const char *fmt
,
665 avail
= sizeof(ab
->tmp
) - ab
->len
;
668 avail
= sizeof(ab
->tmp
) - ab
->len
;
670 len
= vsnprintf(ab
->tmp
+ ab
->len
, avail
, fmt
, args
);
672 /* The printk buffer is 1024 bytes long, so if we get
673 * here and AUDIT_BUFSIZ is at least 1024, then we can
674 * log everything that printk could have logged. */
676 avail
= sizeof(ab
->tmp
) - ab
->len
;
677 len
= vsnprintf(ab
->tmp
+ ab
->len
, avail
, fmt
, args
);
679 ab
->len
+= (len
< avail
) ? len
: avail
;
680 ab
->total
+= (len
< avail
) ? len
: avail
;
683 /* Format a message into the audit buffer. All the work is done in
684 * audit_log_vformat. */
685 void audit_log_format(struct audit_buffer
*ab
, const char *fmt
, ...)
692 audit_log_vformat(ab
, fmt
, args
);
696 /* This is a helper-function to print the d_path without using a static
697 * buffer or allocating another buffer in addition to the one in
699 void audit_log_d_path(struct audit_buffer
*ab
, const char *prefix
,
700 struct dentry
*dentry
, struct vfsmount
*vfsmnt
)
705 if (prefix
) audit_log_format(ab
, " %s", prefix
);
709 avail
= sizeof(ab
->tmp
) - ab
->len
;
710 p
= d_path(dentry
, vfsmnt
, ab
->tmp
+ ab
->len
, avail
);
712 /* FIXME: can we save some information here? */
713 audit_log_format(ab
, "<toolong>");
715 /* path isn't at start of buffer */
716 len
= (ab
->tmp
+ sizeof(ab
->tmp
) - 1) - p
;
717 memmove(ab
->tmp
+ ab
->len
, p
, len
);
723 /* Remove queued messages from the audit_txlist and send them to userspace. */
724 static void audit_tasklet_handler(unsigned long arg
)
727 struct audit_buffer
*ab
;
730 spin_lock_irqsave(&audit_txlist_lock
, flags
);
731 list_splice_init(&audit_txlist
, &list
);
732 spin_unlock_irqrestore(&audit_txlist_lock
, flags
);
734 while (!list_empty(&list
)) {
735 ab
= list_entry(list
.next
, struct audit_buffer
, list
);
737 audit_log_end_fast(ab
);
741 static DECLARE_TASKLET(audit_tasklet
, audit_tasklet_handler
, 0);
743 /* The netlink_* functions cannot be called inside an irq context, so
744 * the audit buffer is places on a queue and a tasklet is scheduled to
745 * remove them from the queue outside the irq context. May be called in
747 void audit_log_end_irq(struct audit_buffer
*ab
)
753 spin_lock_irqsave(&audit_txlist_lock
, flags
);
754 list_add_tail(&ab
->list
, &audit_txlist
);
755 spin_unlock_irqrestore(&audit_txlist_lock
, flags
);
757 tasklet_schedule(&audit_tasklet
);
760 /* Send the message in the audit buffer directly to user space. May not
761 * be called in an irq context. */
762 void audit_log_end_fast(struct audit_buffer
*ab
)
769 if (!audit_rate_check()) {
770 audit_log_lost("rate limit exceeded");
773 if (audit_log_drain(ab
))
777 atomic_dec(&audit_backlog
);
778 spin_lock_irqsave(&audit_freelist_lock
, flags
);
779 if (++audit_freelist_count
> AUDIT_MAXFREE
)
782 list_add(&ab
->list
, &audit_freelist
);
783 spin_unlock_irqrestore(&audit_freelist_lock
, flags
);
786 /* Send or queue the message in the audit buffer, depending on the
787 * current context. (A convenience function that may be called in any
789 void audit_log_end(struct audit_buffer
*ab
)
792 audit_log_end_irq(ab
);
794 audit_log_end_fast(ab
);
797 /* Log an audit record. This is a convenience function that calls
798 * audit_log_start, audit_log_vformat, and audit_log_end. It may be
799 * called in any context. */
800 void audit_log(struct audit_context
*ctx
, const char *fmt
, ...)
802 struct audit_buffer
*ab
;
805 ab
= audit_log_start(ctx
);
808 audit_log_vformat(ab
, fmt
, args
);