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
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/sgrubb/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 DEFINE_SPINLOCK(audit_txlist_lock);
102 static DEFINE_SPINLOCK(audit_freelist_lock);
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;
147 void audit_set_type(struct audit_buffer *ab, int type)
153 struct list_head list;
154 struct audit_rule rule;
157 static void audit_log_end_irq(struct audit_buffer *ab);
158 static void audit_log_end_fast(struct audit_buffer *ab);
160 static void audit_panic(const char *message)
162 switch (audit_failure)
164 case AUDIT_FAIL_SILENT:
166 case AUDIT_FAIL_PRINTK:
167 printk(KERN_ERR "audit: %s\n", message);
169 case AUDIT_FAIL_PANIC:
170 panic("audit: %s\n", message);
175 static inline int audit_rate_check(void)
177 static unsigned long last_check = 0;
178 static int messages = 0;
179 static DEFINE_SPINLOCK(lock);
182 unsigned long elapsed;
185 if (!audit_rate_limit) return 1;
187 spin_lock_irqsave(&lock, flags);
188 if (++messages < audit_rate_limit) {
192 elapsed = now - last_check;
199 spin_unlock_irqrestore(&lock, flags);
204 /* Emit at least 1 message per second, even if audit_rate_check is
206 void audit_log_lost(const char *message)
208 static unsigned long last_msg = 0;
209 static DEFINE_SPINLOCK(lock);
214 atomic_inc(&audit_lost);
216 print = (audit_failure == AUDIT_FAIL_PANIC || !audit_rate_limit);
219 spin_lock_irqsave(&lock, flags);
221 if (now - last_msg > HZ) {
225 spin_unlock_irqrestore(&lock, flags);
230 "audit: audit_lost=%d audit_backlog=%d"
231 " audit_rate_limit=%d audit_backlog_limit=%d\n",
232 atomic_read(&audit_lost),
233 atomic_read(&audit_backlog),
235 audit_backlog_limit);
236 audit_panic(message);
241 static int audit_set_rate_limit(int limit, uid_t loginuid)
243 int old = audit_rate_limit;
244 audit_rate_limit = limit;
245 audit_log(NULL, "audit_rate_limit=%d old=%d by auid %u",
246 audit_rate_limit, old, loginuid);
250 static int audit_set_backlog_limit(int limit, uid_t loginuid)
252 int old = audit_backlog_limit;
253 audit_backlog_limit = limit;
254 audit_log(NULL, "audit_backlog_limit=%d old=%d by auid %u",
255 audit_backlog_limit, old, loginuid);
259 static int audit_set_enabled(int state, uid_t loginuid)
261 int old = audit_enabled;
262 if (state != 0 && state != 1)
264 audit_enabled = state;
265 audit_log(NULL, "audit_enabled=%d old=%d by auid %u",
266 audit_enabled, old, loginuid);
270 static int audit_set_failure(int state, uid_t loginuid)
272 int old = audit_failure;
273 if (state != AUDIT_FAIL_SILENT
274 && state != AUDIT_FAIL_PRINTK
275 && state != AUDIT_FAIL_PANIC)
277 audit_failure = state;
278 audit_log(NULL, "audit_failure=%d old=%d by auid %u",
279 audit_failure, old, loginuid);
284 void audit_send_reply(int pid, int seq, int type, int done, int multi,
285 void *payload, int size)
288 struct nlmsghdr *nlh;
289 int len = NLMSG_SPACE(size);
291 int flags = multi ? NLM_F_MULTI : 0;
292 int t = done ? NLMSG_DONE : type;
294 skb = alloc_skb(len, GFP_KERNEL);
298 nlh = NLMSG_PUT(skb, pid, seq, t, len - sizeof(*nlh));
299 nlh->nlmsg_flags = flags;
300 data = NLMSG_DATA(nlh);
301 memcpy(data, payload, size);
302 netlink_unicast(audit_sock, skb, pid, MSG_DONTWAIT);
305 nlmsg_failure: /* Used by NLMSG_PUT */
311 * Check for appropriate CAP_AUDIT_ capabilities on incoming audit
314 static int audit_netlink_ok(kernel_cap_t eff_cap, u16 msg_type)
324 if (!cap_raised(eff_cap, CAP_AUDIT_CONTROL))
328 if (!cap_raised(eff_cap, CAP_AUDIT_WRITE))
331 default: /* bad msg */
338 static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
342 struct audit_status *status_get, status_set;
344 struct audit_buffer *ab;
345 u16 msg_type = nlh->nlmsg_type;
346 uid_t loginuid; /* loginuid of sender */
348 err = audit_netlink_ok(NETLINK_CB(skb).eff_cap, msg_type);
352 pid = NETLINK_CREDS(skb)->pid;
353 uid = NETLINK_CREDS(skb)->uid;
354 loginuid = NETLINK_CB(skb).loginuid;
355 seq = nlh->nlmsg_seq;
356 data = NLMSG_DATA(nlh);
360 status_set.enabled = audit_enabled;
361 status_set.failure = audit_failure;
362 status_set.pid = audit_pid;
363 status_set.rate_limit = audit_rate_limit;
364 status_set.backlog_limit = audit_backlog_limit;
365 status_set.lost = atomic_read(&audit_lost);
366 status_set.backlog = atomic_read(&audit_backlog);
367 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_GET, 0, 0,
368 &status_set, sizeof(status_set));
371 if (nlh->nlmsg_len < sizeof(struct audit_status))
373 status_get = (struct audit_status *)data;
374 if (status_get->mask & AUDIT_STATUS_ENABLED) {
375 err = audit_set_enabled(status_get->enabled, loginuid);
376 if (err < 0) return err;
378 if (status_get->mask & AUDIT_STATUS_FAILURE) {
379 err = audit_set_failure(status_get->failure, loginuid);
380 if (err < 0) return err;
382 if (status_get->mask & AUDIT_STATUS_PID) {
384 audit_pid = status_get->pid;
385 audit_log(NULL, "audit_pid=%d old=%d by auid %u",
386 audit_pid, old, loginuid);
388 if (status_get->mask & AUDIT_STATUS_RATE_LIMIT)
389 audit_set_rate_limit(status_get->rate_limit, loginuid);
390 if (status_get->mask & AUDIT_STATUS_BACKLOG_LIMIT)
391 audit_set_backlog_limit(status_get->backlog_limit,
395 ab = audit_log_start(NULL);
397 break; /* audit_panic has been called */
399 "user pid=%d uid=%d length=%d loginuid=%u"
403 - ((char *)data - (char *)nlh)),
404 loginuid, (char *)data);
405 ab->type = AUDIT_USER;
411 if (nlh->nlmsg_len < sizeof(struct audit_rule))
415 #ifdef CONFIG_AUDITSYSCALL
416 err = audit_receive_filter(nlh->nlmsg_type, NETLINK_CB(skb).pid,
417 uid, seq, data, loginuid);
427 return err < 0 ? err : 0;
430 /* Get message from skb (based on rtnetlink_rcv_skb). Each message is
431 * processed by audit_receive_msg. Malformed skbs with wrong length are
432 * discarded silently. */
433 static void audit_receive_skb(struct sk_buff *skb)
436 struct nlmsghdr *nlh;
439 while (skb->len >= NLMSG_SPACE(0)) {
440 nlh = (struct nlmsghdr *)skb->data;
441 if (nlh->nlmsg_len < sizeof(*nlh) || skb->len < nlh->nlmsg_len)
443 rlen = NLMSG_ALIGN(nlh->nlmsg_len);
446 if ((err = audit_receive_msg(skb, nlh))) {
447 netlink_ack(skb, nlh, err);
448 } else if (nlh->nlmsg_flags & NLM_F_ACK)
449 netlink_ack(skb, nlh, 0);
454 /* Receive messages from netlink socket. */
455 static void audit_receive(struct sock *sk, int length)
460 down(&audit_netlink_sem);
462 for (qlen = skb_queue_len(&sk->sk_receive_queue); qlen; qlen--) {
463 skb = skb_dequeue(&sk->sk_receive_queue);
464 audit_receive_skb(skb);
467 up(&audit_netlink_sem);
470 /* Move data from tmp buffer into an skb. This is an extra copy, and
471 * that is unfortunate. However, the copy will only occur when a record
472 * is being written to user space, which is already a high-overhead
473 * operation. (Elimination of the copy is possible, for example, by
474 * writing directly into a pre-allocated skb, at the cost of wasting
476 static void audit_log_move(struct audit_buffer *ab)
480 int extra = ab->nlh ? 0 : NLMSG_SPACE(0);
482 /* possible resubmission */
486 skb = skb_peek_tail(&ab->sklist);
487 if (!skb || skb_tailroom(skb) <= ab->len + extra) {
488 skb = alloc_skb(2 * ab->len + extra, GFP_ATOMIC);
490 ab->len = 0; /* Lose information in ab->tmp */
491 audit_log_lost("out of memory in audit_log_move");
494 __skb_queue_tail(&ab->sklist, skb);
496 ab->nlh = (struct nlmsghdr *)skb_put(skb,
499 start = skb_put(skb, ab->len);
500 memcpy(start, ab->tmp, ab->len);
504 /* Iterate over the skbuff in the audit_buffer, sending their contents
506 static inline int audit_log_drain(struct audit_buffer *ab)
510 while ((skb = skb_dequeue(&ab->sklist))) {
515 ab->nlh->nlmsg_len = ab->total;
516 ab->nlh->nlmsg_type = ab->type;
517 ab->nlh->nlmsg_flags = 0;
518 ab->nlh->nlmsg_seq = 0;
519 ab->nlh->nlmsg_pid = ab->pid;
521 skb_get(skb); /* because netlink_* frees */
522 retval = netlink_unicast(audit_sock, skb, audit_pid,
525 if (retval == -EAGAIN &&
526 (atomic_read(&audit_backlog)) < audit_backlog_limit) {
527 skb_queue_head(&ab->sklist, skb);
528 audit_log_end_irq(ab);
532 if (retval == -ECONNREFUSED) {
534 "audit: *NO* daemon at audit_pid=%d\n",
538 audit_log_lost("netlink socket too busy");
540 if (!audit_pid) { /* No daemon */
541 int offset = ab->nlh ? NLMSG_SPACE(0) : 0;
542 int len = skb->len - offset;
543 skb->data[offset + len] = '\0';
544 printk(KERN_ERR "%s\n", skb->data + offset);
552 /* Initialize audit support at boot time. */
553 static int __init audit_init(void)
555 printk(KERN_INFO "audit: initializing netlink socket (%s)\n",
556 audit_default ? "enabled" : "disabled");
557 audit_sock = netlink_kernel_create(NETLINK_AUDIT, audit_receive);
559 audit_panic("cannot initialize netlink socket");
561 audit_initialized = 1;
562 audit_enabled = audit_default;
563 audit_log(NULL, "initialized");
568 /* Without CONFIG_NET, we have no skbuffs. For now, print what we have
570 static void audit_log_move(struct audit_buffer *ab)
572 printk(KERN_ERR "%*.*s\n", ab->len, ab->len, ab->tmp);
576 static inline int audit_log_drain(struct audit_buffer *ab)
581 /* Initialize audit support at boot time. */
582 int __init audit_init(void)
584 printk(KERN_INFO "audit: initializing WITHOUT netlink support\n");
588 audit_initialized = 1;
589 audit_enabled = audit_default;
590 audit_log(NULL, "initialized");
595 __initcall(audit_init);
597 /* Process kernel command-line parameter at boot time. audit=0 or audit=1. */
598 static int __init audit_enable(char *str)
600 audit_default = !!simple_strtol(str, NULL, 0);
601 printk(KERN_INFO "audit: %s%s\n",
602 audit_default ? "enabled" : "disabled",
603 audit_initialized ? "" : " (after initialization)");
604 if (audit_initialized)
605 audit_enabled = audit_default;
609 __setup("audit=", audit_enable);
612 /* Obtain an audit buffer. This routine does locking to obtain the
613 * audit buffer, but then no locking is required for calls to
614 * audit_log_*format. If the tsk is a task that is currently in a
615 * syscall, then the syscall is marked as auditable and an audit record
616 * will be written at syscall exit. If there is no associated task, tsk
618 struct audit_buffer *audit_log_start(struct audit_context *ctx)
620 struct audit_buffer *ab = NULL;
625 if (!audit_initialized)
628 if (audit_backlog_limit
629 && atomic_read(&audit_backlog) > audit_backlog_limit) {
630 if (audit_rate_check())
632 "audit: audit_backlog=%d > "
633 "audit_backlog_limit=%d\n",
634 atomic_read(&audit_backlog),
635 audit_backlog_limit);
636 audit_log_lost("backlog limit exceeded");
640 spin_lock_irqsave(&audit_freelist_lock, flags);
641 if (!list_empty(&audit_freelist)) {
642 ab = list_entry(audit_freelist.next,
643 struct audit_buffer, list);
645 --audit_freelist_count;
647 spin_unlock_irqrestore(&audit_freelist_lock, flags);
650 ab = kmalloc(sizeof(*ab), GFP_ATOMIC);
652 audit_log_lost("out of memory in audit_log_start");
656 atomic_inc(&audit_backlog);
657 skb_queue_head_init(&ab->sklist);
663 ab->type = AUDIT_KERNEL;
666 #ifdef CONFIG_AUDITSYSCALL
668 audit_get_stamp(ab->ctx, &t, &serial);
675 audit_log_format(ab, "audit(%lu.%03lu:%u): ",
676 t.tv_sec, t.tv_nsec/1000000, serial);
681 /* Format an audit message into the audit buffer. If there isn't enough
682 * room in the audit buffer, more room will be allocated and vsnprint
683 * will be called a second time. Currently, we assume that a printk
684 * can't format message larger than 1024 bytes, so we don't either. */
685 static void audit_log_vformat(struct audit_buffer *ab, const char *fmt,
693 avail = sizeof(ab->tmp) - ab->len;
696 avail = sizeof(ab->tmp) - ab->len;
698 len = vsnprintf(ab->tmp + ab->len, avail, fmt, args);
700 /* The printk buffer is 1024 bytes long, so if we get
701 * here and AUDIT_BUFSIZ is at least 1024, then we can
702 * log everything that printk could have logged. */
704 avail = sizeof(ab->tmp) - ab->len;
705 len = vsnprintf(ab->tmp + ab->len, avail, fmt, args);
707 ab->len += (len < avail) ? len : avail;
708 ab->total += (len < avail) ? len : avail;
711 /* Format a message into the audit buffer. All the work is done in
712 * audit_log_vformat. */
713 void audit_log_format(struct audit_buffer *ab, const char *fmt, ...)
720 audit_log_vformat(ab, fmt, args);
724 void audit_log_hex(struct audit_buffer *ab, const unsigned char *buf, size_t len)
728 for (i=0; i<len; i++)
729 audit_log_format(ab, "%02x", buf[i]);
732 void audit_log_untrustedstring(struct audit_buffer *ab, const char *string)
734 const unsigned char *p = string;
737 if (*p == '"' || *p == ' ' || *p < 0x20 || *p > 0x7f) {
738 audit_log_hex(ab, string, strlen(string));
743 audit_log_format(ab, "\"%s\"", string);
747 /* This is a helper-function to print the d_path without using a static
748 * buffer or allocating another buffer in addition to the one in
750 void audit_log_d_path(struct audit_buffer *ab, const char *prefix,
751 struct dentry *dentry, struct vfsmount *vfsmnt)
756 if (prefix) audit_log_format(ab, " %s", prefix);
760 avail = sizeof(ab->tmp) - ab->len;
761 p = d_path(dentry, vfsmnt, ab->tmp + ab->len, avail);
763 /* FIXME: can we save some information here? */
764 audit_log_format(ab, "<toolong>");
766 /* path isn't at start of buffer */
767 len = (ab->tmp + sizeof(ab->tmp) - 1) - p;
768 memmove(ab->tmp + ab->len, p, len);
774 /* Remove queued messages from the audit_txlist and send them to userspace. */
775 static void audit_tasklet_handler(unsigned long arg)
778 struct audit_buffer *ab;
781 spin_lock_irqsave(&audit_txlist_lock, flags);
782 list_splice_init(&audit_txlist, &list);
783 spin_unlock_irqrestore(&audit_txlist_lock, flags);
785 while (!list_empty(&list)) {
786 ab = list_entry(list.next, struct audit_buffer, list);
788 audit_log_end_fast(ab);
792 static DECLARE_TASKLET(audit_tasklet, audit_tasklet_handler, 0);
794 /* The netlink_* functions cannot be called inside an irq context, so
795 * the audit buffer is places on a queue and a tasklet is scheduled to
796 * remove them from the queue outside the irq context. May be called in
798 static void audit_log_end_irq(struct audit_buffer *ab)
804 spin_lock_irqsave(&audit_txlist_lock, flags);
805 list_add_tail(&ab->list, &audit_txlist);
806 spin_unlock_irqrestore(&audit_txlist_lock, flags);
808 tasklet_schedule(&audit_tasklet);
811 /* Send the message in the audit buffer directly to user space. May not
812 * be called in an irq context. */
813 static void audit_log_end_fast(struct audit_buffer *ab)
820 if (!audit_rate_check()) {
821 audit_log_lost("rate limit exceeded");
824 if (audit_log_drain(ab))
828 atomic_dec(&audit_backlog);
829 spin_lock_irqsave(&audit_freelist_lock, flags);
830 if (++audit_freelist_count > AUDIT_MAXFREE)
833 list_add(&ab->list, &audit_freelist);
834 spin_unlock_irqrestore(&audit_freelist_lock, flags);
837 /* Send or queue the message in the audit buffer, depending on the
838 * current context. (A convenience function that may be called in any
840 void audit_log_end(struct audit_buffer *ab)
843 audit_log_end_irq(ab);
845 audit_log_end_fast(ab);
848 /* Log an audit record. This is a convenience function that calls
849 * audit_log_start, audit_log_vformat, and audit_log_end. It may be
850 * called in any context. */
851 void audit_log(struct audit_context *ctx, const char *fmt, ...)
853 struct audit_buffer *ab;
856 ab = audit_log_start(ctx);
859 audit_log_vformat(ab, fmt, args);