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 /* The identity of the user shutting down the audit system. */
83 uid_t audit_sig_uid = -1;
84 pid_t audit_sig_pid = -1;
86 /* Records can be lost in several ways:
87 0) [suppressed in audit_alloc]
88 1) out of memory in audit_log_start [kmalloc of struct audit_buffer]
89 2) out of memory in audit_log_move [alloc_skb]
90 3) suppressed due to audit_rate_limit
91 4) suppressed due to audit_backlog_limit
93 static atomic_t audit_lost = ATOMIC_INIT(0);
95 /* The netlink socket. */
96 static struct sock *audit_sock;
98 /* There are two lists of audit buffers. The txlist contains audit
99 * buffers that cannot be sent immediately to the netlink device because
100 * we are in an irq context (these are sent later in a tasklet).
102 * The second list is a list of pre-allocated audit buffers (if more
103 * than AUDIT_MAXFREE are in use, the audit buffer is freed instead of
104 * being placed on the freelist). */
105 static DEFINE_SPINLOCK(audit_txlist_lock);
106 static DEFINE_SPINLOCK(audit_freelist_lock);
107 static int audit_freelist_count = 0;
108 static LIST_HEAD(audit_txlist);
109 static LIST_HEAD(audit_freelist);
111 /* There are three lists of rules -- one to search at task creation
112 * time, one to search at syscall entry time, and another to search at
113 * syscall exit time. */
114 static LIST_HEAD(audit_tsklist);
115 static LIST_HEAD(audit_entlist);
116 static LIST_HEAD(audit_extlist);
118 /* The netlink socket is only to be read by 1 CPU, which lets us assume
119 * that list additions and deletions never happen simultaneiously in
121 static DECLARE_MUTEX(audit_netlink_sem);
123 /* AUDIT_BUFSIZ is the size of the temporary buffer used for formatting
124 * audit records. Since printk uses a 1024 byte buffer, this buffer
125 * should be at least that large. */
126 #define AUDIT_BUFSIZ 1024
128 /* AUDIT_MAXFREE is the number of empty audit_buffers we keep on the
129 * audit_freelist. Doing so eliminates many kmalloc/kfree calls. */
130 #define AUDIT_MAXFREE (2*NR_CPUS)
132 /* The audit_buffer is used when formatting an audit record. The caller
133 * locks briefly to get the record off the freelist or to allocate the
134 * buffer, and locks briefly to send the buffer to the netlink layer or
135 * to place it on a transmit queue. Multiple audit_buffers can be in
136 * use simultaneously. */
137 struct audit_buffer {
138 struct list_head list;
139 struct sk_buff_head sklist; /* formatted skbs ready to send */
140 struct audit_context *ctx; /* NULL or associated context */
141 int len; /* used area of tmp */
142 char tmp[AUDIT_BUFSIZ];
144 /* Pointer to header and contents */
145 struct nlmsghdr *nlh;
151 void audit_set_type(struct audit_buffer *ab, int type)
157 struct list_head list;
158 struct audit_rule rule;
161 static void audit_log_end_irq(struct audit_buffer *ab);
162 static void audit_log_end_fast(struct audit_buffer *ab);
164 static void audit_panic(const char *message)
166 switch (audit_failure)
168 case AUDIT_FAIL_SILENT:
170 case AUDIT_FAIL_PRINTK:
171 printk(KERN_ERR "audit: %s\n", message);
173 case AUDIT_FAIL_PANIC:
174 panic("audit: %s\n", message);
179 static inline int audit_rate_check(void)
181 static unsigned long last_check = 0;
182 static int messages = 0;
183 static DEFINE_SPINLOCK(lock);
186 unsigned long elapsed;
189 if (!audit_rate_limit) return 1;
191 spin_lock_irqsave(&lock, flags);
192 if (++messages < audit_rate_limit) {
196 elapsed = now - last_check;
203 spin_unlock_irqrestore(&lock, flags);
208 /* Emit at least 1 message per second, even if audit_rate_check is
210 void audit_log_lost(const char *message)
212 static unsigned long last_msg = 0;
213 static DEFINE_SPINLOCK(lock);
218 atomic_inc(&audit_lost);
220 print = (audit_failure == AUDIT_FAIL_PANIC || !audit_rate_limit);
223 spin_lock_irqsave(&lock, flags);
225 if (now - last_msg > HZ) {
229 spin_unlock_irqrestore(&lock, flags);
234 "audit: audit_lost=%d audit_backlog=%d"
235 " audit_rate_limit=%d audit_backlog_limit=%d\n",
236 atomic_read(&audit_lost),
237 atomic_read(&audit_backlog),
239 audit_backlog_limit);
240 audit_panic(message);
245 static int audit_set_rate_limit(int limit, uid_t loginuid)
247 int old = audit_rate_limit;
248 audit_rate_limit = limit;
249 audit_log(NULL, "audit_rate_limit=%d old=%d by auid %u",
250 audit_rate_limit, old, loginuid);
254 static int audit_set_backlog_limit(int limit, uid_t loginuid)
256 int old = audit_backlog_limit;
257 audit_backlog_limit = limit;
258 audit_log(NULL, "audit_backlog_limit=%d old=%d by auid %u",
259 audit_backlog_limit, old, loginuid);
263 static int audit_set_enabled(int state, uid_t loginuid)
265 int old = audit_enabled;
266 if (state != 0 && state != 1)
268 audit_enabled = state;
269 audit_log(NULL, "audit_enabled=%d old=%d by auid %u",
270 audit_enabled, old, loginuid);
274 static int audit_set_failure(int state, uid_t loginuid)
276 int old = audit_failure;
277 if (state != AUDIT_FAIL_SILENT
278 && state != AUDIT_FAIL_PRINTK
279 && state != AUDIT_FAIL_PANIC)
281 audit_failure = state;
282 audit_log(NULL, "audit_failure=%d old=%d by auid %u",
283 audit_failure, old, loginuid);
288 void audit_send_reply(int pid, int seq, int type, int done, int multi,
289 void *payload, int size)
292 struct nlmsghdr *nlh;
293 int len = NLMSG_SPACE(size);
295 int flags = multi ? NLM_F_MULTI : 0;
296 int t = done ? NLMSG_DONE : type;
298 skb = alloc_skb(len, GFP_KERNEL);
302 nlh = NLMSG_PUT(skb, pid, seq, t, len - sizeof(*nlh));
303 nlh->nlmsg_flags = flags;
304 data = NLMSG_DATA(nlh);
305 memcpy(data, payload, size);
306 netlink_unicast(audit_sock, skb, pid, MSG_DONTWAIT);
309 nlmsg_failure: /* Used by NLMSG_PUT */
315 * Check for appropriate CAP_AUDIT_ capabilities on incoming audit
318 static int audit_netlink_ok(kernel_cap_t eff_cap, u16 msg_type)
328 case AUDIT_SIGNAL_INFO:
329 if (!cap_raised(eff_cap, CAP_AUDIT_CONTROL))
333 if (!cap_raised(eff_cap, CAP_AUDIT_WRITE))
336 default: /* bad msg */
343 static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
347 struct audit_status *status_get, status_set;
349 struct audit_buffer *ab;
350 u16 msg_type = nlh->nlmsg_type;
351 uid_t loginuid; /* loginuid of sender */
352 struct audit_sig_info sig_data;
354 err = audit_netlink_ok(NETLINK_CB(skb).eff_cap, msg_type);
358 pid = NETLINK_CREDS(skb)->pid;
359 uid = NETLINK_CREDS(skb)->uid;
360 loginuid = NETLINK_CB(skb).loginuid;
361 seq = nlh->nlmsg_seq;
362 data = NLMSG_DATA(nlh);
366 status_set.enabled = audit_enabled;
367 status_set.failure = audit_failure;
368 status_set.pid = audit_pid;
369 status_set.rate_limit = audit_rate_limit;
370 status_set.backlog_limit = audit_backlog_limit;
371 status_set.lost = atomic_read(&audit_lost);
372 status_set.backlog = atomic_read(&audit_backlog);
373 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_GET, 0, 0,
374 &status_set, sizeof(status_set));
377 if (nlh->nlmsg_len < sizeof(struct audit_status))
379 status_get = (struct audit_status *)data;
380 if (status_get->mask & AUDIT_STATUS_ENABLED) {
381 err = audit_set_enabled(status_get->enabled, loginuid);
382 if (err < 0) return err;
384 if (status_get->mask & AUDIT_STATUS_FAILURE) {
385 err = audit_set_failure(status_get->failure, loginuid);
386 if (err < 0) return err;
388 if (status_get->mask & AUDIT_STATUS_PID) {
390 audit_pid = status_get->pid;
391 audit_log(NULL, "audit_pid=%d old=%d by auid %u",
392 audit_pid, old, loginuid);
394 if (status_get->mask & AUDIT_STATUS_RATE_LIMIT)
395 audit_set_rate_limit(status_get->rate_limit, loginuid);
396 if (status_get->mask & AUDIT_STATUS_BACKLOG_LIMIT)
397 audit_set_backlog_limit(status_get->backlog_limit,
401 ab = audit_log_start(NULL);
403 break; /* audit_panic has been called */
405 "user pid=%d uid=%d length=%d loginuid=%u"
409 - ((char *)data - (char *)nlh)),
410 loginuid, (char *)data);
411 ab->type = AUDIT_USER;
417 if (nlh->nlmsg_len < sizeof(struct audit_rule))
421 #ifdef CONFIG_AUDITSYSCALL
422 err = audit_receive_filter(nlh->nlmsg_type, NETLINK_CB(skb).pid,
423 uid, seq, data, loginuid);
428 case AUDIT_SIGNAL_INFO:
429 sig_data.uid = audit_sig_uid;
430 sig_data.pid = audit_sig_pid;
431 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_SIGNAL_INFO,
432 0, 0, &sig_data, sizeof(sig_data));
439 return err < 0 ? err : 0;
442 /* Get message from skb (based on rtnetlink_rcv_skb). Each message is
443 * processed by audit_receive_msg. Malformed skbs with wrong length are
444 * discarded silently. */
445 static void audit_receive_skb(struct sk_buff *skb)
448 struct nlmsghdr *nlh;
451 while (skb->len >= NLMSG_SPACE(0)) {
452 nlh = (struct nlmsghdr *)skb->data;
453 if (nlh->nlmsg_len < sizeof(*nlh) || skb->len < nlh->nlmsg_len)
455 rlen = NLMSG_ALIGN(nlh->nlmsg_len);
458 if ((err = audit_receive_msg(skb, nlh))) {
459 netlink_ack(skb, nlh, err);
460 } else if (nlh->nlmsg_flags & NLM_F_ACK)
461 netlink_ack(skb, nlh, 0);
466 /* Receive messages from netlink socket. */
467 static void audit_receive(struct sock *sk, int length)
472 down(&audit_netlink_sem);
474 for (qlen = skb_queue_len(&sk->sk_receive_queue); qlen; qlen--) {
475 skb = skb_dequeue(&sk->sk_receive_queue);
476 audit_receive_skb(skb);
479 up(&audit_netlink_sem);
482 /* Move data from tmp buffer into an skb. This is an extra copy, and
483 * that is unfortunate. However, the copy will only occur when a record
484 * is being written to user space, which is already a high-overhead
485 * operation. (Elimination of the copy is possible, for example, by
486 * writing directly into a pre-allocated skb, at the cost of wasting
488 static void audit_log_move(struct audit_buffer *ab)
492 int extra = ab->nlh ? 0 : NLMSG_SPACE(0);
494 /* possible resubmission */
498 skb = skb_peek_tail(&ab->sklist);
499 if (!skb || skb_tailroom(skb) <= ab->len + extra) {
500 skb = alloc_skb(2 * ab->len + extra, GFP_ATOMIC);
502 ab->len = 0; /* Lose information in ab->tmp */
503 audit_log_lost("out of memory in audit_log_move");
506 __skb_queue_tail(&ab->sklist, skb);
508 ab->nlh = (struct nlmsghdr *)skb_put(skb,
511 start = skb_put(skb, ab->len);
512 memcpy(start, ab->tmp, ab->len);
516 /* Iterate over the skbuff in the audit_buffer, sending their contents
518 static inline int audit_log_drain(struct audit_buffer *ab)
522 while ((skb = skb_dequeue(&ab->sklist))) {
527 ab->nlh->nlmsg_len = ab->total;
528 ab->nlh->nlmsg_type = ab->type;
529 ab->nlh->nlmsg_flags = 0;
530 ab->nlh->nlmsg_seq = 0;
531 ab->nlh->nlmsg_pid = ab->pid;
533 skb_get(skb); /* because netlink_* frees */
534 retval = netlink_unicast(audit_sock, skb, audit_pid,
537 if (retval == -EAGAIN &&
538 (atomic_read(&audit_backlog)) < audit_backlog_limit) {
539 skb_queue_head(&ab->sklist, skb);
540 audit_log_end_irq(ab);
544 if (retval == -ECONNREFUSED) {
546 "audit: *NO* daemon at audit_pid=%d\n",
550 audit_log_lost("netlink socket too busy");
552 if (!audit_pid) { /* No daemon */
553 int offset = ab->nlh ? NLMSG_SPACE(0) : 0;
554 int len = skb->len - offset;
555 skb->data[offset + len] = '\0';
556 printk(KERN_ERR "%s\n", skb->data + offset);
564 /* Initialize audit support at boot time. */
565 static int __init audit_init(void)
567 printk(KERN_INFO "audit: initializing netlink socket (%s)\n",
568 audit_default ? "enabled" : "disabled");
569 audit_sock = netlink_kernel_create(NETLINK_AUDIT, audit_receive);
571 audit_panic("cannot initialize netlink socket");
573 audit_initialized = 1;
574 audit_enabled = audit_default;
575 audit_log(NULL, "initialized");
580 /* Without CONFIG_NET, we have no skbuffs. For now, print what we have
582 static void audit_log_move(struct audit_buffer *ab)
584 printk(KERN_ERR "%*.*s\n", ab->len, ab->len, ab->tmp);
588 static inline int audit_log_drain(struct audit_buffer *ab)
593 /* Initialize audit support at boot time. */
594 int __init audit_init(void)
596 printk(KERN_INFO "audit: initializing WITHOUT netlink support\n");
600 audit_initialized = 1;
601 audit_enabled = audit_default;
602 audit_log(NULL, "initialized");
607 __initcall(audit_init);
609 /* Process kernel command-line parameter at boot time. audit=0 or audit=1. */
610 static int __init audit_enable(char *str)
612 audit_default = !!simple_strtol(str, NULL, 0);
613 printk(KERN_INFO "audit: %s%s\n",
614 audit_default ? "enabled" : "disabled",
615 audit_initialized ? "" : " (after initialization)");
616 if (audit_initialized)
617 audit_enabled = audit_default;
621 __setup("audit=", audit_enable);
623 static void audit_buffer_free(struct audit_buffer *ab)
627 atomic_dec(&audit_backlog);
628 spin_lock_irqsave(&audit_freelist_lock, flags);
629 if (++audit_freelist_count > AUDIT_MAXFREE)
632 list_add(&ab->list, &audit_freelist);
633 spin_unlock_irqrestore(&audit_freelist_lock, flags);
636 static struct audit_buffer * audit_buffer_alloc(int gfp_mask)
639 struct audit_buffer *ab = NULL;
641 spin_lock_irqsave(&audit_freelist_lock, flags);
642 if (!list_empty(&audit_freelist)) {
643 ab = list_entry(audit_freelist.next,
644 struct audit_buffer, list);
646 --audit_freelist_count;
648 spin_unlock_irqrestore(&audit_freelist_lock, flags);
651 ab = kmalloc(sizeof(*ab), GFP_ATOMIC);
655 atomic_inc(&audit_backlog);
660 /* Obtain an audit buffer. This routine does locking to obtain the
661 * audit buffer, but then no locking is required for calls to
662 * audit_log_*format. If the tsk is a task that is currently in a
663 * syscall, then the syscall is marked as auditable and an audit record
664 * will be written at syscall exit. If there is no associated task, tsk
666 struct audit_buffer *audit_log_start(struct audit_context *ctx)
668 struct audit_buffer *ab = NULL;
672 if (!audit_initialized)
675 if (audit_backlog_limit
676 && atomic_read(&audit_backlog) > audit_backlog_limit) {
677 if (audit_rate_check())
679 "audit: audit_backlog=%d > "
680 "audit_backlog_limit=%d\n",
681 atomic_read(&audit_backlog),
682 audit_backlog_limit);
683 audit_log_lost("backlog limit exceeded");
687 ab = audit_buffer_alloc(GFP_ATOMIC);
689 audit_log_lost("out of memory in audit_log_start");
693 skb_queue_head_init(&ab->sklist);
699 ab->type = AUDIT_KERNEL;
702 #ifdef CONFIG_AUDITSYSCALL
704 audit_get_stamp(ab->ctx, &t, &serial);
711 audit_log_format(ab, "audit(%lu.%03lu:%u): ",
712 t.tv_sec, t.tv_nsec/1000000, serial);
717 /* Format an audit message into the audit buffer. If there isn't enough
718 * room in the audit buffer, more room will be allocated and vsnprint
719 * will be called a second time. Currently, we assume that a printk
720 * can't format message larger than 1024 bytes, so we don't either. */
721 static void audit_log_vformat(struct audit_buffer *ab, const char *fmt,
729 avail = sizeof(ab->tmp) - ab->len;
732 avail = sizeof(ab->tmp) - ab->len;
734 len = vsnprintf(ab->tmp + ab->len, avail, fmt, args);
736 /* The printk buffer is 1024 bytes long, so if we get
737 * here and AUDIT_BUFSIZ is at least 1024, then we can
738 * log everything that printk could have logged. */
740 avail = sizeof(ab->tmp) - ab->len;
741 len = vsnprintf(ab->tmp + ab->len, avail, fmt, args);
743 ab->len += (len < avail) ? len : avail;
744 ab->total += (len < avail) ? len : avail;
747 /* Format a message into the audit buffer. All the work is done in
748 * audit_log_vformat. */
749 void audit_log_format(struct audit_buffer *ab, const char *fmt, ...)
756 audit_log_vformat(ab, fmt, args);
760 void audit_log_hex(struct audit_buffer *ab, const unsigned char *buf, size_t len)
764 for (i=0; i<len; i++)
765 audit_log_format(ab, "%02x", buf[i]);
768 void audit_log_untrustedstring(struct audit_buffer *ab, const char *string)
770 const unsigned char *p = string;
773 if (*p == '"' || *p == ' ' || *p < 0x20 || *p > 0x7f) {
774 audit_log_hex(ab, string, strlen(string));
779 audit_log_format(ab, "\"%s\"", string);
783 /* This is a helper-function to print the d_path without using a static
784 * buffer or allocating another buffer in addition to the one in
786 void audit_log_d_path(struct audit_buffer *ab, const char *prefix,
787 struct dentry *dentry, struct vfsmount *vfsmnt)
792 if (prefix) audit_log_format(ab, " %s", prefix);
796 avail = sizeof(ab->tmp) - ab->len;
797 p = d_path(dentry, vfsmnt, ab->tmp + ab->len, avail);
799 /* FIXME: can we save some information here? */
800 audit_log_format(ab, "<toolong>");
802 /* path isn't at start of buffer */
803 len = (ab->tmp + sizeof(ab->tmp) - 1) - p;
804 memmove(ab->tmp + ab->len, p, len);
810 /* Remove queued messages from the audit_txlist and send them to userspace. */
811 static void audit_tasklet_handler(unsigned long arg)
814 struct audit_buffer *ab;
817 spin_lock_irqsave(&audit_txlist_lock, flags);
818 list_splice_init(&audit_txlist, &list);
819 spin_unlock_irqrestore(&audit_txlist_lock, flags);
821 while (!list_empty(&list)) {
822 ab = list_entry(list.next, struct audit_buffer, list);
824 audit_log_end_fast(ab);
828 static DECLARE_TASKLET(audit_tasklet, audit_tasklet_handler, 0);
830 /* The netlink_* functions cannot be called inside an irq context, so
831 * the audit buffer is places on a queue and a tasklet is scheduled to
832 * remove them from the queue outside the irq context. May be called in
834 static void audit_log_end_irq(struct audit_buffer *ab)
840 spin_lock_irqsave(&audit_txlist_lock, flags);
841 list_add_tail(&ab->list, &audit_txlist);
842 spin_unlock_irqrestore(&audit_txlist_lock, flags);
844 tasklet_schedule(&audit_tasklet);
847 /* Send the message in the audit buffer directly to user space. May not
848 * be called in an irq context. */
849 static void audit_log_end_fast(struct audit_buffer *ab)
854 if (!audit_rate_check()) {
855 audit_log_lost("rate limit exceeded");
858 if (audit_log_drain(ab))
861 audit_buffer_free(ab);
864 /* Send or queue the message in the audit buffer, depending on the
865 * current context. (A convenience function that may be called in any
867 void audit_log_end(struct audit_buffer *ab)
870 audit_log_end_irq(ab);
872 audit_log_end_fast(ab);
875 /* Log an audit record. This is a convenience function that calls
876 * audit_log_start, audit_log_vformat, and audit_log_end. It may be
877 * called in any context. */
878 void audit_log(struct audit_context *ctx, const char *fmt, ...)
880 struct audit_buffer *ab;
883 ab = audit_log_start(ctx);
886 audit_log_vformat(ab, fmt, args);