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 *skb; /* formatted skb ready to send */
140 struct audit_context *ctx; /* NULL or associated context */
143 void audit_set_type(struct audit_buffer *ab, int type)
145 struct nlmsghdr *nlh = (struct nlmsghdr *)ab->skb->data;
146 nlh->nlmsg_type = type;
149 static void audit_set_pid(struct audit_buffer *ab, pid_t pid)
151 struct nlmsghdr *nlh = (struct nlmsghdr *)ab->skb->data;
152 nlh->nlmsg_pid = pid;
156 struct list_head list;
157 struct audit_rule rule;
160 static void audit_log_end_irq(struct audit_buffer *ab);
161 static void audit_log_end_fast(struct audit_buffer *ab);
163 static void audit_panic(const char *message)
165 switch (audit_failure)
167 case AUDIT_FAIL_SILENT:
169 case AUDIT_FAIL_PRINTK:
170 printk(KERN_ERR "audit: %s\n", message);
172 case AUDIT_FAIL_PANIC:
173 panic("audit: %s\n", message);
178 static inline int audit_rate_check(void)
180 static unsigned long last_check = 0;
181 static int messages = 0;
182 static DEFINE_SPINLOCK(lock);
185 unsigned long elapsed;
188 if (!audit_rate_limit) return 1;
190 spin_lock_irqsave(&lock, flags);
191 if (++messages < audit_rate_limit) {
195 elapsed = now - last_check;
202 spin_unlock_irqrestore(&lock, flags);
207 /* Emit at least 1 message per second, even if audit_rate_check is
209 void audit_log_lost(const char *message)
211 static unsigned long last_msg = 0;
212 static DEFINE_SPINLOCK(lock);
217 atomic_inc(&audit_lost);
219 print = (audit_failure == AUDIT_FAIL_PANIC || !audit_rate_limit);
222 spin_lock_irqsave(&lock, flags);
224 if (now - last_msg > HZ) {
228 spin_unlock_irqrestore(&lock, flags);
233 "audit: audit_lost=%d audit_backlog=%d"
234 " audit_rate_limit=%d audit_backlog_limit=%d\n",
235 atomic_read(&audit_lost),
236 atomic_read(&audit_backlog),
238 audit_backlog_limit);
239 audit_panic(message);
244 static int audit_set_rate_limit(int limit, uid_t loginuid)
246 int old = audit_rate_limit;
247 audit_rate_limit = limit;
248 audit_log(NULL, "audit_rate_limit=%d old=%d by auid %u",
249 audit_rate_limit, old, loginuid);
253 static int audit_set_backlog_limit(int limit, uid_t loginuid)
255 int old = audit_backlog_limit;
256 audit_backlog_limit = limit;
257 audit_log(NULL, "audit_backlog_limit=%d old=%d by auid %u",
258 audit_backlog_limit, old, loginuid);
262 static int audit_set_enabled(int state, uid_t loginuid)
264 int old = audit_enabled;
265 if (state != 0 && state != 1)
267 audit_enabled = state;
268 audit_log(NULL, "audit_enabled=%d old=%d by auid %u",
269 audit_enabled, old, loginuid);
273 static int audit_set_failure(int state, uid_t loginuid)
275 int old = audit_failure;
276 if (state != AUDIT_FAIL_SILENT
277 && state != AUDIT_FAIL_PRINTK
278 && state != AUDIT_FAIL_PANIC)
280 audit_failure = state;
281 audit_log(NULL, "audit_failure=%d old=%d by auid %u",
282 audit_failure, old, loginuid);
286 void audit_send_reply(int pid, int seq, int type, int done, int multi,
287 void *payload, int size)
290 struct nlmsghdr *nlh;
291 int len = NLMSG_SPACE(size);
293 int flags = multi ? NLM_F_MULTI : 0;
294 int t = done ? NLMSG_DONE : type;
296 skb = alloc_skb(len, GFP_KERNEL);
300 nlh = NLMSG_PUT(skb, pid, seq, t, len - sizeof(*nlh));
301 nlh->nlmsg_flags = flags;
302 data = NLMSG_DATA(nlh);
303 memcpy(data, payload, size);
304 netlink_unicast(audit_sock, skb, pid, MSG_DONTWAIT);
307 nlmsg_failure: /* Used by NLMSG_PUT */
313 * Check for appropriate CAP_AUDIT_ capabilities on incoming audit
316 static int audit_netlink_ok(kernel_cap_t eff_cap, u16 msg_type)
326 case AUDIT_SIGNAL_INFO:
327 if (!cap_raised(eff_cap, CAP_AUDIT_CONTROL))
331 if (!cap_raised(eff_cap, CAP_AUDIT_WRITE))
334 default: /* bad msg */
341 static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
345 struct audit_status *status_get, status_set;
347 struct audit_buffer *ab;
348 u16 msg_type = nlh->nlmsg_type;
349 uid_t loginuid; /* loginuid of sender */
350 struct audit_sig_info sig_data;
352 err = audit_netlink_ok(NETLINK_CB(skb).eff_cap, msg_type);
356 pid = NETLINK_CREDS(skb)->pid;
357 uid = NETLINK_CREDS(skb)->uid;
358 loginuid = NETLINK_CB(skb).loginuid;
359 seq = nlh->nlmsg_seq;
360 data = NLMSG_DATA(nlh);
364 status_set.enabled = audit_enabled;
365 status_set.failure = audit_failure;
366 status_set.pid = audit_pid;
367 status_set.rate_limit = audit_rate_limit;
368 status_set.backlog_limit = audit_backlog_limit;
369 status_set.lost = atomic_read(&audit_lost);
370 status_set.backlog = atomic_read(&audit_backlog);
371 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_GET, 0, 0,
372 &status_set, sizeof(status_set));
375 if (nlh->nlmsg_len < sizeof(struct audit_status))
377 status_get = (struct audit_status *)data;
378 if (status_get->mask & AUDIT_STATUS_ENABLED) {
379 err = audit_set_enabled(status_get->enabled, loginuid);
380 if (err < 0) return err;
382 if (status_get->mask & AUDIT_STATUS_FAILURE) {
383 err = audit_set_failure(status_get->failure, loginuid);
384 if (err < 0) return err;
386 if (status_get->mask & AUDIT_STATUS_PID) {
388 audit_pid = status_get->pid;
389 audit_log(NULL, "audit_pid=%d old=%d by auid %u",
390 audit_pid, old, loginuid);
392 if (status_get->mask & AUDIT_STATUS_RATE_LIMIT)
393 audit_set_rate_limit(status_get->rate_limit, loginuid);
394 if (status_get->mask & AUDIT_STATUS_BACKLOG_LIMIT)
395 audit_set_backlog_limit(status_get->backlog_limit,
399 ab = audit_log_start(NULL);
401 break; /* audit_panic has been called */
403 "user pid=%d uid=%d length=%d loginuid=%u"
407 - ((char *)data - (char *)nlh)),
408 loginuid, (char *)data);
409 audit_set_type(ab, AUDIT_USER);
410 audit_set_pid(ab, pid);
415 if (nlh->nlmsg_len < sizeof(struct audit_rule))
419 err = audit_receive_filter(nlh->nlmsg_type, NETLINK_CB(skb).pid,
420 uid, seq, data, loginuid);
422 case AUDIT_SIGNAL_INFO:
423 sig_data.uid = audit_sig_uid;
424 sig_data.pid = audit_sig_pid;
425 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_SIGNAL_INFO,
426 0, 0, &sig_data, sizeof(sig_data));
433 return err < 0 ? err : 0;
436 /* Get message from skb (based on rtnetlink_rcv_skb). Each message is
437 * processed by audit_receive_msg. Malformed skbs with wrong length are
438 * discarded silently. */
439 static void audit_receive_skb(struct sk_buff *skb)
442 struct nlmsghdr *nlh;
445 while (skb->len >= NLMSG_SPACE(0)) {
446 nlh = (struct nlmsghdr *)skb->data;
447 if (nlh->nlmsg_len < sizeof(*nlh) || skb->len < nlh->nlmsg_len)
449 rlen = NLMSG_ALIGN(nlh->nlmsg_len);
452 if ((err = audit_receive_msg(skb, nlh))) {
453 netlink_ack(skb, nlh, err);
454 } else if (nlh->nlmsg_flags & NLM_F_ACK)
455 netlink_ack(skb, nlh, 0);
460 /* Receive messages from netlink socket. */
461 static void audit_receive(struct sock *sk, int length)
466 down(&audit_netlink_sem);
468 for (qlen = skb_queue_len(&sk->sk_receive_queue); qlen; qlen--) {
469 skb = skb_dequeue(&sk->sk_receive_queue);
470 audit_receive_skb(skb);
473 up(&audit_netlink_sem);
476 /* Grab skbuff from the audit_buffer and send to user space. */
477 static inline int audit_log_drain(struct audit_buffer *ab)
479 struct sk_buff *skb = ab->skb;
485 struct nlmsghdr *nlh = (struct nlmsghdr *)skb->data;
486 nlh->nlmsg_len = skb->len - NLMSG_SPACE(0);
487 skb_get(skb); /* because netlink_* frees */
488 retval = netlink_unicast(audit_sock, skb, audit_pid,
491 if (retval == -EAGAIN &&
492 (atomic_read(&audit_backlog)) < audit_backlog_limit) {
493 audit_log_end_irq(ab);
497 if (retval == -ECONNREFUSED) {
499 "audit: *NO* daemon at audit_pid=%d\n",
503 audit_log_lost("netlink socket too busy");
505 if (!audit_pid) { /* No daemon */
506 int offset = NLMSG_SPACE(0);
507 int len = skb->len - offset;
508 skb->data[offset + len] = '\0';
509 printk(KERN_ERR "%s\n", skb->data + offset);
515 /* Initialize audit support at boot time. */
516 static int __init audit_init(void)
518 printk(KERN_INFO "audit: initializing netlink socket (%s)\n",
519 audit_default ? "enabled" : "disabled");
520 audit_sock = netlink_kernel_create(NETLINK_AUDIT, audit_receive);
522 audit_panic("cannot initialize netlink socket");
524 audit_initialized = 1;
525 audit_enabled = audit_default;
526 audit_log(NULL, "initialized");
529 __initcall(audit_init);
531 /* Process kernel command-line parameter at boot time. audit=0 or audit=1. */
532 static int __init audit_enable(char *str)
534 audit_default = !!simple_strtol(str, NULL, 0);
535 printk(KERN_INFO "audit: %s%s\n",
536 audit_default ? "enabled" : "disabled",
537 audit_initialized ? "" : " (after initialization)");
538 if (audit_initialized)
539 audit_enabled = audit_default;
543 __setup("audit=", audit_enable);
545 static void audit_buffer_free(struct audit_buffer *ab)
554 atomic_dec(&audit_backlog);
555 spin_lock_irqsave(&audit_freelist_lock, flags);
556 if (++audit_freelist_count > AUDIT_MAXFREE)
559 list_add(&ab->list, &audit_freelist);
560 spin_unlock_irqrestore(&audit_freelist_lock, flags);
563 static struct audit_buffer * audit_buffer_alloc(struct audit_context *ctx,
567 struct audit_buffer *ab = NULL;
568 struct nlmsghdr *nlh;
570 spin_lock_irqsave(&audit_freelist_lock, flags);
571 if (!list_empty(&audit_freelist)) {
572 ab = list_entry(audit_freelist.next,
573 struct audit_buffer, list);
575 --audit_freelist_count;
577 spin_unlock_irqrestore(&audit_freelist_lock, flags);
580 ab = kmalloc(sizeof(*ab), gfp_mask);
584 atomic_inc(&audit_backlog);
586 ab->skb = alloc_skb(AUDIT_BUFSIZ, gfp_mask);
591 nlh = (struct nlmsghdr *)skb_put(ab->skb, NLMSG_SPACE(0));
592 nlh->nlmsg_type = AUDIT_KERNEL;
593 nlh->nlmsg_flags = 0;
598 audit_buffer_free(ab);
602 /* Obtain an audit buffer. This routine does locking to obtain the
603 * audit buffer, but then no locking is required for calls to
604 * audit_log_*format. If the tsk is a task that is currently in a
605 * syscall, then the syscall is marked as auditable and an audit record
606 * will be written at syscall exit. If there is no associated task, tsk
608 struct audit_buffer *audit_log_start(struct audit_context *ctx)
610 struct audit_buffer *ab = NULL;
614 if (!audit_initialized)
617 if (audit_backlog_limit
618 && atomic_read(&audit_backlog) > audit_backlog_limit) {
619 if (audit_rate_check())
621 "audit: audit_backlog=%d > "
622 "audit_backlog_limit=%d\n",
623 atomic_read(&audit_backlog),
624 audit_backlog_limit);
625 audit_log_lost("backlog limit exceeded");
629 ab = audit_buffer_alloc(ctx, GFP_ATOMIC);
631 audit_log_lost("out of memory in audit_log_start");
635 if (!audit_get_stamp(ab->ctx, &t, &serial)) {
640 audit_log_format(ab, "audit(%lu.%03lu:%u): ",
641 t.tv_sec, t.tv_nsec/1000000, serial);
646 * audit_expand - expand skb in the audit buffer
649 * Returns 0 (no space) on failed expansion, or available space if
652 static inline int audit_expand(struct audit_buffer *ab, int extra)
654 struct sk_buff *skb = ab->skb;
655 int ret = pskb_expand_head(skb, skb_headroom(skb), extra,
658 audit_log_lost("out of memory in audit_expand");
661 return skb_tailroom(skb);
664 /* Format an audit message into the audit buffer. If there isn't enough
665 * room in the audit buffer, more room will be allocated and vsnprint
666 * will be called a second time. Currently, we assume that a printk
667 * can't format message larger than 1024 bytes, so we don't either. */
668 static void audit_log_vformat(struct audit_buffer *ab, const char *fmt,
680 avail = skb_tailroom(skb);
682 avail = audit_expand(ab, AUDIT_BUFSIZ);
686 va_copy(args2, args);
687 len = vsnprintf(skb->tail, avail, fmt, args);
689 /* The printk buffer is 1024 bytes long, so if we get
690 * here and AUDIT_BUFSIZ is at least 1024, then we can
691 * log everything that printk could have logged. */
692 avail = audit_expand(ab, 1+len-avail);
695 len = vsnprintf(skb->tail, avail, fmt, args2);
697 skb_put(skb, (len < avail) ? len : avail);
702 /* Format a message into the audit buffer. All the work is done in
703 * audit_log_vformat. */
704 void audit_log_format(struct audit_buffer *ab, const char *fmt, ...)
711 audit_log_vformat(ab, fmt, args);
715 void audit_log_hex(struct audit_buffer *ab, const unsigned char *buf, size_t len)
719 for (i=0; i<len; i++)
720 audit_log_format(ab, "%02x", buf[i]);
723 void audit_log_untrustedstring(struct audit_buffer *ab, const char *string)
725 const unsigned char *p = string;
728 if (*p == '"' || *p == ' ' || *p < 0x20 || *p > 0x7f) {
729 audit_log_hex(ab, string, strlen(string));
734 audit_log_format(ab, "\"%s\"", string);
738 /* This is a helper-function to print the d_path without using a static
739 * buffer or allocating another buffer in addition to the one in
741 void audit_log_d_path(struct audit_buffer *ab, const char *prefix,
742 struct dentry *dentry, struct vfsmount *vfsmnt)
745 struct sk_buff *skb = ab->skb;
749 audit_log_format(ab, " %s", prefix);
751 avail = skb_tailroom(skb);
752 p = d_path(dentry, vfsmnt, skb->tail, avail);
754 /* FIXME: can we save some information here? */
755 audit_log_format(ab, "<toolong>");
757 /* path isn't at start of buffer */
758 len = ((char *)skb->tail + avail - 1) - p;
759 memmove(skb->tail, p, len);
764 /* Remove queued messages from the audit_txlist and send them to userspace. */
765 static void audit_tasklet_handler(unsigned long arg)
768 struct audit_buffer *ab;
771 spin_lock_irqsave(&audit_txlist_lock, flags);
772 list_splice_init(&audit_txlist, &list);
773 spin_unlock_irqrestore(&audit_txlist_lock, flags);
775 while (!list_empty(&list)) {
776 ab = list_entry(list.next, struct audit_buffer, list);
778 audit_log_end_fast(ab);
782 static DECLARE_TASKLET(audit_tasklet, audit_tasklet_handler, 0);
784 /* The netlink_* functions cannot be called inside an irq context, so
785 * the audit buffer is places on a queue and a tasklet is scheduled to
786 * remove them from the queue outside the irq context. May be called in
788 static void audit_log_end_irq(struct audit_buffer *ab)
794 spin_lock_irqsave(&audit_txlist_lock, flags);
795 list_add_tail(&ab->list, &audit_txlist);
796 spin_unlock_irqrestore(&audit_txlist_lock, flags);
798 tasklet_schedule(&audit_tasklet);
801 /* Send the message in the audit buffer directly to user space. May not
802 * be called in an irq context. */
803 static void audit_log_end_fast(struct audit_buffer *ab)
808 if (!audit_rate_check()) {
809 audit_log_lost("rate limit exceeded");
811 if (audit_log_drain(ab))
814 audit_buffer_free(ab);
817 /* Send or queue the message in the audit buffer, depending on the
818 * current context. (A convenience function that may be called in any
820 void audit_log_end(struct audit_buffer *ab)
823 audit_log_end_irq(ab);
825 audit_log_end_fast(ab);
828 /* Log an audit record. This is a convenience function that calls
829 * audit_log_start, audit_log_vformat, and audit_log_end. It may be
830 * called in any context. */
831 void audit_log(struct audit_context *ctx, const char *fmt, ...)
833 struct audit_buffer *ab;
836 ab = audit_log_start(ctx);
839 audit_log_vformat(ab, fmt, args);