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>
49 #include <linux/err.h>
50 #include <linux/kthread.h>
52 #include <linux/audit.h>
55 #include <linux/skbuff.h>
56 #include <linux/netlink.h>
58 /* No auditing will take place until audit_initialized != 0.
59 * (Initialization happens after skb_init is called.) */
60 static int audit_initialized;
62 /* No syscall auditing will take place unless audit_enabled != 0. */
65 /* Default state when kernel boots without any parameters. */
66 static int audit_default;
68 /* If auditing cannot proceed, audit_failure selects what happens. */
69 static int audit_failure = AUDIT_FAIL_PRINTK;
71 /* If audit records are to be written to the netlink socket, audit_pid
72 * contains the (non-zero) pid. */
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;
80 /* Number of outstanding audit_buffers allowed. */
81 static int audit_backlog_limit = 64;
83 /* The identity of the user shutting down the audit system. */
84 uid_t audit_sig_uid = -1;
85 pid_t audit_sig_pid = -1;
87 /* Records can be lost in several ways:
88 0) [suppressed in audit_alloc]
89 1) out of memory in audit_log_start [kmalloc of struct audit_buffer]
90 2) out of memory in audit_log_move [alloc_skb]
91 3) suppressed due to audit_rate_limit
92 4) suppressed due to audit_backlog_limit
94 static atomic_t audit_lost = ATOMIC_INIT(0);
96 /* The netlink socket. */
97 static struct sock *audit_sock;
99 /* The audit_freelist is a list of pre-allocated audit buffers (if more
100 * than AUDIT_MAXFREE are in use, the audit buffer is freed instead of
101 * being placed on the freelist). */
102 static DEFINE_SPINLOCK(audit_freelist_lock);
103 static int audit_freelist_count = 0;
104 static LIST_HEAD(audit_freelist);
106 static struct sk_buff_head audit_skb_queue;
107 static struct task_struct *kauditd_task;
108 static DECLARE_WAIT_QUEUE_HEAD(kauditd_wait);
110 /* The netlink socket is only to be read by 1 CPU, which lets us assume
111 * that list additions and deletions never happen simultaneously in
113 static DECLARE_MUTEX(audit_netlink_sem);
115 /* AUDIT_BUFSIZ is the size of the temporary buffer used for formatting
116 * audit records. Since printk uses a 1024 byte buffer, this buffer
117 * should be at least that large. */
118 #define AUDIT_BUFSIZ 1024
120 /* AUDIT_MAXFREE is the number of empty audit_buffers we keep on the
121 * audit_freelist. Doing so eliminates many kmalloc/kfree calls. */
122 #define AUDIT_MAXFREE (2*NR_CPUS)
124 /* The audit_buffer is used when formatting an audit record. The caller
125 * locks briefly to get the record off the freelist or to allocate the
126 * buffer, and locks briefly to send the buffer to the netlink layer or
127 * to place it on a transmit queue. Multiple audit_buffers can be in
128 * use simultaneously. */
129 struct audit_buffer {
130 struct list_head list;
131 struct sk_buff *skb; /* formatted skb ready to send */
132 struct audit_context *ctx; /* NULL or associated context */
135 static void audit_set_pid(struct audit_buffer *ab, pid_t pid)
137 struct nlmsghdr *nlh = (struct nlmsghdr *)ab->skb->data;
138 nlh->nlmsg_pid = pid;
142 struct list_head list;
143 struct audit_rule rule;
146 static void audit_panic(const char *message)
148 switch (audit_failure)
150 case AUDIT_FAIL_SILENT:
152 case AUDIT_FAIL_PRINTK:
153 printk(KERN_ERR "audit: %s\n", message);
155 case AUDIT_FAIL_PANIC:
156 panic("audit: %s\n", message);
161 static inline int audit_rate_check(void)
163 static unsigned long last_check = 0;
164 static int messages = 0;
165 static DEFINE_SPINLOCK(lock);
168 unsigned long elapsed;
171 if (!audit_rate_limit) return 1;
173 spin_lock_irqsave(&lock, flags);
174 if (++messages < audit_rate_limit) {
178 elapsed = now - last_check;
185 spin_unlock_irqrestore(&lock, flags);
190 /* Emit at least 1 message per second, even if audit_rate_check is
192 void audit_log_lost(const char *message)
194 static unsigned long last_msg = 0;
195 static DEFINE_SPINLOCK(lock);
200 atomic_inc(&audit_lost);
202 print = (audit_failure == AUDIT_FAIL_PANIC || !audit_rate_limit);
205 spin_lock_irqsave(&lock, flags);
207 if (now - last_msg > HZ) {
211 spin_unlock_irqrestore(&lock, flags);
216 "audit: audit_lost=%d audit_rate_limit=%d audit_backlog_limit=%d\n",
217 atomic_read(&audit_lost),
219 audit_backlog_limit);
220 audit_panic(message);
225 static int audit_set_rate_limit(int limit, uid_t loginuid)
227 int old = audit_rate_limit;
228 audit_rate_limit = limit;
229 audit_log(NULL, AUDIT_CONFIG_CHANGE,
230 "audit_rate_limit=%d old=%d by auid=%u",
231 audit_rate_limit, old, loginuid);
235 static int audit_set_backlog_limit(int limit, uid_t loginuid)
237 int old = audit_backlog_limit;
238 audit_backlog_limit = limit;
239 audit_log(NULL, AUDIT_CONFIG_CHANGE,
240 "audit_backlog_limit=%d old=%d by auid=%u",
241 audit_backlog_limit, old, loginuid);
245 static int audit_set_enabled(int state, uid_t loginuid)
247 int old = audit_enabled;
248 if (state != 0 && state != 1)
250 audit_enabled = state;
251 audit_log(NULL, AUDIT_CONFIG_CHANGE,
252 "audit_enabled=%d old=%d by auid=%u",
253 audit_enabled, old, loginuid);
257 static int audit_set_failure(int state, uid_t loginuid)
259 int old = audit_failure;
260 if (state != AUDIT_FAIL_SILENT
261 && state != AUDIT_FAIL_PRINTK
262 && state != AUDIT_FAIL_PANIC)
264 audit_failure = state;
265 audit_log(NULL, AUDIT_CONFIG_CHANGE,
266 "audit_failure=%d old=%d by auid=%u",
267 audit_failure, old, loginuid);
271 int kauditd_thread(void *dummy)
276 skb = skb_dequeue(&audit_skb_queue);
279 int err = netlink_unicast(audit_sock, skb, audit_pid, 0);
281 BUG_ON(err != -ECONNREFUSED); /* Shoudn't happen */
282 printk(KERN_ERR "audit: *NO* daemon at audit_pid=%d\n", audit_pid);
286 printk(KERN_ERR "%s\n", skb->data + NLMSG_SPACE(0));
290 DECLARE_WAITQUEUE(wait, current);
291 set_current_state(TASK_INTERRUPTIBLE);
292 add_wait_queue(&kauditd_wait, &wait);
294 if (!skb_queue_len(&audit_skb_queue))
297 __set_current_state(TASK_RUNNING);
298 remove_wait_queue(&kauditd_wait, &wait);
303 void audit_send_reply(int pid, int seq, int type, int done, int multi,
304 void *payload, int size)
307 struct nlmsghdr *nlh;
308 int len = NLMSG_SPACE(size);
310 int flags = multi ? NLM_F_MULTI : 0;
311 int t = done ? NLMSG_DONE : type;
313 skb = alloc_skb(len, GFP_KERNEL);
317 nlh = NLMSG_PUT(skb, pid, seq, t, size);
318 nlh->nlmsg_flags = flags;
319 data = NLMSG_DATA(nlh);
320 memcpy(data, payload, size);
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);
327 nlmsg_failure: /* Used by NLMSG_PUT */
333 * Check for appropriate CAP_AUDIT_ capabilities on incoming audit
336 static int audit_netlink_ok(kernel_cap_t eff_cap, u16 msg_type)
346 case AUDIT_SIGNAL_INFO:
347 if (!cap_raised(eff_cap, CAP_AUDIT_CONTROL))
351 case AUDIT_FIRST_USER_MSG...AUDIT_LAST_USER_MSG:
352 if (!cap_raised(eff_cap, CAP_AUDIT_WRITE))
355 default: /* bad msg */
362 static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
366 struct audit_status *status_get, status_set;
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 struct task_struct *tsk;
374 err = audit_netlink_ok(NETLINK_CB(skb).eff_cap, msg_type);
378 /* As soon as there's any sign of userspace auditd, start kauditd to talk to it */
380 kauditd_task = kthread_run(kauditd_thread, NULL, "kauditd");
381 if (IS_ERR(kauditd_task)) {
382 err = PTR_ERR(kauditd_task);
387 pid = NETLINK_CREDS(skb)->pid;
388 uid = NETLINK_CREDS(skb)->uid;
389 loginuid = NETLINK_CB(skb).loginuid;
390 seq = nlh->nlmsg_seq;
391 data = NLMSG_DATA(nlh);
395 status_set.enabled = audit_enabled;
396 status_set.failure = audit_failure;
397 status_set.pid = audit_pid;
398 status_set.rate_limit = audit_rate_limit;
399 status_set.backlog_limit = audit_backlog_limit;
400 status_set.lost = atomic_read(&audit_lost);
401 status_set.backlog = skb_queue_len(&audit_skb_queue);
402 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_GET, 0, 0,
403 &status_set, sizeof(status_set));
406 if (nlh->nlmsg_len < sizeof(struct audit_status))
408 status_get = (struct audit_status *)data;
409 if (status_get->mask & AUDIT_STATUS_ENABLED) {
410 err = audit_set_enabled(status_get->enabled, loginuid);
411 if (err < 0) return err;
413 if (status_get->mask & AUDIT_STATUS_FAILURE) {
414 err = audit_set_failure(status_get->failure, loginuid);
415 if (err < 0) return err;
417 if (status_get->mask & AUDIT_STATUS_PID) {
419 audit_pid = status_get->pid;
420 audit_log(NULL, AUDIT_CONFIG_CHANGE,
421 "audit_pid=%d old=%d by auid=%u",
422 audit_pid, old, loginuid);
424 if (status_get->mask & AUDIT_STATUS_RATE_LIMIT)
425 audit_set_rate_limit(status_get->rate_limit, loginuid);
426 if (status_get->mask & AUDIT_STATUS_BACKLOG_LIMIT)
427 audit_set_backlog_limit(status_get->backlog_limit,
431 case AUDIT_FIRST_USER_MSG...AUDIT_LAST_USER_MSG:
432 read_lock(&tasklist_lock);
433 tsk = find_task_by_pid(pid);
435 get_task_struct(tsk);
436 read_unlock(&tasklist_lock);
440 if (audit_filter_user(tsk, msg_type)) {
441 ab = audit_log_start(NULL, msg_type);
444 "user pid=%d uid=%u auid=%u msg='%.1024s'",
445 pid, uid, loginuid, (char *)data);
446 audit_set_pid(ab, pid);
450 put_task_struct(tsk);
454 if (nlh->nlmsg_len < sizeof(struct audit_rule))
458 err = audit_receive_filter(nlh->nlmsg_type, NETLINK_CB(skb).pid,
459 uid, seq, data, loginuid);
461 case AUDIT_SIGNAL_INFO:
462 sig_data.uid = audit_sig_uid;
463 sig_data.pid = audit_sig_pid;
464 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_SIGNAL_INFO,
465 0, 0, &sig_data, sizeof(sig_data));
472 return err < 0 ? err : 0;
475 /* Get message from skb (based on rtnetlink_rcv_skb). Each message is
476 * processed by audit_receive_msg. Malformed skbs with wrong length are
477 * discarded silently. */
478 static void audit_receive_skb(struct sk_buff *skb)
481 struct nlmsghdr *nlh;
484 while (skb->len >= NLMSG_SPACE(0)) {
485 nlh = (struct nlmsghdr *)skb->data;
486 if (nlh->nlmsg_len < sizeof(*nlh) || skb->len < nlh->nlmsg_len)
488 rlen = NLMSG_ALIGN(nlh->nlmsg_len);
491 if ((err = audit_receive_msg(skb, nlh))) {
492 netlink_ack(skb, nlh, err);
493 } else if (nlh->nlmsg_flags & NLM_F_ACK)
494 netlink_ack(skb, nlh, 0);
499 /* Receive messages from netlink socket. */
500 static void audit_receive(struct sock *sk, int length)
505 down(&audit_netlink_sem);
507 for (qlen = skb_queue_len(&sk->sk_receive_queue); qlen; qlen--) {
508 skb = skb_dequeue(&sk->sk_receive_queue);
509 audit_receive_skb(skb);
512 up(&audit_netlink_sem);
516 /* Initialize audit support at boot time. */
517 static int __init audit_init(void)
519 printk(KERN_INFO "audit: initializing netlink socket (%s)\n",
520 audit_default ? "enabled" : "disabled");
521 audit_sock = netlink_kernel_create(NETLINK_AUDIT, audit_receive);
523 audit_panic("cannot initialize netlink socket");
525 audit_sock->sk_sndtimeo = MAX_SCHEDULE_TIMEOUT;
526 skb_queue_head_init(&audit_skb_queue);
527 audit_initialized = 1;
528 audit_enabled = audit_default;
529 audit_log(NULL, AUDIT_KERNEL, "initialized");
532 __initcall(audit_init);
534 /* Process kernel command-line parameter at boot time. audit=0 or audit=1. */
535 static int __init audit_enable(char *str)
537 audit_default = !!simple_strtol(str, NULL, 0);
538 printk(KERN_INFO "audit: %s%s\n",
539 audit_default ? "enabled" : "disabled",
540 audit_initialized ? "" : " (after initialization)");
541 if (audit_initialized)
542 audit_enabled = audit_default;
546 __setup("audit=", audit_enable);
548 static void audit_buffer_free(struct audit_buffer *ab)
558 spin_lock_irqsave(&audit_freelist_lock, flags);
559 if (++audit_freelist_count > AUDIT_MAXFREE)
562 list_add(&ab->list, &audit_freelist);
563 spin_unlock_irqrestore(&audit_freelist_lock, flags);
566 static struct audit_buffer * audit_buffer_alloc(struct audit_context *ctx,
567 int gfp_mask, int type)
570 struct audit_buffer *ab = NULL;
571 struct nlmsghdr *nlh;
573 spin_lock_irqsave(&audit_freelist_lock, flags);
574 if (!list_empty(&audit_freelist)) {
575 ab = list_entry(audit_freelist.next,
576 struct audit_buffer, list);
578 --audit_freelist_count;
580 spin_unlock_irqrestore(&audit_freelist_lock, flags);
583 ab = kmalloc(sizeof(*ab), gfp_mask);
588 ab->skb = alloc_skb(AUDIT_BUFSIZ, gfp_mask);
593 nlh = (struct nlmsghdr *)skb_put(ab->skb, NLMSG_SPACE(0));
594 nlh->nlmsg_type = type;
595 nlh->nlmsg_flags = 0;
600 audit_buffer_free(ab);
604 /* Compute a serial number for the audit record. Audit records are
605 * written to user-space as soon as they are generated, so a complete
606 * audit record may be written in several pieces. The timestamp of the
607 * record and this serial number are used by the user-space tools to
608 * determine which pieces belong to the same audit record. The
609 * (timestamp,serial) tuple is unique for each syscall and is live from
610 * syscall entry to syscall exit.
612 * Atomic values are only guaranteed to be 24-bit, so we count down.
614 * NOTE: Another possibility is to store the formatted records off the
615 * audit context (for those records that have a context), and emit them
616 * all at syscall exit. However, this could delay the reporting of
617 * significant errors until syscall exit (or never, if the system
619 unsigned int audit_serial(void)
621 static atomic_t serial = ATOMIC_INIT(0xffffff);
625 a = atomic_read(&serial);
626 if (atomic_dec_and_test(&serial))
627 atomic_set(&serial, 0xffffff);
628 b = atomic_read(&serial);
629 } while (b != a - 1);
634 static inline void audit_get_stamp(struct audit_context *ctx,
635 struct timespec *t, unsigned int *serial)
638 auditsc_get_stamp(ctx, t, serial);
641 *serial = audit_serial();
645 /* Obtain an audit buffer. This routine does locking to obtain the
646 * audit buffer, but then no locking is required for calls to
647 * audit_log_*format. If the tsk is a task that is currently in a
648 * syscall, then the syscall is marked as auditable and an audit record
649 * will be written at syscall exit. If there is no associated task, tsk
651 struct audit_buffer *audit_log_start(struct audit_context *ctx, int type)
653 struct audit_buffer *ab = NULL;
657 if (!audit_initialized)
660 if (audit_backlog_limit
661 && skb_queue_len(&audit_skb_queue) > audit_backlog_limit) {
662 if (audit_rate_check())
664 "audit: audit_backlog=%d > "
665 "audit_backlog_limit=%d\n",
666 skb_queue_len(&audit_skb_queue),
667 audit_backlog_limit);
668 audit_log_lost("backlog limit exceeded");
672 ab = audit_buffer_alloc(ctx, GFP_ATOMIC, type);
674 audit_log_lost("out of memory in audit_log_start");
678 audit_get_stamp(ab->ctx, &t, &serial);
680 audit_log_format(ab, "audit(%lu.%03lu:%u): ",
681 t.tv_sec, t.tv_nsec/1000000, serial);
686 * audit_expand - expand skb in the audit buffer
689 * Returns 0 (no space) on failed expansion, or available space if
692 static inline int audit_expand(struct audit_buffer *ab, int extra)
694 struct sk_buff *skb = ab->skb;
695 int ret = pskb_expand_head(skb, skb_headroom(skb), extra,
698 audit_log_lost("out of memory in audit_expand");
701 return skb_tailroom(skb);
704 /* Format an audit message into the audit buffer. If there isn't enough
705 * room in the audit buffer, more room will be allocated and vsnprint
706 * will be called a second time. Currently, we assume that a printk
707 * can't format message larger than 1024 bytes, so we don't either. */
708 static void audit_log_vformat(struct audit_buffer *ab, const char *fmt,
720 avail = skb_tailroom(skb);
722 avail = audit_expand(ab, AUDIT_BUFSIZ);
726 va_copy(args2, args);
727 len = vsnprintf(skb->tail, avail, fmt, args);
729 /* The printk buffer is 1024 bytes long, so if we get
730 * here and AUDIT_BUFSIZ is at least 1024, then we can
731 * log everything that printk could have logged. */
732 avail = audit_expand(ab, max_t(unsigned, AUDIT_BUFSIZ, 1+len-avail));
735 len = vsnprintf(skb->tail, avail, fmt, args2);
743 /* Format a message into the audit buffer. All the work is done in
744 * audit_log_vformat. */
745 void audit_log_format(struct audit_buffer *ab, const char *fmt, ...)
752 audit_log_vformat(ab, fmt, args);
756 /* This function will take the passed buf and convert it into a string of
757 * ascii hex digits. The new string is placed onto the skb. */
758 void audit_log_hex(struct audit_buffer *ab, const unsigned char *buf,
761 int i, avail, new_len;
764 static const unsigned char *hex = "0123456789ABCDEF";
768 avail = skb_tailroom(skb);
770 if (new_len >= avail) {
771 /* Round the buffer request up to the next multiple */
772 new_len = AUDIT_BUFSIZ*(((new_len-avail)/AUDIT_BUFSIZ) + 1);
773 avail = audit_expand(ab, new_len);
779 for (i=0; i<len; i++) {
780 *ptr++ = hex[(buf[i] & 0xF0)>>4]; /* Upper nibble */
781 *ptr++ = hex[buf[i] & 0x0F]; /* Lower nibble */
784 skb_put(skb, len << 1); /* new string is twice the old string */
787 /* This code will escape a string that is passed to it if the string
788 * contains a control character, unprintable character, double quote mark,
789 * or a space. Unescaped strings will start and end with a double quote mark.
790 * Strings that are escaped are printed in hex (2 digits per char). */
791 void audit_log_untrustedstring(struct audit_buffer *ab, const char *string)
793 const unsigned char *p = string;
796 if (*p == '"' || *p < 0x21 || *p > 0x7f) {
797 audit_log_hex(ab, string, strlen(string));
802 audit_log_format(ab, "\"%s\"", string);
805 /* This is a helper-function to print the escaped d_path */
806 void audit_log_d_path(struct audit_buffer *ab, const char *prefix,
807 struct dentry *dentry, struct vfsmount *vfsmnt)
812 audit_log_format(ab, " %s", prefix);
814 /* We will allow 11 spaces for ' (deleted)' to be appended */
815 path = kmalloc(PATH_MAX+11, GFP_KERNEL);
817 audit_log_format(ab, "<no memory>");
820 p = d_path(dentry, vfsmnt, path, PATH_MAX+11);
821 if (IS_ERR(p)) { /* Should never happen since we send PATH_MAX */
822 /* FIXME: can we save some information here? */
823 audit_log_format(ab, "<too long>");
825 audit_log_untrustedstring(ab, p);
829 /* The netlink_* functions cannot be called inside an irq context, so
830 * the audit buffer is places on a queue and a tasklet is scheduled to
831 * remove them from the queue outside the irq context. May be called in
833 void audit_log_end(struct audit_buffer *ab)
837 if (!audit_rate_check()) {
838 audit_log_lost("rate limit exceeded");
841 struct nlmsghdr *nlh = (struct nlmsghdr *)ab->skb->data;
842 nlh->nlmsg_len = ab->skb->len - NLMSG_SPACE(0);
843 skb_queue_tail(&audit_skb_queue, ab->skb);
845 wake_up_interruptible(&kauditd_wait);
847 printk("%s\n", ab->skb->data + NLMSG_SPACE(0));
850 audit_buffer_free(ab);
853 /* Log an audit record. This is a convenience function that calls
854 * audit_log_start, audit_log_vformat, and audit_log_end. It may be
855 * called in any context. */
856 void audit_log(struct audit_context *ctx, int type, const char *fmt, ...)
858 struct audit_buffer *ab;
861 ab = audit_log_start(ctx, type);
864 audit_log_vformat(ab, fmt, args);