1 /* auditsc.c -- System-call auditing support -*- linux-c -*-
2 * Handles all system-call specific auditing features.
4 * Copyright 2003-2004 Red Hat Inc., Durham, North Carolina.
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License as published by
9 * the Free Software Foundation; either version 2 of the License, or
10 * (at your option) any later version.
12 * This program is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 * GNU General Public License for more details.
17 * You should have received a copy of the GNU General Public License
18 * along with this program; if not, write to the Free Software
19 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 * Written by Rickard E. (Rik) Faith <faith@redhat.com>
23 * Many of the ideas implemented here are from Stephen C. Tweedie,
24 * especially the idea of avoiding a copy by using getname.
26 * The method for actual interception of syscall entry and exit (not in
27 * this file -- see entry.S) is based on a GPL'd patch written by
28 * okir@suse.de and Copyright 2003 SuSE Linux AG.
32 #include <linux/init.h>
33 #include <asm/atomic.h>
34 #include <asm/types.h>
36 #include <linux/module.h>
38 #include <linux/audit.h>
39 #include <linux/personality.h>
40 #include <linux/time.h>
41 #include <asm/unistd.h>
44 1 = put_count checking
45 2 = verbose put_count checking
49 /* No syscall auditing will take place unless audit_enabled != 0. */
50 extern int audit_enabled;
52 /* AUDIT_NAMES is the number of slots we reserve in the audit_context
53 * for saving names from getname(). */
54 #define AUDIT_NAMES 20
56 /* AUDIT_NAMES_RESERVED is the number of slots we reserve in the
57 * audit_context from being used for nameless inodes from
59 #define AUDIT_NAMES_RESERVED 7
61 /* At task start time, the audit_state is set in the audit_context using
62 a per-task filter. At syscall entry, the audit_state is augmented by
63 the syscall filter. */
65 AUDIT_DISABLED, /* Do not create per-task audit_context.
66 * No syscall-specific audit records can
68 AUDIT_SETUP_CONTEXT, /* Create the per-task audit_context,
69 * but don't necessarily fill it in at
70 * syscall entry time (i.e., filter
72 AUDIT_BUILD_CONTEXT, /* Create the per-task audit_context,
73 * and always fill it in at syscall
74 * entry time. This makes a full
75 * syscall record available if some
76 * other part of the kernel decides it
77 * should be recorded. */
78 AUDIT_RECORD_CONTEXT /* Create the per-task audit_context,
79 * always fill it in at syscall entry
80 * time, and always write out the audit
81 * record at syscall exit time. */
84 /* When fs/namei.c:getname() is called, we store the pointer in name and
85 * we don't let putname() free it (instead we free all of the saved
86 * pointers at syscall exit time).
88 * Further, in fs/namei.c:path_lookup() we store the inode and device. */
95 /* The per-task audit context. */
96 struct audit_context {
97 int in_syscall; /* 1 if task is in a syscall */
98 enum audit_state state;
99 unsigned int serial; /* serial number for record */
100 struct timespec ctime; /* time of syscall entry */
101 uid_t loginuid; /* login uid (identity) */
102 int major; /* syscall number */
103 unsigned long argv[4]; /* syscall arguments */
104 int return_valid; /* return code is valid */
105 int return_code;/* syscall return code */
106 int auditable; /* 1 if record should be written */
108 struct audit_names names[AUDIT_NAMES];
109 struct audit_context *previous; /* For nested syscalls */
111 /* Save things to print about task_struct */
113 uid_t uid, euid, suid, fsuid;
114 gid_t gid, egid, sgid, fsgid;
115 unsigned long personality;
124 /* There are three lists of rules -- one to search at task creation
125 * time, one to search at syscall entry time, and another to search at
126 * syscall exit time. */
127 static LIST_HEAD(audit_tsklist);
128 static LIST_HEAD(audit_entlist);
129 static LIST_HEAD(audit_extlist);
132 struct list_head list;
134 struct audit_rule rule;
137 /* Check to see if two rules are identical. It is called from
138 * audit_del_rule during AUDIT_DEL. */
139 static int audit_compare_rule(struct audit_rule *a, struct audit_rule *b)
143 if (a->flags != b->flags)
146 if (a->action != b->action)
149 if (a->field_count != b->field_count)
152 for (i = 0; i < a->field_count; i++) {
153 if (a->fields[i] != b->fields[i]
154 || a->values[i] != b->values[i])
158 for (i = 0; i < AUDIT_BITMASK_SIZE; i++)
159 if (a->mask[i] != b->mask[i])
165 /* Note that audit_add_rule and audit_del_rule are called via
166 * audit_receive() in audit.c, and are protected by
167 * audit_netlink_sem. */
168 static inline int audit_add_rule(struct audit_entry *entry,
169 struct list_head *list)
171 if (entry->rule.flags & AUDIT_PREPEND) {
172 entry->rule.flags &= ~AUDIT_PREPEND;
173 list_add_rcu(&entry->list, list);
175 list_add_tail_rcu(&entry->list, list);
180 static void audit_free_rule(void *arg)
185 /* Note that audit_add_rule and audit_del_rule are called via
186 * audit_receive() in audit.c, and are protected by
187 * audit_netlink_sem. */
188 static inline int audit_del_rule(struct audit_rule *rule,
189 struct list_head *list)
191 struct audit_entry *e;
193 /* Do not use the _rcu iterator here, since this is the only
194 * deletion routine. */
195 list_for_each_entry(e, list, list) {
196 if (!audit_compare_rule(rule, &e->rule)) {
197 list_del_rcu(&e->list);
198 call_rcu(&e->rcu, audit_free_rule, e);
202 return -EFAULT; /* No matching rule */
206 /* Copy rule from user-space to kernel-space. Called during
208 static int audit_copy_rule(struct audit_rule *d, struct audit_rule *s)
212 if (s->action != AUDIT_NEVER
213 && s->action != AUDIT_POSSIBLE
214 && s->action != AUDIT_ALWAYS)
216 if (s->field_count < 0 || s->field_count > AUDIT_MAX_FIELDS)
220 d->action = s->action;
221 d->field_count = s->field_count;
222 for (i = 0; i < d->field_count; i++) {
223 d->fields[i] = s->fields[i];
224 d->values[i] = s->values[i];
226 for (i = 0; i < AUDIT_BITMASK_SIZE; i++) d->mask[i] = s->mask[i];
230 int audit_receive_filter(int type, int pid, int uid, int seq, void *data)
233 struct audit_entry *entry;
238 /* The *_rcu iterators not needed here because we are
239 always called with audit_netlink_sem held. */
240 list_for_each_entry(entry, &audit_tsklist, list)
241 audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
242 &entry->rule, sizeof(entry->rule));
243 list_for_each_entry(entry, &audit_entlist, list)
244 audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
245 &entry->rule, sizeof(entry->rule));
246 list_for_each_entry(entry, &audit_extlist, list)
247 audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
248 &entry->rule, sizeof(entry->rule));
249 audit_send_reply(pid, seq, AUDIT_LIST, 1, 1, NULL, 0);
252 if (!capable(CAP_SYS_ADMIN))
254 if (!(entry = kmalloc(sizeof(*entry), GFP_KERNEL)))
256 if (audit_copy_rule(&entry->rule, data)) {
260 flags = entry->rule.flags;
261 if (!err && (flags & AUDIT_PER_TASK))
262 err = audit_add_rule(entry, &audit_tsklist);
263 if (!err && (flags & AUDIT_AT_ENTRY))
264 err = audit_add_rule(entry, &audit_entlist);
265 if (!err && (flags & AUDIT_AT_EXIT))
266 err = audit_add_rule(entry, &audit_extlist);
269 flags =((struct audit_rule *)data)->flags;
270 if (!err && (flags & AUDIT_PER_TASK))
271 err = audit_del_rule(data, &audit_tsklist);
272 if (!err && (flags & AUDIT_AT_ENTRY))
273 err = audit_del_rule(data, &audit_entlist);
274 if (!err && (flags & AUDIT_AT_EXIT))
275 err = audit_del_rule(data, &audit_extlist);
285 /* Compare a task_struct with an audit_rule. Return 1 on match, 0
287 static int audit_filter_rules(struct task_struct *tsk,
288 struct audit_rule *rule,
289 struct audit_context *ctx,
290 enum audit_state *state)
294 for (i = 0; i < rule->field_count; i++) {
295 u32 field = rule->fields[i] & ~AUDIT_NEGATE;
296 u32 value = rule->values[i];
301 result = (tsk->pid == value);
304 result = (tsk->uid == value);
307 result = (tsk->euid == value);
310 result = (tsk->suid == value);
313 result = (tsk->fsuid == value);
316 result = (tsk->gid == value);
319 result = (tsk->egid == value);
322 result = (tsk->sgid == value);
325 result = (tsk->fsgid == value);
328 result = (tsk->personality == value);
332 if (ctx && ctx->return_valid)
333 result = (ctx->return_code == value);
336 if (ctx && ctx->return_valid)
337 result = (ctx->return_code >= 0);
341 for (j = 0; j < ctx->name_count; j++) {
342 if (MAJOR(ctx->names[j].rdev)==value) {
351 for (j = 0; j < ctx->name_count; j++) {
352 if (MINOR(ctx->names[j].rdev)==value) {
361 for (j = 0; j < ctx->name_count; j++) {
362 if (MINOR(ctx->names[j].ino)==value) {
372 result = (ctx->loginuid == value);
379 result = (ctx->argv[field-AUDIT_ARG0]==value);
383 if (rule->fields[i] & AUDIT_NEGATE)
388 switch (rule->action) {
389 case AUDIT_NEVER: *state = AUDIT_DISABLED; break;
390 case AUDIT_POSSIBLE: *state = AUDIT_BUILD_CONTEXT; break;
391 case AUDIT_ALWAYS: *state = AUDIT_RECORD_CONTEXT; break;
396 /* At process creation time, we can determine if system-call auditing is
397 * completely disabled for this task. Since we only have the task
398 * structure at this point, we can only check uid and gid.
400 static enum audit_state audit_filter_task(struct task_struct *tsk)
402 struct audit_entry *e;
403 enum audit_state state;
406 list_for_each_entry_rcu(e, &audit_tsklist, list) {
407 if (audit_filter_rules(tsk, &e->rule, NULL, &state)) {
413 return AUDIT_BUILD_CONTEXT;
416 /* At syscall entry and exit time, this filter is called if the
417 * audit_state is not low enough that auditing cannot take place, but is
418 * also not high enough that we already know we have to write and audit
419 * record (i.e., the state is AUDIT_SETUP_CONTEXT or AUDIT_BUILD_CONTEXT).
421 static enum audit_state audit_filter_syscall(struct task_struct *tsk,
422 struct audit_context *ctx,
423 struct list_head *list)
425 struct audit_entry *e;
426 enum audit_state state;
427 int word = AUDIT_WORD(ctx->major);
428 int bit = AUDIT_BIT(ctx->major);
431 list_for_each_entry_rcu(e, list, list) {
432 if ((e->rule.mask[word] & bit) == bit
433 && audit_filter_rules(tsk, &e->rule, ctx, &state)) {
439 return AUDIT_BUILD_CONTEXT;
442 /* This should be called with task_lock() held. */
443 static inline struct audit_context *audit_get_context(struct task_struct *tsk,
447 struct audit_context *context = tsk->audit_context;
449 if (likely(!context))
451 context->return_valid = return_valid;
452 context->return_code = return_code;
454 if (context->in_syscall && !context->auditable) {
455 enum audit_state state;
456 state = audit_filter_syscall(tsk, context, &audit_extlist);
457 if (state == AUDIT_RECORD_CONTEXT)
458 context->auditable = 1;
461 context->pid = tsk->pid;
462 context->uid = tsk->uid;
463 context->gid = tsk->gid;
464 context->euid = tsk->euid;
465 context->suid = tsk->suid;
466 context->fsuid = tsk->fsuid;
467 context->egid = tsk->egid;
468 context->sgid = tsk->sgid;
469 context->fsgid = tsk->fsgid;
470 context->personality = tsk->personality;
471 tsk->audit_context = NULL;
475 static inline void audit_free_names(struct audit_context *context)
480 if (context->auditable
481 ||context->put_count + context->ino_count != context->name_count) {
482 printk(KERN_ERR "audit.c:%d(:%d): major=%d in_syscall=%d"
483 " name_count=%d put_count=%d"
484 " ino_count=%d [NOT freeing]\n",
486 context->serial, context->major, context->in_syscall,
487 context->name_count, context->put_count,
489 for (i = 0; i < context->name_count; i++)
490 printk(KERN_ERR "names[%d] = %p = %s\n", i,
491 context->names[i].name,
492 context->names[i].name);
498 context->put_count = 0;
499 context->ino_count = 0;
502 for (i = 0; i < context->name_count; i++)
503 if (context->names[i].name)
504 __putname(context->names[i].name);
505 context->name_count = 0;
508 static inline void audit_zero_context(struct audit_context *context,
509 enum audit_state state)
511 uid_t loginuid = context->loginuid;
513 memset(context, 0, sizeof(*context));
514 context->state = state;
515 context->loginuid = loginuid;
518 static inline struct audit_context *audit_alloc_context(enum audit_state state)
520 struct audit_context *context;
522 if (!(context = kmalloc(sizeof(*context), GFP_KERNEL)))
524 audit_zero_context(context, state);
528 /* Filter on the task information and allocate a per-task audit context
529 * if necessary. Doing so turns on system call auditing for the
530 * specified task. This is called from copy_process, so no lock is
532 int audit_alloc(struct task_struct *tsk)
534 struct audit_context *context;
535 enum audit_state state;
537 if (likely(!audit_enabled))
538 return 0; /* Return if not auditing. */
540 state = audit_filter_task(tsk);
541 if (likely(state == AUDIT_DISABLED))
544 if (!(context = audit_alloc_context(state))) {
545 audit_log_lost("out of memory in audit_alloc");
549 /* Preserve login uid */
550 context->loginuid = -1;
551 if (tsk->audit_context)
552 context->loginuid = tsk->audit_context->loginuid;
554 tsk->audit_context = context;
555 set_tsk_thread_flag(tsk, TIF_SYSCALL_AUDIT);
559 static inline void audit_free_context(struct audit_context *context)
561 struct audit_context *previous;
565 previous = context->previous;
566 if (previous || (count && count < 10)) {
568 printk(KERN_ERR "audit(:%d): major=%d name_count=%d:"
569 " freeing multiple contexts (%d)\n",
570 context->serial, context->major,
571 context->name_count, count);
573 audit_free_names(context);
578 printk(KERN_ERR "audit: freed %d contexts\n", count);
581 static void audit_log_exit(struct audit_context *context)
584 struct audit_buffer *ab;
586 ab = audit_log_start(context);
588 return; /* audit_panic has been called */
589 audit_log_format(ab, "syscall=%d", context->major);
590 if (context->personality != PER_LINUX)
591 audit_log_format(ab, " per=%lx", context->personality);
592 if (context->return_valid)
593 audit_log_format(ab, " exit=%u", context->return_code);
595 " a0=%lx a1=%lx a2=%lx a3=%lx items=%d"
596 " pid=%d loginuid=%d uid=%d gid=%d"
597 " euid=%d suid=%d fsuid=%d"
598 " egid=%d sgid=%d fsgid=%d",
608 context->euid, context->suid, context->fsuid,
609 context->egid, context->sgid, context->fsgid);
611 for (i = 0; i < context->name_count; i++) {
612 ab = audit_log_start(context);
614 continue; /* audit_panic has been called */
615 audit_log_format(ab, "item=%d", i);
616 if (context->names[i].name)
617 audit_log_format(ab, " name=%s",
618 context->names[i].name);
619 if (context->names[i].ino != (unsigned long)-1)
620 audit_log_format(ab, " inode=%lu",
621 context->names[i].ino);
622 /* FIXME: should use format_dev_t, but ab structure is
624 if (context->names[i].rdev != -1)
625 audit_log_format(ab, " dev=%02x:%02x",
626 MAJOR(context->names[i].rdev),
627 MINOR(context->names[i].rdev));
632 /* Free a per-task audit context. Called from copy_process and
633 * __put_task_struct. */
634 void audit_free(struct task_struct *tsk)
636 struct audit_context *context;
639 context = audit_get_context(tsk, 0, 0);
642 if (likely(!context))
645 /* Check for system calls that do not go through the exit
646 * function (e.g., exit_group), then free context block. */
647 if (context->in_syscall && context->auditable)
648 audit_log_exit(context);
650 audit_free_context(context);
653 /* Compute a serial number for the audit record. Audit records are
654 * written to user-space as soon as they are generated, so a complete
655 * audit record may be written in several pieces. The timestamp of the
656 * record and this serial number are used by the user-space daemon to
657 * determine which pieces belong to the same audit record. The
658 * (timestamp,serial) tuple is unique for each syscall and is live from
659 * syscall entry to syscall exit.
661 * Atomic values are only guaranteed to be 24-bit, so we count down.
663 * NOTE: Another possibility is to store the formatted records off the
664 * audit context (for those records that have a context), and emit them
665 * all at syscall exit. However, this could delay the reporting of
666 * significant errors until syscall exit (or never, if the system
668 static inline unsigned int audit_serial(void)
670 static atomic_t serial = ATOMIC_INIT(0xffffff);
674 a = atomic_read(&serial);
675 if (atomic_dec_and_test(&serial))
676 atomic_set(&serial, 0xffffff);
677 b = atomic_read(&serial);
678 } while (b != a - 1);
683 /* Fill in audit context at syscall entry. This only happens if the
684 * audit context was created when the task was created and the state or
685 * filters demand the audit context be built. If the state from the
686 * per-task filter or from the per-syscall filter is AUDIT_RECORD_CONTEXT,
687 * then the record will be written at syscall exit time (otherwise, it
688 * will only be written if another part of the kernel requests that it
690 void audit_syscall_entry(struct task_struct *tsk, int major,
691 unsigned long a1, unsigned long a2,
692 unsigned long a3, unsigned long a4)
694 struct audit_context *context = tsk->audit_context;
695 enum audit_state state;
699 /* This happens only on certain architectures that make system
700 * calls in kernel_thread via the entry.S interface, instead of
701 * with direct calls. (If you are porting to a new
702 * architecture, hitting this condition can indicate that you
703 * got the _exit/_leave calls backward in entry.S.)
707 * ppc64 yes (see arch/ppc64/kernel/misc.S)
709 * This also happens with vm86 emulation in a non-nested manner
710 * (entries without exits), so this case must be caught.
712 if (context->in_syscall) {
713 struct audit_context *newctx;
715 #if defined(__NR_vm86) && defined(__NR_vm86old)
716 /* vm86 mode should only be entered once */
717 if (major == __NR_vm86 || major == __NR_vm86old)
722 "audit(:%d) pid=%d in syscall=%d;"
723 " entering syscall=%d\n",
724 context->serial, tsk->pid, context->major, major);
726 newctx = audit_alloc_context(context->state);
728 newctx->previous = context;
730 tsk->audit_context = newctx;
732 /* If we can't alloc a new context, the best we
733 * can do is to leak memory (any pending putname
734 * will be lost). The only other alternative is
735 * to abandon auditing. */
736 audit_zero_context(context, context->state);
739 BUG_ON(context->in_syscall || context->name_count);
744 context->major = major;
745 context->argv[0] = a1;
746 context->argv[1] = a2;
747 context->argv[2] = a3;
748 context->argv[3] = a4;
750 state = context->state;
751 if (state == AUDIT_SETUP_CONTEXT || state == AUDIT_BUILD_CONTEXT)
752 state = audit_filter_syscall(tsk, context, &audit_entlist);
753 if (likely(state == AUDIT_DISABLED))
756 context->serial = audit_serial();
757 context->ctime = CURRENT_TIME;
758 context->in_syscall = 1;
759 context->auditable = !!(state == AUDIT_RECORD_CONTEXT);
762 /* Tear down after system call. If the audit context has been marked as
763 * auditable (either because of the AUDIT_RECORD_CONTEXT state from
764 * filtering, or because some other part of the kernel write an audit
765 * message), then write out the syscall information. In call cases,
766 * free the names stored from getname(). */
767 void audit_syscall_exit(struct task_struct *tsk, int return_code)
769 struct audit_context *context;
771 get_task_struct(tsk);
773 context = audit_get_context(tsk, 1, return_code);
776 /* Not having a context here is ok, since the parent may have
777 * called __put_task_struct. */
778 if (likely(!context))
781 if (context->in_syscall && context->auditable)
782 audit_log_exit(context);
784 context->in_syscall = 0;
785 context->auditable = 0;
786 if (context->previous) {
787 struct audit_context *new_context = context->previous;
788 context->previous = NULL;
789 audit_free_context(context);
790 tsk->audit_context = new_context;
792 audit_free_names(context);
793 audit_zero_context(context, context->state);
794 tsk->audit_context = context;
796 put_task_struct(tsk);
799 /* Add a name to the list. Called from fs/namei.c:getname(). */
800 void audit_getname(const char *name)
802 struct audit_context *context = current->audit_context;
805 if (!context->in_syscall) {
807 printk(KERN_ERR "%s:%d(:%d): ignoring getname(%p)\n",
808 __FILE__, __LINE__, context->serial, name);
813 BUG_ON(context->name_count >= AUDIT_NAMES);
814 context->names[context->name_count].name = name;
815 context->names[context->name_count].ino = (unsigned long)-1;
816 context->names[context->name_count].rdev = -1;
817 ++context->name_count;
820 /* Intercept a putname request. Called from
821 * include/linux/fs.h:putname(). If we have stored the name from
822 * getname in the audit context, then we delay the putname until syscall
824 void audit_putname(const char *name)
826 struct audit_context *context = current->audit_context;
829 if (!context->in_syscall) {
831 printk(KERN_ERR "%s:%d(:%d): __putname(%p)\n",
832 __FILE__, __LINE__, context->serial, name);
833 if (context->name_count) {
835 for (i = 0; i < context->name_count; i++)
836 printk(KERN_ERR "name[%d] = %p = %s\n", i,
837 context->names[i].name,
838 context->names[i].name);
845 ++context->put_count;
846 if (context->put_count > context->name_count) {
847 printk(KERN_ERR "%s:%d(:%d): major=%d"
848 " in_syscall=%d putname(%p) name_count=%d"
851 context->serial, context->major,
852 context->in_syscall, name, context->name_count,
860 /* Store the inode and device from a lookup. Called from
861 * fs/namei.c:path_lookup(). */
862 void audit_inode(const char *name, unsigned long ino, dev_t rdev)
865 struct audit_context *context = current->audit_context;
867 if (!context->in_syscall)
869 if (context->name_count
870 && context->names[context->name_count-1].name
871 && context->names[context->name_count-1].name == name)
872 idx = context->name_count - 1;
873 else if (context->name_count > 1
874 && context->names[context->name_count-2].name
875 && context->names[context->name_count-2].name == name)
876 idx = context->name_count - 2;
878 /* FIXME: how much do we care about inodes that have no
879 * associated name? */
880 if (context->name_count >= AUDIT_NAMES - AUDIT_NAMES_RESERVED)
882 idx = context->name_count++;
883 context->names[idx].name = NULL;
885 ++context->ino_count;
888 context->names[idx].ino = ino;
889 context->names[idx].rdev = rdev;
892 void audit_get_stamp(struct audit_context *ctx,
893 struct timespec *t, int *serial)
896 t->tv_sec = ctx->ctime.tv_sec;
897 t->tv_nsec = ctx->ctime.tv_nsec;
898 *serial = ctx->serial;
906 int audit_set_loginuid(struct audit_context *ctx, uid_t loginuid)
911 ctx->loginuid = loginuid;
916 EXPORT_SYMBOL_GPL(audit_alloc);
917 EXPORT_SYMBOL_GPL(audit_free);
918 EXPORT_SYMBOL_GPL(audit_syscall_entry);
919 EXPORT_SYMBOL_GPL(audit_syscall_exit);
920 EXPORT_SYMBOL_GPL(audit_getname);
921 EXPORT_SYMBOL_GPL(audit_putname);
922 EXPORT_SYMBOL_GPL(audit_inode);