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(struct rcu_head *head)
182 struct audit_entry *e = container_of(head, struct audit_entry, rcu);
186 /* Note that audit_add_rule and audit_del_rule are called via
187 * audit_receive() in audit.c, and are protected by
188 * audit_netlink_sem. */
189 static inline int audit_del_rule(struct audit_rule *rule,
190 struct list_head *list)
192 struct audit_entry *e;
194 /* Do not use the _rcu iterator here, since this is the only
195 * deletion routine. */
196 list_for_each_entry(e, list, list) {
197 if (!audit_compare_rule(rule, &e->rule)) {
198 list_del_rcu(&e->list);
199 call_rcu(&e->rcu, audit_free_rule);
203 return -EFAULT; /* No matching rule */
207 /* Copy rule from user-space to kernel-space. Called during
209 static int audit_copy_rule(struct audit_rule *d, struct audit_rule *s)
213 if (s->action != AUDIT_NEVER
214 && s->action != AUDIT_POSSIBLE
215 && s->action != AUDIT_ALWAYS)
217 if (s->field_count < 0 || s->field_count > AUDIT_MAX_FIELDS)
221 d->action = s->action;
222 d->field_count = s->field_count;
223 for (i = 0; i < d->field_count; i++) {
224 d->fields[i] = s->fields[i];
225 d->values[i] = s->values[i];
227 for (i = 0; i < AUDIT_BITMASK_SIZE; i++) d->mask[i] = s->mask[i];
231 int audit_receive_filter(int type, int pid, int uid, int seq, void *data)
234 struct audit_entry *entry;
239 /* The *_rcu iterators not needed here because we are
240 always called with audit_netlink_sem held. */
241 list_for_each_entry(entry, &audit_tsklist, list)
242 audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
243 &entry->rule, sizeof(entry->rule));
244 list_for_each_entry(entry, &audit_entlist, list)
245 audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
246 &entry->rule, sizeof(entry->rule));
247 list_for_each_entry(entry, &audit_extlist, list)
248 audit_send_reply(pid, seq, AUDIT_LIST, 0, 1,
249 &entry->rule, sizeof(entry->rule));
250 audit_send_reply(pid, seq, AUDIT_LIST, 1, 1, NULL, 0);
253 if (!capable(CAP_SYS_ADMIN))
255 if (!(entry = kmalloc(sizeof(*entry), GFP_KERNEL)))
257 if (audit_copy_rule(&entry->rule, data)) {
261 flags = entry->rule.flags;
262 if (!err && (flags & AUDIT_PER_TASK))
263 err = audit_add_rule(entry, &audit_tsklist);
264 if (!err && (flags & AUDIT_AT_ENTRY))
265 err = audit_add_rule(entry, &audit_entlist);
266 if (!err && (flags & AUDIT_AT_EXIT))
267 err = audit_add_rule(entry, &audit_extlist);
270 flags =((struct audit_rule *)data)->flags;
271 if (!err && (flags & AUDIT_PER_TASK))
272 err = audit_del_rule(data, &audit_tsklist);
273 if (!err && (flags & AUDIT_AT_ENTRY))
274 err = audit_del_rule(data, &audit_entlist);
275 if (!err && (flags & AUDIT_AT_EXIT))
276 err = audit_del_rule(data, &audit_extlist);
286 /* Compare a task_struct with an audit_rule. Return 1 on match, 0
288 static int audit_filter_rules(struct task_struct *tsk,
289 struct audit_rule *rule,
290 struct audit_context *ctx,
291 enum audit_state *state)
295 for (i = 0; i < rule->field_count; i++) {
296 u32 field = rule->fields[i] & ~AUDIT_NEGATE;
297 u32 value = rule->values[i];
302 result = (tsk->pid == value);
305 result = (tsk->uid == value);
308 result = (tsk->euid == value);
311 result = (tsk->suid == value);
314 result = (tsk->fsuid == value);
317 result = (tsk->gid == value);
320 result = (tsk->egid == value);
323 result = (tsk->sgid == value);
326 result = (tsk->fsgid == value);
329 result = (tsk->personality == value);
333 if (ctx && ctx->return_valid)
334 result = (ctx->return_code == value);
337 if (ctx && ctx->return_valid)
338 result = (ctx->return_code >= 0);
342 for (j = 0; j < ctx->name_count; j++) {
343 if (MAJOR(ctx->names[j].rdev)==value) {
352 for (j = 0; j < ctx->name_count; j++) {
353 if (MINOR(ctx->names[j].rdev)==value) {
362 for (j = 0; j < ctx->name_count; j++) {
363 if (MINOR(ctx->names[j].ino)==value) {
373 result = (ctx->loginuid == value);
380 result = (ctx->argv[field-AUDIT_ARG0]==value);
384 if (rule->fields[i] & AUDIT_NEGATE)
389 switch (rule->action) {
390 case AUDIT_NEVER: *state = AUDIT_DISABLED; break;
391 case AUDIT_POSSIBLE: *state = AUDIT_BUILD_CONTEXT; break;
392 case AUDIT_ALWAYS: *state = AUDIT_RECORD_CONTEXT; break;
397 /* At process creation time, we can determine if system-call auditing is
398 * completely disabled for this task. Since we only have the task
399 * structure at this point, we can only check uid and gid.
401 static enum audit_state audit_filter_task(struct task_struct *tsk)
403 struct audit_entry *e;
404 enum audit_state state;
407 list_for_each_entry_rcu(e, &audit_tsklist, list) {
408 if (audit_filter_rules(tsk, &e->rule, NULL, &state)) {
414 return AUDIT_BUILD_CONTEXT;
417 /* At syscall entry and exit time, this filter is called if the
418 * audit_state is not low enough that auditing cannot take place, but is
419 * also not high enough that we already know we have to write and audit
420 * record (i.e., the state is AUDIT_SETUP_CONTEXT or AUDIT_BUILD_CONTEXT).
422 static enum audit_state audit_filter_syscall(struct task_struct *tsk,
423 struct audit_context *ctx,
424 struct list_head *list)
426 struct audit_entry *e;
427 enum audit_state state;
428 int word = AUDIT_WORD(ctx->major);
429 int bit = AUDIT_BIT(ctx->major);
432 list_for_each_entry_rcu(e, list, list) {
433 if ((e->rule.mask[word] & bit) == bit
434 && audit_filter_rules(tsk, &e->rule, ctx, &state)) {
440 return AUDIT_BUILD_CONTEXT;
443 /* This should be called with task_lock() held. */
444 static inline struct audit_context *audit_get_context(struct task_struct *tsk,
448 struct audit_context *context = tsk->audit_context;
450 if (likely(!context))
452 context->return_valid = return_valid;
453 context->return_code = return_code;
455 if (context->in_syscall && !context->auditable) {
456 enum audit_state state;
457 state = audit_filter_syscall(tsk, context, &audit_extlist);
458 if (state == AUDIT_RECORD_CONTEXT)
459 context->auditable = 1;
462 context->pid = tsk->pid;
463 context->uid = tsk->uid;
464 context->gid = tsk->gid;
465 context->euid = tsk->euid;
466 context->suid = tsk->suid;
467 context->fsuid = tsk->fsuid;
468 context->egid = tsk->egid;
469 context->sgid = tsk->sgid;
470 context->fsgid = tsk->fsgid;
471 context->personality = tsk->personality;
472 tsk->audit_context = NULL;
476 static inline void audit_free_names(struct audit_context *context)
481 if (context->auditable
482 ||context->put_count + context->ino_count != context->name_count) {
483 printk(KERN_ERR "audit.c:%d(:%d): major=%d in_syscall=%d"
484 " name_count=%d put_count=%d"
485 " ino_count=%d [NOT freeing]\n",
487 context->serial, context->major, context->in_syscall,
488 context->name_count, context->put_count,
490 for (i = 0; i < context->name_count; i++)
491 printk(KERN_ERR "names[%d] = %p = %s\n", i,
492 context->names[i].name,
493 context->names[i].name);
499 context->put_count = 0;
500 context->ino_count = 0;
503 for (i = 0; i < context->name_count; i++)
504 if (context->names[i].name)
505 __putname(context->names[i].name);
506 context->name_count = 0;
509 static inline void audit_zero_context(struct audit_context *context,
510 enum audit_state state)
512 uid_t loginuid = context->loginuid;
514 memset(context, 0, sizeof(*context));
515 context->state = state;
516 context->loginuid = loginuid;
519 static inline struct audit_context *audit_alloc_context(enum audit_state state)
521 struct audit_context *context;
523 if (!(context = kmalloc(sizeof(*context), GFP_KERNEL)))
525 audit_zero_context(context, state);
529 /* Filter on the task information and allocate a per-task audit context
530 * if necessary. Doing so turns on system call auditing for the
531 * specified task. This is called from copy_process, so no lock is
533 int audit_alloc(struct task_struct *tsk)
535 struct audit_context *context;
536 enum audit_state state;
538 if (likely(!audit_enabled))
539 return 0; /* Return if not auditing. */
541 state = audit_filter_task(tsk);
542 if (likely(state == AUDIT_DISABLED))
545 if (!(context = audit_alloc_context(state))) {
546 audit_log_lost("out of memory in audit_alloc");
550 /* Preserve login uid */
551 context->loginuid = -1;
552 if (tsk->audit_context)
553 context->loginuid = tsk->audit_context->loginuid;
555 tsk->audit_context = context;
556 set_tsk_thread_flag(tsk, TIF_SYSCALL_AUDIT);
560 static inline void audit_free_context(struct audit_context *context)
562 struct audit_context *previous;
566 previous = context->previous;
567 if (previous || (count && count < 10)) {
569 printk(KERN_ERR "audit(:%d): major=%d name_count=%d:"
570 " freeing multiple contexts (%d)\n",
571 context->serial, context->major,
572 context->name_count, count);
574 audit_free_names(context);
579 printk(KERN_ERR "audit: freed %d contexts\n", count);
582 static void audit_log_exit(struct audit_context *context)
585 struct audit_buffer *ab;
587 ab = audit_log_start(context);
589 return; /* audit_panic has been called */
590 audit_log_format(ab, "syscall=%d", context->major);
591 if (context->personality != PER_LINUX)
592 audit_log_format(ab, " per=%lx", context->personality);
593 if (context->return_valid)
594 audit_log_format(ab, " exit=%d", context->return_code);
596 " a0=%lx a1=%lx a2=%lx a3=%lx items=%d"
597 " pid=%d loginuid=%d uid=%d gid=%d"
598 " euid=%d suid=%d fsuid=%d"
599 " egid=%d sgid=%d fsgid=%d",
609 context->euid, context->suid, context->fsuid,
610 context->egid, context->sgid, context->fsgid);
612 for (i = 0; i < context->name_count; i++) {
613 ab = audit_log_start(context);
615 continue; /* audit_panic has been called */
616 audit_log_format(ab, "item=%d", i);
617 if (context->names[i].name)
618 audit_log_format(ab, " name=%s",
619 context->names[i].name);
620 if (context->names[i].ino != (unsigned long)-1)
621 audit_log_format(ab, " inode=%lu",
622 context->names[i].ino);
623 /* FIXME: should use format_dev_t, but ab structure is
625 if (context->names[i].rdev != -1)
626 audit_log_format(ab, " dev=%02x:%02x",
627 MAJOR(context->names[i].rdev),
628 MINOR(context->names[i].rdev));
633 /* Free a per-task audit context. Called from copy_process and
634 * __put_task_struct. */
635 void audit_free(struct task_struct *tsk)
637 struct audit_context *context;
640 context = audit_get_context(tsk, 0, 0);
643 if (likely(!context))
646 /* Check for system calls that do not go through the exit
647 * function (e.g., exit_group), then free context block. */
648 if (context->in_syscall && context->auditable)
649 audit_log_exit(context);
651 audit_free_context(context);
654 /* Compute a serial number for the audit record. Audit records are
655 * written to user-space as soon as they are generated, so a complete
656 * audit record may be written in several pieces. The timestamp of the
657 * record and this serial number are used by the user-space daemon to
658 * determine which pieces belong to the same audit record. The
659 * (timestamp,serial) tuple is unique for each syscall and is live from
660 * syscall entry to syscall exit.
662 * Atomic values are only guaranteed to be 24-bit, so we count down.
664 * NOTE: Another possibility is to store the formatted records off the
665 * audit context (for those records that have a context), and emit them
666 * all at syscall exit. However, this could delay the reporting of
667 * significant errors until syscall exit (or never, if the system
669 static inline unsigned int audit_serial(void)
671 static atomic_t serial = ATOMIC_INIT(0xffffff);
675 a = atomic_read(&serial);
676 if (atomic_dec_and_test(&serial))
677 atomic_set(&serial, 0xffffff);
678 b = atomic_read(&serial);
679 } while (b != a - 1);
684 /* Fill in audit context at syscall entry. This only happens if the
685 * audit context was created when the task was created and the state or
686 * filters demand the audit context be built. If the state from the
687 * per-task filter or from the per-syscall filter is AUDIT_RECORD_CONTEXT,
688 * then the record will be written at syscall exit time (otherwise, it
689 * will only be written if another part of the kernel requests that it
691 void audit_syscall_entry(struct task_struct *tsk, int major,
692 unsigned long a1, unsigned long a2,
693 unsigned long a3, unsigned long a4)
695 struct audit_context *context = tsk->audit_context;
696 enum audit_state state;
700 /* This happens only on certain architectures that make system
701 * calls in kernel_thread via the entry.S interface, instead of
702 * with direct calls. (If you are porting to a new
703 * architecture, hitting this condition can indicate that you
704 * got the _exit/_leave calls backward in entry.S.)
708 * ppc64 yes (see arch/ppc64/kernel/misc.S)
710 * This also happens with vm86 emulation in a non-nested manner
711 * (entries without exits), so this case must be caught.
713 if (context->in_syscall) {
714 struct audit_context *newctx;
716 #if defined(__NR_vm86) && defined(__NR_vm86old)
717 /* vm86 mode should only be entered once */
718 if (major == __NR_vm86 || major == __NR_vm86old)
723 "audit(:%d) pid=%d in syscall=%d;"
724 " entering syscall=%d\n",
725 context->serial, tsk->pid, context->major, major);
727 newctx = audit_alloc_context(context->state);
729 newctx->previous = context;
731 tsk->audit_context = newctx;
733 /* If we can't alloc a new context, the best we
734 * can do is to leak memory (any pending putname
735 * will be lost). The only other alternative is
736 * to abandon auditing. */
737 audit_zero_context(context, context->state);
740 BUG_ON(context->in_syscall || context->name_count);
745 context->major = major;
746 context->argv[0] = a1;
747 context->argv[1] = a2;
748 context->argv[2] = a3;
749 context->argv[3] = a4;
751 state = context->state;
752 if (state == AUDIT_SETUP_CONTEXT || state == AUDIT_BUILD_CONTEXT)
753 state = audit_filter_syscall(tsk, context, &audit_entlist);
754 if (likely(state == AUDIT_DISABLED))
757 context->serial = audit_serial();
758 context->ctime = CURRENT_TIME;
759 context->in_syscall = 1;
760 context->auditable = !!(state == AUDIT_RECORD_CONTEXT);
763 /* Tear down after system call. If the audit context has been marked as
764 * auditable (either because of the AUDIT_RECORD_CONTEXT state from
765 * filtering, or because some other part of the kernel write an audit
766 * message), then write out the syscall information. In call cases,
767 * free the names stored from getname(). */
768 void audit_syscall_exit(struct task_struct *tsk, int return_code)
770 struct audit_context *context;
772 get_task_struct(tsk);
774 context = audit_get_context(tsk, 1, return_code);
777 /* Not having a context here is ok, since the parent may have
778 * called __put_task_struct. */
779 if (likely(!context))
782 if (context->in_syscall && context->auditable)
783 audit_log_exit(context);
785 context->in_syscall = 0;
786 context->auditable = 0;
787 if (context->previous) {
788 struct audit_context *new_context = context->previous;
789 context->previous = NULL;
790 audit_free_context(context);
791 tsk->audit_context = new_context;
793 audit_free_names(context);
794 audit_zero_context(context, context->state);
795 tsk->audit_context = context;
797 put_task_struct(tsk);
800 /* Add a name to the list. Called from fs/namei.c:getname(). */
801 void audit_getname(const char *name)
803 struct audit_context *context = current->audit_context;
806 if (!context->in_syscall) {
808 printk(KERN_ERR "%s:%d(:%d): ignoring getname(%p)\n",
809 __FILE__, __LINE__, context->serial, name);
814 BUG_ON(context->name_count >= AUDIT_NAMES);
815 context->names[context->name_count].name = name;
816 context->names[context->name_count].ino = (unsigned long)-1;
817 context->names[context->name_count].rdev = -1;
818 ++context->name_count;
821 /* Intercept a putname request. Called from
822 * include/linux/fs.h:putname(). If we have stored the name from
823 * getname in the audit context, then we delay the putname until syscall
825 void audit_putname(const char *name)
827 struct audit_context *context = current->audit_context;
830 if (!context->in_syscall) {
832 printk(KERN_ERR "%s:%d(:%d): __putname(%p)\n",
833 __FILE__, __LINE__, context->serial, name);
834 if (context->name_count) {
836 for (i = 0; i < context->name_count; i++)
837 printk(KERN_ERR "name[%d] = %p = %s\n", i,
838 context->names[i].name,
839 context->names[i].name);
846 ++context->put_count;
847 if (context->put_count > context->name_count) {
848 printk(KERN_ERR "%s:%d(:%d): major=%d"
849 " in_syscall=%d putname(%p) name_count=%d"
852 context->serial, context->major,
853 context->in_syscall, name, context->name_count,
860 EXPORT_SYMBOL(audit_putname);
862 /* Store the inode and device from a lookup. Called from
863 * fs/namei.c:path_lookup(). */
864 void audit_inode(const char *name, unsigned long ino, dev_t rdev)
867 struct audit_context *context = current->audit_context;
869 if (!context->in_syscall)
871 if (context->name_count
872 && context->names[context->name_count-1].name
873 && context->names[context->name_count-1].name == name)
874 idx = context->name_count - 1;
875 else if (context->name_count > 1
876 && context->names[context->name_count-2].name
877 && context->names[context->name_count-2].name == name)
878 idx = context->name_count - 2;
880 /* FIXME: how much do we care about inodes that have no
881 * associated name? */
882 if (context->name_count >= AUDIT_NAMES - AUDIT_NAMES_RESERVED)
884 idx = context->name_count++;
885 context->names[idx].name = NULL;
887 ++context->ino_count;
890 context->names[idx].ino = ino;
891 context->names[idx].rdev = rdev;
894 void audit_get_stamp(struct audit_context *ctx,
895 struct timespec *t, int *serial)
898 t->tv_sec = ctx->ctime.tv_sec;
899 t->tv_nsec = ctx->ctime.tv_nsec;
900 *serial = ctx->serial;
908 int audit_set_loginuid(struct audit_context *ctx, uid_t loginuid)
913 ctx->loginuid = loginuid;