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 (!(entry = kmalloc(sizeof(*entry), GFP_KERNEL)))
255 if (audit_copy_rule(&entry->rule, data)) {
259 flags = entry->rule.flags;
260 if (!err && (flags & AUDIT_PER_TASK))
261 err = audit_add_rule(entry, &audit_tsklist);
262 if (!err && (flags & AUDIT_AT_ENTRY))
263 err = audit_add_rule(entry, &audit_entlist);
264 if (!err && (flags & AUDIT_AT_EXIT))
265 err = audit_add_rule(entry, &audit_extlist);
268 flags =((struct audit_rule *)data)->flags;
269 if (!err && (flags & AUDIT_PER_TASK))
270 err = audit_del_rule(data, &audit_tsklist);
271 if (!err && (flags & AUDIT_AT_ENTRY))
272 err = audit_del_rule(data, &audit_entlist);
273 if (!err && (flags & AUDIT_AT_EXIT))
274 err = audit_del_rule(data, &audit_extlist);
284 /* Compare a task_struct with an audit_rule. Return 1 on match, 0
286 static int audit_filter_rules(struct task_struct *tsk,
287 struct audit_rule *rule,
288 struct audit_context *ctx,
289 enum audit_state *state)
293 for (i = 0; i < rule->field_count; i++) {
294 u32 field = rule->fields[i] & ~AUDIT_NEGATE;
295 u32 value = rule->values[i];
300 result = (tsk->pid == value);
303 result = (tsk->uid == value);
306 result = (tsk->euid == value);
309 result = (tsk->suid == value);
312 result = (tsk->fsuid == value);
315 result = (tsk->gid == value);
318 result = (tsk->egid == value);
321 result = (tsk->sgid == value);
324 result = (tsk->fsgid == value);
327 result = (tsk->personality == value);
331 if (ctx && ctx->return_valid)
332 result = (ctx->return_code == value);
335 if (ctx && ctx->return_valid)
336 result = (ctx->return_code >= 0);
340 for (j = 0; j < ctx->name_count; j++) {
341 if (MAJOR(ctx->names[j].rdev)==value) {
350 for (j = 0; j < ctx->name_count; j++) {
351 if (MINOR(ctx->names[j].rdev)==value) {
360 for (j = 0; j < ctx->name_count; j++) {
361 if (ctx->names[j].ino == value) {
371 result = (ctx->loginuid == value);
378 result = (ctx->argv[field-AUDIT_ARG0]==value);
382 if (rule->fields[i] & AUDIT_NEGATE)
387 switch (rule->action) {
388 case AUDIT_NEVER: *state = AUDIT_DISABLED; break;
389 case AUDIT_POSSIBLE: *state = AUDIT_BUILD_CONTEXT; break;
390 case AUDIT_ALWAYS: *state = AUDIT_RECORD_CONTEXT; break;
395 /* At process creation time, we can determine if system-call auditing is
396 * completely disabled for this task. Since we only have the task
397 * structure at this point, we can only check uid and gid.
399 static enum audit_state audit_filter_task(struct task_struct *tsk)
401 struct audit_entry *e;
402 enum audit_state state;
405 list_for_each_entry_rcu(e, &audit_tsklist, list) {
406 if (audit_filter_rules(tsk, &e->rule, NULL, &state)) {
412 return AUDIT_BUILD_CONTEXT;
415 /* At syscall entry and exit time, this filter is called if the
416 * audit_state is not low enough that auditing cannot take place, but is
417 * also not high enough that we already know we have to write and audit
418 * record (i.e., the state is AUDIT_SETUP_CONTEXT or AUDIT_BUILD_CONTEXT).
420 static enum audit_state audit_filter_syscall(struct task_struct *tsk,
421 struct audit_context *ctx,
422 struct list_head *list)
424 struct audit_entry *e;
425 enum audit_state state;
426 int word = AUDIT_WORD(ctx->major);
427 int bit = AUDIT_BIT(ctx->major);
430 list_for_each_entry_rcu(e, list, list) {
431 if ((e->rule.mask[word] & bit) == bit
432 && audit_filter_rules(tsk, &e->rule, ctx, &state)) {
438 return AUDIT_BUILD_CONTEXT;
441 /* This should be called with task_lock() held. */
442 static inline struct audit_context *audit_get_context(struct task_struct *tsk,
446 struct audit_context *context = tsk->audit_context;
448 if (likely(!context))
450 context->return_valid = return_valid;
451 context->return_code = return_code;
453 if (context->in_syscall && !context->auditable) {
454 enum audit_state state;
455 state = audit_filter_syscall(tsk, context, &audit_extlist);
456 if (state == AUDIT_RECORD_CONTEXT)
457 context->auditable = 1;
460 context->pid = tsk->pid;
461 context->uid = tsk->uid;
462 context->gid = tsk->gid;
463 context->euid = tsk->euid;
464 context->suid = tsk->suid;
465 context->fsuid = tsk->fsuid;
466 context->egid = tsk->egid;
467 context->sgid = tsk->sgid;
468 context->fsgid = tsk->fsgid;
469 context->personality = tsk->personality;
470 tsk->audit_context = NULL;
474 static inline void audit_free_names(struct audit_context *context)
479 if (context->auditable
480 ||context->put_count + context->ino_count != context->name_count) {
481 printk(KERN_ERR "audit.c:%d(:%d): major=%d in_syscall=%d"
482 " name_count=%d put_count=%d"
483 " ino_count=%d [NOT freeing]\n",
485 context->serial, context->major, context->in_syscall,
486 context->name_count, context->put_count,
488 for (i = 0; i < context->name_count; i++)
489 printk(KERN_ERR "names[%d] = %p = %s\n", i,
490 context->names[i].name,
491 context->names[i].name);
497 context->put_count = 0;
498 context->ino_count = 0;
501 for (i = 0; i < context->name_count; i++)
502 if (context->names[i].name)
503 __putname(context->names[i].name);
504 context->name_count = 0;
507 static inline void audit_zero_context(struct audit_context *context,
508 enum audit_state state)
510 uid_t loginuid = context->loginuid;
512 memset(context, 0, sizeof(*context));
513 context->state = state;
514 context->loginuid = loginuid;
517 static inline struct audit_context *audit_alloc_context(enum audit_state state)
519 struct audit_context *context;
521 if (!(context = kmalloc(sizeof(*context), GFP_KERNEL)))
523 audit_zero_context(context, state);
527 /* Filter on the task information and allocate a per-task audit context
528 * if necessary. Doing so turns on system call auditing for the
529 * specified task. This is called from copy_process, so no lock is
531 int audit_alloc(struct task_struct *tsk)
533 struct audit_context *context;
534 enum audit_state state;
536 if (likely(!audit_enabled))
537 return 0; /* Return if not auditing. */
539 state = audit_filter_task(tsk);
540 if (likely(state == AUDIT_DISABLED))
543 if (!(context = audit_alloc_context(state))) {
544 audit_log_lost("out of memory in audit_alloc");
548 /* Preserve login uid */
549 context->loginuid = -1;
550 if (current->audit_context)
551 context->loginuid = current->audit_context->loginuid;
553 tsk->audit_context = context;
554 set_tsk_thread_flag(tsk, TIF_SYSCALL_AUDIT);
558 static inline void audit_free_context(struct audit_context *context)
560 struct audit_context *previous;
564 previous = context->previous;
565 if (previous || (count && count < 10)) {
567 printk(KERN_ERR "audit(:%d): major=%d name_count=%d:"
568 " freeing multiple contexts (%d)\n",
569 context->serial, context->major,
570 context->name_count, count);
572 audit_free_names(context);
577 printk(KERN_ERR "audit: freed %d contexts\n", count);
580 static void audit_log_exit(struct audit_context *context)
583 struct audit_buffer *ab;
585 ab = audit_log_start(context);
587 return; /* audit_panic has been called */
588 audit_log_format(ab, "syscall=%d", context->major);
589 if (context->personality != PER_LINUX)
590 audit_log_format(ab, " per=%lx", context->personality);
591 if (context->return_valid)
592 audit_log_format(ab, " exit=%d", context->return_code);
594 " a0=%lx a1=%lx a2=%lx a3=%lx items=%d"
595 " pid=%d loginuid=%d uid=%d gid=%d"
596 " euid=%d suid=%d fsuid=%d"
597 " egid=%d sgid=%d fsgid=%d",
607 context->euid, context->suid, context->fsuid,
608 context->egid, context->sgid, context->fsgid);
610 for (i = 0; i < context->name_count; i++) {
611 ab = audit_log_start(context);
613 continue; /* audit_panic has been called */
614 audit_log_format(ab, "item=%d", i);
615 if (context->names[i].name)
616 audit_log_format(ab, " name=%s",
617 context->names[i].name);
618 if (context->names[i].ino != (unsigned long)-1)
619 audit_log_format(ab, " inode=%lu",
620 context->names[i].ino);
621 /* FIXME: should use format_dev_t, but ab structure is
623 if (context->names[i].rdev != -1)
624 audit_log_format(ab, " dev=%02x:%02x",
625 MAJOR(context->names[i].rdev),
626 MINOR(context->names[i].rdev));
631 /* Free a per-task audit context. Called from copy_process and
632 * __put_task_struct. */
633 void audit_free(struct task_struct *tsk)
635 struct audit_context *context;
638 context = audit_get_context(tsk, 0, 0);
641 if (likely(!context))
644 /* Check for system calls that do not go through the exit
645 * function (e.g., exit_group), then free context block. */
646 if (context->in_syscall && context->auditable)
647 audit_log_exit(context);
649 audit_free_context(context);
652 /* Compute a serial number for the audit record. Audit records are
653 * written to user-space as soon as they are generated, so a complete
654 * audit record may be written in several pieces. The timestamp of the
655 * record and this serial number are used by the user-space daemon to
656 * determine which pieces belong to the same audit record. The
657 * (timestamp,serial) tuple is unique for each syscall and is live from
658 * syscall entry to syscall exit.
660 * Atomic values are only guaranteed to be 24-bit, so we count down.
662 * NOTE: Another possibility is to store the formatted records off the
663 * audit context (for those records that have a context), and emit them
664 * all at syscall exit. However, this could delay the reporting of
665 * significant errors until syscall exit (or never, if the system
667 static inline unsigned int audit_serial(void)
669 static atomic_t serial = ATOMIC_INIT(0xffffff);
673 a = atomic_read(&serial);
674 if (atomic_dec_and_test(&serial))
675 atomic_set(&serial, 0xffffff);
676 b = atomic_read(&serial);
677 } while (b != a - 1);
682 /* Fill in audit context at syscall entry. This only happens if the
683 * audit context was created when the task was created and the state or
684 * filters demand the audit context be built. If the state from the
685 * per-task filter or from the per-syscall filter is AUDIT_RECORD_CONTEXT,
686 * then the record will be written at syscall exit time (otherwise, it
687 * will only be written if another part of the kernel requests that it
689 void audit_syscall_entry(struct task_struct *tsk, int major,
690 unsigned long a1, unsigned long a2,
691 unsigned long a3, unsigned long a4)
693 struct audit_context *context = tsk->audit_context;
694 enum audit_state state;
698 /* This happens only on certain architectures that make system
699 * calls in kernel_thread via the entry.S interface, instead of
700 * with direct calls. (If you are porting to a new
701 * architecture, hitting this condition can indicate that you
702 * got the _exit/_leave calls backward in entry.S.)
706 * ppc64 yes (see arch/ppc64/kernel/misc.S)
708 * This also happens with vm86 emulation in a non-nested manner
709 * (entries without exits), so this case must be caught.
711 if (context->in_syscall) {
712 struct audit_context *newctx;
714 #if defined(__NR_vm86) && defined(__NR_vm86old)
715 /* vm86 mode should only be entered once */
716 if (major == __NR_vm86 || major == __NR_vm86old)
721 "audit(:%d) pid=%d in syscall=%d;"
722 " entering syscall=%d\n",
723 context->serial, tsk->pid, context->major, major);
725 newctx = audit_alloc_context(context->state);
727 newctx->previous = context;
729 tsk->audit_context = newctx;
731 /* If we can't alloc a new context, the best we
732 * can do is to leak memory (any pending putname
733 * will be lost). The only other alternative is
734 * to abandon auditing. */
735 audit_zero_context(context, context->state);
738 BUG_ON(context->in_syscall || context->name_count);
743 context->major = major;
744 context->argv[0] = a1;
745 context->argv[1] = a2;
746 context->argv[2] = a3;
747 context->argv[3] = a4;
749 state = context->state;
750 if (state == AUDIT_SETUP_CONTEXT || state == AUDIT_BUILD_CONTEXT)
751 state = audit_filter_syscall(tsk, context, &audit_entlist);
752 if (likely(state == AUDIT_DISABLED))
755 context->serial = audit_serial();
756 context->ctime = CURRENT_TIME;
757 context->in_syscall = 1;
758 context->auditable = !!(state == AUDIT_RECORD_CONTEXT);
761 /* Tear down after system call. If the audit context has been marked as
762 * auditable (either because of the AUDIT_RECORD_CONTEXT state from
763 * filtering, or because some other part of the kernel write an audit
764 * message), then write out the syscall information. In call cases,
765 * free the names stored from getname(). */
766 void audit_syscall_exit(struct task_struct *tsk, int return_code)
768 struct audit_context *context;
770 get_task_struct(tsk);
772 context = audit_get_context(tsk, 1, return_code);
775 /* Not having a context here is ok, since the parent may have
776 * called __put_task_struct. */
777 if (likely(!context))
780 if (context->in_syscall && context->auditable)
781 audit_log_exit(context);
783 context->in_syscall = 0;
784 context->auditable = 0;
785 if (context->previous) {
786 struct audit_context *new_context = context->previous;
787 context->previous = NULL;
788 audit_free_context(context);
789 tsk->audit_context = new_context;
791 audit_free_names(context);
792 audit_zero_context(context, context->state);
793 tsk->audit_context = context;
795 put_task_struct(tsk);
798 /* Add a name to the list. Called from fs/namei.c:getname(). */
799 void audit_getname(const char *name)
801 struct audit_context *context = current->audit_context;
804 if (!context->in_syscall) {
806 printk(KERN_ERR "%s:%d(:%d): ignoring getname(%p)\n",
807 __FILE__, __LINE__, context->serial, name);
812 BUG_ON(context->name_count >= AUDIT_NAMES);
813 context->names[context->name_count].name = name;
814 context->names[context->name_count].ino = (unsigned long)-1;
815 context->names[context->name_count].rdev = -1;
816 ++context->name_count;
819 /* Intercept a putname request. Called from
820 * include/linux/fs.h:putname(). If we have stored the name from
821 * getname in the audit context, then we delay the putname until syscall
823 void audit_putname(const char *name)
825 struct audit_context *context = current->audit_context;
828 if (!context->in_syscall) {
830 printk(KERN_ERR "%s:%d(:%d): __putname(%p)\n",
831 __FILE__, __LINE__, context->serial, name);
832 if (context->name_count) {
834 for (i = 0; i < context->name_count; i++)
835 printk(KERN_ERR "name[%d] = %p = %s\n", i,
836 context->names[i].name,
837 context->names[i].name);
844 ++context->put_count;
845 if (context->put_count > context->name_count) {
846 printk(KERN_ERR "%s:%d(:%d): major=%d"
847 " in_syscall=%d putname(%p) name_count=%d"
850 context->serial, context->major,
851 context->in_syscall, name, context->name_count,
858 EXPORT_SYMBOL(audit_putname);
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 extern int audit_set_type(struct audit_buffer *ab, int type);
908 int audit_set_loginuid(struct audit_context *ctx, uid_t loginuid)
911 struct audit_buffer *ab;
913 ab = audit_log_start(NULL);
915 audit_log_format(ab, "login pid=%d uid=%u "
916 "old loginuid=%u new loginuid=%u",
917 ctx->pid, ctx->uid, ctx->loginuid, loginuid);
918 audit_set_type(ab, AUDIT_LOGIN);
921 ctx->loginuid = loginuid;
926 uid_t audit_get_loginuid(struct audit_context *ctx)
928 return ctx ? ctx->loginuid : -1;