xref: /linux/kernel/auditsc.c (revision bcefe12eff5dca6fdfa94ed85e5bee66380d5cd9)
1 /* auditsc.c -- System-call auditing support
2  * Handles all system-call specific auditing features.
3  *
4  * Copyright 2003-2004 Red Hat Inc., Durham, North Carolina.
5  * Copyright 2005 Hewlett-Packard Development Company, L.P.
6  * Copyright (C) 2005, 2006 IBM Corporation
7  * All Rights Reserved.
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License as published by
11  * the Free Software Foundation; either version 2 of the License, or
12  * (at your option) any later version.
13  *
14  * This program is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
17  * GNU General Public License for more details.
18  *
19  * You should have received a copy of the GNU General Public License
20  * along with this program; if not, write to the Free Software
21  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
22  *
23  * Written by Rickard E. (Rik) Faith <faith@redhat.com>
24  *
25  * Many of the ideas implemented here are from Stephen C. Tweedie,
26  * especially the idea of avoiding a copy by using getname.
27  *
28  * The method for actual interception of syscall entry and exit (not in
29  * this file -- see entry.S) is based on a GPL'd patch written by
30  * okir@suse.de and Copyright 2003 SuSE Linux AG.
31  *
32  * POSIX message queue support added by George Wilson <ltcgcw@us.ibm.com>,
33  * 2006.
34  *
35  * The support of additional filter rules compares (>, <, >=, <=) was
36  * added by Dustin Kirkland <dustin.kirkland@us.ibm.com>, 2005.
37  *
38  * Modified by Amy Griffis <amy.griffis@hp.com> to collect additional
39  * filesystem information.
40  *
41  * Subject and object context labeling support added by <danjones@us.ibm.com>
42  * and <dustin.kirkland@us.ibm.com> for LSPP certification compliance.
43  */
44 
45 #include <linux/init.h>
46 #include <asm/types.h>
47 #include <asm/atomic.h>
48 #include <linux/fs.h>
49 #include <linux/namei.h>
50 #include <linux/mm.h>
51 #include <linux/module.h>
52 #include <linux/mount.h>
53 #include <linux/socket.h>
54 #include <linux/mqueue.h>
55 #include <linux/audit.h>
56 #include <linux/personality.h>
57 #include <linux/time.h>
58 #include <linux/netlink.h>
59 #include <linux/compiler.h>
60 #include <asm/unistd.h>
61 #include <linux/security.h>
62 #include <linux/list.h>
63 #include <linux/tty.h>
64 #include <linux/binfmts.h>
65 #include <linux/highmem.h>
66 #include <linux/syscalls.h>
67 #include <linux/inotify.h>
68 #include <linux/capability.h>
69 #include <linux/fs_struct.h>
70 
71 #include "audit.h"
72 
73 /* AUDIT_NAMES is the number of slots we reserve in the audit_context
74  * for saving names from getname(). */
75 #define AUDIT_NAMES    20
76 
77 /* Indicates that audit should log the full pathname. */
78 #define AUDIT_NAME_FULL -1
79 
80 /* no execve audit message should be longer than this (userspace limits) */
81 #define MAX_EXECVE_AUDIT_LEN 7500
82 
83 /* number of audit rules */
84 int audit_n_rules;
85 
86 /* determines whether we collect data for signals sent */
87 int audit_signals;
88 
89 struct audit_cap_data {
90 	kernel_cap_t		permitted;
91 	kernel_cap_t		inheritable;
92 	union {
93 		unsigned int	fE;		/* effective bit of a file capability */
94 		kernel_cap_t	effective;	/* effective set of a process */
95 	};
96 };
97 
98 /* When fs/namei.c:getname() is called, we store the pointer in name and
99  * we don't let putname() free it (instead we free all of the saved
100  * pointers at syscall exit time).
101  *
102  * Further, in fs/namei.c:path_lookup() we store the inode and device. */
103 struct audit_names {
104 	const char	*name;
105 	int		name_len;	/* number of name's characters to log */
106 	unsigned	name_put;	/* call __putname() for this name */
107 	unsigned long	ino;
108 	dev_t		dev;
109 	umode_t		mode;
110 	uid_t		uid;
111 	gid_t		gid;
112 	dev_t		rdev;
113 	u32		osid;
114 	struct audit_cap_data fcap;
115 	unsigned int	fcap_ver;
116 };
117 
118 struct audit_aux_data {
119 	struct audit_aux_data	*next;
120 	int			type;
121 };
122 
123 #define AUDIT_AUX_IPCPERM	0
124 
125 /* Number of target pids per aux struct. */
126 #define AUDIT_AUX_PIDS	16
127 
128 struct audit_aux_data_execve {
129 	struct audit_aux_data	d;
130 	int argc;
131 	int envc;
132 	struct mm_struct *mm;
133 };
134 
135 struct audit_aux_data_pids {
136 	struct audit_aux_data	d;
137 	pid_t			target_pid[AUDIT_AUX_PIDS];
138 	uid_t			target_auid[AUDIT_AUX_PIDS];
139 	uid_t			target_uid[AUDIT_AUX_PIDS];
140 	unsigned int		target_sessionid[AUDIT_AUX_PIDS];
141 	u32			target_sid[AUDIT_AUX_PIDS];
142 	char 			target_comm[AUDIT_AUX_PIDS][TASK_COMM_LEN];
143 	int			pid_count;
144 };
145 
146 struct audit_aux_data_bprm_fcaps {
147 	struct audit_aux_data	d;
148 	struct audit_cap_data	fcap;
149 	unsigned int		fcap_ver;
150 	struct audit_cap_data	old_pcap;
151 	struct audit_cap_data	new_pcap;
152 };
153 
154 struct audit_aux_data_capset {
155 	struct audit_aux_data	d;
156 	pid_t			pid;
157 	struct audit_cap_data	cap;
158 };
159 
160 struct audit_tree_refs {
161 	struct audit_tree_refs *next;
162 	struct audit_chunk *c[31];
163 };
164 
165 /* The per-task audit context. */
166 struct audit_context {
167 	int		    dummy;	/* must be the first element */
168 	int		    in_syscall;	/* 1 if task is in a syscall */
169 	enum audit_state    state, current_state;
170 	unsigned int	    serial;     /* serial number for record */
171 	int		    major;      /* syscall number */
172 	struct timespec	    ctime;      /* time of syscall entry */
173 	unsigned long	    argv[4];    /* syscall arguments */
174 	long		    return_code;/* syscall return code */
175 	u64		    prio;
176 	int		    return_valid; /* return code is valid */
177 	int		    name_count;
178 	struct audit_names  names[AUDIT_NAMES];
179 	char *		    filterkey;	/* key for rule that triggered record */
180 	struct path	    pwd;
181 	struct audit_context *previous; /* For nested syscalls */
182 	struct audit_aux_data *aux;
183 	struct audit_aux_data *aux_pids;
184 	struct sockaddr_storage *sockaddr;
185 	size_t sockaddr_len;
186 				/* Save things to print about task_struct */
187 	pid_t		    pid, ppid;
188 	uid_t		    uid, euid, suid, fsuid;
189 	gid_t		    gid, egid, sgid, fsgid;
190 	unsigned long	    personality;
191 	int		    arch;
192 
193 	pid_t		    target_pid;
194 	uid_t		    target_auid;
195 	uid_t		    target_uid;
196 	unsigned int	    target_sessionid;
197 	u32		    target_sid;
198 	char		    target_comm[TASK_COMM_LEN];
199 
200 	struct audit_tree_refs *trees, *first_trees;
201 	struct list_head killed_trees;
202 	int tree_count;
203 
204 	int type;
205 	union {
206 		struct {
207 			int nargs;
208 			long args[6];
209 		} socketcall;
210 		struct {
211 			uid_t			uid;
212 			gid_t			gid;
213 			mode_t			mode;
214 			u32			osid;
215 			int			has_perm;
216 			uid_t			perm_uid;
217 			gid_t			perm_gid;
218 			mode_t			perm_mode;
219 			unsigned long		qbytes;
220 		} ipc;
221 		struct {
222 			mqd_t			mqdes;
223 			struct mq_attr 		mqstat;
224 		} mq_getsetattr;
225 		struct {
226 			mqd_t			mqdes;
227 			int			sigev_signo;
228 		} mq_notify;
229 		struct {
230 			mqd_t			mqdes;
231 			size_t			msg_len;
232 			unsigned int		msg_prio;
233 			struct timespec		abs_timeout;
234 		} mq_sendrecv;
235 		struct {
236 			int			oflag;
237 			mode_t			mode;
238 			struct mq_attr		attr;
239 		} mq_open;
240 		struct {
241 			pid_t			pid;
242 			struct audit_cap_data	cap;
243 		} capset;
244 	};
245 	int fds[2];
246 
247 #if AUDIT_DEBUG
248 	int		    put_count;
249 	int		    ino_count;
250 #endif
251 };
252 
253 #define ACC_MODE(x) ("\004\002\006\006"[(x)&O_ACCMODE])
254 static inline int open_arg(int flags, int mask)
255 {
256 	int n = ACC_MODE(flags);
257 	if (flags & (O_TRUNC | O_CREAT))
258 		n |= AUDIT_PERM_WRITE;
259 	return n & mask;
260 }
261 
262 static int audit_match_perm(struct audit_context *ctx, int mask)
263 {
264 	unsigned n;
265 	if (unlikely(!ctx))
266 		return 0;
267 	n = ctx->major;
268 
269 	switch (audit_classify_syscall(ctx->arch, n)) {
270 	case 0:	/* native */
271 		if ((mask & AUDIT_PERM_WRITE) &&
272 		     audit_match_class(AUDIT_CLASS_WRITE, n))
273 			return 1;
274 		if ((mask & AUDIT_PERM_READ) &&
275 		     audit_match_class(AUDIT_CLASS_READ, n))
276 			return 1;
277 		if ((mask & AUDIT_PERM_ATTR) &&
278 		     audit_match_class(AUDIT_CLASS_CHATTR, n))
279 			return 1;
280 		return 0;
281 	case 1: /* 32bit on biarch */
282 		if ((mask & AUDIT_PERM_WRITE) &&
283 		     audit_match_class(AUDIT_CLASS_WRITE_32, n))
284 			return 1;
285 		if ((mask & AUDIT_PERM_READ) &&
286 		     audit_match_class(AUDIT_CLASS_READ_32, n))
287 			return 1;
288 		if ((mask & AUDIT_PERM_ATTR) &&
289 		     audit_match_class(AUDIT_CLASS_CHATTR_32, n))
290 			return 1;
291 		return 0;
292 	case 2: /* open */
293 		return mask & ACC_MODE(ctx->argv[1]);
294 	case 3: /* openat */
295 		return mask & ACC_MODE(ctx->argv[2]);
296 	case 4: /* socketcall */
297 		return ((mask & AUDIT_PERM_WRITE) && ctx->argv[0] == SYS_BIND);
298 	case 5: /* execve */
299 		return mask & AUDIT_PERM_EXEC;
300 	default:
301 		return 0;
302 	}
303 }
304 
305 static int audit_match_filetype(struct audit_context *ctx, int which)
306 {
307 	unsigned index = which & ~S_IFMT;
308 	mode_t mode = which & S_IFMT;
309 
310 	if (unlikely(!ctx))
311 		return 0;
312 
313 	if (index >= ctx->name_count)
314 		return 0;
315 	if (ctx->names[index].ino == -1)
316 		return 0;
317 	if ((ctx->names[index].mode ^ mode) & S_IFMT)
318 		return 0;
319 	return 1;
320 }
321 
322 /*
323  * We keep a linked list of fixed-sized (31 pointer) arrays of audit_chunk *;
324  * ->first_trees points to its beginning, ->trees - to the current end of data.
325  * ->tree_count is the number of free entries in array pointed to by ->trees.
326  * Original condition is (NULL, NULL, 0); as soon as it grows we never revert to NULL,
327  * "empty" becomes (p, p, 31) afterwards.  We don't shrink the list (and seriously,
328  * it's going to remain 1-element for almost any setup) until we free context itself.
329  * References in it _are_ dropped - at the same time we free/drop aux stuff.
330  */
331 
332 #ifdef CONFIG_AUDIT_TREE
333 static void audit_set_auditable(struct audit_context *ctx)
334 {
335 	if (!ctx->prio) {
336 		ctx->prio = 1;
337 		ctx->current_state = AUDIT_RECORD_CONTEXT;
338 	}
339 }
340 
341 static int put_tree_ref(struct audit_context *ctx, struct audit_chunk *chunk)
342 {
343 	struct audit_tree_refs *p = ctx->trees;
344 	int left = ctx->tree_count;
345 	if (likely(left)) {
346 		p->c[--left] = chunk;
347 		ctx->tree_count = left;
348 		return 1;
349 	}
350 	if (!p)
351 		return 0;
352 	p = p->next;
353 	if (p) {
354 		p->c[30] = chunk;
355 		ctx->trees = p;
356 		ctx->tree_count = 30;
357 		return 1;
358 	}
359 	return 0;
360 }
361 
362 static int grow_tree_refs(struct audit_context *ctx)
363 {
364 	struct audit_tree_refs *p = ctx->trees;
365 	ctx->trees = kzalloc(sizeof(struct audit_tree_refs), GFP_KERNEL);
366 	if (!ctx->trees) {
367 		ctx->trees = p;
368 		return 0;
369 	}
370 	if (p)
371 		p->next = ctx->trees;
372 	else
373 		ctx->first_trees = ctx->trees;
374 	ctx->tree_count = 31;
375 	return 1;
376 }
377 #endif
378 
379 static void unroll_tree_refs(struct audit_context *ctx,
380 		      struct audit_tree_refs *p, int count)
381 {
382 #ifdef CONFIG_AUDIT_TREE
383 	struct audit_tree_refs *q;
384 	int n;
385 	if (!p) {
386 		/* we started with empty chain */
387 		p = ctx->first_trees;
388 		count = 31;
389 		/* if the very first allocation has failed, nothing to do */
390 		if (!p)
391 			return;
392 	}
393 	n = count;
394 	for (q = p; q != ctx->trees; q = q->next, n = 31) {
395 		while (n--) {
396 			audit_put_chunk(q->c[n]);
397 			q->c[n] = NULL;
398 		}
399 	}
400 	while (n-- > ctx->tree_count) {
401 		audit_put_chunk(q->c[n]);
402 		q->c[n] = NULL;
403 	}
404 	ctx->trees = p;
405 	ctx->tree_count = count;
406 #endif
407 }
408 
409 static void free_tree_refs(struct audit_context *ctx)
410 {
411 	struct audit_tree_refs *p, *q;
412 	for (p = ctx->first_trees; p; p = q) {
413 		q = p->next;
414 		kfree(p);
415 	}
416 }
417 
418 static int match_tree_refs(struct audit_context *ctx, struct audit_tree *tree)
419 {
420 #ifdef CONFIG_AUDIT_TREE
421 	struct audit_tree_refs *p;
422 	int n;
423 	if (!tree)
424 		return 0;
425 	/* full ones */
426 	for (p = ctx->first_trees; p != ctx->trees; p = p->next) {
427 		for (n = 0; n < 31; n++)
428 			if (audit_tree_match(p->c[n], tree))
429 				return 1;
430 	}
431 	/* partial */
432 	if (p) {
433 		for (n = ctx->tree_count; n < 31; n++)
434 			if (audit_tree_match(p->c[n], tree))
435 				return 1;
436 	}
437 #endif
438 	return 0;
439 }
440 
441 /* Determine if any context name data matches a rule's watch data */
442 /* Compare a task_struct with an audit_rule.  Return 1 on match, 0
443  * otherwise. */
444 static int audit_filter_rules(struct task_struct *tsk,
445 			      struct audit_krule *rule,
446 			      struct audit_context *ctx,
447 			      struct audit_names *name,
448 			      enum audit_state *state)
449 {
450 	const struct cred *cred = get_task_cred(tsk);
451 	int i, j, need_sid = 1;
452 	u32 sid;
453 
454 	for (i = 0; i < rule->field_count; i++) {
455 		struct audit_field *f = &rule->fields[i];
456 		int result = 0;
457 
458 		switch (f->type) {
459 		case AUDIT_PID:
460 			result = audit_comparator(tsk->pid, f->op, f->val);
461 			break;
462 		case AUDIT_PPID:
463 			if (ctx) {
464 				if (!ctx->ppid)
465 					ctx->ppid = sys_getppid();
466 				result = audit_comparator(ctx->ppid, f->op, f->val);
467 			}
468 			break;
469 		case AUDIT_UID:
470 			result = audit_comparator(cred->uid, f->op, f->val);
471 			break;
472 		case AUDIT_EUID:
473 			result = audit_comparator(cred->euid, f->op, f->val);
474 			break;
475 		case AUDIT_SUID:
476 			result = audit_comparator(cred->suid, f->op, f->val);
477 			break;
478 		case AUDIT_FSUID:
479 			result = audit_comparator(cred->fsuid, f->op, f->val);
480 			break;
481 		case AUDIT_GID:
482 			result = audit_comparator(cred->gid, f->op, f->val);
483 			break;
484 		case AUDIT_EGID:
485 			result = audit_comparator(cred->egid, f->op, f->val);
486 			break;
487 		case AUDIT_SGID:
488 			result = audit_comparator(cred->sgid, f->op, f->val);
489 			break;
490 		case AUDIT_FSGID:
491 			result = audit_comparator(cred->fsgid, f->op, f->val);
492 			break;
493 		case AUDIT_PERS:
494 			result = audit_comparator(tsk->personality, f->op, f->val);
495 			break;
496 		case AUDIT_ARCH:
497 			if (ctx)
498 				result = audit_comparator(ctx->arch, f->op, f->val);
499 			break;
500 
501 		case AUDIT_EXIT:
502 			if (ctx && ctx->return_valid)
503 				result = audit_comparator(ctx->return_code, f->op, f->val);
504 			break;
505 		case AUDIT_SUCCESS:
506 			if (ctx && ctx->return_valid) {
507 				if (f->val)
508 					result = audit_comparator(ctx->return_valid, f->op, AUDITSC_SUCCESS);
509 				else
510 					result = audit_comparator(ctx->return_valid, f->op, AUDITSC_FAILURE);
511 			}
512 			break;
513 		case AUDIT_DEVMAJOR:
514 			if (name)
515 				result = audit_comparator(MAJOR(name->dev),
516 							  f->op, f->val);
517 			else if (ctx) {
518 				for (j = 0; j < ctx->name_count; j++) {
519 					if (audit_comparator(MAJOR(ctx->names[j].dev),	f->op, f->val)) {
520 						++result;
521 						break;
522 					}
523 				}
524 			}
525 			break;
526 		case AUDIT_DEVMINOR:
527 			if (name)
528 				result = audit_comparator(MINOR(name->dev),
529 							  f->op, f->val);
530 			else if (ctx) {
531 				for (j = 0; j < ctx->name_count; j++) {
532 					if (audit_comparator(MINOR(ctx->names[j].dev), f->op, f->val)) {
533 						++result;
534 						break;
535 					}
536 				}
537 			}
538 			break;
539 		case AUDIT_INODE:
540 			if (name)
541 				result = (name->ino == f->val);
542 			else if (ctx) {
543 				for (j = 0; j < ctx->name_count; j++) {
544 					if (audit_comparator(ctx->names[j].ino, f->op, f->val)) {
545 						++result;
546 						break;
547 					}
548 				}
549 			}
550 			break;
551 		case AUDIT_WATCH:
552 			if (name && audit_watch_inode(rule->watch) != (unsigned long)-1)
553 				result = (name->dev == audit_watch_dev(rule->watch) &&
554 					  name->ino == audit_watch_inode(rule->watch));
555 			break;
556 		case AUDIT_DIR:
557 			if (ctx)
558 				result = match_tree_refs(ctx, rule->tree);
559 			break;
560 		case AUDIT_LOGINUID:
561 			result = 0;
562 			if (ctx)
563 				result = audit_comparator(tsk->loginuid, f->op, f->val);
564 			break;
565 		case AUDIT_SUBJ_USER:
566 		case AUDIT_SUBJ_ROLE:
567 		case AUDIT_SUBJ_TYPE:
568 		case AUDIT_SUBJ_SEN:
569 		case AUDIT_SUBJ_CLR:
570 			/* NOTE: this may return negative values indicating
571 			   a temporary error.  We simply treat this as a
572 			   match for now to avoid losing information that
573 			   may be wanted.   An error message will also be
574 			   logged upon error */
575 			if (f->lsm_rule) {
576 				if (need_sid) {
577 					security_task_getsecid(tsk, &sid);
578 					need_sid = 0;
579 				}
580 				result = security_audit_rule_match(sid, f->type,
581 				                                  f->op,
582 				                                  f->lsm_rule,
583 				                                  ctx);
584 			}
585 			break;
586 		case AUDIT_OBJ_USER:
587 		case AUDIT_OBJ_ROLE:
588 		case AUDIT_OBJ_TYPE:
589 		case AUDIT_OBJ_LEV_LOW:
590 		case AUDIT_OBJ_LEV_HIGH:
591 			/* The above note for AUDIT_SUBJ_USER...AUDIT_SUBJ_CLR
592 			   also applies here */
593 			if (f->lsm_rule) {
594 				/* Find files that match */
595 				if (name) {
596 					result = security_audit_rule_match(
597 					           name->osid, f->type, f->op,
598 					           f->lsm_rule, ctx);
599 				} else if (ctx) {
600 					for (j = 0; j < ctx->name_count; j++) {
601 						if (security_audit_rule_match(
602 						      ctx->names[j].osid,
603 						      f->type, f->op,
604 						      f->lsm_rule, ctx)) {
605 							++result;
606 							break;
607 						}
608 					}
609 				}
610 				/* Find ipc objects that match */
611 				if (!ctx || ctx->type != AUDIT_IPC)
612 					break;
613 				if (security_audit_rule_match(ctx->ipc.osid,
614 							      f->type, f->op,
615 							      f->lsm_rule, ctx))
616 					++result;
617 			}
618 			break;
619 		case AUDIT_ARG0:
620 		case AUDIT_ARG1:
621 		case AUDIT_ARG2:
622 		case AUDIT_ARG3:
623 			if (ctx)
624 				result = audit_comparator(ctx->argv[f->type-AUDIT_ARG0], f->op, f->val);
625 			break;
626 		case AUDIT_FILTERKEY:
627 			/* ignore this field for filtering */
628 			result = 1;
629 			break;
630 		case AUDIT_PERM:
631 			result = audit_match_perm(ctx, f->val);
632 			break;
633 		case AUDIT_FILETYPE:
634 			result = audit_match_filetype(ctx, f->val);
635 			break;
636 		}
637 
638 		if (!result) {
639 			put_cred(cred);
640 			return 0;
641 		}
642 	}
643 
644 	if (ctx) {
645 		if (rule->prio <= ctx->prio)
646 			return 0;
647 		if (rule->filterkey) {
648 			kfree(ctx->filterkey);
649 			ctx->filterkey = kstrdup(rule->filterkey, GFP_ATOMIC);
650 		}
651 		ctx->prio = rule->prio;
652 	}
653 	switch (rule->action) {
654 	case AUDIT_NEVER:    *state = AUDIT_DISABLED;	    break;
655 	case AUDIT_ALWAYS:   *state = AUDIT_RECORD_CONTEXT; break;
656 	}
657 	put_cred(cred);
658 	return 1;
659 }
660 
661 /* At process creation time, we can determine if system-call auditing is
662  * completely disabled for this task.  Since we only have the task
663  * structure at this point, we can only check uid and gid.
664  */
665 static enum audit_state audit_filter_task(struct task_struct *tsk, char **key)
666 {
667 	struct audit_entry *e;
668 	enum audit_state   state;
669 
670 	rcu_read_lock();
671 	list_for_each_entry_rcu(e, &audit_filter_list[AUDIT_FILTER_TASK], list) {
672 		if (audit_filter_rules(tsk, &e->rule, NULL, NULL, &state)) {
673 			if (state == AUDIT_RECORD_CONTEXT)
674 				*key = kstrdup(e->rule.filterkey, GFP_ATOMIC);
675 			rcu_read_unlock();
676 			return state;
677 		}
678 	}
679 	rcu_read_unlock();
680 	return AUDIT_BUILD_CONTEXT;
681 }
682 
683 /* At syscall entry and exit time, this filter is called if the
684  * audit_state is not low enough that auditing cannot take place, but is
685  * also not high enough that we already know we have to write an audit
686  * record (i.e., the state is AUDIT_SETUP_CONTEXT or AUDIT_BUILD_CONTEXT).
687  */
688 static enum audit_state audit_filter_syscall(struct task_struct *tsk,
689 					     struct audit_context *ctx,
690 					     struct list_head *list)
691 {
692 	struct audit_entry *e;
693 	enum audit_state state;
694 
695 	if (audit_pid && tsk->tgid == audit_pid)
696 		return AUDIT_DISABLED;
697 
698 	rcu_read_lock();
699 	if (!list_empty(list)) {
700 		int word = AUDIT_WORD(ctx->major);
701 		int bit  = AUDIT_BIT(ctx->major);
702 
703 		list_for_each_entry_rcu(e, list, list) {
704 			if ((e->rule.mask[word] & bit) == bit &&
705 			    audit_filter_rules(tsk, &e->rule, ctx, NULL,
706 					       &state)) {
707 				rcu_read_unlock();
708 				ctx->current_state = state;
709 				return state;
710 			}
711 		}
712 	}
713 	rcu_read_unlock();
714 	return AUDIT_BUILD_CONTEXT;
715 }
716 
717 /* At syscall exit time, this filter is called if any audit_names[] have been
718  * collected during syscall processing.  We only check rules in sublists at hash
719  * buckets applicable to the inode numbers in audit_names[].
720  * Regarding audit_state, same rules apply as for audit_filter_syscall().
721  */
722 void audit_filter_inodes(struct task_struct *tsk, struct audit_context *ctx)
723 {
724 	int i;
725 	struct audit_entry *e;
726 	enum audit_state state;
727 
728 	if (audit_pid && tsk->tgid == audit_pid)
729 		return;
730 
731 	rcu_read_lock();
732 	for (i = 0; i < ctx->name_count; i++) {
733 		int word = AUDIT_WORD(ctx->major);
734 		int bit  = AUDIT_BIT(ctx->major);
735 		struct audit_names *n = &ctx->names[i];
736 		int h = audit_hash_ino((u32)n->ino);
737 		struct list_head *list = &audit_inode_hash[h];
738 
739 		if (list_empty(list))
740 			continue;
741 
742 		list_for_each_entry_rcu(e, list, list) {
743 			if ((e->rule.mask[word] & bit) == bit &&
744 			    audit_filter_rules(tsk, &e->rule, ctx, n, &state)) {
745 				rcu_read_unlock();
746 				ctx->current_state = state;
747 				return;
748 			}
749 		}
750 	}
751 	rcu_read_unlock();
752 }
753 
754 static inline struct audit_context *audit_get_context(struct task_struct *tsk,
755 						      int return_valid,
756 						      long return_code)
757 {
758 	struct audit_context *context = tsk->audit_context;
759 
760 	if (likely(!context))
761 		return NULL;
762 	context->return_valid = return_valid;
763 
764 	/*
765 	 * we need to fix up the return code in the audit logs if the actual
766 	 * return codes are later going to be fixed up by the arch specific
767 	 * signal handlers
768 	 *
769 	 * This is actually a test for:
770 	 * (rc == ERESTARTSYS ) || (rc == ERESTARTNOINTR) ||
771 	 * (rc == ERESTARTNOHAND) || (rc == ERESTART_RESTARTBLOCK)
772 	 *
773 	 * but is faster than a bunch of ||
774 	 */
775 	if (unlikely(return_code <= -ERESTARTSYS) &&
776 	    (return_code >= -ERESTART_RESTARTBLOCK) &&
777 	    (return_code != -ENOIOCTLCMD))
778 		context->return_code = -EINTR;
779 	else
780 		context->return_code  = return_code;
781 
782 	if (context->in_syscall && !context->dummy) {
783 		audit_filter_syscall(tsk, context, &audit_filter_list[AUDIT_FILTER_EXIT]);
784 		audit_filter_inodes(tsk, context);
785 	}
786 
787 	tsk->audit_context = NULL;
788 	return context;
789 }
790 
791 static inline void audit_free_names(struct audit_context *context)
792 {
793 	int i;
794 
795 #if AUDIT_DEBUG == 2
796 	if (context->put_count + context->ino_count != context->name_count) {
797 		printk(KERN_ERR "%s:%d(:%d): major=%d in_syscall=%d"
798 		       " name_count=%d put_count=%d"
799 		       " ino_count=%d [NOT freeing]\n",
800 		       __FILE__, __LINE__,
801 		       context->serial, context->major, context->in_syscall,
802 		       context->name_count, context->put_count,
803 		       context->ino_count);
804 		for (i = 0; i < context->name_count; i++) {
805 			printk(KERN_ERR "names[%d] = %p = %s\n", i,
806 			       context->names[i].name,
807 			       context->names[i].name ?: "(null)");
808 		}
809 		dump_stack();
810 		return;
811 	}
812 #endif
813 #if AUDIT_DEBUG
814 	context->put_count  = 0;
815 	context->ino_count  = 0;
816 #endif
817 
818 	for (i = 0; i < context->name_count; i++) {
819 		if (context->names[i].name && context->names[i].name_put)
820 			__putname(context->names[i].name);
821 	}
822 	context->name_count = 0;
823 	path_put(&context->pwd);
824 	context->pwd.dentry = NULL;
825 	context->pwd.mnt = NULL;
826 }
827 
828 static inline void audit_free_aux(struct audit_context *context)
829 {
830 	struct audit_aux_data *aux;
831 
832 	while ((aux = context->aux)) {
833 		context->aux = aux->next;
834 		kfree(aux);
835 	}
836 	while ((aux = context->aux_pids)) {
837 		context->aux_pids = aux->next;
838 		kfree(aux);
839 	}
840 }
841 
842 static inline void audit_zero_context(struct audit_context *context,
843 				      enum audit_state state)
844 {
845 	memset(context, 0, sizeof(*context));
846 	context->state      = state;
847 	context->prio = state == AUDIT_RECORD_CONTEXT ? ~0ULL : 0;
848 }
849 
850 static inline struct audit_context *audit_alloc_context(enum audit_state state)
851 {
852 	struct audit_context *context;
853 
854 	if (!(context = kmalloc(sizeof(*context), GFP_KERNEL)))
855 		return NULL;
856 	audit_zero_context(context, state);
857 	INIT_LIST_HEAD(&context->killed_trees);
858 	return context;
859 }
860 
861 /**
862  * audit_alloc - allocate an audit context block for a task
863  * @tsk: task
864  *
865  * Filter on the task information and allocate a per-task audit context
866  * if necessary.  Doing so turns on system call auditing for the
867  * specified task.  This is called from copy_process, so no lock is
868  * needed.
869  */
870 int audit_alloc(struct task_struct *tsk)
871 {
872 	struct audit_context *context;
873 	enum audit_state     state;
874 	char *key = NULL;
875 
876 	if (likely(!audit_ever_enabled))
877 		return 0; /* Return if not auditing. */
878 
879 	state = audit_filter_task(tsk, &key);
880 	if (likely(state == AUDIT_DISABLED))
881 		return 0;
882 
883 	if (!(context = audit_alloc_context(state))) {
884 		kfree(key);
885 		audit_log_lost("out of memory in audit_alloc");
886 		return -ENOMEM;
887 	}
888 	context->filterkey = key;
889 
890 	tsk->audit_context  = context;
891 	set_tsk_thread_flag(tsk, TIF_SYSCALL_AUDIT);
892 	return 0;
893 }
894 
895 static inline void audit_free_context(struct audit_context *context)
896 {
897 	struct audit_context *previous;
898 	int		     count = 0;
899 
900 	do {
901 		previous = context->previous;
902 		if (previous || (count &&  count < 10)) {
903 			++count;
904 			printk(KERN_ERR "audit(:%d): major=%d name_count=%d:"
905 			       " freeing multiple contexts (%d)\n",
906 			       context->serial, context->major,
907 			       context->name_count, count);
908 		}
909 		audit_free_names(context);
910 		unroll_tree_refs(context, NULL, 0);
911 		free_tree_refs(context);
912 		audit_free_aux(context);
913 		kfree(context->filterkey);
914 		kfree(context->sockaddr);
915 		kfree(context);
916 		context  = previous;
917 	} while (context);
918 	if (count >= 10)
919 		printk(KERN_ERR "audit: freed %d contexts\n", count);
920 }
921 
922 void audit_log_task_context(struct audit_buffer *ab)
923 {
924 	char *ctx = NULL;
925 	unsigned len;
926 	int error;
927 	u32 sid;
928 
929 	security_task_getsecid(current, &sid);
930 	if (!sid)
931 		return;
932 
933 	error = security_secid_to_secctx(sid, &ctx, &len);
934 	if (error) {
935 		if (error != -EINVAL)
936 			goto error_path;
937 		return;
938 	}
939 
940 	audit_log_format(ab, " subj=%s", ctx);
941 	security_release_secctx(ctx, len);
942 	return;
943 
944 error_path:
945 	audit_panic("error in audit_log_task_context");
946 	return;
947 }
948 
949 EXPORT_SYMBOL(audit_log_task_context);
950 
951 static void audit_log_task_info(struct audit_buffer *ab, struct task_struct *tsk)
952 {
953 	char name[sizeof(tsk->comm)];
954 	struct mm_struct *mm = tsk->mm;
955 	struct vm_area_struct *vma;
956 
957 	/* tsk == current */
958 
959 	get_task_comm(name, tsk);
960 	audit_log_format(ab, " comm=");
961 	audit_log_untrustedstring(ab, name);
962 
963 	if (mm) {
964 		down_read(&mm->mmap_sem);
965 		vma = mm->mmap;
966 		while (vma) {
967 			if ((vma->vm_flags & VM_EXECUTABLE) &&
968 			    vma->vm_file) {
969 				audit_log_d_path(ab, "exe=",
970 						 &vma->vm_file->f_path);
971 				break;
972 			}
973 			vma = vma->vm_next;
974 		}
975 		up_read(&mm->mmap_sem);
976 	}
977 	audit_log_task_context(ab);
978 }
979 
980 static int audit_log_pid_context(struct audit_context *context, pid_t pid,
981 				 uid_t auid, uid_t uid, unsigned int sessionid,
982 				 u32 sid, char *comm)
983 {
984 	struct audit_buffer *ab;
985 	char *ctx = NULL;
986 	u32 len;
987 	int rc = 0;
988 
989 	ab = audit_log_start(context, GFP_KERNEL, AUDIT_OBJ_PID);
990 	if (!ab)
991 		return rc;
992 
993 	audit_log_format(ab, "opid=%d oauid=%d ouid=%d oses=%d", pid, auid,
994 			 uid, sessionid);
995 	if (security_secid_to_secctx(sid, &ctx, &len)) {
996 		audit_log_format(ab, " obj=(none)");
997 		rc = 1;
998 	} else {
999 		audit_log_format(ab, " obj=%s", ctx);
1000 		security_release_secctx(ctx, len);
1001 	}
1002 	audit_log_format(ab, " ocomm=");
1003 	audit_log_untrustedstring(ab, comm);
1004 	audit_log_end(ab);
1005 
1006 	return rc;
1007 }
1008 
1009 /*
1010  * to_send and len_sent accounting are very loose estimates.  We aren't
1011  * really worried about a hard cap to MAX_EXECVE_AUDIT_LEN so much as being
1012  * within about 500 bytes (next page boundry)
1013  *
1014  * why snprintf?  an int is up to 12 digits long.  if we just assumed when
1015  * logging that a[%d]= was going to be 16 characters long we would be wasting
1016  * space in every audit message.  In one 7500 byte message we can log up to
1017  * about 1000 min size arguments.  That comes down to about 50% waste of space
1018  * if we didn't do the snprintf to find out how long arg_num_len was.
1019  */
1020 static int audit_log_single_execve_arg(struct audit_context *context,
1021 					struct audit_buffer **ab,
1022 					int arg_num,
1023 					size_t *len_sent,
1024 					const char __user *p,
1025 					char *buf)
1026 {
1027 	char arg_num_len_buf[12];
1028 	const char __user *tmp_p = p;
1029 	/* how many digits are in arg_num? 5 is the length of ' a=""' */
1030 	size_t arg_num_len = snprintf(arg_num_len_buf, 12, "%d", arg_num) + 5;
1031 	size_t len, len_left, to_send;
1032 	size_t max_execve_audit_len = MAX_EXECVE_AUDIT_LEN;
1033 	unsigned int i, has_cntl = 0, too_long = 0;
1034 	int ret;
1035 
1036 	/* strnlen_user includes the null we don't want to send */
1037 	len_left = len = strnlen_user(p, MAX_ARG_STRLEN) - 1;
1038 
1039 	/*
1040 	 * We just created this mm, if we can't find the strings
1041 	 * we just copied into it something is _very_ wrong. Similar
1042 	 * for strings that are too long, we should not have created
1043 	 * any.
1044 	 */
1045 	if (unlikely((len == -1) || len > MAX_ARG_STRLEN - 1)) {
1046 		WARN_ON(1);
1047 		send_sig(SIGKILL, current, 0);
1048 		return -1;
1049 	}
1050 
1051 	/* walk the whole argument looking for non-ascii chars */
1052 	do {
1053 		if (len_left > MAX_EXECVE_AUDIT_LEN)
1054 			to_send = MAX_EXECVE_AUDIT_LEN;
1055 		else
1056 			to_send = len_left;
1057 		ret = copy_from_user(buf, tmp_p, to_send);
1058 		/*
1059 		 * There is no reason for this copy to be short. We just
1060 		 * copied them here, and the mm hasn't been exposed to user-
1061 		 * space yet.
1062 		 */
1063 		if (ret) {
1064 			WARN_ON(1);
1065 			send_sig(SIGKILL, current, 0);
1066 			return -1;
1067 		}
1068 		buf[to_send] = '\0';
1069 		has_cntl = audit_string_contains_control(buf, to_send);
1070 		if (has_cntl) {
1071 			/*
1072 			 * hex messages get logged as 2 bytes, so we can only
1073 			 * send half as much in each message
1074 			 */
1075 			max_execve_audit_len = MAX_EXECVE_AUDIT_LEN / 2;
1076 			break;
1077 		}
1078 		len_left -= to_send;
1079 		tmp_p += to_send;
1080 	} while (len_left > 0);
1081 
1082 	len_left = len;
1083 
1084 	if (len > max_execve_audit_len)
1085 		too_long = 1;
1086 
1087 	/* rewalk the argument actually logging the message */
1088 	for (i = 0; len_left > 0; i++) {
1089 		int room_left;
1090 
1091 		if (len_left > max_execve_audit_len)
1092 			to_send = max_execve_audit_len;
1093 		else
1094 			to_send = len_left;
1095 
1096 		/* do we have space left to send this argument in this ab? */
1097 		room_left = MAX_EXECVE_AUDIT_LEN - arg_num_len - *len_sent;
1098 		if (has_cntl)
1099 			room_left -= (to_send * 2);
1100 		else
1101 			room_left -= to_send;
1102 		if (room_left < 0) {
1103 			*len_sent = 0;
1104 			audit_log_end(*ab);
1105 			*ab = audit_log_start(context, GFP_KERNEL, AUDIT_EXECVE);
1106 			if (!*ab)
1107 				return 0;
1108 		}
1109 
1110 		/*
1111 		 * first record needs to say how long the original string was
1112 		 * so we can be sure nothing was lost.
1113 		 */
1114 		if ((i == 0) && (too_long))
1115 			audit_log_format(*ab, " a%d_len=%zu", arg_num,
1116 					 has_cntl ? 2*len : len);
1117 
1118 		/*
1119 		 * normally arguments are small enough to fit and we already
1120 		 * filled buf above when we checked for control characters
1121 		 * so don't bother with another copy_from_user
1122 		 */
1123 		if (len >= max_execve_audit_len)
1124 			ret = copy_from_user(buf, p, to_send);
1125 		else
1126 			ret = 0;
1127 		if (ret) {
1128 			WARN_ON(1);
1129 			send_sig(SIGKILL, current, 0);
1130 			return -1;
1131 		}
1132 		buf[to_send] = '\0';
1133 
1134 		/* actually log it */
1135 		audit_log_format(*ab, " a%d", arg_num);
1136 		if (too_long)
1137 			audit_log_format(*ab, "[%d]", i);
1138 		audit_log_format(*ab, "=");
1139 		if (has_cntl)
1140 			audit_log_n_hex(*ab, buf, to_send);
1141 		else
1142 			audit_log_string(*ab, buf);
1143 
1144 		p += to_send;
1145 		len_left -= to_send;
1146 		*len_sent += arg_num_len;
1147 		if (has_cntl)
1148 			*len_sent += to_send * 2;
1149 		else
1150 			*len_sent += to_send;
1151 	}
1152 	/* include the null we didn't log */
1153 	return len + 1;
1154 }
1155 
1156 static void audit_log_execve_info(struct audit_context *context,
1157 				  struct audit_buffer **ab,
1158 				  struct audit_aux_data_execve *axi)
1159 {
1160 	int i;
1161 	size_t len, len_sent = 0;
1162 	const char __user *p;
1163 	char *buf;
1164 
1165 	if (axi->mm != current->mm)
1166 		return; /* execve failed, no additional info */
1167 
1168 	p = (const char __user *)axi->mm->arg_start;
1169 
1170 	audit_log_format(*ab, "argc=%d", axi->argc);
1171 
1172 	/*
1173 	 * we need some kernel buffer to hold the userspace args.  Just
1174 	 * allocate one big one rather than allocating one of the right size
1175 	 * for every single argument inside audit_log_single_execve_arg()
1176 	 * should be <8k allocation so should be pretty safe.
1177 	 */
1178 	buf = kmalloc(MAX_EXECVE_AUDIT_LEN + 1, GFP_KERNEL);
1179 	if (!buf) {
1180 		audit_panic("out of memory for argv string\n");
1181 		return;
1182 	}
1183 
1184 	for (i = 0; i < axi->argc; i++) {
1185 		len = audit_log_single_execve_arg(context, ab, i,
1186 						  &len_sent, p, buf);
1187 		if (len <= 0)
1188 			break;
1189 		p += len;
1190 	}
1191 	kfree(buf);
1192 }
1193 
1194 static void audit_log_cap(struct audit_buffer *ab, char *prefix, kernel_cap_t *cap)
1195 {
1196 	int i;
1197 
1198 	audit_log_format(ab, " %s=", prefix);
1199 	CAP_FOR_EACH_U32(i) {
1200 		audit_log_format(ab, "%08x", cap->cap[(_KERNEL_CAPABILITY_U32S-1) - i]);
1201 	}
1202 }
1203 
1204 static void audit_log_fcaps(struct audit_buffer *ab, struct audit_names *name)
1205 {
1206 	kernel_cap_t *perm = &name->fcap.permitted;
1207 	kernel_cap_t *inh = &name->fcap.inheritable;
1208 	int log = 0;
1209 
1210 	if (!cap_isclear(*perm)) {
1211 		audit_log_cap(ab, "cap_fp", perm);
1212 		log = 1;
1213 	}
1214 	if (!cap_isclear(*inh)) {
1215 		audit_log_cap(ab, "cap_fi", inh);
1216 		log = 1;
1217 	}
1218 
1219 	if (log)
1220 		audit_log_format(ab, " cap_fe=%d cap_fver=%x", name->fcap.fE, name->fcap_ver);
1221 }
1222 
1223 static void show_special(struct audit_context *context, int *call_panic)
1224 {
1225 	struct audit_buffer *ab;
1226 	int i;
1227 
1228 	ab = audit_log_start(context, GFP_KERNEL, context->type);
1229 	if (!ab)
1230 		return;
1231 
1232 	switch (context->type) {
1233 	case AUDIT_SOCKETCALL: {
1234 		int nargs = context->socketcall.nargs;
1235 		audit_log_format(ab, "nargs=%d", nargs);
1236 		for (i = 0; i < nargs; i++)
1237 			audit_log_format(ab, " a%d=%lx", i,
1238 				context->socketcall.args[i]);
1239 		break; }
1240 	case AUDIT_IPC: {
1241 		u32 osid = context->ipc.osid;
1242 
1243 		audit_log_format(ab, "ouid=%u ogid=%u mode=%#o",
1244 			 context->ipc.uid, context->ipc.gid, context->ipc.mode);
1245 		if (osid) {
1246 			char *ctx = NULL;
1247 			u32 len;
1248 			if (security_secid_to_secctx(osid, &ctx, &len)) {
1249 				audit_log_format(ab, " osid=%u", osid);
1250 				*call_panic = 1;
1251 			} else {
1252 				audit_log_format(ab, " obj=%s", ctx);
1253 				security_release_secctx(ctx, len);
1254 			}
1255 		}
1256 		if (context->ipc.has_perm) {
1257 			audit_log_end(ab);
1258 			ab = audit_log_start(context, GFP_KERNEL,
1259 					     AUDIT_IPC_SET_PERM);
1260 			audit_log_format(ab,
1261 				"qbytes=%lx ouid=%u ogid=%u mode=%#o",
1262 				context->ipc.qbytes,
1263 				context->ipc.perm_uid,
1264 				context->ipc.perm_gid,
1265 				context->ipc.perm_mode);
1266 			if (!ab)
1267 				return;
1268 		}
1269 		break; }
1270 	case AUDIT_MQ_OPEN: {
1271 		audit_log_format(ab,
1272 			"oflag=0x%x mode=%#o mq_flags=0x%lx mq_maxmsg=%ld "
1273 			"mq_msgsize=%ld mq_curmsgs=%ld",
1274 			context->mq_open.oflag, context->mq_open.mode,
1275 			context->mq_open.attr.mq_flags,
1276 			context->mq_open.attr.mq_maxmsg,
1277 			context->mq_open.attr.mq_msgsize,
1278 			context->mq_open.attr.mq_curmsgs);
1279 		break; }
1280 	case AUDIT_MQ_SENDRECV: {
1281 		audit_log_format(ab,
1282 			"mqdes=%d msg_len=%zd msg_prio=%u "
1283 			"abs_timeout_sec=%ld abs_timeout_nsec=%ld",
1284 			context->mq_sendrecv.mqdes,
1285 			context->mq_sendrecv.msg_len,
1286 			context->mq_sendrecv.msg_prio,
1287 			context->mq_sendrecv.abs_timeout.tv_sec,
1288 			context->mq_sendrecv.abs_timeout.tv_nsec);
1289 		break; }
1290 	case AUDIT_MQ_NOTIFY: {
1291 		audit_log_format(ab, "mqdes=%d sigev_signo=%d",
1292 				context->mq_notify.mqdes,
1293 				context->mq_notify.sigev_signo);
1294 		break; }
1295 	case AUDIT_MQ_GETSETATTR: {
1296 		struct mq_attr *attr = &context->mq_getsetattr.mqstat;
1297 		audit_log_format(ab,
1298 			"mqdes=%d mq_flags=0x%lx mq_maxmsg=%ld mq_msgsize=%ld "
1299 			"mq_curmsgs=%ld ",
1300 			context->mq_getsetattr.mqdes,
1301 			attr->mq_flags, attr->mq_maxmsg,
1302 			attr->mq_msgsize, attr->mq_curmsgs);
1303 		break; }
1304 	case AUDIT_CAPSET: {
1305 		audit_log_format(ab, "pid=%d", context->capset.pid);
1306 		audit_log_cap(ab, "cap_pi", &context->capset.cap.inheritable);
1307 		audit_log_cap(ab, "cap_pp", &context->capset.cap.permitted);
1308 		audit_log_cap(ab, "cap_pe", &context->capset.cap.effective);
1309 		break; }
1310 	}
1311 	audit_log_end(ab);
1312 }
1313 
1314 static void audit_log_exit(struct audit_context *context, struct task_struct *tsk)
1315 {
1316 	const struct cred *cred;
1317 	int i, call_panic = 0;
1318 	struct audit_buffer *ab;
1319 	struct audit_aux_data *aux;
1320 	const char *tty;
1321 
1322 	/* tsk == current */
1323 	context->pid = tsk->pid;
1324 	if (!context->ppid)
1325 		context->ppid = sys_getppid();
1326 	cred = current_cred();
1327 	context->uid   = cred->uid;
1328 	context->gid   = cred->gid;
1329 	context->euid  = cred->euid;
1330 	context->suid  = cred->suid;
1331 	context->fsuid = cred->fsuid;
1332 	context->egid  = cred->egid;
1333 	context->sgid  = cred->sgid;
1334 	context->fsgid = cred->fsgid;
1335 	context->personality = tsk->personality;
1336 
1337 	ab = audit_log_start(context, GFP_KERNEL, AUDIT_SYSCALL);
1338 	if (!ab)
1339 		return;		/* audit_panic has been called */
1340 	audit_log_format(ab, "arch=%x syscall=%d",
1341 			 context->arch, context->major);
1342 	if (context->personality != PER_LINUX)
1343 		audit_log_format(ab, " per=%lx", context->personality);
1344 	if (context->return_valid)
1345 		audit_log_format(ab, " success=%s exit=%ld",
1346 				 (context->return_valid==AUDITSC_SUCCESS)?"yes":"no",
1347 				 context->return_code);
1348 
1349 	spin_lock_irq(&tsk->sighand->siglock);
1350 	if (tsk->signal && tsk->signal->tty && tsk->signal->tty->name)
1351 		tty = tsk->signal->tty->name;
1352 	else
1353 		tty = "(none)";
1354 	spin_unlock_irq(&tsk->sighand->siglock);
1355 
1356 	audit_log_format(ab,
1357 		  " a0=%lx a1=%lx a2=%lx a3=%lx items=%d"
1358 		  " ppid=%d pid=%d auid=%u uid=%u gid=%u"
1359 		  " euid=%u suid=%u fsuid=%u"
1360 		  " egid=%u sgid=%u fsgid=%u tty=%s ses=%u",
1361 		  context->argv[0],
1362 		  context->argv[1],
1363 		  context->argv[2],
1364 		  context->argv[3],
1365 		  context->name_count,
1366 		  context->ppid,
1367 		  context->pid,
1368 		  tsk->loginuid,
1369 		  context->uid,
1370 		  context->gid,
1371 		  context->euid, context->suid, context->fsuid,
1372 		  context->egid, context->sgid, context->fsgid, tty,
1373 		  tsk->sessionid);
1374 
1375 
1376 	audit_log_task_info(ab, tsk);
1377 	audit_log_key(ab, context->filterkey);
1378 	audit_log_end(ab);
1379 
1380 	for (aux = context->aux; aux; aux = aux->next) {
1381 
1382 		ab = audit_log_start(context, GFP_KERNEL, aux->type);
1383 		if (!ab)
1384 			continue; /* audit_panic has been called */
1385 
1386 		switch (aux->type) {
1387 
1388 		case AUDIT_EXECVE: {
1389 			struct audit_aux_data_execve *axi = (void *)aux;
1390 			audit_log_execve_info(context, &ab, axi);
1391 			break; }
1392 
1393 		case AUDIT_BPRM_FCAPS: {
1394 			struct audit_aux_data_bprm_fcaps *axs = (void *)aux;
1395 			audit_log_format(ab, "fver=%x", axs->fcap_ver);
1396 			audit_log_cap(ab, "fp", &axs->fcap.permitted);
1397 			audit_log_cap(ab, "fi", &axs->fcap.inheritable);
1398 			audit_log_format(ab, " fe=%d", axs->fcap.fE);
1399 			audit_log_cap(ab, "old_pp", &axs->old_pcap.permitted);
1400 			audit_log_cap(ab, "old_pi", &axs->old_pcap.inheritable);
1401 			audit_log_cap(ab, "old_pe", &axs->old_pcap.effective);
1402 			audit_log_cap(ab, "new_pp", &axs->new_pcap.permitted);
1403 			audit_log_cap(ab, "new_pi", &axs->new_pcap.inheritable);
1404 			audit_log_cap(ab, "new_pe", &axs->new_pcap.effective);
1405 			break; }
1406 
1407 		}
1408 		audit_log_end(ab);
1409 	}
1410 
1411 	if (context->type)
1412 		show_special(context, &call_panic);
1413 
1414 	if (context->fds[0] >= 0) {
1415 		ab = audit_log_start(context, GFP_KERNEL, AUDIT_FD_PAIR);
1416 		if (ab) {
1417 			audit_log_format(ab, "fd0=%d fd1=%d",
1418 					context->fds[0], context->fds[1]);
1419 			audit_log_end(ab);
1420 		}
1421 	}
1422 
1423 	if (context->sockaddr_len) {
1424 		ab = audit_log_start(context, GFP_KERNEL, AUDIT_SOCKADDR);
1425 		if (ab) {
1426 			audit_log_format(ab, "saddr=");
1427 			audit_log_n_hex(ab, (void *)context->sockaddr,
1428 					context->sockaddr_len);
1429 			audit_log_end(ab);
1430 		}
1431 	}
1432 
1433 	for (aux = context->aux_pids; aux; aux = aux->next) {
1434 		struct audit_aux_data_pids *axs = (void *)aux;
1435 
1436 		for (i = 0; i < axs->pid_count; i++)
1437 			if (audit_log_pid_context(context, axs->target_pid[i],
1438 						  axs->target_auid[i],
1439 						  axs->target_uid[i],
1440 						  axs->target_sessionid[i],
1441 						  axs->target_sid[i],
1442 						  axs->target_comm[i]))
1443 				call_panic = 1;
1444 	}
1445 
1446 	if (context->target_pid &&
1447 	    audit_log_pid_context(context, context->target_pid,
1448 				  context->target_auid, context->target_uid,
1449 				  context->target_sessionid,
1450 				  context->target_sid, context->target_comm))
1451 			call_panic = 1;
1452 
1453 	if (context->pwd.dentry && context->pwd.mnt) {
1454 		ab = audit_log_start(context, GFP_KERNEL, AUDIT_CWD);
1455 		if (ab) {
1456 			audit_log_d_path(ab, "cwd=", &context->pwd);
1457 			audit_log_end(ab);
1458 		}
1459 	}
1460 	for (i = 0; i < context->name_count; i++) {
1461 		struct audit_names *n = &context->names[i];
1462 
1463 		ab = audit_log_start(context, GFP_KERNEL, AUDIT_PATH);
1464 		if (!ab)
1465 			continue; /* audit_panic has been called */
1466 
1467 		audit_log_format(ab, "item=%d", i);
1468 
1469 		if (n->name) {
1470 			switch(n->name_len) {
1471 			case AUDIT_NAME_FULL:
1472 				/* log the full path */
1473 				audit_log_format(ab, " name=");
1474 				audit_log_untrustedstring(ab, n->name);
1475 				break;
1476 			case 0:
1477 				/* name was specified as a relative path and the
1478 				 * directory component is the cwd */
1479 				audit_log_d_path(ab, "name=", &context->pwd);
1480 				break;
1481 			default:
1482 				/* log the name's directory component */
1483 				audit_log_format(ab, " name=");
1484 				audit_log_n_untrustedstring(ab, n->name,
1485 							    n->name_len);
1486 			}
1487 		} else
1488 			audit_log_format(ab, " name=(null)");
1489 
1490 		if (n->ino != (unsigned long)-1) {
1491 			audit_log_format(ab, " inode=%lu"
1492 					 " dev=%02x:%02x mode=%#o"
1493 					 " ouid=%u ogid=%u rdev=%02x:%02x",
1494 					 n->ino,
1495 					 MAJOR(n->dev),
1496 					 MINOR(n->dev),
1497 					 n->mode,
1498 					 n->uid,
1499 					 n->gid,
1500 					 MAJOR(n->rdev),
1501 					 MINOR(n->rdev));
1502 		}
1503 		if (n->osid != 0) {
1504 			char *ctx = NULL;
1505 			u32 len;
1506 			if (security_secid_to_secctx(
1507 				n->osid, &ctx, &len)) {
1508 				audit_log_format(ab, " osid=%u", n->osid);
1509 				call_panic = 2;
1510 			} else {
1511 				audit_log_format(ab, " obj=%s", ctx);
1512 				security_release_secctx(ctx, len);
1513 			}
1514 		}
1515 
1516 		audit_log_fcaps(ab, n);
1517 
1518 		audit_log_end(ab);
1519 	}
1520 
1521 	/* Send end of event record to help user space know we are finished */
1522 	ab = audit_log_start(context, GFP_KERNEL, AUDIT_EOE);
1523 	if (ab)
1524 		audit_log_end(ab);
1525 	if (call_panic)
1526 		audit_panic("error converting sid to string");
1527 }
1528 
1529 /**
1530  * audit_free - free a per-task audit context
1531  * @tsk: task whose audit context block to free
1532  *
1533  * Called from copy_process and do_exit
1534  */
1535 void audit_free(struct task_struct *tsk)
1536 {
1537 	struct audit_context *context;
1538 
1539 	context = audit_get_context(tsk, 0, 0);
1540 	if (likely(!context))
1541 		return;
1542 
1543 	/* Check for system calls that do not go through the exit
1544 	 * function (e.g., exit_group), then free context block.
1545 	 * We use GFP_ATOMIC here because we might be doing this
1546 	 * in the context of the idle thread */
1547 	/* that can happen only if we are called from do_exit() */
1548 	if (context->in_syscall && context->current_state == AUDIT_RECORD_CONTEXT)
1549 		audit_log_exit(context, tsk);
1550 	if (!list_empty(&context->killed_trees))
1551 		audit_kill_trees(&context->killed_trees);
1552 
1553 	audit_free_context(context);
1554 }
1555 
1556 /**
1557  * audit_syscall_entry - fill in an audit record at syscall entry
1558  * @arch: architecture type
1559  * @major: major syscall type (function)
1560  * @a1: additional syscall register 1
1561  * @a2: additional syscall register 2
1562  * @a3: additional syscall register 3
1563  * @a4: additional syscall register 4
1564  *
1565  * Fill in audit context at syscall entry.  This only happens if the
1566  * audit context was created when the task was created and the state or
1567  * filters demand the audit context be built.  If the state from the
1568  * per-task filter or from the per-syscall filter is AUDIT_RECORD_CONTEXT,
1569  * then the record will be written at syscall exit time (otherwise, it
1570  * will only be written if another part of the kernel requests that it
1571  * be written).
1572  */
1573 void audit_syscall_entry(int arch, int major,
1574 			 unsigned long a1, unsigned long a2,
1575 			 unsigned long a3, unsigned long a4)
1576 {
1577 	struct task_struct *tsk = current;
1578 	struct audit_context *context = tsk->audit_context;
1579 	enum audit_state     state;
1580 
1581 	if (unlikely(!context))
1582 		return;
1583 
1584 	/*
1585 	 * This happens only on certain architectures that make system
1586 	 * calls in kernel_thread via the entry.S interface, instead of
1587 	 * with direct calls.  (If you are porting to a new
1588 	 * architecture, hitting this condition can indicate that you
1589 	 * got the _exit/_leave calls backward in entry.S.)
1590 	 *
1591 	 * i386     no
1592 	 * x86_64   no
1593 	 * ppc64    yes (see arch/powerpc/platforms/iseries/misc.S)
1594 	 *
1595 	 * This also happens with vm86 emulation in a non-nested manner
1596 	 * (entries without exits), so this case must be caught.
1597 	 */
1598 	if (context->in_syscall) {
1599 		struct audit_context *newctx;
1600 
1601 #if AUDIT_DEBUG
1602 		printk(KERN_ERR
1603 		       "audit(:%d) pid=%d in syscall=%d;"
1604 		       " entering syscall=%d\n",
1605 		       context->serial, tsk->pid, context->major, major);
1606 #endif
1607 		newctx = audit_alloc_context(context->state);
1608 		if (newctx) {
1609 			newctx->previous   = context;
1610 			context		   = newctx;
1611 			tsk->audit_context = newctx;
1612 		} else	{
1613 			/* If we can't alloc a new context, the best we
1614 			 * can do is to leak memory (any pending putname
1615 			 * will be lost).  The only other alternative is
1616 			 * to abandon auditing. */
1617 			audit_zero_context(context, context->state);
1618 		}
1619 	}
1620 	BUG_ON(context->in_syscall || context->name_count);
1621 
1622 	if (!audit_enabled)
1623 		return;
1624 
1625 	context->arch	    = arch;
1626 	context->major      = major;
1627 	context->argv[0]    = a1;
1628 	context->argv[1]    = a2;
1629 	context->argv[2]    = a3;
1630 	context->argv[3]    = a4;
1631 
1632 	state = context->state;
1633 	context->dummy = !audit_n_rules;
1634 	if (!context->dummy && state == AUDIT_BUILD_CONTEXT) {
1635 		context->prio = 0;
1636 		state = audit_filter_syscall(tsk, context, &audit_filter_list[AUDIT_FILTER_ENTRY]);
1637 	}
1638 	if (likely(state == AUDIT_DISABLED))
1639 		return;
1640 
1641 	context->serial     = 0;
1642 	context->ctime      = CURRENT_TIME;
1643 	context->in_syscall = 1;
1644 	context->current_state  = state;
1645 	context->ppid       = 0;
1646 }
1647 
1648 void audit_finish_fork(struct task_struct *child)
1649 {
1650 	struct audit_context *ctx = current->audit_context;
1651 	struct audit_context *p = child->audit_context;
1652 	if (!p || !ctx)
1653 		return;
1654 	if (!ctx->in_syscall || ctx->current_state != AUDIT_RECORD_CONTEXT)
1655 		return;
1656 	p->arch = ctx->arch;
1657 	p->major = ctx->major;
1658 	memcpy(p->argv, ctx->argv, sizeof(ctx->argv));
1659 	p->ctime = ctx->ctime;
1660 	p->dummy = ctx->dummy;
1661 	p->in_syscall = ctx->in_syscall;
1662 	p->filterkey = kstrdup(ctx->filterkey, GFP_KERNEL);
1663 	p->ppid = current->pid;
1664 	p->prio = ctx->prio;
1665 	p->current_state = ctx->current_state;
1666 }
1667 
1668 /**
1669  * audit_syscall_exit - deallocate audit context after a system call
1670  * @valid: success/failure flag
1671  * @return_code: syscall return value
1672  *
1673  * Tear down after system call.  If the audit context has been marked as
1674  * auditable (either because of the AUDIT_RECORD_CONTEXT state from
1675  * filtering, or because some other part of the kernel write an audit
1676  * message), then write out the syscall information.  In call cases,
1677  * free the names stored from getname().
1678  */
1679 void audit_syscall_exit(int valid, long return_code)
1680 {
1681 	struct task_struct *tsk = current;
1682 	struct audit_context *context;
1683 
1684 	context = audit_get_context(tsk, valid, return_code);
1685 
1686 	if (likely(!context))
1687 		return;
1688 
1689 	if (context->in_syscall && context->current_state == AUDIT_RECORD_CONTEXT)
1690 		audit_log_exit(context, tsk);
1691 
1692 	context->in_syscall = 0;
1693 	context->prio = context->state == AUDIT_RECORD_CONTEXT ? ~0ULL : 0;
1694 
1695 	if (!list_empty(&context->killed_trees))
1696 		audit_kill_trees(&context->killed_trees);
1697 
1698 	if (context->previous) {
1699 		struct audit_context *new_context = context->previous;
1700 		context->previous  = NULL;
1701 		audit_free_context(context);
1702 		tsk->audit_context = new_context;
1703 	} else {
1704 		audit_free_names(context);
1705 		unroll_tree_refs(context, NULL, 0);
1706 		audit_free_aux(context);
1707 		context->aux = NULL;
1708 		context->aux_pids = NULL;
1709 		context->target_pid = 0;
1710 		context->target_sid = 0;
1711 		context->sockaddr_len = 0;
1712 		context->type = 0;
1713 		context->fds[0] = -1;
1714 		if (context->state != AUDIT_RECORD_CONTEXT) {
1715 			kfree(context->filterkey);
1716 			context->filterkey = NULL;
1717 		}
1718 		tsk->audit_context = context;
1719 	}
1720 }
1721 
1722 static inline void handle_one(const struct inode *inode)
1723 {
1724 #ifdef CONFIG_AUDIT_TREE
1725 	struct audit_context *context;
1726 	struct audit_tree_refs *p;
1727 	struct audit_chunk *chunk;
1728 	int count;
1729 	if (likely(list_empty(&inode->inotify_watches)))
1730 		return;
1731 	context = current->audit_context;
1732 	p = context->trees;
1733 	count = context->tree_count;
1734 	rcu_read_lock();
1735 	chunk = audit_tree_lookup(inode);
1736 	rcu_read_unlock();
1737 	if (!chunk)
1738 		return;
1739 	if (likely(put_tree_ref(context, chunk)))
1740 		return;
1741 	if (unlikely(!grow_tree_refs(context))) {
1742 		printk(KERN_WARNING "out of memory, audit has lost a tree reference\n");
1743 		audit_set_auditable(context);
1744 		audit_put_chunk(chunk);
1745 		unroll_tree_refs(context, p, count);
1746 		return;
1747 	}
1748 	put_tree_ref(context, chunk);
1749 #endif
1750 }
1751 
1752 static void handle_path(const struct dentry *dentry)
1753 {
1754 #ifdef CONFIG_AUDIT_TREE
1755 	struct audit_context *context;
1756 	struct audit_tree_refs *p;
1757 	const struct dentry *d, *parent;
1758 	struct audit_chunk *drop;
1759 	unsigned long seq;
1760 	int count;
1761 
1762 	context = current->audit_context;
1763 	p = context->trees;
1764 	count = context->tree_count;
1765 retry:
1766 	drop = NULL;
1767 	d = dentry;
1768 	rcu_read_lock();
1769 	seq = read_seqbegin(&rename_lock);
1770 	for(;;) {
1771 		struct inode *inode = d->d_inode;
1772 		if (inode && unlikely(!list_empty(&inode->inotify_watches))) {
1773 			struct audit_chunk *chunk;
1774 			chunk = audit_tree_lookup(inode);
1775 			if (chunk) {
1776 				if (unlikely(!put_tree_ref(context, chunk))) {
1777 					drop = chunk;
1778 					break;
1779 				}
1780 			}
1781 		}
1782 		parent = d->d_parent;
1783 		if (parent == d)
1784 			break;
1785 		d = parent;
1786 	}
1787 	if (unlikely(read_seqretry(&rename_lock, seq) || drop)) {  /* in this order */
1788 		rcu_read_unlock();
1789 		if (!drop) {
1790 			/* just a race with rename */
1791 			unroll_tree_refs(context, p, count);
1792 			goto retry;
1793 		}
1794 		audit_put_chunk(drop);
1795 		if (grow_tree_refs(context)) {
1796 			/* OK, got more space */
1797 			unroll_tree_refs(context, p, count);
1798 			goto retry;
1799 		}
1800 		/* too bad */
1801 		printk(KERN_WARNING
1802 			"out of memory, audit has lost a tree reference\n");
1803 		unroll_tree_refs(context, p, count);
1804 		audit_set_auditable(context);
1805 		return;
1806 	}
1807 	rcu_read_unlock();
1808 #endif
1809 }
1810 
1811 /**
1812  * audit_getname - add a name to the list
1813  * @name: name to add
1814  *
1815  * Add a name to the list of audit names for this context.
1816  * Called from fs/namei.c:getname().
1817  */
1818 void __audit_getname(const char *name)
1819 {
1820 	struct audit_context *context = current->audit_context;
1821 
1822 	if (IS_ERR(name) || !name)
1823 		return;
1824 
1825 	if (!context->in_syscall) {
1826 #if AUDIT_DEBUG == 2
1827 		printk(KERN_ERR "%s:%d(:%d): ignoring getname(%p)\n",
1828 		       __FILE__, __LINE__, context->serial, name);
1829 		dump_stack();
1830 #endif
1831 		return;
1832 	}
1833 	BUG_ON(context->name_count >= AUDIT_NAMES);
1834 	context->names[context->name_count].name = name;
1835 	context->names[context->name_count].name_len = AUDIT_NAME_FULL;
1836 	context->names[context->name_count].name_put = 1;
1837 	context->names[context->name_count].ino  = (unsigned long)-1;
1838 	context->names[context->name_count].osid = 0;
1839 	++context->name_count;
1840 	if (!context->pwd.dentry) {
1841 		read_lock(&current->fs->lock);
1842 		context->pwd = current->fs->pwd;
1843 		path_get(&current->fs->pwd);
1844 		read_unlock(&current->fs->lock);
1845 	}
1846 
1847 }
1848 
1849 /* audit_putname - intercept a putname request
1850  * @name: name to intercept and delay for putname
1851  *
1852  * If we have stored the name from getname in the audit context,
1853  * then we delay the putname until syscall exit.
1854  * Called from include/linux/fs.h:putname().
1855  */
1856 void audit_putname(const char *name)
1857 {
1858 	struct audit_context *context = current->audit_context;
1859 
1860 	BUG_ON(!context);
1861 	if (!context->in_syscall) {
1862 #if AUDIT_DEBUG == 2
1863 		printk(KERN_ERR "%s:%d(:%d): __putname(%p)\n",
1864 		       __FILE__, __LINE__, context->serial, name);
1865 		if (context->name_count) {
1866 			int i;
1867 			for (i = 0; i < context->name_count; i++)
1868 				printk(KERN_ERR "name[%d] = %p = %s\n", i,
1869 				       context->names[i].name,
1870 				       context->names[i].name ?: "(null)");
1871 		}
1872 #endif
1873 		__putname(name);
1874 	}
1875 #if AUDIT_DEBUG
1876 	else {
1877 		++context->put_count;
1878 		if (context->put_count > context->name_count) {
1879 			printk(KERN_ERR "%s:%d(:%d): major=%d"
1880 			       " in_syscall=%d putname(%p) name_count=%d"
1881 			       " put_count=%d\n",
1882 			       __FILE__, __LINE__,
1883 			       context->serial, context->major,
1884 			       context->in_syscall, name, context->name_count,
1885 			       context->put_count);
1886 			dump_stack();
1887 		}
1888 	}
1889 #endif
1890 }
1891 
1892 static int audit_inc_name_count(struct audit_context *context,
1893 				const struct inode *inode)
1894 {
1895 	if (context->name_count >= AUDIT_NAMES) {
1896 		if (inode)
1897 			printk(KERN_DEBUG "name_count maxed, losing inode data: "
1898 			       "dev=%02x:%02x, inode=%lu\n",
1899 			       MAJOR(inode->i_sb->s_dev),
1900 			       MINOR(inode->i_sb->s_dev),
1901 			       inode->i_ino);
1902 
1903 		else
1904 			printk(KERN_DEBUG "name_count maxed, losing inode data\n");
1905 		return 1;
1906 	}
1907 	context->name_count++;
1908 #if AUDIT_DEBUG
1909 	context->ino_count++;
1910 #endif
1911 	return 0;
1912 }
1913 
1914 
1915 static inline int audit_copy_fcaps(struct audit_names *name, const struct dentry *dentry)
1916 {
1917 	struct cpu_vfs_cap_data caps;
1918 	int rc;
1919 
1920 	memset(&name->fcap.permitted, 0, sizeof(kernel_cap_t));
1921 	memset(&name->fcap.inheritable, 0, sizeof(kernel_cap_t));
1922 	name->fcap.fE = 0;
1923 	name->fcap_ver = 0;
1924 
1925 	if (!dentry)
1926 		return 0;
1927 
1928 	rc = get_vfs_caps_from_disk(dentry, &caps);
1929 	if (rc)
1930 		return rc;
1931 
1932 	name->fcap.permitted = caps.permitted;
1933 	name->fcap.inheritable = caps.inheritable;
1934 	name->fcap.fE = !!(caps.magic_etc & VFS_CAP_FLAGS_EFFECTIVE);
1935 	name->fcap_ver = (caps.magic_etc & VFS_CAP_REVISION_MASK) >> VFS_CAP_REVISION_SHIFT;
1936 
1937 	return 0;
1938 }
1939 
1940 
1941 /* Copy inode data into an audit_names. */
1942 static void audit_copy_inode(struct audit_names *name, const struct dentry *dentry,
1943 			     const struct inode *inode)
1944 {
1945 	name->ino   = inode->i_ino;
1946 	name->dev   = inode->i_sb->s_dev;
1947 	name->mode  = inode->i_mode;
1948 	name->uid   = inode->i_uid;
1949 	name->gid   = inode->i_gid;
1950 	name->rdev  = inode->i_rdev;
1951 	security_inode_getsecid(inode, &name->osid);
1952 	audit_copy_fcaps(name, dentry);
1953 }
1954 
1955 /**
1956  * audit_inode - store the inode and device from a lookup
1957  * @name: name being audited
1958  * @dentry: dentry being audited
1959  *
1960  * Called from fs/namei.c:path_lookup().
1961  */
1962 void __audit_inode(const char *name, const struct dentry *dentry)
1963 {
1964 	int idx;
1965 	struct audit_context *context = current->audit_context;
1966 	const struct inode *inode = dentry->d_inode;
1967 
1968 	if (!context->in_syscall)
1969 		return;
1970 	if (context->name_count
1971 	    && context->names[context->name_count-1].name
1972 	    && context->names[context->name_count-1].name == name)
1973 		idx = context->name_count - 1;
1974 	else if (context->name_count > 1
1975 		 && context->names[context->name_count-2].name
1976 		 && context->names[context->name_count-2].name == name)
1977 		idx = context->name_count - 2;
1978 	else {
1979 		/* FIXME: how much do we care about inodes that have no
1980 		 * associated name? */
1981 		if (audit_inc_name_count(context, inode))
1982 			return;
1983 		idx = context->name_count - 1;
1984 		context->names[idx].name = NULL;
1985 	}
1986 	handle_path(dentry);
1987 	audit_copy_inode(&context->names[idx], dentry, inode);
1988 }
1989 
1990 /**
1991  * audit_inode_child - collect inode info for created/removed objects
1992  * @dname: inode's dentry name
1993  * @dentry: dentry being audited
1994  * @parent: inode of dentry parent
1995  *
1996  * For syscalls that create or remove filesystem objects, audit_inode
1997  * can only collect information for the filesystem object's parent.
1998  * This call updates the audit context with the child's information.
1999  * Syscalls that create a new filesystem object must be hooked after
2000  * the object is created.  Syscalls that remove a filesystem object
2001  * must be hooked prior, in order to capture the target inode during
2002  * unsuccessful attempts.
2003  */
2004 void __audit_inode_child(const char *dname, const struct dentry *dentry,
2005 			 const struct inode *parent)
2006 {
2007 	int idx;
2008 	struct audit_context *context = current->audit_context;
2009 	const char *found_parent = NULL, *found_child = NULL;
2010 	const struct inode *inode = dentry->d_inode;
2011 	int dirlen = 0;
2012 
2013 	if (!context->in_syscall)
2014 		return;
2015 
2016 	if (inode)
2017 		handle_one(inode);
2018 	/* determine matching parent */
2019 	if (!dname)
2020 		goto add_names;
2021 
2022 	/* parent is more likely, look for it first */
2023 	for (idx = 0; idx < context->name_count; idx++) {
2024 		struct audit_names *n = &context->names[idx];
2025 
2026 		if (!n->name)
2027 			continue;
2028 
2029 		if (n->ino == parent->i_ino &&
2030 		    !audit_compare_dname_path(dname, n->name, &dirlen)) {
2031 			n->name_len = dirlen; /* update parent data in place */
2032 			found_parent = n->name;
2033 			goto add_names;
2034 		}
2035 	}
2036 
2037 	/* no matching parent, look for matching child */
2038 	for (idx = 0; idx < context->name_count; idx++) {
2039 		struct audit_names *n = &context->names[idx];
2040 
2041 		if (!n->name)
2042 			continue;
2043 
2044 		/* strcmp() is the more likely scenario */
2045 		if (!strcmp(dname, n->name) ||
2046 		     !audit_compare_dname_path(dname, n->name, &dirlen)) {
2047 			if (inode)
2048 				audit_copy_inode(n, NULL, inode);
2049 			else
2050 				n->ino = (unsigned long)-1;
2051 			found_child = n->name;
2052 			goto add_names;
2053 		}
2054 	}
2055 
2056 add_names:
2057 	if (!found_parent) {
2058 		if (audit_inc_name_count(context, parent))
2059 			return;
2060 		idx = context->name_count - 1;
2061 		context->names[idx].name = NULL;
2062 		audit_copy_inode(&context->names[idx], NULL, parent);
2063 	}
2064 
2065 	if (!found_child) {
2066 		if (audit_inc_name_count(context, inode))
2067 			return;
2068 		idx = context->name_count - 1;
2069 
2070 		/* Re-use the name belonging to the slot for a matching parent
2071 		 * directory. All names for this context are relinquished in
2072 		 * audit_free_names() */
2073 		if (found_parent) {
2074 			context->names[idx].name = found_parent;
2075 			context->names[idx].name_len = AUDIT_NAME_FULL;
2076 			/* don't call __putname() */
2077 			context->names[idx].name_put = 0;
2078 		} else {
2079 			context->names[idx].name = NULL;
2080 		}
2081 
2082 		if (inode)
2083 			audit_copy_inode(&context->names[idx], NULL, inode);
2084 		else
2085 			context->names[idx].ino = (unsigned long)-1;
2086 	}
2087 }
2088 EXPORT_SYMBOL_GPL(__audit_inode_child);
2089 
2090 /**
2091  * auditsc_get_stamp - get local copies of audit_context values
2092  * @ctx: audit_context for the task
2093  * @t: timespec to store time recorded in the audit_context
2094  * @serial: serial value that is recorded in the audit_context
2095  *
2096  * Also sets the context as auditable.
2097  */
2098 int auditsc_get_stamp(struct audit_context *ctx,
2099 		       struct timespec *t, unsigned int *serial)
2100 {
2101 	if (!ctx->in_syscall)
2102 		return 0;
2103 	if (!ctx->serial)
2104 		ctx->serial = audit_serial();
2105 	t->tv_sec  = ctx->ctime.tv_sec;
2106 	t->tv_nsec = ctx->ctime.tv_nsec;
2107 	*serial    = ctx->serial;
2108 	if (!ctx->prio) {
2109 		ctx->prio = 1;
2110 		ctx->current_state = AUDIT_RECORD_CONTEXT;
2111 	}
2112 	return 1;
2113 }
2114 
2115 /* global counter which is incremented every time something logs in */
2116 static atomic_t session_id = ATOMIC_INIT(0);
2117 
2118 /**
2119  * audit_set_loginuid - set a task's audit_context loginuid
2120  * @task: task whose audit context is being modified
2121  * @loginuid: loginuid value
2122  *
2123  * Returns 0.
2124  *
2125  * Called (set) from fs/proc/base.c::proc_loginuid_write().
2126  */
2127 int audit_set_loginuid(struct task_struct *task, uid_t loginuid)
2128 {
2129 	unsigned int sessionid = atomic_inc_return(&session_id);
2130 	struct audit_context *context = task->audit_context;
2131 
2132 	if (context && context->in_syscall) {
2133 		struct audit_buffer *ab;
2134 
2135 		ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_LOGIN);
2136 		if (ab) {
2137 			audit_log_format(ab, "login pid=%d uid=%u "
2138 				"old auid=%u new auid=%u"
2139 				" old ses=%u new ses=%u",
2140 				task->pid, task_uid(task),
2141 				task->loginuid, loginuid,
2142 				task->sessionid, sessionid);
2143 			audit_log_end(ab);
2144 		}
2145 	}
2146 	task->sessionid = sessionid;
2147 	task->loginuid = loginuid;
2148 	return 0;
2149 }
2150 
2151 /**
2152  * __audit_mq_open - record audit data for a POSIX MQ open
2153  * @oflag: open flag
2154  * @mode: mode bits
2155  * @attr: queue attributes
2156  *
2157  */
2158 void __audit_mq_open(int oflag, mode_t mode, struct mq_attr *attr)
2159 {
2160 	struct audit_context *context = current->audit_context;
2161 
2162 	if (attr)
2163 		memcpy(&context->mq_open.attr, attr, sizeof(struct mq_attr));
2164 	else
2165 		memset(&context->mq_open.attr, 0, sizeof(struct mq_attr));
2166 
2167 	context->mq_open.oflag = oflag;
2168 	context->mq_open.mode = mode;
2169 
2170 	context->type = AUDIT_MQ_OPEN;
2171 }
2172 
2173 /**
2174  * __audit_mq_sendrecv - record audit data for a POSIX MQ timed send/receive
2175  * @mqdes: MQ descriptor
2176  * @msg_len: Message length
2177  * @msg_prio: Message priority
2178  * @abs_timeout: Message timeout in absolute time
2179  *
2180  */
2181 void __audit_mq_sendrecv(mqd_t mqdes, size_t msg_len, unsigned int msg_prio,
2182 			const struct timespec *abs_timeout)
2183 {
2184 	struct audit_context *context = current->audit_context;
2185 	struct timespec *p = &context->mq_sendrecv.abs_timeout;
2186 
2187 	if (abs_timeout)
2188 		memcpy(p, abs_timeout, sizeof(struct timespec));
2189 	else
2190 		memset(p, 0, sizeof(struct timespec));
2191 
2192 	context->mq_sendrecv.mqdes = mqdes;
2193 	context->mq_sendrecv.msg_len = msg_len;
2194 	context->mq_sendrecv.msg_prio = msg_prio;
2195 
2196 	context->type = AUDIT_MQ_SENDRECV;
2197 }
2198 
2199 /**
2200  * __audit_mq_notify - record audit data for a POSIX MQ notify
2201  * @mqdes: MQ descriptor
2202  * @notification: Notification event
2203  *
2204  */
2205 
2206 void __audit_mq_notify(mqd_t mqdes, const struct sigevent *notification)
2207 {
2208 	struct audit_context *context = current->audit_context;
2209 
2210 	if (notification)
2211 		context->mq_notify.sigev_signo = notification->sigev_signo;
2212 	else
2213 		context->mq_notify.sigev_signo = 0;
2214 
2215 	context->mq_notify.mqdes = mqdes;
2216 	context->type = AUDIT_MQ_NOTIFY;
2217 }
2218 
2219 /**
2220  * __audit_mq_getsetattr - record audit data for a POSIX MQ get/set attribute
2221  * @mqdes: MQ descriptor
2222  * @mqstat: MQ flags
2223  *
2224  */
2225 void __audit_mq_getsetattr(mqd_t mqdes, struct mq_attr *mqstat)
2226 {
2227 	struct audit_context *context = current->audit_context;
2228 	context->mq_getsetattr.mqdes = mqdes;
2229 	context->mq_getsetattr.mqstat = *mqstat;
2230 	context->type = AUDIT_MQ_GETSETATTR;
2231 }
2232 
2233 /**
2234  * audit_ipc_obj - record audit data for ipc object
2235  * @ipcp: ipc permissions
2236  *
2237  */
2238 void __audit_ipc_obj(struct kern_ipc_perm *ipcp)
2239 {
2240 	struct audit_context *context = current->audit_context;
2241 	context->ipc.uid = ipcp->uid;
2242 	context->ipc.gid = ipcp->gid;
2243 	context->ipc.mode = ipcp->mode;
2244 	context->ipc.has_perm = 0;
2245 	security_ipc_getsecid(ipcp, &context->ipc.osid);
2246 	context->type = AUDIT_IPC;
2247 }
2248 
2249 /**
2250  * audit_ipc_set_perm - record audit data for new ipc permissions
2251  * @qbytes: msgq bytes
2252  * @uid: msgq user id
2253  * @gid: msgq group id
2254  * @mode: msgq mode (permissions)
2255  *
2256  * Called only after audit_ipc_obj().
2257  */
2258 void __audit_ipc_set_perm(unsigned long qbytes, uid_t uid, gid_t gid, mode_t mode)
2259 {
2260 	struct audit_context *context = current->audit_context;
2261 
2262 	context->ipc.qbytes = qbytes;
2263 	context->ipc.perm_uid = uid;
2264 	context->ipc.perm_gid = gid;
2265 	context->ipc.perm_mode = mode;
2266 	context->ipc.has_perm = 1;
2267 }
2268 
2269 int audit_bprm(struct linux_binprm *bprm)
2270 {
2271 	struct audit_aux_data_execve *ax;
2272 	struct audit_context *context = current->audit_context;
2273 
2274 	if (likely(!audit_enabled || !context || context->dummy))
2275 		return 0;
2276 
2277 	ax = kmalloc(sizeof(*ax), GFP_KERNEL);
2278 	if (!ax)
2279 		return -ENOMEM;
2280 
2281 	ax->argc = bprm->argc;
2282 	ax->envc = bprm->envc;
2283 	ax->mm = bprm->mm;
2284 	ax->d.type = AUDIT_EXECVE;
2285 	ax->d.next = context->aux;
2286 	context->aux = (void *)ax;
2287 	return 0;
2288 }
2289 
2290 
2291 /**
2292  * audit_socketcall - record audit data for sys_socketcall
2293  * @nargs: number of args
2294  * @args: args array
2295  *
2296  */
2297 void audit_socketcall(int nargs, unsigned long *args)
2298 {
2299 	struct audit_context *context = current->audit_context;
2300 
2301 	if (likely(!context || context->dummy))
2302 		return;
2303 
2304 	context->type = AUDIT_SOCKETCALL;
2305 	context->socketcall.nargs = nargs;
2306 	memcpy(context->socketcall.args, args, nargs * sizeof(unsigned long));
2307 }
2308 
2309 /**
2310  * __audit_fd_pair - record audit data for pipe and socketpair
2311  * @fd1: the first file descriptor
2312  * @fd2: the second file descriptor
2313  *
2314  */
2315 void __audit_fd_pair(int fd1, int fd2)
2316 {
2317 	struct audit_context *context = current->audit_context;
2318 	context->fds[0] = fd1;
2319 	context->fds[1] = fd2;
2320 }
2321 
2322 /**
2323  * audit_sockaddr - record audit data for sys_bind, sys_connect, sys_sendto
2324  * @len: data length in user space
2325  * @a: data address in kernel space
2326  *
2327  * Returns 0 for success or NULL context or < 0 on error.
2328  */
2329 int audit_sockaddr(int len, void *a)
2330 {
2331 	struct audit_context *context = current->audit_context;
2332 
2333 	if (likely(!context || context->dummy))
2334 		return 0;
2335 
2336 	if (!context->sockaddr) {
2337 		void *p = kmalloc(sizeof(struct sockaddr_storage), GFP_KERNEL);
2338 		if (!p)
2339 			return -ENOMEM;
2340 		context->sockaddr = p;
2341 	}
2342 
2343 	context->sockaddr_len = len;
2344 	memcpy(context->sockaddr, a, len);
2345 	return 0;
2346 }
2347 
2348 void __audit_ptrace(struct task_struct *t)
2349 {
2350 	struct audit_context *context = current->audit_context;
2351 
2352 	context->target_pid = t->pid;
2353 	context->target_auid = audit_get_loginuid(t);
2354 	context->target_uid = task_uid(t);
2355 	context->target_sessionid = audit_get_sessionid(t);
2356 	security_task_getsecid(t, &context->target_sid);
2357 	memcpy(context->target_comm, t->comm, TASK_COMM_LEN);
2358 }
2359 
2360 /**
2361  * audit_signal_info - record signal info for shutting down audit subsystem
2362  * @sig: signal value
2363  * @t: task being signaled
2364  *
2365  * If the audit subsystem is being terminated, record the task (pid)
2366  * and uid that is doing that.
2367  */
2368 int __audit_signal_info(int sig, struct task_struct *t)
2369 {
2370 	struct audit_aux_data_pids *axp;
2371 	struct task_struct *tsk = current;
2372 	struct audit_context *ctx = tsk->audit_context;
2373 	uid_t uid = current_uid(), t_uid = task_uid(t);
2374 
2375 	if (audit_pid && t->tgid == audit_pid) {
2376 		if (sig == SIGTERM || sig == SIGHUP || sig == SIGUSR1 || sig == SIGUSR2) {
2377 			audit_sig_pid = tsk->pid;
2378 			if (tsk->loginuid != -1)
2379 				audit_sig_uid = tsk->loginuid;
2380 			else
2381 				audit_sig_uid = uid;
2382 			security_task_getsecid(tsk, &audit_sig_sid);
2383 		}
2384 		if (!audit_signals || audit_dummy_context())
2385 			return 0;
2386 	}
2387 
2388 	/* optimize the common case by putting first signal recipient directly
2389 	 * in audit_context */
2390 	if (!ctx->target_pid) {
2391 		ctx->target_pid = t->tgid;
2392 		ctx->target_auid = audit_get_loginuid(t);
2393 		ctx->target_uid = t_uid;
2394 		ctx->target_sessionid = audit_get_sessionid(t);
2395 		security_task_getsecid(t, &ctx->target_sid);
2396 		memcpy(ctx->target_comm, t->comm, TASK_COMM_LEN);
2397 		return 0;
2398 	}
2399 
2400 	axp = (void *)ctx->aux_pids;
2401 	if (!axp || axp->pid_count == AUDIT_AUX_PIDS) {
2402 		axp = kzalloc(sizeof(*axp), GFP_ATOMIC);
2403 		if (!axp)
2404 			return -ENOMEM;
2405 
2406 		axp->d.type = AUDIT_OBJ_PID;
2407 		axp->d.next = ctx->aux_pids;
2408 		ctx->aux_pids = (void *)axp;
2409 	}
2410 	BUG_ON(axp->pid_count >= AUDIT_AUX_PIDS);
2411 
2412 	axp->target_pid[axp->pid_count] = t->tgid;
2413 	axp->target_auid[axp->pid_count] = audit_get_loginuid(t);
2414 	axp->target_uid[axp->pid_count] = t_uid;
2415 	axp->target_sessionid[axp->pid_count] = audit_get_sessionid(t);
2416 	security_task_getsecid(t, &axp->target_sid[axp->pid_count]);
2417 	memcpy(axp->target_comm[axp->pid_count], t->comm, TASK_COMM_LEN);
2418 	axp->pid_count++;
2419 
2420 	return 0;
2421 }
2422 
2423 /**
2424  * __audit_log_bprm_fcaps - store information about a loading bprm and relevant fcaps
2425  * @bprm: pointer to the bprm being processed
2426  * @new: the proposed new credentials
2427  * @old: the old credentials
2428  *
2429  * Simply check if the proc already has the caps given by the file and if not
2430  * store the priv escalation info for later auditing at the end of the syscall
2431  *
2432  * -Eric
2433  */
2434 int __audit_log_bprm_fcaps(struct linux_binprm *bprm,
2435 			   const struct cred *new, const struct cred *old)
2436 {
2437 	struct audit_aux_data_bprm_fcaps *ax;
2438 	struct audit_context *context = current->audit_context;
2439 	struct cpu_vfs_cap_data vcaps;
2440 	struct dentry *dentry;
2441 
2442 	ax = kmalloc(sizeof(*ax), GFP_KERNEL);
2443 	if (!ax)
2444 		return -ENOMEM;
2445 
2446 	ax->d.type = AUDIT_BPRM_FCAPS;
2447 	ax->d.next = context->aux;
2448 	context->aux = (void *)ax;
2449 
2450 	dentry = dget(bprm->file->f_dentry);
2451 	get_vfs_caps_from_disk(dentry, &vcaps);
2452 	dput(dentry);
2453 
2454 	ax->fcap.permitted = vcaps.permitted;
2455 	ax->fcap.inheritable = vcaps.inheritable;
2456 	ax->fcap.fE = !!(vcaps.magic_etc & VFS_CAP_FLAGS_EFFECTIVE);
2457 	ax->fcap_ver = (vcaps.magic_etc & VFS_CAP_REVISION_MASK) >> VFS_CAP_REVISION_SHIFT;
2458 
2459 	ax->old_pcap.permitted   = old->cap_permitted;
2460 	ax->old_pcap.inheritable = old->cap_inheritable;
2461 	ax->old_pcap.effective   = old->cap_effective;
2462 
2463 	ax->new_pcap.permitted   = new->cap_permitted;
2464 	ax->new_pcap.inheritable = new->cap_inheritable;
2465 	ax->new_pcap.effective   = new->cap_effective;
2466 	return 0;
2467 }
2468 
2469 /**
2470  * __audit_log_capset - store information about the arguments to the capset syscall
2471  * @pid: target pid of the capset call
2472  * @new: the new credentials
2473  * @old: the old (current) credentials
2474  *
2475  * Record the aguments userspace sent to sys_capset for later printing by the
2476  * audit system if applicable
2477  */
2478 void __audit_log_capset(pid_t pid,
2479 		       const struct cred *new, const struct cred *old)
2480 {
2481 	struct audit_context *context = current->audit_context;
2482 	context->capset.pid = pid;
2483 	context->capset.cap.effective   = new->cap_effective;
2484 	context->capset.cap.inheritable = new->cap_effective;
2485 	context->capset.cap.permitted   = new->cap_permitted;
2486 	context->type = AUDIT_CAPSET;
2487 }
2488 
2489 /**
2490  * audit_core_dumps - record information about processes that end abnormally
2491  * @signr: signal value
2492  *
2493  * If a process ends with a core dump, something fishy is going on and we
2494  * should record the event for investigation.
2495  */
2496 void audit_core_dumps(long signr)
2497 {
2498 	struct audit_buffer *ab;
2499 	u32 sid;
2500 	uid_t auid = audit_get_loginuid(current), uid;
2501 	gid_t gid;
2502 	unsigned int sessionid = audit_get_sessionid(current);
2503 
2504 	if (!audit_enabled)
2505 		return;
2506 
2507 	if (signr == SIGQUIT)	/* don't care for those */
2508 		return;
2509 
2510 	ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_ANOM_ABEND);
2511 	current_uid_gid(&uid, &gid);
2512 	audit_log_format(ab, "auid=%u uid=%u gid=%u ses=%u",
2513 			 auid, uid, gid, sessionid);
2514 	security_task_getsecid(current, &sid);
2515 	if (sid) {
2516 		char *ctx = NULL;
2517 		u32 len;
2518 
2519 		if (security_secid_to_secctx(sid, &ctx, &len))
2520 			audit_log_format(ab, " ssid=%u", sid);
2521 		else {
2522 			audit_log_format(ab, " subj=%s", ctx);
2523 			security_release_secctx(ctx, len);
2524 		}
2525 	}
2526 	audit_log_format(ab, " pid=%d comm=", current->pid);
2527 	audit_log_untrustedstring(ab, current->comm);
2528 	audit_log_format(ab, " sig=%ld", signr);
2529 	audit_log_end(ab);
2530 }
2531 
2532 struct list_head *audit_killed_trees(void)
2533 {
2534 	struct audit_context *ctx = current->audit_context;
2535 	if (likely(!ctx || !ctx->in_syscall))
2536 		return NULL;
2537 	return &ctx->killed_trees;
2538 }
2539