page allocator: check only once if the zonelist is suitable for the allocation
[sfrench/cifs-2.6.git] / security / selinux / hooks.c
1 /*
2  *  NSA Security-Enhanced Linux (SELinux) security module
3  *
4  *  This file contains the SELinux hook function implementations.
5  *
6  *  Authors:  Stephen Smalley, <sds@epoch.ncsc.mil>
7  *            Chris Vance, <cvance@nai.com>
8  *            Wayne Salamon, <wsalamon@nai.com>
9  *            James Morris <jmorris@redhat.com>
10  *
11  *  Copyright (C) 2001,2002 Networks Associates Technology, Inc.
12  *  Copyright (C) 2003-2008 Red Hat, Inc., James Morris <jmorris@redhat.com>
13  *                                         Eric Paris <eparis@redhat.com>
14  *  Copyright (C) 2004-2005 Trusted Computer Solutions, Inc.
15  *                          <dgoeddel@trustedcs.com>
16  *  Copyright (C) 2006, 2007 Hewlett-Packard Development Company, L.P.
17  *              Paul Moore <paul.moore@hp.com>
18  *  Copyright (C) 2007 Hitachi Software Engineering Co., Ltd.
19  *                     Yuichi Nakamura <ynakam@hitachisoft.jp>
20  *
21  *      This program is free software; you can redistribute it and/or modify
22  *      it under the terms of the GNU General Public License version 2,
23  *      as published by the Free Software Foundation.
24  */
25
26 #include <linux/init.h>
27 #include <linux/kernel.h>
28 #include <linux/tracehook.h>
29 #include <linux/errno.h>
30 #include <linux/sched.h>
31 #include <linux/security.h>
32 #include <linux/xattr.h>
33 #include <linux/capability.h>
34 #include <linux/unistd.h>
35 #include <linux/mm.h>
36 #include <linux/mman.h>
37 #include <linux/slab.h>
38 #include <linux/pagemap.h>
39 #include <linux/swap.h>
40 #include <linux/spinlock.h>
41 #include <linux/syscalls.h>
42 #include <linux/file.h>
43 #include <linux/fdtable.h>
44 #include <linux/namei.h>
45 #include <linux/mount.h>
46 #include <linux/proc_fs.h>
47 #include <linux/netfilter_ipv4.h>
48 #include <linux/netfilter_ipv6.h>
49 #include <linux/tty.h>
50 #include <net/icmp.h>
51 #include <net/ip.h>             /* for local_port_range[] */
52 #include <net/tcp.h>            /* struct or_callable used in sock_rcv_skb */
53 #include <net/net_namespace.h>
54 #include <net/netlabel.h>
55 #include <linux/uaccess.h>
56 #include <asm/ioctls.h>
57 #include <asm/atomic.h>
58 #include <linux/bitops.h>
59 #include <linux/interrupt.h>
60 #include <linux/netdevice.h>    /* for network interface checks */
61 #include <linux/netlink.h>
62 #include <linux/tcp.h>
63 #include <linux/udp.h>
64 #include <linux/dccp.h>
65 #include <linux/quota.h>
66 #include <linux/un.h>           /* for Unix socket types */
67 #include <net/af_unix.h>        /* for Unix socket types */
68 #include <linux/parser.h>
69 #include <linux/nfs_mount.h>
70 #include <net/ipv6.h>
71 #include <linux/hugetlb.h>
72 #include <linux/personality.h>
73 #include <linux/sysctl.h>
74 #include <linux/audit.h>
75 #include <linux/string.h>
76 #include <linux/selinux.h>
77 #include <linux/mutex.h>
78 #include <linux/posix-timers.h>
79
80 #include "avc.h"
81 #include "objsec.h"
82 #include "netif.h"
83 #include "netnode.h"
84 #include "netport.h"
85 #include "xfrm.h"
86 #include "netlabel.h"
87 #include "audit.h"
88
89 #define XATTR_SELINUX_SUFFIX "selinux"
90 #define XATTR_NAME_SELINUX XATTR_SECURITY_PREFIX XATTR_SELINUX_SUFFIX
91
92 #define NUM_SEL_MNT_OPTS 5
93
94 extern unsigned int policydb_loaded_version;
95 extern int selinux_nlmsg_lookup(u16 sclass, u16 nlmsg_type, u32 *perm);
96 extern struct security_operations *security_ops;
97
98 /* SECMARK reference count */
99 atomic_t selinux_secmark_refcount = ATOMIC_INIT(0);
100
101 #ifdef CONFIG_SECURITY_SELINUX_DEVELOP
102 int selinux_enforcing;
103
104 static int __init enforcing_setup(char *str)
105 {
106         unsigned long enforcing;
107         if (!strict_strtoul(str, 0, &enforcing))
108                 selinux_enforcing = enforcing ? 1 : 0;
109         return 1;
110 }
111 __setup("enforcing=", enforcing_setup);
112 #endif
113
114 #ifdef CONFIG_SECURITY_SELINUX_BOOTPARAM
115 int selinux_enabled = CONFIG_SECURITY_SELINUX_BOOTPARAM_VALUE;
116
117 static int __init selinux_enabled_setup(char *str)
118 {
119         unsigned long enabled;
120         if (!strict_strtoul(str, 0, &enabled))
121                 selinux_enabled = enabled ? 1 : 0;
122         return 1;
123 }
124 __setup("selinux=", selinux_enabled_setup);
125 #else
126 int selinux_enabled = 1;
127 #endif
128
129
130 /*
131  * Minimal support for a secondary security module,
132  * just to allow the use of the capability module.
133  */
134 static struct security_operations *secondary_ops;
135
136 /* Lists of inode and superblock security structures initialized
137    before the policy was loaded. */
138 static LIST_HEAD(superblock_security_head);
139 static DEFINE_SPINLOCK(sb_security_lock);
140
141 static struct kmem_cache *sel_inode_cache;
142
143 /**
144  * selinux_secmark_enabled - Check to see if SECMARK is currently enabled
145  *
146  * Description:
147  * This function checks the SECMARK reference counter to see if any SECMARK
148  * targets are currently configured, if the reference counter is greater than
149  * zero SECMARK is considered to be enabled.  Returns true (1) if SECMARK is
150  * enabled, false (0) if SECMARK is disabled.
151  *
152  */
153 static int selinux_secmark_enabled(void)
154 {
155         return (atomic_read(&selinux_secmark_refcount) > 0);
156 }
157
158 /*
159  * initialise the security for the init task
160  */
161 static void cred_init_security(void)
162 {
163         struct cred *cred = (struct cred *) current->real_cred;
164         struct task_security_struct *tsec;
165
166         tsec = kzalloc(sizeof(struct task_security_struct), GFP_KERNEL);
167         if (!tsec)
168                 panic("SELinux:  Failed to initialize initial task.\n");
169
170         tsec->osid = tsec->sid = SECINITSID_KERNEL;
171         cred->security = tsec;
172 }
173
174 /*
175  * get the security ID of a set of credentials
176  */
177 static inline u32 cred_sid(const struct cred *cred)
178 {
179         const struct task_security_struct *tsec;
180
181         tsec = cred->security;
182         return tsec->sid;
183 }
184
185 /*
186  * get the objective security ID of a task
187  */
188 static inline u32 task_sid(const struct task_struct *task)
189 {
190         u32 sid;
191
192         rcu_read_lock();
193         sid = cred_sid(__task_cred(task));
194         rcu_read_unlock();
195         return sid;
196 }
197
198 /*
199  * get the subjective security ID of the current task
200  */
201 static inline u32 current_sid(void)
202 {
203         const struct task_security_struct *tsec = current_cred()->security;
204
205         return tsec->sid;
206 }
207
208 /* Allocate and free functions for each kind of security blob. */
209
210 static int inode_alloc_security(struct inode *inode)
211 {
212         struct inode_security_struct *isec;
213         u32 sid = current_sid();
214
215         isec = kmem_cache_zalloc(sel_inode_cache, GFP_NOFS);
216         if (!isec)
217                 return -ENOMEM;
218
219         mutex_init(&isec->lock);
220         INIT_LIST_HEAD(&isec->list);
221         isec->inode = inode;
222         isec->sid = SECINITSID_UNLABELED;
223         isec->sclass = SECCLASS_FILE;
224         isec->task_sid = sid;
225         inode->i_security = isec;
226
227         return 0;
228 }
229
230 static void inode_free_security(struct inode *inode)
231 {
232         struct inode_security_struct *isec = inode->i_security;
233         struct superblock_security_struct *sbsec = inode->i_sb->s_security;
234
235         spin_lock(&sbsec->isec_lock);
236         if (!list_empty(&isec->list))
237                 list_del_init(&isec->list);
238         spin_unlock(&sbsec->isec_lock);
239
240         inode->i_security = NULL;
241         kmem_cache_free(sel_inode_cache, isec);
242 }
243
244 static int file_alloc_security(struct file *file)
245 {
246         struct file_security_struct *fsec;
247         u32 sid = current_sid();
248
249         fsec = kzalloc(sizeof(struct file_security_struct), GFP_KERNEL);
250         if (!fsec)
251                 return -ENOMEM;
252
253         fsec->sid = sid;
254         fsec->fown_sid = sid;
255         file->f_security = fsec;
256
257         return 0;
258 }
259
260 static void file_free_security(struct file *file)
261 {
262         struct file_security_struct *fsec = file->f_security;
263         file->f_security = NULL;
264         kfree(fsec);
265 }
266
267 static int superblock_alloc_security(struct super_block *sb)
268 {
269         struct superblock_security_struct *sbsec;
270
271         sbsec = kzalloc(sizeof(struct superblock_security_struct), GFP_KERNEL);
272         if (!sbsec)
273                 return -ENOMEM;
274
275         mutex_init(&sbsec->lock);
276         INIT_LIST_HEAD(&sbsec->list);
277         INIT_LIST_HEAD(&sbsec->isec_head);
278         spin_lock_init(&sbsec->isec_lock);
279         sbsec->sb = sb;
280         sbsec->sid = SECINITSID_UNLABELED;
281         sbsec->def_sid = SECINITSID_FILE;
282         sbsec->mntpoint_sid = SECINITSID_UNLABELED;
283         sb->s_security = sbsec;
284
285         return 0;
286 }
287
288 static void superblock_free_security(struct super_block *sb)
289 {
290         struct superblock_security_struct *sbsec = sb->s_security;
291
292         spin_lock(&sb_security_lock);
293         if (!list_empty(&sbsec->list))
294                 list_del_init(&sbsec->list);
295         spin_unlock(&sb_security_lock);
296
297         sb->s_security = NULL;
298         kfree(sbsec);
299 }
300
301 static int sk_alloc_security(struct sock *sk, int family, gfp_t priority)
302 {
303         struct sk_security_struct *ssec;
304
305         ssec = kzalloc(sizeof(*ssec), priority);
306         if (!ssec)
307                 return -ENOMEM;
308
309         ssec->peer_sid = SECINITSID_UNLABELED;
310         ssec->sid = SECINITSID_UNLABELED;
311         sk->sk_security = ssec;
312
313         selinux_netlbl_sk_security_reset(ssec);
314
315         return 0;
316 }
317
318 static void sk_free_security(struct sock *sk)
319 {
320         struct sk_security_struct *ssec = sk->sk_security;
321
322         sk->sk_security = NULL;
323         selinux_netlbl_sk_security_free(ssec);
324         kfree(ssec);
325 }
326
327 /* The security server must be initialized before
328    any labeling or access decisions can be provided. */
329 extern int ss_initialized;
330
331 /* The file system's label must be initialized prior to use. */
332
333 static char *labeling_behaviors[6] = {
334         "uses xattr",
335         "uses transition SIDs",
336         "uses task SIDs",
337         "uses genfs_contexts",
338         "not configured for labeling",
339         "uses mountpoint labeling",
340 };
341
342 static int inode_doinit_with_dentry(struct inode *inode, struct dentry *opt_dentry);
343
344 static inline int inode_doinit(struct inode *inode)
345 {
346         return inode_doinit_with_dentry(inode, NULL);
347 }
348
349 enum {
350         Opt_error = -1,
351         Opt_context = 1,
352         Opt_fscontext = 2,
353         Opt_defcontext = 3,
354         Opt_rootcontext = 4,
355         Opt_labelsupport = 5,
356 };
357
358 static const match_table_t tokens = {
359         {Opt_context, CONTEXT_STR "%s"},
360         {Opt_fscontext, FSCONTEXT_STR "%s"},
361         {Opt_defcontext, DEFCONTEXT_STR "%s"},
362         {Opt_rootcontext, ROOTCONTEXT_STR "%s"},
363         {Opt_labelsupport, LABELSUPP_STR},
364         {Opt_error, NULL},
365 };
366
367 #define SEL_MOUNT_FAIL_MSG "SELinux:  duplicate or incompatible mount options\n"
368
369 static int may_context_mount_sb_relabel(u32 sid,
370                         struct superblock_security_struct *sbsec,
371                         const struct cred *cred)
372 {
373         const struct task_security_struct *tsec = cred->security;
374         int rc;
375
376         rc = avc_has_perm(tsec->sid, sbsec->sid, SECCLASS_FILESYSTEM,
377                           FILESYSTEM__RELABELFROM, NULL);
378         if (rc)
379                 return rc;
380
381         rc = avc_has_perm(tsec->sid, sid, SECCLASS_FILESYSTEM,
382                           FILESYSTEM__RELABELTO, NULL);
383         return rc;
384 }
385
386 static int may_context_mount_inode_relabel(u32 sid,
387                         struct superblock_security_struct *sbsec,
388                         const struct cred *cred)
389 {
390         const struct task_security_struct *tsec = cred->security;
391         int rc;
392         rc = avc_has_perm(tsec->sid, sbsec->sid, SECCLASS_FILESYSTEM,
393                           FILESYSTEM__RELABELFROM, NULL);
394         if (rc)
395                 return rc;
396
397         rc = avc_has_perm(sid, sbsec->sid, SECCLASS_FILESYSTEM,
398                           FILESYSTEM__ASSOCIATE, NULL);
399         return rc;
400 }
401
402 static int sb_finish_set_opts(struct super_block *sb)
403 {
404         struct superblock_security_struct *sbsec = sb->s_security;
405         struct dentry *root = sb->s_root;
406         struct inode *root_inode = root->d_inode;
407         int rc = 0;
408
409         if (sbsec->behavior == SECURITY_FS_USE_XATTR) {
410                 /* Make sure that the xattr handler exists and that no
411                    error other than -ENODATA is returned by getxattr on
412                    the root directory.  -ENODATA is ok, as this may be
413                    the first boot of the SELinux kernel before we have
414                    assigned xattr values to the filesystem. */
415                 if (!root_inode->i_op->getxattr) {
416                         printk(KERN_WARNING "SELinux: (dev %s, type %s) has no "
417                                "xattr support\n", sb->s_id, sb->s_type->name);
418                         rc = -EOPNOTSUPP;
419                         goto out;
420                 }
421                 rc = root_inode->i_op->getxattr(root, XATTR_NAME_SELINUX, NULL, 0);
422                 if (rc < 0 && rc != -ENODATA) {
423                         if (rc == -EOPNOTSUPP)
424                                 printk(KERN_WARNING "SELinux: (dev %s, type "
425                                        "%s) has no security xattr handler\n",
426                                        sb->s_id, sb->s_type->name);
427                         else
428                                 printk(KERN_WARNING "SELinux: (dev %s, type "
429                                        "%s) getxattr errno %d\n", sb->s_id,
430                                        sb->s_type->name, -rc);
431                         goto out;
432                 }
433         }
434
435         sbsec->flags |= (SE_SBINITIALIZED | SE_SBLABELSUPP);
436
437         if (sbsec->behavior > ARRAY_SIZE(labeling_behaviors))
438                 printk(KERN_ERR "SELinux: initialized (dev %s, type %s), unknown behavior\n",
439                        sb->s_id, sb->s_type->name);
440         else
441                 printk(KERN_DEBUG "SELinux: initialized (dev %s, type %s), %s\n",
442                        sb->s_id, sb->s_type->name,
443                        labeling_behaviors[sbsec->behavior-1]);
444
445         if (sbsec->behavior == SECURITY_FS_USE_GENFS ||
446             sbsec->behavior == SECURITY_FS_USE_MNTPOINT ||
447             sbsec->behavior == SECURITY_FS_USE_NONE ||
448             sbsec->behavior > ARRAY_SIZE(labeling_behaviors))
449                 sbsec->flags &= ~SE_SBLABELSUPP;
450
451         /* Initialize the root inode. */
452         rc = inode_doinit_with_dentry(root_inode, root);
453
454         /* Initialize any other inodes associated with the superblock, e.g.
455            inodes created prior to initial policy load or inodes created
456            during get_sb by a pseudo filesystem that directly
457            populates itself. */
458         spin_lock(&sbsec->isec_lock);
459 next_inode:
460         if (!list_empty(&sbsec->isec_head)) {
461                 struct inode_security_struct *isec =
462                                 list_entry(sbsec->isec_head.next,
463                                            struct inode_security_struct, list);
464                 struct inode *inode = isec->inode;
465                 spin_unlock(&sbsec->isec_lock);
466                 inode = igrab(inode);
467                 if (inode) {
468                         if (!IS_PRIVATE(inode))
469                                 inode_doinit(inode);
470                         iput(inode);
471                 }
472                 spin_lock(&sbsec->isec_lock);
473                 list_del_init(&isec->list);
474                 goto next_inode;
475         }
476         spin_unlock(&sbsec->isec_lock);
477 out:
478         return rc;
479 }
480
481 /*
482  * This function should allow an FS to ask what it's mount security
483  * options were so it can use those later for submounts, displaying
484  * mount options, or whatever.
485  */
486 static int selinux_get_mnt_opts(const struct super_block *sb,
487                                 struct security_mnt_opts *opts)
488 {
489         int rc = 0, i;
490         struct superblock_security_struct *sbsec = sb->s_security;
491         char *context = NULL;
492         u32 len;
493         char tmp;
494
495         security_init_mnt_opts(opts);
496
497         if (!(sbsec->flags & SE_SBINITIALIZED))
498                 return -EINVAL;
499
500         if (!ss_initialized)
501                 return -EINVAL;
502
503         tmp = sbsec->flags & SE_MNTMASK;
504         /* count the number of mount options for this sb */
505         for (i = 0; i < 8; i++) {
506                 if (tmp & 0x01)
507                         opts->num_mnt_opts++;
508                 tmp >>= 1;
509         }
510         /* Check if the Label support flag is set */
511         if (sbsec->flags & SE_SBLABELSUPP)
512                 opts->num_mnt_opts++;
513
514         opts->mnt_opts = kcalloc(opts->num_mnt_opts, sizeof(char *), GFP_ATOMIC);
515         if (!opts->mnt_opts) {
516                 rc = -ENOMEM;
517                 goto out_free;
518         }
519
520         opts->mnt_opts_flags = kcalloc(opts->num_mnt_opts, sizeof(int), GFP_ATOMIC);
521         if (!opts->mnt_opts_flags) {
522                 rc = -ENOMEM;
523                 goto out_free;
524         }
525
526         i = 0;
527         if (sbsec->flags & FSCONTEXT_MNT) {
528                 rc = security_sid_to_context(sbsec->sid, &context, &len);
529                 if (rc)
530                         goto out_free;
531                 opts->mnt_opts[i] = context;
532                 opts->mnt_opts_flags[i++] = FSCONTEXT_MNT;
533         }
534         if (sbsec->flags & CONTEXT_MNT) {
535                 rc = security_sid_to_context(sbsec->mntpoint_sid, &context, &len);
536                 if (rc)
537                         goto out_free;
538                 opts->mnt_opts[i] = context;
539                 opts->mnt_opts_flags[i++] = CONTEXT_MNT;
540         }
541         if (sbsec->flags & DEFCONTEXT_MNT) {
542                 rc = security_sid_to_context(sbsec->def_sid, &context, &len);
543                 if (rc)
544                         goto out_free;
545                 opts->mnt_opts[i] = context;
546                 opts->mnt_opts_flags[i++] = DEFCONTEXT_MNT;
547         }
548         if (sbsec->flags & ROOTCONTEXT_MNT) {
549                 struct inode *root = sbsec->sb->s_root->d_inode;
550                 struct inode_security_struct *isec = root->i_security;
551
552                 rc = security_sid_to_context(isec->sid, &context, &len);
553                 if (rc)
554                         goto out_free;
555                 opts->mnt_opts[i] = context;
556                 opts->mnt_opts_flags[i++] = ROOTCONTEXT_MNT;
557         }
558         if (sbsec->flags & SE_SBLABELSUPP) {
559                 opts->mnt_opts[i] = NULL;
560                 opts->mnt_opts_flags[i++] = SE_SBLABELSUPP;
561         }
562
563         BUG_ON(i != opts->num_mnt_opts);
564
565         return 0;
566
567 out_free:
568         security_free_mnt_opts(opts);
569         return rc;
570 }
571
572 static int bad_option(struct superblock_security_struct *sbsec, char flag,
573                       u32 old_sid, u32 new_sid)
574 {
575         char mnt_flags = sbsec->flags & SE_MNTMASK;
576
577         /* check if the old mount command had the same options */
578         if (sbsec->flags & SE_SBINITIALIZED)
579                 if (!(sbsec->flags & flag) ||
580                     (old_sid != new_sid))
581                         return 1;
582
583         /* check if we were passed the same options twice,
584          * aka someone passed context=a,context=b
585          */
586         if (!(sbsec->flags & SE_SBINITIALIZED))
587                 if (mnt_flags & flag)
588                         return 1;
589         return 0;
590 }
591
592 /*
593  * Allow filesystems with binary mount data to explicitly set mount point
594  * labeling information.
595  */
596 static int selinux_set_mnt_opts(struct super_block *sb,
597                                 struct security_mnt_opts *opts)
598 {
599         const struct cred *cred = current_cred();
600         int rc = 0, i;
601         struct superblock_security_struct *sbsec = sb->s_security;
602         const char *name = sb->s_type->name;
603         struct inode *inode = sbsec->sb->s_root->d_inode;
604         struct inode_security_struct *root_isec = inode->i_security;
605         u32 fscontext_sid = 0, context_sid = 0, rootcontext_sid = 0;
606         u32 defcontext_sid = 0;
607         char **mount_options = opts->mnt_opts;
608         int *flags = opts->mnt_opts_flags;
609         int num_opts = opts->num_mnt_opts;
610
611         mutex_lock(&sbsec->lock);
612
613         if (!ss_initialized) {
614                 if (!num_opts) {
615                         /* Defer initialization until selinux_complete_init,
616                            after the initial policy is loaded and the security
617                            server is ready to handle calls. */
618                         spin_lock(&sb_security_lock);
619                         if (list_empty(&sbsec->list))
620                                 list_add(&sbsec->list, &superblock_security_head);
621                         spin_unlock(&sb_security_lock);
622                         goto out;
623                 }
624                 rc = -EINVAL;
625                 printk(KERN_WARNING "SELinux: Unable to set superblock options "
626                         "before the security server is initialized\n");
627                 goto out;
628         }
629
630         /*
631          * Binary mount data FS will come through this function twice.  Once
632          * from an explicit call and once from the generic calls from the vfs.
633          * Since the generic VFS calls will not contain any security mount data
634          * we need to skip the double mount verification.
635          *
636          * This does open a hole in which we will not notice if the first
637          * mount using this sb set explict options and a second mount using
638          * this sb does not set any security options.  (The first options
639          * will be used for both mounts)
640          */
641         if ((sbsec->flags & SE_SBINITIALIZED) && (sb->s_type->fs_flags & FS_BINARY_MOUNTDATA)
642             && (num_opts == 0))
643                 goto out;
644
645         /*
646          * parse the mount options, check if they are valid sids.
647          * also check if someone is trying to mount the same sb more
648          * than once with different security options.
649          */
650         for (i = 0; i < num_opts; i++) {
651                 u32 sid;
652
653                 if (flags[i] == SE_SBLABELSUPP)
654                         continue;
655                 rc = security_context_to_sid(mount_options[i],
656                                              strlen(mount_options[i]), &sid);
657                 if (rc) {
658                         printk(KERN_WARNING "SELinux: security_context_to_sid"
659                                "(%s) failed for (dev %s, type %s) errno=%d\n",
660                                mount_options[i], sb->s_id, name, rc);
661                         goto out;
662                 }
663                 switch (flags[i]) {
664                 case FSCONTEXT_MNT:
665                         fscontext_sid = sid;
666
667                         if (bad_option(sbsec, FSCONTEXT_MNT, sbsec->sid,
668                                         fscontext_sid))
669                                 goto out_double_mount;
670
671                         sbsec->flags |= FSCONTEXT_MNT;
672                         break;
673                 case CONTEXT_MNT:
674                         context_sid = sid;
675
676                         if (bad_option(sbsec, CONTEXT_MNT, sbsec->mntpoint_sid,
677                                         context_sid))
678                                 goto out_double_mount;
679
680                         sbsec->flags |= CONTEXT_MNT;
681                         break;
682                 case ROOTCONTEXT_MNT:
683                         rootcontext_sid = sid;
684
685                         if (bad_option(sbsec, ROOTCONTEXT_MNT, root_isec->sid,
686                                         rootcontext_sid))
687                                 goto out_double_mount;
688
689                         sbsec->flags |= ROOTCONTEXT_MNT;
690
691                         break;
692                 case DEFCONTEXT_MNT:
693                         defcontext_sid = sid;
694
695                         if (bad_option(sbsec, DEFCONTEXT_MNT, sbsec->def_sid,
696                                         defcontext_sid))
697                                 goto out_double_mount;
698
699                         sbsec->flags |= DEFCONTEXT_MNT;
700
701                         break;
702                 default:
703                         rc = -EINVAL;
704                         goto out;
705                 }
706         }
707
708         if (sbsec->flags & SE_SBINITIALIZED) {
709                 /* previously mounted with options, but not on this attempt? */
710                 if ((sbsec->flags & SE_MNTMASK) && !num_opts)
711                         goto out_double_mount;
712                 rc = 0;
713                 goto out;
714         }
715
716         if (strcmp(sb->s_type->name, "proc") == 0)
717                 sbsec->flags |= SE_SBPROC;
718
719         /* Determine the labeling behavior to use for this filesystem type. */
720         rc = security_fs_use((sbsec->flags & SE_SBPROC) ? "proc" : sb->s_type->name, &sbsec->behavior, &sbsec->sid);
721         if (rc) {
722                 printk(KERN_WARNING "%s: security_fs_use(%s) returned %d\n",
723                        __func__, sb->s_type->name, rc);
724                 goto out;
725         }
726
727         /* sets the context of the superblock for the fs being mounted. */
728         if (fscontext_sid) {
729                 rc = may_context_mount_sb_relabel(fscontext_sid, sbsec, cred);
730                 if (rc)
731                         goto out;
732
733                 sbsec->sid = fscontext_sid;
734         }
735
736         /*
737          * Switch to using mount point labeling behavior.
738          * sets the label used on all file below the mountpoint, and will set
739          * the superblock context if not already set.
740          */
741         if (context_sid) {
742                 if (!fscontext_sid) {
743                         rc = may_context_mount_sb_relabel(context_sid, sbsec,
744                                                           cred);
745                         if (rc)
746                                 goto out;
747                         sbsec->sid = context_sid;
748                 } else {
749                         rc = may_context_mount_inode_relabel(context_sid, sbsec,
750                                                              cred);
751                         if (rc)
752                                 goto out;
753                 }
754                 if (!rootcontext_sid)
755                         rootcontext_sid = context_sid;
756
757                 sbsec->mntpoint_sid = context_sid;
758                 sbsec->behavior = SECURITY_FS_USE_MNTPOINT;
759         }
760
761         if (rootcontext_sid) {
762                 rc = may_context_mount_inode_relabel(rootcontext_sid, sbsec,
763                                                      cred);
764                 if (rc)
765                         goto out;
766
767                 root_isec->sid = rootcontext_sid;
768                 root_isec->initialized = 1;
769         }
770
771         if (defcontext_sid) {
772                 if (sbsec->behavior != SECURITY_FS_USE_XATTR) {
773                         rc = -EINVAL;
774                         printk(KERN_WARNING "SELinux: defcontext option is "
775                                "invalid for this filesystem type\n");
776                         goto out;
777                 }
778
779                 if (defcontext_sid != sbsec->def_sid) {
780                         rc = may_context_mount_inode_relabel(defcontext_sid,
781                                                              sbsec, cred);
782                         if (rc)
783                                 goto out;
784                 }
785
786                 sbsec->def_sid = defcontext_sid;
787         }
788
789         rc = sb_finish_set_opts(sb);
790 out:
791         mutex_unlock(&sbsec->lock);
792         return rc;
793 out_double_mount:
794         rc = -EINVAL;
795         printk(KERN_WARNING "SELinux: mount invalid.  Same superblock, different "
796                "security settings for (dev %s, type %s)\n", sb->s_id, name);
797         goto out;
798 }
799
800 static void selinux_sb_clone_mnt_opts(const struct super_block *oldsb,
801                                         struct super_block *newsb)
802 {
803         const struct superblock_security_struct *oldsbsec = oldsb->s_security;
804         struct superblock_security_struct *newsbsec = newsb->s_security;
805
806         int set_fscontext =     (oldsbsec->flags & FSCONTEXT_MNT);
807         int set_context =       (oldsbsec->flags & CONTEXT_MNT);
808         int set_rootcontext =   (oldsbsec->flags & ROOTCONTEXT_MNT);
809
810         /*
811          * if the parent was able to be mounted it clearly had no special lsm
812          * mount options.  thus we can safely put this sb on the list and deal
813          * with it later
814          */
815         if (!ss_initialized) {
816                 spin_lock(&sb_security_lock);
817                 if (list_empty(&newsbsec->list))
818                         list_add(&newsbsec->list, &superblock_security_head);
819                 spin_unlock(&sb_security_lock);
820                 return;
821         }
822
823         /* how can we clone if the old one wasn't set up?? */
824         BUG_ON(!(oldsbsec->flags & SE_SBINITIALIZED));
825
826         /* if fs is reusing a sb, just let its options stand... */
827         if (newsbsec->flags & SE_SBINITIALIZED)
828                 return;
829
830         mutex_lock(&newsbsec->lock);
831
832         newsbsec->flags = oldsbsec->flags;
833
834         newsbsec->sid = oldsbsec->sid;
835         newsbsec->def_sid = oldsbsec->def_sid;
836         newsbsec->behavior = oldsbsec->behavior;
837
838         if (set_context) {
839                 u32 sid = oldsbsec->mntpoint_sid;
840
841                 if (!set_fscontext)
842                         newsbsec->sid = sid;
843                 if (!set_rootcontext) {
844                         struct inode *newinode = newsb->s_root->d_inode;
845                         struct inode_security_struct *newisec = newinode->i_security;
846                         newisec->sid = sid;
847                 }
848                 newsbsec->mntpoint_sid = sid;
849         }
850         if (set_rootcontext) {
851                 const struct inode *oldinode = oldsb->s_root->d_inode;
852                 const struct inode_security_struct *oldisec = oldinode->i_security;
853                 struct inode *newinode = newsb->s_root->d_inode;
854                 struct inode_security_struct *newisec = newinode->i_security;
855
856                 newisec->sid = oldisec->sid;
857         }
858
859         sb_finish_set_opts(newsb);
860         mutex_unlock(&newsbsec->lock);
861 }
862
863 static int selinux_parse_opts_str(char *options,
864                                   struct security_mnt_opts *opts)
865 {
866         char *p;
867         char *context = NULL, *defcontext = NULL;
868         char *fscontext = NULL, *rootcontext = NULL;
869         int rc, num_mnt_opts = 0;
870
871         opts->num_mnt_opts = 0;
872
873         /* Standard string-based options. */
874         while ((p = strsep(&options, "|")) != NULL) {
875                 int token;
876                 substring_t args[MAX_OPT_ARGS];
877
878                 if (!*p)
879                         continue;
880
881                 token = match_token(p, tokens, args);
882
883                 switch (token) {
884                 case Opt_context:
885                         if (context || defcontext) {
886                                 rc = -EINVAL;
887                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
888                                 goto out_err;
889                         }
890                         context = match_strdup(&args[0]);
891                         if (!context) {
892                                 rc = -ENOMEM;
893                                 goto out_err;
894                         }
895                         break;
896
897                 case Opt_fscontext:
898                         if (fscontext) {
899                                 rc = -EINVAL;
900                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
901                                 goto out_err;
902                         }
903                         fscontext = match_strdup(&args[0]);
904                         if (!fscontext) {
905                                 rc = -ENOMEM;
906                                 goto out_err;
907                         }
908                         break;
909
910                 case Opt_rootcontext:
911                         if (rootcontext) {
912                                 rc = -EINVAL;
913                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
914                                 goto out_err;
915                         }
916                         rootcontext = match_strdup(&args[0]);
917                         if (!rootcontext) {
918                                 rc = -ENOMEM;
919                                 goto out_err;
920                         }
921                         break;
922
923                 case Opt_defcontext:
924                         if (context || defcontext) {
925                                 rc = -EINVAL;
926                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
927                                 goto out_err;
928                         }
929                         defcontext = match_strdup(&args[0]);
930                         if (!defcontext) {
931                                 rc = -ENOMEM;
932                                 goto out_err;
933                         }
934                         break;
935                 case Opt_labelsupport:
936                         break;
937                 default:
938                         rc = -EINVAL;
939                         printk(KERN_WARNING "SELinux:  unknown mount option\n");
940                         goto out_err;
941
942                 }
943         }
944
945         rc = -ENOMEM;
946         opts->mnt_opts = kcalloc(NUM_SEL_MNT_OPTS, sizeof(char *), GFP_ATOMIC);
947         if (!opts->mnt_opts)
948                 goto out_err;
949
950         opts->mnt_opts_flags = kcalloc(NUM_SEL_MNT_OPTS, sizeof(int), GFP_ATOMIC);
951         if (!opts->mnt_opts_flags) {
952                 kfree(opts->mnt_opts);
953                 goto out_err;
954         }
955
956         if (fscontext) {
957                 opts->mnt_opts[num_mnt_opts] = fscontext;
958                 opts->mnt_opts_flags[num_mnt_opts++] = FSCONTEXT_MNT;
959         }
960         if (context) {
961                 opts->mnt_opts[num_mnt_opts] = context;
962                 opts->mnt_opts_flags[num_mnt_opts++] = CONTEXT_MNT;
963         }
964         if (rootcontext) {
965                 opts->mnt_opts[num_mnt_opts] = rootcontext;
966                 opts->mnt_opts_flags[num_mnt_opts++] = ROOTCONTEXT_MNT;
967         }
968         if (defcontext) {
969                 opts->mnt_opts[num_mnt_opts] = defcontext;
970                 opts->mnt_opts_flags[num_mnt_opts++] = DEFCONTEXT_MNT;
971         }
972
973         opts->num_mnt_opts = num_mnt_opts;
974         return 0;
975
976 out_err:
977         kfree(context);
978         kfree(defcontext);
979         kfree(fscontext);
980         kfree(rootcontext);
981         return rc;
982 }
983 /*
984  * string mount options parsing and call set the sbsec
985  */
986 static int superblock_doinit(struct super_block *sb, void *data)
987 {
988         int rc = 0;
989         char *options = data;
990         struct security_mnt_opts opts;
991
992         security_init_mnt_opts(&opts);
993
994         if (!data)
995                 goto out;
996
997         BUG_ON(sb->s_type->fs_flags & FS_BINARY_MOUNTDATA);
998
999         rc = selinux_parse_opts_str(options, &opts);
1000         if (rc)
1001                 goto out_err;
1002
1003 out:
1004         rc = selinux_set_mnt_opts(sb, &opts);
1005
1006 out_err:
1007         security_free_mnt_opts(&opts);
1008         return rc;
1009 }
1010
1011 static void selinux_write_opts(struct seq_file *m,
1012                                struct security_mnt_opts *opts)
1013 {
1014         int i;
1015         char *prefix;
1016
1017         for (i = 0; i < opts->num_mnt_opts; i++) {
1018                 char *has_comma;
1019
1020                 if (opts->mnt_opts[i])
1021                         has_comma = strchr(opts->mnt_opts[i], ',');
1022                 else
1023                         has_comma = NULL;
1024
1025                 switch (opts->mnt_opts_flags[i]) {
1026                 case CONTEXT_MNT:
1027                         prefix = CONTEXT_STR;
1028                         break;
1029                 case FSCONTEXT_MNT:
1030                         prefix = FSCONTEXT_STR;
1031                         break;
1032                 case ROOTCONTEXT_MNT:
1033                         prefix = ROOTCONTEXT_STR;
1034                         break;
1035                 case DEFCONTEXT_MNT:
1036                         prefix = DEFCONTEXT_STR;
1037                         break;
1038                 case SE_SBLABELSUPP:
1039                         seq_putc(m, ',');
1040                         seq_puts(m, LABELSUPP_STR);
1041                         continue;
1042                 default:
1043                         BUG();
1044                 };
1045                 /* we need a comma before each option */
1046                 seq_putc(m, ',');
1047                 seq_puts(m, prefix);
1048                 if (has_comma)
1049                         seq_putc(m, '\"');
1050                 seq_puts(m, opts->mnt_opts[i]);
1051                 if (has_comma)
1052                         seq_putc(m, '\"');
1053         }
1054 }
1055
1056 static int selinux_sb_show_options(struct seq_file *m, struct super_block *sb)
1057 {
1058         struct security_mnt_opts opts;
1059         int rc;
1060
1061         rc = selinux_get_mnt_opts(sb, &opts);
1062         if (rc) {
1063                 /* before policy load we may get EINVAL, don't show anything */
1064                 if (rc == -EINVAL)
1065                         rc = 0;
1066                 return rc;
1067         }
1068
1069         selinux_write_opts(m, &opts);
1070
1071         security_free_mnt_opts(&opts);
1072
1073         return rc;
1074 }
1075
1076 static inline u16 inode_mode_to_security_class(umode_t mode)
1077 {
1078         switch (mode & S_IFMT) {
1079         case S_IFSOCK:
1080                 return SECCLASS_SOCK_FILE;
1081         case S_IFLNK:
1082                 return SECCLASS_LNK_FILE;
1083         case S_IFREG:
1084                 return SECCLASS_FILE;
1085         case S_IFBLK:
1086                 return SECCLASS_BLK_FILE;
1087         case S_IFDIR:
1088                 return SECCLASS_DIR;
1089         case S_IFCHR:
1090                 return SECCLASS_CHR_FILE;
1091         case S_IFIFO:
1092                 return SECCLASS_FIFO_FILE;
1093
1094         }
1095
1096         return SECCLASS_FILE;
1097 }
1098
1099 static inline int default_protocol_stream(int protocol)
1100 {
1101         return (protocol == IPPROTO_IP || protocol == IPPROTO_TCP);
1102 }
1103
1104 static inline int default_protocol_dgram(int protocol)
1105 {
1106         return (protocol == IPPROTO_IP || protocol == IPPROTO_UDP);
1107 }
1108
1109 static inline u16 socket_type_to_security_class(int family, int type, int protocol)
1110 {
1111         switch (family) {
1112         case PF_UNIX:
1113                 switch (type) {
1114                 case SOCK_STREAM:
1115                 case SOCK_SEQPACKET:
1116                         return SECCLASS_UNIX_STREAM_SOCKET;
1117                 case SOCK_DGRAM:
1118                         return SECCLASS_UNIX_DGRAM_SOCKET;
1119                 }
1120                 break;
1121         case PF_INET:
1122         case PF_INET6:
1123                 switch (type) {
1124                 case SOCK_STREAM:
1125                         if (default_protocol_stream(protocol))
1126                                 return SECCLASS_TCP_SOCKET;
1127                         else
1128                                 return SECCLASS_RAWIP_SOCKET;
1129                 case SOCK_DGRAM:
1130                         if (default_protocol_dgram(protocol))
1131                                 return SECCLASS_UDP_SOCKET;
1132                         else
1133                                 return SECCLASS_RAWIP_SOCKET;
1134                 case SOCK_DCCP:
1135                         return SECCLASS_DCCP_SOCKET;
1136                 default:
1137                         return SECCLASS_RAWIP_SOCKET;
1138                 }
1139                 break;
1140         case PF_NETLINK:
1141                 switch (protocol) {
1142                 case NETLINK_ROUTE:
1143                         return SECCLASS_NETLINK_ROUTE_SOCKET;
1144                 case NETLINK_FIREWALL:
1145                         return SECCLASS_NETLINK_FIREWALL_SOCKET;
1146                 case NETLINK_INET_DIAG:
1147                         return SECCLASS_NETLINK_TCPDIAG_SOCKET;
1148                 case NETLINK_NFLOG:
1149                         return SECCLASS_NETLINK_NFLOG_SOCKET;
1150                 case NETLINK_XFRM:
1151                         return SECCLASS_NETLINK_XFRM_SOCKET;
1152                 case NETLINK_SELINUX:
1153                         return SECCLASS_NETLINK_SELINUX_SOCKET;
1154                 case NETLINK_AUDIT:
1155                         return SECCLASS_NETLINK_AUDIT_SOCKET;
1156                 case NETLINK_IP6_FW:
1157                         return SECCLASS_NETLINK_IP6FW_SOCKET;
1158                 case NETLINK_DNRTMSG:
1159                         return SECCLASS_NETLINK_DNRT_SOCKET;
1160                 case NETLINK_KOBJECT_UEVENT:
1161                         return SECCLASS_NETLINK_KOBJECT_UEVENT_SOCKET;
1162                 default:
1163                         return SECCLASS_NETLINK_SOCKET;
1164                 }
1165         case PF_PACKET:
1166                 return SECCLASS_PACKET_SOCKET;
1167         case PF_KEY:
1168                 return SECCLASS_KEY_SOCKET;
1169         case PF_APPLETALK:
1170                 return SECCLASS_APPLETALK_SOCKET;
1171         }
1172
1173         return SECCLASS_SOCKET;
1174 }
1175
1176 #ifdef CONFIG_PROC_FS
1177 static int selinux_proc_get_sid(struct proc_dir_entry *de,
1178                                 u16 tclass,
1179                                 u32 *sid)
1180 {
1181         int buflen, rc;
1182         char *buffer, *path, *end;
1183
1184         buffer = (char *)__get_free_page(GFP_KERNEL);
1185         if (!buffer)
1186                 return -ENOMEM;
1187
1188         buflen = PAGE_SIZE;
1189         end = buffer+buflen;
1190         *--end = '\0';
1191         buflen--;
1192         path = end-1;
1193         *path = '/';
1194         while (de && de != de->parent) {
1195                 buflen -= de->namelen + 1;
1196                 if (buflen < 0)
1197                         break;
1198                 end -= de->namelen;
1199                 memcpy(end, de->name, de->namelen);
1200                 *--end = '/';
1201                 path = end;
1202                 de = de->parent;
1203         }
1204         rc = security_genfs_sid("proc", path, tclass, sid);
1205         free_page((unsigned long)buffer);
1206         return rc;
1207 }
1208 #else
1209 static int selinux_proc_get_sid(struct proc_dir_entry *de,
1210                                 u16 tclass,
1211                                 u32 *sid)
1212 {
1213         return -EINVAL;
1214 }
1215 #endif
1216
1217 /* The inode's security attributes must be initialized before first use. */
1218 static int inode_doinit_with_dentry(struct inode *inode, struct dentry *opt_dentry)
1219 {
1220         struct superblock_security_struct *sbsec = NULL;
1221         struct inode_security_struct *isec = inode->i_security;
1222         u32 sid;
1223         struct dentry *dentry;
1224 #define INITCONTEXTLEN 255
1225         char *context = NULL;
1226         unsigned len = 0;
1227         int rc = 0;
1228
1229         if (isec->initialized)
1230                 goto out;
1231
1232         mutex_lock(&isec->lock);
1233         if (isec->initialized)
1234                 goto out_unlock;
1235
1236         sbsec = inode->i_sb->s_security;
1237         if (!(sbsec->flags & SE_SBINITIALIZED)) {
1238                 /* Defer initialization until selinux_complete_init,
1239                    after the initial policy is loaded and the security
1240                    server is ready to handle calls. */
1241                 spin_lock(&sbsec->isec_lock);
1242                 if (list_empty(&isec->list))
1243                         list_add(&isec->list, &sbsec->isec_head);
1244                 spin_unlock(&sbsec->isec_lock);
1245                 goto out_unlock;
1246         }
1247
1248         switch (sbsec->behavior) {
1249         case SECURITY_FS_USE_XATTR:
1250                 if (!inode->i_op->getxattr) {
1251                         isec->sid = sbsec->def_sid;
1252                         break;
1253                 }
1254
1255                 /* Need a dentry, since the xattr API requires one.
1256                    Life would be simpler if we could just pass the inode. */
1257                 if (opt_dentry) {
1258                         /* Called from d_instantiate or d_splice_alias. */
1259                         dentry = dget(opt_dentry);
1260                 } else {
1261                         /* Called from selinux_complete_init, try to find a dentry. */
1262                         dentry = d_find_alias(inode);
1263                 }
1264                 if (!dentry) {
1265                         /*
1266                          * this is can be hit on boot when a file is accessed
1267                          * before the policy is loaded.  When we load policy we
1268                          * may find inodes that have no dentry on the
1269                          * sbsec->isec_head list.  No reason to complain as these
1270                          * will get fixed up the next time we go through
1271                          * inode_doinit with a dentry, before these inodes could
1272                          * be used again by userspace.
1273                          */
1274                         goto out_unlock;
1275                 }
1276
1277                 len = INITCONTEXTLEN;
1278                 context = kmalloc(len+1, GFP_NOFS);
1279                 if (!context) {
1280                         rc = -ENOMEM;
1281                         dput(dentry);
1282                         goto out_unlock;
1283                 }
1284                 context[len] = '\0';
1285                 rc = inode->i_op->getxattr(dentry, XATTR_NAME_SELINUX,
1286                                            context, len);
1287                 if (rc == -ERANGE) {
1288                         /* Need a larger buffer.  Query for the right size. */
1289                         rc = inode->i_op->getxattr(dentry, XATTR_NAME_SELINUX,
1290                                                    NULL, 0);
1291                         if (rc < 0) {
1292                                 dput(dentry);
1293                                 goto out_unlock;
1294                         }
1295                         kfree(context);
1296                         len = rc;
1297                         context = kmalloc(len+1, GFP_NOFS);
1298                         if (!context) {
1299                                 rc = -ENOMEM;
1300                                 dput(dentry);
1301                                 goto out_unlock;
1302                         }
1303                         context[len] = '\0';
1304                         rc = inode->i_op->getxattr(dentry,
1305                                                    XATTR_NAME_SELINUX,
1306                                                    context, len);
1307                 }
1308                 dput(dentry);
1309                 if (rc < 0) {
1310                         if (rc != -ENODATA) {
1311                                 printk(KERN_WARNING "SELinux: %s:  getxattr returned "
1312                                        "%d for dev=%s ino=%ld\n", __func__,
1313                                        -rc, inode->i_sb->s_id, inode->i_ino);
1314                                 kfree(context);
1315                                 goto out_unlock;
1316                         }
1317                         /* Map ENODATA to the default file SID */
1318                         sid = sbsec->def_sid;
1319                         rc = 0;
1320                 } else {
1321                         rc = security_context_to_sid_default(context, rc, &sid,
1322                                                              sbsec->def_sid,
1323                                                              GFP_NOFS);
1324                         if (rc) {
1325                                 char *dev = inode->i_sb->s_id;
1326                                 unsigned long ino = inode->i_ino;
1327
1328                                 if (rc == -EINVAL) {
1329                                         if (printk_ratelimit())
1330                                                 printk(KERN_NOTICE "SELinux: inode=%lu on dev=%s was found to have an invalid "
1331                                                         "context=%s.  This indicates you may need to relabel the inode or the "
1332                                                         "filesystem in question.\n", ino, dev, context);
1333                                 } else {
1334                                         printk(KERN_WARNING "SELinux: %s:  context_to_sid(%s) "
1335                                                "returned %d for dev=%s ino=%ld\n",
1336                                                __func__, context, -rc, dev, ino);
1337                                 }
1338                                 kfree(context);
1339                                 /* Leave with the unlabeled SID */
1340                                 rc = 0;
1341                                 break;
1342                         }
1343                 }
1344                 kfree(context);
1345                 isec->sid = sid;
1346                 break;
1347         case SECURITY_FS_USE_TASK:
1348                 isec->sid = isec->task_sid;
1349                 break;
1350         case SECURITY_FS_USE_TRANS:
1351                 /* Default to the fs SID. */
1352                 isec->sid = sbsec->sid;
1353
1354                 /* Try to obtain a transition SID. */
1355                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
1356                 rc = security_transition_sid(isec->task_sid,
1357                                              sbsec->sid,
1358                                              isec->sclass,
1359                                              &sid);
1360                 if (rc)
1361                         goto out_unlock;
1362                 isec->sid = sid;
1363                 break;
1364         case SECURITY_FS_USE_MNTPOINT:
1365                 isec->sid = sbsec->mntpoint_sid;
1366                 break;
1367         default:
1368                 /* Default to the fs superblock SID. */
1369                 isec->sid = sbsec->sid;
1370
1371                 if ((sbsec->flags & SE_SBPROC) && !S_ISLNK(inode->i_mode)) {
1372                         struct proc_inode *proci = PROC_I(inode);
1373                         if (proci->pde) {
1374                                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
1375                                 rc = selinux_proc_get_sid(proci->pde,
1376                                                           isec->sclass,
1377                                                           &sid);
1378                                 if (rc)
1379                                         goto out_unlock;
1380                                 isec->sid = sid;
1381                         }
1382                 }
1383                 break;
1384         }
1385
1386         isec->initialized = 1;
1387
1388 out_unlock:
1389         mutex_unlock(&isec->lock);
1390 out:
1391         if (isec->sclass == SECCLASS_FILE)
1392                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
1393         return rc;
1394 }
1395
1396 /* Convert a Linux signal to an access vector. */
1397 static inline u32 signal_to_av(int sig)
1398 {
1399         u32 perm = 0;
1400
1401         switch (sig) {
1402         case SIGCHLD:
1403                 /* Commonly granted from child to parent. */
1404                 perm = PROCESS__SIGCHLD;
1405                 break;
1406         case SIGKILL:
1407                 /* Cannot be caught or ignored */
1408                 perm = PROCESS__SIGKILL;
1409                 break;
1410         case SIGSTOP:
1411                 /* Cannot be caught or ignored */
1412                 perm = PROCESS__SIGSTOP;
1413                 break;
1414         default:
1415                 /* All other signals. */
1416                 perm = PROCESS__SIGNAL;
1417                 break;
1418         }
1419
1420         return perm;
1421 }
1422
1423 /*
1424  * Check permission between a pair of credentials
1425  * fork check, ptrace check, etc.
1426  */
1427 static int cred_has_perm(const struct cred *actor,
1428                          const struct cred *target,
1429                          u32 perms)
1430 {
1431         u32 asid = cred_sid(actor), tsid = cred_sid(target);
1432
1433         return avc_has_perm(asid, tsid, SECCLASS_PROCESS, perms, NULL);
1434 }
1435
1436 /*
1437  * Check permission between a pair of tasks, e.g. signal checks,
1438  * fork check, ptrace check, etc.
1439  * tsk1 is the actor and tsk2 is the target
1440  * - this uses the default subjective creds of tsk1
1441  */
1442 static int task_has_perm(const struct task_struct *tsk1,
1443                          const struct task_struct *tsk2,
1444                          u32 perms)
1445 {
1446         const struct task_security_struct *__tsec1, *__tsec2;
1447         u32 sid1, sid2;
1448
1449         rcu_read_lock();
1450         __tsec1 = __task_cred(tsk1)->security;  sid1 = __tsec1->sid;
1451         __tsec2 = __task_cred(tsk2)->security;  sid2 = __tsec2->sid;
1452         rcu_read_unlock();
1453         return avc_has_perm(sid1, sid2, SECCLASS_PROCESS, perms, NULL);
1454 }
1455
1456 /*
1457  * Check permission between current and another task, e.g. signal checks,
1458  * fork check, ptrace check, etc.
1459  * current is the actor and tsk2 is the target
1460  * - this uses current's subjective creds
1461  */
1462 static int current_has_perm(const struct task_struct *tsk,
1463                             u32 perms)
1464 {
1465         u32 sid, tsid;
1466
1467         sid = current_sid();
1468         tsid = task_sid(tsk);
1469         return avc_has_perm(sid, tsid, SECCLASS_PROCESS, perms, NULL);
1470 }
1471
1472 #if CAP_LAST_CAP > 63
1473 #error Fix SELinux to handle capabilities > 63.
1474 #endif
1475
1476 /* Check whether a task is allowed to use a capability. */
1477 static int task_has_capability(struct task_struct *tsk,
1478                                const struct cred *cred,
1479                                int cap, int audit)
1480 {
1481         struct avc_audit_data ad;
1482         struct av_decision avd;
1483         u16 sclass;
1484         u32 sid = cred_sid(cred);
1485         u32 av = CAP_TO_MASK(cap);
1486         int rc;
1487
1488         AVC_AUDIT_DATA_INIT(&ad, CAP);
1489         ad.tsk = tsk;
1490         ad.u.cap = cap;
1491
1492         switch (CAP_TO_INDEX(cap)) {
1493         case 0:
1494                 sclass = SECCLASS_CAPABILITY;
1495                 break;
1496         case 1:
1497                 sclass = SECCLASS_CAPABILITY2;
1498                 break;
1499         default:
1500                 printk(KERN_ERR
1501                        "SELinux:  out of range capability %d\n", cap);
1502                 BUG();
1503         }
1504
1505         rc = avc_has_perm_noaudit(sid, sid, sclass, av, 0, &avd);
1506         if (audit == SECURITY_CAP_AUDIT)
1507                 avc_audit(sid, sid, sclass, av, &avd, rc, &ad);
1508         return rc;
1509 }
1510
1511 /* Check whether a task is allowed to use a system operation. */
1512 static int task_has_system(struct task_struct *tsk,
1513                            u32 perms)
1514 {
1515         u32 sid = task_sid(tsk);
1516
1517         return avc_has_perm(sid, SECINITSID_KERNEL,
1518                             SECCLASS_SYSTEM, perms, NULL);
1519 }
1520
1521 /* Check whether a task has a particular permission to an inode.
1522    The 'adp' parameter is optional and allows other audit
1523    data to be passed (e.g. the dentry). */
1524 static int inode_has_perm(const struct cred *cred,
1525                           struct inode *inode,
1526                           u32 perms,
1527                           struct avc_audit_data *adp)
1528 {
1529         struct inode_security_struct *isec;
1530         struct avc_audit_data ad;
1531         u32 sid;
1532
1533         if (unlikely(IS_PRIVATE(inode)))
1534                 return 0;
1535
1536         sid = cred_sid(cred);
1537         isec = inode->i_security;
1538
1539         if (!adp) {
1540                 adp = &ad;
1541                 AVC_AUDIT_DATA_INIT(&ad, FS);
1542                 ad.u.fs.inode = inode;
1543         }
1544
1545         return avc_has_perm(sid, isec->sid, isec->sclass, perms, adp);
1546 }
1547
1548 /* Same as inode_has_perm, but pass explicit audit data containing
1549    the dentry to help the auditing code to more easily generate the
1550    pathname if needed. */
1551 static inline int dentry_has_perm(const struct cred *cred,
1552                                   struct vfsmount *mnt,
1553                                   struct dentry *dentry,
1554                                   u32 av)
1555 {
1556         struct inode *inode = dentry->d_inode;
1557         struct avc_audit_data ad;
1558
1559         AVC_AUDIT_DATA_INIT(&ad, FS);
1560         ad.u.fs.path.mnt = mnt;
1561         ad.u.fs.path.dentry = dentry;
1562         return inode_has_perm(cred, inode, av, &ad);
1563 }
1564
1565 /* Check whether a task can use an open file descriptor to
1566    access an inode in a given way.  Check access to the
1567    descriptor itself, and then use dentry_has_perm to
1568    check a particular permission to the file.
1569    Access to the descriptor is implicitly granted if it
1570    has the same SID as the process.  If av is zero, then
1571    access to the file is not checked, e.g. for cases
1572    where only the descriptor is affected like seek. */
1573 static int file_has_perm(const struct cred *cred,
1574                          struct file *file,
1575                          u32 av)
1576 {
1577         struct file_security_struct *fsec = file->f_security;
1578         struct inode *inode = file->f_path.dentry->d_inode;
1579         struct avc_audit_data ad;
1580         u32 sid = cred_sid(cred);
1581         int rc;
1582
1583         AVC_AUDIT_DATA_INIT(&ad, FS);
1584         ad.u.fs.path = file->f_path;
1585
1586         if (sid != fsec->sid) {
1587                 rc = avc_has_perm(sid, fsec->sid,
1588                                   SECCLASS_FD,
1589                                   FD__USE,
1590                                   &ad);
1591                 if (rc)
1592                         goto out;
1593         }
1594
1595         /* av is zero if only checking access to the descriptor. */
1596         rc = 0;
1597         if (av)
1598                 rc = inode_has_perm(cred, inode, av, &ad);
1599
1600 out:
1601         return rc;
1602 }
1603
1604 /* Check whether a task can create a file. */
1605 static int may_create(struct inode *dir,
1606                       struct dentry *dentry,
1607                       u16 tclass)
1608 {
1609         const struct cred *cred = current_cred();
1610         const struct task_security_struct *tsec = cred->security;
1611         struct inode_security_struct *dsec;
1612         struct superblock_security_struct *sbsec;
1613         u32 sid, newsid;
1614         struct avc_audit_data ad;
1615         int rc;
1616
1617         dsec = dir->i_security;
1618         sbsec = dir->i_sb->s_security;
1619
1620         sid = tsec->sid;
1621         newsid = tsec->create_sid;
1622
1623         AVC_AUDIT_DATA_INIT(&ad, FS);
1624         ad.u.fs.path.dentry = dentry;
1625
1626         rc = avc_has_perm(sid, dsec->sid, SECCLASS_DIR,
1627                           DIR__ADD_NAME | DIR__SEARCH,
1628                           &ad);
1629         if (rc)
1630                 return rc;
1631
1632         if (!newsid || !(sbsec->flags & SE_SBLABELSUPP)) {
1633                 rc = security_transition_sid(sid, dsec->sid, tclass, &newsid);
1634                 if (rc)
1635                         return rc;
1636         }
1637
1638         rc = avc_has_perm(sid, newsid, tclass, FILE__CREATE, &ad);
1639         if (rc)
1640                 return rc;
1641
1642         return avc_has_perm(newsid, sbsec->sid,
1643                             SECCLASS_FILESYSTEM,
1644                             FILESYSTEM__ASSOCIATE, &ad);
1645 }
1646
1647 /* Check whether a task can create a key. */
1648 static int may_create_key(u32 ksid,
1649                           struct task_struct *ctx)
1650 {
1651         u32 sid = task_sid(ctx);
1652
1653         return avc_has_perm(sid, ksid, SECCLASS_KEY, KEY__CREATE, NULL);
1654 }
1655
1656 #define MAY_LINK        0
1657 #define MAY_UNLINK      1
1658 #define MAY_RMDIR       2
1659
1660 /* Check whether a task can link, unlink, or rmdir a file/directory. */
1661 static int may_link(struct inode *dir,
1662                     struct dentry *dentry,
1663                     int kind)
1664
1665 {
1666         struct inode_security_struct *dsec, *isec;
1667         struct avc_audit_data ad;
1668         u32 sid = current_sid();
1669         u32 av;
1670         int rc;
1671
1672         dsec = dir->i_security;
1673         isec = dentry->d_inode->i_security;
1674
1675         AVC_AUDIT_DATA_INIT(&ad, FS);
1676         ad.u.fs.path.dentry = dentry;
1677
1678         av = DIR__SEARCH;
1679         av |= (kind ? DIR__REMOVE_NAME : DIR__ADD_NAME);
1680         rc = avc_has_perm(sid, dsec->sid, SECCLASS_DIR, av, &ad);
1681         if (rc)
1682                 return rc;
1683
1684         switch (kind) {
1685         case MAY_LINK:
1686                 av = FILE__LINK;
1687                 break;
1688         case MAY_UNLINK:
1689                 av = FILE__UNLINK;
1690                 break;
1691         case MAY_RMDIR:
1692                 av = DIR__RMDIR;
1693                 break;
1694         default:
1695                 printk(KERN_WARNING "SELinux: %s:  unrecognized kind %d\n",
1696                         __func__, kind);
1697                 return 0;
1698         }
1699
1700         rc = avc_has_perm(sid, isec->sid, isec->sclass, av, &ad);
1701         return rc;
1702 }
1703
1704 static inline int may_rename(struct inode *old_dir,
1705                              struct dentry *old_dentry,
1706                              struct inode *new_dir,
1707                              struct dentry *new_dentry)
1708 {
1709         struct inode_security_struct *old_dsec, *new_dsec, *old_isec, *new_isec;
1710         struct avc_audit_data ad;
1711         u32 sid = current_sid();
1712         u32 av;
1713         int old_is_dir, new_is_dir;
1714         int rc;
1715
1716         old_dsec = old_dir->i_security;
1717         old_isec = old_dentry->d_inode->i_security;
1718         old_is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
1719         new_dsec = new_dir->i_security;
1720
1721         AVC_AUDIT_DATA_INIT(&ad, FS);
1722
1723         ad.u.fs.path.dentry = old_dentry;
1724         rc = avc_has_perm(sid, old_dsec->sid, SECCLASS_DIR,
1725                           DIR__REMOVE_NAME | DIR__SEARCH, &ad);
1726         if (rc)
1727                 return rc;
1728         rc = avc_has_perm(sid, old_isec->sid,
1729                           old_isec->sclass, FILE__RENAME, &ad);
1730         if (rc)
1731                 return rc;
1732         if (old_is_dir && new_dir != old_dir) {
1733                 rc = avc_has_perm(sid, old_isec->sid,
1734                                   old_isec->sclass, DIR__REPARENT, &ad);
1735                 if (rc)
1736                         return rc;
1737         }
1738
1739         ad.u.fs.path.dentry = new_dentry;
1740         av = DIR__ADD_NAME | DIR__SEARCH;
1741         if (new_dentry->d_inode)
1742                 av |= DIR__REMOVE_NAME;
1743         rc = avc_has_perm(sid, new_dsec->sid, SECCLASS_DIR, av, &ad);
1744         if (rc)
1745                 return rc;
1746         if (new_dentry->d_inode) {
1747                 new_isec = new_dentry->d_inode->i_security;
1748                 new_is_dir = S_ISDIR(new_dentry->d_inode->i_mode);
1749                 rc = avc_has_perm(sid, new_isec->sid,
1750                                   new_isec->sclass,
1751                                   (new_is_dir ? DIR__RMDIR : FILE__UNLINK), &ad);
1752                 if (rc)
1753                         return rc;
1754         }
1755
1756         return 0;
1757 }
1758
1759 /* Check whether a task can perform a filesystem operation. */
1760 static int superblock_has_perm(const struct cred *cred,
1761                                struct super_block *sb,
1762                                u32 perms,
1763                                struct avc_audit_data *ad)
1764 {
1765         struct superblock_security_struct *sbsec;
1766         u32 sid = cred_sid(cred);
1767
1768         sbsec = sb->s_security;
1769         return avc_has_perm(sid, sbsec->sid, SECCLASS_FILESYSTEM, perms, ad);
1770 }
1771
1772 /* Convert a Linux mode and permission mask to an access vector. */
1773 static inline u32 file_mask_to_av(int mode, int mask)
1774 {
1775         u32 av = 0;
1776
1777         if ((mode & S_IFMT) != S_IFDIR) {
1778                 if (mask & MAY_EXEC)
1779                         av |= FILE__EXECUTE;
1780                 if (mask & MAY_READ)
1781                         av |= FILE__READ;
1782
1783                 if (mask & MAY_APPEND)
1784                         av |= FILE__APPEND;
1785                 else if (mask & MAY_WRITE)
1786                         av |= FILE__WRITE;
1787
1788         } else {
1789                 if (mask & MAY_EXEC)
1790                         av |= DIR__SEARCH;
1791                 if (mask & MAY_WRITE)
1792                         av |= DIR__WRITE;
1793                 if (mask & MAY_READ)
1794                         av |= DIR__READ;
1795         }
1796
1797         return av;
1798 }
1799
1800 /* Convert a Linux file to an access vector. */
1801 static inline u32 file_to_av(struct file *file)
1802 {
1803         u32 av = 0;
1804
1805         if (file->f_mode & FMODE_READ)
1806                 av |= FILE__READ;
1807         if (file->f_mode & FMODE_WRITE) {
1808                 if (file->f_flags & O_APPEND)
1809                         av |= FILE__APPEND;
1810                 else
1811                         av |= FILE__WRITE;
1812         }
1813         if (!av) {
1814                 /*
1815                  * Special file opened with flags 3 for ioctl-only use.
1816                  */
1817                 av = FILE__IOCTL;
1818         }
1819
1820         return av;
1821 }
1822
1823 /*
1824  * Convert a file to an access vector and include the correct open
1825  * open permission.
1826  */
1827 static inline u32 open_file_to_av(struct file *file)
1828 {
1829         u32 av = file_to_av(file);
1830
1831         if (selinux_policycap_openperm) {
1832                 mode_t mode = file->f_path.dentry->d_inode->i_mode;
1833                 /*
1834                  * lnk files and socks do not really have an 'open'
1835                  */
1836                 if (S_ISREG(mode))
1837                         av |= FILE__OPEN;
1838                 else if (S_ISCHR(mode))
1839                         av |= CHR_FILE__OPEN;
1840                 else if (S_ISBLK(mode))
1841                         av |= BLK_FILE__OPEN;
1842                 else if (S_ISFIFO(mode))
1843                         av |= FIFO_FILE__OPEN;
1844                 else if (S_ISDIR(mode))
1845                         av |= DIR__OPEN;
1846                 else if (S_ISSOCK(mode))
1847                         av |= SOCK_FILE__OPEN;
1848                 else
1849                         printk(KERN_ERR "SELinux: WARNING: inside %s with "
1850                                 "unknown mode:%o\n", __func__, mode);
1851         }
1852         return av;
1853 }
1854
1855 /* Hook functions begin here. */
1856
1857 static int selinux_ptrace_may_access(struct task_struct *child,
1858                                      unsigned int mode)
1859 {
1860         int rc;
1861
1862         rc = cap_ptrace_may_access(child, mode);
1863         if (rc)
1864                 return rc;
1865
1866         if (mode == PTRACE_MODE_READ) {
1867                 u32 sid = current_sid();
1868                 u32 csid = task_sid(child);
1869                 return avc_has_perm(sid, csid, SECCLASS_FILE, FILE__READ, NULL);
1870         }
1871
1872         return current_has_perm(child, PROCESS__PTRACE);
1873 }
1874
1875 static int selinux_ptrace_traceme(struct task_struct *parent)
1876 {
1877         int rc;
1878
1879         rc = cap_ptrace_traceme(parent);
1880         if (rc)
1881                 return rc;
1882
1883         return task_has_perm(parent, current, PROCESS__PTRACE);
1884 }
1885
1886 static int selinux_capget(struct task_struct *target, kernel_cap_t *effective,
1887                           kernel_cap_t *inheritable, kernel_cap_t *permitted)
1888 {
1889         int error;
1890
1891         error = current_has_perm(target, PROCESS__GETCAP);
1892         if (error)
1893                 return error;
1894
1895         return cap_capget(target, effective, inheritable, permitted);
1896 }
1897
1898 static int selinux_capset(struct cred *new, const struct cred *old,
1899                           const kernel_cap_t *effective,
1900                           const kernel_cap_t *inheritable,
1901                           const kernel_cap_t *permitted)
1902 {
1903         int error;
1904
1905         error = cap_capset(new, old,
1906                                       effective, inheritable, permitted);
1907         if (error)
1908                 return error;
1909
1910         return cred_has_perm(old, new, PROCESS__SETCAP);
1911 }
1912
1913 /*
1914  * (This comment used to live with the selinux_task_setuid hook,
1915  * which was removed).
1916  *
1917  * Since setuid only affects the current process, and since the SELinux
1918  * controls are not based on the Linux identity attributes, SELinux does not
1919  * need to control this operation.  However, SELinux does control the use of
1920  * the CAP_SETUID and CAP_SETGID capabilities using the capable hook.
1921  */
1922
1923 static int selinux_capable(struct task_struct *tsk, const struct cred *cred,
1924                            int cap, int audit)
1925 {
1926         int rc;
1927
1928         rc = cap_capable(tsk, cred, cap, audit);
1929         if (rc)
1930                 return rc;
1931
1932         return task_has_capability(tsk, cred, cap, audit);
1933 }
1934
1935 static int selinux_sysctl_get_sid(ctl_table *table, u16 tclass, u32 *sid)
1936 {
1937         int buflen, rc;
1938         char *buffer, *path, *end;
1939
1940         rc = -ENOMEM;
1941         buffer = (char *)__get_free_page(GFP_KERNEL);
1942         if (!buffer)
1943                 goto out;
1944
1945         buflen = PAGE_SIZE;
1946         end = buffer+buflen;
1947         *--end = '\0';
1948         buflen--;
1949         path = end-1;
1950         *path = '/';
1951         while (table) {
1952                 const char *name = table->procname;
1953                 size_t namelen = strlen(name);
1954                 buflen -= namelen + 1;
1955                 if (buflen < 0)
1956                         goto out_free;
1957                 end -= namelen;
1958                 memcpy(end, name, namelen);
1959                 *--end = '/';
1960                 path = end;
1961                 table = table->parent;
1962         }
1963         buflen -= 4;
1964         if (buflen < 0)
1965                 goto out_free;
1966         end -= 4;
1967         memcpy(end, "/sys", 4);
1968         path = end;
1969         rc = security_genfs_sid("proc", path, tclass, sid);
1970 out_free:
1971         free_page((unsigned long)buffer);
1972 out:
1973         return rc;
1974 }
1975
1976 static int selinux_sysctl(ctl_table *table, int op)
1977 {
1978         int error = 0;
1979         u32 av;
1980         u32 tsid, sid;
1981         int rc;
1982
1983         sid = current_sid();
1984
1985         rc = selinux_sysctl_get_sid(table, (op == 0001) ?
1986                                     SECCLASS_DIR : SECCLASS_FILE, &tsid);
1987         if (rc) {
1988                 /* Default to the well-defined sysctl SID. */
1989                 tsid = SECINITSID_SYSCTL;
1990         }
1991
1992         /* The op values are "defined" in sysctl.c, thereby creating
1993          * a bad coupling between this module and sysctl.c */
1994         if (op == 001) {
1995                 error = avc_has_perm(sid, tsid,
1996                                      SECCLASS_DIR, DIR__SEARCH, NULL);
1997         } else {
1998                 av = 0;
1999                 if (op & 004)
2000                         av |= FILE__READ;
2001                 if (op & 002)
2002                         av |= FILE__WRITE;
2003                 if (av)
2004                         error = avc_has_perm(sid, tsid,
2005                                              SECCLASS_FILE, av, NULL);
2006         }
2007
2008         return error;
2009 }
2010
2011 static int selinux_quotactl(int cmds, int type, int id, struct super_block *sb)
2012 {
2013         const struct cred *cred = current_cred();
2014         int rc = 0;
2015
2016         if (!sb)
2017                 return 0;
2018
2019         switch (cmds) {
2020         case Q_SYNC:
2021         case Q_QUOTAON:
2022         case Q_QUOTAOFF:
2023         case Q_SETINFO:
2024         case Q_SETQUOTA:
2025                 rc = superblock_has_perm(cred, sb, FILESYSTEM__QUOTAMOD, NULL);
2026                 break;
2027         case Q_GETFMT:
2028         case Q_GETINFO:
2029         case Q_GETQUOTA:
2030                 rc = superblock_has_perm(cred, sb, FILESYSTEM__QUOTAGET, NULL);
2031                 break;
2032         default:
2033                 rc = 0;  /* let the kernel handle invalid cmds */
2034                 break;
2035         }
2036         return rc;
2037 }
2038
2039 static int selinux_quota_on(struct dentry *dentry)
2040 {
2041         const struct cred *cred = current_cred();
2042
2043         return dentry_has_perm(cred, NULL, dentry, FILE__QUOTAON);
2044 }
2045
2046 static int selinux_syslog(int type)
2047 {
2048         int rc;
2049
2050         rc = cap_syslog(type);
2051         if (rc)
2052                 return rc;
2053
2054         switch (type) {
2055         case 3:         /* Read last kernel messages */
2056         case 10:        /* Return size of the log buffer */
2057                 rc = task_has_system(current, SYSTEM__SYSLOG_READ);
2058                 break;
2059         case 6:         /* Disable logging to console */
2060         case 7:         /* Enable logging to console */
2061         case 8:         /* Set level of messages printed to console */
2062                 rc = task_has_system(current, SYSTEM__SYSLOG_CONSOLE);
2063                 break;
2064         case 0:         /* Close log */
2065         case 1:         /* Open log */
2066         case 2:         /* Read from log */
2067         case 4:         /* Read/clear last kernel messages */
2068         case 5:         /* Clear ring buffer */
2069         default:
2070                 rc = task_has_system(current, SYSTEM__SYSLOG_MOD);
2071                 break;
2072         }
2073         return rc;
2074 }
2075
2076 /*
2077  * Check that a process has enough memory to allocate a new virtual
2078  * mapping. 0 means there is enough memory for the allocation to
2079  * succeed and -ENOMEM implies there is not.
2080  *
2081  * Do not audit the selinux permission check, as this is applied to all
2082  * processes that allocate mappings.
2083  */
2084 static int selinux_vm_enough_memory(struct mm_struct *mm, long pages)
2085 {
2086         int rc, cap_sys_admin = 0;
2087
2088         rc = selinux_capable(current, current_cred(), CAP_SYS_ADMIN,
2089                              SECURITY_CAP_NOAUDIT);
2090         if (rc == 0)
2091                 cap_sys_admin = 1;
2092
2093         return __vm_enough_memory(mm, pages, cap_sys_admin);
2094 }
2095
2096 /* binprm security operations */
2097
2098 static int selinux_bprm_set_creds(struct linux_binprm *bprm)
2099 {
2100         const struct task_security_struct *old_tsec;
2101         struct task_security_struct *new_tsec;
2102         struct inode_security_struct *isec;
2103         struct avc_audit_data ad;
2104         struct inode *inode = bprm->file->f_path.dentry->d_inode;
2105         int rc;
2106
2107         rc = cap_bprm_set_creds(bprm);
2108         if (rc)
2109                 return rc;
2110
2111         /* SELinux context only depends on initial program or script and not
2112          * the script interpreter */
2113         if (bprm->cred_prepared)
2114                 return 0;
2115
2116         old_tsec = current_security();
2117         new_tsec = bprm->cred->security;
2118         isec = inode->i_security;
2119
2120         /* Default to the current task SID. */
2121         new_tsec->sid = old_tsec->sid;
2122         new_tsec->osid = old_tsec->sid;
2123
2124         /* Reset fs, key, and sock SIDs on execve. */
2125         new_tsec->create_sid = 0;
2126         new_tsec->keycreate_sid = 0;
2127         new_tsec->sockcreate_sid = 0;
2128
2129         if (old_tsec->exec_sid) {
2130                 new_tsec->sid = old_tsec->exec_sid;
2131                 /* Reset exec SID on execve. */
2132                 new_tsec->exec_sid = 0;
2133         } else {
2134                 /* Check for a default transition on this program. */
2135                 rc = security_transition_sid(old_tsec->sid, isec->sid,
2136                                              SECCLASS_PROCESS, &new_tsec->sid);
2137                 if (rc)
2138                         return rc;
2139         }
2140
2141         AVC_AUDIT_DATA_INIT(&ad, FS);
2142         ad.u.fs.path = bprm->file->f_path;
2143
2144         if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
2145                 new_tsec->sid = old_tsec->sid;
2146
2147         if (new_tsec->sid == old_tsec->sid) {
2148                 rc = avc_has_perm(old_tsec->sid, isec->sid,
2149                                   SECCLASS_FILE, FILE__EXECUTE_NO_TRANS, &ad);
2150                 if (rc)
2151                         return rc;
2152         } else {
2153                 /* Check permissions for the transition. */
2154                 rc = avc_has_perm(old_tsec->sid, new_tsec->sid,
2155                                   SECCLASS_PROCESS, PROCESS__TRANSITION, &ad);
2156                 if (rc)
2157                         return rc;
2158
2159                 rc = avc_has_perm(new_tsec->sid, isec->sid,
2160                                   SECCLASS_FILE, FILE__ENTRYPOINT, &ad);
2161                 if (rc)
2162                         return rc;
2163
2164                 /* Check for shared state */
2165                 if (bprm->unsafe & LSM_UNSAFE_SHARE) {
2166                         rc = avc_has_perm(old_tsec->sid, new_tsec->sid,
2167                                           SECCLASS_PROCESS, PROCESS__SHARE,
2168                                           NULL);
2169                         if (rc)
2170                                 return -EPERM;
2171                 }
2172
2173                 /* Make sure that anyone attempting to ptrace over a task that
2174                  * changes its SID has the appropriate permit */
2175                 if (bprm->unsafe &
2176                     (LSM_UNSAFE_PTRACE | LSM_UNSAFE_PTRACE_CAP)) {
2177                         struct task_struct *tracer;
2178                         struct task_security_struct *sec;
2179                         u32 ptsid = 0;
2180
2181                         rcu_read_lock();
2182                         tracer = tracehook_tracer_task(current);
2183                         if (likely(tracer != NULL)) {
2184                                 sec = __task_cred(tracer)->security;
2185                                 ptsid = sec->sid;
2186                         }
2187                         rcu_read_unlock();
2188
2189                         if (ptsid != 0) {
2190                                 rc = avc_has_perm(ptsid, new_tsec->sid,
2191                                                   SECCLASS_PROCESS,
2192                                                   PROCESS__PTRACE, NULL);
2193                                 if (rc)
2194                                         return -EPERM;
2195                         }
2196                 }
2197
2198                 /* Clear any possibly unsafe personality bits on exec: */
2199                 bprm->per_clear |= PER_CLEAR_ON_SETID;
2200         }
2201
2202         return 0;
2203 }
2204
2205 static int selinux_bprm_secureexec(struct linux_binprm *bprm)
2206 {
2207         const struct cred *cred = current_cred();
2208         const struct task_security_struct *tsec = cred->security;
2209         u32 sid, osid;
2210         int atsecure = 0;
2211
2212         sid = tsec->sid;
2213         osid = tsec->osid;
2214
2215         if (osid != sid) {
2216                 /* Enable secure mode for SIDs transitions unless
2217                    the noatsecure permission is granted between
2218                    the two SIDs, i.e. ahp returns 0. */
2219                 atsecure = avc_has_perm(osid, sid,
2220                                         SECCLASS_PROCESS,
2221                                         PROCESS__NOATSECURE, NULL);
2222         }
2223
2224         return (atsecure || cap_bprm_secureexec(bprm));
2225 }
2226
2227 extern struct vfsmount *selinuxfs_mount;
2228 extern struct dentry *selinux_null;
2229
2230 /* Derived from fs/exec.c:flush_old_files. */
2231 static inline void flush_unauthorized_files(const struct cred *cred,
2232                                             struct files_struct *files)
2233 {
2234         struct avc_audit_data ad;
2235         struct file *file, *devnull = NULL;
2236         struct tty_struct *tty;
2237         struct fdtable *fdt;
2238         long j = -1;
2239         int drop_tty = 0;
2240
2241         tty = get_current_tty();
2242         if (tty) {
2243                 file_list_lock();
2244                 if (!list_empty(&tty->tty_files)) {
2245                         struct inode *inode;
2246
2247                         /* Revalidate access to controlling tty.
2248                            Use inode_has_perm on the tty inode directly rather
2249                            than using file_has_perm, as this particular open
2250                            file may belong to another process and we are only
2251                            interested in the inode-based check here. */
2252                         file = list_first_entry(&tty->tty_files, struct file, f_u.fu_list);
2253                         inode = file->f_path.dentry->d_inode;
2254                         if (inode_has_perm(cred, inode,
2255                                            FILE__READ | FILE__WRITE, NULL)) {
2256                                 drop_tty = 1;
2257                         }
2258                 }
2259                 file_list_unlock();
2260                 tty_kref_put(tty);
2261         }
2262         /* Reset controlling tty. */
2263         if (drop_tty)
2264                 no_tty();
2265
2266         /* Revalidate access to inherited open files. */
2267
2268         AVC_AUDIT_DATA_INIT(&ad, FS);
2269
2270         spin_lock(&files->file_lock);
2271         for (;;) {
2272                 unsigned long set, i;
2273                 int fd;
2274
2275                 j++;
2276                 i = j * __NFDBITS;
2277                 fdt = files_fdtable(files);
2278                 if (i >= fdt->max_fds)
2279                         break;
2280                 set = fdt->open_fds->fds_bits[j];
2281                 if (!set)
2282                         continue;
2283                 spin_unlock(&files->file_lock);
2284                 for ( ; set ; i++, set >>= 1) {
2285                         if (set & 1) {
2286                                 file = fget(i);
2287                                 if (!file)
2288                                         continue;
2289                                 if (file_has_perm(cred,
2290                                                   file,
2291                                                   file_to_av(file))) {
2292                                         sys_close(i);
2293                                         fd = get_unused_fd();
2294                                         if (fd != i) {
2295                                                 if (fd >= 0)
2296                                                         put_unused_fd(fd);
2297                                                 fput(file);
2298                                                 continue;
2299                                         }
2300                                         if (devnull) {
2301                                                 get_file(devnull);
2302                                         } else {
2303                                                 devnull = dentry_open(
2304                                                         dget(selinux_null),
2305                                                         mntget(selinuxfs_mount),
2306                                                         O_RDWR, cred);
2307                                                 if (IS_ERR(devnull)) {
2308                                                         devnull = NULL;
2309                                                         put_unused_fd(fd);
2310                                                         fput(file);
2311                                                         continue;
2312                                                 }
2313                                         }
2314                                         fd_install(fd, devnull);
2315                                 }
2316                                 fput(file);
2317                         }
2318                 }
2319                 spin_lock(&files->file_lock);
2320
2321         }
2322         spin_unlock(&files->file_lock);
2323 }
2324
2325 /*
2326  * Prepare a process for imminent new credential changes due to exec
2327  */
2328 static void selinux_bprm_committing_creds(struct linux_binprm *bprm)
2329 {
2330         struct task_security_struct *new_tsec;
2331         struct rlimit *rlim, *initrlim;
2332         int rc, i;
2333
2334         new_tsec = bprm->cred->security;
2335         if (new_tsec->sid == new_tsec->osid)
2336                 return;
2337
2338         /* Close files for which the new task SID is not authorized. */
2339         flush_unauthorized_files(bprm->cred, current->files);
2340
2341         /* Always clear parent death signal on SID transitions. */
2342         current->pdeath_signal = 0;
2343
2344         /* Check whether the new SID can inherit resource limits from the old
2345          * SID.  If not, reset all soft limits to the lower of the current
2346          * task's hard limit and the init task's soft limit.
2347          *
2348          * Note that the setting of hard limits (even to lower them) can be
2349          * controlled by the setrlimit check.  The inclusion of the init task's
2350          * soft limit into the computation is to avoid resetting soft limits
2351          * higher than the default soft limit for cases where the default is
2352          * lower than the hard limit, e.g. RLIMIT_CORE or RLIMIT_STACK.
2353          */
2354         rc = avc_has_perm(new_tsec->osid, new_tsec->sid, SECCLASS_PROCESS,
2355                           PROCESS__RLIMITINH, NULL);
2356         if (rc) {
2357                 for (i = 0; i < RLIM_NLIMITS; i++) {
2358                         rlim = current->signal->rlim + i;
2359                         initrlim = init_task.signal->rlim + i;
2360                         rlim->rlim_cur = min(rlim->rlim_max, initrlim->rlim_cur);
2361                 }
2362                 update_rlimit_cpu(rlim->rlim_cur);
2363         }
2364 }
2365
2366 /*
2367  * Clean up the process immediately after the installation of new credentials
2368  * due to exec
2369  */
2370 static void selinux_bprm_committed_creds(struct linux_binprm *bprm)
2371 {
2372         const struct task_security_struct *tsec = current_security();
2373         struct itimerval itimer;
2374         u32 osid, sid;
2375         int rc, i;
2376
2377         osid = tsec->osid;
2378         sid = tsec->sid;
2379
2380         if (sid == osid)
2381                 return;
2382
2383         /* Check whether the new SID can inherit signal state from the old SID.
2384          * If not, clear itimers to avoid subsequent signal generation and
2385          * flush and unblock signals.
2386          *
2387          * This must occur _after_ the task SID has been updated so that any
2388          * kill done after the flush will be checked against the new SID.
2389          */
2390         rc = avc_has_perm(osid, sid, SECCLASS_PROCESS, PROCESS__SIGINH, NULL);
2391         if (rc) {
2392                 memset(&itimer, 0, sizeof itimer);
2393                 for (i = 0; i < 3; i++)
2394                         do_setitimer(i, &itimer, NULL);
2395                 spin_lock_irq(&current->sighand->siglock);
2396                 if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
2397                         __flush_signals(current);
2398                         flush_signal_handlers(current, 1);
2399                         sigemptyset(&current->blocked);
2400                 }
2401                 spin_unlock_irq(&current->sighand->siglock);
2402         }
2403
2404         /* Wake up the parent if it is waiting so that it can recheck
2405          * wait permission to the new task SID. */
2406         read_lock(&tasklist_lock);
2407         wake_up_interruptible(&current->real_parent->signal->wait_chldexit);
2408         read_unlock(&tasklist_lock);
2409 }
2410
2411 /* superblock security operations */
2412
2413 static int selinux_sb_alloc_security(struct super_block *sb)
2414 {
2415         return superblock_alloc_security(sb);
2416 }
2417
2418 static void selinux_sb_free_security(struct super_block *sb)
2419 {
2420         superblock_free_security(sb);
2421 }
2422
2423 static inline int match_prefix(char *prefix, int plen, char *option, int olen)
2424 {
2425         if (plen > olen)
2426                 return 0;
2427
2428         return !memcmp(prefix, option, plen);
2429 }
2430
2431 static inline int selinux_option(char *option, int len)
2432 {
2433         return (match_prefix(CONTEXT_STR, sizeof(CONTEXT_STR)-1, option, len) ||
2434                 match_prefix(FSCONTEXT_STR, sizeof(FSCONTEXT_STR)-1, option, len) ||
2435                 match_prefix(DEFCONTEXT_STR, sizeof(DEFCONTEXT_STR)-1, option, len) ||
2436                 match_prefix(ROOTCONTEXT_STR, sizeof(ROOTCONTEXT_STR)-1, option, len) ||
2437                 match_prefix(LABELSUPP_STR, sizeof(LABELSUPP_STR)-1, option, len));
2438 }
2439
2440 static inline void take_option(char **to, char *from, int *first, int len)
2441 {
2442         if (!*first) {
2443                 **to = ',';
2444                 *to += 1;
2445         } else
2446                 *first = 0;
2447         memcpy(*to, from, len);
2448         *to += len;
2449 }
2450
2451 static inline void take_selinux_option(char **to, char *from, int *first,
2452                                        int len)
2453 {
2454         int current_size = 0;
2455
2456         if (!*first) {
2457                 **to = '|';
2458                 *to += 1;
2459         } else
2460                 *first = 0;
2461
2462         while (current_size < len) {
2463                 if (*from != '"') {
2464                         **to = *from;
2465                         *to += 1;
2466                 }
2467                 from += 1;
2468                 current_size += 1;
2469         }
2470 }
2471
2472 static int selinux_sb_copy_data(char *orig, char *copy)
2473 {
2474         int fnosec, fsec, rc = 0;
2475         char *in_save, *in_curr, *in_end;
2476         char *sec_curr, *nosec_save, *nosec;
2477         int open_quote = 0;
2478
2479         in_curr = orig;
2480         sec_curr = copy;
2481
2482         nosec = (char *)get_zeroed_page(GFP_KERNEL);
2483         if (!nosec) {
2484                 rc = -ENOMEM;
2485                 goto out;
2486         }
2487
2488         nosec_save = nosec;
2489         fnosec = fsec = 1;
2490         in_save = in_end = orig;
2491
2492         do {
2493                 if (*in_end == '"')
2494                         open_quote = !open_quote;
2495                 if ((*in_end == ',' && open_quote == 0) ||
2496                                 *in_end == '\0') {
2497                         int len = in_end - in_curr;
2498
2499                         if (selinux_option(in_curr, len))
2500                                 take_selinux_option(&sec_curr, in_curr, &fsec, len);
2501                         else
2502                                 take_option(&nosec, in_curr, &fnosec, len);
2503
2504                         in_curr = in_end + 1;
2505                 }
2506         } while (*in_end++);
2507
2508         strcpy(in_save, nosec_save);
2509         free_page((unsigned long)nosec_save);
2510 out:
2511         return rc;
2512 }
2513
2514 static int selinux_sb_kern_mount(struct super_block *sb, int flags, void *data)
2515 {
2516         const struct cred *cred = current_cred();
2517         struct avc_audit_data ad;
2518         int rc;
2519
2520         rc = superblock_doinit(sb, data);
2521         if (rc)
2522                 return rc;
2523
2524         /* Allow all mounts performed by the kernel */
2525         if (flags & MS_KERNMOUNT)
2526                 return 0;
2527
2528         AVC_AUDIT_DATA_INIT(&ad, FS);
2529         ad.u.fs.path.dentry = sb->s_root;
2530         return superblock_has_perm(cred, sb, FILESYSTEM__MOUNT, &ad);
2531 }
2532
2533 static int selinux_sb_statfs(struct dentry *dentry)
2534 {
2535         const struct cred *cred = current_cred();
2536         struct avc_audit_data ad;
2537
2538         AVC_AUDIT_DATA_INIT(&ad, FS);
2539         ad.u.fs.path.dentry = dentry->d_sb->s_root;
2540         return superblock_has_perm(cred, dentry->d_sb, FILESYSTEM__GETATTR, &ad);
2541 }
2542
2543 static int selinux_mount(char *dev_name,
2544                          struct path *path,
2545                          char *type,
2546                          unsigned long flags,
2547                          void *data)
2548 {
2549         const struct cred *cred = current_cred();
2550
2551         if (flags & MS_REMOUNT)
2552                 return superblock_has_perm(cred, path->mnt->mnt_sb,
2553                                            FILESYSTEM__REMOUNT, NULL);
2554         else
2555                 return dentry_has_perm(cred, path->mnt, path->dentry,
2556                                        FILE__MOUNTON);
2557 }
2558
2559 static int selinux_umount(struct vfsmount *mnt, int flags)
2560 {
2561         const struct cred *cred = current_cred();
2562
2563         return superblock_has_perm(cred, mnt->mnt_sb,
2564                                    FILESYSTEM__UNMOUNT, NULL);
2565 }
2566
2567 /* inode security operations */
2568
2569 static int selinux_inode_alloc_security(struct inode *inode)
2570 {
2571         return inode_alloc_security(inode);
2572 }
2573
2574 static void selinux_inode_free_security(struct inode *inode)
2575 {
2576         inode_free_security(inode);
2577 }
2578
2579 static int selinux_inode_init_security(struct inode *inode, struct inode *dir,
2580                                        char **name, void **value,
2581                                        size_t *len)
2582 {
2583         const struct cred *cred = current_cred();
2584         const struct task_security_struct *tsec = cred->security;
2585         struct inode_security_struct *dsec;
2586         struct superblock_security_struct *sbsec;
2587         u32 sid, newsid, clen;
2588         int rc;
2589         char *namep = NULL, *context;
2590
2591         dsec = dir->i_security;
2592         sbsec = dir->i_sb->s_security;
2593
2594         sid = tsec->sid;
2595         newsid = tsec->create_sid;
2596
2597         if (!newsid || !(sbsec->flags & SE_SBLABELSUPP)) {
2598                 rc = security_transition_sid(sid, dsec->sid,
2599                                              inode_mode_to_security_class(inode->i_mode),
2600                                              &newsid);
2601                 if (rc) {
2602                         printk(KERN_WARNING "%s:  "
2603                                "security_transition_sid failed, rc=%d (dev=%s "
2604                                "ino=%ld)\n",
2605                                __func__,
2606                                -rc, inode->i_sb->s_id, inode->i_ino);
2607                         return rc;
2608                 }
2609         }
2610
2611         /* Possibly defer initialization to selinux_complete_init. */
2612         if (sbsec->flags & SE_SBINITIALIZED) {
2613                 struct inode_security_struct *isec = inode->i_security;
2614                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
2615                 isec->sid = newsid;
2616                 isec->initialized = 1;
2617         }
2618
2619         if (!ss_initialized || !(sbsec->flags & SE_SBLABELSUPP))
2620                 return -EOPNOTSUPP;
2621
2622         if (name) {
2623                 namep = kstrdup(XATTR_SELINUX_SUFFIX, GFP_NOFS);
2624                 if (!namep)
2625                         return -ENOMEM;
2626                 *name = namep;
2627         }
2628
2629         if (value && len) {
2630                 rc = security_sid_to_context_force(newsid, &context, &clen);
2631                 if (rc) {
2632                         kfree(namep);
2633                         return rc;
2634                 }
2635                 *value = context;
2636                 *len = clen;
2637         }
2638
2639         return 0;
2640 }
2641
2642 static int selinux_inode_create(struct inode *dir, struct dentry *dentry, int mask)
2643 {
2644         return may_create(dir, dentry, SECCLASS_FILE);
2645 }
2646
2647 static int selinux_inode_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2648 {
2649         return may_link(dir, old_dentry, MAY_LINK);
2650 }
2651
2652 static int selinux_inode_unlink(struct inode *dir, struct dentry *dentry)
2653 {
2654         return may_link(dir, dentry, MAY_UNLINK);
2655 }
2656
2657 static int selinux_inode_symlink(struct inode *dir, struct dentry *dentry, const char *name)
2658 {
2659         return may_create(dir, dentry, SECCLASS_LNK_FILE);
2660 }
2661
2662 static int selinux_inode_mkdir(struct inode *dir, struct dentry *dentry, int mask)
2663 {
2664         return may_create(dir, dentry, SECCLASS_DIR);
2665 }
2666
2667 static int selinux_inode_rmdir(struct inode *dir, struct dentry *dentry)
2668 {
2669         return may_link(dir, dentry, MAY_RMDIR);
2670 }
2671
2672 static int selinux_inode_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2673 {
2674         return may_create(dir, dentry, inode_mode_to_security_class(mode));
2675 }
2676
2677 static int selinux_inode_rename(struct inode *old_inode, struct dentry *old_dentry,
2678                                 struct inode *new_inode, struct dentry *new_dentry)
2679 {
2680         return may_rename(old_inode, old_dentry, new_inode, new_dentry);
2681 }
2682
2683 static int selinux_inode_readlink(struct dentry *dentry)
2684 {
2685         const struct cred *cred = current_cred();
2686
2687         return dentry_has_perm(cred, NULL, dentry, FILE__READ);
2688 }
2689
2690 static int selinux_inode_follow_link(struct dentry *dentry, struct nameidata *nameidata)
2691 {
2692         const struct cred *cred = current_cred();
2693
2694         return dentry_has_perm(cred, NULL, dentry, FILE__READ);
2695 }
2696
2697 static int selinux_inode_permission(struct inode *inode, int mask)
2698 {
2699         const struct cred *cred = current_cred();
2700
2701         if (!mask) {
2702                 /* No permission to check.  Existence test. */
2703                 return 0;
2704         }
2705
2706         return inode_has_perm(cred, inode,
2707                               file_mask_to_av(inode->i_mode, mask), NULL);
2708 }
2709
2710 static int selinux_inode_setattr(struct dentry *dentry, struct iattr *iattr)
2711 {
2712         const struct cred *cred = current_cred();
2713
2714         if (iattr->ia_valid & ATTR_FORCE)
2715                 return 0;
2716
2717         if (iattr->ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID |
2718                                ATTR_ATIME_SET | ATTR_MTIME_SET))
2719                 return dentry_has_perm(cred, NULL, dentry, FILE__SETATTR);
2720
2721         return dentry_has_perm(cred, NULL, dentry, FILE__WRITE);
2722 }
2723
2724 static int selinux_inode_getattr(struct vfsmount *mnt, struct dentry *dentry)
2725 {
2726         const struct cred *cred = current_cred();
2727
2728         return dentry_has_perm(cred, mnt, dentry, FILE__GETATTR);
2729 }
2730
2731 static int selinux_inode_setotherxattr(struct dentry *dentry, const char *name)
2732 {
2733         const struct cred *cred = current_cred();
2734
2735         if (!strncmp(name, XATTR_SECURITY_PREFIX,
2736                      sizeof XATTR_SECURITY_PREFIX - 1)) {
2737                 if (!strcmp(name, XATTR_NAME_CAPS)) {
2738                         if (!capable(CAP_SETFCAP))
2739                                 return -EPERM;
2740                 } else if (!capable(CAP_SYS_ADMIN)) {
2741                         /* A different attribute in the security namespace.
2742                            Restrict to administrator. */
2743                         return -EPERM;
2744                 }
2745         }
2746
2747         /* Not an attribute we recognize, so just check the
2748            ordinary setattr permission. */
2749         return dentry_has_perm(cred, NULL, dentry, FILE__SETATTR);
2750 }
2751
2752 static int selinux_inode_setxattr(struct dentry *dentry, const char *name,
2753                                   const void *value, size_t size, int flags)
2754 {
2755         struct inode *inode = dentry->d_inode;
2756         struct inode_security_struct *isec = inode->i_security;
2757         struct superblock_security_struct *sbsec;
2758         struct avc_audit_data ad;
2759         u32 newsid, sid = current_sid();
2760         int rc = 0;
2761
2762         if (strcmp(name, XATTR_NAME_SELINUX))
2763                 return selinux_inode_setotherxattr(dentry, name);
2764
2765         sbsec = inode->i_sb->s_security;
2766         if (!(sbsec->flags & SE_SBLABELSUPP))
2767                 return -EOPNOTSUPP;
2768
2769         if (!is_owner_or_cap(inode))
2770                 return -EPERM;
2771
2772         AVC_AUDIT_DATA_INIT(&ad, FS);
2773         ad.u.fs.path.dentry = dentry;
2774
2775         rc = avc_has_perm(sid, isec->sid, isec->sclass,
2776                           FILE__RELABELFROM, &ad);
2777         if (rc)
2778                 return rc;
2779
2780         rc = security_context_to_sid(value, size, &newsid);
2781         if (rc == -EINVAL) {
2782                 if (!capable(CAP_MAC_ADMIN))
2783                         return rc;
2784                 rc = security_context_to_sid_force(value, size, &newsid);
2785         }
2786         if (rc)
2787                 return rc;
2788
2789         rc = avc_has_perm(sid, newsid, isec->sclass,
2790                           FILE__RELABELTO, &ad);
2791         if (rc)
2792                 return rc;
2793
2794         rc = security_validate_transition(isec->sid, newsid, sid,
2795                                           isec->sclass);
2796         if (rc)
2797                 return rc;
2798
2799         return avc_has_perm(newsid,
2800                             sbsec->sid,
2801                             SECCLASS_FILESYSTEM,
2802                             FILESYSTEM__ASSOCIATE,
2803                             &ad);
2804 }
2805
2806 static void selinux_inode_post_setxattr(struct dentry *dentry, const char *name,
2807                                         const void *value, size_t size,
2808                                         int flags)
2809 {
2810         struct inode *inode = dentry->d_inode;
2811         struct inode_security_struct *isec = inode->i_security;
2812         u32 newsid;
2813         int rc;
2814
2815         if (strcmp(name, XATTR_NAME_SELINUX)) {
2816                 /* Not an attribute we recognize, so nothing to do. */
2817                 return;
2818         }
2819
2820         rc = security_context_to_sid_force(value, size, &newsid);
2821         if (rc) {
2822                 printk(KERN_ERR "SELinux:  unable to map context to SID"
2823                        "for (%s, %lu), rc=%d\n",
2824                        inode->i_sb->s_id, inode->i_ino, -rc);
2825                 return;
2826         }
2827
2828         isec->sid = newsid;
2829         return;
2830 }
2831
2832 static int selinux_inode_getxattr(struct dentry *dentry, const char *name)
2833 {
2834         const struct cred *cred = current_cred();
2835
2836         return dentry_has_perm(cred, NULL, dentry, FILE__GETATTR);
2837 }
2838
2839 static int selinux_inode_listxattr(struct dentry *dentry)
2840 {
2841         const struct cred *cred = current_cred();
2842
2843         return dentry_has_perm(cred, NULL, dentry, FILE__GETATTR);
2844 }
2845
2846 static int selinux_inode_removexattr(struct dentry *dentry, const char *name)
2847 {
2848         if (strcmp(name, XATTR_NAME_SELINUX))
2849                 return selinux_inode_setotherxattr(dentry, name);
2850
2851         /* No one is allowed to remove a SELinux security label.
2852            You can change the label, but all data must be labeled. */
2853         return -EACCES;
2854 }
2855
2856 /*
2857  * Copy the inode security context value to the user.
2858  *
2859  * Permission check is handled by selinux_inode_getxattr hook.
2860  */
2861 static int selinux_inode_getsecurity(const struct inode *inode, const char *name, void **buffer, bool alloc)
2862 {
2863         u32 size;
2864         int error;
2865         char *context = NULL;
2866         struct inode_security_struct *isec = inode->i_security;
2867
2868         if (strcmp(name, XATTR_SELINUX_SUFFIX))
2869                 return -EOPNOTSUPP;
2870
2871         /*
2872          * If the caller has CAP_MAC_ADMIN, then get the raw context
2873          * value even if it is not defined by current policy; otherwise,
2874          * use the in-core value under current policy.
2875          * Use the non-auditing forms of the permission checks since
2876          * getxattr may be called by unprivileged processes commonly
2877          * and lack of permission just means that we fall back to the
2878          * in-core context value, not a denial.
2879          */
2880         error = selinux_capable(current, current_cred(), CAP_MAC_ADMIN,
2881                                 SECURITY_CAP_NOAUDIT);
2882         if (!error)
2883                 error = security_sid_to_context_force(isec->sid, &context,
2884                                                       &size);
2885         else
2886                 error = security_sid_to_context(isec->sid, &context, &size);
2887         if (error)
2888                 return error;
2889         error = size;
2890         if (alloc) {
2891                 *buffer = context;
2892                 goto out_nofree;
2893         }
2894         kfree(context);
2895 out_nofree:
2896         return error;
2897 }
2898
2899 static int selinux_inode_setsecurity(struct inode *inode, const char *name,
2900                                      const void *value, size_t size, int flags)
2901 {
2902         struct inode_security_struct *isec = inode->i_security;
2903         u32 newsid;
2904         int rc;
2905
2906         if (strcmp(name, XATTR_SELINUX_SUFFIX))
2907                 return -EOPNOTSUPP;
2908
2909         if (!value || !size)
2910                 return -EACCES;
2911
2912         rc = security_context_to_sid((void *)value, size, &newsid);
2913         if (rc)
2914                 return rc;
2915
2916         isec->sid = newsid;
2917         return 0;
2918 }
2919
2920 static int selinux_inode_listsecurity(struct inode *inode, char *buffer, size_t buffer_size)
2921 {
2922         const int len = sizeof(XATTR_NAME_SELINUX);
2923         if (buffer && len <= buffer_size)
2924                 memcpy(buffer, XATTR_NAME_SELINUX, len);
2925         return len;
2926 }
2927
2928 static void selinux_inode_getsecid(const struct inode *inode, u32 *secid)
2929 {
2930         struct inode_security_struct *isec = inode->i_security;
2931         *secid = isec->sid;
2932 }
2933
2934 /* file security operations */
2935
2936 static int selinux_revalidate_file_permission(struct file *file, int mask)
2937 {
2938         const struct cred *cred = current_cred();
2939         struct inode *inode = file->f_path.dentry->d_inode;
2940
2941         if (!mask) {
2942                 /* No permission to check.  Existence test. */
2943                 return 0;
2944         }
2945
2946         /* file_mask_to_av won't add FILE__WRITE if MAY_APPEND is set */
2947         if ((file->f_flags & O_APPEND) && (mask & MAY_WRITE))
2948                 mask |= MAY_APPEND;
2949
2950         return file_has_perm(cred, file,
2951                              file_mask_to_av(inode->i_mode, mask));
2952 }
2953
2954 static int selinux_file_permission(struct file *file, int mask)
2955 {
2956         if (!mask)
2957                 /* No permission to check.  Existence test. */
2958                 return 0;
2959
2960         return selinux_revalidate_file_permission(file, mask);
2961 }
2962
2963 static int selinux_file_alloc_security(struct file *file)
2964 {
2965         return file_alloc_security(file);
2966 }
2967
2968 static void selinux_file_free_security(struct file *file)
2969 {
2970         file_free_security(file);
2971 }
2972
2973 static int selinux_file_ioctl(struct file *file, unsigned int cmd,
2974                               unsigned long arg)
2975 {
2976         const struct cred *cred = current_cred();
2977         u32 av = 0;
2978
2979         if (_IOC_DIR(cmd) & _IOC_WRITE)
2980                 av |= FILE__WRITE;
2981         if (_IOC_DIR(cmd) & _IOC_READ)
2982                 av |= FILE__READ;
2983         if (!av)
2984                 av = FILE__IOCTL;
2985
2986         return file_has_perm(cred, file, av);
2987 }
2988
2989 static int file_map_prot_check(struct file *file, unsigned long prot, int shared)
2990 {
2991         const struct cred *cred = current_cred();
2992         int rc = 0;
2993
2994 #ifndef CONFIG_PPC32
2995         if ((prot & PROT_EXEC) && (!file || (!shared && (prot & PROT_WRITE)))) {
2996                 /*
2997                  * We are making executable an anonymous mapping or a
2998                  * private file mapping that will also be writable.
2999                  * This has an additional check.
3000                  */
3001                 rc = cred_has_perm(cred, cred, PROCESS__EXECMEM);
3002                 if (rc)
3003                         goto error;
3004         }
3005 #endif
3006
3007         if (file) {
3008                 /* read access is always possible with a mapping */
3009                 u32 av = FILE__READ;
3010
3011                 /* write access only matters if the mapping is shared */
3012                 if (shared && (prot & PROT_WRITE))
3013                         av |= FILE__WRITE;
3014
3015                 if (prot & PROT_EXEC)
3016                         av |= FILE__EXECUTE;
3017
3018                 return file_has_perm(cred, file, av);
3019         }
3020
3021 error:
3022         return rc;
3023 }
3024
3025 static int selinux_file_mmap(struct file *file, unsigned long reqprot,
3026                              unsigned long prot, unsigned long flags,
3027                              unsigned long addr, unsigned long addr_only)
3028 {
3029         int rc = 0;
3030         u32 sid = current_sid();
3031
3032         if (addr < mmap_min_addr)
3033                 rc = avc_has_perm(sid, sid, SECCLASS_MEMPROTECT,
3034                                   MEMPROTECT__MMAP_ZERO, NULL);
3035         if (rc || addr_only)
3036                 return rc;
3037
3038         if (selinux_checkreqprot)
3039                 prot = reqprot;
3040
3041         return file_map_prot_check(file, prot,
3042                                    (flags & MAP_TYPE) == MAP_SHARED);
3043 }
3044
3045 static int selinux_file_mprotect(struct vm_area_struct *vma,
3046                                  unsigned long reqprot,
3047                                  unsigned long prot)
3048 {
3049         const struct cred *cred = current_cred();
3050
3051         if (selinux_checkreqprot)
3052                 prot = reqprot;
3053
3054 #ifndef CONFIG_PPC32
3055         if ((prot & PROT_EXEC) && !(vma->vm_flags & VM_EXEC)) {
3056                 int rc = 0;
3057                 if (vma->vm_start >= vma->vm_mm->start_brk &&
3058                     vma->vm_end <= vma->vm_mm->brk) {
3059                         rc = cred_has_perm(cred, cred, PROCESS__EXECHEAP);
3060                 } else if (!vma->vm_file &&
3061                            vma->vm_start <= vma->vm_mm->start_stack &&
3062                            vma->vm_end >= vma->vm_mm->start_stack) {
3063                         rc = current_has_perm(current, PROCESS__EXECSTACK);
3064                 } else if (vma->vm_file && vma->anon_vma) {
3065                         /*
3066                          * We are making executable a file mapping that has
3067                          * had some COW done. Since pages might have been
3068                          * written, check ability to execute the possibly
3069                          * modified content.  This typically should only
3070                          * occur for text relocations.
3071                          */
3072                         rc = file_has_perm(cred, vma->vm_file, FILE__EXECMOD);
3073                 }
3074                 if (rc)
3075                         return rc;
3076         }
3077 #endif
3078
3079         return file_map_prot_check(vma->vm_file, prot, vma->vm_flags&VM_SHARED);
3080 }
3081
3082 static int selinux_file_lock(struct file *file, unsigned int cmd)
3083 {
3084         const struct cred *cred = current_cred();
3085
3086         return file_has_perm(cred, file, FILE__LOCK);
3087 }
3088
3089 static int selinux_file_fcntl(struct file *file, unsigned int cmd,
3090                               unsigned long arg)
3091 {
3092         const struct cred *cred = current_cred();
3093         int err = 0;
3094
3095         switch (cmd) {
3096         case F_SETFL:
3097                 if (!file->f_path.dentry || !file->f_path.dentry->d_inode) {
3098                         err = -EINVAL;
3099                         break;
3100                 }
3101
3102                 if ((file->f_flags & O_APPEND) && !(arg & O_APPEND)) {
3103                         err = file_has_perm(cred, file, FILE__WRITE);
3104                         break;
3105                 }
3106                 /* fall through */
3107         case F_SETOWN:
3108         case F_SETSIG:
3109         case F_GETFL:
3110         case F_GETOWN:
3111         case F_GETSIG:
3112                 /* Just check FD__USE permission */
3113                 err = file_has_perm(cred, file, 0);
3114                 break;
3115         case F_GETLK:
3116         case F_SETLK:
3117         case F_SETLKW:
3118 #if BITS_PER_LONG == 32
3119         case F_GETLK64:
3120         case F_SETLK64:
3121         case F_SETLKW64:
3122 #endif
3123                 if (!file->f_path.dentry || !file->f_path.dentry->d_inode) {
3124                         err = -EINVAL;
3125                         break;
3126                 }
3127                 err = file_has_perm(cred, file, FILE__LOCK);
3128                 break;
3129         }
3130
3131         return err;
3132 }
3133
3134 static int selinux_file_set_fowner(struct file *file)
3135 {
3136         struct file_security_struct *fsec;
3137
3138         fsec = file->f_security;
3139         fsec->fown_sid = current_sid();
3140
3141         return 0;
3142 }
3143
3144 static int selinux_file_send_sigiotask(struct task_struct *tsk,
3145                                        struct fown_struct *fown, int signum)
3146 {
3147         struct file *file;
3148         u32 sid = task_sid(tsk);
3149         u32 perm;
3150         struct file_security_struct *fsec;
3151
3152         /* struct fown_struct is never outside the context of a struct file */
3153         file = container_of(fown, struct file, f_owner);
3154
3155         fsec = file->f_security;
3156
3157         if (!signum)
3158                 perm = signal_to_av(SIGIO); /* as per send_sigio_to_task */
3159         else
3160                 perm = signal_to_av(signum);
3161
3162         return avc_has_perm(fsec->fown_sid, sid,
3163                             SECCLASS_PROCESS, perm, NULL);
3164 }
3165
3166 static int selinux_file_receive(struct file *file)
3167 {
3168         const struct cred *cred = current_cred();
3169
3170         return file_has_perm(cred, file, file_to_av(file));
3171 }
3172
3173 static int selinux_dentry_open(struct file *file, const struct cred *cred)
3174 {
3175         struct file_security_struct *fsec;
3176         struct inode *inode;
3177         struct inode_security_struct *isec;
3178
3179         inode = file->f_path.dentry->d_inode;
3180         fsec = file->f_security;
3181         isec = inode->i_security;
3182         /*
3183          * Save inode label and policy sequence number
3184          * at open-time so that selinux_file_permission
3185          * can determine whether revalidation is necessary.
3186          * Task label is already saved in the file security
3187          * struct as its SID.
3188          */
3189         fsec->isid = isec->sid;