apparmor: clarify CRYPTO dependency
[muen/linux.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, 2009 Hewlett-Packard Development Company, L.P.
17  *      Paul Moore <paul@paul-moore.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/kd.h>
28 #include <linux/kernel.h>
29 #include <linux/tracehook.h>
30 #include <linux/errno.h>
31 #include <linux/sched.h>
32 #include <linux/lsm_hooks.h>
33 #include <linux/xattr.h>
34 #include <linux/capability.h>
35 #include <linux/unistd.h>
36 #include <linux/mm.h>
37 #include <linux/mman.h>
38 #include <linux/slab.h>
39 #include <linux/pagemap.h>
40 #include <linux/proc_fs.h>
41 #include <linux/swap.h>
42 #include <linux/spinlock.h>
43 #include <linux/syscalls.h>
44 #include <linux/dcache.h>
45 #include <linux/file.h>
46 #include <linux/fdtable.h>
47 #include <linux/namei.h>
48 #include <linux/mount.h>
49 #include <linux/netfilter_ipv4.h>
50 #include <linux/netfilter_ipv6.h>
51 #include <linux/tty.h>
52 #include <net/icmp.h>
53 #include <net/ip.h>             /* for local_port_range[] */
54 #include <net/tcp.h>            /* struct or_callable used in sock_rcv_skb */
55 #include <net/inet_connection_sock.h>
56 #include <net/net_namespace.h>
57 #include <net/netlabel.h>
58 #include <linux/uaccess.h>
59 #include <asm/ioctls.h>
60 #include <linux/atomic.h>
61 #include <linux/bitops.h>
62 #include <linux/interrupt.h>
63 #include <linux/netdevice.h>    /* for network interface checks */
64 #include <net/netlink.h>
65 #include <linux/tcp.h>
66 #include <linux/udp.h>
67 #include <linux/dccp.h>
68 #include <linux/quota.h>
69 #include <linux/un.h>           /* for Unix socket types */
70 #include <net/af_unix.h>        /* for Unix socket types */
71 #include <linux/parser.h>
72 #include <linux/nfs_mount.h>
73 #include <net/ipv6.h>
74 #include <linux/hugetlb.h>
75 #include <linux/personality.h>
76 #include <linux/audit.h>
77 #include <linux/string.h>
78 #include <linux/selinux.h>
79 #include <linux/mutex.h>
80 #include <linux/posix-timers.h>
81 #include <linux/syslog.h>
82 #include <linux/user_namespace.h>
83 #include <linux/export.h>
84 #include <linux/msg.h>
85 #include <linux/shm.h>
86
87 #include "avc.h"
88 #include "objsec.h"
89 #include "netif.h"
90 #include "netnode.h"
91 #include "netport.h"
92 #include "xfrm.h"
93 #include "netlabel.h"
94 #include "audit.h"
95 #include "avc_ss.h"
96
97 /* SECMARK reference count */
98 static atomic_t selinux_secmark_refcount = ATOMIC_INIT(0);
99
100 #ifdef CONFIG_SECURITY_SELINUX_DEVELOP
101 int selinux_enforcing;
102
103 static int __init enforcing_setup(char *str)
104 {
105         unsigned long enforcing;
106         if (!kstrtoul(str, 0, &enforcing))
107                 selinux_enforcing = enforcing ? 1 : 0;
108         return 1;
109 }
110 __setup("enforcing=", enforcing_setup);
111 #endif
112
113 #ifdef CONFIG_SECURITY_SELINUX_BOOTPARAM
114 int selinux_enabled = CONFIG_SECURITY_SELINUX_BOOTPARAM_VALUE;
115
116 static int __init selinux_enabled_setup(char *str)
117 {
118         unsigned long enabled;
119         if (!kstrtoul(str, 0, &enabled))
120                 selinux_enabled = enabled ? 1 : 0;
121         return 1;
122 }
123 __setup("selinux=", selinux_enabled_setup);
124 #else
125 int selinux_enabled = 1;
126 #endif
127
128 static struct kmem_cache *sel_inode_cache;
129
130 /**
131  * selinux_secmark_enabled - Check to see if SECMARK is currently enabled
132  *
133  * Description:
134  * This function checks the SECMARK reference counter to see if any SECMARK
135  * targets are currently configured, if the reference counter is greater than
136  * zero SECMARK is considered to be enabled.  Returns true (1) if SECMARK is
137  * enabled, false (0) if SECMARK is disabled.  If the always_check_network
138  * policy capability is enabled, SECMARK is always considered enabled.
139  *
140  */
141 static int selinux_secmark_enabled(void)
142 {
143         return (selinux_policycap_alwaysnetwork || atomic_read(&selinux_secmark_refcount));
144 }
145
146 /**
147  * selinux_peerlbl_enabled - Check to see if peer labeling is currently enabled
148  *
149  * Description:
150  * This function checks if NetLabel or labeled IPSEC is enabled.  Returns true
151  * (1) if any are enabled or false (0) if neither are enabled.  If the
152  * always_check_network policy capability is enabled, peer labeling
153  * is always considered enabled.
154  *
155  */
156 static int selinux_peerlbl_enabled(void)
157 {
158         return (selinux_policycap_alwaysnetwork || netlbl_enabled() || selinux_xfrm_enabled());
159 }
160
161 static int selinux_netcache_avc_callback(u32 event)
162 {
163         if (event == AVC_CALLBACK_RESET) {
164                 sel_netif_flush();
165                 sel_netnode_flush();
166                 sel_netport_flush();
167                 synchronize_net();
168         }
169         return 0;
170 }
171
172 /*
173  * initialise the security for the init task
174  */
175 static void cred_init_security(void)
176 {
177         struct cred *cred = (struct cred *) current->real_cred;
178         struct task_security_struct *tsec;
179
180         tsec = kzalloc(sizeof(struct task_security_struct), GFP_KERNEL);
181         if (!tsec)
182                 panic("SELinux:  Failed to initialize initial task.\n");
183
184         tsec->osid = tsec->sid = SECINITSID_KERNEL;
185         cred->security = tsec;
186 }
187
188 /*
189  * get the security ID of a set of credentials
190  */
191 static inline u32 cred_sid(const struct cred *cred)
192 {
193         const struct task_security_struct *tsec;
194
195         tsec = cred->security;
196         return tsec->sid;
197 }
198
199 /*
200  * get the objective security ID of a task
201  */
202 static inline u32 task_sid(const struct task_struct *task)
203 {
204         u32 sid;
205
206         rcu_read_lock();
207         sid = cred_sid(__task_cred(task));
208         rcu_read_unlock();
209         return sid;
210 }
211
212 /*
213  * get the subjective security ID of the current task
214  */
215 static inline u32 current_sid(void)
216 {
217         const struct task_security_struct *tsec = current_security();
218
219         return tsec->sid;
220 }
221
222 /* Allocate and free functions for each kind of security blob. */
223
224 static int inode_alloc_security(struct inode *inode)
225 {
226         struct inode_security_struct *isec;
227         u32 sid = current_sid();
228
229         isec = kmem_cache_zalloc(sel_inode_cache, GFP_NOFS);
230         if (!isec)
231                 return -ENOMEM;
232
233         mutex_init(&isec->lock);
234         INIT_LIST_HEAD(&isec->list);
235         isec->inode = inode;
236         isec->sid = SECINITSID_UNLABELED;
237         isec->sclass = SECCLASS_FILE;
238         isec->task_sid = sid;
239         inode->i_security = isec;
240
241         return 0;
242 }
243
244 static void inode_free_rcu(struct rcu_head *head)
245 {
246         struct inode_security_struct *isec;
247
248         isec = container_of(head, struct inode_security_struct, rcu);
249         kmem_cache_free(sel_inode_cache, isec);
250 }
251
252 static void inode_free_security(struct inode *inode)
253 {
254         struct inode_security_struct *isec = inode->i_security;
255         struct superblock_security_struct *sbsec = inode->i_sb->s_security;
256
257         /*
258          * As not all inode security structures are in a list, we check for
259          * empty list outside of the lock to make sure that we won't waste
260          * time taking a lock doing nothing.
261          *
262          * The list_del_init() function can be safely called more than once.
263          * It should not be possible for this function to be called with
264          * concurrent list_add(), but for better safety against future changes
265          * in the code, we use list_empty_careful() here.
266          */
267         if (!list_empty_careful(&isec->list)) {
268                 spin_lock(&sbsec->isec_lock);
269                 list_del_init(&isec->list);
270                 spin_unlock(&sbsec->isec_lock);
271         }
272
273         /*
274          * The inode may still be referenced in a path walk and
275          * a call to selinux_inode_permission() can be made
276          * after inode_free_security() is called. Ideally, the VFS
277          * wouldn't do this, but fixing that is a much harder
278          * job. For now, simply free the i_security via RCU, and
279          * leave the current inode->i_security pointer intact.
280          * The inode will be freed after the RCU grace period too.
281          */
282         call_rcu(&isec->rcu, inode_free_rcu);
283 }
284
285 static int file_alloc_security(struct file *file)
286 {
287         struct file_security_struct *fsec;
288         u32 sid = current_sid();
289
290         fsec = kzalloc(sizeof(struct file_security_struct), GFP_KERNEL);
291         if (!fsec)
292                 return -ENOMEM;
293
294         fsec->sid = sid;
295         fsec->fown_sid = sid;
296         file->f_security = fsec;
297
298         return 0;
299 }
300
301 static void file_free_security(struct file *file)
302 {
303         struct file_security_struct *fsec = file->f_security;
304         file->f_security = NULL;
305         kfree(fsec);
306 }
307
308 static int superblock_alloc_security(struct super_block *sb)
309 {
310         struct superblock_security_struct *sbsec;
311
312         sbsec = kzalloc(sizeof(struct superblock_security_struct), GFP_KERNEL);
313         if (!sbsec)
314                 return -ENOMEM;
315
316         mutex_init(&sbsec->lock);
317         INIT_LIST_HEAD(&sbsec->isec_head);
318         spin_lock_init(&sbsec->isec_lock);
319         sbsec->sb = sb;
320         sbsec->sid = SECINITSID_UNLABELED;
321         sbsec->def_sid = SECINITSID_FILE;
322         sbsec->mntpoint_sid = SECINITSID_UNLABELED;
323         sb->s_security = sbsec;
324
325         return 0;
326 }
327
328 static void superblock_free_security(struct super_block *sb)
329 {
330         struct superblock_security_struct *sbsec = sb->s_security;
331         sb->s_security = NULL;
332         kfree(sbsec);
333 }
334
335 /* The file system's label must be initialized prior to use. */
336
337 static const char *labeling_behaviors[7] = {
338         "uses xattr",
339         "uses transition SIDs",
340         "uses task SIDs",
341         "uses genfs_contexts",
342         "not configured for labeling",
343         "uses mountpoint labeling",
344         "uses native labeling",
345 };
346
347 static int inode_doinit_with_dentry(struct inode *inode, struct dentry *opt_dentry);
348
349 static inline int inode_doinit(struct inode *inode)
350 {
351         return inode_doinit_with_dentry(inode, NULL);
352 }
353
354 enum {
355         Opt_error = -1,
356         Opt_context = 1,
357         Opt_fscontext = 2,
358         Opt_defcontext = 3,
359         Opt_rootcontext = 4,
360         Opt_labelsupport = 5,
361         Opt_nextmntopt = 6,
362 };
363
364 #define NUM_SEL_MNT_OPTS        (Opt_nextmntopt - 1)
365
366 static const match_table_t tokens = {
367         {Opt_context, CONTEXT_STR "%s"},
368         {Opt_fscontext, FSCONTEXT_STR "%s"},
369         {Opt_defcontext, DEFCONTEXT_STR "%s"},
370         {Opt_rootcontext, ROOTCONTEXT_STR "%s"},
371         {Opt_labelsupport, LABELSUPP_STR},
372         {Opt_error, NULL},
373 };
374
375 #define SEL_MOUNT_FAIL_MSG "SELinux:  duplicate or incompatible mount options\n"
376
377 static int may_context_mount_sb_relabel(u32 sid,
378                         struct superblock_security_struct *sbsec,
379                         const struct cred *cred)
380 {
381         const struct task_security_struct *tsec = cred->security;
382         int rc;
383
384         rc = avc_has_perm(tsec->sid, sbsec->sid, SECCLASS_FILESYSTEM,
385                           FILESYSTEM__RELABELFROM, NULL);
386         if (rc)
387                 return rc;
388
389         rc = avc_has_perm(tsec->sid, sid, SECCLASS_FILESYSTEM,
390                           FILESYSTEM__RELABELTO, NULL);
391         return rc;
392 }
393
394 static int may_context_mount_inode_relabel(u32 sid,
395                         struct superblock_security_struct *sbsec,
396                         const struct cred *cred)
397 {
398         const struct task_security_struct *tsec = cred->security;
399         int rc;
400         rc = avc_has_perm(tsec->sid, sbsec->sid, SECCLASS_FILESYSTEM,
401                           FILESYSTEM__RELABELFROM, NULL);
402         if (rc)
403                 return rc;
404
405         rc = avc_has_perm(sid, sbsec->sid, SECCLASS_FILESYSTEM,
406                           FILESYSTEM__ASSOCIATE, NULL);
407         return rc;
408 }
409
410 static int selinux_is_sblabel_mnt(struct super_block *sb)
411 {
412         struct superblock_security_struct *sbsec = sb->s_security;
413
414         return sbsec->behavior == SECURITY_FS_USE_XATTR ||
415                 sbsec->behavior == SECURITY_FS_USE_TRANS ||
416                 sbsec->behavior == SECURITY_FS_USE_TASK ||
417                 sbsec->behavior == SECURITY_FS_USE_NATIVE ||
418                 /* Special handling. Genfs but also in-core setxattr handler */
419                 !strcmp(sb->s_type->name, "sysfs") ||
420                 !strcmp(sb->s_type->name, "pstore") ||
421                 !strcmp(sb->s_type->name, "debugfs") ||
422                 !strcmp(sb->s_type->name, "rootfs");
423 }
424
425 static int sb_finish_set_opts(struct super_block *sb)
426 {
427         struct superblock_security_struct *sbsec = sb->s_security;
428         struct dentry *root = sb->s_root;
429         struct inode *root_inode = d_backing_inode(root);
430         int rc = 0;
431
432         if (sbsec->behavior == SECURITY_FS_USE_XATTR) {
433                 /* Make sure that the xattr handler exists and that no
434                    error other than -ENODATA is returned by getxattr on
435                    the root directory.  -ENODATA is ok, as this may be
436                    the first boot of the SELinux kernel before we have
437                    assigned xattr values to the filesystem. */
438                 if (!root_inode->i_op->getxattr) {
439                         printk(KERN_WARNING "SELinux: (dev %s, type %s) has no "
440                                "xattr support\n", sb->s_id, sb->s_type->name);
441                         rc = -EOPNOTSUPP;
442                         goto out;
443                 }
444                 rc = root_inode->i_op->getxattr(root, XATTR_NAME_SELINUX, NULL, 0);
445                 if (rc < 0 && rc != -ENODATA) {
446                         if (rc == -EOPNOTSUPP)
447                                 printk(KERN_WARNING "SELinux: (dev %s, type "
448                                        "%s) has no security xattr handler\n",
449                                        sb->s_id, sb->s_type->name);
450                         else
451                                 printk(KERN_WARNING "SELinux: (dev %s, type "
452                                        "%s) getxattr errno %d\n", sb->s_id,
453                                        sb->s_type->name, -rc);
454                         goto out;
455                 }
456         }
457
458         if (sbsec->behavior > ARRAY_SIZE(labeling_behaviors))
459                 printk(KERN_ERR "SELinux: initialized (dev %s, type %s), unknown behavior\n",
460                        sb->s_id, sb->s_type->name);
461
462         sbsec->flags |= SE_SBINITIALIZED;
463         if (selinux_is_sblabel_mnt(sb))
464                 sbsec->flags |= SBLABEL_MNT;
465
466         /* Initialize the root inode. */
467         rc = inode_doinit_with_dentry(root_inode, root);
468
469         /* Initialize any other inodes associated with the superblock, e.g.
470            inodes created prior to initial policy load or inodes created
471            during get_sb by a pseudo filesystem that directly
472            populates itself. */
473         spin_lock(&sbsec->isec_lock);
474 next_inode:
475         if (!list_empty(&sbsec->isec_head)) {
476                 struct inode_security_struct *isec =
477                                 list_entry(sbsec->isec_head.next,
478                                            struct inode_security_struct, list);
479                 struct inode *inode = isec->inode;
480                 list_del_init(&isec->list);
481                 spin_unlock(&sbsec->isec_lock);
482                 inode = igrab(inode);
483                 if (inode) {
484                         if (!IS_PRIVATE(inode))
485                                 inode_doinit(inode);
486                         iput(inode);
487                 }
488                 spin_lock(&sbsec->isec_lock);
489                 goto next_inode;
490         }
491         spin_unlock(&sbsec->isec_lock);
492 out:
493         return rc;
494 }
495
496 /*
497  * This function should allow an FS to ask what it's mount security
498  * options were so it can use those later for submounts, displaying
499  * mount options, or whatever.
500  */
501 static int selinux_get_mnt_opts(const struct super_block *sb,
502                                 struct security_mnt_opts *opts)
503 {
504         int rc = 0, i;
505         struct superblock_security_struct *sbsec = sb->s_security;
506         char *context = NULL;
507         u32 len;
508         char tmp;
509
510         security_init_mnt_opts(opts);
511
512         if (!(sbsec->flags & SE_SBINITIALIZED))
513                 return -EINVAL;
514
515         if (!ss_initialized)
516                 return -EINVAL;
517
518         /* make sure we always check enough bits to cover the mask */
519         BUILD_BUG_ON(SE_MNTMASK >= (1 << NUM_SEL_MNT_OPTS));
520
521         tmp = sbsec->flags & SE_MNTMASK;
522         /* count the number of mount options for this sb */
523         for (i = 0; i < NUM_SEL_MNT_OPTS; i++) {
524                 if (tmp & 0x01)
525                         opts->num_mnt_opts++;
526                 tmp >>= 1;
527         }
528         /* Check if the Label support flag is set */
529         if (sbsec->flags & SBLABEL_MNT)
530                 opts->num_mnt_opts++;
531
532         opts->mnt_opts = kcalloc(opts->num_mnt_opts, sizeof(char *), GFP_ATOMIC);
533         if (!opts->mnt_opts) {
534                 rc = -ENOMEM;
535                 goto out_free;
536         }
537
538         opts->mnt_opts_flags = kcalloc(opts->num_mnt_opts, sizeof(int), GFP_ATOMIC);
539         if (!opts->mnt_opts_flags) {
540                 rc = -ENOMEM;
541                 goto out_free;
542         }
543
544         i = 0;
545         if (sbsec->flags & FSCONTEXT_MNT) {
546                 rc = security_sid_to_context(sbsec->sid, &context, &len);
547                 if (rc)
548                         goto out_free;
549                 opts->mnt_opts[i] = context;
550                 opts->mnt_opts_flags[i++] = FSCONTEXT_MNT;
551         }
552         if (sbsec->flags & CONTEXT_MNT) {
553                 rc = security_sid_to_context(sbsec->mntpoint_sid, &context, &len);
554                 if (rc)
555                         goto out_free;
556                 opts->mnt_opts[i] = context;
557                 opts->mnt_opts_flags[i++] = CONTEXT_MNT;
558         }
559         if (sbsec->flags & DEFCONTEXT_MNT) {
560                 rc = security_sid_to_context(sbsec->def_sid, &context, &len);
561                 if (rc)
562                         goto out_free;
563                 opts->mnt_opts[i] = context;
564                 opts->mnt_opts_flags[i++] = DEFCONTEXT_MNT;
565         }
566         if (sbsec->flags & ROOTCONTEXT_MNT) {
567                 struct inode *root = d_backing_inode(sbsec->sb->s_root);
568                 struct inode_security_struct *isec = root->i_security;
569
570                 rc = security_sid_to_context(isec->sid, &context, &len);
571                 if (rc)
572                         goto out_free;
573                 opts->mnt_opts[i] = context;
574                 opts->mnt_opts_flags[i++] = ROOTCONTEXT_MNT;
575         }
576         if (sbsec->flags & SBLABEL_MNT) {
577                 opts->mnt_opts[i] = NULL;
578                 opts->mnt_opts_flags[i++] = SBLABEL_MNT;
579         }
580
581         BUG_ON(i != opts->num_mnt_opts);
582
583         return 0;
584
585 out_free:
586         security_free_mnt_opts(opts);
587         return rc;
588 }
589
590 static int bad_option(struct superblock_security_struct *sbsec, char flag,
591                       u32 old_sid, u32 new_sid)
592 {
593         char mnt_flags = sbsec->flags & SE_MNTMASK;
594
595         /* check if the old mount command had the same options */
596         if (sbsec->flags & SE_SBINITIALIZED)
597                 if (!(sbsec->flags & flag) ||
598                     (old_sid != new_sid))
599                         return 1;
600
601         /* check if we were passed the same options twice,
602          * aka someone passed context=a,context=b
603          */
604         if (!(sbsec->flags & SE_SBINITIALIZED))
605                 if (mnt_flags & flag)
606                         return 1;
607         return 0;
608 }
609
610 /*
611  * Allow filesystems with binary mount data to explicitly set mount point
612  * labeling information.
613  */
614 static int selinux_set_mnt_opts(struct super_block *sb,
615                                 struct security_mnt_opts *opts,
616                                 unsigned long kern_flags,
617                                 unsigned long *set_kern_flags)
618 {
619         const struct cred *cred = current_cred();
620         int rc = 0, i;
621         struct superblock_security_struct *sbsec = sb->s_security;
622         const char *name = sb->s_type->name;
623         struct inode *inode = d_backing_inode(sbsec->sb->s_root);
624         struct inode_security_struct *root_isec = inode->i_security;
625         u32 fscontext_sid = 0, context_sid = 0, rootcontext_sid = 0;
626         u32 defcontext_sid = 0;
627         char **mount_options = opts->mnt_opts;
628         int *flags = opts->mnt_opts_flags;
629         int num_opts = opts->num_mnt_opts;
630
631         mutex_lock(&sbsec->lock);
632
633         if (!ss_initialized) {
634                 if (!num_opts) {
635                         /* Defer initialization until selinux_complete_init,
636                            after the initial policy is loaded and the security
637                            server is ready to handle calls. */
638                         goto out;
639                 }
640                 rc = -EINVAL;
641                 printk(KERN_WARNING "SELinux: Unable to set superblock options "
642                         "before the security server is initialized\n");
643                 goto out;
644         }
645         if (kern_flags && !set_kern_flags) {
646                 /* Specifying internal flags without providing a place to
647                  * place the results is not allowed */
648                 rc = -EINVAL;
649                 goto out;
650         }
651
652         /*
653          * Binary mount data FS will come through this function twice.  Once
654          * from an explicit call and once from the generic calls from the vfs.
655          * Since the generic VFS calls will not contain any security mount data
656          * we need to skip the double mount verification.
657          *
658          * This does open a hole in which we will not notice if the first
659          * mount using this sb set explict options and a second mount using
660          * this sb does not set any security options.  (The first options
661          * will be used for both mounts)
662          */
663         if ((sbsec->flags & SE_SBINITIALIZED) && (sb->s_type->fs_flags & FS_BINARY_MOUNTDATA)
664             && (num_opts == 0))
665                 goto out;
666
667         /*
668          * parse the mount options, check if they are valid sids.
669          * also check if someone is trying to mount the same sb more
670          * than once with different security options.
671          */
672         for (i = 0; i < num_opts; i++) {
673                 u32 sid;
674
675                 if (flags[i] == SBLABEL_MNT)
676                         continue;
677                 rc = security_context_to_sid(mount_options[i],
678                                              strlen(mount_options[i]), &sid, GFP_KERNEL);
679                 if (rc) {
680                         printk(KERN_WARNING "SELinux: security_context_to_sid"
681                                "(%s) failed for (dev %s, type %s) errno=%d\n",
682                                mount_options[i], sb->s_id, name, rc);
683                         goto out;
684                 }
685                 switch (flags[i]) {
686                 case FSCONTEXT_MNT:
687                         fscontext_sid = sid;
688
689                         if (bad_option(sbsec, FSCONTEXT_MNT, sbsec->sid,
690                                         fscontext_sid))
691                                 goto out_double_mount;
692
693                         sbsec->flags |= FSCONTEXT_MNT;
694                         break;
695                 case CONTEXT_MNT:
696                         context_sid = sid;
697
698                         if (bad_option(sbsec, CONTEXT_MNT, sbsec->mntpoint_sid,
699                                         context_sid))
700                                 goto out_double_mount;
701
702                         sbsec->flags |= CONTEXT_MNT;
703                         break;
704                 case ROOTCONTEXT_MNT:
705                         rootcontext_sid = sid;
706
707                         if (bad_option(sbsec, ROOTCONTEXT_MNT, root_isec->sid,
708                                         rootcontext_sid))
709                                 goto out_double_mount;
710
711                         sbsec->flags |= ROOTCONTEXT_MNT;
712
713                         break;
714                 case DEFCONTEXT_MNT:
715                         defcontext_sid = sid;
716
717                         if (bad_option(sbsec, DEFCONTEXT_MNT, sbsec->def_sid,
718                                         defcontext_sid))
719                                 goto out_double_mount;
720
721                         sbsec->flags |= DEFCONTEXT_MNT;
722
723                         break;
724                 default:
725                         rc = -EINVAL;
726                         goto out;
727                 }
728         }
729
730         if (sbsec->flags & SE_SBINITIALIZED) {
731                 /* previously mounted with options, but not on this attempt? */
732                 if ((sbsec->flags & SE_MNTMASK) && !num_opts)
733                         goto out_double_mount;
734                 rc = 0;
735                 goto out;
736         }
737
738         if (strcmp(sb->s_type->name, "proc") == 0)
739                 sbsec->flags |= SE_SBPROC | SE_SBGENFS;
740
741         if (!strcmp(sb->s_type->name, "debugfs") ||
742             !strcmp(sb->s_type->name, "sysfs") ||
743             !strcmp(sb->s_type->name, "pstore"))
744                 sbsec->flags |= SE_SBGENFS;
745
746         if (!sbsec->behavior) {
747                 /*
748                  * Determine the labeling behavior to use for this
749                  * filesystem type.
750                  */
751                 rc = security_fs_use(sb);
752                 if (rc) {
753                         printk(KERN_WARNING
754                                 "%s: security_fs_use(%s) returned %d\n",
755                                         __func__, sb->s_type->name, rc);
756                         goto out;
757                 }
758         }
759         /* sets the context of the superblock for the fs being mounted. */
760         if (fscontext_sid) {
761                 rc = may_context_mount_sb_relabel(fscontext_sid, sbsec, cred);
762                 if (rc)
763                         goto out;
764
765                 sbsec->sid = fscontext_sid;
766         }
767
768         /*
769          * Switch to using mount point labeling behavior.
770          * sets the label used on all file below the mountpoint, and will set
771          * the superblock context if not already set.
772          */
773         if (kern_flags & SECURITY_LSM_NATIVE_LABELS && !context_sid) {
774                 sbsec->behavior = SECURITY_FS_USE_NATIVE;
775                 *set_kern_flags |= SECURITY_LSM_NATIVE_LABELS;
776         }
777
778         if (context_sid) {
779                 if (!fscontext_sid) {
780                         rc = may_context_mount_sb_relabel(context_sid, sbsec,
781                                                           cred);
782                         if (rc)
783                                 goto out;
784                         sbsec->sid = context_sid;
785                 } else {
786                         rc = may_context_mount_inode_relabel(context_sid, sbsec,
787                                                              cred);
788                         if (rc)
789                                 goto out;
790                 }
791                 if (!rootcontext_sid)
792                         rootcontext_sid = context_sid;
793
794                 sbsec->mntpoint_sid = context_sid;
795                 sbsec->behavior = SECURITY_FS_USE_MNTPOINT;
796         }
797
798         if (rootcontext_sid) {
799                 rc = may_context_mount_inode_relabel(rootcontext_sid, sbsec,
800                                                      cred);
801                 if (rc)
802                         goto out;
803
804                 root_isec->sid = rootcontext_sid;
805                 root_isec->initialized = 1;
806         }
807
808         if (defcontext_sid) {
809                 if (sbsec->behavior != SECURITY_FS_USE_XATTR &&
810                         sbsec->behavior != SECURITY_FS_USE_NATIVE) {
811                         rc = -EINVAL;
812                         printk(KERN_WARNING "SELinux: defcontext option is "
813                                "invalid for this filesystem type\n");
814                         goto out;
815                 }
816
817                 if (defcontext_sid != sbsec->def_sid) {
818                         rc = may_context_mount_inode_relabel(defcontext_sid,
819                                                              sbsec, cred);
820                         if (rc)
821                                 goto out;
822                 }
823
824                 sbsec->def_sid = defcontext_sid;
825         }
826
827         rc = sb_finish_set_opts(sb);
828 out:
829         mutex_unlock(&sbsec->lock);
830         return rc;
831 out_double_mount:
832         rc = -EINVAL;
833         printk(KERN_WARNING "SELinux: mount invalid.  Same superblock, different "
834                "security settings for (dev %s, type %s)\n", sb->s_id, name);
835         goto out;
836 }
837
838 static int selinux_cmp_sb_context(const struct super_block *oldsb,
839                                     const struct super_block *newsb)
840 {
841         struct superblock_security_struct *old = oldsb->s_security;
842         struct superblock_security_struct *new = newsb->s_security;
843         char oldflags = old->flags & SE_MNTMASK;
844         char newflags = new->flags & SE_MNTMASK;
845
846         if (oldflags != newflags)
847                 goto mismatch;
848         if ((oldflags & FSCONTEXT_MNT) && old->sid != new->sid)
849                 goto mismatch;
850         if ((oldflags & CONTEXT_MNT) && old->mntpoint_sid != new->mntpoint_sid)
851                 goto mismatch;
852         if ((oldflags & DEFCONTEXT_MNT) && old->def_sid != new->def_sid)
853                 goto mismatch;
854         if (oldflags & ROOTCONTEXT_MNT) {
855                 struct inode_security_struct *oldroot = d_backing_inode(oldsb->s_root)->i_security;
856                 struct inode_security_struct *newroot = d_backing_inode(newsb->s_root)->i_security;
857                 if (oldroot->sid != newroot->sid)
858                         goto mismatch;
859         }
860         return 0;
861 mismatch:
862         printk(KERN_WARNING "SELinux: mount invalid.  Same superblock, "
863                             "different security settings for (dev %s, "
864                             "type %s)\n", newsb->s_id, newsb->s_type->name);
865         return -EBUSY;
866 }
867
868 static int selinux_sb_clone_mnt_opts(const struct super_block *oldsb,
869                                         struct super_block *newsb)
870 {
871         const struct superblock_security_struct *oldsbsec = oldsb->s_security;
872         struct superblock_security_struct *newsbsec = newsb->s_security;
873
874         int set_fscontext =     (oldsbsec->flags & FSCONTEXT_MNT);
875         int set_context =       (oldsbsec->flags & CONTEXT_MNT);
876         int set_rootcontext =   (oldsbsec->flags & ROOTCONTEXT_MNT);
877
878         /*
879          * if the parent was able to be mounted it clearly had no special lsm
880          * mount options.  thus we can safely deal with this superblock later
881          */
882         if (!ss_initialized)
883                 return 0;
884
885         /* how can we clone if the old one wasn't set up?? */
886         BUG_ON(!(oldsbsec->flags & SE_SBINITIALIZED));
887
888         /* if fs is reusing a sb, make sure that the contexts match */
889         if (newsbsec->flags & SE_SBINITIALIZED)
890                 return selinux_cmp_sb_context(oldsb, newsb);
891
892         mutex_lock(&newsbsec->lock);
893
894         newsbsec->flags = oldsbsec->flags;
895
896         newsbsec->sid = oldsbsec->sid;
897         newsbsec->def_sid = oldsbsec->def_sid;
898         newsbsec->behavior = oldsbsec->behavior;
899
900         if (set_context) {
901                 u32 sid = oldsbsec->mntpoint_sid;
902
903                 if (!set_fscontext)
904                         newsbsec->sid = sid;
905                 if (!set_rootcontext) {
906                         struct inode *newinode = d_backing_inode(newsb->s_root);
907                         struct inode_security_struct *newisec = newinode->i_security;
908                         newisec->sid = sid;
909                 }
910                 newsbsec->mntpoint_sid = sid;
911         }
912         if (set_rootcontext) {
913                 const struct inode *oldinode = d_backing_inode(oldsb->s_root);
914                 const struct inode_security_struct *oldisec = oldinode->i_security;
915                 struct inode *newinode = d_backing_inode(newsb->s_root);
916                 struct inode_security_struct *newisec = newinode->i_security;
917
918                 newisec->sid = oldisec->sid;
919         }
920
921         sb_finish_set_opts(newsb);
922         mutex_unlock(&newsbsec->lock);
923         return 0;
924 }
925
926 static int selinux_parse_opts_str(char *options,
927                                   struct security_mnt_opts *opts)
928 {
929         char *p;
930         char *context = NULL, *defcontext = NULL;
931         char *fscontext = NULL, *rootcontext = NULL;
932         int rc, num_mnt_opts = 0;
933
934         opts->num_mnt_opts = 0;
935
936         /* Standard string-based options. */
937         while ((p = strsep(&options, "|")) != NULL) {
938                 int token;
939                 substring_t args[MAX_OPT_ARGS];
940
941                 if (!*p)
942                         continue;
943
944                 token = match_token(p, tokens, args);
945
946                 switch (token) {
947                 case Opt_context:
948                         if (context || defcontext) {
949                                 rc = -EINVAL;
950                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
951                                 goto out_err;
952                         }
953                         context = match_strdup(&args[0]);
954                         if (!context) {
955                                 rc = -ENOMEM;
956                                 goto out_err;
957                         }
958                         break;
959
960                 case Opt_fscontext:
961                         if (fscontext) {
962                                 rc = -EINVAL;
963                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
964                                 goto out_err;
965                         }
966                         fscontext = match_strdup(&args[0]);
967                         if (!fscontext) {
968                                 rc = -ENOMEM;
969                                 goto out_err;
970                         }
971                         break;
972
973                 case Opt_rootcontext:
974                         if (rootcontext) {
975                                 rc = -EINVAL;
976                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
977                                 goto out_err;
978                         }
979                         rootcontext = match_strdup(&args[0]);
980                         if (!rootcontext) {
981                                 rc = -ENOMEM;
982                                 goto out_err;
983                         }
984                         break;
985
986                 case Opt_defcontext:
987                         if (context || defcontext) {
988                                 rc = -EINVAL;
989                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
990                                 goto out_err;
991                         }
992                         defcontext = match_strdup(&args[0]);
993                         if (!defcontext) {
994                                 rc = -ENOMEM;
995                                 goto out_err;
996                         }
997                         break;
998                 case Opt_labelsupport:
999                         break;
1000                 default:
1001                         rc = -EINVAL;
1002                         printk(KERN_WARNING "SELinux:  unknown mount option\n");
1003                         goto out_err;
1004
1005                 }
1006         }
1007
1008         rc = -ENOMEM;
1009         opts->mnt_opts = kcalloc(NUM_SEL_MNT_OPTS, sizeof(char *), GFP_ATOMIC);
1010         if (!opts->mnt_opts)
1011                 goto out_err;
1012
1013         opts->mnt_opts_flags = kcalloc(NUM_SEL_MNT_OPTS, sizeof(int), GFP_ATOMIC);
1014         if (!opts->mnt_opts_flags) {
1015                 kfree(opts->mnt_opts);
1016                 goto out_err;
1017         }
1018
1019         if (fscontext) {
1020                 opts->mnt_opts[num_mnt_opts] = fscontext;
1021                 opts->mnt_opts_flags[num_mnt_opts++] = FSCONTEXT_MNT;
1022         }
1023         if (context) {
1024                 opts->mnt_opts[num_mnt_opts] = context;
1025                 opts->mnt_opts_flags[num_mnt_opts++] = CONTEXT_MNT;
1026         }
1027         if (rootcontext) {
1028                 opts->mnt_opts[num_mnt_opts] = rootcontext;
1029                 opts->mnt_opts_flags[num_mnt_opts++] = ROOTCONTEXT_MNT;
1030         }
1031         if (defcontext) {
1032                 opts->mnt_opts[num_mnt_opts] = defcontext;
1033                 opts->mnt_opts_flags[num_mnt_opts++] = DEFCONTEXT_MNT;
1034         }
1035
1036         opts->num_mnt_opts = num_mnt_opts;
1037         return 0;
1038
1039 out_err:
1040         kfree(context);
1041         kfree(defcontext);
1042         kfree(fscontext);
1043         kfree(rootcontext);
1044         return rc;
1045 }
1046 /*
1047  * string mount options parsing and call set the sbsec
1048  */
1049 static int superblock_doinit(struct super_block *sb, void *data)
1050 {
1051         int rc = 0;
1052         char *options = data;
1053         struct security_mnt_opts opts;
1054
1055         security_init_mnt_opts(&opts);
1056
1057         if (!data)
1058                 goto out;
1059
1060         BUG_ON(sb->s_type->fs_flags & FS_BINARY_MOUNTDATA);
1061
1062         rc = selinux_parse_opts_str(options, &opts);
1063         if (rc)
1064                 goto out_err;
1065
1066 out:
1067         rc = selinux_set_mnt_opts(sb, &opts, 0, NULL);
1068
1069 out_err:
1070         security_free_mnt_opts(&opts);
1071         return rc;
1072 }
1073
1074 static void selinux_write_opts(struct seq_file *m,
1075                                struct security_mnt_opts *opts)
1076 {
1077         int i;
1078         char *prefix;
1079
1080         for (i = 0; i < opts->num_mnt_opts; i++) {
1081                 char *has_comma;
1082
1083                 if (opts->mnt_opts[i])
1084                         has_comma = strchr(opts->mnt_opts[i], ',');
1085                 else
1086                         has_comma = NULL;
1087
1088                 switch (opts->mnt_opts_flags[i]) {
1089                 case CONTEXT_MNT:
1090                         prefix = CONTEXT_STR;
1091                         break;
1092                 case FSCONTEXT_MNT:
1093                         prefix = FSCONTEXT_STR;
1094                         break;
1095                 case ROOTCONTEXT_MNT:
1096                         prefix = ROOTCONTEXT_STR;
1097                         break;
1098                 case DEFCONTEXT_MNT:
1099                         prefix = DEFCONTEXT_STR;
1100                         break;
1101                 case SBLABEL_MNT:
1102                         seq_putc(m, ',');
1103                         seq_puts(m, LABELSUPP_STR);
1104                         continue;
1105                 default:
1106                         BUG();
1107                         return;
1108                 };
1109                 /* we need a comma before each option */
1110                 seq_putc(m, ',');
1111                 seq_puts(m, prefix);
1112                 if (has_comma)
1113                         seq_putc(m, '\"');
1114                 seq_escape(m, opts->mnt_opts[i], "\"\n\\");
1115                 if (has_comma)
1116                         seq_putc(m, '\"');
1117         }
1118 }
1119
1120 static int selinux_sb_show_options(struct seq_file *m, struct super_block *sb)
1121 {
1122         struct security_mnt_opts opts;
1123         int rc;
1124
1125         rc = selinux_get_mnt_opts(sb, &opts);
1126         if (rc) {
1127                 /* before policy load we may get EINVAL, don't show anything */
1128                 if (rc == -EINVAL)
1129                         rc = 0;
1130                 return rc;
1131         }
1132
1133         selinux_write_opts(m, &opts);
1134
1135         security_free_mnt_opts(&opts);
1136
1137         return rc;
1138 }
1139
1140 static inline u16 inode_mode_to_security_class(umode_t mode)
1141 {
1142         switch (mode & S_IFMT) {
1143         case S_IFSOCK:
1144                 return SECCLASS_SOCK_FILE;
1145         case S_IFLNK:
1146                 return SECCLASS_LNK_FILE;
1147         case S_IFREG:
1148                 return SECCLASS_FILE;
1149         case S_IFBLK:
1150                 return SECCLASS_BLK_FILE;
1151         case S_IFDIR:
1152                 return SECCLASS_DIR;
1153         case S_IFCHR:
1154                 return SECCLASS_CHR_FILE;
1155         case S_IFIFO:
1156                 return SECCLASS_FIFO_FILE;
1157
1158         }
1159
1160         return SECCLASS_FILE;
1161 }
1162
1163 static inline int default_protocol_stream(int protocol)
1164 {
1165         return (protocol == IPPROTO_IP || protocol == IPPROTO_TCP);
1166 }
1167
1168 static inline int default_protocol_dgram(int protocol)
1169 {
1170         return (protocol == IPPROTO_IP || protocol == IPPROTO_UDP);
1171 }
1172
1173 static inline u16 socket_type_to_security_class(int family, int type, int protocol)
1174 {
1175         switch (family) {
1176         case PF_UNIX:
1177                 switch (type) {
1178                 case SOCK_STREAM:
1179                 case SOCK_SEQPACKET:
1180                         return SECCLASS_UNIX_STREAM_SOCKET;
1181                 case SOCK_DGRAM:
1182                         return SECCLASS_UNIX_DGRAM_SOCKET;
1183                 }
1184                 break;
1185         case PF_INET:
1186         case PF_INET6:
1187                 switch (type) {
1188                 case SOCK_STREAM:
1189                         if (default_protocol_stream(protocol))
1190                                 return SECCLASS_TCP_SOCKET;
1191                         else
1192                                 return SECCLASS_RAWIP_SOCKET;
1193                 case SOCK_DGRAM:
1194                         if (default_protocol_dgram(protocol))
1195                                 return SECCLASS_UDP_SOCKET;
1196                         else
1197                                 return SECCLASS_RAWIP_SOCKET;
1198                 case SOCK_DCCP:
1199                         return SECCLASS_DCCP_SOCKET;
1200                 default:
1201                         return SECCLASS_RAWIP_SOCKET;
1202                 }
1203                 break;
1204         case PF_NETLINK:
1205                 switch (protocol) {
1206                 case NETLINK_ROUTE:
1207                         return SECCLASS_NETLINK_ROUTE_SOCKET;
1208                 case NETLINK_SOCK_DIAG:
1209                         return SECCLASS_NETLINK_TCPDIAG_SOCKET;
1210                 case NETLINK_NFLOG:
1211                         return SECCLASS_NETLINK_NFLOG_SOCKET;
1212                 case NETLINK_XFRM:
1213                         return SECCLASS_NETLINK_XFRM_SOCKET;
1214                 case NETLINK_SELINUX:
1215                         return SECCLASS_NETLINK_SELINUX_SOCKET;
1216                 case NETLINK_ISCSI:
1217                         return SECCLASS_NETLINK_ISCSI_SOCKET;
1218                 case NETLINK_AUDIT:
1219                         return SECCLASS_NETLINK_AUDIT_SOCKET;
1220                 case NETLINK_FIB_LOOKUP:
1221                         return SECCLASS_NETLINK_FIB_LOOKUP_SOCKET;
1222                 case NETLINK_CONNECTOR:
1223                         return SECCLASS_NETLINK_CONNECTOR_SOCKET;
1224                 case NETLINK_NETFILTER:
1225                         return SECCLASS_NETLINK_NETFILTER_SOCKET;
1226                 case NETLINK_DNRTMSG:
1227                         return SECCLASS_NETLINK_DNRT_SOCKET;
1228                 case NETLINK_KOBJECT_UEVENT:
1229                         return SECCLASS_NETLINK_KOBJECT_UEVENT_SOCKET;
1230                 case NETLINK_GENERIC:
1231                         return SECCLASS_NETLINK_GENERIC_SOCKET;
1232                 case NETLINK_SCSITRANSPORT:
1233                         return SECCLASS_NETLINK_SCSITRANSPORT_SOCKET;
1234                 case NETLINK_RDMA:
1235                         return SECCLASS_NETLINK_RDMA_SOCKET;
1236                 case NETLINK_CRYPTO:
1237                         return SECCLASS_NETLINK_CRYPTO_SOCKET;
1238                 default:
1239                         return SECCLASS_NETLINK_SOCKET;
1240                 }
1241         case PF_PACKET:
1242                 return SECCLASS_PACKET_SOCKET;
1243         case PF_KEY:
1244                 return SECCLASS_KEY_SOCKET;
1245         case PF_APPLETALK:
1246                 return SECCLASS_APPLETALK_SOCKET;
1247         }
1248
1249         return SECCLASS_SOCKET;
1250 }
1251
1252 static int selinux_genfs_get_sid(struct dentry *dentry,
1253                                  u16 tclass,
1254                                  u16 flags,
1255                                  u32 *sid)
1256 {
1257         int rc;
1258         struct super_block *sb = dentry->d_inode->i_sb;
1259         char *buffer, *path;
1260
1261         buffer = (char *)__get_free_page(GFP_KERNEL);
1262         if (!buffer)
1263                 return -ENOMEM;
1264
1265         path = dentry_path_raw(dentry, buffer, PAGE_SIZE);
1266         if (IS_ERR(path))
1267                 rc = PTR_ERR(path);
1268         else {
1269                 if (flags & SE_SBPROC) {
1270                         /* each process gets a /proc/PID/ entry. Strip off the
1271                          * PID part to get a valid selinux labeling.
1272                          * e.g. /proc/1/net/rpc/nfs -> /net/rpc/nfs */
1273                         while (path[1] >= '0' && path[1] <= '9') {
1274                                 path[1] = '/';
1275                                 path++;
1276                         }
1277                 }
1278                 rc = security_genfs_sid(sb->s_type->name, path, tclass, sid);
1279         }
1280         free_page((unsigned long)buffer);
1281         return rc;
1282 }
1283
1284 /* The inode's security attributes must be initialized before first use. */
1285 static int inode_doinit_with_dentry(struct inode *inode, struct dentry *opt_dentry)
1286 {
1287         struct superblock_security_struct *sbsec = NULL;
1288         struct inode_security_struct *isec = inode->i_security;
1289         u32 sid;
1290         struct dentry *dentry;
1291 #define INITCONTEXTLEN 255
1292         char *context = NULL;
1293         unsigned len = 0;
1294         int rc = 0;
1295
1296         if (isec->initialized)
1297                 goto out;
1298
1299         mutex_lock(&isec->lock);
1300         if (isec->initialized)
1301                 goto out_unlock;
1302
1303         sbsec = inode->i_sb->s_security;
1304         if (!(sbsec->flags & SE_SBINITIALIZED)) {
1305                 /* Defer initialization until selinux_complete_init,
1306                    after the initial policy is loaded and the security
1307                    server is ready to handle calls. */
1308                 spin_lock(&sbsec->isec_lock);
1309                 if (list_empty(&isec->list))
1310                         list_add(&isec->list, &sbsec->isec_head);
1311                 spin_unlock(&sbsec->isec_lock);
1312                 goto out_unlock;
1313         }
1314
1315         switch (sbsec->behavior) {
1316         case SECURITY_FS_USE_NATIVE:
1317                 break;
1318         case SECURITY_FS_USE_XATTR:
1319                 if (!inode->i_op->getxattr) {
1320                         isec->sid = sbsec->def_sid;
1321                         break;
1322                 }
1323
1324                 /* Need a dentry, since the xattr API requires one.
1325                    Life would be simpler if we could just pass the inode. */
1326                 if (opt_dentry) {
1327                         /* Called from d_instantiate or d_splice_alias. */
1328                         dentry = dget(opt_dentry);
1329                 } else {
1330                         /* Called from selinux_complete_init, try to find a dentry. */
1331                         dentry = d_find_alias(inode);
1332                 }
1333                 if (!dentry) {
1334                         /*
1335                          * this is can be hit on boot when a file is accessed
1336                          * before the policy is loaded.  When we load policy we
1337                          * may find inodes that have no dentry on the
1338                          * sbsec->isec_head list.  No reason to complain as these
1339                          * will get fixed up the next time we go through
1340                          * inode_doinit with a dentry, before these inodes could
1341                          * be used again by userspace.
1342                          */
1343                         goto out_unlock;
1344                 }
1345
1346                 len = INITCONTEXTLEN;
1347                 context = kmalloc(len+1, GFP_NOFS);
1348                 if (!context) {
1349                         rc = -ENOMEM;
1350                         dput(dentry);
1351                         goto out_unlock;
1352                 }
1353                 context[len] = '\0';
1354                 rc = inode->i_op->getxattr(dentry, XATTR_NAME_SELINUX,
1355                                            context, len);
1356                 if (rc == -ERANGE) {
1357                         kfree(context);
1358
1359                         /* Need a larger buffer.  Query for the right size. */
1360                         rc = inode->i_op->getxattr(dentry, XATTR_NAME_SELINUX,
1361                                                    NULL, 0);
1362                         if (rc < 0) {
1363                                 dput(dentry);
1364                                 goto out_unlock;
1365                         }
1366                         len = rc;
1367                         context = kmalloc(len+1, GFP_NOFS);
1368                         if (!context) {
1369                                 rc = -ENOMEM;
1370                                 dput(dentry);
1371                                 goto out_unlock;
1372                         }
1373                         context[len] = '\0';
1374                         rc = inode->i_op->getxattr(dentry,
1375                                                    XATTR_NAME_SELINUX,
1376                                                    context, len);
1377                 }
1378                 dput(dentry);
1379                 if (rc < 0) {
1380                         if (rc != -ENODATA) {
1381                                 printk(KERN_WARNING "SELinux: %s:  getxattr returned "
1382                                        "%d for dev=%s ino=%ld\n", __func__,
1383                                        -rc, inode->i_sb->s_id, inode->i_ino);
1384                                 kfree(context);
1385                                 goto out_unlock;
1386                         }
1387                         /* Map ENODATA to the default file SID */
1388                         sid = sbsec->def_sid;
1389                         rc = 0;
1390                 } else {
1391                         rc = security_context_to_sid_default(context, rc, &sid,
1392                                                              sbsec->def_sid,
1393                                                              GFP_NOFS);
1394                         if (rc) {
1395                                 char *dev = inode->i_sb->s_id;
1396                                 unsigned long ino = inode->i_ino;
1397
1398                                 if (rc == -EINVAL) {
1399                                         if (printk_ratelimit())
1400                                                 printk(KERN_NOTICE "SELinux: inode=%lu on dev=%s was found to have an invalid "
1401                                                         "context=%s.  This indicates you may need to relabel the inode or the "
1402                                                         "filesystem in question.\n", ino, dev, context);
1403                                 } else {
1404                                         printk(KERN_WARNING "SELinux: %s:  context_to_sid(%s) "
1405                                                "returned %d for dev=%s ino=%ld\n",
1406                                                __func__, context, -rc, dev, ino);
1407                                 }
1408                                 kfree(context);
1409                                 /* Leave with the unlabeled SID */
1410                                 rc = 0;
1411                                 break;
1412                         }
1413                 }
1414                 kfree(context);
1415                 isec->sid = sid;
1416                 break;
1417         case SECURITY_FS_USE_TASK:
1418                 isec->sid = isec->task_sid;
1419                 break;
1420         case SECURITY_FS_USE_TRANS:
1421                 /* Default to the fs SID. */
1422                 isec->sid = sbsec->sid;
1423
1424                 /* Try to obtain a transition SID. */
1425                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
1426                 rc = security_transition_sid(isec->task_sid, sbsec->sid,
1427                                              isec->sclass, NULL, &sid);
1428                 if (rc)
1429                         goto out_unlock;
1430                 isec->sid = sid;
1431                 break;
1432         case SECURITY_FS_USE_MNTPOINT:
1433                 isec->sid = sbsec->mntpoint_sid;
1434                 break;
1435         default:
1436                 /* Default to the fs superblock SID. */
1437                 isec->sid = sbsec->sid;
1438
1439                 if ((sbsec->flags & SE_SBGENFS) && !S_ISLNK(inode->i_mode)) {
1440                         /* We must have a dentry to determine the label on
1441                          * procfs inodes */
1442                         if (opt_dentry)
1443                                 /* Called from d_instantiate or
1444                                  * d_splice_alias. */
1445                                 dentry = dget(opt_dentry);
1446                         else
1447                                 /* Called from selinux_complete_init, try to
1448                                  * find a dentry. */
1449                                 dentry = d_find_alias(inode);
1450                         /*
1451                          * This can be hit on boot when a file is accessed
1452                          * before the policy is loaded.  When we load policy we
1453                          * may find inodes that have no dentry on the
1454                          * sbsec->isec_head list.  No reason to complain as
1455                          * these will get fixed up the next time we go through
1456                          * inode_doinit() with a dentry, before these inodes
1457                          * could be used again by userspace.
1458                          */
1459                         if (!dentry)
1460                                 goto out_unlock;
1461                         isec->sclass = inode_mode_to_security_class(inode->i_mode);
1462                         rc = selinux_genfs_get_sid(dentry, isec->sclass,
1463                                                    sbsec->flags, &sid);
1464                         dput(dentry);
1465                         if (rc)
1466                                 goto out_unlock;
1467                         isec->sid = sid;
1468                 }
1469                 break;
1470         }
1471
1472         isec->initialized = 1;
1473
1474 out_unlock:
1475         mutex_unlock(&isec->lock);
1476 out:
1477         if (isec->sclass == SECCLASS_FILE)
1478                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
1479         return rc;
1480 }
1481
1482 /* Convert a Linux signal to an access vector. */
1483 static inline u32 signal_to_av(int sig)
1484 {
1485         u32 perm = 0;
1486
1487         switch (sig) {
1488         case SIGCHLD:
1489                 /* Commonly granted from child to parent. */
1490                 perm = PROCESS__SIGCHLD;
1491                 break;
1492         case SIGKILL:
1493                 /* Cannot be caught or ignored */
1494                 perm = PROCESS__SIGKILL;
1495                 break;
1496         case SIGSTOP:
1497                 /* Cannot be caught or ignored */
1498                 perm = PROCESS__SIGSTOP;
1499                 break;
1500         default:
1501                 /* All other signals. */
1502                 perm = PROCESS__SIGNAL;
1503                 break;
1504         }
1505
1506         return perm;
1507 }
1508
1509 /*
1510  * Check permission between a pair of credentials
1511  * fork check, ptrace check, etc.
1512  */
1513 static int cred_has_perm(const struct cred *actor,
1514                          const struct cred *target,
1515                          u32 perms)
1516 {
1517         u32 asid = cred_sid(actor), tsid = cred_sid(target);
1518
1519         return avc_has_perm(asid, tsid, SECCLASS_PROCESS, perms, NULL);
1520 }
1521
1522 /*
1523  * Check permission between a pair of tasks, e.g. signal checks,
1524  * fork check, ptrace check, etc.
1525  * tsk1 is the actor and tsk2 is the target
1526  * - this uses the default subjective creds of tsk1
1527  */
1528 static int task_has_perm(const struct task_struct *tsk1,
1529                          const struct task_struct *tsk2,
1530                          u32 perms)
1531 {
1532         const struct task_security_struct *__tsec1, *__tsec2;
1533         u32 sid1, sid2;
1534
1535         rcu_read_lock();
1536         __tsec1 = __task_cred(tsk1)->security;  sid1 = __tsec1->sid;
1537         __tsec2 = __task_cred(tsk2)->security;  sid2 = __tsec2->sid;
1538         rcu_read_unlock();
1539         return avc_has_perm(sid1, sid2, SECCLASS_PROCESS, perms, NULL);
1540 }
1541
1542 /*
1543  * Check permission between current and another task, e.g. signal checks,
1544  * fork check, ptrace check, etc.
1545  * current is the actor and tsk2 is the target
1546  * - this uses current's subjective creds
1547  */
1548 static int current_has_perm(const struct task_struct *tsk,
1549                             u32 perms)
1550 {
1551         u32 sid, tsid;
1552
1553         sid = current_sid();
1554         tsid = task_sid(tsk);
1555         return avc_has_perm(sid, tsid, SECCLASS_PROCESS, perms, NULL);
1556 }
1557
1558 #if CAP_LAST_CAP > 63
1559 #error Fix SELinux to handle capabilities > 63.
1560 #endif
1561
1562 /* Check whether a task is allowed to use a capability. */
1563 static int cred_has_capability(const struct cred *cred,
1564                                int cap, int audit)
1565 {
1566         struct common_audit_data ad;
1567         struct av_decision avd;
1568         u16 sclass;
1569         u32 sid = cred_sid(cred);
1570         u32 av = CAP_TO_MASK(cap);
1571         int rc;
1572
1573         ad.type = LSM_AUDIT_DATA_CAP;
1574         ad.u.cap = cap;
1575
1576         switch (CAP_TO_INDEX(cap)) {
1577         case 0:
1578                 sclass = SECCLASS_CAPABILITY;
1579                 break;
1580         case 1:
1581                 sclass = SECCLASS_CAPABILITY2;
1582                 break;
1583         default:
1584                 printk(KERN_ERR
1585                        "SELinux:  out of range capability %d\n", cap);
1586                 BUG();
1587                 return -EINVAL;
1588         }
1589
1590         rc = avc_has_perm_noaudit(sid, sid, sclass, av, 0, &avd);
1591         if (audit == SECURITY_CAP_AUDIT) {
1592                 int rc2 = avc_audit(sid, sid, sclass, av, &avd, rc, &ad, 0);
1593                 if (rc2)
1594                         return rc2;
1595         }
1596         return rc;
1597 }
1598
1599 /* Check whether a task is allowed to use a system operation. */
1600 static int task_has_system(struct task_struct *tsk,
1601                            u32 perms)
1602 {
1603         u32 sid = task_sid(tsk);
1604
1605         return avc_has_perm(sid, SECINITSID_KERNEL,
1606                             SECCLASS_SYSTEM, perms, NULL);
1607 }
1608
1609 /* Check whether a task has a particular permission to an inode.
1610    The 'adp' parameter is optional and allows other audit
1611    data to be passed (e.g. the dentry). */
1612 static int inode_has_perm(const struct cred *cred,
1613                           struct inode *inode,
1614                           u32 perms,
1615                           struct common_audit_data *adp)
1616 {
1617         struct inode_security_struct *isec;
1618         u32 sid;
1619
1620         validate_creds(cred);
1621
1622         if (unlikely(IS_PRIVATE(inode)))
1623                 return 0;
1624
1625         sid = cred_sid(cred);
1626         isec = inode->i_security;
1627
1628         return avc_has_perm(sid, isec->sid, isec->sclass, perms, adp);
1629 }
1630
1631 /* Same as inode_has_perm, but pass explicit audit data containing
1632    the dentry to help the auditing code to more easily generate the
1633    pathname if needed. */
1634 static inline int dentry_has_perm(const struct cred *cred,
1635                                   struct dentry *dentry,
1636                                   u32 av)
1637 {
1638         struct inode *inode = d_backing_inode(dentry);
1639         struct common_audit_data ad;
1640
1641         ad.type = LSM_AUDIT_DATA_DENTRY;
1642         ad.u.dentry = dentry;
1643         return inode_has_perm(cred, inode, av, &ad);
1644 }
1645
1646 /* Same as inode_has_perm, but pass explicit audit data containing
1647    the path to help the auditing code to more easily generate the
1648    pathname if needed. */
1649 static inline int path_has_perm(const struct cred *cred,
1650                                 const struct path *path,
1651                                 u32 av)
1652 {
1653         struct inode *inode = d_backing_inode(path->dentry);
1654         struct common_audit_data ad;
1655
1656         ad.type = LSM_AUDIT_DATA_PATH;
1657         ad.u.path = *path;
1658         return inode_has_perm(cred, inode, av, &ad);
1659 }
1660
1661 /* Same as path_has_perm, but uses the inode from the file struct. */
1662 static inline int file_path_has_perm(const struct cred *cred,
1663                                      struct file *file,
1664                                      u32 av)
1665 {
1666         struct common_audit_data ad;
1667
1668         ad.type = LSM_AUDIT_DATA_PATH;
1669         ad.u.path = file->f_path;
1670         return inode_has_perm(cred, file_inode(file), av, &ad);
1671 }
1672
1673 /* Check whether a task can use an open file descriptor to
1674    access an inode in a given way.  Check access to the
1675    descriptor itself, and then use dentry_has_perm to
1676    check a particular permission to the file.
1677    Access to the descriptor is implicitly granted if it
1678    has the same SID as the process.  If av is zero, then
1679    access to the file is not checked, e.g. for cases
1680    where only the descriptor is affected like seek. */
1681 static int file_has_perm(const struct cred *cred,
1682                          struct file *file,
1683                          u32 av)
1684 {
1685         struct file_security_struct *fsec = file->f_security;
1686         struct inode *inode = file_inode(file);
1687         struct common_audit_data ad;
1688         u32 sid = cred_sid(cred);
1689         int rc;
1690
1691         ad.type = LSM_AUDIT_DATA_PATH;
1692         ad.u.path = file->f_path;
1693
1694         if (sid != fsec->sid) {
1695                 rc = avc_has_perm(sid, fsec->sid,
1696                                   SECCLASS_FD,
1697                                   FD__USE,
1698                                   &ad);
1699                 if (rc)
1700                         goto out;
1701         }
1702
1703         /* av is zero if only checking access to the descriptor. */
1704         rc = 0;
1705         if (av)
1706                 rc = inode_has_perm(cred, inode, av, &ad);
1707
1708 out:
1709         return rc;
1710 }
1711
1712 /*
1713  * Determine the label for an inode that might be unioned.
1714  */
1715 static int selinux_determine_inode_label(const struct inode *dir,
1716                                          const struct qstr *name,
1717                                          u16 tclass,
1718                                          u32 *_new_isid)
1719 {
1720         const struct superblock_security_struct *sbsec = dir->i_sb->s_security;
1721         const struct inode_security_struct *dsec = dir->i_security;
1722         const struct task_security_struct *tsec = current_security();
1723
1724         if ((sbsec->flags & SE_SBINITIALIZED) &&
1725             (sbsec->behavior == SECURITY_FS_USE_MNTPOINT)) {
1726                 *_new_isid = sbsec->mntpoint_sid;
1727         } else if ((sbsec->flags & SBLABEL_MNT) &&
1728                    tsec->create_sid) {
1729                 *_new_isid = tsec->create_sid;
1730         } else {
1731                 return security_transition_sid(tsec->sid, dsec->sid, tclass,
1732                                                name, _new_isid);
1733         }
1734
1735         return 0;
1736 }
1737
1738 /* Check whether a task can create a file. */
1739 static int may_create(struct inode *dir,
1740                       struct dentry *dentry,
1741                       u16 tclass)
1742 {
1743         const struct task_security_struct *tsec = current_security();
1744         struct inode_security_struct *dsec;
1745         struct superblock_security_struct *sbsec;
1746         u32 sid, newsid;
1747         struct common_audit_data ad;
1748         int rc;
1749
1750         dsec = dir->i_security;
1751         sbsec = dir->i_sb->s_security;
1752
1753         sid = tsec->sid;
1754
1755         ad.type = LSM_AUDIT_DATA_DENTRY;
1756         ad.u.dentry = dentry;
1757
1758         rc = avc_has_perm(sid, dsec->sid, SECCLASS_DIR,
1759                           DIR__ADD_NAME | DIR__SEARCH,
1760                           &ad);
1761         if (rc)
1762                 return rc;
1763
1764         rc = selinux_determine_inode_label(dir, &dentry->d_name, tclass,
1765                                            &newsid);
1766         if (rc)
1767                 return rc;
1768
1769         rc = avc_has_perm(sid, newsid, tclass, FILE__CREATE, &ad);
1770         if (rc)
1771                 return rc;
1772
1773         return avc_has_perm(newsid, sbsec->sid,
1774                             SECCLASS_FILESYSTEM,
1775                             FILESYSTEM__ASSOCIATE, &ad);
1776 }
1777
1778 /* Check whether a task can create a key. */
1779 static int may_create_key(u32 ksid,
1780                           struct task_struct *ctx)
1781 {
1782         u32 sid = task_sid(ctx);
1783
1784         return avc_has_perm(sid, ksid, SECCLASS_KEY, KEY__CREATE, NULL);
1785 }
1786
1787 #define MAY_LINK        0
1788 #define MAY_UNLINK      1
1789 #define MAY_RMDIR       2
1790
1791 /* Check whether a task can link, unlink, or rmdir a file/directory. */
1792 static int may_link(struct inode *dir,
1793                     struct dentry *dentry,
1794                     int kind)
1795
1796 {
1797         struct inode_security_struct *dsec, *isec;
1798         struct common_audit_data ad;
1799         u32 sid = current_sid();
1800         u32 av;
1801         int rc;
1802
1803         dsec = dir->i_security;
1804         isec = d_backing_inode(dentry)->i_security;
1805
1806         ad.type = LSM_AUDIT_DATA_DENTRY;
1807         ad.u.dentry = dentry;
1808
1809         av = DIR__SEARCH;
1810         av |= (kind ? DIR__REMOVE_NAME : DIR__ADD_NAME);
1811         rc = avc_has_perm(sid, dsec->sid, SECCLASS_DIR, av, &ad);
1812         if (rc)
1813                 return rc;
1814
1815         switch (kind) {
1816         case MAY_LINK:
1817                 av = FILE__LINK;
1818                 break;
1819         case MAY_UNLINK:
1820                 av = FILE__UNLINK;
1821                 break;
1822         case MAY_RMDIR:
1823                 av = DIR__RMDIR;
1824                 break;
1825         default:
1826                 printk(KERN_WARNING "SELinux: %s:  unrecognized kind %d\n",
1827                         __func__, kind);
1828                 return 0;
1829         }
1830
1831         rc = avc_has_perm(sid, isec->sid, isec->sclass, av, &ad);
1832         return rc;
1833 }
1834
1835 static inline int may_rename(struct inode *old_dir,
1836                              struct dentry *old_dentry,
1837                              struct inode *new_dir,
1838                              struct dentry *new_dentry)
1839 {
1840         struct inode_security_struct *old_dsec, *new_dsec, *old_isec, *new_isec;
1841         struct common_audit_data ad;
1842         u32 sid = current_sid();
1843         u32 av;
1844         int old_is_dir, new_is_dir;
1845         int rc;
1846
1847         old_dsec = old_dir->i_security;
1848         old_isec = d_backing_inode(old_dentry)->i_security;
1849         old_is_dir = d_is_dir(old_dentry);
1850         new_dsec = new_dir->i_security;
1851
1852         ad.type = LSM_AUDIT_DATA_DENTRY;
1853
1854         ad.u.dentry = old_dentry;
1855         rc = avc_has_perm(sid, old_dsec->sid, SECCLASS_DIR,
1856                           DIR__REMOVE_NAME | DIR__SEARCH, &ad);
1857         if (rc)
1858                 return rc;
1859         rc = avc_has_perm(sid, old_isec->sid,
1860                           old_isec->sclass, FILE__RENAME, &ad);
1861         if (rc)
1862                 return rc;
1863         if (old_is_dir && new_dir != old_dir) {
1864                 rc = avc_has_perm(sid, old_isec->sid,
1865                                   old_isec->sclass, DIR__REPARENT, &ad);
1866                 if (rc)
1867                         return rc;
1868         }
1869
1870         ad.u.dentry = new_dentry;
1871         av = DIR__ADD_NAME | DIR__SEARCH;
1872         if (d_is_positive(new_dentry))
1873                 av |= DIR__REMOVE_NAME;
1874         rc = avc_has_perm(sid, new_dsec->sid, SECCLASS_DIR, av, &ad);
1875         if (rc)
1876                 return rc;
1877         if (d_is_positive(new_dentry)) {
1878                 new_isec = d_backing_inode(new_dentry)->i_security;
1879                 new_is_dir = d_is_dir(new_dentry);
1880                 rc = avc_has_perm(sid, new_isec->sid,
1881                                   new_isec->sclass,
1882                                   (new_is_dir ? DIR__RMDIR : FILE__UNLINK), &ad);
1883                 if (rc)
1884                         return rc;
1885         }
1886
1887         return 0;
1888 }
1889
1890 /* Check whether a task can perform a filesystem operation. */
1891 static int superblock_has_perm(const struct cred *cred,
1892                                struct super_block *sb,
1893                                u32 perms,
1894                                struct common_audit_data *ad)
1895 {
1896         struct superblock_security_struct *sbsec;
1897         u32 sid = cred_sid(cred);
1898
1899         sbsec = sb->s_security;
1900         return avc_has_perm(sid, sbsec->sid, SECCLASS_FILESYSTEM, perms, ad);
1901 }
1902
1903 /* Convert a Linux mode and permission mask to an access vector. */
1904 static inline u32 file_mask_to_av(int mode, int mask)
1905 {
1906         u32 av = 0;
1907
1908         if (!S_ISDIR(mode)) {
1909                 if (mask & MAY_EXEC)
1910                         av |= FILE__EXECUTE;
1911                 if (mask & MAY_READ)
1912                         av |= FILE__READ;
1913
1914                 if (mask & MAY_APPEND)
1915                         av |= FILE__APPEND;
1916                 else if (mask & MAY_WRITE)
1917                         av |= FILE__WRITE;
1918
1919         } else {
1920                 if (mask & MAY_EXEC)
1921                         av |= DIR__SEARCH;
1922                 if (mask & MAY_WRITE)
1923                         av |= DIR__WRITE;
1924                 if (mask & MAY_READ)
1925                         av |= DIR__READ;
1926         }
1927
1928         return av;
1929 }
1930
1931 /* Convert a Linux file to an access vector. */
1932 static inline u32 file_to_av(struct file *file)
1933 {
1934         u32 av = 0;
1935
1936         if (file->f_mode & FMODE_READ)
1937                 av |= FILE__READ;
1938         if (file->f_mode & FMODE_WRITE) {
1939                 if (file->f_flags & O_APPEND)
1940                         av |= FILE__APPEND;
1941                 else
1942                         av |= FILE__WRITE;
1943         }
1944         if (!av) {
1945                 /*
1946                  * Special file opened with flags 3 for ioctl-only use.
1947                  */
1948                 av = FILE__IOCTL;
1949         }
1950
1951         return av;
1952 }
1953
1954 /*
1955  * Convert a file to an access vector and include the correct open
1956  * open permission.
1957  */
1958 static inline u32 open_file_to_av(struct file *file)
1959 {
1960         u32 av = file_to_av(file);
1961
1962         if (selinux_policycap_openperm)
1963                 av |= FILE__OPEN;
1964
1965         return av;
1966 }
1967
1968 /* Hook functions begin here. */
1969
1970 static int selinux_binder_set_context_mgr(struct task_struct *mgr)
1971 {
1972         u32 mysid = current_sid();
1973         u32 mgrsid = task_sid(mgr);
1974
1975         return avc_has_perm(mysid, mgrsid, SECCLASS_BINDER,
1976                             BINDER__SET_CONTEXT_MGR, NULL);
1977 }
1978
1979 static int selinux_binder_transaction(struct task_struct *from,
1980                                       struct task_struct *to)
1981 {
1982         u32 mysid = current_sid();
1983         u32 fromsid = task_sid(from);
1984         u32 tosid = task_sid(to);
1985         int rc;
1986
1987         if (mysid != fromsid) {
1988                 rc = avc_has_perm(mysid, fromsid, SECCLASS_BINDER,
1989                                   BINDER__IMPERSONATE, NULL);
1990                 if (rc)
1991                         return rc;
1992         }
1993
1994         return avc_has_perm(fromsid, tosid, SECCLASS_BINDER, BINDER__CALL,
1995                             NULL);
1996 }
1997
1998 static int selinux_binder_transfer_binder(struct task_struct *from,
1999                                           struct task_struct *to)
2000 {
2001         u32 fromsid = task_sid(from);
2002         u32 tosid = task_sid(to);
2003
2004         return avc_has_perm(fromsid, tosid, SECCLASS_BINDER, BINDER__TRANSFER,
2005                             NULL);
2006 }
2007
2008 static int selinux_binder_transfer_file(struct task_struct *from,
2009                                         struct task_struct *to,
2010                                         struct file *file)
2011 {
2012         u32 sid = task_sid(to);
2013         struct file_security_struct *fsec = file->f_security;
2014         struct inode *inode = d_backing_inode(file->f_path.dentry);
2015         struct inode_security_struct *isec = inode->i_security;
2016         struct common_audit_data ad;
2017         int rc;
2018
2019         ad.type = LSM_AUDIT_DATA_PATH;
2020         ad.u.path = file->f_path;
2021
2022         if (sid != fsec->sid) {
2023                 rc = avc_has_perm(sid, fsec->sid,
2024                                   SECCLASS_FD,
2025                                   FD__USE,
2026                                   &ad);
2027                 if (rc)
2028                         return rc;
2029         }
2030
2031         if (unlikely(IS_PRIVATE(inode)))
2032                 return 0;
2033
2034         return avc_has_perm(sid, isec->sid, isec->sclass, file_to_av(file),
2035                             &ad);
2036 }
2037
2038 static int selinux_ptrace_access_check(struct task_struct *child,
2039                                      unsigned int mode)
2040 {
2041         if (mode & PTRACE_MODE_READ) {
2042                 u32 sid = current_sid();
2043                 u32 csid = task_sid(child);
2044                 return avc_has_perm(sid, csid, SECCLASS_FILE, FILE__READ, NULL);
2045         }
2046
2047         return current_has_perm(child, PROCESS__PTRACE);
2048 }
2049
2050 static int selinux_ptrace_traceme(struct task_struct *parent)
2051 {
2052         return task_has_perm(parent, current, PROCESS__PTRACE);
2053 }
2054
2055 static int selinux_capget(struct task_struct *target, kernel_cap_t *effective,
2056                           kernel_cap_t *inheritable, kernel_cap_t *permitted)
2057 {
2058         return current_has_perm(target, PROCESS__GETCAP);
2059 }
2060
2061 static int selinux_capset(struct cred *new, const struct cred *old,
2062                           const kernel_cap_t *effective,
2063                           const kernel_cap_t *inheritable,
2064                           const kernel_cap_t *permitted)
2065 {
2066         return cred_has_perm(old, new, PROCESS__SETCAP);
2067 }
2068
2069 /*
2070  * (This comment used to live with the selinux_task_setuid hook,
2071  * which was removed).
2072  *
2073  * Since setuid only affects the current process, and since the SELinux
2074  * controls are not based on the Linux identity attributes, SELinux does not
2075  * need to control this operation.  However, SELinux does control the use of
2076  * the CAP_SETUID and CAP_SETGID capabilities using the capable hook.
2077  */
2078
2079 static int selinux_capable(const struct cred *cred, struct user_namespace *ns,
2080                            int cap, int audit)
2081 {
2082         return cred_has_capability(cred, cap, audit);
2083 }
2084
2085 static int selinux_quotactl(int cmds, int type, int id, struct super_block *sb)
2086 {
2087         const struct cred *cred = current_cred();
2088         int rc = 0;
2089
2090         if (!sb)
2091                 return 0;
2092
2093         switch (cmds) {
2094         case Q_SYNC:
2095         case Q_QUOTAON:
2096         case Q_QUOTAOFF:
2097         case Q_SETINFO:
2098         case Q_SETQUOTA:
2099                 rc = superblock_has_perm(cred, sb, FILESYSTEM__QUOTAMOD, NULL);
2100                 break;
2101         case Q_GETFMT:
2102         case Q_GETINFO:
2103         case Q_GETQUOTA:
2104                 rc = superblock_has_perm(cred, sb, FILESYSTEM__QUOTAGET, NULL);
2105                 break;
2106         default:
2107                 rc = 0;  /* let the kernel handle invalid cmds */
2108                 break;
2109         }
2110         return rc;
2111 }
2112
2113 static int selinux_quota_on(struct dentry *dentry)
2114 {
2115         const struct cred *cred = current_cred();
2116
2117         return dentry_has_perm(cred, dentry, FILE__QUOTAON);
2118 }
2119
2120 static int selinux_syslog(int type)
2121 {
2122         int rc;
2123
2124         switch (type) {
2125         case SYSLOG_ACTION_READ_ALL:    /* Read last kernel messages */
2126         case SYSLOG_ACTION_SIZE_BUFFER: /* Return size of the log buffer */
2127                 rc = task_has_system(current, SYSTEM__SYSLOG_READ);
2128                 break;
2129         case SYSLOG_ACTION_CONSOLE_OFF: /* Disable logging to console */
2130         case SYSLOG_ACTION_CONSOLE_ON:  /* Enable logging to console */
2131         /* Set level of messages printed to console */
2132         case SYSLOG_ACTION_CONSOLE_LEVEL:
2133                 rc = task_has_system(current, SYSTEM__SYSLOG_CONSOLE);
2134                 break;
2135         case SYSLOG_ACTION_CLOSE:       /* Close log */
2136         case SYSLOG_ACTION_OPEN:        /* Open log */
2137         case SYSLOG_ACTION_READ:        /* Read from log */
2138         case SYSLOG_ACTION_READ_CLEAR:  /* Read/clear last kernel messages */
2139         case SYSLOG_ACTION_CLEAR:       /* Clear ring buffer */
2140         default:
2141                 rc = task_has_system(current, SYSTEM__SYSLOG_MOD);
2142                 break;
2143         }
2144         return rc;
2145 }
2146
2147 /*
2148  * Check that a process has enough memory to allocate a new virtual
2149  * mapping. 0 means there is enough memory for the allocation to
2150  * succeed and -ENOMEM implies there is not.
2151  *
2152  * Do not audit the selinux permission check, as this is applied to all
2153  * processes that allocate mappings.
2154  */
2155 static int selinux_vm_enough_memory(struct mm_struct *mm, long pages)
2156 {
2157         int rc, cap_sys_admin = 0;
2158
2159         rc = cred_has_capability(current_cred(), CAP_SYS_ADMIN,
2160                                         SECURITY_CAP_NOAUDIT);
2161         if (rc == 0)
2162                 cap_sys_admin = 1;
2163
2164         return cap_sys_admin;
2165 }
2166
2167 /* binprm security operations */
2168
2169 static int check_nnp_nosuid(const struct linux_binprm *bprm,
2170                             const struct task_security_struct *old_tsec,
2171                             const struct task_security_struct *new_tsec)
2172 {
2173         int nnp = (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS);
2174         int nosuid = (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID);
2175         int rc;
2176
2177         if (!nnp && !nosuid)
2178                 return 0; /* neither NNP nor nosuid */
2179
2180         if (new_tsec->sid == old_tsec->sid)
2181                 return 0; /* No change in credentials */
2182
2183         /*
2184          * The only transitions we permit under NNP or nosuid
2185          * are transitions to bounded SIDs, i.e. SIDs that are
2186          * guaranteed to only be allowed a subset of the permissions
2187          * of the current SID.
2188          */
2189         rc = security_bounded_transition(old_tsec->sid, new_tsec->sid);
2190         if (rc) {
2191                 /*
2192                  * On failure, preserve the errno values for NNP vs nosuid.
2193                  * NNP:  Operation not permitted for caller.
2194                  * nosuid:  Permission denied to file.
2195                  */
2196                 if (nnp)
2197                         return -EPERM;
2198                 else
2199                         return -EACCES;
2200         }
2201         return 0;
2202 }
2203
2204 static int selinux_bprm_set_creds(struct linux_binprm *bprm)
2205 {
2206         const struct task_security_struct *old_tsec;
2207         struct task_security_struct *new_tsec;
2208         struct inode_security_struct *isec;
2209         struct common_audit_data ad;
2210         struct inode *inode = file_inode(bprm->file);
2211         int rc;
2212
2213         /* SELinux context only depends on initial program or script and not
2214          * the script interpreter */
2215         if (bprm->cred_prepared)
2216                 return 0;
2217
2218         old_tsec = current_security();
2219         new_tsec = bprm->cred->security;
2220         isec = inode->i_security;
2221
2222         /* Default to the current task SID. */
2223         new_tsec->sid = old_tsec->sid;
2224         new_tsec->osid = old_tsec->sid;
2225
2226         /* Reset fs, key, and sock SIDs on execve. */
2227         new_tsec->create_sid = 0;
2228         new_tsec->keycreate_sid = 0;
2229         new_tsec->sockcreate_sid = 0;
2230
2231         if (old_tsec->exec_sid) {
2232                 new_tsec->sid = old_tsec->exec_sid;
2233                 /* Reset exec SID on execve. */
2234                 new_tsec->exec_sid = 0;
2235
2236                 /* Fail on NNP or nosuid if not an allowed transition. */
2237                 rc = check_nnp_nosuid(bprm, old_tsec, new_tsec);
2238                 if (rc)
2239                         return rc;
2240         } else {
2241                 /* Check for a default transition on this program. */
2242                 rc = security_transition_sid(old_tsec->sid, isec->sid,
2243                                              SECCLASS_PROCESS, NULL,
2244                                              &new_tsec->sid);
2245                 if (rc)
2246                         return rc;
2247
2248                 /*
2249                  * Fallback to old SID on NNP or nosuid if not an allowed
2250                  * transition.
2251                  */
2252                 rc = check_nnp_nosuid(bprm, old_tsec, new_tsec);
2253                 if (rc)
2254                         new_tsec->sid = old_tsec->sid;
2255         }
2256
2257         ad.type = LSM_AUDIT_DATA_PATH;
2258         ad.u.path = bprm->file->f_path;
2259
2260         if (new_tsec->sid == old_tsec->sid) {
2261                 rc = avc_has_perm(old_tsec->sid, isec->sid,
2262                                   SECCLASS_FILE, FILE__EXECUTE_NO_TRANS, &ad);
2263                 if (rc)
2264                         return rc;
2265         } else {
2266                 /* Check permissions for the transition. */
2267                 rc = avc_has_perm(old_tsec->sid, new_tsec->sid,
2268                                   SECCLASS_PROCESS, PROCESS__TRANSITION, &ad);
2269                 if (rc)
2270                         return rc;
2271
2272                 rc = avc_has_perm(new_tsec->sid, isec->sid,
2273                                   SECCLASS_FILE, FILE__ENTRYPOINT, &ad);
2274                 if (rc)
2275                         return rc;
2276
2277                 /* Check for shared state */
2278                 if (bprm->unsafe & LSM_UNSAFE_SHARE) {
2279                         rc = avc_has_perm(old_tsec->sid, new_tsec->sid,
2280                                           SECCLASS_PROCESS, PROCESS__SHARE,
2281                                           NULL);
2282                         if (rc)
2283                                 return -EPERM;
2284                 }
2285
2286                 /* Make sure that anyone attempting to ptrace over a task that
2287                  * changes its SID has the appropriate permit */
2288                 if (bprm->unsafe &
2289                     (LSM_UNSAFE_PTRACE | LSM_UNSAFE_PTRACE_CAP)) {
2290                         struct task_struct *tracer;
2291                         struct task_security_struct *sec;
2292                         u32 ptsid = 0;
2293
2294                         rcu_read_lock();
2295                         tracer = ptrace_parent(current);
2296                         if (likely(tracer != NULL)) {
2297                                 sec = __task_cred(tracer)->security;
2298                                 ptsid = sec->sid;
2299                         }
2300                         rcu_read_unlock();
2301
2302                         if (ptsid != 0) {
2303                                 rc = avc_has_perm(ptsid, new_tsec->sid,
2304                                                   SECCLASS_PROCESS,
2305                                                   PROCESS__PTRACE, NULL);
2306                                 if (rc)
2307                                         return -EPERM;
2308                         }
2309                 }
2310
2311                 /* Clear any possibly unsafe personality bits on exec: */
2312                 bprm->per_clear |= PER_CLEAR_ON_SETID;
2313         }
2314
2315         return 0;
2316 }
2317
2318 static int selinux_bprm_secureexec(struct linux_binprm *bprm)
2319 {
2320         const struct task_security_struct *tsec = current_security();
2321         u32 sid, osid;
2322         int atsecure = 0;
2323
2324         sid = tsec->sid;
2325         osid = tsec->osid;
2326
2327         if (osid != sid) {
2328                 /* Enable secure mode for SIDs transitions unless
2329                    the noatsecure permission is granted between
2330                    the two SIDs, i.e. ahp returns 0. */
2331                 atsecure = avc_has_perm(osid, sid,
2332                                         SECCLASS_PROCESS,
2333                                         PROCESS__NOATSECURE, NULL);
2334         }
2335
2336         return !!atsecure;
2337 }
2338
2339 static int match_file(const void *p, struct file *file, unsigned fd)
2340 {
2341         return file_has_perm(p, file, file_to_av(file)) ? fd + 1 : 0;
2342 }
2343
2344 /* Derived from fs/exec.c:flush_old_files. */
2345 static inline void flush_unauthorized_files(const struct cred *cred,
2346                                             struct files_struct *files)
2347 {
2348         struct file *file, *devnull = NULL;
2349         struct tty_struct *tty;
2350         int drop_tty = 0;
2351         unsigned n;
2352
2353         tty = get_current_tty();
2354         if (tty) {
2355                 spin_lock(&tty_files_lock);
2356                 if (!list_empty(&tty->tty_files)) {
2357                         struct tty_file_private *file_priv;
2358
2359                         /* Revalidate access to controlling tty.
2360                            Use file_path_has_perm on the tty path directly
2361                            rather than using file_has_perm, as this particular
2362                            open file may belong to another process and we are
2363                            only interested in the inode-based check here. */
2364                         file_priv = list_first_entry(&tty->tty_files,
2365                                                 struct tty_file_private, list);
2366                         file = file_priv->file;
2367                         if (file_path_has_perm(cred, file, FILE__READ | FILE__WRITE))
2368                                 drop_tty = 1;
2369                 }
2370                 spin_unlock(&tty_files_lock);
2371                 tty_kref_put(tty);
2372         }
2373         /* Reset controlling tty. */
2374         if (drop_tty)
2375                 no_tty();
2376
2377         /* Revalidate access to inherited open files. */
2378         n = iterate_fd(files, 0, match_file, cred);
2379         if (!n) /* none found? */
2380                 return;
2381
2382         devnull = dentry_open(&selinux_null, O_RDWR, cred);
2383         if (IS_ERR(devnull))
2384                 devnull = NULL;
2385         /* replace all the matching ones with this */
2386         do {
2387                 replace_fd(n - 1, devnull, 0);
2388         } while ((n = iterate_fd(files, n, match_file, cred)) != 0);
2389         if (devnull)
2390                 fput(devnull);
2391 }
2392
2393 /*
2394  * Prepare a process for imminent new credential changes due to exec
2395  */
2396 static void selinux_bprm_committing_creds(struct linux_binprm *bprm)
2397 {
2398         struct task_security_struct *new_tsec;
2399         struct rlimit *rlim, *initrlim;
2400         int rc, i;
2401
2402         new_tsec = bprm->cred->security;
2403         if (new_tsec->sid == new_tsec->osid)
2404                 return;
2405
2406         /* Close files for which the new task SID is not authorized. */
2407         flush_unauthorized_files(bprm->cred, current->files);
2408
2409         /* Always clear parent death signal on SID transitions. */
2410         current->pdeath_signal = 0;
2411
2412         /* Check whether the new SID can inherit resource limits from the old
2413          * SID.  If not, reset all soft limits to the lower of the current
2414          * task's hard limit and the init task's soft limit.
2415          *
2416          * Note that the setting of hard limits (even to lower them) can be
2417          * controlled by the setrlimit check.  The inclusion of the init task's
2418          * soft limit into the computation is to avoid resetting soft limits
2419          * higher than the default soft limit for cases where the default is
2420          * lower than the hard limit, e.g. RLIMIT_CORE or RLIMIT_STACK.
2421          */
2422         rc = avc_has_perm(new_tsec->osid, new_tsec->sid, SECCLASS_PROCESS,
2423                           PROCESS__RLIMITINH, NULL);
2424         if (rc) {
2425                 /* protect against do_prlimit() */
2426                 task_lock(current);
2427                 for (i = 0; i < RLIM_NLIMITS; i++) {
2428                         rlim = current->signal->rlim + i;
2429                         initrlim = init_task.signal->rlim + i;
2430                         rlim->rlim_cur = min(rlim->rlim_max, initrlim->rlim_cur);
2431                 }
2432                 task_unlock(current);
2433                 update_rlimit_cpu(current, rlimit(RLIMIT_CPU));
2434         }
2435 }
2436
2437 /*
2438  * Clean up the process immediately after the installation of new credentials
2439  * due to exec
2440  */
2441 static void selinux_bprm_committed_creds(struct linux_binprm *bprm)
2442 {
2443         const struct task_security_struct *tsec = current_security();
2444         struct itimerval itimer;
2445         u32 osid, sid;
2446         int rc, i;
2447
2448         osid = tsec->osid;
2449         sid = tsec->sid;
2450
2451         if (sid == osid)
2452                 return;
2453
2454         /* Check whether the new SID can inherit signal state from the old SID.
2455          * If not, clear itimers to avoid subsequent signal generation and
2456          * flush and unblock signals.
2457          *
2458          * This must occur _after_ the task SID has been updated so that any
2459          * kill done after the flush will be checked against the new SID.
2460          */
2461         rc = avc_has_perm(osid, sid, SECCLASS_PROCESS, PROCESS__SIGINH, NULL);
2462         if (rc) {
2463                 memset(&itimer, 0, sizeof itimer);
2464                 for (i = 0; i < 3; i++)
2465                         do_setitimer(i, &itimer, NULL);
2466                 spin_lock_irq(&current->sighand->siglock);
2467                 if (!fatal_signal_pending(current)) {
2468                         flush_sigqueue(&current->pending);
2469                         flush_sigqueue(&current->signal->shared_pending);
2470                         flush_signal_handlers(current, 1);
2471                         sigemptyset(&current->blocked);
2472                         recalc_sigpending();
2473                 }
2474                 spin_unlock_irq(&current->sighand->siglock);
2475         }
2476
2477         /* Wake up the parent if it is waiting so that it can recheck
2478          * wait permission to the new task SID. */
2479         read_lock(&tasklist_lock);
2480         __wake_up_parent(current, current->real_parent);
2481         read_unlock(&tasklist_lock);
2482 }
2483
2484 /* superblock security operations */
2485
2486 static int selinux_sb_alloc_security(struct super_block *sb)
2487 {
2488         return superblock_alloc_security(sb);
2489 }
2490
2491 static void selinux_sb_free_security(struct super_block *sb)
2492 {
2493         superblock_free_security(sb);
2494 }
2495
2496 static inline int match_prefix(char *prefix, int plen, char *option, int olen)
2497 {
2498         if (plen > olen)
2499                 return 0;
2500
2501         return !memcmp(prefix, option, plen);
2502 }
2503
2504 static inline int selinux_option(char *option, int len)
2505 {
2506         return (match_prefix(CONTEXT_STR, sizeof(CONTEXT_STR)-1, option, len) ||
2507                 match_prefix(FSCONTEXT_STR, sizeof(FSCONTEXT_STR)-1, option, len) ||
2508                 match_prefix(DEFCONTEXT_STR, sizeof(DEFCONTEXT_STR)-1, option, len) ||
2509                 match_prefix(ROOTCONTEXT_STR, sizeof(ROOTCONTEXT_STR)-1, option, len) ||
2510                 match_prefix(LABELSUPP_STR, sizeof(LABELSUPP_STR)-1, option, len));
2511 }
2512
2513 static inline void take_option(char **to, char *from, int *first, int len)
2514 {
2515         if (!*first) {
2516                 **to = ',';
2517                 *to += 1;
2518         } else
2519                 *first = 0;
2520         memcpy(*to, from, len);
2521         *to += len;
2522 }
2523
2524 static inline void take_selinux_option(char **to, char *from, int *first,
2525                                        int len)
2526 {
2527         int current_size = 0;
2528
2529         if (!*first) {
2530                 **to = '|';
2531                 *to += 1;
2532         } else
2533                 *first = 0;
2534
2535         while (current_size < len) {
2536                 if (*from != '"') {
2537                         **to = *from;
2538                         *to += 1;
2539                 }
2540                 from += 1;
2541                 current_size += 1;
2542         }
2543 }
2544
2545 static int selinux_sb_copy_data(char *orig, char *copy)
2546 {
2547         int fnosec, fsec, rc = 0;
2548         char *in_save, *in_curr, *in_end;
2549         char *sec_curr, *nosec_save, *nosec;
2550         int open_quote = 0;
2551
2552         in_curr = orig;
2553         sec_curr = copy;
2554
2555         nosec = (char *)get_zeroed_page(GFP_KERNEL);
2556         if (!nosec) {
2557                 rc = -ENOMEM;
2558                 goto out;
2559         }
2560
2561         nosec_save = nosec;
2562         fnosec = fsec = 1;
2563         in_save = in_end = orig;
2564
2565         do {
2566                 if (*in_end == '"')
2567                         open_quote = !open_quote;
2568                 if ((*in_end == ',' && open_quote == 0) ||
2569                                 *in_end == '\0') {
2570                         int len = in_end - in_curr;
2571
2572                         if (selinux_option(in_curr, len))
2573                                 take_selinux_option(&sec_curr, in_curr, &fsec, len);
2574                         else
2575                                 take_option(&nosec, in_curr, &fnosec, len);
2576
2577                         in_curr = in_end + 1;
2578                 }
2579         } while (*in_end++);
2580
2581         strcpy(in_save, nosec_save);
2582         free_page((unsigned long)nosec_save);
2583 out:
2584         return rc;
2585 }
2586
2587 static int selinux_sb_remount(struct super_block *sb, void *data)
2588 {
2589         int rc, i, *flags;
2590         struct security_mnt_opts opts;
2591         char *secdata, **mount_options;
2592         struct superblock_security_struct *sbsec = sb->s_security;
2593
2594         if (!(sbsec->flags & SE_SBINITIALIZED))
2595                 return 0;
2596
2597         if (!data)
2598                 return 0;
2599
2600         if (sb->s_type->fs_flags & FS_BINARY_MOUNTDATA)
2601                 return 0;
2602
2603         security_init_mnt_opts(&opts);
2604         secdata = alloc_secdata();
2605         if (!secdata)
2606                 return -ENOMEM;
2607         rc = selinux_sb_copy_data(data, secdata);
2608         if (rc)
2609                 goto out_free_secdata;
2610
2611         rc = selinux_parse_opts_str(secdata, &opts);
2612         if (rc)
2613                 goto out_free_secdata;
2614
2615         mount_options = opts.mnt_opts;
2616         flags = opts.mnt_opts_flags;
2617
2618         for (i = 0; i < opts.num_mnt_opts; i++) {
2619                 u32 sid;
2620                 size_t len;
2621
2622                 if (flags[i] == SBLABEL_MNT)
2623                         continue;
2624                 len = strlen(mount_options[i]);
2625                 rc = security_context_to_sid(mount_options[i], len, &sid,
2626                                              GFP_KERNEL);
2627                 if (rc) {
2628                         printk(KERN_WARNING "SELinux: security_context_to_sid"
2629                                "(%s) failed for (dev %s, type %s) errno=%d\n",
2630                                mount_options[i], sb->s_id, sb->s_type->name, rc);
2631                         goto out_free_opts;
2632                 }
2633                 rc = -EINVAL;
2634                 switch (flags[i]) {
2635                 case FSCONTEXT_MNT:
2636                         if (bad_option(sbsec, FSCONTEXT_MNT, sbsec->sid, sid))
2637                                 goto out_bad_option;
2638                         break;
2639                 case CONTEXT_MNT:
2640                         if (bad_option(sbsec, CONTEXT_MNT, sbsec->mntpoint_sid, sid))
2641                                 goto out_bad_option;
2642                         break;
2643                 case ROOTCONTEXT_MNT: {
2644                         struct inode_security_struct *root_isec;
2645                         root_isec = d_backing_inode(sb->s_root)->i_security;
2646
2647                         if (bad_option(sbsec, ROOTCONTEXT_MNT, root_isec->sid, sid))
2648                                 goto out_bad_option;
2649                         break;
2650                 }
2651                 case DEFCONTEXT_MNT:
2652                         if (bad_option(sbsec, DEFCONTEXT_MNT, sbsec->def_sid, sid))
2653                                 goto out_bad_option;
2654                         break;
2655                 default:
2656                         goto out_free_opts;
2657                 }
2658         }
2659
2660         rc = 0;
2661 out_free_opts:
2662         security_free_mnt_opts(&opts);
2663 out_free_secdata:
2664         free_secdata(secdata);
2665         return rc;
2666 out_bad_option:
2667         printk(KERN_WARNING "SELinux: unable to change security options "
2668                "during remount (dev %s, type=%s)\n", sb->s_id,
2669                sb->s_type->name);
2670         goto out_free_opts;
2671 }
2672
2673 static int selinux_sb_kern_mount(struct super_block *sb, int flags, void *data)
2674 {
2675         const struct cred *cred = current_cred();
2676         struct common_audit_data ad;
2677         int rc;
2678
2679         rc = superblock_doinit(sb, data);
2680         if (rc)
2681                 return rc;
2682
2683         /* Allow all mounts performed by the kernel */
2684         if (flags & MS_KERNMOUNT)
2685                 return 0;
2686
2687         ad.type = LSM_AUDIT_DATA_DENTRY;
2688         ad.u.dentry = sb->s_root;
2689         return superblock_has_perm(cred, sb, FILESYSTEM__MOUNT, &ad);
2690 }
2691
2692 static int selinux_sb_statfs(struct dentry *dentry)
2693 {
2694         const struct cred *cred = current_cred();
2695         struct common_audit_data ad;
2696
2697         ad.type = LSM_AUDIT_DATA_DENTRY;
2698         ad.u.dentry = dentry->d_sb->s_root;
2699         return superblock_has_perm(cred, dentry->d_sb, FILESYSTEM__GETATTR, &ad);
2700 }
2701
2702 static int selinux_mount(const char *dev_name,
2703                          struct path *path,
2704                          const char *type,
2705                          unsigned long flags,
2706                          void *data)
2707 {
2708         const struct cred *cred = current_cred();
2709
2710         if (flags & MS_REMOUNT)
2711                 return superblock_has_perm(cred, path->dentry->d_sb,
2712                                            FILESYSTEM__REMOUNT, NULL);
2713         else
2714                 return path_has_perm(cred, path, FILE__MOUNTON);
2715 }
2716
2717 static int selinux_umount(struct vfsmount *mnt, int flags)
2718 {
2719         const struct cred *cred = current_cred();
2720
2721         return superblock_has_perm(cred, mnt->mnt_sb,
2722                                    FILESYSTEM__UNMOUNT, NULL);
2723 }
2724
2725 /* inode security operations */
2726
2727 static int selinux_inode_alloc_security(struct inode *inode)
2728 {
2729         return inode_alloc_security(inode);
2730 }
2731
2732 static void selinux_inode_free_security(struct inode *inode)
2733 {
2734         inode_free_security(inode);
2735 }
2736
2737 static int selinux_dentry_init_security(struct dentry *dentry, int mode,
2738                                         struct qstr *name, void **ctx,
2739                                         u32 *ctxlen)
2740 {
2741         u32 newsid;
2742         int rc;
2743
2744         rc = selinux_determine_inode_label(d_inode(dentry->d_parent), name,
2745                                            inode_mode_to_security_class(mode),
2746                                            &newsid);
2747         if (rc)
2748                 return rc;
2749
2750         return security_sid_to_context(newsid, (char **)ctx, ctxlen);
2751 }
2752
2753 static int selinux_inode_init_security(struct inode *inode, struct inode *dir,
2754                                        const struct qstr *qstr,
2755                                        const char **name,
2756                                        void **value, size_t *len)
2757 {
2758         const struct task_security_struct *tsec = current_security();
2759         struct inode_security_struct *dsec;
2760         struct superblock_security_struct *sbsec;
2761         u32 sid, newsid, clen;
2762         int rc;
2763         char *context;
2764
2765         dsec = dir->i_security;
2766         sbsec = dir->i_sb->s_security;
2767
2768         sid = tsec->sid;
2769         newsid = tsec->create_sid;
2770
2771         rc = selinux_determine_inode_label(
2772                 dir, qstr,
2773                 inode_mode_to_security_class(inode->i_mode),
2774                 &newsid);
2775         if (rc)
2776                 return rc;
2777
2778         /* Possibly defer initialization to selinux_complete_init. */
2779         if (sbsec->flags & SE_SBINITIALIZED) {
2780                 struct inode_security_struct *isec = inode->i_security;
2781                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
2782                 isec->sid = newsid;
2783                 isec->initialized = 1;
2784         }
2785
2786         if (!ss_initialized || !(sbsec->flags & SBLABEL_MNT))
2787                 return -EOPNOTSUPP;
2788
2789         if (name)
2790                 *name = XATTR_SELINUX_SUFFIX;
2791
2792         if (value && len) {
2793                 rc = security_sid_to_context_force(newsid, &context, &clen);
2794                 if (rc)
2795                         return rc;
2796                 *value = context;
2797                 *len = clen;
2798         }
2799
2800         return 0;
2801 }
2802
2803 static int selinux_inode_create(struct inode *dir, struct dentry *dentry, umode_t mode)
2804 {
2805         return may_create(dir, dentry, SECCLASS_FILE);
2806 }
2807
2808 static int selinux_inode_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2809 {
2810         return may_link(dir, old_dentry, MAY_LINK);
2811 }
2812
2813 static int selinux_inode_unlink(struct inode *dir, struct dentry *dentry)
2814 {
2815         return may_link(dir, dentry, MAY_UNLINK);
2816 }
2817
2818 static int selinux_inode_symlink(struct inode *dir, struct dentry *dentry, const char *name)
2819 {
2820         return may_create(dir, dentry, SECCLASS_LNK_FILE);
2821 }
2822
2823 static int selinux_inode_mkdir(struct inode *dir, struct dentry *dentry, umode_t mask)
2824 {
2825         return may_create(dir, dentry, SECCLASS_DIR);
2826 }
2827
2828 static int selinux_inode_rmdir(struct inode *dir, struct dentry *dentry)
2829 {
2830         return may_link(dir, dentry, MAY_RMDIR);
2831 }
2832
2833 static int selinux_inode_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2834 {
2835         return may_create(dir, dentry, inode_mode_to_security_class(mode));
2836 }
2837
2838 static int selinux_inode_rename(struct inode *old_inode, struct dentry *old_dentry,
2839                                 struct inode *new_inode, struct dentry *new_dentry)
2840 {
2841         return may_rename(old_inode, old_dentry, new_inode, new_dentry);
2842 }
2843
2844 static int selinux_inode_readlink(struct dentry *dentry)
2845 {
2846         const struct cred *cred = current_cred();
2847
2848         return dentry_has_perm(cred, dentry, FILE__READ);
2849 }
2850
2851 static int selinux_inode_follow_link(struct dentry *dentry, struct inode *inode,
2852                                      bool rcu)
2853 {
2854         const struct cred *cred = current_cred();
2855         struct common_audit_data ad;
2856         struct inode_security_struct *isec;
2857         u32 sid;
2858
2859         validate_creds(cred);
2860
2861         ad.type = LSM_AUDIT_DATA_DENTRY;
2862         ad.u.dentry = dentry;
2863         sid = cred_sid(cred);
2864         isec = inode->i_security;
2865
2866         return avc_has_perm_flags(sid, isec->sid, isec->sclass, FILE__READ, &ad,
2867                                   rcu ? MAY_NOT_BLOCK : 0);
2868 }
2869
2870 static noinline int audit_inode_permission(struct inode *inode,
2871                                            u32 perms, u32 audited, u32 denied,
2872                                            int result,
2873                                            unsigned flags)
2874 {
2875         struct common_audit_data ad;
2876         struct inode_security_struct *isec = inode->i_security;
2877         int rc;
2878
2879         ad.type = LSM_AUDIT_DATA_INODE;
2880         ad.u.inode = inode;
2881
2882         rc = slow_avc_audit(current_sid(), isec->sid, isec->sclass, perms,
2883                             audited, denied, result, &ad, flags);
2884         if (rc)
2885                 return rc;
2886         return 0;
2887 }
2888
2889 static int selinux_inode_permission(struct inode *inode, int mask)
2890 {
2891         const struct cred *cred = current_cred();
2892         u32 perms;
2893         bool from_access;
2894         unsigned flags = mask & MAY_NOT_BLOCK;
2895         struct inode_security_struct *isec;
2896         u32 sid;
2897         struct av_decision avd;
2898         int rc, rc2;
2899         u32 audited, denied;
2900
2901         from_access = mask & MAY_ACCESS;
2902         mask &= (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND);
2903
2904         /* No permission to check.  Existence test. */
2905         if (!mask)
2906                 return 0;
2907
2908         validate_creds(cred);
2909
2910         if (unlikely(IS_PRIVATE(inode)))
2911                 return 0;
2912
2913         perms = file_mask_to_av(inode->i_mode, mask);
2914
2915         sid = cred_sid(cred);
2916         isec = inode->i_security;
2917
2918         rc = avc_has_perm_noaudit(sid, isec->sid, isec->sclass, perms, 0, &avd);
2919         audited = avc_audit_required(perms, &avd, rc,
2920                                      from_access ? FILE__AUDIT_ACCESS : 0,
2921                                      &denied);
2922         if (likely(!audited))
2923                 return rc;
2924
2925         rc2 = audit_inode_permission(inode, perms, audited, denied, rc, flags);
2926         if (rc2)
2927                 return rc2;
2928         return rc;
2929 }
2930
2931 static int selinux_inode_setattr(struct dentry *dentry, struct iattr *iattr)
2932 {
2933         const struct cred *cred = current_cred();
2934         unsigned int ia_valid = iattr->ia_valid;
2935         __u32 av = FILE__WRITE;
2936
2937         /* ATTR_FORCE is just used for ATTR_KILL_S[UG]ID. */
2938         if (ia_valid & ATTR_FORCE) {
2939                 ia_valid &= ~(ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_MODE |
2940                               ATTR_FORCE);
2941                 if (!ia_valid)
2942                         return 0;
2943         }
2944
2945         if (ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID |
2946                         ATTR_ATIME_SET | ATTR_MTIME_SET | ATTR_TIMES_SET))
2947                 return dentry_has_perm(cred, dentry, FILE__SETATTR);
2948
2949         if (selinux_policycap_openperm && (ia_valid & ATTR_SIZE))
2950                 av |= FILE__OPEN;
2951
2952         return dentry_has_perm(cred, dentry, av);
2953 }
2954
2955 static int selinux_inode_getattr(const struct path *path)
2956 {
2957         return path_has_perm(current_cred(), path, FILE__GETATTR);
2958 }
2959
2960 static int selinux_inode_setotherxattr(struct dentry *dentry, const char *name)
2961 {
2962         const struct cred *cred = current_cred();
2963
2964         if (!strncmp(name, XATTR_SECURITY_PREFIX,
2965                      sizeof XATTR_SECURITY_PREFIX - 1)) {
2966                 if (!strcmp(name, XATTR_NAME_CAPS)) {
2967                         if (!capable(CAP_SETFCAP))
2968                                 return -EPERM;
2969                 } else if (!capable(CAP_SYS_ADMIN)) {
2970                         /* A different attribute in the security namespace.
2971                            Restrict to administrator. */
2972                         return -EPERM;
2973                 }
2974         }
2975
2976         /* Not an attribute we recognize, so just check the
2977            ordinary setattr permission. */
2978         return dentry_has_perm(cred, dentry, FILE__SETATTR);
2979 }
2980
2981 static int selinux_inode_setxattr(struct dentry *dentry, const char *name,
2982                                   const void *value, size_t size, int flags)
2983 {
2984         struct inode *inode = d_backing_inode(dentry);
2985         struct inode_security_struct *isec = inode->i_security;
2986         struct superblock_security_struct *sbsec;
2987         struct common_audit_data ad;
2988         u32 newsid, sid = current_sid();
2989         int rc = 0;
2990
2991         if (strcmp(name, XATTR_NAME_SELINUX))
2992                 return selinux_inode_setotherxattr(dentry, name);
2993
2994         sbsec = inode->i_sb->s_security;
2995         if (!(sbsec->flags & SBLABEL_MNT))
2996                 return -EOPNOTSUPP;
2997
2998         if (!inode_owner_or_capable(inode))
2999                 return -EPERM;
3000
3001         ad.type = LSM_AUDIT_DATA_DENTRY;
3002         ad.u.dentry = dentry;
3003
3004         rc = avc_has_perm(sid, isec->sid, isec->sclass,
3005                           FILE__RELABELFROM, &ad);
3006         if (rc)
3007                 return rc;
3008
3009         rc = security_context_to_sid(value, size, &newsid, GFP_KERNEL);
3010         if (rc == -EINVAL) {
3011                 if (!capable(CAP_MAC_ADMIN)) {
3012                         struct audit_buffer *ab;
3013                         size_t audit_size;
3014                         const char *str;
3015
3016                         /* We strip a nul only if it is at the end, otherwise the
3017                          * context contains a nul and we should audit that */
3018                         if (value) {
3019                                 str = value;
3020                                 if (str[size - 1] == '\0')
3021                                         audit_size = size - 1;
3022                                 else
3023                                         audit_size = size;
3024                         } else {
3025                                 str = "";
3026                                 audit_size = 0;
3027                         }
3028                         ab = audit_log_start(current->audit_context, GFP_ATOMIC, AUDIT_SELINUX_ERR);
3029                         audit_log_format(ab, "op=setxattr invalid_context=");
3030                         audit_log_n_untrustedstring(ab, value, audit_size);
3031                         audit_log_end(ab);
3032
3033                         return rc;
3034                 }
3035                 rc = security_context_to_sid_force(value, size, &newsid);
3036         }
3037         if (rc)
3038                 return rc;
3039
3040         rc = avc_has_perm(sid, newsid, isec->sclass,
3041                           FILE__RELABELTO, &ad);
3042         if (rc)
3043                 return rc;
3044
3045         rc = security_validate_transition(isec->sid, newsid, sid,
3046                                           isec->sclass);
3047         if (rc)
3048                 return rc;
3049
3050         return avc_has_perm(newsid,
3051                             sbsec->sid,
3052                             SECCLASS_FILESYSTEM,
3053                             FILESYSTEM__ASSOCIATE,
3054                             &ad);
3055 }
3056
3057 static void selinux_inode_post_setxattr(struct dentry *dentry, const char *name,
3058                                         const void *value, size_t size,
3059                                         int flags)
3060 {
3061         struct inode *inode = d_backing_inode(dentry);
3062         struct inode_security_struct *isec = inode->i_security;
3063         u32 newsid;
3064         int rc;
3065
3066         if (strcmp(name, XATTR_NAME_SELINUX)) {
3067                 /* Not an attribute we recognize, so nothing to do. */
3068                 return;
3069         }
3070
3071         rc = security_context_to_sid_force(value, size, &newsid);
3072         if (rc) {
3073                 printk(KERN_ERR "SELinux:  unable to map context to SID"
3074                        "for (%s, %lu), rc=%d\n",
3075                        inode->i_sb->s_id, inode->i_ino, -rc);
3076                 return;
3077         }
3078
3079         isec->sclass = inode_mode_to_security_class(inode->i_mode);
3080         isec->sid = newsid;
3081         isec->initialized = 1;
3082
3083         return;
3084 }
3085
3086 static int selinux_inode_getxattr(struct dentry *dentry, const char *name)
3087 {
3088         const struct cred *cred = current_cred();
3089
3090         return dentry_has_perm(cred, dentry, FILE__GETATTR);
3091 }
3092
3093 static int selinux_inode_listxattr(struct dentry *dentry)
3094 {
3095         const struct cred *cred = current_cred();
3096
3097         return dentry_has_perm(cred, dentry, FILE__GETATTR);
3098 }
3099
3100 static int selinux_inode_removexattr(struct dentry *dentry, const char *name)
3101 {
3102         if (strcmp(name, XATTR_NAME_SELINUX))
3103                 return selinux_inode_setotherxattr(dentry, name);
3104
3105         /* No one is allowed to remove a SELinux security label.
3106            You can change the label, but all data must be labeled. */
3107         return -EACCES;
3108 }
3109
3110 /*
3111  * Copy the inode security context value to the user.
3112  *
3113  * Permission check is handled by selinux_inode_getxattr hook.
3114  */
3115 static int selinux_inode_getsecurity(const struct inode *inode, const char *name, void **buffer, bool alloc)
3116 {
3117         u32 size;
3118         int error;
3119         char *context = NULL;
3120         struct inode_security_struct *isec = inode->i_security;
3121
3122         if (strcmp(name, XATTR_SELINUX_SUFFIX))
3123                 return -EOPNOTSUPP;
3124
3125         /*
3126          * If the caller has CAP_MAC_ADMIN, then get the raw context
3127          * value even if it is not defined by current policy; otherwise,
3128          * use the in-core value under current policy.
3129          * Use the non-auditing forms of the permission checks since
3130          * getxattr may be called by unprivileged processes commonly
3131          * and lack of permission just means that we fall back to the
3132          * in-core context value, not a denial.
3133          */
3134         error = cap_capable(current_cred(), &init_user_ns, CAP_MAC_ADMIN,
3135                             SECURITY_CAP_NOAUDIT);
3136         if (!error)
3137                 error = cred_has_capability(current_cred(), CAP_MAC_ADMIN,
3138                                             SECURITY_CAP_NOAUDIT);
3139         if (!error)
3140                 error = security_sid_to_context_force(isec->sid, &context,
3141                                                       &size);
3142         else
3143                 error = security_sid_to_context(isec->sid, &context, &size);
3144         if (error)
3145                 return error;
3146         error = size;
3147         if (alloc) {
3148                 *buffer = context;
3149                 goto out_nofree;
3150         }
3151         kfree(context);
3152 out_nofree:
3153         return error;
3154 }
3155
3156 static int selinux_inode_setsecurity(struct inode *inode, const char *name,
3157                                      const void *value, size_t size, int flags)
3158 {
3159         struct inode_security_struct *isec = inode->i_security;
3160         u32 newsid;
3161         int rc;
3162
3163         if (strcmp(name, XATTR_SELINUX_SUFFIX))
3164                 return -EOPNOTSUPP;
3165
3166         if (!value || !size)
3167                 return -EACCES;
3168
3169         rc = security_context_to_sid((void *)value, size, &newsid, GFP_KERNEL);
3170         if (rc)
3171                 return rc;
3172
3173         isec->sclass = inode_mode_to_security_class(inode->i_mode);
3174         isec->sid = newsid;
3175         isec->initialized = 1;
3176         return 0;
3177 }
3178
3179 static int selinux_inode_listsecurity(struct inode *inode, char *buffer, size_t buffer_size)
3180 {
3181         const int len = sizeof(XATTR_NAME_SELINUX);
3182         if (buffer && len <= buffer_size)
3183                 memcpy(buffer, XATTR_NAME_SELINUX, len);
3184         return len;
3185 }
3186
3187 static void selinux_inode_getsecid(const struct inode *inode, u32 *secid)
3188 {
3189         struct inode_security_struct *isec = inode->i_security;
3190         *secid = isec->sid;
3191 }
3192
3193 /* file security operations */
3194
3195 static int selinux_revalidate_file_permission(struct file *file, int mask)
3196 {
3197         const struct cred *cred = current_cred();
3198         struct inode *inode = file_inode(file);
3199
3200         /* file_mask_to_av won't add FILE__WRITE if MAY_APPEND is set */
3201         if ((file->f_flags & O_APPEND) && (mask & MAY_WRITE))
3202                 mask |= MAY_APPEND;
3203
3204         return file_has_perm(cred, file,
3205                              file_mask_to_av(inode->i_mode, mask));
3206 }
3207
3208 static int selinux_file_permission(struct file *file, int mask)
3209 {
3210         struct inode *inode = file_inode(file);
3211         struct file_security_struct *fsec = file->f_security;
3212         struct inode_security_struct *isec = inode->i_security;
3213         u32 sid = current_sid();
3214
3215         if (!mask)
3216                 /* No permission to check.  Existence test. */
3217                 return 0;
3218
3219         if (sid == fsec->sid && fsec->isid == isec->sid &&
3220             fsec->pseqno == avc_policy_seqno())
3221                 /* No change since file_open check. */
3222                 return 0;
3223
3224         return selinux_revalidate_file_permission(file, mask);
3225 }
3226
3227 static int selinux_file_alloc_security(struct file *file)
3228 {
3229         return file_alloc_security(file);
3230 }
3231
3232 static void selinux_file_free_security(struct file *file)
3233 {
3234         file_free_security(file);
3235 }
3236
3237 /*
3238  * Check whether a task has the ioctl permission and cmd
3239  * operation to an inode.
3240  */
3241 int ioctl_has_perm(const struct cred *cred, struct file *file,
3242                 u32 requested, u16 cmd)
3243 {
3244         struct common_audit_data ad;
3245         struct file_security_struct *fsec = file->f_security;
3246         struct inode *inode = file_inode(file);
3247         struct inode_security_struct *isec = inode->i_security;
3248         struct lsm_ioctlop_audit ioctl;
3249         u32 ssid = cred_sid(cred);
3250         int rc;
3251         u8 driver = cmd >> 8;
3252         u8 xperm = cmd & 0xff;
3253
3254         ad.type = LSM_AUDIT_DATA_IOCTL_OP;
3255         ad.u.op = &ioctl;
3256         ad.u.op->cmd = cmd;
3257         ad.u.op->path = file->f_path;
3258
3259         if (ssid != fsec->sid) {
3260                 rc = avc_has_perm(ssid, fsec->sid,
3261                                 SECCLASS_FD,
3262                                 FD__USE,
3263                                 &ad);
3264                 if (rc)
3265                         goto out;
3266         }
3267
3268         if (unlikely(IS_PRIVATE(inode)))
3269                 return 0;
3270
3271         rc = avc_has_extended_perms(ssid, isec->sid, isec->sclass,
3272                         requested, driver, xperm, &ad);
3273 out:
3274         return rc;
3275 }
3276
3277 static int selinux_file_ioctl(struct file *file, unsigned int cmd,
3278                               unsigned long arg)
3279 {
3280         const struct cred *cred = current_cred();
3281         int error = 0;
3282
3283         switch (cmd) {
3284         case FIONREAD:
3285         /* fall through */
3286         case FIBMAP:
3287         /* fall through */
3288         case FIGETBSZ:
3289         /* fall through */
3290         case FS_IOC_GETFLAGS:
3291         /* fall through */
3292         case FS_IOC_GETVERSION:
3293                 error = file_has_perm(cred, file, FILE__GETATTR);
3294                 break;
3295
3296         case FS_IOC_SETFLAGS:
3297         /* fall through */
3298         case FS_IOC_SETVERSION:
3299                 error = file_has_perm(cred, file, FILE__SETATTR);
3300                 break;
3301
3302         /* sys_ioctl() checks */
3303         case FIONBIO:
3304         /* fall through */
3305         case FIOASYNC:
3306                 error = file_has_perm(cred, file, 0);
3307                 break;
3308
3309         case KDSKBENT:
3310         case KDSKBSENT:
3311                 error = cred_has_capability(cred, CAP_SYS_TTY_CONFIG,
3312                                             SECURITY_CAP_AUDIT);
3313                 break;
3314
3315         /* default case assumes that the command will go
3316          * to the file's ioctl() function.
3317          */
3318         default:
3319                 error = ioctl_has_perm(cred, file, FILE__IOCTL, (u16) cmd);
3320         }
3321         return error;
3322 }
3323
3324 static int default_noexec;
3325
3326 static int file_map_prot_check(struct file *file, unsigned long prot, int shared)
3327 {
3328         const struct cred *cred = current_cred();
3329         int rc = 0;
3330
3331         if (default_noexec &&
3332             (prot & PROT_EXEC) && (!file || IS_PRIVATE(file_inode(file)) ||
3333                                    (!shared && (prot & PROT_WRITE)))) {
3334                 /*
3335                  * We are making executable an anonymous mapping or a
3336                  * private file mapping that will also be writable.
3337                  * This has an additional check.
3338                  */
3339                 rc = cred_has_perm(cred, cred, PROCESS__EXECMEM);
3340                 if (rc)
3341                         goto error;
3342         }
3343
3344