2b5ee5fbd652de22a951d7618e6daad3d7cb82f5
[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@tycho.nsa.gov>
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  *  Copyright (C) 2016 Mellanox Technologies
21  *
22  *      This program is free software; you can redistribute it and/or modify
23  *      it under the terms of the GNU General Public License version 2,
24  *      as published by the Free Software Foundation.
25  */
26
27 #include <linux/init.h>
28 #include <linux/kd.h>
29 #include <linux/kernel.h>
30 #include <linux/tracehook.h>
31 #include <linux/errno.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/task.h>
34 #include <linux/lsm_hooks.h>
35 #include <linux/xattr.h>
36 #include <linux/capability.h>
37 #include <linux/unistd.h>
38 #include <linux/mm.h>
39 #include <linux/mman.h>
40 #include <linux/slab.h>
41 #include <linux/pagemap.h>
42 #include <linux/proc_fs.h>
43 #include <linux/swap.h>
44 #include <linux/spinlock.h>
45 #include <linux/syscalls.h>
46 #include <linux/dcache.h>
47 #include <linux/file.h>
48 #include <linux/fdtable.h>
49 #include <linux/namei.h>
50 #include <linux/mount.h>
51 #include <linux/netfilter_ipv4.h>
52 #include <linux/netfilter_ipv6.h>
53 #include <linux/tty.h>
54 #include <net/icmp.h>
55 #include <net/ip.h>             /* for local_port_range[] */
56 #include <net/tcp.h>            /* struct or_callable used in sock_rcv_skb */
57 #include <net/inet_connection_sock.h>
58 #include <net/net_namespace.h>
59 #include <net/netlabel.h>
60 #include <linux/uaccess.h>
61 #include <asm/ioctls.h>
62 #include <linux/atomic.h>
63 #include <linux/bitops.h>
64 #include <linux/interrupt.h>
65 #include <linux/netdevice.h>    /* for network interface checks */
66 #include <net/netlink.h>
67 #include <linux/tcp.h>
68 #include <linux/udp.h>
69 #include <linux/dccp.h>
70 #include <linux/sctp.h>
71 #include <net/sctp/structs.h>
72 #include <linux/quota.h>
73 #include <linux/un.h>           /* for Unix socket types */
74 #include <net/af_unix.h>        /* for Unix socket types */
75 #include <linux/parser.h>
76 #include <linux/nfs_mount.h>
77 #include <net/ipv6.h>
78 #include <linux/hugetlb.h>
79 #include <linux/personality.h>
80 #include <linux/audit.h>
81 #include <linux/string.h>
82 #include <linux/selinux.h>
83 #include <linux/mutex.h>
84 #include <linux/posix-timers.h>
85 #include <linux/syslog.h>
86 #include <linux/user_namespace.h>
87 #include <linux/export.h>
88 #include <linux/msg.h>
89 #include <linux/shm.h>
90 #include <linux/bpf.h>
91
92 #include "avc.h"
93 #include "objsec.h"
94 #include "netif.h"
95 #include "netnode.h"
96 #include "netport.h"
97 #include "ibpkey.h"
98 #include "xfrm.h"
99 #include "netlabel.h"
100 #include "audit.h"
101 #include "avc_ss.h"
102
103 struct selinux_state selinux_state;
104
105 /* SECMARK reference count */
106 static atomic_t selinux_secmark_refcount = ATOMIC_INIT(0);
107
108 #ifdef CONFIG_SECURITY_SELINUX_DEVELOP
109 static int selinux_enforcing_boot;
110
111 static int __init enforcing_setup(char *str)
112 {
113         unsigned long enforcing;
114         if (!kstrtoul(str, 0, &enforcing))
115                 selinux_enforcing_boot = enforcing ? 1 : 0;
116         return 1;
117 }
118 __setup("enforcing=", enforcing_setup);
119 #else
120 #define selinux_enforcing_boot 1
121 #endif
122
123 #ifdef CONFIG_SECURITY_SELINUX_BOOTPARAM
124 int selinux_enabled = CONFIG_SECURITY_SELINUX_BOOTPARAM_VALUE;
125
126 static int __init selinux_enabled_setup(char *str)
127 {
128         unsigned long enabled;
129         if (!kstrtoul(str, 0, &enabled))
130                 selinux_enabled = enabled ? 1 : 0;
131         return 1;
132 }
133 __setup("selinux=", selinux_enabled_setup);
134 #else
135 int selinux_enabled = 1;
136 #endif
137
138 static unsigned int selinux_checkreqprot_boot =
139         CONFIG_SECURITY_SELINUX_CHECKREQPROT_VALUE;
140
141 static int __init checkreqprot_setup(char *str)
142 {
143         unsigned long checkreqprot;
144
145         if (!kstrtoul(str, 0, &checkreqprot))
146                 selinux_checkreqprot_boot = checkreqprot ? 1 : 0;
147         return 1;
148 }
149 __setup("checkreqprot=", checkreqprot_setup);
150
151 static struct kmem_cache *sel_inode_cache;
152 static struct kmem_cache *file_security_cache;
153
154 /**
155  * selinux_secmark_enabled - Check to see if SECMARK is currently enabled
156  *
157  * Description:
158  * This function checks the SECMARK reference counter to see if any SECMARK
159  * targets are currently configured, if the reference counter is greater than
160  * zero SECMARK is considered to be enabled.  Returns true (1) if SECMARK is
161  * enabled, false (0) if SECMARK is disabled.  If the always_check_network
162  * policy capability is enabled, SECMARK is always considered enabled.
163  *
164  */
165 static int selinux_secmark_enabled(void)
166 {
167         return (selinux_policycap_alwaysnetwork() ||
168                 atomic_read(&selinux_secmark_refcount));
169 }
170
171 /**
172  * selinux_peerlbl_enabled - Check to see if peer labeling is currently enabled
173  *
174  * Description:
175  * This function checks if NetLabel or labeled IPSEC is enabled.  Returns true
176  * (1) if any are enabled or false (0) if neither are enabled.  If the
177  * always_check_network policy capability is enabled, peer labeling
178  * is always considered enabled.
179  *
180  */
181 static int selinux_peerlbl_enabled(void)
182 {
183         return (selinux_policycap_alwaysnetwork() ||
184                 netlbl_enabled() || selinux_xfrm_enabled());
185 }
186
187 static int selinux_netcache_avc_callback(u32 event)
188 {
189         if (event == AVC_CALLBACK_RESET) {
190                 sel_netif_flush();
191                 sel_netnode_flush();
192                 sel_netport_flush();
193                 synchronize_net();
194         }
195         return 0;
196 }
197
198 static int selinux_lsm_notifier_avc_callback(u32 event)
199 {
200         if (event == AVC_CALLBACK_RESET) {
201                 sel_ib_pkey_flush();
202                 call_lsm_notifier(LSM_POLICY_CHANGE, NULL);
203         }
204
205         return 0;
206 }
207
208 /*
209  * initialise the security for the init task
210  */
211 static void cred_init_security(void)
212 {
213         struct cred *cred = (struct cred *) current->real_cred;
214         struct task_security_struct *tsec;
215
216         tsec = kzalloc(sizeof(struct task_security_struct), GFP_KERNEL);
217         if (!tsec)
218                 panic("SELinux:  Failed to initialize initial task.\n");
219
220         tsec->osid = tsec->sid = SECINITSID_KERNEL;
221         cred->security = tsec;
222 }
223
224 /*
225  * get the security ID of a set of credentials
226  */
227 static inline u32 cred_sid(const struct cred *cred)
228 {
229         const struct task_security_struct *tsec;
230
231         tsec = cred->security;
232         return tsec->sid;
233 }
234
235 /*
236  * get the objective security ID of a task
237  */
238 static inline u32 task_sid(const struct task_struct *task)
239 {
240         u32 sid;
241
242         rcu_read_lock();
243         sid = cred_sid(__task_cred(task));
244         rcu_read_unlock();
245         return sid;
246 }
247
248 /* Allocate and free functions for each kind of security blob. */
249
250 static int inode_alloc_security(struct inode *inode)
251 {
252         struct inode_security_struct *isec;
253         u32 sid = current_sid();
254
255         isec = kmem_cache_zalloc(sel_inode_cache, GFP_NOFS);
256         if (!isec)
257                 return -ENOMEM;
258
259         spin_lock_init(&isec->lock);
260         INIT_LIST_HEAD(&isec->list);
261         isec->inode = inode;
262         isec->sid = SECINITSID_UNLABELED;
263         isec->sclass = SECCLASS_FILE;
264         isec->task_sid = sid;
265         isec->initialized = LABEL_INVALID;
266         inode->i_security = isec;
267
268         return 0;
269 }
270
271 static int inode_doinit_with_dentry(struct inode *inode, struct dentry *opt_dentry);
272
273 /*
274  * Try reloading inode security labels that have been marked as invalid.  The
275  * @may_sleep parameter indicates when sleeping and thus reloading labels is
276  * allowed; when set to false, returns -ECHILD when the label is
277  * invalid.  The @dentry parameter should be set to a dentry of the inode.
278  */
279 static int __inode_security_revalidate(struct inode *inode,
280                                        struct dentry *dentry,
281                                        bool may_sleep)
282 {
283         struct inode_security_struct *isec = inode->i_security;
284
285         might_sleep_if(may_sleep);
286
287         if (selinux_state.initialized &&
288             isec->initialized != LABEL_INITIALIZED) {
289                 if (!may_sleep)
290                         return -ECHILD;
291
292                 /*
293                  * Try reloading the inode security label.  This will fail if
294                  * @opt_dentry is NULL and no dentry for this inode can be
295                  * found; in that case, continue using the old label.
296                  */
297                 inode_doinit_with_dentry(inode, dentry);
298         }
299         return 0;
300 }
301
302 static struct inode_security_struct *inode_security_novalidate(struct inode *inode)
303 {
304         return inode->i_security;
305 }
306
307 static struct inode_security_struct *inode_security_rcu(struct inode *inode, bool rcu)
308 {
309         int error;
310
311         error = __inode_security_revalidate(inode, NULL, !rcu);
312         if (error)
313                 return ERR_PTR(error);
314         return inode->i_security;
315 }
316
317 /*
318  * Get the security label of an inode.
319  */
320 static struct inode_security_struct *inode_security(struct inode *inode)
321 {
322         __inode_security_revalidate(inode, NULL, true);
323         return inode->i_security;
324 }
325
326 static struct inode_security_struct *backing_inode_security_novalidate(struct dentry *dentry)
327 {
328         struct inode *inode = d_backing_inode(dentry);
329
330         return inode->i_security;
331 }
332
333 /*
334  * Get the security label of a dentry's backing inode.
335  */
336 static struct inode_security_struct *backing_inode_security(struct dentry *dentry)
337 {
338         struct inode *inode = d_backing_inode(dentry);
339
340         __inode_security_revalidate(inode, dentry, true);
341         return inode->i_security;
342 }
343
344 static void inode_free_rcu(struct rcu_head *head)
345 {
346         struct inode_security_struct *isec;
347
348         isec = container_of(head, struct inode_security_struct, rcu);
349         kmem_cache_free(sel_inode_cache, isec);
350 }
351
352 static void inode_free_security(struct inode *inode)
353 {
354         struct inode_security_struct *isec = inode->i_security;
355         struct superblock_security_struct *sbsec = inode->i_sb->s_security;
356
357         /*
358          * As not all inode security structures are in a list, we check for
359          * empty list outside of the lock to make sure that we won't waste
360          * time taking a lock doing nothing.
361          *
362          * The list_del_init() function can be safely called more than once.
363          * It should not be possible for this function to be called with
364          * concurrent list_add(), but for better safety against future changes
365          * in the code, we use list_empty_careful() here.
366          */
367         if (!list_empty_careful(&isec->list)) {
368                 spin_lock(&sbsec->isec_lock);
369                 list_del_init(&isec->list);
370                 spin_unlock(&sbsec->isec_lock);
371         }
372
373         /*
374          * The inode may still be referenced in a path walk and
375          * a call to selinux_inode_permission() can be made
376          * after inode_free_security() is called. Ideally, the VFS
377          * wouldn't do this, but fixing that is a much harder
378          * job. For now, simply free the i_security via RCU, and
379          * leave the current inode->i_security pointer intact.
380          * The inode will be freed after the RCU grace period too.
381          */
382         call_rcu(&isec->rcu, inode_free_rcu);
383 }
384
385 static int file_alloc_security(struct file *file)
386 {
387         struct file_security_struct *fsec;
388         u32 sid = current_sid();
389
390         fsec = kmem_cache_zalloc(file_security_cache, GFP_KERNEL);
391         if (!fsec)
392                 return -ENOMEM;
393
394         fsec->sid = sid;
395         fsec->fown_sid = sid;
396         file->f_security = fsec;
397
398         return 0;
399 }
400
401 static void file_free_security(struct file *file)
402 {
403         struct file_security_struct *fsec = file->f_security;
404         file->f_security = NULL;
405         kmem_cache_free(file_security_cache, fsec);
406 }
407
408 static int superblock_alloc_security(struct super_block *sb)
409 {
410         struct superblock_security_struct *sbsec;
411
412         sbsec = kzalloc(sizeof(struct superblock_security_struct), GFP_KERNEL);
413         if (!sbsec)
414                 return -ENOMEM;
415
416         mutex_init(&sbsec->lock);
417         INIT_LIST_HEAD(&sbsec->isec_head);
418         spin_lock_init(&sbsec->isec_lock);
419         sbsec->sb = sb;
420         sbsec->sid = SECINITSID_UNLABELED;
421         sbsec->def_sid = SECINITSID_FILE;
422         sbsec->mntpoint_sid = SECINITSID_UNLABELED;
423         sb->s_security = sbsec;
424
425         return 0;
426 }
427
428 static void superblock_free_security(struct super_block *sb)
429 {
430         struct superblock_security_struct *sbsec = sb->s_security;
431         sb->s_security = NULL;
432         kfree(sbsec);
433 }
434
435 static inline int inode_doinit(struct inode *inode)
436 {
437         return inode_doinit_with_dentry(inode, NULL);
438 }
439
440 enum {
441         Opt_error = -1,
442         Opt_context = 1,
443         Opt_fscontext = 2,
444         Opt_defcontext = 3,
445         Opt_rootcontext = 4,
446         Opt_labelsupport = 5,
447         Opt_nextmntopt = 6,
448 };
449
450 #define NUM_SEL_MNT_OPTS        (Opt_nextmntopt - 1)
451
452 static const match_table_t tokens = {
453         {Opt_context, CONTEXT_STR "%s"},
454         {Opt_fscontext, FSCONTEXT_STR "%s"},
455         {Opt_defcontext, DEFCONTEXT_STR "%s"},
456         {Opt_rootcontext, ROOTCONTEXT_STR "%s"},
457         {Opt_labelsupport, LABELSUPP_STR},
458         {Opt_error, NULL},
459 };
460
461 #define SEL_MOUNT_FAIL_MSG "SELinux:  duplicate or incompatible mount options\n"
462
463 static int may_context_mount_sb_relabel(u32 sid,
464                         struct superblock_security_struct *sbsec,
465                         const struct cred *cred)
466 {
467         const struct task_security_struct *tsec = cred->security;
468         int rc;
469
470         rc = avc_has_perm(&selinux_state,
471                           tsec->sid, sbsec->sid, SECCLASS_FILESYSTEM,
472                           FILESYSTEM__RELABELFROM, NULL);
473         if (rc)
474                 return rc;
475
476         rc = avc_has_perm(&selinux_state,
477                           tsec->sid, sid, SECCLASS_FILESYSTEM,
478                           FILESYSTEM__RELABELTO, NULL);
479         return rc;
480 }
481
482 static int may_context_mount_inode_relabel(u32 sid,
483                         struct superblock_security_struct *sbsec,
484                         const struct cred *cred)
485 {
486         const struct task_security_struct *tsec = cred->security;
487         int rc;
488         rc = avc_has_perm(&selinux_state,
489                           tsec->sid, sbsec->sid, SECCLASS_FILESYSTEM,
490                           FILESYSTEM__RELABELFROM, NULL);
491         if (rc)
492                 return rc;
493
494         rc = avc_has_perm(&selinux_state,
495                           sid, sbsec->sid, SECCLASS_FILESYSTEM,
496                           FILESYSTEM__ASSOCIATE, NULL);
497         return rc;
498 }
499
500 static int selinux_is_sblabel_mnt(struct super_block *sb)
501 {
502         struct superblock_security_struct *sbsec = sb->s_security;
503
504         return sbsec->behavior == SECURITY_FS_USE_XATTR ||
505                 sbsec->behavior == SECURITY_FS_USE_TRANS ||
506                 sbsec->behavior == SECURITY_FS_USE_TASK ||
507                 sbsec->behavior == SECURITY_FS_USE_NATIVE ||
508                 /* Special handling. Genfs but also in-core setxattr handler */
509                 !strcmp(sb->s_type->name, "sysfs") ||
510                 !strcmp(sb->s_type->name, "pstore") ||
511                 !strcmp(sb->s_type->name, "debugfs") ||
512                 !strcmp(sb->s_type->name, "tracefs") ||
513                 !strcmp(sb->s_type->name, "rootfs") ||
514                 (selinux_policycap_cgroupseclabel() &&
515                  (!strcmp(sb->s_type->name, "cgroup") ||
516                   !strcmp(sb->s_type->name, "cgroup2")));
517 }
518
519 static int sb_finish_set_opts(struct super_block *sb)
520 {
521         struct superblock_security_struct *sbsec = sb->s_security;
522         struct dentry *root = sb->s_root;
523         struct inode *root_inode = d_backing_inode(root);
524         int rc = 0;
525
526         if (sbsec->behavior == SECURITY_FS_USE_XATTR) {
527                 /* Make sure that the xattr handler exists and that no
528                    error other than -ENODATA is returned by getxattr on
529                    the root directory.  -ENODATA is ok, as this may be
530                    the first boot of the SELinux kernel before we have
531                    assigned xattr values to the filesystem. */
532                 if (!(root_inode->i_opflags & IOP_XATTR)) {
533                         printk(KERN_WARNING "SELinux: (dev %s, type %s) has no "
534                                "xattr support\n", sb->s_id, sb->s_type->name);
535                         rc = -EOPNOTSUPP;
536                         goto out;
537                 }
538
539                 rc = __vfs_getxattr(root, root_inode, XATTR_NAME_SELINUX, NULL, 0);
540                 if (rc < 0 && rc != -ENODATA) {
541                         if (rc == -EOPNOTSUPP)
542                                 printk(KERN_WARNING "SELinux: (dev %s, type "
543                                        "%s) has no security xattr handler\n",
544                                        sb->s_id, sb->s_type->name);
545                         else
546                                 printk(KERN_WARNING "SELinux: (dev %s, type "
547                                        "%s) getxattr errno %d\n", sb->s_id,
548                                        sb->s_type->name, -rc);
549                         goto out;
550                 }
551         }
552
553         sbsec->flags |= SE_SBINITIALIZED;
554
555         /*
556          * Explicitly set or clear SBLABEL_MNT.  It's not sufficient to simply
557          * leave the flag untouched because sb_clone_mnt_opts might be handing
558          * us a superblock that needs the flag to be cleared.
559          */
560         if (selinux_is_sblabel_mnt(sb))
561                 sbsec->flags |= SBLABEL_MNT;
562         else
563                 sbsec->flags &= ~SBLABEL_MNT;
564
565         /* Initialize the root inode. */
566         rc = inode_doinit_with_dentry(root_inode, root);
567
568         /* Initialize any other inodes associated with the superblock, e.g.
569            inodes created prior to initial policy load or inodes created
570            during get_sb by a pseudo filesystem that directly
571            populates itself. */
572         spin_lock(&sbsec->isec_lock);
573 next_inode:
574         if (!list_empty(&sbsec->isec_head)) {
575                 struct inode_security_struct *isec =
576                                 list_entry(sbsec->isec_head.next,
577                                            struct inode_security_struct, list);
578                 struct inode *inode = isec->inode;
579                 list_del_init(&isec->list);
580                 spin_unlock(&sbsec->isec_lock);
581                 inode = igrab(inode);
582                 if (inode) {
583                         if (!IS_PRIVATE(inode))
584                                 inode_doinit(inode);
585                         iput(inode);
586                 }
587                 spin_lock(&sbsec->isec_lock);
588                 goto next_inode;
589         }
590         spin_unlock(&sbsec->isec_lock);
591 out:
592         return rc;
593 }
594
595 /*
596  * This function should allow an FS to ask what it's mount security
597  * options were so it can use those later for submounts, displaying
598  * mount options, or whatever.
599  */
600 static int selinux_get_mnt_opts(const struct super_block *sb,
601                                 struct security_mnt_opts *opts)
602 {
603         int rc = 0, i;
604         struct superblock_security_struct *sbsec = sb->s_security;
605         char *context = NULL;
606         u32 len;
607         char tmp;
608
609         security_init_mnt_opts(opts);
610
611         if (!(sbsec->flags & SE_SBINITIALIZED))
612                 return -EINVAL;
613
614         if (!selinux_state.initialized)
615                 return -EINVAL;
616
617         /* make sure we always check enough bits to cover the mask */
618         BUILD_BUG_ON(SE_MNTMASK >= (1 << NUM_SEL_MNT_OPTS));
619
620         tmp = sbsec->flags & SE_MNTMASK;
621         /* count the number of mount options for this sb */
622         for (i = 0; i < NUM_SEL_MNT_OPTS; i++) {
623                 if (tmp & 0x01)
624                         opts->num_mnt_opts++;
625                 tmp >>= 1;
626         }
627         /* Check if the Label support flag is set */
628         if (sbsec->flags & SBLABEL_MNT)
629                 opts->num_mnt_opts++;
630
631         opts->mnt_opts = kcalloc(opts->num_mnt_opts, sizeof(char *), GFP_ATOMIC);
632         if (!opts->mnt_opts) {
633                 rc = -ENOMEM;
634                 goto out_free;
635         }
636
637         opts->mnt_opts_flags = kcalloc(opts->num_mnt_opts, sizeof(int), GFP_ATOMIC);
638         if (!opts->mnt_opts_flags) {
639                 rc = -ENOMEM;
640                 goto out_free;
641         }
642
643         i = 0;
644         if (sbsec->flags & FSCONTEXT_MNT) {
645                 rc = security_sid_to_context(&selinux_state, sbsec->sid,
646                                              &context, &len);
647                 if (rc)
648                         goto out_free;
649                 opts->mnt_opts[i] = context;
650                 opts->mnt_opts_flags[i++] = FSCONTEXT_MNT;
651         }
652         if (sbsec->flags & CONTEXT_MNT) {
653                 rc = security_sid_to_context(&selinux_state,
654                                              sbsec->mntpoint_sid,
655                                              &context, &len);
656                 if (rc)
657                         goto out_free;
658                 opts->mnt_opts[i] = context;
659                 opts->mnt_opts_flags[i++] = CONTEXT_MNT;
660         }
661         if (sbsec->flags & DEFCONTEXT_MNT) {
662                 rc = security_sid_to_context(&selinux_state, sbsec->def_sid,
663                                              &context, &len);
664                 if (rc)
665                         goto out_free;
666                 opts->mnt_opts[i] = context;
667                 opts->mnt_opts_flags[i++] = DEFCONTEXT_MNT;
668         }
669         if (sbsec->flags & ROOTCONTEXT_MNT) {
670                 struct dentry *root = sbsec->sb->s_root;
671                 struct inode_security_struct *isec = backing_inode_security(root);
672
673                 rc = security_sid_to_context(&selinux_state, isec->sid,
674                                              &context, &len);
675                 if (rc)
676                         goto out_free;
677                 opts->mnt_opts[i] = context;
678                 opts->mnt_opts_flags[i++] = ROOTCONTEXT_MNT;
679         }
680         if (sbsec->flags & SBLABEL_MNT) {
681                 opts->mnt_opts[i] = NULL;
682                 opts->mnt_opts_flags[i++] = SBLABEL_MNT;
683         }
684
685         BUG_ON(i != opts->num_mnt_opts);
686
687         return 0;
688
689 out_free:
690         security_free_mnt_opts(opts);
691         return rc;
692 }
693
694 static int bad_option(struct superblock_security_struct *sbsec, char flag,
695                       u32 old_sid, u32 new_sid)
696 {
697         char mnt_flags = sbsec->flags & SE_MNTMASK;
698
699         /* check if the old mount command had the same options */
700         if (sbsec->flags & SE_SBINITIALIZED)
701                 if (!(sbsec->flags & flag) ||
702                     (old_sid != new_sid))
703                         return 1;
704
705         /* check if we were passed the same options twice,
706          * aka someone passed context=a,context=b
707          */
708         if (!(sbsec->flags & SE_SBINITIALIZED))
709                 if (mnt_flags & flag)
710                         return 1;
711         return 0;
712 }
713
714 /*
715  * Allow filesystems with binary mount data to explicitly set mount point
716  * labeling information.
717  */
718 static int selinux_set_mnt_opts(struct super_block *sb,
719                                 struct security_mnt_opts *opts,
720                                 unsigned long kern_flags,
721                                 unsigned long *set_kern_flags)
722 {
723         const struct cred *cred = current_cred();
724         int rc = 0, i;
725         struct superblock_security_struct *sbsec = sb->s_security;
726         const char *name = sb->s_type->name;
727         struct dentry *root = sbsec->sb->s_root;
728         struct inode_security_struct *root_isec;
729         u32 fscontext_sid = 0, context_sid = 0, rootcontext_sid = 0;
730         u32 defcontext_sid = 0;
731         char **mount_options = opts->mnt_opts;
732         int *flags = opts->mnt_opts_flags;
733         int num_opts = opts->num_mnt_opts;
734
735         mutex_lock(&sbsec->lock);
736
737         if (!selinux_state.initialized) {
738                 if (!num_opts) {
739                         /* Defer initialization until selinux_complete_init,
740                            after the initial policy is loaded and the security
741                            server is ready to handle calls. */
742                         goto out;
743                 }
744                 rc = -EINVAL;
745                 printk(KERN_WARNING "SELinux: Unable to set superblock options "
746                         "before the security server is initialized\n");
747                 goto out;
748         }
749         if (kern_flags && !set_kern_flags) {
750                 /* Specifying internal flags without providing a place to
751                  * place the results is not allowed */
752                 rc = -EINVAL;
753                 goto out;
754         }
755
756         /*
757          * Binary mount data FS will come through this function twice.  Once
758          * from an explicit call and once from the generic calls from the vfs.
759          * Since the generic VFS calls will not contain any security mount data
760          * we need to skip the double mount verification.
761          *
762          * This does open a hole in which we will not notice if the first
763          * mount using this sb set explict options and a second mount using
764          * this sb does not set any security options.  (The first options
765          * will be used for both mounts)
766          */
767         if ((sbsec->flags & SE_SBINITIALIZED) && (sb->s_type->fs_flags & FS_BINARY_MOUNTDATA)
768             && (num_opts == 0))
769                 goto out;
770
771         root_isec = backing_inode_security_novalidate(root);
772
773         /*
774          * parse the mount options, check if they are valid sids.
775          * also check if someone is trying to mount the same sb more
776          * than once with different security options.
777          */
778         for (i = 0; i < num_opts; i++) {
779                 u32 sid;
780
781                 if (flags[i] == SBLABEL_MNT)
782                         continue;
783                 rc = security_context_str_to_sid(&selinux_state,
784                                                  mount_options[i], &sid,
785                                                  GFP_KERNEL);
786                 if (rc) {
787                         printk(KERN_WARNING "SELinux: security_context_str_to_sid"
788                                "(%s) failed for (dev %s, type %s) errno=%d\n",
789                                mount_options[i], sb->s_id, name, rc);
790                         goto out;
791                 }
792                 switch (flags[i]) {
793                 case FSCONTEXT_MNT:
794                         fscontext_sid = sid;
795
796                         if (bad_option(sbsec, FSCONTEXT_MNT, sbsec->sid,
797                                         fscontext_sid))
798                                 goto out_double_mount;
799
800                         sbsec->flags |= FSCONTEXT_MNT;
801                         break;
802                 case CONTEXT_MNT:
803                         context_sid = sid;
804
805                         if (bad_option(sbsec, CONTEXT_MNT, sbsec->mntpoint_sid,
806                                         context_sid))
807                                 goto out_double_mount;
808
809                         sbsec->flags |= CONTEXT_MNT;
810                         break;
811                 case ROOTCONTEXT_MNT:
812                         rootcontext_sid = sid;
813
814                         if (bad_option(sbsec, ROOTCONTEXT_MNT, root_isec->sid,
815                                         rootcontext_sid))
816                                 goto out_double_mount;
817
818                         sbsec->flags |= ROOTCONTEXT_MNT;
819
820                         break;
821                 case DEFCONTEXT_MNT:
822                         defcontext_sid = sid;
823
824                         if (bad_option(sbsec, DEFCONTEXT_MNT, sbsec->def_sid,
825                                         defcontext_sid))
826                                 goto out_double_mount;
827
828                         sbsec->flags |= DEFCONTEXT_MNT;
829
830                         break;
831                 default:
832                         rc = -EINVAL;
833                         goto out;
834                 }
835         }
836
837         if (sbsec->flags & SE_SBINITIALIZED) {
838                 /* previously mounted with options, but not on this attempt? */
839                 if ((sbsec->flags & SE_MNTMASK) && !num_opts)
840                         goto out_double_mount;
841                 rc = 0;
842                 goto out;
843         }
844
845         if (strcmp(sb->s_type->name, "proc") == 0)
846                 sbsec->flags |= SE_SBPROC | SE_SBGENFS;
847
848         if (!strcmp(sb->s_type->name, "debugfs") ||
849             !strcmp(sb->s_type->name, "tracefs") ||
850             !strcmp(sb->s_type->name, "sysfs") ||
851             !strcmp(sb->s_type->name, "pstore") ||
852             !strcmp(sb->s_type->name, "cgroup") ||
853             !strcmp(sb->s_type->name, "cgroup2"))
854                 sbsec->flags |= SE_SBGENFS;
855
856         if (!sbsec->behavior) {
857                 /*
858                  * Determine the labeling behavior to use for this
859                  * filesystem type.
860                  */
861                 rc = security_fs_use(&selinux_state, sb);
862                 if (rc) {
863                         printk(KERN_WARNING
864                                 "%s: security_fs_use(%s) returned %d\n",
865                                         __func__, sb->s_type->name, rc);
866                         goto out;
867                 }
868         }
869
870         /*
871          * If this is a user namespace mount and the filesystem type is not
872          * explicitly whitelisted, then no contexts are allowed on the command
873          * line and security labels must be ignored.
874          */
875         if (sb->s_user_ns != &init_user_ns &&
876             strcmp(sb->s_type->name, "tmpfs") &&
877             strcmp(sb->s_type->name, "ramfs") &&
878             strcmp(sb->s_type->name, "devpts")) {
879                 if (context_sid || fscontext_sid || rootcontext_sid ||
880                     defcontext_sid) {
881                         rc = -EACCES;
882                         goto out;
883                 }
884                 if (sbsec->behavior == SECURITY_FS_USE_XATTR) {
885                         sbsec->behavior = SECURITY_FS_USE_MNTPOINT;
886                         rc = security_transition_sid(&selinux_state,
887                                                      current_sid(),
888                                                      current_sid(),
889                                                      SECCLASS_FILE, NULL,
890                                                      &sbsec->mntpoint_sid);
891                         if (rc)
892                                 goto out;
893                 }
894                 goto out_set_opts;
895         }
896
897         /* sets the context of the superblock for the fs being mounted. */
898         if (fscontext_sid) {
899                 rc = may_context_mount_sb_relabel(fscontext_sid, sbsec, cred);
900                 if (rc)
901                         goto out;
902
903                 sbsec->sid = fscontext_sid;
904         }
905
906         /*
907          * Switch to using mount point labeling behavior.
908          * sets the label used on all file below the mountpoint, and will set
909          * the superblock context if not already set.
910          */
911         if (kern_flags & SECURITY_LSM_NATIVE_LABELS && !context_sid) {
912                 sbsec->behavior = SECURITY_FS_USE_NATIVE;
913                 *set_kern_flags |= SECURITY_LSM_NATIVE_LABELS;
914         }
915
916         if (context_sid) {
917                 if (!fscontext_sid) {
918                         rc = may_context_mount_sb_relabel(context_sid, sbsec,
919                                                           cred);
920                         if (rc)
921                                 goto out;
922                         sbsec->sid = context_sid;
923                 } else {
924                         rc = may_context_mount_inode_relabel(context_sid, sbsec,
925                                                              cred);
926                         if (rc)
927                                 goto out;
928                 }
929                 if (!rootcontext_sid)
930                         rootcontext_sid = context_sid;
931
932                 sbsec->mntpoint_sid = context_sid;
933                 sbsec->behavior = SECURITY_FS_USE_MNTPOINT;
934         }
935
936         if (rootcontext_sid) {
937                 rc = may_context_mount_inode_relabel(rootcontext_sid, sbsec,
938                                                      cred);
939                 if (rc)
940                         goto out;
941
942                 root_isec->sid = rootcontext_sid;
943                 root_isec->initialized = LABEL_INITIALIZED;
944         }
945
946         if (defcontext_sid) {
947                 if (sbsec->behavior != SECURITY_FS_USE_XATTR &&
948                         sbsec->behavior != SECURITY_FS_USE_NATIVE) {
949                         rc = -EINVAL;
950                         printk(KERN_WARNING "SELinux: defcontext option is "
951                                "invalid for this filesystem type\n");
952                         goto out;
953                 }
954
955                 if (defcontext_sid != sbsec->def_sid) {
956                         rc = may_context_mount_inode_relabel(defcontext_sid,
957                                                              sbsec, cred);
958                         if (rc)
959                                 goto out;
960                 }
961
962                 sbsec->def_sid = defcontext_sid;
963         }
964
965 out_set_opts:
966         rc = sb_finish_set_opts(sb);
967 out:
968         mutex_unlock(&sbsec->lock);
969         return rc;
970 out_double_mount:
971         rc = -EINVAL;
972         printk(KERN_WARNING "SELinux: mount invalid.  Same superblock, different "
973                "security settings for (dev %s, type %s)\n", sb->s_id, name);
974         goto out;
975 }
976
977 static int selinux_cmp_sb_context(const struct super_block *oldsb,
978                                     const struct super_block *newsb)
979 {
980         struct superblock_security_struct *old = oldsb->s_security;
981         struct superblock_security_struct *new = newsb->s_security;
982         char oldflags = old->flags & SE_MNTMASK;
983         char newflags = new->flags & SE_MNTMASK;
984
985         if (oldflags != newflags)
986                 goto mismatch;
987         if ((oldflags & FSCONTEXT_MNT) && old->sid != new->sid)
988                 goto mismatch;
989         if ((oldflags & CONTEXT_MNT) && old->mntpoint_sid != new->mntpoint_sid)
990                 goto mismatch;
991         if ((oldflags & DEFCONTEXT_MNT) && old->def_sid != new->def_sid)
992                 goto mismatch;
993         if (oldflags & ROOTCONTEXT_MNT) {
994                 struct inode_security_struct *oldroot = backing_inode_security(oldsb->s_root);
995                 struct inode_security_struct *newroot = backing_inode_security(newsb->s_root);
996                 if (oldroot->sid != newroot->sid)
997                         goto mismatch;
998         }
999         return 0;
1000 mismatch:
1001         printk(KERN_WARNING "SELinux: mount invalid.  Same superblock, "
1002                             "different security settings for (dev %s, "
1003                             "type %s)\n", newsb->s_id, newsb->s_type->name);
1004         return -EBUSY;
1005 }
1006
1007 static int selinux_sb_clone_mnt_opts(const struct super_block *oldsb,
1008                                         struct super_block *newsb,
1009                                         unsigned long kern_flags,
1010                                         unsigned long *set_kern_flags)
1011 {
1012         int rc = 0;
1013         const struct superblock_security_struct *oldsbsec = oldsb->s_security;
1014         struct superblock_security_struct *newsbsec = newsb->s_security;
1015
1016         int set_fscontext =     (oldsbsec->flags & FSCONTEXT_MNT);
1017         int set_context =       (oldsbsec->flags & CONTEXT_MNT);
1018         int set_rootcontext =   (oldsbsec->flags & ROOTCONTEXT_MNT);
1019
1020         /*
1021          * if the parent was able to be mounted it clearly had no special lsm
1022          * mount options.  thus we can safely deal with this superblock later
1023          */
1024         if (!selinux_state.initialized)
1025                 return 0;
1026
1027         /*
1028          * Specifying internal flags without providing a place to
1029          * place the results is not allowed.
1030          */
1031         if (kern_flags && !set_kern_flags)
1032                 return -EINVAL;
1033
1034         /* how can we clone if the old one wasn't set up?? */
1035         BUG_ON(!(oldsbsec->flags & SE_SBINITIALIZED));
1036
1037         /* if fs is reusing a sb, make sure that the contexts match */
1038         if (newsbsec->flags & SE_SBINITIALIZED)
1039                 return selinux_cmp_sb_context(oldsb, newsb);
1040
1041         mutex_lock(&newsbsec->lock);
1042
1043         newsbsec->flags = oldsbsec->flags;
1044
1045         newsbsec->sid = oldsbsec->sid;
1046         newsbsec->def_sid = oldsbsec->def_sid;
1047         newsbsec->behavior = oldsbsec->behavior;
1048
1049         if (newsbsec->behavior == SECURITY_FS_USE_NATIVE &&
1050                 !(kern_flags & SECURITY_LSM_NATIVE_LABELS) && !set_context) {
1051                 rc = security_fs_use(&selinux_state, newsb);
1052                 if (rc)
1053                         goto out;
1054         }
1055
1056         if (kern_flags & SECURITY_LSM_NATIVE_LABELS && !set_context) {
1057                 newsbsec->behavior = SECURITY_FS_USE_NATIVE;
1058                 *set_kern_flags |= SECURITY_LSM_NATIVE_LABELS;
1059         }
1060
1061         if (set_context) {
1062                 u32 sid = oldsbsec->mntpoint_sid;
1063
1064                 if (!set_fscontext)
1065                         newsbsec->sid = sid;
1066                 if (!set_rootcontext) {
1067                         struct inode_security_struct *newisec = backing_inode_security(newsb->s_root);
1068                         newisec->sid = sid;
1069                 }
1070                 newsbsec->mntpoint_sid = sid;
1071         }
1072         if (set_rootcontext) {
1073                 const struct inode_security_struct *oldisec = backing_inode_security(oldsb->s_root);
1074                 struct inode_security_struct *newisec = backing_inode_security(newsb->s_root);
1075
1076                 newisec->sid = oldisec->sid;
1077         }
1078
1079         sb_finish_set_opts(newsb);
1080 out:
1081         mutex_unlock(&newsbsec->lock);
1082         return rc;
1083 }
1084
1085 static int selinux_parse_opts_str(char *options,
1086                                   struct security_mnt_opts *opts)
1087 {
1088         char *p;
1089         char *context = NULL, *defcontext = NULL;
1090         char *fscontext = NULL, *rootcontext = NULL;
1091         int rc, num_mnt_opts = 0;
1092
1093         opts->num_mnt_opts = 0;
1094
1095         /* Standard string-based options. */
1096         while ((p = strsep(&options, "|")) != NULL) {
1097                 int token;
1098                 substring_t args[MAX_OPT_ARGS];
1099
1100                 if (!*p)
1101                         continue;
1102
1103                 token = match_token(p, tokens, args);
1104
1105                 switch (token) {
1106                 case Opt_context:
1107                         if (context || defcontext) {
1108                                 rc = -EINVAL;
1109                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
1110                                 goto out_err;
1111                         }
1112                         context = match_strdup(&args[0]);
1113                         if (!context) {
1114                                 rc = -ENOMEM;
1115                                 goto out_err;
1116                         }
1117                         break;
1118
1119                 case Opt_fscontext:
1120                         if (fscontext) {
1121                                 rc = -EINVAL;
1122                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
1123                                 goto out_err;
1124                         }
1125                         fscontext = match_strdup(&args[0]);
1126                         if (!fscontext) {
1127                                 rc = -ENOMEM;
1128                                 goto out_err;
1129                         }
1130                         break;
1131
1132                 case Opt_rootcontext:
1133                         if (rootcontext) {
1134                                 rc = -EINVAL;
1135                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
1136                                 goto out_err;
1137                         }
1138                         rootcontext = match_strdup(&args[0]);
1139                         if (!rootcontext) {
1140                                 rc = -ENOMEM;
1141                                 goto out_err;
1142                         }
1143                         break;
1144
1145                 case Opt_defcontext:
1146                         if (context || defcontext) {
1147                                 rc = -EINVAL;
1148                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
1149                                 goto out_err;
1150                         }
1151                         defcontext = match_strdup(&args[0]);
1152                         if (!defcontext) {
1153                                 rc = -ENOMEM;
1154                                 goto out_err;
1155                         }
1156                         break;
1157                 case Opt_labelsupport:
1158                         break;
1159                 default:
1160                         rc = -EINVAL;
1161                         printk(KERN_WARNING "SELinux:  unknown mount option\n");
1162                         goto out_err;
1163
1164                 }
1165         }
1166
1167         rc = -ENOMEM;
1168         opts->mnt_opts = kcalloc(NUM_SEL_MNT_OPTS, sizeof(char *), GFP_KERNEL);
1169         if (!opts->mnt_opts)
1170                 goto out_err;
1171
1172         opts->mnt_opts_flags = kcalloc(NUM_SEL_MNT_OPTS, sizeof(int),
1173                                        GFP_KERNEL);
1174         if (!opts->mnt_opts_flags)
1175                 goto out_err;
1176
1177         if (fscontext) {
1178                 opts->mnt_opts[num_mnt_opts] = fscontext;
1179                 opts->mnt_opts_flags[num_mnt_opts++] = FSCONTEXT_MNT;
1180         }
1181         if (context) {
1182                 opts->mnt_opts[num_mnt_opts] = context;
1183                 opts->mnt_opts_flags[num_mnt_opts++] = CONTEXT_MNT;
1184         }
1185         if (rootcontext) {
1186                 opts->mnt_opts[num_mnt_opts] = rootcontext;
1187                 opts->mnt_opts_flags[num_mnt_opts++] = ROOTCONTEXT_MNT;
1188         }
1189         if (defcontext) {
1190                 opts->mnt_opts[num_mnt_opts] = defcontext;
1191                 opts->mnt_opts_flags[num_mnt_opts++] = DEFCONTEXT_MNT;
1192         }
1193
1194         opts->num_mnt_opts = num_mnt_opts;
1195         return 0;
1196
1197 out_err:
1198         security_free_mnt_opts(opts);
1199         kfree(context);
1200         kfree(defcontext);
1201         kfree(fscontext);
1202         kfree(rootcontext);
1203         return rc;
1204 }
1205 /*
1206  * string mount options parsing and call set the sbsec
1207  */
1208 static int superblock_doinit(struct super_block *sb, void *data)
1209 {
1210         int rc = 0;
1211         char *options = data;
1212         struct security_mnt_opts opts;
1213
1214         security_init_mnt_opts(&opts);
1215
1216         if (!data)
1217                 goto out;
1218
1219         BUG_ON(sb->s_type->fs_flags & FS_BINARY_MOUNTDATA);
1220
1221         rc = selinux_parse_opts_str(options, &opts);
1222         if (rc)
1223                 goto out_err;
1224
1225 out:
1226         rc = selinux_set_mnt_opts(sb, &opts, 0, NULL);
1227
1228 out_err:
1229         security_free_mnt_opts(&opts);
1230         return rc;
1231 }
1232
1233 static void selinux_write_opts(struct seq_file *m,
1234                                struct security_mnt_opts *opts)
1235 {
1236         int i;
1237         char *prefix;
1238
1239         for (i = 0; i < opts->num_mnt_opts; i++) {
1240                 char *has_comma;
1241
1242                 if (opts->mnt_opts[i])
1243                         has_comma = strchr(opts->mnt_opts[i], ',');
1244                 else
1245                         has_comma = NULL;
1246
1247                 switch (opts->mnt_opts_flags[i]) {
1248                 case CONTEXT_MNT:
1249                         prefix = CONTEXT_STR;
1250                         break;
1251                 case FSCONTEXT_MNT:
1252                         prefix = FSCONTEXT_STR;
1253                         break;
1254                 case ROOTCONTEXT_MNT:
1255                         prefix = ROOTCONTEXT_STR;
1256                         break;
1257                 case DEFCONTEXT_MNT:
1258                         prefix = DEFCONTEXT_STR;
1259                         break;
1260                 case SBLABEL_MNT:
1261                         seq_putc(m, ',');
1262                         seq_puts(m, LABELSUPP_STR);
1263                         continue;
1264                 default:
1265                         BUG();
1266                         return;
1267                 };
1268                 /* we need a comma before each option */
1269                 seq_putc(m, ',');
1270                 seq_puts(m, prefix);
1271                 if (has_comma)
1272                         seq_putc(m, '\"');
1273                 seq_escape(m, opts->mnt_opts[i], "\"\n\\");
1274                 if (has_comma)
1275                         seq_putc(m, '\"');
1276         }
1277 }
1278
1279 static int selinux_sb_show_options(struct seq_file *m, struct super_block *sb)
1280 {
1281         struct security_mnt_opts opts;
1282         int rc;
1283
1284         rc = selinux_get_mnt_opts(sb, &opts);
1285         if (rc) {
1286                 /* before policy load we may get EINVAL, don't show anything */
1287                 if (rc == -EINVAL)
1288                         rc = 0;
1289                 return rc;
1290         }
1291
1292         selinux_write_opts(m, &opts);
1293
1294         security_free_mnt_opts(&opts);
1295
1296         return rc;
1297 }
1298
1299 static inline u16 inode_mode_to_security_class(umode_t mode)
1300 {
1301         switch (mode & S_IFMT) {
1302         case S_IFSOCK:
1303                 return SECCLASS_SOCK_FILE;
1304         case S_IFLNK:
1305                 return SECCLASS_LNK_FILE;
1306         case S_IFREG:
1307                 return SECCLASS_FILE;
1308         case S_IFBLK:
1309                 return SECCLASS_BLK_FILE;
1310         case S_IFDIR:
1311                 return SECCLASS_DIR;
1312         case S_IFCHR:
1313                 return SECCLASS_CHR_FILE;
1314         case S_IFIFO:
1315                 return SECCLASS_FIFO_FILE;
1316
1317         }
1318
1319         return SECCLASS_FILE;
1320 }
1321
1322 static inline int default_protocol_stream(int protocol)
1323 {
1324         return (protocol == IPPROTO_IP || protocol == IPPROTO_TCP);
1325 }
1326
1327 static inline int default_protocol_dgram(int protocol)
1328 {
1329         return (protocol == IPPROTO_IP || protocol == IPPROTO_UDP);
1330 }
1331
1332 static inline u16 socket_type_to_security_class(int family, int type, int protocol)
1333 {
1334         int extsockclass = selinux_policycap_extsockclass();
1335
1336         switch (family) {
1337         case PF_UNIX:
1338                 switch (type) {
1339                 case SOCK_STREAM:
1340                 case SOCK_SEQPACKET:
1341                         return SECCLASS_UNIX_STREAM_SOCKET;
1342                 case SOCK_DGRAM:
1343                 case SOCK_RAW:
1344                         return SECCLASS_UNIX_DGRAM_SOCKET;
1345                 }
1346                 break;
1347         case PF_INET:
1348         case PF_INET6:
1349                 switch (type) {
1350                 case SOCK_STREAM:
1351                 case SOCK_SEQPACKET:
1352                         if (default_protocol_stream(protocol))
1353                                 return SECCLASS_TCP_SOCKET;
1354                         else if (extsockclass && protocol == IPPROTO_SCTP)
1355                                 return SECCLASS_SCTP_SOCKET;
1356                         else
1357                                 return SECCLASS_RAWIP_SOCKET;
1358                 case SOCK_DGRAM:
1359                         if (default_protocol_dgram(protocol))
1360                                 return SECCLASS_UDP_SOCKET;
1361                         else if (extsockclass && (protocol == IPPROTO_ICMP ||
1362                                                   protocol == IPPROTO_ICMPV6))
1363                                 return SECCLASS_ICMP_SOCKET;
1364                         else
1365                                 return SECCLASS_RAWIP_SOCKET;
1366                 case SOCK_DCCP:
1367                         return SECCLASS_DCCP_SOCKET;
1368                 default:
1369                         return SECCLASS_RAWIP_SOCKET;
1370                 }
1371                 break;
1372         case PF_NETLINK:
1373                 switch (protocol) {
1374                 case NETLINK_ROUTE:
1375                         return SECCLASS_NETLINK_ROUTE_SOCKET;
1376                 case NETLINK_SOCK_DIAG:
1377                         return SECCLASS_NETLINK_TCPDIAG_SOCKET;
1378                 case NETLINK_NFLOG:
1379                         return SECCLASS_NETLINK_NFLOG_SOCKET;
1380                 case NETLINK_XFRM:
1381                         return SECCLASS_NETLINK_XFRM_SOCKET;
1382                 case NETLINK_SELINUX:
1383                         return SECCLASS_NETLINK_SELINUX_SOCKET;
1384                 case NETLINK_ISCSI:
1385                         return SECCLASS_NETLINK_ISCSI_SOCKET;
1386                 case NETLINK_AUDIT:
1387                         return SECCLASS_NETLINK_AUDIT_SOCKET;
1388                 case NETLINK_FIB_LOOKUP:
1389                         return SECCLASS_NETLINK_FIB_LOOKUP_SOCKET;
1390                 case NETLINK_CONNECTOR:
1391                         return SECCLASS_NETLINK_CONNECTOR_SOCKET;
1392                 case NETLINK_NETFILTER:
1393                         return SECCLASS_NETLINK_NETFILTER_SOCKET;
1394                 case NETLINK_DNRTMSG:
1395                         return SECCLASS_NETLINK_DNRT_SOCKET;
1396                 case NETLINK_KOBJECT_UEVENT:
1397                         return SECCLASS_NETLINK_KOBJECT_UEVENT_SOCKET;
1398                 case NETLINK_GENERIC:
1399                         return SECCLASS_NETLINK_GENERIC_SOCKET;
1400                 case NETLINK_SCSITRANSPORT:
1401                         return SECCLASS_NETLINK_SCSITRANSPORT_SOCKET;
1402                 case NETLINK_RDMA:
1403                         return SECCLASS_NETLINK_RDMA_SOCKET;
1404                 case NETLINK_CRYPTO:
1405                         return SECCLASS_NETLINK_CRYPTO_SOCKET;
1406                 default:
1407                         return SECCLASS_NETLINK_SOCKET;
1408                 }
1409         case PF_PACKET:
1410                 return SECCLASS_PACKET_SOCKET;
1411         case PF_KEY:
1412                 return SECCLASS_KEY_SOCKET;
1413         case PF_APPLETALK:
1414                 return SECCLASS_APPLETALK_SOCKET;
1415         }
1416
1417         if (extsockclass) {
1418                 switch (family) {
1419                 case PF_AX25:
1420                         return SECCLASS_AX25_SOCKET;
1421                 case PF_IPX:
1422                         return SECCLASS_IPX_SOCKET;
1423                 case PF_NETROM:
1424                         return SECCLASS_NETROM_SOCKET;
1425                 case PF_ATMPVC:
1426                         return SECCLASS_ATMPVC_SOCKET;
1427                 case PF_X25:
1428                         return SECCLASS_X25_SOCKET;
1429                 case PF_ROSE:
1430                         return SECCLASS_ROSE_SOCKET;
1431                 case PF_DECnet:
1432                         return SECCLASS_DECNET_SOCKET;
1433                 case PF_ATMSVC:
1434                         return SECCLASS_ATMSVC_SOCKET;
1435                 case PF_RDS:
1436                         return SECCLASS_RDS_SOCKET;
1437                 case PF_IRDA:
1438                         return SECCLASS_IRDA_SOCKET;
1439                 case PF_PPPOX:
1440                         return SECCLASS_PPPOX_SOCKET;
1441                 case PF_LLC:
1442                         return SECCLASS_LLC_SOCKET;
1443                 case PF_CAN:
1444                         return SECCLASS_CAN_SOCKET;
1445                 case PF_TIPC:
1446                         return SECCLASS_TIPC_SOCKET;
1447                 case PF_BLUETOOTH:
1448                         return SECCLASS_BLUETOOTH_SOCKET;
1449                 case PF_IUCV:
1450                         return SECCLASS_IUCV_SOCKET;
1451                 case PF_RXRPC:
1452                         return SECCLASS_RXRPC_SOCKET;
1453                 case PF_ISDN:
1454                         return SECCLASS_ISDN_SOCKET;
1455                 case PF_PHONET:
1456                         return SECCLASS_PHONET_SOCKET;
1457                 case PF_IEEE802154:
1458                         return SECCLASS_IEEE802154_SOCKET;
1459                 case PF_CAIF:
1460                         return SECCLASS_CAIF_SOCKET;
1461                 case PF_ALG:
1462                         return SECCLASS_ALG_SOCKET;
1463                 case PF_NFC:
1464                         return SECCLASS_NFC_SOCKET;
1465                 case PF_VSOCK:
1466                         return SECCLASS_VSOCK_SOCKET;
1467                 case PF_KCM:
1468                         return SECCLASS_KCM_SOCKET;
1469                 case PF_QIPCRTR:
1470                         return SECCLASS_QIPCRTR_SOCKET;
1471                 case PF_SMC:
1472                         return SECCLASS_SMC_SOCKET;
1473                 case PF_XDP:
1474                         return SECCLASS_XDP_SOCKET;
1475 #if PF_MAX > 45
1476 #error New address family defined, please update this function.
1477 #endif
1478                 }
1479         }
1480
1481         return SECCLASS_SOCKET;
1482 }
1483
1484 static int selinux_genfs_get_sid(struct dentry *dentry,
1485                                  u16 tclass,
1486                                  u16 flags,
1487                                  u32 *sid)
1488 {
1489         int rc;
1490         struct super_block *sb = dentry->d_sb;
1491         char *buffer, *path;
1492
1493         buffer = (char *)__get_free_page(GFP_KERNEL);
1494         if (!buffer)
1495                 return -ENOMEM;
1496
1497         path = dentry_path_raw(dentry, buffer, PAGE_SIZE);
1498         if (IS_ERR(path))
1499                 rc = PTR_ERR(path);
1500         else {
1501                 if (flags & SE_SBPROC) {
1502                         /* each process gets a /proc/PID/ entry. Strip off the
1503                          * PID part to get a valid selinux labeling.
1504                          * e.g. /proc/1/net/rpc/nfs -> /net/rpc/nfs */
1505                         while (path[1] >= '0' && path[1] <= '9') {
1506                                 path[1] = '/';
1507                                 path++;
1508                         }
1509                 }
1510                 rc = security_genfs_sid(&selinux_state, sb->s_type->name,
1511                                         path, tclass, sid);
1512         }
1513         free_page((unsigned long)buffer);
1514         return rc;
1515 }
1516
1517 /* The inode's security attributes must be initialized before first use. */
1518 static int inode_doinit_with_dentry(struct inode *inode, struct dentry *opt_dentry)
1519 {
1520         struct superblock_security_struct *sbsec = NULL;
1521         struct inode_security_struct *isec = inode->i_security;
1522         u32 task_sid, sid = 0;
1523         u16 sclass;
1524         struct dentry *dentry;
1525 #define INITCONTEXTLEN 255
1526         char *context = NULL;
1527         unsigned len = 0;
1528         int rc = 0;
1529
1530         if (isec->initialized == LABEL_INITIALIZED)
1531                 return 0;
1532
1533         spin_lock(&isec->lock);
1534         if (isec->initialized == LABEL_INITIALIZED)
1535                 goto out_unlock;
1536
1537         if (isec->sclass == SECCLASS_FILE)
1538                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
1539
1540         sbsec = inode->i_sb->s_security;
1541         if (!(sbsec->flags & SE_SBINITIALIZED)) {
1542                 /* Defer initialization until selinux_complete_init,
1543                    after the initial policy is loaded and the security
1544                    server is ready to handle calls. */
1545                 spin_lock(&sbsec->isec_lock);
1546                 if (list_empty(&isec->list))
1547                         list_add(&isec->list, &sbsec->isec_head);
1548                 spin_unlock(&sbsec->isec_lock);
1549                 goto out_unlock;
1550         }
1551
1552         sclass = isec->sclass;
1553         task_sid = isec->task_sid;
1554         sid = isec->sid;
1555         isec->initialized = LABEL_PENDING;
1556         spin_unlock(&isec->lock);
1557
1558         switch (sbsec->behavior) {
1559         case SECURITY_FS_USE_NATIVE:
1560                 break;
1561         case SECURITY_FS_USE_XATTR:
1562                 if (!(inode->i_opflags & IOP_XATTR)) {
1563                         sid = sbsec->def_sid;
1564                         break;
1565                 }
1566                 /* Need a dentry, since the xattr API requires one.
1567                    Life would be simpler if we could just pass the inode. */
1568                 if (opt_dentry) {
1569                         /* Called from d_instantiate or d_splice_alias. */
1570                         dentry = dget(opt_dentry);
1571                 } else {
1572                         /*
1573                          * Called from selinux_complete_init, try to find a dentry.
1574                          * Some filesystems really want a connected one, so try
1575                          * that first.  We could split SECURITY_FS_USE_XATTR in
1576                          * two, depending upon that...
1577                          */
1578                         dentry = d_find_alias(inode);
1579                         if (!dentry)
1580                                 dentry = d_find_any_alias(inode);
1581                 }
1582                 if (!dentry) {
1583                         /*
1584                          * this is can be hit on boot when a file is accessed
1585                          * before the policy is loaded.  When we load policy we
1586                          * may find inodes that have no dentry on the
1587                          * sbsec->isec_head list.  No reason to complain as these
1588                          * will get fixed up the next time we go through
1589                          * inode_doinit with a dentry, before these inodes could
1590                          * be used again by userspace.
1591                          */
1592                         goto out;
1593                 }
1594
1595                 len = INITCONTEXTLEN;
1596                 context = kmalloc(len+1, GFP_NOFS);
1597                 if (!context) {
1598                         rc = -ENOMEM;
1599                         dput(dentry);
1600                         goto out;
1601                 }
1602                 context[len] = '\0';
1603                 rc = __vfs_getxattr(dentry, inode, XATTR_NAME_SELINUX, context, len);
1604                 if (rc == -ERANGE) {
1605                         kfree(context);
1606
1607                         /* Need a larger buffer.  Query for the right size. */
1608                         rc = __vfs_getxattr(dentry, inode, XATTR_NAME_SELINUX, NULL, 0);
1609                         if (rc < 0) {
1610                                 dput(dentry);
1611                                 goto out;
1612                         }
1613                         len = rc;
1614                         context = kmalloc(len+1, GFP_NOFS);
1615                         if (!context) {
1616                                 rc = -ENOMEM;
1617                                 dput(dentry);
1618                                 goto out;
1619                         }
1620                         context[len] = '\0';
1621                         rc = __vfs_getxattr(dentry, inode, XATTR_NAME_SELINUX, context, len);
1622                 }
1623                 dput(dentry);
1624                 if (rc < 0) {
1625                         if (rc != -ENODATA) {
1626                                 printk(KERN_WARNING "SELinux: %s:  getxattr returned "
1627                                        "%d for dev=%s ino=%ld\n", __func__,
1628                                        -rc, inode->i_sb->s_id, inode->i_ino);
1629                                 kfree(context);
1630                                 goto out;
1631                         }
1632                         /* Map ENODATA to the default file SID */
1633                         sid = sbsec->def_sid;
1634                         rc = 0;
1635                 } else {
1636                         rc = security_context_to_sid_default(&selinux_state,
1637                                                              context, rc, &sid,
1638                                                              sbsec->def_sid,
1639                                                              GFP_NOFS);
1640                         if (rc) {
1641                                 char *dev = inode->i_sb->s_id;
1642                                 unsigned long ino = inode->i_ino;
1643
1644                                 if (rc == -EINVAL) {
1645                                         if (printk_ratelimit())
1646                                                 printk(KERN_NOTICE "SELinux: inode=%lu on dev=%s was found to have an invalid "
1647                                                         "context=%s.  This indicates you may need to relabel the inode or the "
1648                                                         "filesystem in question.\n", ino, dev, context);
1649                                 } else {
1650                                         printk(KERN_WARNING "SELinux: %s:  context_to_sid(%s) "
1651                                                "returned %d for dev=%s ino=%ld\n",
1652                                                __func__, context, -rc, dev, ino);
1653                                 }
1654                                 kfree(context);
1655                                 /* Leave with the unlabeled SID */
1656                                 rc = 0;
1657                                 break;
1658                         }
1659                 }
1660                 kfree(context);
1661                 break;
1662         case SECURITY_FS_USE_TASK:
1663                 sid = task_sid;
1664                 break;
1665         case SECURITY_FS_USE_TRANS:
1666                 /* Default to the fs SID. */
1667                 sid = sbsec->sid;
1668
1669                 /* Try to obtain a transition SID. */
1670                 rc = security_transition_sid(&selinux_state, task_sid, sid,
1671                                              sclass, NULL, &sid);
1672                 if (rc)
1673                         goto out;
1674                 break;
1675         case SECURITY_FS_USE_MNTPOINT:
1676                 sid = sbsec->mntpoint_sid;
1677                 break;
1678         default:
1679                 /* Default to the fs superblock SID. */
1680                 sid = sbsec->sid;
1681
1682                 if ((sbsec->flags & SE_SBGENFS) && !S_ISLNK(inode->i_mode)) {
1683                         /* We must have a dentry to determine the label on
1684                          * procfs inodes */
1685                         if (opt_dentry) {
1686                                 /* Called from d_instantiate or
1687                                  * d_splice_alias. */
1688                                 dentry = dget(opt_dentry);
1689                         } else {
1690                                 /* Called from selinux_complete_init, try to
1691                                  * find a dentry.  Some filesystems really want
1692                                  * a connected one, so try that first.
1693                                  */
1694                                 dentry = d_find_alias(inode);
1695                                 if (!dentry)
1696                                         dentry = d_find_any_alias(inode);
1697                         }
1698                         /*
1699                          * This can be hit on boot when a file is accessed
1700                          * before the policy is loaded.  When we load policy we
1701                          * may find inodes that have no dentry on the
1702                          * sbsec->isec_head list.  No reason to complain as
1703                          * these will get fixed up the next time we go through
1704                          * inode_doinit() with a dentry, before these inodes
1705                          * could be used again by userspace.
1706                          */
1707                         if (!dentry)
1708                                 goto out;
1709                         rc = selinux_genfs_get_sid(dentry, sclass,
1710                                                    sbsec->flags, &sid);
1711                         dput(dentry);
1712                         if (rc)
1713                                 goto out;
1714                 }
1715                 break;
1716         }
1717
1718 out:
1719         spin_lock(&isec->lock);
1720         if (isec->initialized == LABEL_PENDING) {
1721                 if (!sid || rc) {
1722                         isec->initialized = LABEL_INVALID;
1723                         goto out_unlock;
1724                 }
1725
1726                 isec->initialized = LABEL_INITIALIZED;
1727                 isec->sid = sid;
1728         }
1729
1730 out_unlock:
1731         spin_unlock(&isec->lock);
1732         return rc;
1733 }
1734
1735 /* Convert a Linux signal to an access vector. */
1736 static inline u32 signal_to_av(int sig)
1737 {
1738         u32 perm = 0;
1739
1740         switch (sig) {
1741         case SIGCHLD:
1742                 /* Commonly granted from child to parent. */
1743                 perm = PROCESS__SIGCHLD;
1744                 break;
1745         case SIGKILL:
1746                 /* Cannot be caught or ignored */
1747                 perm = PROCESS__SIGKILL;
1748                 break;
1749         case SIGSTOP:
1750                 /* Cannot be caught or ignored */
1751                 perm = PROCESS__SIGSTOP;
1752                 break;
1753         default:
1754                 /* All other signals. */
1755                 perm = PROCESS__SIGNAL;
1756                 break;
1757         }
1758
1759         return perm;
1760 }
1761
1762 #if CAP_LAST_CAP > 63
1763 #error Fix SELinux to handle capabilities > 63.
1764 #endif
1765
1766 /* Check whether a task is allowed to use a capability. */
1767 static int cred_has_capability(const struct cred *cred,
1768                                int cap, int audit, bool initns)
1769 {
1770         struct common_audit_data ad;
1771         struct av_decision avd;
1772         u16 sclass;
1773         u32 sid = cred_sid(cred);
1774         u32 av = CAP_TO_MASK(cap);
1775         int rc;
1776
1777         ad.type = LSM_AUDIT_DATA_CAP;
1778         ad.u.cap = cap;
1779
1780         switch (CAP_TO_INDEX(cap)) {
1781         case 0:
1782                 sclass = initns ? SECCLASS_CAPABILITY : SECCLASS_CAP_USERNS;
1783                 break;
1784         case 1:
1785                 sclass = initns ? SECCLASS_CAPABILITY2 : SECCLASS_CAP2_USERNS;
1786                 break;
1787         default:
1788                 printk(KERN_ERR
1789                        "SELinux:  out of range capability %d\n", cap);
1790                 BUG();
1791                 return -EINVAL;
1792         }
1793
1794         rc = avc_has_perm_noaudit(&selinux_state,
1795                                   sid, sid, sclass, av, 0, &avd);
1796         if (audit == SECURITY_CAP_AUDIT) {
1797                 int rc2 = avc_audit(&selinux_state,
1798                                     sid, sid, sclass, av, &avd, rc, &ad, 0);
1799                 if (rc2)
1800                         return rc2;
1801         }
1802         return rc;
1803 }
1804
1805 /* Check whether a task has a particular permission to an inode.
1806    The 'adp' parameter is optional and allows other audit
1807    data to be passed (e.g. the dentry). */
1808 static int inode_has_perm(const struct cred *cred,
1809                           struct inode *inode,
1810                           u32 perms,
1811                           struct common_audit_data *adp)
1812 {
1813         struct inode_security_struct *isec;
1814         u32 sid;
1815
1816         validate_creds(cred);
1817
1818         if (unlikely(IS_PRIVATE(inode)))
1819                 return 0;
1820
1821         sid = cred_sid(cred);
1822         isec = inode->i_security;
1823
1824         return avc_has_perm(&selinux_state,
1825                             sid, isec->sid, isec->sclass, perms, adp);
1826 }
1827
1828 /* Same as inode_has_perm, but pass explicit audit data containing
1829    the dentry to help the auditing code to more easily generate the
1830    pathname if needed. */
1831 static inline int dentry_has_perm(const struct cred *cred,
1832                                   struct dentry *dentry,
1833                                   u32 av)
1834 {
1835         struct inode *inode = d_backing_inode(dentry);
1836         struct common_audit_data ad;
1837
1838         ad.type = LSM_AUDIT_DATA_DENTRY;
1839         ad.u.dentry = dentry;
1840         __inode_security_revalidate(inode, dentry, true);
1841         return inode_has_perm(cred, inode, av, &ad);
1842 }
1843
1844 /* Same as inode_has_perm, but pass explicit audit data containing
1845    the path to help the auditing code to more easily generate the
1846    pathname if needed. */
1847 static inline int path_has_perm(const struct cred *cred,
1848                                 const struct path *path,
1849                                 u32 av)
1850 {
1851         struct inode *inode = d_backing_inode(path->dentry);
1852         struct common_audit_data ad;
1853
1854         ad.type = LSM_AUDIT_DATA_PATH;
1855         ad.u.path = *path;
1856         __inode_security_revalidate(inode, path->dentry, true);
1857         return inode_has_perm(cred, inode, av, &ad);
1858 }
1859
1860 /* Same as path_has_perm, but uses the inode from the file struct. */
1861 static inline int file_path_has_perm(const struct cred *cred,
1862                                      struct file *file,
1863                                      u32 av)
1864 {
1865         struct common_audit_data ad;
1866
1867         ad.type = LSM_AUDIT_DATA_FILE;
1868         ad.u.file = file;
1869         return inode_has_perm(cred, file_inode(file), av, &ad);
1870 }
1871
1872 #ifdef CONFIG_BPF_SYSCALL
1873 static int bpf_fd_pass(struct file *file, u32 sid);
1874 #endif
1875
1876 /* Check whether a task can use an open file descriptor to
1877    access an inode in a given way.  Check access to the
1878    descriptor itself, and then use dentry_has_perm to
1879    check a particular permission to the file.
1880    Access to the descriptor is implicitly granted if it
1881    has the same SID as the process.  If av is zero, then
1882    access to the file is not checked, e.g. for cases
1883    where only the descriptor is affected like seek. */
1884 static int file_has_perm(const struct cred *cred,
1885                          struct file *file,
1886                          u32 av)
1887 {
1888         struct file_security_struct *fsec = file->f_security;
1889         struct inode *inode = file_inode(file);
1890         struct common_audit_data ad;
1891         u32 sid = cred_sid(cred);
1892         int rc;
1893
1894         ad.type = LSM_AUDIT_DATA_FILE;
1895         ad.u.file = file;
1896
1897         if (sid != fsec->sid) {
1898                 rc = avc_has_perm(&selinux_state,
1899                                   sid, fsec->sid,
1900                                   SECCLASS_FD,
1901                                   FD__USE,
1902                                   &ad);
1903                 if (rc)
1904                         goto out;
1905         }
1906
1907 #ifdef CONFIG_BPF_SYSCALL
1908         rc = bpf_fd_pass(file, cred_sid(cred));
1909         if (rc)
1910                 return rc;
1911 #endif
1912
1913         /* av is zero if only checking access to the descriptor. */
1914         rc = 0;
1915         if (av)
1916                 rc = inode_has_perm(cred, inode, av, &ad);
1917
1918 out:
1919         return rc;
1920 }
1921
1922 /*
1923  * Determine the label for an inode that might be unioned.
1924  */
1925 static int
1926 selinux_determine_inode_label(const struct task_security_struct *tsec,
1927                                  struct inode *dir,
1928                                  const struct qstr *name, u16 tclass,
1929                                  u32 *_new_isid)
1930 {
1931         const struct superblock_security_struct *sbsec = dir->i_sb->s_security;
1932
1933         if ((sbsec->flags & SE_SBINITIALIZED) &&
1934             (sbsec->behavior == SECURITY_FS_USE_MNTPOINT)) {
1935                 *_new_isid = sbsec->mntpoint_sid;
1936         } else if ((sbsec->flags & SBLABEL_MNT) &&
1937                    tsec->create_sid) {
1938                 *_new_isid = tsec->create_sid;
1939         } else {
1940                 const struct inode_security_struct *dsec = inode_security(dir);
1941                 return security_transition_sid(&selinux_state, tsec->sid,
1942                                                dsec->sid, tclass,
1943                                                name, _new_isid);
1944         }
1945
1946         return 0;
1947 }
1948
1949 /* Check whether a task can create a file. */
1950 static int may_create(struct inode *dir,
1951                       struct dentry *dentry,
1952                       u16 tclass)
1953 {
1954         const struct task_security_struct *tsec = current_security();
1955         struct inode_security_struct *dsec;
1956         struct superblock_security_struct *sbsec;
1957         u32 sid, newsid;
1958         struct common_audit_data ad;
1959         int rc;
1960
1961         dsec = inode_security(dir);
1962         sbsec = dir->i_sb->s_security;
1963
1964         sid = tsec->sid;
1965
1966         ad.type = LSM_AUDIT_DATA_DENTRY;
1967         ad.u.dentry = dentry;
1968
1969         rc = avc_has_perm(&selinux_state,
1970                           sid, dsec->sid, SECCLASS_DIR,
1971                           DIR__ADD_NAME | DIR__SEARCH,
1972                           &ad);
1973         if (rc)
1974                 return rc;
1975
1976         rc = selinux_determine_inode_label(current_security(), dir,
1977                                            &dentry->d_name, tclass, &newsid);
1978         if (rc)
1979                 return rc;
1980
1981         rc = avc_has_perm(&selinux_state,
1982                           sid, newsid, tclass, FILE__CREATE, &ad);
1983         if (rc)
1984                 return rc;
1985
1986         return avc_has_perm(&selinux_state,
1987                             newsid, sbsec->sid,
1988                             SECCLASS_FILESYSTEM,
1989                             FILESYSTEM__ASSOCIATE, &ad);
1990 }
1991
1992 #define MAY_LINK        0
1993 #define MAY_UNLINK      1
1994 #define MAY_RMDIR       2
1995
1996 /* Check whether a task can link, unlink, or rmdir a file/directory. */
1997 static int may_link(struct inode *dir,
1998                     struct dentry *dentry,
1999                     int kind)
2000
2001 {
2002         struct inode_security_struct *dsec, *isec;
2003         struct common_audit_data ad;
2004         u32 sid = current_sid();
2005         u32 av;
2006         int rc;
2007
2008         dsec = inode_security(dir);
2009         isec = backing_inode_security(dentry);
2010
2011         ad.type = LSM_AUDIT_DATA_DENTRY;
2012         ad.u.dentry = dentry;
2013
2014         av = DIR__SEARCH;
2015         av |= (kind ? DIR__REMOVE_NAME : DIR__ADD_NAME);
2016         rc = avc_has_perm(&selinux_state,
2017                           sid, dsec->sid, SECCLASS_DIR, av, &ad);
2018         if (rc)
2019                 return rc;
2020
2021         switch (kind) {
2022         case MAY_LINK:
2023                 av = FILE__LINK;
2024                 break;
2025         case MAY_UNLINK:
2026                 av = FILE__UNLINK;
2027                 break;
2028         case MAY_RMDIR:
2029                 av = DIR__RMDIR;
2030                 break;
2031         default:
2032                 printk(KERN_WARNING "SELinux: %s:  unrecognized kind %d\n",
2033                         __func__, kind);
2034                 return 0;
2035         }
2036
2037         rc = avc_has_perm(&selinux_state,
2038                           sid, isec->sid, isec->sclass, av, &ad);
2039         return rc;
2040 }
2041
2042 static inline int may_rename(struct inode *old_dir,
2043                              struct dentry *old_dentry,
2044                              struct inode *new_dir,
2045                              struct dentry *new_dentry)
2046 {
2047         struct inode_security_struct *old_dsec, *new_dsec, *old_isec, *new_isec;
2048         struct common_audit_data ad;
2049         u32 sid = current_sid();
2050         u32 av;
2051         int old_is_dir, new_is_dir;
2052         int rc;
2053
2054         old_dsec = inode_security(old_dir);
2055         old_isec = backing_inode_security(old_dentry);
2056         old_is_dir = d_is_dir(old_dentry);
2057         new_dsec = inode_security(new_dir);
2058
2059         ad.type = LSM_AUDIT_DATA_DENTRY;
2060
2061         ad.u.dentry = old_dentry;
2062         rc = avc_has_perm(&selinux_state,
2063                           sid, old_dsec->sid, SECCLASS_DIR,
2064                           DIR__REMOVE_NAME | DIR__SEARCH, &ad);
2065         if (rc)
2066                 return rc;
2067         rc = avc_has_perm(&selinux_state,
2068                           sid, old_isec->sid,
2069                           old_isec->sclass, FILE__RENAME, &ad);
2070         if (rc)
2071                 return rc;
2072         if (old_is_dir && new_dir != old_dir) {
2073                 rc = avc_has_perm(&selinux_state,
2074                                   sid, old_isec->sid,
2075                                   old_isec->sclass, DIR__REPARENT, &ad);
2076                 if (rc)
2077                         return rc;
2078         }
2079
2080         ad.u.dentry = new_dentry;
2081         av = DIR__ADD_NAME | DIR__SEARCH;
2082         if (d_is_positive(new_dentry))
2083                 av |= DIR__REMOVE_NAME;
2084         rc = avc_has_perm(&selinux_state,
2085                           sid, new_dsec->sid, SECCLASS_DIR, av, &ad);
2086         if (rc)
2087                 return rc;
2088         if (d_is_positive(new_dentry)) {
2089                 new_isec = backing_inode_security(new_dentry);
2090                 new_is_dir = d_is_dir(new_dentry);
2091                 rc = avc_has_perm(&selinux_state,
2092                                   sid, new_isec->sid,
2093                                   new_isec->sclass,
2094                                   (new_is_dir ? DIR__RMDIR : FILE__UNLINK), &ad);
2095                 if (rc)
2096                         return rc;
2097         }
2098
2099         return 0;
2100 }
2101
2102 /* Check whether a task can perform a filesystem operation. */
2103 static int superblock_has_perm(const struct cred *cred,
2104                                struct super_block *sb,
2105                                u32 perms,
2106                                struct common_audit_data *ad)
2107 {
2108         struct superblock_security_struct *sbsec;
2109         u32 sid = cred_sid(cred);
2110
2111         sbsec = sb->s_security;
2112         return avc_has_perm(&selinux_state,
2113                             sid, sbsec->sid, SECCLASS_FILESYSTEM, perms, ad);
2114 }
2115
2116 /* Convert a Linux mode and permission mask to an access vector. */
2117 static inline u32 file_mask_to_av(int mode, int mask)
2118 {
2119         u32 av = 0;
2120
2121         if (!S_ISDIR(mode)) {
2122                 if (mask & MAY_EXEC)
2123                         av |= FILE__EXECUTE;
2124                 if (mask & MAY_READ)
2125                         av |= FILE__READ;
2126
2127                 if (mask & MAY_APPEND)
2128                         av |= FILE__APPEND;
2129                 else if (mask & MAY_WRITE)
2130                         av |= FILE__WRITE;
2131
2132         } else {
2133                 if (mask & MAY_EXEC)
2134                         av |= DIR__SEARCH;
2135                 if (mask & MAY_WRITE)
2136                         av |= DIR__WRITE;
2137                 if (mask & MAY_READ)
2138                         av |= DIR__READ;
2139         }
2140
2141         return av;
2142 }
2143
2144 /* Convert a Linux file to an access vector. */
2145 static inline u32 file_to_av(struct file *file)
2146 {
2147         u32 av = 0;
2148
2149         if (file->f_mode & FMODE_READ)
2150                 av |= FILE__READ;
2151         if (file->f_mode & FMODE_WRITE) {
2152                 if (file->f_flags & O_APPEND)
2153                         av |= FILE__APPEND;
2154                 else
2155                         av |= FILE__WRITE;
2156         }
2157         if (!av) {
2158                 /*
2159                  * Special file opened with flags 3 for ioctl-only use.
2160                  */
2161                 av = FILE__IOCTL;
2162         }
2163
2164         return av;
2165 }
2166
2167 /*
2168  * Convert a file to an access vector and include the correct open
2169  * open permission.
2170  */
2171 static inline u32 open_file_to_av(struct file *file)
2172 {
2173         u32 av = file_to_av(file);
2174         struct inode *inode = file_inode(file);
2175
2176         if (selinux_policycap_openperm() &&
2177             inode->i_sb->s_magic != SOCKFS_MAGIC)
2178                 av |= FILE__OPEN;
2179
2180         return av;
2181 }
2182
2183 /* Hook functions begin here. */
2184
2185 static int selinux_binder_set_context_mgr(struct task_struct *mgr)
2186 {
2187         u32 mysid = current_sid();
2188         u32 mgrsid = task_sid(mgr);
2189
2190         return avc_has_perm(&selinux_state,
2191                             mysid, mgrsid, SECCLASS_BINDER,
2192                             BINDER__SET_CONTEXT_MGR, NULL);
2193 }
2194
2195 static int selinux_binder_transaction(struct task_struct *from,
2196                                       struct task_struct *to)
2197 {
2198         u32 mysid = current_sid();
2199         u32 fromsid = task_sid(from);
2200         u32 tosid = task_sid(to);
2201         int rc;
2202
2203         if (mysid != fromsid) {
2204                 rc = avc_has_perm(&selinux_state,
2205                                   mysid, fromsid, SECCLASS_BINDER,
2206                                   BINDER__IMPERSONATE, NULL);
2207                 if (rc)
2208                         return rc;
2209         }
2210
2211         return avc_has_perm(&selinux_state,
2212                             fromsid, tosid, SECCLASS_BINDER, BINDER__CALL,
2213                             NULL);
2214 }
2215
2216 static int selinux_binder_transfer_binder(struct task_struct *from,
2217                                           struct task_struct *to)
2218 {
2219         u32 fromsid = task_sid(from);
2220         u32 tosid = task_sid(to);
2221
2222         return avc_has_perm(&selinux_state,
2223                             fromsid, tosid, SECCLASS_BINDER, BINDER__TRANSFER,
2224                             NULL);
2225 }
2226
2227 static int selinux_binder_transfer_file(struct task_struct *from,
2228                                         struct task_struct *to,
2229                                         struct file *file)
2230 {
2231         u32 sid = task_sid(to);
2232         struct file_security_struct *fsec = file->f_security;
2233         struct dentry *dentry = file->f_path.dentry;
2234         struct inode_security_struct *isec;
2235         struct common_audit_data ad;
2236         int rc;
2237
2238         ad.type = LSM_AUDIT_DATA_PATH;
2239         ad.u.path = file->f_path;
2240
2241         if (sid != fsec->sid) {
2242                 rc = avc_has_perm(&selinux_state,
2243                                   sid, fsec->sid,
2244                                   SECCLASS_FD,
2245                                   FD__USE,
2246                                   &ad);
2247                 if (rc)
2248                         return rc;
2249         }
2250
2251 #ifdef CONFIG_BPF_SYSCALL
2252         rc = bpf_fd_pass(file, sid);
2253         if (rc)
2254                 return rc;
2255 #endif
2256
2257         if (unlikely(IS_PRIVATE(d_backing_inode(dentry))))
2258                 return 0;
2259
2260         isec = backing_inode_security(dentry);
2261         return avc_has_perm(&selinux_state,
2262                             sid, isec->sid, isec->sclass, file_to_av(file),
2263                             &ad);
2264 }
2265
2266 static int selinux_ptrace_access_check(struct task_struct *child,
2267                                      unsigned int mode)
2268 {
2269         u32 sid = current_sid();
2270         u32 csid = task_sid(child);
2271
2272         if (mode & PTRACE_MODE_READ)
2273                 return avc_has_perm(&selinux_state,
2274                                     sid, csid, SECCLASS_FILE, FILE__READ, NULL);
2275
2276         return avc_has_perm(&selinux_state,
2277                             sid, csid, SECCLASS_PROCESS, PROCESS__PTRACE, NULL);
2278 }
2279
2280 static int selinux_ptrace_traceme(struct task_struct *parent)
2281 {
2282         return avc_has_perm(&selinux_state,
2283                             task_sid(parent), current_sid(), SECCLASS_PROCESS,
2284                             PROCESS__PTRACE, NULL);
2285 }
2286
2287 static int selinux_capget(struct task_struct *target, kernel_cap_t *effective,
2288                           kernel_cap_t *inheritable, kernel_cap_t *permitted)
2289 {
2290         return avc_has_perm(&selinux_state,
2291                             current_sid(), task_sid(target), SECCLASS_PROCESS,
2292                             PROCESS__GETCAP, NULL);
2293 }
2294
2295 static int selinux_capset(struct cred *new, const struct cred *old,
2296                           const kernel_cap_t *effective,
2297                           const kernel_cap_t *inheritable,
2298                           const kernel_cap_t *permitted)
2299 {
2300         return avc_has_perm(&selinux_state,
2301                             cred_sid(old), cred_sid(new), SECCLASS_PROCESS,
2302                             PROCESS__SETCAP, NULL);
2303 }
2304
2305 /*
2306  * (This comment used to live with the selinux_task_setuid hook,
2307  * which was removed).
2308  *
2309  * Since setuid only affects the current process, and since the SELinux
2310  * controls are not based on the Linux identity attributes, SELinux does not
2311  * need to control this operation.  However, SELinux does control the use of
2312  * the CAP_SETUID and CAP_SETGID capabilities using the capable hook.
2313  */
2314
2315 static int selinux_capable(const struct cred *cred, struct user_namespace *ns,
2316                            int cap, int audit)
2317 {
2318         return cred_has_capability(cred, cap, audit, ns == &init_user_ns);
2319 }
2320
2321 static int selinux_quotactl(int cmds, int type, int id, struct super_block *sb)
2322 {
2323         const struct cred *cred = current_cred();
2324         int rc = 0;
2325
2326         if (!sb)
2327                 return 0;
2328
2329         switch (cmds) {
2330         case Q_SYNC:
2331         case Q_QUOTAON:
2332         case Q_QUOTAOFF:
2333         case Q_SETINFO:
2334         case Q_SETQUOTA:
2335                 rc = superblock_has_perm(cred, sb, FILESYSTEM__QUOTAMOD, NULL);
2336                 break;
2337         case Q_GETFMT:
2338         case Q_GETINFO:
2339         case Q_GETQUOTA:
2340                 rc = superblock_has_perm(cred, sb, FILESYSTEM__QUOTAGET, NULL);
2341                 break;
2342         default:
2343                 rc = 0;  /* let the kernel handle invalid cmds */
2344                 break;
2345         }
2346         return rc;
2347 }
2348
2349 static int selinux_quota_on(struct dentry *dentry)
2350 {
2351         const struct cred *cred = current_cred();
2352
2353         return dentry_has_perm(cred, dentry, FILE__QUOTAON);
2354 }
2355
2356 static int selinux_syslog(int type)
2357 {
2358         switch (type) {
2359         case SYSLOG_ACTION_READ_ALL:    /* Read last kernel messages */
2360         case SYSLOG_ACTION_SIZE_BUFFER: /* Return size of the log buffer */
2361                 return avc_has_perm(&selinux_state,
2362                                     current_sid(), SECINITSID_KERNEL,
2363                                     SECCLASS_SYSTEM, SYSTEM__SYSLOG_READ, NULL);
2364         case SYSLOG_ACTION_CONSOLE_OFF: /* Disable logging to console */
2365         case SYSLOG_ACTION_CONSOLE_ON:  /* Enable logging to console */
2366         /* Set level of messages printed to console */
2367         case SYSLOG_ACTION_CONSOLE_LEVEL:
2368                 return avc_has_perm(&selinux_state,
2369                                     current_sid(), SECINITSID_KERNEL,
2370                                     SECCLASS_SYSTEM, SYSTEM__SYSLOG_CONSOLE,
2371                                     NULL);
2372         }
2373         /* All other syslog types */
2374         return avc_has_perm(&selinux_state,
2375                             current_sid(), SECINITSID_KERNEL,
2376                             SECCLASS_SYSTEM, SYSTEM__SYSLOG_MOD, NULL);
2377 }
2378
2379 /*
2380  * Check that a process has enough memory to allocate a new virtual
2381  * mapping. 0 means there is enough memory for the allocation to
2382  * succeed and -ENOMEM implies there is not.
2383  *
2384  * Do not audit the selinux permission check, as this is applied to all
2385  * processes that allocate mappings.
2386  */
2387 static int selinux_vm_enough_memory(struct mm_struct *mm, long pages)
2388 {
2389         int rc, cap_sys_admin = 0;
2390
2391         rc = cred_has_capability(current_cred(), CAP_SYS_ADMIN,
2392                                  SECURITY_CAP_NOAUDIT, true);
2393         if (rc == 0)
2394                 cap_sys_admin = 1;
2395
2396         return cap_sys_admin;
2397 }
2398
2399 /* binprm security operations */
2400
2401 static u32 ptrace_parent_sid(void)
2402 {
2403         u32 sid = 0;
2404         struct task_struct *tracer;
2405
2406         rcu_read_lock();
2407         tracer = ptrace_parent(current);
2408         if (tracer)
2409                 sid = task_sid(tracer);
2410         rcu_read_unlock();
2411
2412         return sid;
2413 }
2414
2415 static int check_nnp_nosuid(const struct linux_binprm *bprm,
2416                             const struct task_security_struct *old_tsec,
2417                             const struct task_security_struct *new_tsec)
2418 {
2419         int nnp = (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS);
2420         int nosuid = !mnt_may_suid(bprm->file->f_path.mnt);
2421         int rc;
2422         u32 av;
2423
2424         if (!nnp && !nosuid)
2425                 return 0; /* neither NNP nor nosuid */
2426
2427         if (new_tsec->sid == old_tsec->sid)
2428                 return 0; /* No change in credentials */
2429
2430         /*
2431          * If the policy enables the nnp_nosuid_transition policy capability,
2432          * then we permit transitions under NNP or nosuid if the
2433          * policy allows the corresponding permission between
2434          * the old and new contexts.
2435          */
2436         if (selinux_policycap_nnp_nosuid_transition()) {
2437                 av = 0;
2438                 if (nnp)
2439                         av |= PROCESS2__NNP_TRANSITION;
2440                 if (nosuid)
2441                         av |= PROCESS2__NOSUID_TRANSITION;
2442                 rc = avc_has_perm(&selinux_state,
2443                                   old_tsec->sid, new_tsec->sid,
2444                                   SECCLASS_PROCESS2, av, NULL);
2445                 if (!rc)
2446                         return 0;
2447         }
2448
2449         /*
2450          * We also permit NNP or nosuid transitions to bounded SIDs,
2451          * i.e. SIDs that are guaranteed to only be allowed a subset
2452          * of the permissions of the current SID.
2453          */
2454         rc = security_bounded_transition(&selinux_state, old_tsec->sid,
2455                                          new_tsec->sid);
2456         if (!rc)
2457                 return 0;
2458
2459         /*
2460          * On failure, preserve the errno values for NNP vs nosuid.
2461          * NNP:  Operation not permitted for caller.
2462          * nosuid:  Permission denied to file.
2463          */
2464         if (nnp)
2465                 return -EPERM;
2466         return -EACCES;
2467 }
2468
2469 static int selinux_bprm_set_creds(struct linux_binprm *bprm)
2470 {
2471         const struct task_security_struct *old_tsec;
2472         struct task_security_struct *new_tsec;
2473         struct inode_security_struct *isec;
2474         struct common_audit_data ad;
2475         struct inode *inode = file_inode(bprm->file);
2476         int rc;
2477
2478         /* SELinux context only depends on initial program or script and not
2479          * the script interpreter */
2480         if (bprm->called_set_creds)
2481                 return 0;
2482
2483         old_tsec = current_security();
2484         new_tsec = bprm->cred->security;
2485         isec = inode_security(inode);
2486
2487         /* Default to the current task SID. */
2488         new_tsec->sid = old_tsec->sid;
2489         new_tsec->osid = old_tsec->sid;
2490
2491         /* Reset fs, key, and sock SIDs on execve. */
2492         new_tsec->create_sid = 0;
2493         new_tsec->keycreate_sid = 0;
2494         new_tsec->sockcreate_sid = 0;
2495
2496         if (old_tsec->exec_sid) {
2497                 new_tsec->sid = old_tsec->exec_sid;
2498                 /* Reset exec SID on execve. */
2499                 new_tsec->exec_sid = 0;
2500
2501                 /* Fail on NNP or nosuid if not an allowed transition. */
2502                 rc = check_nnp_nosuid(bprm, old_tsec, new_tsec);
2503                 if (rc)
2504                         return rc;
2505         } else {
2506                 /* Check for a default transition on this program. */
2507                 rc = security_transition_sid(&selinux_state, old_tsec->sid,
2508                                              isec->sid, SECCLASS_PROCESS, NULL,
2509                                              &new_tsec->sid);
2510                 if (rc)
2511                         return rc;
2512
2513                 /*
2514                  * Fallback to old SID on NNP or nosuid if not an allowed
2515                  * transition.
2516                  */
2517                 rc = check_nnp_nosuid(bprm, old_tsec, new_tsec);
2518                 if (rc)
2519                         new_tsec->sid = old_tsec->sid;
2520         }
2521
2522         ad.type = LSM_AUDIT_DATA_FILE;
2523         ad.u.file = bprm->file;
2524
2525         if (new_tsec->sid == old_tsec->sid) {
2526                 rc = avc_has_perm(&selinux_state,
2527                                   old_tsec->sid, isec->sid,
2528                                   SECCLASS_FILE, FILE__EXECUTE_NO_TRANS, &ad);
2529                 if (rc)
2530                         return rc;
2531         } else {
2532                 /* Check permissions for the transition. */
2533                 rc = avc_has_perm(&selinux_state,
2534                                   old_tsec->sid, new_tsec->sid,
2535                                   SECCLASS_PROCESS, PROCESS__TRANSITION, &ad);
2536                 if (rc)
2537                         return rc;
2538
2539                 rc = avc_has_perm(&selinux_state,
2540                                   new_tsec->sid, isec->sid,
2541                                   SECCLASS_FILE, FILE__ENTRYPOINT, &ad);
2542                 if (rc)
2543                         return rc;
2544
2545                 /* Check for shared state */
2546                 if (bprm->unsafe & LSM_UNSAFE_SHARE) {
2547                         rc = avc_has_perm(&selinux_state,
2548                                           old_tsec->sid, new_tsec->sid,
2549                                           SECCLASS_PROCESS, PROCESS__SHARE,
2550                                           NULL);
2551                         if (rc)
2552                                 return -EPERM;
2553                 }
2554
2555                 /* Make sure that anyone attempting to ptrace over a task that
2556                  * changes its SID has the appropriate permit */
2557                 if (bprm->unsafe & LSM_UNSAFE_PTRACE) {
2558                         u32 ptsid = ptrace_parent_sid();
2559                         if (ptsid != 0) {
2560                                 rc = avc_has_perm(&selinux_state,
2561                                                   ptsid, new_tsec->sid,
2562                                                   SECCLASS_PROCESS,
2563                                                   PROCESS__PTRACE, NULL);
2564                                 if (rc)
2565                                         return -EPERM;
2566                         }
2567                 }
2568
2569                 /* Clear any possibly unsafe personality bits on exec: */
2570                 bprm->per_clear |= PER_CLEAR_ON_SETID;
2571
2572                 /* Enable secure mode for SIDs transitions unless
2573                    the noatsecure permission is granted between
2574                    the two SIDs, i.e. ahp returns 0. */
2575                 rc = avc_has_perm(&selinux_state,
2576                                   old_tsec->sid, new_tsec->sid,
2577                                   SECCLASS_PROCESS, PROCESS__NOATSECURE,
2578                                   NULL);
2579                 bprm->secureexec |= !!rc;
2580         }
2581
2582         return 0;
2583 }
2584
2585 static int match_file(const void *p, struct file *file, unsigned fd)
2586 {
2587         return file_has_perm(p, file, file_to_av(file)) ? fd + 1 : 0;
2588 }
2589
2590 /* Derived from fs/exec.c:flush_old_files. */
2591 static inline void flush_unauthorized_files(const struct cred *cred,
2592                                             struct files_struct *files)
2593 {
2594         struct file *file, *devnull = NULL;
2595         struct tty_struct *tty;
2596         int drop_tty = 0;
2597         unsigned n;
2598
2599         tty = get_current_tty();
2600         if (tty) {
2601                 spin_lock(&tty->files_lock);
2602                 if (!list_empty(&tty->tty_files)) {
2603                         struct tty_file_private *file_priv;
2604
2605                         /* Revalidate access to controlling tty.
2606                            Use file_path_has_perm on the tty path directly
2607                            rather than using file_has_perm, as this particular
2608                            open file may belong to another process and we are
2609                            only interested in the inode-based check here. */
2610                         file_priv = list_first_entry(&tty->tty_files,
2611                                                 struct tty_file_private, list);
2612                         file = file_priv->file;
2613                         if (file_path_has_perm(cred, file, FILE__READ | FILE__WRITE))
2614                                 drop_tty = 1;
2615                 }
2616                 spin_unlock(&tty->files_lock);
2617                 tty_kref_put(tty);
2618         }
2619         /* Reset controlling tty. */
2620         if (drop_tty)
2621                 no_tty();
2622
2623         /* Revalidate access to inherited open files. */
2624         n = iterate_fd(files, 0, match_file, cred);
2625         if (!n) /* none found? */
2626                 return;
2627
2628         devnull = dentry_open(&selinux_null, O_RDWR, cred);
2629         if (IS_ERR(devnull))
2630                 devnull = NULL;
2631         /* replace all the matching ones with this */
2632         do {
2633                 replace_fd(n - 1, devnull, 0);
2634         } while ((n = iterate_fd(files, n, match_file, cred)) != 0);
2635         if (devnull)
2636                 fput(devnull);
2637 }
2638
2639 /*
2640  * Prepare a process for imminent new credential changes due to exec
2641  */
2642 static void selinux_bprm_committing_creds(struct linux_binprm *bprm)
2643 {
2644         struct task_security_struct *new_tsec;
2645         struct rlimit *rlim, *initrlim;
2646         int rc, i;
2647
2648         new_tsec = bprm->cred->security;
2649         if (new_tsec->sid == new_tsec->osid)
2650                 return;
2651
2652         /* Close files for which the new task SID is not authorized. */
2653         flush_unauthorized_files(bprm->cred, current->files);
2654
2655         /* Always clear parent death signal on SID transitions. */
2656         current->pdeath_signal = 0;
2657
2658         /* Check whether the new SID can inherit resource limits from the old
2659          * SID.  If not, reset all soft limits to the lower of the current
2660          * task's hard limit and the init task's soft limit.
2661          *
2662          * Note that the setting of hard limits (even to lower them) can be
2663          * controlled by the setrlimit check.  The inclusion of the init task's
2664          * soft limit into the computation is to avoid resetting soft limits
2665          * higher than the default soft limit for cases where the default is
2666          * lower than the hard limit, e.g. RLIMIT_CORE or RLIMIT_STACK.
2667          */
2668         rc = avc_has_perm(&selinux_state,
2669                           new_tsec->osid, new_tsec->sid, SECCLASS_PROCESS,
2670                           PROCESS__RLIMITINH, NULL);
2671         if (rc) {
2672                 /* protect against do_prlimit() */
2673                 task_lock(current);
2674                 for (i = 0; i < RLIM_NLIMITS; i++) {
2675                         rlim = current->signal->rlim + i;
2676                         initrlim = init_task.signal->rlim + i;
2677                         rlim->rlim_cur = min(rlim->rlim_max, initrlim->rlim_cur);
2678                 }
2679                 task_unlock(current);
2680                 if (IS_ENABLED(CONFIG_POSIX_TIMERS))
2681                         update_rlimit_cpu(current, rlimit(RLIMIT_CPU));
2682         }
2683 }
2684
2685 /*
2686  * Clean up the process immediately after the installation of new credentials
2687  * due to exec
2688  */
2689 static void selinux_bprm_committed_creds(struct linux_binprm *bprm)
2690 {
2691         const struct task_security_struct *tsec = current_security();
2692         struct itimerval itimer;
2693         u32 osid, sid;
2694         int rc, i;
2695
2696         osid = tsec->osid;
2697         sid = tsec->sid;
2698
2699         if (sid == osid)
2700                 return;
2701
2702         /* Check whether the new SID can inherit signal state from the old SID.
2703          * If not, clear itimers to avoid subsequent signal generation and
2704          * flush and unblock signals.
2705          *
2706          * This must occur _after_ the task SID has been updated so that any
2707          * kill done after the flush will be checked against the new SID.
2708          */
2709         rc = avc_has_perm(&selinux_state,
2710                           osid, sid, SECCLASS_PROCESS, PROCESS__SIGINH, NULL);
2711         if (rc) {
2712                 if (IS_ENABLED(CONFIG_POSIX_TIMERS)) {
2713                         memset(&itimer, 0, sizeof itimer);
2714                         for (i = 0; i < 3; i++)
2715                                 do_setitimer(i, &itimer, NULL);
2716                 }
2717                 spin_lock_irq(&current->sighand->siglock);
2718                 if (!fatal_signal_pending(current)) {
2719                         flush_sigqueue(&current->pending);
2720                         flush_sigqueue(&current->signal->shared_pending);
2721                         flush_signal_handlers(current, 1);
2722                         sigemptyset(&current->blocked);
2723                         recalc_sigpending();
2724                 }
2725                 spin_unlock_irq(&current->sighand->siglock);
2726         }
2727
2728         /* Wake up the parent if it is waiting so that it can recheck
2729          * wait permission to the new task SID. */
2730         read_lock(&tasklist_lock);
2731         __wake_up_parent(current, current->real_parent);
2732         read_unlock(&tasklist_lock);
2733 }
2734
2735 /* superblock security operations */
2736
2737 static int selinux_sb_alloc_security(struct super_block *sb)
2738 {
2739         return superblock_alloc_security(sb);
2740 }
2741
2742 static void selinux_sb_free_security(struct super_block *sb)
2743 {
2744         superblock_free_security(sb);
2745 }
2746
2747 static inline int match_prefix(char *prefix, int plen, char *option, int olen)
2748 {
2749         if (plen > olen)
2750                 return 0;
2751
2752         return !memcmp(prefix, option, plen);
2753 }
2754
2755 static inline int selinux_option(char *option, int len)
2756 {
2757         return (match_prefix(CONTEXT_STR, sizeof(CONTEXT_STR)-1, option, len) ||
2758                 match_prefix(FSCONTEXT_STR, sizeof(FSCONTEXT_STR)-1, option, len) ||
2759                 match_prefix(DEFCONTEXT_STR, sizeof(DEFCONTEXT_STR)-1, option, len) ||
2760                 match_prefix(ROOTCONTEXT_STR, sizeof(ROOTCONTEXT_STR)-1, option, len) ||
2761                 match_prefix(LABELSUPP_STR, sizeof(LABELSUPP_STR)-1, option, len));
2762 }
2763
2764 static inline void take_option(char **to, char *from, int *first, int len)
2765 {
2766         if (!*first) {
2767                 **to = ',';
2768                 *to += 1;
2769         } else
2770                 *first = 0;
2771         memcpy(*to, from, len);
2772         *to += len;
2773 }
2774
2775 static inline void take_selinux_option(char **to, char *from, int *first,
2776                                        int len)
2777 {
2778         int current_size = 0;
2779
2780         if (!*first) {
2781                 **to = '|';
2782                 *to += 1;
2783         } else
2784                 *first = 0;
2785
2786         while (current_size < len) {
2787                 if (*from != '"') {
2788                         **to = *from;
2789                         *to += 1;
2790                 }
2791                 from += 1;
2792                 current_size += 1;
2793         }
2794 }
2795
2796 static int selinux_sb_copy_data(char *orig, char *copy)
2797 {
2798         int fnosec, fsec, rc = 0;
2799         char *in_save, *in_curr, *in_end;
2800         char *sec_curr, *nosec_save, *nosec;
2801         int open_quote = 0;
2802
2803         in_curr = orig;
2804         sec_curr = copy;
2805
2806         nosec = (char *)get_zeroed_page(GFP_KERNEL);
2807         if (!nosec) {
2808                 rc = -ENOMEM;
2809                 goto out;
2810         }
2811
2812         nosec_save = nosec;
2813         fnosec = fsec = 1;
2814         in_save = in_end = orig;
2815
2816         do {
2817                 if (*in_end == '"')
2818                         open_quote = !open_quote;
2819                 if ((*in_end == ',' && open_quote == 0) ||
2820                                 *in_end == '\0') {
2821                         int len = in_end - in_curr;
2822
2823                         if (selinux_option(in_curr, len))
2824                                 take_selinux_option(&sec_curr, in_curr, &fsec, len);
2825                         else
2826                                 take_option(&nosec, in_curr, &fnosec, len);
2827
2828                         in_curr = in_end + 1;
2829                 }
2830         } while (*in_end++);
2831
2832         strcpy(in_save, nosec_save);
2833         free_page((unsigned long)nosec_save);
2834 out:
2835         return rc;
2836 }
2837
2838 static int selinux_sb_remount(struct super_block *sb, void *data)
2839 {
2840         int rc, i, *flags;
2841         struct security_mnt_opts opts;
2842         char *secdata, **mount_options;
2843         struct superblock_security_struct *sbsec = sb->s_security;
2844
2845         if (!(sbsec->flags & SE_SBINITIALIZED))
2846                 return 0;
2847
2848         if (!data)
2849                 return 0;
2850
2851         if (sb->s_type->fs_flags & FS_BINARY_MOUNTDATA)
2852                 return 0;
2853
2854         security_init_mnt_opts(&opts);
2855         secdata = alloc_secdata();
2856         if (!secdata)
2857                 return -ENOMEM;
2858         rc = selinux_sb_copy_data(data, secdata);
2859         if (rc)
2860                 goto out_free_secdata;
2861
2862         rc = selinux_parse_opts_str(secdata, &opts);
2863         if (rc)
2864                 goto out_free_secdata;
2865
2866         mount_options = opts.mnt_opts;
2867         flags = opts.mnt_opts_flags;
2868
2869         for (i = 0; i < opts.num_mnt_opts; i++) {
2870                 u32 sid;
2871
2872                 if (flags[i] == SBLABEL_MNT)
2873                         continue;
2874                 rc = security_context_str_to_sid(&selinux_state,
2875                                                  mount_options[i], &sid,
2876                                                  GFP_KERNEL);
2877                 if (rc) {
2878                         printk(KERN_WARNING "SELinux: security_context_str_to_sid"
2879                                "(%s) failed for (dev %s, type %s) errno=%d\n",
2880                                mount_options[i], sb->s_id, sb->s_type->name, rc);
2881                         goto out_free_opts;
2882                 }
2883                 rc = -EINVAL;
2884                 switch (flags[i]) {
2885                 case FSCONTEXT_MNT:
2886                         if (bad_option(sbsec, FSCONTEXT_MNT, sbsec->sid, sid))
2887                                 goto out_bad_option;
2888                         break;
2889                 case CONTEXT_MNT:
2890                         if (bad_option(sbsec, CONTEXT_MNT, sbsec->mntpoint_sid, sid))
2891                                 goto out_bad_option;
2892                         break;
2893                 case ROOTCONTEXT_MNT: {
2894                         struct inode_security_struct *root_isec;
2895                         root_isec = backing_inode_security(sb->s_root);
2896
2897                         if (bad_option(sbsec, ROOTCONTEXT_MNT, root_isec->sid, sid))
2898                                 goto out_bad_option;
2899                         break;
2900                 }
2901                 case DEFCONTEXT_MNT:
2902                         if (bad_option(sbsec, DEFCONTEXT_MNT, sbsec->def_sid, sid))
2903                                 goto out_bad_option;
2904                         break;
2905                 default:
2906                         goto out_free_opts;
2907                 }
2908         }
2909
2910         rc = 0;
2911 out_free_opts:
2912         security_free_mnt_opts(&opts);
2913 out_free_secdata:
2914         free_secdata(secdata);
2915         return rc;
2916 out_bad_option:
2917         printk(KERN_WARNING "SELinux: unable to change security options "
2918                "during remount (dev %s, type=%s)\n", sb->s_id,
2919                sb->s_type->name);
2920         goto out_free_opts;
2921 }
2922
2923 static int selinux_sb_kern_mount(struct super_block *sb, int flags, void *data)
2924 {
2925         const struct cred *cred = current_cred();
2926         struct common_audit_data ad;
2927         int rc;
2928
2929         rc = superblock_doinit(sb, data);
2930         if (rc)
2931                 return rc;
2932
2933         /* Allow all mounts performed by the kernel */
2934         if (flags & MS_KERNMOUNT)
2935                 return 0;
2936
2937         ad.type = LSM_AUDIT_DATA_DENTRY;
2938         ad.u.dentry = sb->s_root;
2939         return superblock_has_perm(cred, sb, FILESYSTEM__MOUNT, &ad);
2940 }
2941
2942 static int selinux_sb_statfs(struct dentry *dentry)
2943 {
2944         const struct cred *cred = current_cred();
2945         struct common_audit_data ad;
2946
2947         ad.type = LSM_AUDIT_DATA_DENTRY;
2948         ad.u.dentry = dentry->d_sb->s_root;
2949         return superblock_has_perm(cred, dentry->d_sb, FILESYSTEM__GETATTR, &ad);
2950 }
2951
2952 static int selinux_mount(const char *dev_name,
2953                          const struct path *path,
2954                          const char *type,
2955                          unsigned long flags,
2956                          void *data)
2957 {
2958         const struct cred *cred = current_cred();
2959
2960         if (flags & MS_REMOUNT)
2961                 return superblock_has_perm(cred, path->dentry->d_sb,
2962                                            FILESYSTEM__REMOUNT, NULL);
2963         else
2964                 return path_has_perm(cred, path, FILE__MOUNTON);
2965 }
2966
2967 static int selinux_umount(struct vfsmount *mnt, int flags)
2968 {
2969         const struct cred *cred = current_cred();
2970
2971         return superblock_has_perm(cred, mnt->mnt_sb,
2972                                    FILESYSTEM__UNMOUNT, NULL);
2973 }
2974
2975 /* inode security operations */
2976
2977 static int selinux_inode_alloc_security(struct inode *inode)
2978 {
2979         return inode_alloc_security(inode);
2980 }
2981
2982 static void selinux_inode_free_security(struct inode *inode)
2983 {
2984         inode_free_security(inode);
2985 }
2986
2987 static int selinux_dentry_init_security(struct dentry *dentry, int mode,
2988                                         const struct qstr *name, void **ctx,
2989                                         u32 *ctxlen)
2990 {
2991         u32 newsid;
2992         int rc;
2993
2994         rc = selinux_determine_inode_label(current_security(),
2995                                            d_inode(dentry->d_parent), name,
2996                                            inode_mode_to_security_class(mode),
2997                                            &newsid);
2998         if (rc)
2999                 return rc;
3000
3001         return security_sid_to_context(&selinux_state, newsid, (char **)ctx,
3002                                        ctxlen);
3003 }
3004
3005 static int selinux_dentry_create_files_as(struct dentry *dentry, int mode,
3006                                           struct qstr *name,
3007                                           const struct cred *old,
3008                                           struct cred *new)
3009 {
3010         u32 newsid;
3011         int rc;
3012         struct task_security_struct *tsec;
3013
3014         rc = selinux_determine_inode_label(old->security,
3015                                            d_inode(dentry->d_parent), name,
3016                                            inode_mode_to_security_class(mode),
3017                                            &newsid);
3018         if (rc)
3019                 return rc;
3020
3021         tsec = new->security;
3022         tsec->create_sid = newsid;
3023         return 0;
3024 }
3025
3026 static int selinux_inode_init_security(struct inode *inode, struct inode *dir,
3027                                        const struct qstr *qstr,
3028                                        const char **name,
3029                                        void **value, size_t *len)
3030 {
3031         const struct task_security_struct *tsec = current_security();
3032         struct superblock_security_struct *sbsec;
3033         u32 newsid, clen;
3034         int rc;
3035         char *context;
3036
3037         sbsec = dir->i_sb->s_security;
3038
3039         newsid = tsec->create_sid;
3040
3041         rc = selinux_determine_inode_label(current_security(),
3042                 dir, qstr,
3043                 inode_mode_to_security_class(inode->i_mode),
3044                 &newsid);
3045         if (rc)
3046                 return rc;
3047
3048         /* Possibly defer initialization to selinux_complete_init. */
3049         if (sbsec->flags & SE_SBINITIALIZED) {
3050                 struct inode_security_struct *isec = inode->i_security;
3051                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
3052                 isec->sid = newsid;
3053                 isec->initialized = LABEL_INITIALIZED;
3054         }
3055
3056         if (!selinux_state.initialized || !(sbsec->flags & SBLABEL_MNT))
3057                 return -EOPNOTSUPP;
3058
3059         if (name)
3060                 *name = XATTR_SELINUX_SUFFIX;
3061
3062         if (value && len) {
3063                 rc = security_sid_to_context_force(&selinux_state, newsid,
3064                                                    &context, &clen);
3065                 if (rc)
3066                         return rc;
3067                 *value = context;
3068                 *len = clen;
3069         }
3070
3071         return 0;
3072 }
3073
3074 static int selinux_inode_create(struct inode *dir, struct dentry *dentry, umode_t mode)
3075 {
3076         return may_create(dir, dentry, SECCLASS_FILE);
3077 }
3078
3079 static int selinux_inode_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3080 {
3081         return may_link(dir, old_dentry, MAY_LINK);
3082 }
3083
3084 static int selinux_inode_unlink(struct inode *dir, struct dentry *dentry)
3085 {
3086         return may_link(dir, dentry, MAY_UNLINK);
3087 }
3088
3089 static int selinux_inode_symlink(struct inode *dir, struct dentry *dentry, const char *name)
3090 {
3091         return may_create(dir, dentry, SECCLASS_LNK_FILE);
3092 }
3093
3094 static int selinux_inode_mkdir(struct inode *dir, struct dentry *dentry, umode_t mask)
3095 {
3096         return may_create(dir, dentry, SECCLASS_DIR);
3097 }
3098
3099 static int selinux_inode_rmdir(struct inode *dir, struct dentry *dentry)
3100 {
3101         return may_link(dir, dentry, MAY_RMDIR);
3102 }
3103
3104 static int selinux_inode_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3105 {
3106         return may_create(dir, dentry, inode_mode_to_security_class(mode));
3107 }
3108
3109 static int selinux_inode_rename(struct inode *old_inode, struct dentry *old_dentry,
3110                                 struct inode *new_inode, struct dentry *new_dentry)
3111 {
3112         return may_rename(old_inode, old_dentry, new_inode, new_dentry);
3113 }
3114
3115 static int selinux_inode_readlink(struct dentry *dentry)
3116 {
3117         const struct cred *cred = current_cred();
3118
3119         return dentry_has_perm(cred, dentry, FILE__READ);
3120 }
3121
3122 static int selinux_inode_follow_link(struct dentry *dentry, struct inode *inode,
3123                                      bool rcu)
3124 {
3125         const struct cred *cred = current_cred();
3126         struct common_audit_data ad;
3127         struct inode_security_struct *isec;
3128         u32 sid;
3129
3130         validate_creds(cred);
3131
3132         ad.type = LSM_AUDIT_DATA_DENTRY;
3133         ad.u.dentry = dentry;
3134         sid = cred_sid(cred);
3135         isec = inode_security_rcu(inode, rcu);
3136         if (IS_ERR(isec))
3137                 return PTR_ERR(isec);
3138
3139         return avc_has_perm_flags(&selinux_state,
3140                                   sid, isec->sid, isec->sclass, FILE__READ, &ad,
3141                                   rcu ? MAY_NOT_BLOCK : 0);
3142 }
3143
3144 static noinline int audit_inode_permission(struct inode *inode,
3145                                            u32 perms, u32 audited, u32 denied,
3146                                            int result,
3147                                            unsigned flags)
3148 {
3149         struct common_audit_data ad;
3150         struct inode_security_struct *isec = inode->i_security;
3151         int rc;
3152
3153         ad.type = LSM_AUDIT_DATA_INODE;
3154         ad.u.inode = inode;
3155
3156         rc = slow_avc_audit(&selinux_state,
3157                             current_sid(), isec->sid, isec->sclass, perms,
3158                             audited, denied, result, &ad, flags);
3159         if (rc)
3160                 return rc;
3161         return 0;
3162 }
3163
3164 static int selinux_inode_permission(struct inode *inode, int mask)
3165 {
3166         const struct cred *cred = current_cred();
3167         u32 perms;
3168         bool from_access;
3169         unsigned flags = mask & MAY_NOT_BLOCK;
3170         struct inode_security_struct *isec;
3171         u32 sid;
3172         struct av_decision avd;
3173         int rc, rc2;
3174         u32 audited, denied;
3175
3176         from_access = mask & MAY_ACCESS;
3177         mask &= (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND);
3178
3179         /* No permission to check.  Existence test. */
3180         if (!mask)
3181                 return 0;
3182
3183         validate_creds(cred);
3184
3185         if (unlikely(IS_PRIVATE(inode)))
3186                 return 0;
3187
3188         perms = file_mask_to_av(inode->i_mode, mask);
3189
3190         sid = cred_sid(cred);
3191         isec = inode_security_rcu(inode, flags & MAY_NOT_BLOCK);
3192         if (IS_ERR(isec))
3193                 return PTR_ERR(isec);
3194
3195         rc = avc_has_perm_noaudit(&selinux_state,
3196                                   sid, isec->sid, isec->sclass, perms, 0, &avd);
3197         audited = avc_audit_required(perms, &avd, rc,
3198                                      from_access ? FILE__AUDIT_ACCESS : 0,
3199                                      &denied);
3200         if (likely(!audited))
3201                 return rc;
3202
3203         rc2 = audit_inode_permission(inode, perms, audited, denied, rc, flags);
3204         if (rc2)
3205                 return rc2;
3206         return rc;
3207 }
3208
3209 static int selinux_inode_setattr(struct dentry *dentry, struct iattr *iattr)
3210 {
3211         const struct cred *cred = current_cred();
3212         struct inode *inode = d_backing_inode(dentry);
3213         unsigned int ia_valid = iattr->ia_valid;
3214         __u32 av = FILE__WRITE;
3215
3216         /* ATTR_FORCE is just used for ATTR_KILL_S[UG]ID. */
3217         if (ia_valid & ATTR_FORCE) {
3218                 ia_valid &= ~(ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_MODE |
3219                               ATTR_FORCE);
3220                 if (!ia_valid)
3221                         return 0;
3222         }
3223
3224         if (ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID |
3225                         ATTR_ATIME_SET | ATTR_MTIME_SET | ATTR_TIMES_SET))
3226                 return dentry_has_perm(cred, dentry, FILE__SETATTR);
3227
3228         if (selinux_policycap_openperm() &&
3229             inode->i_sb->s_magic != SOCKFS_MAGIC &&
3230             (ia_valid & ATTR_SIZE) &&
3231             !(ia_valid & ATTR_FILE))
3232                 av |= FILE__OPEN;
3233
3234         return dentry_has_perm(cred, dentry, av);
3235 }
3236
3237 static int selinux_inode_getattr(const struct path *path)
3238 {
3239         return path_has_perm(current_cred(), path, FILE__GETATTR);
3240 }
3241
3242 static bool has_cap_mac_admin(bool audit)
3243 {
3244         const struct cred *cred = current_cred();
3245         int cap_audit = audit ? SECURITY_CAP_AUDIT : SECURITY_CAP_NOAUDIT;
3246
3247         if (cap_capable(cred, &init_user_ns, CAP_MAC_ADMIN, cap_audit))
3248                 return false;
3249         if (cred_has_capability(cred, CAP_MAC_ADMIN, cap_audit, true))
3250                 return false;
3251         return true;
3252 }
3253
3254 static int selinux_inode_setxattr(struct dentry *dentry, const char *name,
3255                                   const void *value, size_t size, int flags)
3256 {
3257         struct inode *inode = d_backing_inode(dentry);
3258         struct inode_security_struct *isec;
3259         struct superblock_security_struct *sbsec;
3260         struct common_audit_data ad;
3261         u32 newsid, sid = current_sid();
3262         int rc = 0;
3263
3264         if (strcmp(name, XATTR_NAME_SELINUX)) {
3265                 rc = cap_inode_setxattr(dentry, name, value, size, flags);
3266                 if (rc)
3267                         return rc;
3268
3269                 /* Not an attribute we recognize, so just check the
3270                    ordinary setattr permission. */
3271                 return dentry_has_perm(current_cred(), dentry, FILE__SETATTR);
3272         }
3273
3274         sbsec = inode->i_sb->s_security;
3275         if (!(sbsec->flags & SBLABEL_MNT))
3276                 return -EOPNOTSUPP;
3277
3278         if (!inode_owner_or_capable(inode))
3279                 return -EPERM;
3280
3281         ad.type = LSM_AUDIT_DATA_DENTRY;
3282         ad.u.dentry = dentry;
3283
3284         isec = backing_inode_security(dentry);
3285         rc = avc_has_perm(&selinux_state,
3286                           sid, isec->sid, isec->sclass,
3287                           FILE__RELABELFROM, &ad);
3288         if (rc)
3289                 return rc;
3290
3291         rc = security_context_to_sid(&selinux_state, value, size, &newsid,
3292                                      GFP_KERNEL);
3293         if (rc == -EINVAL) {
3294                 if (!has_cap_mac_admin(true)) {
3295                         struct audit_buffer *ab;
3296                         size_t audit_size;
3297
3298                         /* We strip a nul only if it is at the end, otherwise the
3299                          * context contains a nul and we should audit that */
3300                         if (value) {
3301                                 const char *str = value;
3302
3303                                 if (str[size - 1] == '\0')
3304                                         audit_size = size - 1;
3305                                 else
3306                                         audit_size = size;
3307                         } else {
3308                                 audit_size = 0;
3309                         }
3310                         ab = audit_log_start(audit_context(),
3311                                              GFP_ATOMIC, AUDIT_SELINUX_ERR);
3312                         audit_log_format(ab, "op=setxattr invalid_context=");
3313                         audit_log_n_untrustedstring(ab, value, audit_size);
3314                         audit_log_end(ab);
3315
3316                         return rc;
3317                 }
3318                 rc = security_context_to_sid_force(&selinux_state, value,
3319                                                    size, &newsid);
3320         }
3321         if (rc)
3322                 return rc;
3323
3324         rc = avc_has_perm(&selinux_state,
3325                           sid, newsid, isec->sclass,
3326                           FILE__RELABELTO, &ad);
3327         if (rc)
3328                 return rc;
3329
3330         rc = security_validate_transition(&selinux_state, isec->sid, newsid,
3331                                           sid, isec->sclass);
3332         if (rc)
3333                 return rc;
3334
3335         return avc_has_perm(&selinux_state,
3336                             newsid,
3337                             sbsec->sid,
3338                             SECCLASS_FILESYSTEM,
3339                             FILESYSTEM__ASSOCIATE,
3340                             &ad);
3341 }
3342
3343 static void selinux_inode_post_setxattr(struct dentry *dentry, const char *name,
3344                                         const void *value, size_t size,
3345                                         int flags)
3346 {
3347         struct inode *inode = d_backing_inode(dentry);
3348         struct inode_security_struct *isec;
3349         u32 newsid;
3350         int rc;
3351
3352         if (strcmp(name, XATTR_NAME_SELINUX)) {
3353                 /* Not an attribute we recognize, so nothing to do. */
3354                 return;
3355         }
3356
3357         rc = security_context_to_sid_force(&selinux_state, value, size,
3358                                            &newsid);
3359         if (rc) {
3360                 printk(KERN_ERR "SELinux:  unable to map context to SID"
3361                        "for (%s, %lu), rc=%d\n",
3362                        inode->i_sb->s_id, inode->i_ino, -rc);
3363                 return;
3364         }
3365
3366         isec = backing_inode_security(dentry);
3367         spin_lock(&isec->lock);
3368         isec->sclass = inode_mode_to_security_class(inode->i_mode);
3369         isec->sid = newsid;
3370         isec->initialized = LABEL_INITIALIZED;
3371         spin_unlock(&isec->lock);
3372
3373         return;
3374 }
3375
3376 static int selinux_inode_getxattr(struct dentry *dentry, const char *name)
3377 {
3378         const struct cred *cred = current_cred();
3379
3380         return dentry_has_perm(cred, dentry, FILE__GETATTR);
3381 }
3382
3383 static int selinux_inode_listxattr(struct dentry *dentry)
3384 {
3385         const struct cred *cred = current_cred();
3386
3387         return dentry_has_perm(cred, dentry, FILE__GETATTR);
3388 }
3389
3390 static int selinux_inode_removexattr(struct dentry *dentry, const char *name)
3391 {
3392         if (strcmp(name, XATTR_NAME_SELINUX)) {
3393                 int rc = cap_inode_removexattr(dentry, name);
3394                 if (rc)
3395                         return rc;
3396
3397                 /* Not an attribute we recognize, so just check the
3398                    ordinary setattr permission. */
3399                 return dentry_has_perm(current_cred(), dentry, FILE__SETATTR);
3400         }
3401
3402         /* No one is allowed to remove a SELinux security label.
3403            You can change the label, but all data must be labeled. */
3404         return -EACCES;
3405 }
3406
3407 /*
3408  * Copy the inode security context value to the user.
3409  *
3410  * Permission check is handled by selinux_inode_getxattr hook.
3411  */
3412 static int selinux_inode_getsecurity(struct inode *inode, const char *name, void **buffer, bool alloc)
3413 {
3414         u32 size;
3415         int error;
3416         char *context = NULL;
3417         struct inode_security_struct *isec;
3418
3419         if (strcmp(name, XATTR_SELINUX_SUFFIX))
3420                 return -EOPNOTSUPP;
3421
3422         /*
3423          * If the caller has CAP_MAC_ADMIN, then get the raw context
3424          * value even if it is not defined by current policy; otherwise,
3425          * use the in-core value under current policy.
3426          * Use the non-auditing forms of the permission checks since
3427          * getxattr may be called by unprivileged processes commonly
3428          * and lack of permission just means that we fall back to the
3429          * in-core context value, not a denial.
3430          */
3431         isec = inode_security(inode);
3432         if (has_cap_mac_admin(false))
3433                 error = security_sid_to_context_force(&selinux_state,
3434                                                       isec->sid, &context,
3435                                                       &size);
3436         else
3437                 error = security_sid_to_context(&selinux_state, isec->sid,
3438                                                 &context, &size);
3439         if (error)
3440                 return error;
3441         error = size;
3442         if (alloc) {
3443                 *buffer = context;
3444                 goto out_nofree;
3445         }
3446         kfree(context);
3447 out_nofree:
3448         return error;
3449 }
3450
3451 static int selinux_inode_setsecurity(struct inode *inode, const char *name,
3452                                      const void *value, size_t size, int flags)
3453 {
3454         struct inode_security_struct *isec = inode_security_novalidate(inode);
3455         u32 newsid;
3456         int rc;
3457
3458         if (strcmp(name, XATTR_SELINUX_SUFFIX))
3459                 return -EOPNOTSUPP;
3460
3461         if (!value || !size)
3462                 return -EACCES;
3463
3464         rc = security_context_to_sid(&selinux_state, value, size, &newsid,
3465                                      GFP_KERNEL);
3466         if (rc)
3467                 return rc;
3468
3469         spin_lock(&isec->lock);
3470         isec->sclass = inode_mode_to_security_class(inode->i_mode);
3471         isec->sid = newsid;
3472         isec->initialized = LABEL_INITIALIZED;
3473         spin_unlock(&isec->lock);
3474         return 0;
3475 }
3476
3477 static int selinux_inode_listsecurity(struct inode *inode, char *buffer, size_t buffer_size)
3478 {
3479         const int len = sizeof(XATTR_NAME_SELINUX);
3480         if (buffer && len <= buffer_size)
3481                 memcpy(buffer, XATTR_NAME_SELINUX, len);
3482         return len;
3483 }
3484
3485 static void selinux_inode_getsecid(struct inode *inode, u32 *secid)
3486 {
3487         struct inode_security_struct *isec = inode_security_novalidate(inode);
3488         *secid = isec->sid;
3489 }
3490
3491 static int selinux_inode_copy_up(struct dentry *src, struct cred **new)
3492 {
3493         u32 sid;
3494         struct task_security_struct *tsec;
3495         struct cred *new_creds = *new;
3496
3497         if (new_creds == NULL) {
3498                 new_creds = prepare_creds();
3499                 if (!new_creds)
3500                         return -ENOMEM;
3501         }
3502
3503         tsec = new_creds->security;
3504         /* Get label from overlay inode and set it in create_sid */
3505         selinux_inode_getsecid(d_inode(src), &sid);
3506         tsec->create_sid = sid;
3507         *new = new_creds;
3508         return 0;
3509 }
3510
3511 static int selinux_inode_copy_up_xattr(const char *name)
3512 {
3513         /* The copy_up hook above sets the initial context on an inode, but we
3514          * don't then want to overwrite it by blindly copying all the lower
3515          * xattrs up.  Instead, we have to filter out SELinux-related xattrs.
3516          */
3517         if (strcmp(name, XATTR_NAME_SELINUX) == 0)
3518                 return 1; /* Discard */
3519         /*
3520          * Any other attribute apart from SELINUX is not claimed, supported
3521          * by selinux.
3522          */
3523         return -EOPNOTSUPP;
3524 }
3525
3526 /* file security operations */
3527
3528 static int selinux_revalidate_file_permission(struct file *file, int mask)
3529 {
3530         const struct cred *cred = current_cred();
3531         struct inode *inode = file_inode(file);
3532
3533         /* file_mask_to_av won't add FILE__WRITE if MAY_APPEND is set */
3534         if ((file->f_flags & O_APPEND) && (mask & MAY_WRITE))
3535                 mask |= MAY_APPEND;
3536
3537         return file_has_perm(cred, file,
3538                              file_mask_to_av(inode->i_mode, mask));
3539 }
3540
3541 static int selinux_file_permission(struct file *file, int mask)
3542 {
3543         struct inode *inode = file_inode(file);
3544         struct file_security_struct *fsec = file->f_security;
3545         struct inode_security_struct *isec;
3546         u32 sid = current_sid();
3547
3548         if (!mask)
3549                 /* No permission to check.  Existence test. */
3550                 return 0;
3551
3552         isec = inode_security(inode);
3553         if (sid == fsec->sid && fsec->isid == isec->sid &&
3554             fsec->pseqno == avc_policy_seqno(&selinux_state))
3555                 /* No change since file_open check. */
3556                 return 0;
3557
3558         return selinux_revalidate_file_permission(file, mask);
3559 }
3560
3561 static int selinux_file_alloc_security(struct file *file)
3562 {
3563         return file_alloc_security(file);
3564 }
3565
3566 static void selinux_file_free_security(struct file *file)
3567 {
3568         file_free_security(file);
3569 }
3570
3571 /*
3572  * Check whether a task has the ioctl permission and cmd
3573  * operation to an inode.
3574  */
3575 static int ioctl_has_perm(const struct cred *cred, struct file *file,
3576                 u32 requested, u16 cmd)
3577 {
3578         struct common_audit_data ad;
3579         struct file_security_struct *fsec = file->f_security;
3580         struct inode *inode = file_inode(file);
3581         struct inode_security_struct *isec;
3582         struct lsm_ioctlop_audit ioctl;
3583         u32 ssid = cred_sid(cred);
3584         int rc;
3585         u8 driver = cmd >> 8;
3586         u8 xperm = cmd & 0xff;
3587
3588         ad.type = LSM_AUDIT_DATA_IOCTL_OP;
3589         ad.u.op = &ioctl;
3590         ad.u.op->cmd = cmd;
3591         ad.u.op->path = file->f_path;
3592
3593         if (ssid != fsec->sid) {
3594                 rc = avc_has_perm(&selinux_state,
3595                                   ssid, fsec->sid,
3596                                 SECCLASS_FD,
3597                                 FD__USE,
3598                                 &ad);
3599                 if (rc)
3600                         goto out;
3601         }
3602
3603         if (unlikely(IS_PRIVATE(inode)))
3604                 return 0;
3605
3606         isec = inode_security(inode);
3607         rc = avc_has_extended_perms(&selinux_state,
3608                                     ssid, isec->sid, isec->sclass,
3609                                     requested, driver, xperm, &ad);
3610 out:
3611         return rc;
3612 }
3613
3614 static int selinux_file_ioctl(struct file *file, unsigned int cmd,
3615                               unsigned long arg)
3616 {
3617         const struct cred *cred = current_cred();
3618         int error = 0;
3619
3620         switch (cmd) {
3621         case FIONREAD:
3622         /* fall through */
3623         case FIBMAP:
3624         /* fall through */
3625         case FIGETBSZ:
3626         /* fall through */
3627         case FS_IOC_GETFLAGS:
3628         /* fall through */
3629         case FS_IOC_GETVERSION:
3630                 error = file_has_perm(cred, file, FILE__GETATTR);
3631                 break;
3632
3633         case FS_IOC_SETFLAGS:
3634         /* fall through */
3635         case FS_IOC_SETVERSION:
3636                 error = file_has_perm(cred, file, FILE__SETATTR);
3637                 break;
3638
3639         /* sys_ioctl() checks */
3640         case FIONBIO:
3641         /* fall through */
3642         case FIOASYNC:
3643                 error = file_has_perm(cred, file, 0);
3644                 break;
3645
3646         case KDSKBENT:
3647         case KDSKBSENT:
3648                 error = cred_has_capability(cred, CAP_SYS_TTY_CONFIG,
3649                                             SECURITY_CAP_AUDIT, true);
3650                 break;
3651
3652         /* default case assumes that the command will go
3653          * to the file's ioctl() function.
3654          */
3655         default:
3656                 error = ioctl_has_perm(cred, file, FILE__IOCTL, (u16) cmd);
3657         }
3658         return error;
3659 }
3660
3661 static int default_noexec;
3662
3663 static int file_map_prot_check(struct file *file, unsigned long prot, int shared)
3664 {
3665         const struct cred *cred = current_cred();
3666         u32 sid = cred_sid(cred);
3667         int rc = 0;
3668
3669         if (default_noexec &&
3670             (prot & PROT_EXEC) && (!file || IS_PRIVATE(file_inode(file)) ||
3671                                    (!shared && (prot & PROT_WRITE)))) {
3672                 /*
3673                  * We are making executable an anonymous mapping or a
3674                  * private file mapping that will also be writable.
3675                  * This has an additional check.
3676                  */
3677                 rc = avc_has_perm(&selinux_state,
3678                                   sid, sid, SECCLASS_PROCESS,
3679                                   PROCESS__EXECMEM, NULL);
3680                 if (rc)
3681                         goto error;
3682         }
3683
3684         if (file) {
3685                 /* read access is always possible with a mapping */
3686                 u32 av = FILE__READ;
3687
3688                 /* write access only matters if the mapping is shared */
3689                 if (shared && (prot & PROT_WRITE))
3690                         av |= FILE__WRITE;
3691
3692                 if (prot & PROT_EXEC)
3693                         av |= FILE__EXECUTE;
3694
3695                 return file_has_perm(cred, file, av);
3696         }
3697
3698 error:
3699         return rc;
3700 }
3701
3702 static int selinux_mmap_addr(unsigned long addr)
3703 {
3704         int rc = 0;
3705
3706         if (addr < CONFIG_LSM_MMAP_MIN_ADDR) {
3707                 u32 sid = current_sid();
3708                 rc = avc_has_perm(&selinux_state,
3709                                   sid, sid, SECCLASS_MEMPROTECT,
3710                                   MEMPROTECT__MMAP_ZERO, NULL);
3711         }
3712
3713         return rc;
3714 }
3715
3716 static int selinux_mmap_file(struct file *file, unsigned long reqprot,
3717                              unsigned long prot, unsigned long flags)
3718 {
3719         struct common_audit_data ad;
3720         int rc;
3721
3722         if (file) {
3723                 ad.type = LSM_AUDIT_DATA_FILE;
3724                 ad.u.file = file;
3725                 rc = inode_has_perm(current_cred(), file_inode(file),
3726                                     FILE__MAP, &ad);
3727                 if (rc)
3728                         return rc;
3729         }
3730
3731         if (selinux_state.checkreqprot)
3732                 prot = reqprot;
3733
3734         return file_map_prot_check(file, prot,
3735                                    (flags & MAP_TYPE) == MAP_SHARED);
3736 }
3737
3738 static int selinux_file_mprotect(struct vm_area_struct *vma,
3739                                  unsigned long reqprot,
3740                                  unsigned long prot)
3741 {
3742         const struct cred *cred = current_cred();
3743         u32 sid = cred_sid(cred);
3744
3745         if (selinux_state.checkreqprot)
3746                 prot = reqprot;
3747
3748         if (default_noexec &&
3749             (prot & PROT_EXEC) && !(vma->vm_flags & VM_EXEC)) {
3750                 int rc = 0;
3751                 if (vma->vm_start >= vma->vm_mm->start_brk &&
3752                     vma->vm_end <= vma->vm_mm->brk) {
3753                         rc = avc_has_perm(&selinux_state,
3754                                           sid, sid, SECCLASS_PROCESS,
3755                                           PROCESS__EXECHEAP, NULL);
3756                 } else if (!vma->vm_file &&
3757                            ((vma->vm_start <= vma->vm_mm->start_stack &&
3758                              vma->vm_end >= vma->vm_mm->start_stack) ||
3759                             vma_is_stack_for_current(vma))) {
3760                         rc = avc_has_perm(&selinux_state,
3761                                           sid, sid, SECCLASS_PROCESS,
3762                                           PROCESS__EXECSTACK, NULL);
3763                 } else if (vma->vm_file && vma->anon_vma) {
3764                         /*
3765                          * We are making executable a file mapping that has
3766                          * had some COW done. Since pages might have been
3767                          * written, check ability to execute the possibly
3768                          * modified content.  This typically should only
3769                          * occur for text relocations.
3770                          */
3771                         rc = file_has_perm(cred, vma->vm_file, FILE__EXECMOD);
3772                 }
3773                 if (rc)
3774                         return rc;
3775         }
3776
3777         return file_map_prot_check(vma->vm_file, prot, vma->vm_flags&VM_SHARED);
3778 }
3779
3780 static int selinux_file_lock(struct file *file, unsigned int cmd)
3781 {
3782         const struct cred *cred = current_cred();
3783
3784         return file_has_perm(cred, file, FILE__LOCK);
3785 }
3786
3787 static int selinux_file_fcntl(struct file *file, unsigned int cmd,
3788                               unsigned long arg)
3789 {
3790         const struct cred *cred = current_cred();
3791         int err = 0;
3792
3793         switch (cmd) {
3794         case F_SETFL:
3795                 if ((file->f_flags & O_APPEND) && !(arg & O_APPEND)) {
3796                         err = file_has_perm(cred, file, FILE__WRITE);
3797                         break;
3798                 }
3799                 /* fall through */
3800         case F_SETOWN:
3801         case F_SETSIG:
3802         case F_GETFL:
3803         case F_GETOWN:
3804         case F_GETSIG:
3805         case F_GETOWNER_UIDS:
3806                 /* Just check FD__USE permission */
3807                 err = file_has_perm(cred, file, 0);
3808                 break;
3809         case F_GETLK:
3810         case F_SETLK:
3811         case F_SETLKW:
3812         case F_OFD_GETLK:
3813         case F_OFD_SETLK:
3814         case F_OFD_SETLKW:
3815 #if BITS_PER_LONG == 32
3816         case F_GETLK64:
3817         case F_SETLK64:
3818         case F_SETLKW64:
3819 #endif
3820                 err = file_has_perm(cred, file, FILE__LOCK);
3821                 break;
3822         }
3823
3824         return err;
3825 }
3826
3827 static void selinux_file_set_fowner(struct file *file)
3828 {
3829         struct file_security_struct *fsec;
3830
3831         fsec = file->f_security;
3832         fsec->fown_sid = current_sid();
3833 }
3834
3835 static int selinux_file_send_sigiotask(struct task_struct *tsk,
3836                                        struct fown_struct *fown, int signum)
3837 {
3838         struct file *file;
3839         u32 sid = task_sid(tsk);
3840         u32 perm;
3841         struct file_security_struct *fsec;
3842
3843         /* struct fown_struct is never outside the context of a struct file */
3844         file = container_of(fown, struct file, f_owner);
3845
3846         fsec = file->f_security;
3847
3848         if (!signum)