Linux 5.9-rc6
[sfrench/cifs-2.6.git] / security / selinux / hooks.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  NSA Security-Enhanced Linux (SELinux) security module
4  *
5  *  This file contains the SELinux hook function implementations.
6  *
7  *  Authors:  Stephen Smalley, <sds@tycho.nsa.gov>
8  *            Chris Vance, <cvance@nai.com>
9  *            Wayne Salamon, <wsalamon@nai.com>
10  *            James Morris <jmorris@redhat.com>
11  *
12  *  Copyright (C) 2001,2002 Networks Associates Technology, Inc.
13  *  Copyright (C) 2003-2008 Red Hat, Inc., James Morris <jmorris@redhat.com>
14  *                                         Eric Paris <eparis@redhat.com>
15  *  Copyright (C) 2004-2005 Trusted Computer Solutions, Inc.
16  *                          <dgoeddel@trustedcs.com>
17  *  Copyright (C) 2006, 2007, 2009 Hewlett-Packard Development Company, L.P.
18  *      Paul Moore <paul@paul-moore.com>
19  *  Copyright (C) 2007 Hitachi Software Engineering Co., Ltd.
20  *                     Yuichi Nakamura <ynakam@hitachisoft.jp>
21  *  Copyright (C) 2016 Mellanox Technologies
22  */
23
24 #include <linux/init.h>
25 #include <linux/kd.h>
26 #include <linux/kernel.h>
27 #include <linux/tracehook.h>
28 #include <linux/errno.h>
29 #include <linux/sched/signal.h>
30 #include <linux/sched/task.h>
31 #include <linux/lsm_hooks.h>
32 #include <linux/xattr.h>
33 #include <linux/capability.h>
34 #include <linux/unistd.h>
35 #include <linux/mm.h>
36 #include <linux/mman.h>
37 #include <linux/slab.h>
38 #include <linux/pagemap.h>
39 #include <linux/proc_fs.h>
40 #include <linux/swap.h>
41 #include <linux/spinlock.h>
42 #include <linux/syscalls.h>
43 #include <linux/dcache.h>
44 #include <linux/file.h>
45 #include <linux/fdtable.h>
46 #include <linux/namei.h>
47 #include <linux/mount.h>
48 #include <linux/fs_context.h>
49 #include <linux/fs_parser.h>
50 #include <linux/netfilter_ipv4.h>
51 #include <linux/netfilter_ipv6.h>
52 #include <linux/tty.h>
53 #include <net/icmp.h>
54 #include <net/ip.h>             /* for local_port_range[] */
55 #include <net/tcp.h>            /* struct or_callable used in sock_rcv_skb */
56 #include <net/inet_connection_sock.h>
57 #include <net/net_namespace.h>
58 #include <net/netlabel.h>
59 #include <linux/uaccess.h>
60 #include <asm/ioctls.h>
61 #include <linux/atomic.h>
62 #include <linux/bitops.h>
63 #include <linux/interrupt.h>
64 #include <linux/netdevice.h>    /* for network interface checks */
65 #include <net/netlink.h>
66 #include <linux/tcp.h>
67 #include <linux/udp.h>
68 #include <linux/dccp.h>
69 #include <linux/sctp.h>
70 #include <net/sctp/structs.h>
71 #include <linux/quota.h>
72 #include <linux/un.h>           /* for Unix socket types */
73 #include <net/af_unix.h>        /* for Unix socket types */
74 #include <linux/parser.h>
75 #include <linux/nfs_mount.h>
76 #include <net/ipv6.h>
77 #include <linux/hugetlb.h>
78 #include <linux/personality.h>
79 #include <linux/audit.h>
80 #include <linux/string.h>
81 #include <linux/mutex.h>
82 #include <linux/posix-timers.h>
83 #include <linux/syslog.h>
84 #include <linux/user_namespace.h>
85 #include <linux/export.h>
86 #include <linux/msg.h>
87 #include <linux/shm.h>
88 #include <linux/bpf.h>
89 #include <linux/kernfs.h>
90 #include <linux/stringhash.h>   /* for hashlen_string() */
91 #include <uapi/linux/mount.h>
92 #include <linux/fsnotify.h>
93 #include <linux/fanotify.h>
94
95 #include "avc.h"
96 #include "objsec.h"
97 #include "netif.h"
98 #include "netnode.h"
99 #include "netport.h"
100 #include "ibpkey.h"
101 #include "xfrm.h"
102 #include "netlabel.h"
103 #include "audit.h"
104 #include "avc_ss.h"
105
106 struct selinux_state selinux_state;
107
108 /* SECMARK reference count */
109 static atomic_t selinux_secmark_refcount = ATOMIC_INIT(0);
110
111 #ifdef CONFIG_SECURITY_SELINUX_DEVELOP
112 static int selinux_enforcing_boot __initdata;
113
114 static int __init enforcing_setup(char *str)
115 {
116         unsigned long enforcing;
117         if (!kstrtoul(str, 0, &enforcing))
118                 selinux_enforcing_boot = enforcing ? 1 : 0;
119         return 1;
120 }
121 __setup("enforcing=", enforcing_setup);
122 #else
123 #define selinux_enforcing_boot 1
124 #endif
125
126 int selinux_enabled_boot __initdata = 1;
127 #ifdef CONFIG_SECURITY_SELINUX_BOOTPARAM
128 static int __init selinux_enabled_setup(char *str)
129 {
130         unsigned long enabled;
131         if (!kstrtoul(str, 0, &enabled))
132                 selinux_enabled_boot = enabled ? 1 : 0;
133         return 1;
134 }
135 __setup("selinux=", selinux_enabled_setup);
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                 if (checkreqprot)
148                         pr_warn("SELinux: checkreqprot set to 1 via kernel parameter.  This is deprecated and will be rejected in a future kernel release.\n");
149         }
150         return 1;
151 }
152 __setup("checkreqprot=", checkreqprot_setup);
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_blocking_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 = selinux_cred(cred);
217         tsec->osid = tsec->sid = SECINITSID_KERNEL;
218 }
219
220 /*
221  * get the security ID of a set of credentials
222  */
223 static inline u32 cred_sid(const struct cred *cred)
224 {
225         const struct task_security_struct *tsec;
226
227         tsec = selinux_cred(cred);
228         return tsec->sid;
229 }
230
231 /*
232  * get the objective security ID of a task
233  */
234 static inline u32 task_sid(const struct task_struct *task)
235 {
236         u32 sid;
237
238         rcu_read_lock();
239         sid = cred_sid(__task_cred(task));
240         rcu_read_unlock();
241         return sid;
242 }
243
244 static int inode_doinit_with_dentry(struct inode *inode, struct dentry *opt_dentry);
245
246 /*
247  * Try reloading inode security labels that have been marked as invalid.  The
248  * @may_sleep parameter indicates when sleeping and thus reloading labels is
249  * allowed; when set to false, returns -ECHILD when the label is
250  * invalid.  The @dentry parameter should be set to a dentry of the inode.
251  */
252 static int __inode_security_revalidate(struct inode *inode,
253                                        struct dentry *dentry,
254                                        bool may_sleep)
255 {
256         struct inode_security_struct *isec = selinux_inode(inode);
257
258         might_sleep_if(may_sleep);
259
260         if (selinux_initialized(&selinux_state) &&
261             isec->initialized != LABEL_INITIALIZED) {
262                 if (!may_sleep)
263                         return -ECHILD;
264
265                 /*
266                  * Try reloading the inode security label.  This will fail if
267                  * @opt_dentry is NULL and no dentry for this inode can be
268                  * found; in that case, continue using the old label.
269                  */
270                 inode_doinit_with_dentry(inode, dentry);
271         }
272         return 0;
273 }
274
275 static struct inode_security_struct *inode_security_novalidate(struct inode *inode)
276 {
277         return selinux_inode(inode);
278 }
279
280 static struct inode_security_struct *inode_security_rcu(struct inode *inode, bool rcu)
281 {
282         int error;
283
284         error = __inode_security_revalidate(inode, NULL, !rcu);
285         if (error)
286                 return ERR_PTR(error);
287         return selinux_inode(inode);
288 }
289
290 /*
291  * Get the security label of an inode.
292  */
293 static struct inode_security_struct *inode_security(struct inode *inode)
294 {
295         __inode_security_revalidate(inode, NULL, true);
296         return selinux_inode(inode);
297 }
298
299 static struct inode_security_struct *backing_inode_security_novalidate(struct dentry *dentry)
300 {
301         struct inode *inode = d_backing_inode(dentry);
302
303         return selinux_inode(inode);
304 }
305
306 /*
307  * Get the security label of a dentry's backing inode.
308  */
309 static struct inode_security_struct *backing_inode_security(struct dentry *dentry)
310 {
311         struct inode *inode = d_backing_inode(dentry);
312
313         __inode_security_revalidate(inode, dentry, true);
314         return selinux_inode(inode);
315 }
316
317 static void inode_free_security(struct inode *inode)
318 {
319         struct inode_security_struct *isec = selinux_inode(inode);
320         struct superblock_security_struct *sbsec;
321
322         if (!isec)
323                 return;
324         sbsec = inode->i_sb->s_security;
325         /*
326          * As not all inode security structures are in a list, we check for
327          * empty list outside of the lock to make sure that we won't waste
328          * time taking a lock doing nothing.
329          *
330          * The list_del_init() function can be safely called more than once.
331          * It should not be possible for this function to be called with
332          * concurrent list_add(), but for better safety against future changes
333          * in the code, we use list_empty_careful() here.
334          */
335         if (!list_empty_careful(&isec->list)) {
336                 spin_lock(&sbsec->isec_lock);
337                 list_del_init(&isec->list);
338                 spin_unlock(&sbsec->isec_lock);
339         }
340 }
341
342 static void superblock_free_security(struct super_block *sb)
343 {
344         struct superblock_security_struct *sbsec = sb->s_security;
345         sb->s_security = NULL;
346         kfree(sbsec);
347 }
348
349 struct selinux_mnt_opts {
350         const char *fscontext, *context, *rootcontext, *defcontext;
351 };
352
353 static void selinux_free_mnt_opts(void *mnt_opts)
354 {
355         struct selinux_mnt_opts *opts = mnt_opts;
356         kfree(opts->fscontext);
357         kfree(opts->context);
358         kfree(opts->rootcontext);
359         kfree(opts->defcontext);
360         kfree(opts);
361 }
362
363 enum {
364         Opt_error = -1,
365         Opt_context = 0,
366         Opt_defcontext = 1,
367         Opt_fscontext = 2,
368         Opt_rootcontext = 3,
369         Opt_seclabel = 4,
370 };
371
372 #define A(s, has_arg) {#s, sizeof(#s) - 1, Opt_##s, has_arg}
373 static struct {
374         const char *name;
375         int len;
376         int opt;
377         bool has_arg;
378 } tokens[] = {
379         A(context, true),
380         A(fscontext, true),
381         A(defcontext, true),
382         A(rootcontext, true),
383         A(seclabel, false),
384 };
385 #undef A
386
387 static int match_opt_prefix(char *s, int l, char **arg)
388 {
389         int i;
390
391         for (i = 0; i < ARRAY_SIZE(tokens); i++) {
392                 size_t len = tokens[i].len;
393                 if (len > l || memcmp(s, tokens[i].name, len))
394                         continue;
395                 if (tokens[i].has_arg) {
396                         if (len == l || s[len] != '=')
397                                 continue;
398                         *arg = s + len + 1;
399                 } else if (len != l)
400                         continue;
401                 return tokens[i].opt;
402         }
403         return Opt_error;
404 }
405
406 #define SEL_MOUNT_FAIL_MSG "SELinux:  duplicate or incompatible mount options\n"
407
408 static int may_context_mount_sb_relabel(u32 sid,
409                         struct superblock_security_struct *sbsec,
410                         const struct cred *cred)
411 {
412         const struct task_security_struct *tsec = selinux_cred(cred);
413         int rc;
414
415         rc = avc_has_perm(&selinux_state,
416                           tsec->sid, sbsec->sid, SECCLASS_FILESYSTEM,
417                           FILESYSTEM__RELABELFROM, NULL);
418         if (rc)
419                 return rc;
420
421         rc = avc_has_perm(&selinux_state,
422                           tsec->sid, sid, SECCLASS_FILESYSTEM,
423                           FILESYSTEM__RELABELTO, NULL);
424         return rc;
425 }
426
427 static int may_context_mount_inode_relabel(u32 sid,
428                         struct superblock_security_struct *sbsec,
429                         const struct cred *cred)
430 {
431         const struct task_security_struct *tsec = selinux_cred(cred);
432         int rc;
433         rc = avc_has_perm(&selinux_state,
434                           tsec->sid, sbsec->sid, SECCLASS_FILESYSTEM,
435                           FILESYSTEM__RELABELFROM, NULL);
436         if (rc)
437                 return rc;
438
439         rc = avc_has_perm(&selinux_state,
440                           sid, sbsec->sid, SECCLASS_FILESYSTEM,
441                           FILESYSTEM__ASSOCIATE, NULL);
442         return rc;
443 }
444
445 static int selinux_is_genfs_special_handling(struct super_block *sb)
446 {
447         /* Special handling. Genfs but also in-core setxattr handler */
448         return  !strcmp(sb->s_type->name, "sysfs") ||
449                 !strcmp(sb->s_type->name, "pstore") ||
450                 !strcmp(sb->s_type->name, "debugfs") ||
451                 !strcmp(sb->s_type->name, "tracefs") ||
452                 !strcmp(sb->s_type->name, "rootfs") ||
453                 (selinux_policycap_cgroupseclabel() &&
454                  (!strcmp(sb->s_type->name, "cgroup") ||
455                   !strcmp(sb->s_type->name, "cgroup2")));
456 }
457
458 static int selinux_is_sblabel_mnt(struct super_block *sb)
459 {
460         struct superblock_security_struct *sbsec = sb->s_security;
461
462         /*
463          * IMPORTANT: Double-check logic in this function when adding a new
464          * SECURITY_FS_USE_* definition!
465          */
466         BUILD_BUG_ON(SECURITY_FS_USE_MAX != 7);
467
468         switch (sbsec->behavior) {
469         case SECURITY_FS_USE_XATTR:
470         case SECURITY_FS_USE_TRANS:
471         case SECURITY_FS_USE_TASK:
472         case SECURITY_FS_USE_NATIVE:
473                 return 1;
474
475         case SECURITY_FS_USE_GENFS:
476                 return selinux_is_genfs_special_handling(sb);
477
478         /* Never allow relabeling on context mounts */
479         case SECURITY_FS_USE_MNTPOINT:
480         case SECURITY_FS_USE_NONE:
481         default:
482                 return 0;
483         }
484 }
485
486 static int sb_finish_set_opts(struct super_block *sb)
487 {
488         struct superblock_security_struct *sbsec = sb->s_security;
489         struct dentry *root = sb->s_root;
490         struct inode *root_inode = d_backing_inode(root);
491         int rc = 0;
492
493         if (sbsec->behavior == SECURITY_FS_USE_XATTR) {
494                 /* Make sure that the xattr handler exists and that no
495                    error other than -ENODATA is returned by getxattr on
496                    the root directory.  -ENODATA is ok, as this may be
497                    the first boot of the SELinux kernel before we have
498                    assigned xattr values to the filesystem. */
499                 if (!(root_inode->i_opflags & IOP_XATTR)) {
500                         pr_warn("SELinux: (dev %s, type %s) has no "
501                                "xattr support\n", sb->s_id, sb->s_type->name);
502                         rc = -EOPNOTSUPP;
503                         goto out;
504                 }
505
506                 rc = __vfs_getxattr(root, root_inode, XATTR_NAME_SELINUX, NULL, 0);
507                 if (rc < 0 && rc != -ENODATA) {
508                         if (rc == -EOPNOTSUPP)
509                                 pr_warn("SELinux: (dev %s, type "
510                                        "%s) has no security xattr handler\n",
511                                        sb->s_id, sb->s_type->name);
512                         else
513                                 pr_warn("SELinux: (dev %s, type "
514                                        "%s) getxattr errno %d\n", sb->s_id,
515                                        sb->s_type->name, -rc);
516                         goto out;
517                 }
518         }
519
520         sbsec->flags |= SE_SBINITIALIZED;
521
522         /*
523          * Explicitly set or clear SBLABEL_MNT.  It's not sufficient to simply
524          * leave the flag untouched because sb_clone_mnt_opts might be handing
525          * us a superblock that needs the flag to be cleared.
526          */
527         if (selinux_is_sblabel_mnt(sb))
528                 sbsec->flags |= SBLABEL_MNT;
529         else
530                 sbsec->flags &= ~SBLABEL_MNT;
531
532         /* Initialize the root inode. */
533         rc = inode_doinit_with_dentry(root_inode, root);
534
535         /* Initialize any other inodes associated with the superblock, e.g.
536            inodes created prior to initial policy load or inodes created
537            during get_sb by a pseudo filesystem that directly
538            populates itself. */
539         spin_lock(&sbsec->isec_lock);
540         while (!list_empty(&sbsec->isec_head)) {
541                 struct inode_security_struct *isec =
542                                 list_first_entry(&sbsec->isec_head,
543                                            struct inode_security_struct, list);
544                 struct inode *inode = isec->inode;
545                 list_del_init(&isec->list);
546                 spin_unlock(&sbsec->isec_lock);
547                 inode = igrab(inode);
548                 if (inode) {
549                         if (!IS_PRIVATE(inode))
550                                 inode_doinit_with_dentry(inode, NULL);
551                         iput(inode);
552                 }
553                 spin_lock(&sbsec->isec_lock);
554         }
555         spin_unlock(&sbsec->isec_lock);
556 out:
557         return rc;
558 }
559
560 static int bad_option(struct superblock_security_struct *sbsec, char flag,
561                       u32 old_sid, u32 new_sid)
562 {
563         char mnt_flags = sbsec->flags & SE_MNTMASK;
564
565         /* check if the old mount command had the same options */
566         if (sbsec->flags & SE_SBINITIALIZED)
567                 if (!(sbsec->flags & flag) ||
568                     (old_sid != new_sid))
569                         return 1;
570
571         /* check if we were passed the same options twice,
572          * aka someone passed context=a,context=b
573          */
574         if (!(sbsec->flags & SE_SBINITIALIZED))
575                 if (mnt_flags & flag)
576                         return 1;
577         return 0;
578 }
579
580 static int parse_sid(struct super_block *sb, const char *s, u32 *sid)
581 {
582         int rc = security_context_str_to_sid(&selinux_state, s,
583                                              sid, GFP_KERNEL);
584         if (rc)
585                 pr_warn("SELinux: security_context_str_to_sid"
586                        "(%s) failed for (dev %s, type %s) errno=%d\n",
587                        s, sb->s_id, sb->s_type->name, rc);
588         return rc;
589 }
590
591 /*
592  * Allow filesystems with binary mount data to explicitly set mount point
593  * labeling information.
594  */
595 static int selinux_set_mnt_opts(struct super_block *sb,
596                                 void *mnt_opts,
597                                 unsigned long kern_flags,
598                                 unsigned long *set_kern_flags)
599 {
600         const struct cred *cred = current_cred();
601         struct superblock_security_struct *sbsec = sb->s_security;
602         struct dentry *root = sbsec->sb->s_root;
603         struct selinux_mnt_opts *opts = mnt_opts;
604         struct inode_security_struct *root_isec;
605         u32 fscontext_sid = 0, context_sid = 0, rootcontext_sid = 0;
606         u32 defcontext_sid = 0;
607         int rc = 0;
608
609         mutex_lock(&sbsec->lock);
610
611         if (!selinux_initialized(&selinux_state)) {
612                 if (!opts) {
613                         /* Defer initialization until selinux_complete_init,
614                            after the initial policy is loaded and the security
615                            server is ready to handle calls. */
616                         goto out;
617                 }
618                 rc = -EINVAL;
619                 pr_warn("SELinux: Unable to set superblock options "
620                         "before the security server is initialized\n");
621                 goto out;
622         }
623         if (kern_flags && !set_kern_flags) {
624                 /* Specifying internal flags without providing a place to
625                  * place the results is not allowed */
626                 rc = -EINVAL;
627                 goto out;
628         }
629
630         /*
631          * Binary mount data FS will come through this function twice.  Once
632          * from an explicit call and once from the generic calls from the vfs.
633          * Since the generic VFS calls will not contain any security mount data
634          * we need to skip the double mount verification.
635          *
636          * This does open a hole in which we will not notice if the first
637          * mount using this sb set explict options and a second mount using
638          * this sb does not set any security options.  (The first options
639          * will be used for both mounts)
640          */
641         if ((sbsec->flags & SE_SBINITIALIZED) && (sb->s_type->fs_flags & FS_BINARY_MOUNTDATA)
642             && !opts)
643                 goto out;
644
645         root_isec = backing_inode_security_novalidate(root);
646
647         /*
648          * parse the mount options, check if they are valid sids.
649          * also check if someone is trying to mount the same sb more
650          * than once with different security options.
651          */
652         if (opts) {
653                 if (opts->fscontext) {
654                         rc = parse_sid(sb, opts->fscontext, &fscontext_sid);
655                         if (rc)
656                                 goto out;
657                         if (bad_option(sbsec, FSCONTEXT_MNT, sbsec->sid,
658                                         fscontext_sid))
659                                 goto out_double_mount;
660                         sbsec->flags |= FSCONTEXT_MNT;
661                 }
662                 if (opts->context) {
663                         rc = parse_sid(sb, opts->context, &context_sid);
664                         if (rc)
665                                 goto out;
666                         if (bad_option(sbsec, CONTEXT_MNT, sbsec->mntpoint_sid,
667                                         context_sid))
668                                 goto out_double_mount;
669                         sbsec->flags |= CONTEXT_MNT;
670                 }
671                 if (opts->rootcontext) {
672                         rc = parse_sid(sb, opts->rootcontext, &rootcontext_sid);
673                         if (rc)
674                                 goto out;
675                         if (bad_option(sbsec, ROOTCONTEXT_MNT, root_isec->sid,
676                                         rootcontext_sid))
677                                 goto out_double_mount;
678                         sbsec->flags |= ROOTCONTEXT_MNT;
679                 }
680                 if (opts->defcontext) {
681                         rc = parse_sid(sb, opts->defcontext, &defcontext_sid);
682                         if (rc)
683                                 goto out;
684                         if (bad_option(sbsec, DEFCONTEXT_MNT, sbsec->def_sid,
685                                         defcontext_sid))
686                                 goto out_double_mount;
687                         sbsec->flags |= DEFCONTEXT_MNT;
688                 }
689         }
690
691         if (sbsec->flags & SE_SBINITIALIZED) {
692                 /* previously mounted with options, but not on this attempt? */
693                 if ((sbsec->flags & SE_MNTMASK) && !opts)
694                         goto out_double_mount;
695                 rc = 0;
696                 goto out;
697         }
698
699         if (strcmp(sb->s_type->name, "proc") == 0)
700                 sbsec->flags |= SE_SBPROC | SE_SBGENFS;
701
702         if (!strcmp(sb->s_type->name, "debugfs") ||
703             !strcmp(sb->s_type->name, "tracefs") ||
704             !strcmp(sb->s_type->name, "binder") ||
705             !strcmp(sb->s_type->name, "bpf") ||
706             !strcmp(sb->s_type->name, "pstore"))
707                 sbsec->flags |= SE_SBGENFS;
708
709         if (!strcmp(sb->s_type->name, "sysfs") ||
710             !strcmp(sb->s_type->name, "cgroup") ||
711             !strcmp(sb->s_type->name, "cgroup2"))
712                 sbsec->flags |= SE_SBGENFS | SE_SBGENFS_XATTR;
713
714         if (!sbsec->behavior) {
715                 /*
716                  * Determine the labeling behavior to use for this
717                  * filesystem type.
718                  */
719                 rc = security_fs_use(&selinux_state, sb);
720                 if (rc) {
721                         pr_warn("%s: security_fs_use(%s) returned %d\n",
722                                         __func__, sb->s_type->name, rc);
723                         goto out;
724                 }
725         }
726
727         /*
728          * If this is a user namespace mount and the filesystem type is not
729          * explicitly whitelisted, then no contexts are allowed on the command
730          * line and security labels must be ignored.
731          */
732         if (sb->s_user_ns != &init_user_ns &&
733             strcmp(sb->s_type->name, "tmpfs") &&
734             strcmp(sb->s_type->name, "ramfs") &&
735             strcmp(sb->s_type->name, "devpts")) {
736                 if (context_sid || fscontext_sid || rootcontext_sid ||
737                     defcontext_sid) {
738                         rc = -EACCES;
739                         goto out;
740                 }
741                 if (sbsec->behavior == SECURITY_FS_USE_XATTR) {
742                         sbsec->behavior = SECURITY_FS_USE_MNTPOINT;
743                         rc = security_transition_sid(&selinux_state,
744                                                      current_sid(),
745                                                      current_sid(),
746                                                      SECCLASS_FILE, NULL,
747                                                      &sbsec->mntpoint_sid);
748                         if (rc)
749                                 goto out;
750                 }
751                 goto out_set_opts;
752         }
753
754         /* sets the context of the superblock for the fs being mounted. */
755         if (fscontext_sid) {
756                 rc = may_context_mount_sb_relabel(fscontext_sid, sbsec, cred);
757                 if (rc)
758                         goto out;
759
760                 sbsec->sid = fscontext_sid;
761         }
762
763         /*
764          * Switch to using mount point labeling behavior.
765          * sets the label used on all file below the mountpoint, and will set
766          * the superblock context if not already set.
767          */
768         if (kern_flags & SECURITY_LSM_NATIVE_LABELS && !context_sid) {
769                 sbsec->behavior = SECURITY_FS_USE_NATIVE;
770                 *set_kern_flags |= SECURITY_LSM_NATIVE_LABELS;
771         }
772
773         if (context_sid) {
774                 if (!fscontext_sid) {
775                         rc = may_context_mount_sb_relabel(context_sid, sbsec,
776                                                           cred);
777                         if (rc)
778                                 goto out;
779                         sbsec->sid = context_sid;
780                 } else {
781                         rc = may_context_mount_inode_relabel(context_sid, sbsec,
782                                                              cred);
783                         if (rc)
784                                 goto out;
785                 }
786                 if (!rootcontext_sid)
787                         rootcontext_sid = context_sid;
788
789                 sbsec->mntpoint_sid = context_sid;
790                 sbsec->behavior = SECURITY_FS_USE_MNTPOINT;
791         }
792
793         if (rootcontext_sid) {
794                 rc = may_context_mount_inode_relabel(rootcontext_sid, sbsec,
795                                                      cred);
796                 if (rc)
797                         goto out;
798
799                 root_isec->sid = rootcontext_sid;
800                 root_isec->initialized = LABEL_INITIALIZED;
801         }
802
803         if (defcontext_sid) {
804                 if (sbsec->behavior != SECURITY_FS_USE_XATTR &&
805                         sbsec->behavior != SECURITY_FS_USE_NATIVE) {
806                         rc = -EINVAL;
807                         pr_warn("SELinux: defcontext option is "
808                                "invalid for this filesystem type\n");
809                         goto out;
810                 }
811
812                 if (defcontext_sid != sbsec->def_sid) {
813                         rc = may_context_mount_inode_relabel(defcontext_sid,
814                                                              sbsec, cred);
815                         if (rc)
816                                 goto out;
817                 }
818
819                 sbsec->def_sid = defcontext_sid;
820         }
821
822 out_set_opts:
823         rc = sb_finish_set_opts(sb);
824 out:
825         mutex_unlock(&sbsec->lock);
826         return rc;
827 out_double_mount:
828         rc = -EINVAL;
829         pr_warn("SELinux: mount invalid.  Same superblock, different "
830                "security settings for (dev %s, type %s)\n", sb->s_id,
831                sb->s_type->name);
832         goto out;
833 }
834
835 static int selinux_cmp_sb_context(const struct super_block *oldsb,
836                                     const struct super_block *newsb)
837 {
838         struct superblock_security_struct *old = oldsb->s_security;
839         struct superblock_security_struct *new = newsb->s_security;
840         char oldflags = old->flags & SE_MNTMASK;
841         char newflags = new->flags & SE_MNTMASK;
842
843         if (oldflags != newflags)
844                 goto mismatch;
845         if ((oldflags & FSCONTEXT_MNT) && old->sid != new->sid)
846                 goto mismatch;
847         if ((oldflags & CONTEXT_MNT) && old->mntpoint_sid != new->mntpoint_sid)
848                 goto mismatch;
849         if ((oldflags & DEFCONTEXT_MNT) && old->def_sid != new->def_sid)
850                 goto mismatch;
851         if (oldflags & ROOTCONTEXT_MNT) {
852                 struct inode_security_struct *oldroot = backing_inode_security(oldsb->s_root);
853                 struct inode_security_struct *newroot = backing_inode_security(newsb->s_root);
854                 if (oldroot->sid != newroot->sid)
855                         goto mismatch;
856         }
857         return 0;
858 mismatch:
859         pr_warn("SELinux: mount invalid.  Same superblock, "
860                             "different security settings for (dev %s, "
861                             "type %s)\n", newsb->s_id, newsb->s_type->name);
862         return -EBUSY;
863 }
864
865 static int selinux_sb_clone_mnt_opts(const struct super_block *oldsb,
866                                         struct super_block *newsb,
867                                         unsigned long kern_flags,
868                                         unsigned long *set_kern_flags)
869 {
870         int rc = 0;
871         const struct superblock_security_struct *oldsbsec = oldsb->s_security;
872         struct superblock_security_struct *newsbsec = newsb->s_security;
873
874         int set_fscontext =     (oldsbsec->flags & FSCONTEXT_MNT);
875         int set_context =       (oldsbsec->flags & CONTEXT_MNT);
876         int set_rootcontext =   (oldsbsec->flags & ROOTCONTEXT_MNT);
877
878         /*
879          * if the parent was able to be mounted it clearly had no special lsm
880          * mount options.  thus we can safely deal with this superblock later
881          */
882         if (!selinux_initialized(&selinux_state))
883                 return 0;
884
885         /*
886          * Specifying internal flags without providing a place to
887          * place the results is not allowed.
888          */
889         if (kern_flags && !set_kern_flags)
890                 return -EINVAL;
891
892         /* how can we clone if the old one wasn't set up?? */
893         BUG_ON(!(oldsbsec->flags & SE_SBINITIALIZED));
894
895         /* if fs is reusing a sb, make sure that the contexts match */
896         if (newsbsec->flags & SE_SBINITIALIZED) {
897                 if ((kern_flags & SECURITY_LSM_NATIVE_LABELS) && !set_context)
898                         *set_kern_flags |= SECURITY_LSM_NATIVE_LABELS;
899                 return selinux_cmp_sb_context(oldsb, newsb);
900         }
901
902         mutex_lock(&newsbsec->lock);
903
904         newsbsec->flags = oldsbsec->flags;
905
906         newsbsec->sid = oldsbsec->sid;
907         newsbsec->def_sid = oldsbsec->def_sid;
908         newsbsec->behavior = oldsbsec->behavior;
909
910         if (newsbsec->behavior == SECURITY_FS_USE_NATIVE &&
911                 !(kern_flags & SECURITY_LSM_NATIVE_LABELS) && !set_context) {
912                 rc = security_fs_use(&selinux_state, newsb);
913                 if (rc)
914                         goto out;
915         }
916
917         if (kern_flags & SECURITY_LSM_NATIVE_LABELS && !set_context) {
918                 newsbsec->behavior = SECURITY_FS_USE_NATIVE;
919                 *set_kern_flags |= SECURITY_LSM_NATIVE_LABELS;
920         }
921
922         if (set_context) {
923                 u32 sid = oldsbsec->mntpoint_sid;
924
925                 if (!set_fscontext)
926                         newsbsec->sid = sid;
927                 if (!set_rootcontext) {
928                         struct inode_security_struct *newisec = backing_inode_security(newsb->s_root);
929                         newisec->sid = sid;
930                 }
931                 newsbsec->mntpoint_sid = sid;
932         }
933         if (set_rootcontext) {
934                 const struct inode_security_struct *oldisec = backing_inode_security(oldsb->s_root);
935                 struct inode_security_struct *newisec = backing_inode_security(newsb->s_root);
936
937                 newisec->sid = oldisec->sid;
938         }
939
940         sb_finish_set_opts(newsb);
941 out:
942         mutex_unlock(&newsbsec->lock);
943         return rc;
944 }
945
946 static int selinux_add_opt(int token, const char *s, void **mnt_opts)
947 {
948         struct selinux_mnt_opts *opts = *mnt_opts;
949
950         if (token == Opt_seclabel)      /* eaten and completely ignored */
951                 return 0;
952
953         if (!opts) {
954                 opts = kzalloc(sizeof(struct selinux_mnt_opts), GFP_KERNEL);
955                 if (!opts)
956                         return -ENOMEM;
957                 *mnt_opts = opts;
958         }
959         if (!s)
960                 return -ENOMEM;
961         switch (token) {
962         case Opt_context:
963                 if (opts->context || opts->defcontext)
964                         goto Einval;
965                 opts->context = s;
966                 break;
967         case Opt_fscontext:
968                 if (opts->fscontext)
969                         goto Einval;
970                 opts->fscontext = s;
971                 break;
972         case Opt_rootcontext:
973                 if (opts->rootcontext)
974                         goto Einval;
975                 opts->rootcontext = s;
976                 break;
977         case Opt_defcontext:
978                 if (opts->context || opts->defcontext)
979                         goto Einval;
980                 opts->defcontext = s;
981                 break;
982         }
983         return 0;
984 Einval:
985         pr_warn(SEL_MOUNT_FAIL_MSG);
986         return -EINVAL;
987 }
988
989 static int selinux_add_mnt_opt(const char *option, const char *val, int len,
990                                void **mnt_opts)
991 {
992         int token = Opt_error;
993         int rc, i;
994
995         for (i = 0; i < ARRAY_SIZE(tokens); i++) {
996                 if (strcmp(option, tokens[i].name) == 0) {
997                         token = tokens[i].opt;
998                         break;
999                 }
1000         }
1001
1002         if (token == Opt_error)
1003                 return -EINVAL;
1004
1005         if (token != Opt_seclabel) {
1006                 val = kmemdup_nul(val, len, GFP_KERNEL);
1007                 if (!val) {
1008                         rc = -ENOMEM;
1009                         goto free_opt;
1010                 }
1011         }
1012         rc = selinux_add_opt(token, val, mnt_opts);
1013         if (unlikely(rc)) {
1014                 kfree(val);
1015                 goto free_opt;
1016         }
1017         return rc;
1018
1019 free_opt:
1020         if (*mnt_opts) {
1021                 selinux_free_mnt_opts(*mnt_opts);
1022                 *mnt_opts = NULL;
1023         }
1024         return rc;
1025 }
1026
1027 static int show_sid(struct seq_file *m, u32 sid)
1028 {
1029         char *context = NULL;
1030         u32 len;
1031         int rc;
1032
1033         rc = security_sid_to_context(&selinux_state, sid,
1034                                              &context, &len);
1035         if (!rc) {
1036                 bool has_comma = context && strchr(context, ',');
1037
1038                 seq_putc(m, '=');
1039                 if (has_comma)
1040                         seq_putc(m, '\"');
1041                 seq_escape(m, context, "\"\n\\");
1042                 if (has_comma)
1043                         seq_putc(m, '\"');
1044         }
1045         kfree(context);
1046         return rc;
1047 }
1048
1049 static int selinux_sb_show_options(struct seq_file *m, struct super_block *sb)
1050 {
1051         struct superblock_security_struct *sbsec = sb->s_security;
1052         int rc;
1053
1054         if (!(sbsec->flags & SE_SBINITIALIZED))
1055                 return 0;
1056
1057         if (!selinux_initialized(&selinux_state))
1058                 return 0;
1059
1060         if (sbsec->flags & FSCONTEXT_MNT) {
1061                 seq_putc(m, ',');
1062                 seq_puts(m, FSCONTEXT_STR);
1063                 rc = show_sid(m, sbsec->sid);
1064                 if (rc)
1065                         return rc;
1066         }
1067         if (sbsec->flags & CONTEXT_MNT) {
1068                 seq_putc(m, ',');
1069                 seq_puts(m, CONTEXT_STR);
1070                 rc = show_sid(m, sbsec->mntpoint_sid);
1071                 if (rc)
1072                         return rc;
1073         }
1074         if (sbsec->flags & DEFCONTEXT_MNT) {
1075                 seq_putc(m, ',');
1076                 seq_puts(m, DEFCONTEXT_STR);
1077                 rc = show_sid(m, sbsec->def_sid);
1078                 if (rc)
1079                         return rc;
1080         }
1081         if (sbsec->flags & ROOTCONTEXT_MNT) {
1082                 struct dentry *root = sbsec->sb->s_root;
1083                 struct inode_security_struct *isec = backing_inode_security(root);
1084                 seq_putc(m, ',');
1085                 seq_puts(m, ROOTCONTEXT_STR);
1086                 rc = show_sid(m, isec->sid);
1087                 if (rc)
1088                         return rc;
1089         }
1090         if (sbsec->flags & SBLABEL_MNT) {
1091                 seq_putc(m, ',');
1092                 seq_puts(m, SECLABEL_STR);
1093         }
1094         return 0;
1095 }
1096
1097 static inline u16 inode_mode_to_security_class(umode_t mode)
1098 {
1099         switch (mode & S_IFMT) {
1100         case S_IFSOCK:
1101                 return SECCLASS_SOCK_FILE;
1102         case S_IFLNK:
1103                 return SECCLASS_LNK_FILE;
1104         case S_IFREG:
1105                 return SECCLASS_FILE;
1106         case S_IFBLK:
1107                 return SECCLASS_BLK_FILE;
1108         case S_IFDIR:
1109                 return SECCLASS_DIR;
1110         case S_IFCHR:
1111                 return SECCLASS_CHR_FILE;
1112         case S_IFIFO:
1113                 return SECCLASS_FIFO_FILE;
1114
1115         }
1116
1117         return SECCLASS_FILE;
1118 }
1119
1120 static inline int default_protocol_stream(int protocol)
1121 {
1122         return (protocol == IPPROTO_IP || protocol == IPPROTO_TCP);
1123 }
1124
1125 static inline int default_protocol_dgram(int protocol)
1126 {
1127         return (protocol == IPPROTO_IP || protocol == IPPROTO_UDP);
1128 }
1129
1130 static inline u16 socket_type_to_security_class(int family, int type, int protocol)
1131 {
1132         int extsockclass = selinux_policycap_extsockclass();
1133
1134         switch (family) {
1135         case PF_UNIX:
1136                 switch (type) {
1137                 case SOCK_STREAM:
1138                 case SOCK_SEQPACKET:
1139                         return SECCLASS_UNIX_STREAM_SOCKET;
1140                 case SOCK_DGRAM:
1141                 case SOCK_RAW:
1142                         return SECCLASS_UNIX_DGRAM_SOCKET;
1143                 }
1144                 break;
1145         case PF_INET:
1146         case PF_INET6:
1147                 switch (type) {
1148                 case SOCK_STREAM:
1149                 case SOCK_SEQPACKET:
1150                         if (default_protocol_stream(protocol))
1151                                 return SECCLASS_TCP_SOCKET;
1152                         else if (extsockclass && protocol == IPPROTO_SCTP)
1153                                 return SECCLASS_SCTP_SOCKET;
1154                         else
1155                                 return SECCLASS_RAWIP_SOCKET;
1156                 case SOCK_DGRAM:
1157                         if (default_protocol_dgram(protocol))
1158                                 return SECCLASS_UDP_SOCKET;
1159                         else if (extsockclass && (protocol == IPPROTO_ICMP ||
1160                                                   protocol == IPPROTO_ICMPV6))
1161                                 return SECCLASS_ICMP_SOCKET;
1162                         else
1163                                 return SECCLASS_RAWIP_SOCKET;
1164                 case SOCK_DCCP:
1165                         return SECCLASS_DCCP_SOCKET;
1166                 default:
1167                         return SECCLASS_RAWIP_SOCKET;
1168                 }
1169                 break;
1170         case PF_NETLINK:
1171                 switch (protocol) {
1172                 case NETLINK_ROUTE:
1173                         return SECCLASS_NETLINK_ROUTE_SOCKET;
1174                 case NETLINK_SOCK_DIAG:
1175                         return SECCLASS_NETLINK_TCPDIAG_SOCKET;
1176                 case NETLINK_NFLOG:
1177                         return SECCLASS_NETLINK_NFLOG_SOCKET;
1178                 case NETLINK_XFRM:
1179                         return SECCLASS_NETLINK_XFRM_SOCKET;
1180                 case NETLINK_SELINUX:
1181                         return SECCLASS_NETLINK_SELINUX_SOCKET;
1182                 case NETLINK_ISCSI:
1183                         return SECCLASS_NETLINK_ISCSI_SOCKET;
1184                 case NETLINK_AUDIT:
1185                         return SECCLASS_NETLINK_AUDIT_SOCKET;
1186                 case NETLINK_FIB_LOOKUP:
1187                         return SECCLASS_NETLINK_FIB_LOOKUP_SOCKET;
1188                 case NETLINK_CONNECTOR:
1189                         return SECCLASS_NETLINK_CONNECTOR_SOCKET;
1190                 case NETLINK_NETFILTER:
1191                         return SECCLASS_NETLINK_NETFILTER_SOCKET;
1192                 case NETLINK_DNRTMSG:
1193                         return SECCLASS_NETLINK_DNRT_SOCKET;
1194                 case NETLINK_KOBJECT_UEVENT:
1195                         return SECCLASS_NETLINK_KOBJECT_UEVENT_SOCKET;
1196                 case NETLINK_GENERIC:
1197                         return SECCLASS_NETLINK_GENERIC_SOCKET;
1198                 case NETLINK_SCSITRANSPORT:
1199                         return SECCLASS_NETLINK_SCSITRANSPORT_SOCKET;
1200                 case NETLINK_RDMA:
1201                         return SECCLASS_NETLINK_RDMA_SOCKET;
1202                 case NETLINK_CRYPTO:
1203                         return SECCLASS_NETLINK_CRYPTO_SOCKET;
1204                 default:
1205                         return SECCLASS_NETLINK_SOCKET;
1206                 }
1207         case PF_PACKET:
1208                 return SECCLASS_PACKET_SOCKET;
1209         case PF_KEY:
1210                 return SECCLASS_KEY_SOCKET;
1211         case PF_APPLETALK:
1212                 return SECCLASS_APPLETALK_SOCKET;
1213         }
1214
1215         if (extsockclass) {
1216                 switch (family) {
1217                 case PF_AX25:
1218                         return SECCLASS_AX25_SOCKET;
1219                 case PF_IPX:
1220                         return SECCLASS_IPX_SOCKET;
1221                 case PF_NETROM:
1222                         return SECCLASS_NETROM_SOCKET;
1223                 case PF_ATMPVC:
1224                         return SECCLASS_ATMPVC_SOCKET;
1225                 case PF_X25:
1226                         return SECCLASS_X25_SOCKET;
1227                 case PF_ROSE:
1228                         return SECCLASS_ROSE_SOCKET;
1229                 case PF_DECnet:
1230                         return SECCLASS_DECNET_SOCKET;
1231                 case PF_ATMSVC:
1232                         return SECCLASS_ATMSVC_SOCKET;
1233                 case PF_RDS:
1234                         return SECCLASS_RDS_SOCKET;
1235                 case PF_IRDA:
1236                         return SECCLASS_IRDA_SOCKET;
1237                 case PF_PPPOX:
1238                         return SECCLASS_PPPOX_SOCKET;
1239                 case PF_LLC:
1240                         return SECCLASS_LLC_SOCKET;
1241                 case PF_CAN:
1242                         return SECCLASS_CAN_SOCKET;
1243                 case PF_TIPC:
1244                         return SECCLASS_TIPC_SOCKET;
1245                 case PF_BLUETOOTH:
1246                         return SECCLASS_BLUETOOTH_SOCKET;
1247                 case PF_IUCV:
1248                         return SECCLASS_IUCV_SOCKET;
1249                 case PF_RXRPC:
1250                         return SECCLASS_RXRPC_SOCKET;
1251                 case PF_ISDN:
1252                         return SECCLASS_ISDN_SOCKET;
1253                 case PF_PHONET:
1254                         return SECCLASS_PHONET_SOCKET;
1255                 case PF_IEEE802154:
1256                         return SECCLASS_IEEE802154_SOCKET;
1257                 case PF_CAIF:
1258                         return SECCLASS_CAIF_SOCKET;
1259                 case PF_ALG:
1260                         return SECCLASS_ALG_SOCKET;
1261                 case PF_NFC:
1262                         return SECCLASS_NFC_SOCKET;
1263                 case PF_VSOCK:
1264                         return SECCLASS_VSOCK_SOCKET;
1265                 case PF_KCM:
1266                         return SECCLASS_KCM_SOCKET;
1267                 case PF_QIPCRTR:
1268                         return SECCLASS_QIPCRTR_SOCKET;
1269                 case PF_SMC:
1270                         return SECCLASS_SMC_SOCKET;
1271                 case PF_XDP:
1272                         return SECCLASS_XDP_SOCKET;
1273 #if PF_MAX > 45
1274 #error New address family defined, please update this function.
1275 #endif
1276                 }
1277         }
1278
1279         return SECCLASS_SOCKET;
1280 }
1281
1282 static int selinux_genfs_get_sid(struct dentry *dentry,
1283                                  u16 tclass,
1284                                  u16 flags,
1285                                  u32 *sid)
1286 {
1287         int rc;
1288         struct super_block *sb = dentry->d_sb;
1289         char *buffer, *path;
1290
1291         buffer = (char *)__get_free_page(GFP_KERNEL);
1292         if (!buffer)
1293                 return -ENOMEM;
1294
1295         path = dentry_path_raw(dentry, buffer, PAGE_SIZE);
1296         if (IS_ERR(path))
1297                 rc = PTR_ERR(path);
1298         else {
1299                 if (flags & SE_SBPROC) {
1300                         /* each process gets a /proc/PID/ entry. Strip off the
1301                          * PID part to get a valid selinux labeling.
1302                          * e.g. /proc/1/net/rpc/nfs -> /net/rpc/nfs */
1303                         while (path[1] >= '0' && path[1] <= '9') {
1304                                 path[1] = '/';
1305                                 path++;
1306                         }
1307                 }
1308                 rc = security_genfs_sid(&selinux_state, sb->s_type->name,
1309                                         path, tclass, sid);
1310                 if (rc == -ENOENT) {
1311                         /* No match in policy, mark as unlabeled. */
1312                         *sid = SECINITSID_UNLABELED;
1313                         rc = 0;
1314                 }
1315         }
1316         free_page((unsigned long)buffer);
1317         return rc;
1318 }
1319
1320 static int inode_doinit_use_xattr(struct inode *inode, struct dentry *dentry,
1321                                   u32 def_sid, u32 *sid)
1322 {
1323 #define INITCONTEXTLEN 255
1324         char *context;
1325         unsigned int len;
1326         int rc;
1327
1328         len = INITCONTEXTLEN;
1329         context = kmalloc(len + 1, GFP_NOFS);
1330         if (!context)
1331                 return -ENOMEM;
1332
1333         context[len] = '\0';
1334         rc = __vfs_getxattr(dentry, inode, XATTR_NAME_SELINUX, context, len);
1335         if (rc == -ERANGE) {
1336                 kfree(context);
1337
1338                 /* Need a larger buffer.  Query for the right size. */
1339                 rc = __vfs_getxattr(dentry, inode, XATTR_NAME_SELINUX, NULL, 0);
1340                 if (rc < 0)
1341                         return rc;
1342
1343                 len = rc;
1344                 context = kmalloc(len + 1, GFP_NOFS);
1345                 if (!context)
1346                         return -ENOMEM;
1347
1348                 context[len] = '\0';
1349                 rc = __vfs_getxattr(dentry, inode, XATTR_NAME_SELINUX,
1350                                     context, len);
1351         }
1352         if (rc < 0) {
1353                 kfree(context);
1354                 if (rc != -ENODATA) {
1355                         pr_warn("SELinux: %s:  getxattr returned %d for dev=%s ino=%ld\n",
1356                                 __func__, -rc, inode->i_sb->s_id, inode->i_ino);
1357                         return rc;
1358                 }
1359                 *sid = def_sid;
1360                 return 0;
1361         }
1362
1363         rc = security_context_to_sid_default(&selinux_state, context, rc, sid,
1364                                              def_sid, GFP_NOFS);
1365         if (rc) {
1366                 char *dev = inode->i_sb->s_id;
1367                 unsigned long ino = inode->i_ino;
1368
1369                 if (rc == -EINVAL) {
1370                         pr_notice_ratelimited("SELinux: inode=%lu on dev=%s was found to have an invalid context=%s.  This indicates you may need to relabel the inode or the filesystem in question.\n",
1371                                               ino, dev, context);
1372                 } else {
1373                         pr_warn("SELinux: %s:  context_to_sid(%s) returned %d for dev=%s ino=%ld\n",
1374                                 __func__, context, -rc, dev, ino);
1375                 }
1376         }
1377         kfree(context);
1378         return 0;
1379 }
1380
1381 /* The inode's security attributes must be initialized before first use. */
1382 static int inode_doinit_with_dentry(struct inode *inode, struct dentry *opt_dentry)
1383 {
1384         struct superblock_security_struct *sbsec = NULL;
1385         struct inode_security_struct *isec = selinux_inode(inode);
1386         u32 task_sid, sid = 0;
1387         u16 sclass;
1388         struct dentry *dentry;
1389         int rc = 0;
1390
1391         if (isec->initialized == LABEL_INITIALIZED)
1392                 return 0;
1393
1394         spin_lock(&isec->lock);
1395         if (isec->initialized == LABEL_INITIALIZED)
1396                 goto out_unlock;
1397
1398         if (isec->sclass == SECCLASS_FILE)
1399                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
1400
1401         sbsec = inode->i_sb->s_security;
1402         if (!(sbsec->flags & SE_SBINITIALIZED)) {
1403                 /* Defer initialization until selinux_complete_init,
1404                    after the initial policy is loaded and the security
1405                    server is ready to handle calls. */
1406                 spin_lock(&sbsec->isec_lock);
1407                 if (list_empty(&isec->list))
1408                         list_add(&isec->list, &sbsec->isec_head);
1409                 spin_unlock(&sbsec->isec_lock);
1410                 goto out_unlock;
1411         }
1412
1413         sclass = isec->sclass;
1414         task_sid = isec->task_sid;
1415         sid = isec->sid;
1416         isec->initialized = LABEL_PENDING;
1417         spin_unlock(&isec->lock);
1418
1419         switch (sbsec->behavior) {
1420         case SECURITY_FS_USE_NATIVE:
1421                 break;
1422         case SECURITY_FS_USE_XATTR:
1423                 if (!(inode->i_opflags & IOP_XATTR)) {
1424                         sid = sbsec->def_sid;
1425                         break;
1426                 }
1427                 /* Need a dentry, since the xattr API requires one.
1428                    Life would be simpler if we could just pass the inode. */
1429                 if (opt_dentry) {
1430                         /* Called from d_instantiate or d_splice_alias. */
1431                         dentry = dget(opt_dentry);
1432                 } else {
1433                         /*
1434                          * Called from selinux_complete_init, try to find a dentry.
1435                          * Some filesystems really want a connected one, so try
1436                          * that first.  We could split SECURITY_FS_USE_XATTR in
1437                          * two, depending upon that...
1438                          */
1439                         dentry = d_find_alias(inode);
1440                         if (!dentry)
1441                                 dentry = d_find_any_alias(inode);
1442                 }
1443                 if (!dentry) {
1444                         /*
1445                          * this is can be hit on boot when a file is accessed
1446                          * before the policy is loaded.  When we load policy we
1447                          * may find inodes that have no dentry on the
1448                          * sbsec->isec_head list.  No reason to complain as these
1449                          * will get fixed up the next time we go through
1450                          * inode_doinit with a dentry, before these inodes could
1451                          * be used again by userspace.
1452                          */
1453                         goto out;
1454                 }
1455
1456                 rc = inode_doinit_use_xattr(inode, dentry, sbsec->def_sid,
1457                                             &sid);
1458                 dput(dentry);
1459                 if (rc)
1460                         goto out;
1461                 break;
1462         case SECURITY_FS_USE_TASK:
1463                 sid = task_sid;
1464                 break;
1465         case SECURITY_FS_USE_TRANS:
1466                 /* Default to the fs SID. */
1467                 sid = sbsec->sid;
1468
1469                 /* Try to obtain a transition SID. */
1470                 rc = security_transition_sid(&selinux_state, task_sid, sid,
1471                                              sclass, NULL, &sid);
1472                 if (rc)
1473                         goto out;
1474                 break;
1475         case SECURITY_FS_USE_MNTPOINT:
1476                 sid = sbsec->mntpoint_sid;
1477                 break;
1478         default:
1479                 /* Default to the fs superblock SID. */
1480                 sid = sbsec->sid;
1481
1482                 if ((sbsec->flags & SE_SBGENFS) &&
1483                      (!S_ISLNK(inode->i_mode) ||
1484                       selinux_policycap_genfs_seclabel_symlinks())) {
1485                         /* We must have a dentry to determine the label on
1486                          * procfs inodes */
1487                         if (opt_dentry) {
1488                                 /* Called from d_instantiate or
1489                                  * d_splice_alias. */
1490                                 dentry = dget(opt_dentry);
1491                         } else {
1492                                 /* Called from selinux_complete_init, try to
1493                                  * find a dentry.  Some filesystems really want
1494                                  * a connected one, so try that first.
1495                                  */
1496                                 dentry = d_find_alias(inode);
1497                                 if (!dentry)
1498                                         dentry = d_find_any_alias(inode);
1499                         }
1500                         /*
1501                          * This can be hit on boot when a file is accessed
1502                          * before the policy is loaded.  When we load policy we
1503                          * may find inodes that have no dentry on the
1504                          * sbsec->isec_head list.  No reason to complain as
1505                          * these will get fixed up the next time we go through
1506                          * inode_doinit() with a dentry, before these inodes
1507                          * could be used again by userspace.
1508                          */
1509                         if (!dentry)
1510                                 goto out;
1511                         rc = selinux_genfs_get_sid(dentry, sclass,
1512                                                    sbsec->flags, &sid);
1513                         if (rc) {
1514                                 dput(dentry);
1515                                 goto out;
1516                         }
1517
1518                         if ((sbsec->flags & SE_SBGENFS_XATTR) &&
1519                             (inode->i_opflags & IOP_XATTR)) {
1520                                 rc = inode_doinit_use_xattr(inode, dentry,
1521                                                             sid, &sid);
1522                                 if (rc) {
1523                                         dput(dentry);
1524                                         goto out;
1525                                 }
1526                         }
1527                         dput(dentry);
1528                 }
1529                 break;
1530         }
1531
1532 out:
1533         spin_lock(&isec->lock);
1534         if (isec->initialized == LABEL_PENDING) {
1535                 if (!sid || rc) {
1536                         isec->initialized = LABEL_INVALID;
1537                         goto out_unlock;
1538                 }
1539
1540                 isec->initialized = LABEL_INITIALIZED;
1541                 isec->sid = sid;
1542         }
1543
1544 out_unlock:
1545         spin_unlock(&isec->lock);
1546         return rc;
1547 }
1548
1549 /* Convert a Linux signal to an access vector. */
1550 static inline u32 signal_to_av(int sig)
1551 {
1552         u32 perm = 0;
1553
1554         switch (sig) {
1555         case SIGCHLD:
1556                 /* Commonly granted from child to parent. */
1557                 perm = PROCESS__SIGCHLD;
1558                 break;
1559         case SIGKILL:
1560                 /* Cannot be caught or ignored */
1561                 perm = PROCESS__SIGKILL;
1562                 break;
1563         case SIGSTOP:
1564                 /* Cannot be caught or ignored */
1565                 perm = PROCESS__SIGSTOP;
1566                 break;
1567         default:
1568                 /* All other signals. */
1569                 perm = PROCESS__SIGNAL;
1570                 break;
1571         }
1572
1573         return perm;
1574 }
1575
1576 #if CAP_LAST_CAP > 63
1577 #error Fix SELinux to handle capabilities > 63.
1578 #endif
1579
1580 /* Check whether a task is allowed to use a capability. */
1581 static int cred_has_capability(const struct cred *cred,
1582                                int cap, unsigned int opts, bool initns)
1583 {
1584         struct common_audit_data ad;
1585         struct av_decision avd;
1586         u16 sclass;
1587         u32 sid = cred_sid(cred);
1588         u32 av = CAP_TO_MASK(cap);
1589         int rc;
1590
1591         ad.type = LSM_AUDIT_DATA_CAP;
1592         ad.u.cap = cap;
1593
1594         switch (CAP_TO_INDEX(cap)) {
1595         case 0:
1596                 sclass = initns ? SECCLASS_CAPABILITY : SECCLASS_CAP_USERNS;
1597                 break;
1598         case 1:
1599                 sclass = initns ? SECCLASS_CAPABILITY2 : SECCLASS_CAP2_USERNS;
1600                 break;
1601         default:
1602                 pr_err("SELinux:  out of range capability %d\n", cap);
1603                 BUG();
1604                 return -EINVAL;
1605         }
1606
1607         rc = avc_has_perm_noaudit(&selinux_state,
1608                                   sid, sid, sclass, av, 0, &avd);
1609         if (!(opts & CAP_OPT_NOAUDIT)) {
1610                 int rc2 = avc_audit(&selinux_state,
1611                                     sid, sid, sclass, av, &avd, rc, &ad, 0);
1612                 if (rc2)
1613                         return rc2;
1614         }
1615         return rc;
1616 }
1617
1618 /* Check whether a task has a particular permission to an inode.
1619    The 'adp' parameter is optional and allows other audit
1620    data to be passed (e.g. the dentry). */
1621 static int inode_has_perm(const struct cred *cred,
1622                           struct inode *inode,
1623                           u32 perms,
1624                           struct common_audit_data *adp)
1625 {
1626         struct inode_security_struct *isec;
1627         u32 sid;
1628
1629         validate_creds(cred);
1630
1631         if (unlikely(IS_PRIVATE(inode)))
1632                 return 0;
1633
1634         sid = cred_sid(cred);
1635         isec = selinux_inode(inode);
1636
1637         return avc_has_perm(&selinux_state,
1638                             sid, isec->sid, isec->sclass, perms, adp);
1639 }
1640
1641 /* Same as inode_has_perm, but pass explicit audit data containing
1642    the dentry to help the auditing code to more easily generate the
1643    pathname if needed. */
1644 static inline int dentry_has_perm(const struct cred *cred,
1645                                   struct dentry *dentry,
1646                                   u32 av)
1647 {
1648         struct inode *inode = d_backing_inode(dentry);
1649         struct common_audit_data ad;
1650
1651         ad.type = LSM_AUDIT_DATA_DENTRY;
1652         ad.u.dentry = dentry;
1653         __inode_security_revalidate(inode, dentry, true);
1654         return inode_has_perm(cred, inode, av, &ad);
1655 }
1656
1657 /* Same as inode_has_perm, but pass explicit audit data containing
1658    the path to help the auditing code to more easily generate the
1659    pathname if needed. */
1660 static inline int path_has_perm(const struct cred *cred,
1661                                 const struct path *path,
1662                                 u32 av)
1663 {
1664         struct inode *inode = d_backing_inode(path->dentry);
1665         struct common_audit_data ad;
1666
1667         ad.type = LSM_AUDIT_DATA_PATH;
1668         ad.u.path = *path;
1669         __inode_security_revalidate(inode, path->dentry, true);
1670         return inode_has_perm(cred, inode, av, &ad);
1671 }
1672
1673 /* Same as path_has_perm, but uses the inode from the file struct. */
1674 static inline int file_path_has_perm(const struct cred *cred,
1675                                      struct file *file,
1676                                      u32 av)
1677 {
1678         struct common_audit_data ad;
1679
1680         ad.type = LSM_AUDIT_DATA_FILE;
1681         ad.u.file = file;
1682         return inode_has_perm(cred, file_inode(file), av, &ad);
1683 }
1684
1685 #ifdef CONFIG_BPF_SYSCALL
1686 static int bpf_fd_pass(struct file *file, u32 sid);
1687 #endif
1688
1689 /* Check whether a task can use an open file descriptor to
1690    access an inode in a given way.  Check access to the
1691    descriptor itself, and then use dentry_has_perm to
1692    check a particular permission to the file.
1693    Access to the descriptor is implicitly granted if it
1694    has the same SID as the process.  If av is zero, then
1695    access to the file is not checked, e.g. for cases
1696    where only the descriptor is affected like seek. */
1697 static int file_has_perm(const struct cred *cred,
1698                          struct file *file,
1699                          u32 av)
1700 {
1701         struct file_security_struct *fsec = selinux_file(file);
1702         struct inode *inode = file_inode(file);
1703         struct common_audit_data ad;
1704         u32 sid = cred_sid(cred);
1705         int rc;
1706
1707         ad.type = LSM_AUDIT_DATA_FILE;
1708         ad.u.file = file;
1709
1710         if (sid != fsec->sid) {
1711                 rc = avc_has_perm(&selinux_state,
1712                                   sid, fsec->sid,
1713                                   SECCLASS_FD,
1714                                   FD__USE,
1715                                   &ad);
1716                 if (rc)
1717                         goto out;
1718         }
1719
1720 #ifdef CONFIG_BPF_SYSCALL
1721         rc = bpf_fd_pass(file, cred_sid(cred));
1722         if (rc)
1723                 return rc;
1724 #endif
1725
1726         /* av is zero if only checking access to the descriptor. */
1727         rc = 0;
1728         if (av)
1729                 rc = inode_has_perm(cred, inode, av, &ad);
1730
1731 out:
1732         return rc;
1733 }
1734
1735 /*
1736  * Determine the label for an inode that might be unioned.
1737  */
1738 static int
1739 selinux_determine_inode_label(const struct task_security_struct *tsec,
1740                                  struct inode *dir,
1741                                  const struct qstr *name, u16 tclass,
1742                                  u32 *_new_isid)
1743 {
1744         const struct superblock_security_struct *sbsec = dir->i_sb->s_security;
1745
1746         if ((sbsec->flags & SE_SBINITIALIZED) &&
1747             (sbsec->behavior == SECURITY_FS_USE_MNTPOINT)) {
1748                 *_new_isid = sbsec->mntpoint_sid;
1749         } else if ((sbsec->flags & SBLABEL_MNT) &&
1750                    tsec->create_sid) {
1751                 *_new_isid = tsec->create_sid;
1752         } else {
1753                 const struct inode_security_struct *dsec = inode_security(dir);
1754                 return security_transition_sid(&selinux_state, tsec->sid,
1755                                                dsec->sid, tclass,
1756                                                name, _new_isid);
1757         }
1758
1759         return 0;
1760 }
1761
1762 /* Check whether a task can create a file. */
1763 static int may_create(struct inode *dir,
1764                       struct dentry *dentry,
1765                       u16 tclass)
1766 {
1767         const struct task_security_struct *tsec = selinux_cred(current_cred());
1768         struct inode_security_struct *dsec;
1769         struct superblock_security_struct *sbsec;
1770         u32 sid, newsid;
1771         struct common_audit_data ad;
1772         int rc;
1773
1774         dsec = inode_security(dir);
1775         sbsec = dir->i_sb->s_security;
1776
1777         sid = tsec->sid;
1778
1779         ad.type = LSM_AUDIT_DATA_DENTRY;
1780         ad.u.dentry = dentry;
1781
1782         rc = avc_has_perm(&selinux_state,
1783                           sid, dsec->sid, SECCLASS_DIR,
1784                           DIR__ADD_NAME | DIR__SEARCH,
1785                           &ad);
1786         if (rc)
1787                 return rc;
1788
1789         rc = selinux_determine_inode_label(tsec, dir, &dentry->d_name, tclass,
1790                                            &newsid);
1791         if (rc)
1792                 return rc;
1793
1794         rc = avc_has_perm(&selinux_state,
1795                           sid, newsid, tclass, FILE__CREATE, &ad);
1796         if (rc)
1797                 return rc;
1798
1799         return avc_has_perm(&selinux_state,
1800                             newsid, sbsec->sid,
1801                             SECCLASS_FILESYSTEM,
1802                             FILESYSTEM__ASSOCIATE, &ad);
1803 }
1804
1805 #define MAY_LINK        0
1806 #define MAY_UNLINK      1
1807 #define MAY_RMDIR       2
1808
1809 /* Check whether a task can link, unlink, or rmdir a file/directory. */
1810 static int may_link(struct inode *dir,
1811                     struct dentry *dentry,
1812                     int kind)
1813
1814 {
1815         struct inode_security_struct *dsec, *isec;
1816         struct common_audit_data ad;
1817         u32 sid = current_sid();
1818         u32 av;
1819         int rc;
1820
1821         dsec = inode_security(dir);
1822         isec = backing_inode_security(dentry);
1823
1824         ad.type = LSM_AUDIT_DATA_DENTRY;
1825         ad.u.dentry = dentry;
1826
1827         av = DIR__SEARCH;
1828         av |= (kind ? DIR__REMOVE_NAME : DIR__ADD_NAME);
1829         rc = avc_has_perm(&selinux_state,
1830                           sid, dsec->sid, SECCLASS_DIR, av, &ad);
1831         if (rc)
1832                 return rc;
1833
1834         switch (kind) {
1835         case MAY_LINK:
1836                 av = FILE__LINK;
1837                 break;
1838         case MAY_UNLINK:
1839                 av = FILE__UNLINK;
1840                 break;
1841         case MAY_RMDIR:
1842                 av = DIR__RMDIR;
1843                 break;
1844         default:
1845                 pr_warn("SELinux: %s:  unrecognized kind %d\n",
1846                         __func__, kind);
1847                 return 0;
1848         }
1849
1850         rc = avc_has_perm(&selinux_state,
1851                           sid, isec->sid, isec->sclass, av, &ad);
1852         return rc;
1853 }
1854
1855 static inline int may_rename(struct inode *old_dir,
1856                              struct dentry *old_dentry,
1857                              struct inode *new_dir,
1858                              struct dentry *new_dentry)
1859 {
1860         struct inode_security_struct *old_dsec, *new_dsec, *old_isec, *new_isec;
1861         struct common_audit_data ad;
1862         u32 sid = current_sid();
1863         u32 av;
1864         int old_is_dir, new_is_dir;
1865         int rc;
1866
1867         old_dsec = inode_security(old_dir);
1868         old_isec = backing_inode_security(old_dentry);
1869         old_is_dir = d_is_dir(old_dentry);
1870         new_dsec = inode_security(new_dir);
1871
1872         ad.type = LSM_AUDIT_DATA_DENTRY;
1873
1874         ad.u.dentry = old_dentry;
1875         rc = avc_has_perm(&selinux_state,
1876                           sid, old_dsec->sid, SECCLASS_DIR,
1877                           DIR__REMOVE_NAME | DIR__SEARCH, &ad);
1878         if (rc)
1879                 return rc;
1880         rc = avc_has_perm(&selinux_state,
1881                           sid, old_isec->sid,
1882                           old_isec->sclass, FILE__RENAME, &ad);
1883         if (rc)
1884                 return rc;
1885         if (old_is_dir && new_dir != old_dir) {
1886                 rc = avc_has_perm(&selinux_state,
1887                                   sid, old_isec->sid,
1888                                   old_isec->sclass, DIR__REPARENT, &ad);
1889                 if (rc)
1890                         return rc;
1891         }
1892
1893         ad.u.dentry = new_dentry;
1894         av = DIR__ADD_NAME | DIR__SEARCH;
1895         if (d_is_positive(new_dentry))
1896                 av |= DIR__REMOVE_NAME;
1897         rc = avc_has_perm(&selinux_state,
1898                           sid, new_dsec->sid, SECCLASS_DIR, av, &ad);
1899         if (rc)
1900                 return rc;
1901         if (d_is_positive(new_dentry)) {
1902                 new_isec = backing_inode_security(new_dentry);
1903                 new_is_dir = d_is_dir(new_dentry);
1904                 rc = avc_has_perm(&selinux_state,
1905                                   sid, new_isec->sid,
1906                                   new_isec->sclass,
1907                                   (new_is_dir ? DIR__RMDIR : FILE__UNLINK), &ad);
1908                 if (rc)
1909                         return rc;
1910         }
1911
1912         return 0;
1913 }
1914
1915 /* Check whether a task can perform a filesystem operation. */
1916 static int superblock_has_perm(const struct cred *cred,
1917                                struct super_block *sb,
1918                                u32 perms,
1919                                struct common_audit_data *ad)
1920 {
1921         struct superblock_security_struct *sbsec;
1922         u32 sid = cred_sid(cred);
1923
1924         sbsec = sb->s_security;
1925         return avc_has_perm(&selinux_state,
1926                             sid, sbsec->sid, SECCLASS_FILESYSTEM, perms, ad);
1927 }
1928
1929 /* Convert a Linux mode and permission mask to an access vector. */
1930 static inline u32 file_mask_to_av(int mode, int mask)
1931 {
1932         u32 av = 0;
1933
1934         if (!S_ISDIR(mode)) {
1935                 if (mask & MAY_EXEC)
1936                         av |= FILE__EXECUTE;
1937                 if (mask & MAY_READ)
1938                         av |= FILE__READ;
1939
1940                 if (mask & MAY_APPEND)
1941                         av |= FILE__APPEND;
1942                 else if (mask & MAY_WRITE)
1943                         av |= FILE__WRITE;
1944
1945         } else {
1946                 if (mask & MAY_EXEC)
1947                         av |= DIR__SEARCH;
1948                 if (mask & MAY_WRITE)
1949                         av |= DIR__WRITE;
1950                 if (mask & MAY_READ)
1951                         av |= DIR__READ;
1952         }
1953
1954         return av;
1955 }
1956
1957 /* Convert a Linux file to an access vector. */
1958 static inline u32 file_to_av(struct file *file)
1959 {
1960         u32 av = 0;
1961
1962         if (file->f_mode & FMODE_READ)
1963                 av |= FILE__READ;
1964         if (file->f_mode & FMODE_WRITE) {
1965                 if (file->f_flags & O_APPEND)
1966                         av |= FILE__APPEND;
1967                 else
1968                         av |= FILE__WRITE;
1969         }
1970         if (!av) {
1971                 /*
1972                  * Special file opened with flags 3 for ioctl-only use.
1973                  */
1974                 av = FILE__IOCTL;
1975         }
1976
1977         return av;
1978 }
1979
1980 /*
1981  * Convert a file to an access vector and include the correct open
1982  * open permission.
1983  */
1984 static inline u32 open_file_to_av(struct file *file)
1985 {
1986         u32 av = file_to_av(file);
1987         struct inode *inode = file_inode(file);
1988
1989         if (selinux_policycap_openperm() &&
1990             inode->i_sb->s_magic != SOCKFS_MAGIC)
1991                 av |= FILE__OPEN;
1992
1993         return av;
1994 }
1995
1996 /* Hook functions begin here. */
1997
1998 static int selinux_binder_set_context_mgr(struct task_struct *mgr)
1999 {
2000         u32 mysid = current_sid();
2001         u32 mgrsid = task_sid(mgr);
2002
2003         return avc_has_perm(&selinux_state,
2004                             mysid, mgrsid, SECCLASS_BINDER,
2005                             BINDER__SET_CONTEXT_MGR, NULL);
2006 }
2007
2008 static int selinux_binder_transaction(struct task_struct *from,
2009                                       struct task_struct *to)
2010 {
2011         u32 mysid = current_sid();
2012         u32 fromsid = task_sid(from);
2013         u32 tosid = task_sid(to);
2014         int rc;
2015
2016         if (mysid != fromsid) {
2017                 rc = avc_has_perm(&selinux_state,
2018                                   mysid, fromsid, SECCLASS_BINDER,
2019                                   BINDER__IMPERSONATE, NULL);
2020                 if (rc)
2021                         return rc;
2022         }
2023
2024         return avc_has_perm(&selinux_state,
2025                             fromsid, tosid, SECCLASS_BINDER, BINDER__CALL,
2026                             NULL);
2027 }
2028
2029 static int selinux_binder_transfer_binder(struct task_struct *from,
2030                                           struct task_struct *to)
2031 {
2032         u32 fromsid = task_sid(from);
2033         u32 tosid = task_sid(to);
2034
2035         return avc_has_perm(&selinux_state,
2036                             fromsid, tosid, SECCLASS_BINDER, BINDER__TRANSFER,
2037                             NULL);
2038 }
2039
2040 static int selinux_binder_transfer_file(struct task_struct *from,
2041                                         struct task_struct *to,
2042                                         struct file *file)
2043 {
2044         u32 sid = task_sid(to);
2045         struct file_security_struct *fsec = selinux_file(file);
2046         struct dentry *dentry = file->f_path.dentry;
2047         struct inode_security_struct *isec;
2048         struct common_audit_data ad;
2049         int rc;
2050
2051         ad.type = LSM_AUDIT_DATA_PATH;
2052         ad.u.path = file->f_path;
2053
2054         if (sid != fsec->sid) {
2055                 rc = avc_has_perm(&selinux_state,
2056                                   sid, fsec->sid,
2057                                   SECCLASS_FD,
2058                                   FD__USE,
2059                                   &ad);
2060                 if (rc)
2061                         return rc;
2062         }
2063
2064 #ifdef CONFIG_BPF_SYSCALL
2065         rc = bpf_fd_pass(file, sid);
2066         if (rc)
2067                 return rc;
2068 #endif
2069
2070         if (unlikely(IS_PRIVATE(d_backing_inode(dentry))))
2071                 return 0;
2072
2073         isec = backing_inode_security(dentry);
2074         return avc_has_perm(&selinux_state,
2075                             sid, isec->sid, isec->sclass, file_to_av(file),
2076                             &ad);
2077 }
2078
2079 static int selinux_ptrace_access_check(struct task_struct *child,
2080                                      unsigned int mode)
2081 {
2082         u32 sid = current_sid();
2083         u32 csid = task_sid(child);
2084
2085         if (mode & PTRACE_MODE_READ)
2086                 return avc_has_perm(&selinux_state,
2087                                     sid, csid, SECCLASS_FILE, FILE__READ, NULL);
2088
2089         return avc_has_perm(&selinux_state,
2090                             sid, csid, SECCLASS_PROCESS, PROCESS__PTRACE, NULL);
2091 }
2092
2093 static int selinux_ptrace_traceme(struct task_struct *parent)
2094 {
2095         return avc_has_perm(&selinux_state,
2096                             task_sid(parent), current_sid(), SECCLASS_PROCESS,
2097                             PROCESS__PTRACE, NULL);
2098 }
2099
2100 static int selinux_capget(struct task_struct *target, kernel_cap_t *effective,
2101                           kernel_cap_t *inheritable, kernel_cap_t *permitted)
2102 {
2103         return avc_has_perm(&selinux_state,
2104                             current_sid(), task_sid(target), SECCLASS_PROCESS,
2105                             PROCESS__GETCAP, NULL);
2106 }
2107
2108 static int selinux_capset(struct cred *new, const struct cred *old,
2109                           const kernel_cap_t *effective,
2110                           const kernel_cap_t *inheritable,
2111                           const kernel_cap_t *permitted)
2112 {
2113         return avc_has_perm(&selinux_state,
2114                             cred_sid(old), cred_sid(new), SECCLASS_PROCESS,
2115                             PROCESS__SETCAP, NULL);
2116 }
2117
2118 /*
2119  * (This comment used to live with the selinux_task_setuid hook,
2120  * which was removed).
2121  *
2122  * Since setuid only affects the current process, and since the SELinux
2123  * controls are not based on the Linux identity attributes, SELinux does not
2124  * need to control this operation.  However, SELinux does control the use of
2125  * the CAP_SETUID and CAP_SETGID capabilities using the capable hook.
2126  */
2127
2128 static int selinux_capable(const struct cred *cred, struct user_namespace *ns,
2129                            int cap, unsigned int opts)
2130 {
2131         return cred_has_capability(cred, cap, opts, ns == &init_user_ns);
2132 }
2133
2134 static int selinux_quotactl(int cmds, int type, int id, struct super_block *sb)
2135 {
2136         const struct cred *cred = current_cred();
2137         int rc = 0;
2138
2139         if (!sb)
2140                 return 0;
2141
2142         switch (cmds) {
2143         case Q_SYNC:
2144         case Q_QUOTAON:
2145         case Q_QUOTAOFF:
2146         case Q_SETINFO:
2147         case Q_SETQUOTA:
2148         case Q_XQUOTAOFF:
2149         case Q_XQUOTAON:
2150         case Q_XSETQLIM:
2151                 rc = superblock_has_perm(cred, sb, FILESYSTEM__QUOTAMOD, NULL);
2152                 break;
2153         case Q_GETFMT:
2154         case Q_GETINFO:
2155         case Q_GETQUOTA:
2156         case Q_XGETQUOTA:
2157         case Q_XGETQSTAT:
2158         case Q_XGETQSTATV:
2159         case Q_XGETNEXTQUOTA:
2160                 rc = superblock_has_perm(cred, sb, FILESYSTEM__QUOTAGET, NULL);
2161                 break;
2162         default:
2163                 rc = 0;  /* let the kernel handle invalid cmds */
2164                 break;
2165         }
2166         return rc;
2167 }
2168
2169 static int selinux_quota_on(struct dentry *dentry)
2170 {
2171         const struct cred *cred = current_cred();
2172
2173         return dentry_has_perm(cred, dentry, FILE__QUOTAON);
2174 }
2175
2176 static int selinux_syslog(int type)
2177 {
2178         switch (type) {
2179         case SYSLOG_ACTION_READ_ALL:    /* Read last kernel messages */
2180         case SYSLOG_ACTION_SIZE_BUFFER: /* Return size of the log buffer */
2181                 return avc_has_perm(&selinux_state,
2182                                     current_sid(), SECINITSID_KERNEL,
2183                                     SECCLASS_SYSTEM, SYSTEM__SYSLOG_READ, NULL);
2184         case SYSLOG_ACTION_CONSOLE_OFF: /* Disable logging to console */
2185         case SYSLOG_ACTION_CONSOLE_ON:  /* Enable logging to console */
2186         /* Set level of messages printed to console */
2187         case SYSLOG_ACTION_CONSOLE_LEVEL:
2188                 return avc_has_perm(&selinux_state,
2189                                     current_sid(), SECINITSID_KERNEL,
2190                                     SECCLASS_SYSTEM, SYSTEM__SYSLOG_CONSOLE,
2191                                     NULL);
2192         }
2193         /* All other syslog types */
2194         return avc_has_perm(&selinux_state,
2195                             current_sid(), SECINITSID_KERNEL,
2196                             SECCLASS_SYSTEM, SYSTEM__SYSLOG_MOD, NULL);
2197 }
2198
2199 /*
2200  * Check that a process has enough memory to allocate a new virtual
2201  * mapping. 0 means there is enough memory for the allocation to
2202  * succeed and -ENOMEM implies there is not.
2203  *
2204  * Do not audit the selinux permission check, as this is applied to all
2205  * processes that allocate mappings.
2206  */
2207 static int selinux_vm_enough_memory(struct mm_struct *mm, long pages)
2208 {
2209         int rc, cap_sys_admin = 0;
2210
2211         rc = cred_has_capability(current_cred(), CAP_SYS_ADMIN,
2212                                  CAP_OPT_NOAUDIT, true);
2213         if (rc == 0)
2214                 cap_sys_admin = 1;
2215
2216         return cap_sys_admin;
2217 }
2218
2219 /* binprm security operations */
2220
2221 static u32 ptrace_parent_sid(void)
2222 {
2223         u32 sid = 0;
2224         struct task_struct *tracer;
2225
2226         rcu_read_lock();
2227         tracer = ptrace_parent(current);
2228         if (tracer)
2229                 sid = task_sid(tracer);
2230         rcu_read_unlock();
2231
2232         return sid;
2233 }
2234
2235 static int check_nnp_nosuid(const struct linux_binprm *bprm,
2236                             const struct task_security_struct *old_tsec,
2237                             const struct task_security_struct *new_tsec)
2238 {
2239         int nnp = (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS);
2240         int nosuid = !mnt_may_suid(bprm->file->f_path.mnt);
2241         int rc;
2242         u32 av;
2243
2244         if (!nnp && !nosuid)
2245                 return 0; /* neither NNP nor nosuid */
2246
2247         if (new_tsec->sid == old_tsec->sid)
2248                 return 0; /* No change in credentials */
2249
2250         /*
2251          * If the policy enables the nnp_nosuid_transition policy capability,
2252          * then we permit transitions under NNP or nosuid if the
2253          * policy allows the corresponding permission between
2254          * the old and new contexts.
2255          */
2256         if (selinux_policycap_nnp_nosuid_transition()) {
2257                 av = 0;
2258                 if (nnp)
2259                         av |= PROCESS2__NNP_TRANSITION;
2260                 if (nosuid)
2261                         av |= PROCESS2__NOSUID_TRANSITION;
2262                 rc = avc_has_perm(&selinux_state,
2263                                   old_tsec->sid, new_tsec->sid,
2264                                   SECCLASS_PROCESS2, av, NULL);
2265                 if (!rc)
2266                         return 0;
2267         }
2268
2269         /*
2270          * We also permit NNP or nosuid transitions to bounded SIDs,
2271          * i.e. SIDs that are guaranteed to only be allowed a subset
2272          * of the permissions of the current SID.
2273          */
2274         rc = security_bounded_transition(&selinux_state, old_tsec->sid,
2275                                          new_tsec->sid);
2276         if (!rc)
2277                 return 0;
2278
2279         /*
2280          * On failure, preserve the errno values for NNP vs nosuid.
2281          * NNP:  Operation not permitted for caller.
2282          * nosuid:  Permission denied to file.
2283          */
2284         if (nnp)
2285                 return -EPERM;
2286         return -EACCES;
2287 }
2288
2289 static int selinux_bprm_creds_for_exec(struct linux_binprm *bprm)
2290 {
2291         const struct task_security_struct *old_tsec;
2292         struct task_security_struct *new_tsec;
2293         struct inode_security_struct *isec;
2294         struct common_audit_data ad;
2295         struct inode *inode = file_inode(bprm->file);
2296         int rc;
2297
2298         /* SELinux context only depends on initial program or script and not
2299          * the script interpreter */
2300
2301         old_tsec = selinux_cred(current_cred());
2302         new_tsec = selinux_cred(bprm->cred);
2303         isec = inode_security(inode);
2304
2305         /* Default to the current task SID. */
2306         new_tsec->sid = old_tsec->sid;
2307         new_tsec->osid = old_tsec->sid;
2308
2309         /* Reset fs, key, and sock SIDs on execve. */
2310         new_tsec->create_sid = 0;
2311         new_tsec->keycreate_sid = 0;
2312         new_tsec->sockcreate_sid = 0;
2313
2314         if (old_tsec->exec_sid) {
2315                 new_tsec->sid = old_tsec->exec_sid;
2316                 /* Reset exec SID on execve. */
2317                 new_tsec->exec_sid = 0;
2318
2319                 /* Fail on NNP or nosuid if not an allowed transition. */
2320                 rc = check_nnp_nosuid(bprm, old_tsec, new_tsec);
2321                 if (rc)
2322                         return rc;
2323         } else {
2324                 /* Check for a default transition on this program. */
2325                 rc = security_transition_sid(&selinux_state, old_tsec->sid,
2326                                              isec->sid, SECCLASS_PROCESS, NULL,
2327                                              &new_tsec->sid);
2328                 if (rc)
2329                         return rc;
2330
2331                 /*
2332                  * Fallback to old SID on NNP or nosuid if not an allowed
2333                  * transition.
2334                  */
2335                 rc = check_nnp_nosuid(bprm, old_tsec, new_tsec);
2336                 if (rc)
2337                         new_tsec->sid = old_tsec->sid;
2338         }
2339
2340         ad.type = LSM_AUDIT_DATA_FILE;
2341         ad.u.file = bprm->file;
2342
2343         if (new_tsec->sid == old_tsec->sid) {
2344                 rc = avc_has_perm(&selinux_state,
2345                                   old_tsec->sid, isec->sid,
2346                                   SECCLASS_FILE, FILE__EXECUTE_NO_TRANS, &ad);
2347                 if (rc)
2348                         return rc;
2349         } else {
2350                 /* Check permissions for the transition. */
2351                 rc = avc_has_perm(&selinux_state,
2352                                   old_tsec->sid, new_tsec->sid,
2353                                   SECCLASS_PROCESS, PROCESS__TRANSITION, &ad);
2354                 if (rc)
2355                         return rc;
2356
2357                 rc = avc_has_perm(&selinux_state,
2358                                   new_tsec->sid, isec->sid,
2359                                   SECCLASS_FILE, FILE__ENTRYPOINT, &ad);
2360                 if (rc)
2361                         return rc;
2362
2363                 /* Check for shared state */
2364                 if (bprm->unsafe & LSM_UNSAFE_SHARE) {
2365                         rc = avc_has_perm(&selinux_state,
2366                                           old_tsec->sid, new_tsec->sid,
2367                                           SECCLASS_PROCESS, PROCESS__SHARE,
2368                                           NULL);
2369                         if (rc)
2370                                 return -EPERM;
2371                 }
2372
2373                 /* Make sure that anyone attempting to ptrace over a task that
2374                  * changes its SID has the appropriate permit */
2375                 if (bprm->unsafe & LSM_UNSAFE_PTRACE) {
2376                         u32 ptsid = ptrace_parent_sid();
2377                         if (ptsid != 0) {
2378                                 rc = avc_has_perm(&selinux_state,
2379                                                   ptsid, new_tsec->sid,
2380                                                   SECCLASS_PROCESS,
2381                                                   PROCESS__PTRACE, NULL);
2382                                 if (rc)
2383                                         return -EPERM;
2384                         }
2385                 }
2386
2387                 /* Clear any possibly unsafe personality bits on exec: */
2388                 bprm->per_clear |= PER_CLEAR_ON_SETID;
2389
2390                 /* Enable secure mode for SIDs transitions unless
2391                    the noatsecure permission is granted between
2392                    the two SIDs, i.e. ahp returns 0. */
2393                 rc = avc_has_perm(&selinux_state,
2394                                   old_tsec->sid, new_tsec->sid,
2395                                   SECCLASS_PROCESS, PROCESS__NOATSECURE,
2396                                   NULL);
2397                 bprm->secureexec |= !!rc;
2398         }
2399
2400         return 0;
2401 }
2402
2403 static int match_file(const void *p, struct file *file, unsigned fd)
2404 {
2405         return file_has_perm(p, file, file_to_av(file)) ? fd + 1 : 0;
2406 }
2407
2408 /* Derived from fs/exec.c:flush_old_files. */
2409 static inline void flush_unauthorized_files(const struct cred *cred,
2410                                             struct files_struct *files)
2411 {
2412         struct file *file, *devnull = NULL;
2413         struct tty_struct *tty;
2414         int drop_tty = 0;
2415         unsigned n;
2416
2417         tty = get_current_tty();
2418         if (tty) {
2419                 spin_lock(&tty->files_lock);
2420                 if (!list_empty(&tty->tty_files)) {
2421                         struct tty_file_private *file_priv;
2422
2423                         /* Revalidate access to controlling tty.
2424                            Use file_path_has_perm on the tty path directly
2425                            rather than using file_has_perm, as this particular
2426                            open file may belong to another process and we are
2427                            only interested in the inode-based check here. */
2428                         file_priv = list_first_entry(&tty->tty_files,
2429                                                 struct tty_file_private, list);
2430                         file = file_priv->file;
2431                         if (file_path_has_perm(cred, file, FILE__READ | FILE__WRITE))
2432                                 drop_tty = 1;
2433                 }
2434                 spin_unlock(&tty->files_lock);
2435                 tty_kref_put(tty);
2436         }
2437         /* Reset controlling tty. */
2438         if (drop_tty)
2439                 no_tty();
2440
2441         /* Revalidate access to inherited open files. */
2442         n = iterate_fd(files, 0, match_file, cred);
2443         if (!n) /* none found? */
2444                 return;
2445
2446         devnull = dentry_open(&selinux_null, O_RDWR, cred);
2447         if (IS_ERR(devnull))
2448                 devnull = NULL;
2449         /* replace all the matching ones with this */
2450         do {
2451                 replace_fd(n - 1, devnull, 0);
2452         } while ((n = iterate_fd(files, n, match_file, cred)) != 0);
2453         if (devnull)
2454                 fput(devnull);
2455 }
2456
2457 /*
2458  * Prepare a process for imminent new credential changes due to exec
2459  */
2460 static void selinux_bprm_committing_creds(struct linux_binprm *bprm)
2461 {
2462         struct task_security_struct *new_tsec;
2463         struct rlimit *rlim, *initrlim;
2464         int rc, i;
2465
2466         new_tsec = selinux_cred(bprm->cred);
2467         if (new_tsec->sid == new_tsec->osid)
2468                 return;
2469
2470         /* Close files for which the new task SID is not authorized. */
2471         flush_unauthorized_files(bprm->cred, current->files);
2472
2473         /* Always clear parent death signal on SID transitions. */
2474         current->pdeath_signal = 0;
2475
2476         /* Check whether the new SID can inherit resource limits from the old
2477          * SID.  If not, reset all soft limits to the lower of the current
2478          * task's hard limit and the init task's soft limit.
2479          *
2480          * Note that the setting of hard limits (even to lower them) can be
2481          * controlled by the setrlimit check.  The inclusion of the init task's
2482          * soft limit into the computation is to avoid resetting soft limits
2483          * higher than the default soft limit for cases where the default is
2484          * lower than the hard limit, e.g. RLIMIT_CORE or RLIMIT_STACK.
2485          */
2486         rc = avc_has_perm(&selinux_state,
2487                           new_tsec->osid, new_tsec->sid, SECCLASS_PROCESS,
2488                           PROCESS__RLIMITINH, NULL);
2489         if (rc) {
2490                 /* protect against do_prlimit() */
2491                 task_lock(current);
2492                 for (i = 0; i < RLIM_NLIMITS; i++) {
2493                         rlim = current->signal->rlim + i;
2494                         initrlim = init_task.signal->rlim + i;
2495                         rlim->rlim_cur = min(rlim->rlim_max, initrlim->rlim_cur);
2496                 }
2497                 task_unlock(current);
2498                 if (IS_ENABLED(CONFIG_POSIX_TIMERS))
2499                         update_rlimit_cpu(current, rlimit(RLIMIT_CPU));
2500         }
2501 }
2502
2503 /*
2504  * Clean up the process immediately after the installation of new credentials
2505  * due to exec
2506  */
2507 static void selinux_bprm_committed_creds(struct linux_binprm *bprm)
2508 {
2509         const struct task_security_struct *tsec = selinux_cred(current_cred());
2510         u32 osid, sid;
2511         int rc;
2512
2513         osid = tsec->osid;
2514         sid = tsec->sid;
2515
2516         if (sid == osid)
2517                 return;
2518
2519         /* Check whether the new SID can inherit signal state from the old SID.
2520          * If not, clear itimers to avoid subsequent signal generation and
2521          * flush and unblock signals.
2522          *
2523          * This must occur _after_ the task SID has been updated so that any
2524          * kill done after the flush will be checked against the new SID.
2525          */
2526         rc = avc_has_perm(&selinux_state,
2527                           osid, sid, SECCLASS_PROCESS, PROCESS__SIGINH, NULL);
2528         if (rc) {
2529                 clear_itimer();
2530
2531                 spin_lock_irq(&current->sighand->siglock);
2532                 if (!fatal_signal_pending(current)) {
2533                         flush_sigqueue(&current->pending);
2534                         flush_sigqueue(&current->signal->shared_pending);
2535                         flush_signal_handlers(current, 1);
2536                         sigemptyset(&current->blocked);
2537                         recalc_sigpending();
2538                 }
2539                 spin_unlock_irq(&current->sighand->siglock);
2540         }
2541
2542         /* Wake up the parent if it is waiting so that it can recheck
2543          * wait permission to the new task SID. */
2544         read_lock(&tasklist_lock);
2545         __wake_up_parent(current, current->real_parent);
2546         read_unlock(&tasklist_lock);
2547 }
2548
2549 /* superblock security operations */
2550
2551 static int selinux_sb_alloc_security(struct super_block *sb)
2552 {
2553         struct superblock_security_struct *sbsec;
2554
2555         sbsec = kzalloc(sizeof(struct superblock_security_struct), GFP_KERNEL);
2556         if (!sbsec)
2557                 return -ENOMEM;
2558
2559         mutex_init(&sbsec->lock);
2560         INIT_LIST_HEAD(&sbsec->isec_head);
2561         spin_lock_init(&sbsec->isec_lock);
2562         sbsec->sb = sb;
2563         sbsec->sid = SECINITSID_UNLABELED;
2564         sbsec->def_sid = SECINITSID_FILE;
2565         sbsec->mntpoint_sid = SECINITSID_UNLABELED;
2566         sb->s_security = sbsec;
2567
2568         return 0;
2569 }
2570
2571 static void selinux_sb_free_security(struct super_block *sb)
2572 {
2573         superblock_free_security(sb);
2574 }
2575
2576 static inline int opt_len(const char *s)
2577 {
2578         bool open_quote = false;
2579         int len;
2580         char c;
2581
2582         for (len = 0; (c = s[len]) != '\0'; len++) {
2583                 if (c == '"')
2584                         open_quote = !open_quote;
2585                 if (c == ',' && !open_quote)
2586                         break;
2587         }
2588         return len;
2589 }
2590
2591 static int selinux_sb_eat_lsm_opts(char *options, void **mnt_opts)
2592 {
2593         char *from = options;
2594         char *to = options;
2595         bool first = true;
2596         int rc;
2597
2598         while (1) {
2599                 int len = opt_len(from);
2600                 int token;
2601                 char *arg = NULL;
2602
2603                 token = match_opt_prefix(from, len, &arg);
2604
2605                 if (token != Opt_error) {
2606                         char *p, *q;
2607
2608                         /* strip quotes */
2609                         if (arg) {
2610                                 for (p = q = arg; p < from + len; p++) {
2611                                         char c = *p;
2612                                         if (c != '"')
2613                                                 *q++ = c;
2614                                 }
2615                                 arg = kmemdup_nul(arg, q - arg, GFP_KERNEL);
2616                                 if (!arg) {
2617                                         rc = -ENOMEM;
2618                                         goto free_opt;
2619                                 }
2620                         }
2621                         rc = selinux_add_opt(token, arg, mnt_opts);
2622                         if (unlikely(rc)) {
2623                                 kfree(arg);
2624                                 goto free_opt;
2625                         }
2626                 } else {
2627                         if (!first) {   // copy with preceding comma
2628                                 from--;
2629                                 len++;
2630                         }
2631                         if (to != from)
2632                                 memmove(to, from, len);
2633                         to += len;
2634                         first = false;
2635                 }
2636                 if (!from[len])
2637                         break;
2638                 from += len + 1;
2639         }
2640         *to = '\0';
2641         return 0;
2642
2643 free_opt:
2644         if (*mnt_opts) {
2645                 selinux_free_mnt_opts(*mnt_opts);
2646                 *mnt_opts = NULL;
2647         }
2648         return rc;
2649 }
2650
2651 static int selinux_sb_remount(struct super_block *sb, void *mnt_opts)
2652 {
2653         struct selinux_mnt_opts *opts = mnt_opts;
2654         struct superblock_security_struct *sbsec = sb->s_security;
2655         u32 sid;
2656         int rc;
2657
2658         if (!(sbsec->flags & SE_SBINITIALIZED))
2659                 return 0;
2660
2661         if (!opts)
2662                 return 0;
2663
2664         if (opts->fscontext) {
2665                 rc = parse_sid(sb, opts->fscontext, &sid);
2666                 if (rc)
2667                         return rc;
2668                 if (bad_option(sbsec, FSCONTEXT_MNT, sbsec->sid, sid))
2669                         goto out_bad_option;
2670         }
2671         if (opts->context) {
2672                 rc = parse_sid(sb, opts->context, &sid);
2673                 if (rc)
2674                         return rc;
2675                 if (bad_option(sbsec, CONTEXT_MNT, sbsec->mntpoint_sid, sid))
2676                         goto out_bad_option;
2677         }
2678         if (opts->rootcontext) {
2679                 struct inode_security_struct *root_isec;
2680                 root_isec = backing_inode_security(sb->s_root);
2681                 rc = parse_sid(sb, opts->rootcontext, &sid);
2682                 if (rc)
2683                         return rc;
2684                 if (bad_option(sbsec, ROOTCONTEXT_MNT, root_isec->sid, sid))
2685                         goto out_bad_option;
2686         }
2687         if (opts->defcontext) {
2688                 rc = parse_sid(sb, opts->defcontext, &sid);
2689                 if (rc)
2690                         return rc;
2691                 if (bad_option(sbsec, DEFCONTEXT_MNT, sbsec->def_sid, sid))
2692                         goto out_bad_option;
2693         }
2694         return 0;
2695
2696 out_bad_option:
2697         pr_warn("SELinux: unable to change security options "
2698                "during remount (dev %s, type=%s)\n", sb->s_id,
2699                sb->s_type->name);
2700         return -EINVAL;
2701 }
2702
2703 static int selinux_sb_kern_mount(struct super_block *sb)
2704 {
2705         const struct cred *cred = current_cred();
2706         struct common_audit_data ad;
2707
2708         ad.type = LSM_AUDIT_DATA_DENTRY;
2709         ad.u.dentry = sb->s_root;
2710         return superblock_has_perm(cred, sb, FILESYSTEM__MOUNT, &ad);
2711 }
2712
2713 static int selinux_sb_statfs(struct dentry *dentry)
2714 {
2715         const struct cred *cred = current_cred();
2716         struct common_audit_data ad;
2717
2718         ad.type = LSM_AUDIT_DATA_DENTRY;
2719         ad.u.dentry = dentry->d_sb->s_root;
2720         return superblock_has_perm(cred, dentry->d_sb, FILESYSTEM__GETATTR, &ad);
2721 }
2722
2723 static int selinux_mount(const char *dev_name,
2724                          const struct path *path,
2725                          const char *type,
2726                          unsigned long flags,
2727                          void *data)
2728 {
2729         const struct cred *cred = current_cred();
2730
2731         if (flags & MS_REMOUNT)
2732                 return superblock_has_perm(cred, path->dentry->d_sb,
2733                                            FILESYSTEM__REMOUNT, NULL);
2734         else
2735                 return path_has_perm(cred, path, FILE__MOUNTON);
2736 }
2737
2738 static int selinux_move_mount(const struct path *from_path,
2739                               const struct path *to_path)
2740 {
2741         const struct cred *cred = current_cred();
2742
2743         return path_has_perm(cred, to_path, FILE__MOUNTON);
2744 }
2745
2746 static int selinux_umount(struct vfsmount *mnt, int flags)
2747 {
2748         const struct cred *cred = current_cred();
2749
2750         return superblock_has_perm(cred, mnt->mnt_sb,
2751                                    FILESYSTEM__UNMOUNT, NULL);
2752 }
2753
2754 static int selinux_fs_context_dup(struct fs_context *fc,
2755                                   struct fs_context *src_fc)
2756 {
2757         const struct selinux_mnt_opts *src = src_fc->security;
2758         struct selinux_mnt_opts *opts;
2759
2760         if (!src)
2761                 return 0;
2762
2763         fc->security = kzalloc(sizeof(struct selinux_mnt_opts), GFP_KERNEL);
2764         if (!fc->security)
2765                 return -ENOMEM;
2766
2767         opts = fc->security;
2768
2769         if (src->fscontext) {
2770                 opts->fscontext = kstrdup(src->fscontext, GFP_KERNEL);
2771                 if (!opts->fscontext)
2772                         return -ENOMEM;
2773         }
2774         if (src->context) {
2775                 opts->context = kstrdup(src->context, GFP_KERNEL);
2776                 if (!opts->context)
2777                         return -ENOMEM;
2778         }
2779         if (src->rootcontext) {
2780                 opts->rootcontext = kstrdup(src->rootcontext, GFP_KERNEL);
2781                 if (!opts->rootcontext)
2782                         return -ENOMEM;
2783         }
2784         if (src->defcontext) {
2785                 opts->defcontext = kstrdup(src->defcontext, GFP_KERNEL);
2786                 if (!opts->defcontext)
2787                         return -ENOMEM;
2788         }
2789         return 0;
2790 }
2791
2792 static const struct fs_parameter_spec selinux_fs_parameters[] = {
2793         fsparam_string(CONTEXT_STR,     Opt_context),
2794         fsparam_string(DEFCONTEXT_STR,  Opt_defcontext),
2795         fsparam_string(FSCONTEXT_STR,   Opt_fscontext),
2796         fsparam_string(ROOTCONTEXT_STR, Opt_rootcontext),
2797         fsparam_flag  (SECLABEL_STR,    Opt_seclabel),
2798         {}
2799 };
2800
2801 static int selinux_fs_context_parse_param(struct fs_context *fc,
2802                                           struct fs_parameter *param)
2803 {
2804         struct fs_parse_result result;
2805         int opt, rc;
2806
2807         opt = fs_parse(fc, selinux_fs_parameters, param, &result);
2808         if (opt < 0)
2809                 return opt;
2810
2811         rc = selinux_add_opt(opt, param->string, &fc->security);
2812         if (!rc) {
2813                 param->string = NULL;
2814                 rc = 1;
2815         }
2816         return rc;
2817 }
2818
2819 /* inode security operations */
2820
2821 static int selinux_inode_alloc_security(struct inode *inode)
2822 {
2823         struct inode_security_struct *isec = selinux_inode(inode);
2824         u32 sid = current_sid();
2825
2826         spin_lock_init(&isec->lock);
2827         INIT_LIST_HEAD(&isec->list);
2828         isec->inode = inode;
2829         isec->sid = SECINITSID_UNLABELED;
2830         isec->sclass = SECCLASS_FILE;
2831         isec->task_sid = sid;
2832         isec->initialized = LABEL_INVALID;
2833
2834         return 0;
2835 }
2836
2837 static void selinux_inode_free_security(struct inode *inode)
2838 {
2839         inode_free_security(inode);
2840 }
2841
2842 static int selinux_dentry_init_security(struct dentry *dentry, int mode,
2843                                         const struct qstr *name, void **ctx,
2844                                         u32 *ctxlen)
2845 {
2846         u32 newsid;
2847         int rc;
2848
2849         rc = selinux_determine_inode_label(selinux_cred(current_cred()),
2850                                            d_inode(dentry->d_parent), name,
2851                                            inode_mode_to_security_class(mode),
2852                                            &newsid);
2853         if (rc)
2854                 return rc;
2855
2856         return security_sid_to_context(&selinux_state, newsid, (char **)ctx,
2857                                        ctxlen);
2858 }
2859
2860 static int selinux_dentry_create_files_as(struct dentry *dentry, int mode,
2861                                           struct qstr *name,
2862                                           const struct cred *old,
2863                                           struct cred *new)
2864 {
2865         u32 newsid;
2866         int rc;
2867         struct task_security_struct *tsec;
2868
2869         rc = selinux_determine_inode_label(selinux_cred(old),
2870                                            d_inode(dentry->d_parent), name,
2871                                            inode_mode_to_security_class(mode),
2872                                            &newsid);
2873         if (rc)
2874                 return rc;
2875
2876         tsec = selinux_cred(new);
2877         tsec->create_sid = newsid;
2878         return 0;
2879 }
2880
2881 static int selinux_inode_init_security(struct inode *inode, struct inode *dir,
2882                                        const struct qstr *qstr,
2883                                        const char **name,
2884                                        void **value, size_t *len)
2885 {
2886         const struct task_security_struct *tsec = selinux_cred(current_cred());
2887         struct superblock_security_struct *sbsec;
2888         u32 newsid, clen;
2889         int rc;
2890         char *context;
2891
2892         sbsec = dir->i_sb->s_security;
2893
2894         newsid = tsec->create_sid;
2895
2896         rc = selinux_determine_inode_label(tsec, dir, qstr,
2897                 inode_mode_to_security_class(inode->i_mode),
2898                 &newsid);
2899         if (rc)
2900                 return rc;
2901
2902         /* Possibly defer initialization to selinux_complete_init. */
2903         if (sbsec->flags & SE_SBINITIALIZED) {
2904                 struct inode_security_struct *isec = selinux_inode(inode);
2905                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
2906                 isec->sid = newsid;
2907                 isec->initialized = LABEL_INITIALIZED;
2908         }
2909
2910         if (!selinux_initialized(&selinux_state) ||
2911             !(sbsec->flags & SBLABEL_MNT))
2912                 return -EOPNOTSUPP;
2913
2914         if (name)
2915                 *name = XATTR_SELINUX_SUFFIX;
2916
2917         if (value && len) {
2918                 rc = security_sid_to_context_force(&selinux_state, newsid,
2919                                                    &context, &clen);
2920                 if (rc)
2921                         return rc;
2922                 *value = context;
2923                 *len = clen;
2924         }
2925
2926         return 0;
2927 }
2928
2929 static int selinux_inode_create(struct inode *dir, struct dentry *dentry, umode_t mode)
2930 {
2931         return may_create(dir, dentry, SECCLASS_FILE);
2932 }
2933
2934 static int selinux_inode_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2935 {
2936         return may_link(dir, old_dentry, MAY_LINK);
2937 }
2938
2939 static int selinux_inode_unlink(struct inode *dir, struct dentry *dentry)
2940 {
2941         return may_link(dir, dentry, MAY_UNLINK);
2942 }
2943
2944 static int selinux_inode_symlink(struct inode *dir, struct dentry *dentry, const char *name)
2945 {
2946         return may_create(dir, dentry, SECCLASS_LNK_FILE);
2947 }
2948
2949 static int selinux_inode_mkdir(struct inode *dir, struct dentry *dentry, umode_t mask)
2950 {
2951         return may_create(dir, dentry, SECCLASS_DIR);
2952 }
2953
2954 static int selinux_inode_rmdir(struct inode *dir, struct dentry *dentry)
2955 {
2956         return may_link(dir, dentry, MAY_RMDIR);
2957 }
2958
2959 static int selinux_inode_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2960 {
2961         return may_create(dir, dentry, inode_mode_to_security_class(mode));
2962 }
2963
2964 static int selinux_inode_rename(struct inode *old_inode, struct dentry *old_dentry,
2965                                 struct inode *new_inode, struct dentry *new_dentry)
2966 {
2967         return may_rename(old_inode, old_dentry, new_inode, new_dentry);
2968 }
2969
2970 static int selinux_inode_readlink(struct dentry *dentry)
2971 {
2972         const struct cred *cred = current_cred();
2973
2974         return dentry_has_perm(cred, dentry, FILE__READ);
2975 }
2976
2977 static int selinux_inode_follow_link(struct dentry *dentry, struct inode *inode,
2978                                      bool rcu)
2979 {
2980         const struct cred *cred = current_cred();
2981         struct common_audit_data ad;
2982         struct inode_security_struct *isec;
2983         u32 sid;
2984
2985         validate_creds(cred);
2986
2987         ad.type = LSM_AUDIT_DATA_DENTRY;
2988         ad.u.dentry = dentry;
2989         sid = cred_sid(cred);
2990         isec = inode_security_rcu(inode, rcu);
2991         if (IS_ERR(isec))
2992                 return PTR_ERR(isec);
2993
2994         return avc_has_perm_flags(&selinux_state,
2995                                   sid, isec->sid, isec->sclass, FILE__READ, &ad,
2996                                   rcu ? MAY_NOT_BLOCK : 0);
2997 }
2998
2999 static noinline int audit_inode_permission(struct inode *inode,
3000                                            u32 perms, u32 audited, u32 denied,
3001                                            int result)
3002 {
3003         struct common_audit_data ad;
3004         struct inode_security_struct *isec = selinux_inode(inode);
3005         int rc;
3006
3007         ad.type = LSM_AUDIT_DATA_INODE;
3008         ad.u.inode = inode;
3009
3010         rc = slow_avc_audit(&selinux_state,
3011                             current_sid(), isec->sid, isec->sclass, perms,
3012                             audited, denied, result, &ad);
3013         if (rc)
3014                 return rc;
3015         return 0;
3016 }
3017
3018 static int selinux_inode_permission(struct inode *inode, int mask)
3019 {
3020         const struct cred *cred = current_cred();
3021         u32 perms;
3022         bool from_access;
3023         bool no_block = mask & MAY_NOT_BLOCK;
3024         struct inode_security_struct *isec;
3025         u32 sid;
3026         struct av_decision avd;
3027         int rc, rc2;
3028         u32 audited, denied;
3029
3030         from_access = mask & MAY_ACCESS;
3031         mask &= (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND);
3032
3033         /* No permission to check.  Existence test. */
3034         if (!mask)
3035                 return 0;
3036
3037         validate_creds(cred);
3038
3039         if (unlikely(IS_PRIVATE(inode)))
3040                 return 0;
3041
3042         perms = file_mask_to_av(inode->i_mode, mask);
3043
3044         sid = cred_sid(cred);
3045         isec = inode_security_rcu(inode, no_block);
3046         if (IS_ERR(isec))
3047                 return PTR_ERR(isec);
3048
3049         rc = avc_has_perm_noaudit(&selinux_state,
3050                                   sid, isec->sid, isec->sclass, perms,
3051                                   no_block ? AVC_NONBLOCKING : 0,
3052                                   &avd);
3053         audited = avc_audit_required(perms, &avd, rc,
3054                                      from_access ? FILE__AUDIT_ACCESS : 0,
3055                                      &denied);
3056         if (likely(!audited))
3057                 return rc;
3058
3059         /* fall back to ref-walk if we have to generate audit */
3060         if (no_block)
3061                 return -ECHILD;
3062
3063         rc2 = audit_inode_permission(inode, perms, audited, denied, rc);
3064         if (rc2)
3065                 return rc2;
3066         return rc;
3067 }
3068
3069 static int selinux_inode_setattr(struct dentry *dentry, struct iattr *iattr)
3070 {
3071         const struct cred *cred = current_cred();
3072         struct inode *inode = d_backing_inode(dentry);
3073         unsigned int ia_valid = iattr->ia_valid;
3074         __u32 av = FILE__WRITE;
3075
3076         /* ATTR_FORCE is just used for ATTR_KILL_S[UG]ID. */
3077         if (ia_valid & ATTR_FORCE) {
3078                 ia_valid &= ~(ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_MODE |
3079                               ATTR_FORCE);
3080                 if (!ia_valid)
3081                         return 0;
3082         }
3083
3084         if (ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID |
3085                         ATTR_ATIME_SET | ATTR_MTIME_SET | ATTR_TIMES_SET))
3086                 return dentry_has_perm(cred, dentry, FILE__SETATTR);
3087
3088         if (selinux_policycap_openperm() &&
3089             inode->i_sb->s_magic != SOCKFS_MAGIC &&
3090             (ia_valid & ATTR_SIZE) &&
3091             !(ia_valid & ATTR_FILE))
3092                 av |= FILE__OPEN;
3093
3094         return dentry_has_perm(cred, dentry, av);
3095 }
3096
3097 static int selinux_inode_getattr(const struct path *path)
3098 {
3099         return path_has_perm(current_cred(), path, FILE__GETATTR);
3100 }
3101
3102 static bool has_cap_mac_admin(bool audit)
3103 {
3104         const struct cred *cred = current_cred();
3105         unsigned int opts = audit ? CAP_OPT_NONE : CAP_OPT_NOAUDIT;
3106
3107         if (cap_capable(cred, &init_user_ns, CAP_MAC_ADMIN, opts))
3108                 return false;
3109         if (cred_has_capability(cred, CAP_MAC_ADMIN, opts, true))
3110                 return false;
3111         return true;
3112 }
3113
3114 static int selinux_inode_setxattr(struct dentry *dentry, const char *name,
3115                                   const void *value, size_t size, int flags)
3116 {
3117         struct inode *inode = d_backing_inode(dentry);
3118         struct inode_security_struct *isec;
3119         struct superblock_security_struct *sbsec;
3120         struct common_audit_data ad;
3121         u32 newsid, sid = current_sid();
3122         int rc = 0;
3123
3124         if (strcmp(name, XATTR_NAME_SELINUX)) {
3125                 rc = cap_inode_setxattr(dentry, name, value, size, flags);
3126                 if (rc)
3127                         return rc;
3128
3129                 /* Not an attribute we recognize, so just check the
3130                    ordinary setattr permission. */
3131                 return dentry_has_perm(current_cred(), dentry, FILE__SETATTR);
3132         }
3133
3134         if (!selinux_initialized(&selinux_state))
3135                 return (inode_owner_or_capable(inode) ? 0 : -EPERM);
3136
3137         sbsec = inode->i_sb->s_security;
3138         if (!(sbsec->flags & SBLABEL_MNT))
3139                 return -EOPNOTSUPP;
3140
3141         if (!inode_owner_or_capable(inode))
3142                 return -EPERM;
3143
3144         ad.type = LSM_AUDIT_DATA_DENTRY;
3145         ad.u.dentry = dentry;
3146
3147         isec = backing_inode_security(dentry);
3148         rc = avc_has_perm(&selinux_state,
3149                           sid, isec->sid, isec->sclass,
3150                           FILE__RELABELFROM, &ad);
3151         if (rc)
3152                 return rc;
3153
3154         rc = security_context_to_sid(&selinux_state, value, size, &newsid,
3155                                      GFP_KERNEL);
3156         if (rc == -EINVAL) {
3157                 if (!has_cap_mac_admin(true)) {
3158                         struct audit_buffer *ab;
3159                         size_t audit_size;
3160
3161                         /* We strip a nul only if it is at the end, otherwise the
3162                          * context contains a nul and we should audit that */
3163                         if (value) {
3164                                 const char *str = value;
3165
3166                                 if (str[size - 1] == '\0')
3167                                         audit_size = size - 1;
3168                                 else
3169                                         audit_size = size;
3170                         } else {
3171                                 audit_size = 0;
3172                         }
3173                         ab = audit_log_start(audit_context(),
3174                                              GFP_ATOMIC, AUDIT_SELINUX_ERR);
3175                         audit_log_format(ab, "op=setxattr invalid_context=");
3176                         audit_log_n_untrustedstring(ab, value, audit_size);
3177                         audit_log_end(ab);
3178
3179                         return rc;
3180                 }
3181                 rc = security_context_to_sid_force(&selinux_state, value,
3182                                                    size, &newsid);
3183         }
3184         if (rc)
3185                 return rc;
3186
3187         rc = avc_has_perm(&selinux_state,
3188                           sid, newsid, isec->sclass,
3189                           FILE__RELABELTO, &ad);
3190         if (rc)
3191                 return rc;
3192
3193         rc = security_validate_transition(&selinux_state, isec->sid, newsid,
3194                                           sid, isec->sclass);
3195         if (rc)
3196                 return rc;
3197
3198         return avc_has_perm(&selinux_state,
3199                             newsid,
3200                             sbsec->sid,
3201                             SECCLASS_FILESYSTEM,
3202                             FILESYSTEM__ASSOCIATE,
3203                             &ad);
3204 }
3205
3206 static void selinux_inode_post_setxattr(struct dentry *dentry, const char *name,
3207                                         const void *value, size_t size,
3208                                         int flags)
3209 {
3210         struct inode *inode = d_backing_inode(dentry);
3211         struct inode_security_struct *isec;
3212         u32 newsid;
3213         int rc;
3214
3215         if (strcmp(name, XATTR_NAME_SELINUX)) {
3216                 /* Not an attribute we recognize, so nothing to do. */
3217                 return;
3218         }
3219
3220         if (!selinux_initialized(&selinux_state)) {
3221                 /* If we haven't even been initialized, then we can't validate
3222                  * against a policy, so leave the label as invalid. It may
3223                  * resolve to a valid label on the next revalidation try if
3224                  * we've since initialized.
3225                  */
3226                 return;
3227         }
3228
3229         rc = security_context_to_sid_force(&selinux_state, value, size,
3230                                            &newsid);
3231         if (rc) {
3232                 pr_err("SELinux:  unable to map context to SID"
3233                        "for (%s, %lu), rc=%d\n",
3234                        inode->i_sb->s_id, inode->i_ino, -rc);
3235                 return;
3236         }
3237
3238         isec = backing_inode_security(dentry);
3239         spin_lock(&isec->lock);
3240         isec->sclass = inode_mode_to_security_class(inode->i_mode);
3241         isec->sid = newsid;
3242         isec->initialized = LABEL_INITIALIZED;
3243         spin_unlock(&isec->lock);
3244
3245         return;
3246 }
3247
3248 static int selinux_inode_getxattr(struct dentry *dentry, const char *name)
3249 {
3250         const struct cred *cred = current_cred();
3251
3252         return dentry_has_perm(cred, dentry, FILE__GETATTR);
3253 }
3254
3255 static int selinux_inode_listxattr(struct dentry *dentry)
3256 {
3257         const struct cred *cred = current_cred();
3258
3259         return dentry_has_perm(cred, dentry, FILE__GETATTR);
3260 }
3261
3262 static int selinux_inode_removexattr(struct dentry *dentry, const char *name)
3263 {
3264         if (strcmp(name, XATTR_NAME_SELINUX)) {
3265                 int rc = cap_inode_removexattr(dentry, name);
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         /* No one is allowed to remove a SELinux security label.
3275            You can change the label, but all data must be labeled. */
3276         return -EACCES;
3277 }
3278
3279 static int selinux_path_notify(const struct path *path, u64 mask,
3280                                                 unsigned int obj_type)
3281 {
3282         int ret;
3283         u32 perm;
3284
3285         struct common_audit_data ad;
3286
3287         ad.type = LSM_AUDIT_DATA_PATH;
3288         ad.u.path = *path;
3289
3290         /*
3291          * Set permission needed based on the type of mark being set.
3292          * Performs an additional check for sb watches.
3293          */
3294         switch (obj_type) {
3295         case FSNOTIFY_OBJ_TYPE_VFSMOUNT:
3296                 perm = FILE__WATCH_MOUNT;
3297                 break;
3298         case FSNOTIFY_OBJ_TYPE_SB:
3299                 perm = FILE__WATCH_SB;
3300                 ret = superblock_has_perm(current_cred(), path->dentry->d_sb,
3301                                                 FILESYSTEM__WATCH, &ad);
3302                 if (ret)
3303                         return ret;
3304                 break;
3305         case FSNOTIFY_OBJ_TYPE_INODE:
3306                 perm = FILE__WATCH;
3307                 break;
3308         default:
3309                 return -EINVAL;
3310         }
3311
3312         /* blocking watches require the file:watch_with_perm permission */
3313         if (mask & (ALL_FSNOTIFY_PERM_EVENTS))
3314                 perm |= FILE__WATCH_WITH_PERM;
3315
3316         /* watches on read-like events need the file:watch_reads permission */
3317         if (mask & (FS_ACCESS | FS_ACCESS_PERM | FS_CLOSE_NOWRITE))
3318                 perm |= FILE__WATCH_READS;
3319
3320         return path_has_perm(current_cred(), path, perm);
3321 }
3322
3323 /*
3324  * Copy the inode security context value to the user.
3325  *
3326  * Permission check is handled by selinux_inode_getxattr hook.
3327  */
3328 static int selinux_inode_getsecurity(struct inode *inode, const char *name, void **buffer, bool alloc)
3329 {
3330         u32 size;
3331         int error;
3332         char *context = NULL;
3333         struct inode_security_struct *isec;
3334
3335         if (strcmp(name, XATTR_SELINUX_SUFFIX))
3336                 return -EOPNOTSUPP;
3337
3338         /*
3339          * If the caller has CAP_MAC_ADMIN, then get the raw context
3340          * value even if it is not defined by current policy; otherwise,
3341          * use the in-core value under current policy.
3342          * Use the non-auditing forms of the permission checks since
3343          * getxattr may be called by unprivileged processes commonly
3344          * and lack of permission just means that we fall back to the
3345          * in-core context value, not a denial.
3346          */
3347         isec = inode_security(inode);
3348         if (has_cap_mac_admin(false))
3349                 error = security_sid_to_context_force(&selinux_state,
3350                                                       isec->sid, &context,
3351                                                       &size);
3352         else
3353                 error = security_sid_to_context(&selinux_state, isec->sid,
3354                                                 &context, &size);
3355         if (error)
3356                 return error;
3357         error = size;
3358         if (alloc) {
3359                 *buffer = context;
3360                 goto out_nofree;
3361         }
3362         kfree(context);
3363 out_nofree:
3364         return error;
3365 }
3366
3367 static int selinux_inode_setsecurity(struct inode *inode, const char *name,
3368                                      const void *value, size_t size, int flags)
3369 {
3370         struct inode_security_struct *isec = inode_security_novalidate(inode);
3371         struct superblock_security_struct *sbsec = inode->i_sb->s_security;
3372         u32 newsid;
3373         int rc;
3374
3375         if (strcmp(name, XATTR_SELINUX_SUFFIX))
3376                 return -EOPNOTSUPP;
3377
3378         if (!(sbsec->flags & SBLABEL_MNT))
3379                 return -EOPNOTSUPP;
3380
3381         if (!value || !size)
3382                 return -EACCES;
3383
3384         rc = security_context_to_sid(&selinux_state, value, size, &newsid,
3385                                      GFP_KERNEL);
3386         if (rc)
3387                 return rc;
3388
3389         spin_lock(&isec->lock);
3390         isec->sclass = inode_mode_to_security_class(inode->i_mode);
3391         isec->sid = newsid;
3392         isec->initialized = LABEL_INITIALIZED;
3393         spin_unlock(&isec->lock);
3394         return 0;
3395 }
3396
3397 static int selinux_inode_listsecurity(struct inode *inode, char *buffer, size_t buffer_size)
3398 {
3399         const int len = sizeof(XATTR_NAME_SELINUX);
3400         if (buffer && len <= buffer_size)
3401                 memcpy(buffer, XATTR_NAME_SELINUX, len);
3402         return len;
3403 }
3404
3405 static void selinux_inode_getsecid(struct inode *inode, u32 *secid)
3406 {
3407         struct inode_security_struct *isec = inode_security_novalidate(inode);
3408         *secid = isec->sid;
3409 }
3410
3411 static int selinux_inode_copy_up(struct dentry *src, struct cred **new)
3412 {
3413         u32 sid;
3414         struct task_security_struct *tsec;
3415         struct cred *new_creds = *new;
3416
3417         if (new_creds == NULL) {
3418                 new_creds = prepare_creds();
3419                 if (!new_creds)
3420                         return -ENOMEM;
3421         }
3422
3423         tsec = selinux_cred(new_creds);
3424         /* Get label from overlay inode and set it in create_sid */
3425         selinux_inode_getsecid(d_inode(src), &sid);
3426         tsec->create_sid = sid;
3427         *new = new_creds;
3428         return 0;
3429 }
3430
3431 static int selinux_inode_copy_up_xattr(const char *name)
3432 {
3433         /* The copy_up hook above sets the initial context on an inode, but we
3434          * don't then want to overwrite it by blindly copying all the lower
3435          * xattrs up.  Instead, we have to filter out SELinux-related xattrs.
3436          */
3437         if (strcmp(name, XATTR_NAME_SELINUX) == 0)
3438                 return 1; /* Discard */
3439         /*
3440          * Any other attribute apart from SELINUX is not claimed, supported
3441          * by selinux.
3442          */
3443         return -EOPNOTSUPP;
3444 }
3445
3446 /* kernfs node operations */
3447
3448 static int selinux_kernfs_init_security(struct kernfs_node *kn_dir,
3449                                         struct kernfs_node *kn)
3450 {
3451         const struct task_security_struct *tsec = selinux_cred(current_cred());
3452         u32 parent_sid, newsid, clen;
3453         int rc;
3454         char *context;
3455
3456         rc = kernfs_xattr_get(kn_dir, XATTR_NAME_SELINUX, NULL, 0);
3457         if (rc == -ENODATA)
3458                 return 0;
3459         else if (rc < 0)
3460                 return rc;
3461
3462         clen = (u32)rc;
3463         context = kmalloc(clen, GFP_KERNEL);
3464         if (!context)
3465                 return -ENOMEM;
3466
3467         rc = kernfs_xattr_get(kn_dir, XATTR_NAME_SELINUX, context, clen);
3468         if (rc < 0) {
3469                 kfree(context);
3470                 return rc;
3471         }
3472
3473         rc = security_context_to_sid(&selinux_state, context, clen, &parent_sid,
3474                                      GFP_KERNEL);
3475         kfree(context);
3476         if (rc)
3477                 return rc;
3478
3479         if (tsec->create_sid) {
3480                 newsid = tsec->create_sid;
3481         } else {
3482                 u16 secclass = inode_mode_to_security_class(kn->mode);
3483                 struct qstr q;
3484
3485                 q.name = kn->name;
3486                 q.hash_len = hashlen_string(kn_dir, kn->name);
3487
3488                 rc = security_transition_sid(&selinux_state, tsec->sid,
3489                                              parent_sid, secclass, &q,
3490                                              &newsid);
3491                 if (rc)
3492                         return rc;
3493         }
3494
3495         rc = security_sid_to_context_force(&selinux_state, newsid,
3496                                            &context, &clen);
3497         if (rc)
3498                 return rc;
3499
3500         rc = kernfs_xattr_set(kn, XATTR_NAME_SELINUX, context, clen,
3501                               XATTR_CREATE);
3502         kfree(context);
3503         return rc;
3504 }
3505
3506
3507 /* file security operations */
3508
3509 static int selinux_revalidate_file_permission(struct file *file, int mask)
3510 {
3511         const struct cred *cred = current_cred();
3512         struct inode *inode = file_inode(file);
3513
3514         /* file_mask_to_av won't add FILE__WRITE if MAY_APPEND is set */
3515         if ((file->f_flags & O_APPEND) && (mask & MAY_WRITE))
3516                 mask |= MAY_APPEND;
3517
3518         return file_has_perm(cred, file,
3519                              file_mask_to_av(inode->i_mode, mask));
3520 }
3521
3522 static int selinux_file_permission(struct file *file, int mask)
3523 {
3524         struct inode *inode = file_inode(file);
3525         struct file_security_struct *fsec = selinux_file(file);
3526         struct inode_security_struct *isec;
3527         u32 sid = current_sid();
3528
3529         if (!mask)
3530                 /* No permission to check.  Existence test. */
3531                 return 0;
3532
3533         isec = inode_security(inode);
3534         if (sid == fsec->sid && fsec->isid == isec->sid &&
3535             fsec->pseqno == avc_policy_seqno(&selinux_state))
3536                 /* No change since file_open check. */
3537                 return 0;
3538
3539         return selinux_revalidate_file_permission(file, mask);
3540 }
3541
3542 static int selinux_file_alloc_security(struct file *file)
3543 {
3544         struct file_security_struct *fsec = selinux_file(file);