security/selinux/hooks.c

   1 /*
   2  *  NSA Security-Enhanced Linux (SELinux) security module
   3  *
   4  *  This file contains the SELinux hook function implementations.
   5  *
   6  *  Authors:  Stephen Smalley, <sds@tycho.nsa.gov>
   7  *            Chris Vance, <cvance@nai.com>
   8  *            Wayne Salamon, <wsalamon@nai.com>
   9  *            James Morris <jmorris@redhat.com>
  10  *
  11  *  Copyright (C) 2001,2002 Networks Associates Technology, Inc.
  12  *  Copyright (C) 2003-2008 Red Hat, Inc., James Morris <jmorris@redhat.com>
  13  *                                         Eric Paris <eparis@redhat.com>
  14  *  Copyright (C) 2004-2005 Trusted Computer Solutions, Inc.
  15  *                          <dgoeddel@trustedcs.com>
  16  *  Copyright (C) 2006, 2007, 2009 Hewlett-Packard Development Company, L.P.
  17  *      Paul Moore <paul@paul-moore.com>
  18  *  Copyright (C) 2007 Hitachi Software Engineering Co., Ltd.
  19  *                     Yuichi Nakamura <ynakam@hitachisoft.jp>
  20  *  Copyright (C) 2016 Mellanox Technologies
  21  *
  22  *      This program is free software; you can redistribute it and/or modify
  23  *      it under the terms of the GNU General Public License version 2,
  24  *      as published by the Free Software Foundation.
  25  */
  26
  27 #include <linux/init.h>
  28 #include <linux/kd.h>
  29 #include <linux/kernel.h>
  30 #include <linux/tracehook.h>
  31 #include <linux/errno.h>
  32 #include <linux/sched/signal.h>
  33 #include <linux/sched/task.h>
  34 #include <linux/lsm_hooks.h>
  35 #include <linux/xattr.h>
  36 #include <linux/capability.h>
  37 #include <linux/unistd.h>
  38 #include <linux/mm.h>
  39 #include <linux/mman.h>
  40 #include <linux/slab.h>
  41 #include <linux/pagemap.h>
  42 #include <linux/proc_fs.h>
  43 #include <linux/swap.h>
  44 #include <linux/spinlock.h>
  45 #include <linux/syscalls.h>
  46 #include <linux/dcache.h>
  47 #include <linux/file.h>
  48 #include <linux/fdtable.h>
  49 #include <linux/namei.h>
  50 #include <linux/mount.h>
  51 #include <linux/netfilter_ipv4.h>
  52 #include <linux/netfilter_ipv6.h>
  53 #include <linux/tty.h>
  54 #include <net/icmp.h>
  55 #include <net/ip.h>             /* for local_port_range[] */
  56 #include <net/tcp.h>            /* struct or_callable used in sock_rcv_skb */
  57 #include <net/inet_connection_sock.h>
  58 #include <net/net_namespace.h>
  59 #include <net/netlabel.h>
  60 #include <linux/uaccess.h>
  61 #include <asm/ioctls.h>
  62 #include <linux/atomic.h>
  63 #include <linux/bitops.h>
  64 #include <linux/interrupt.h>
  65 #include <linux/netdevice.h>    /* for network interface checks */
  66 #include <net/netlink.h>
  67 #include <linux/tcp.h>
  68 #include <linux/udp.h>
  69 #include <linux/dccp.h>
  70 #include <linux/sctp.h>
  71 #include <net/sctp/structs.h>
  72 #include <linux/quota.h>
  73 #include <linux/un.h>           /* for Unix socket types */
  74 #include <net/af_unix.h>        /* for Unix socket types */
  75 #include <linux/parser.h>
  76 #include <linux/nfs_mount.h>
  77 #include <net/ipv6.h>
  78 #include <linux/hugetlb.h>
  79 #include <linux/personality.h>
  80 #include <linux/audit.h>
  81 #include <linux/string.h>
  82 #include <linux/selinux.h>
  83 #include <linux/mutex.h>
  84 #include <linux/posix-timers.h>
  85 #include <linux/syslog.h>
  86 #include <linux/user_namespace.h>
  87 #include <linux/export.h>
  88 #include <linux/msg.h>
  89 #include <linux/shm.h>
  90 #include <linux/bpf.h>
  91
  92 #include "avc.h"
  93 #include "objsec.h"
  94 #include "netif.h"
  95 #include "netnode.h"
  96 #include "netport.h"
  97 #include "ibpkey.h"
  98 #include "xfrm.h"
  99 #include "netlabel.h"
 100 #include "audit.h"
 101 #include "avc_ss.h"
 102
 103 struct selinux_state selinux_state;
 104
 105 /* SECMARK reference count */
 106 static atomic_t selinux_secmark_refcount = ATOMIC_INIT(0);
 107
 108 #ifdef CONFIG_SECURITY_SELINUX_DEVELOP
 109 static int selinux_enforcing_boot;
 110
 111 static int __init enforcing_setup(char *str)
 112 {
 113         unsigned long enforcing;
 114         if (!kstrtoul(str, 0, &enforcing))
 115                 selinux_enforcing_boot = enforcing ? 1 : 0;
 116         return 1;
 117 }
 118 __setup("enforcing=", enforcing_setup);
 119 #else
 120 #define selinux_enforcing_boot 1
 121 #endif
 122
 123 #ifdef CONFIG_SECURITY_SELINUX_BOOTPARAM
 124 int selinux_enabled = CONFIG_SECURITY_SELINUX_BOOTPARAM_VALUE;
 125
 126 static int __init selinux_enabled_setup(char *str)
 127 {
 128         unsigned long enabled;
 129         if (!kstrtoul(str, 0, &enabled))
 130                 selinux_enabled = enabled ? 1 : 0;
 131         return 1;
 132 }
 133 __setup("selinux=", selinux_enabled_setup);
 134 #else
 135 int selinux_enabled = 1;
 136 #endif
 137
 138 static unsigned int selinux_checkreqprot_boot =
 139         CONFIG_SECURITY_SELINUX_CHECKREQPROT_VALUE;
 140
 141 static int __init checkreqprot_setup(char *str)
 142 {
 143         unsigned long checkreqprot;
 144
 145         if (!kstrtoul(str, 0, &checkreqprot))
 146                 selinux_checkreqprot_boot = checkreqprot ? 1 : 0;
 147         return 1;
 148 }
 149 __setup("checkreqprot=", checkreqprot_setup);
 150
 151 static struct kmem_cache *sel_inode_cache;
 152 static struct kmem_cache *file_security_cache;
 153
 154 /**
 155  * selinux_secmark_enabled - Check to see if SECMARK is currently enabled
 156  *
 157  * Description:
 158  * This function checks the SECMARK reference counter to see if any SECMARK
 159  * targets are currently configured, if the reference counter is greater than
 160  * zero SECMARK is considered to be enabled.  Returns true (1) if SECMARK is
 161  * enabled, false (0) if SECMARK is disabled.  If the always_check_network
 162  * policy capability is enabled, SECMARK is always considered enabled.
 163  *
 164  */
 165 static int selinux_secmark_enabled(void)
 166 {
 167         return (selinux_policycap_alwaysnetwork() ||
 168                 atomic_read(&selinux_secmark_refcount));
 169 }
 170
 171 /**
 172  * selinux_peerlbl_enabled - Check to see if peer labeling is currently enabled
 173  *
 174  * Description:
 175  * This function checks if NetLabel or labeled IPSEC is enabled.  Returns true
 176  * (1) if any are enabled or false (0) if neither are enabled.  If the
 177  * always_check_network policy capability is enabled, peer labeling
 178  * is always considered enabled.
 179  *
 180  */
 181 static int selinux_peerlbl_enabled(void)
 182 {
 183         return (selinux_policycap_alwaysnetwork() ||
 184                 netlbl_enabled() || selinux_xfrm_enabled());
 185 }
 186
 187 static int selinux_netcache_avc_callback(u32 event)
 188 {
 189         if (event == AVC_CALLBACK_RESET) {
 190                 sel_netif_flush();
 191                 sel_netnode_flush();
 192                 sel_netport_flush();
 193                 synchronize_net();
 194         }
 195         return 0;
 196 }
 197
 198 static int selinux_lsm_notifier_avc_callback(u32 event)
 199 {
 200         if (event == AVC_CALLBACK_RESET) {
 201                 sel_ib_pkey_flush();
 202                 call_lsm_notifier(LSM_POLICY_CHANGE, NULL);
 203         }
 204
 205         return 0;
 206 }
 207
 208 /*
 209  * initialise the security for the init task
 210  */
 211 static void cred_init_security(void)
 212 {
 213         struct cred *cred = (struct cred *) current->real_cred;
 214         struct task_security_struct *tsec;
 215
 216         tsec = kzalloc(sizeof(struct task_security_struct), GFP_KERNEL);
 217         if (!tsec)
 218                 panic("SELinux:  Failed to initialize initial task.\n");
 219
 220         tsec->osid = tsec->sid = SECINITSID_KERNEL;
 221         cred->security = tsec;
 222 }
 223
 224 /*
 225  * get the security ID of a set of credentials
 226  */
 227 static inline u32 cred_sid(const struct cred *cred)
 228 {
 229         const struct task_security_struct *tsec;
 230
 231         tsec = cred->security;
 232         return tsec->sid;
 233 }
 234
 235 /*
 236  * get the objective security ID of a task
 237  */
 238 static inline u32 task_sid(const struct task_struct *task)
 239 {
 240         u32 sid;
 241
 242         rcu_read_lock();
 243         sid = cred_sid(__task_cred(task));
 244         rcu_read_unlock();
 245         return sid;
 246 }
 247
 248 /* Allocate and free functions for each kind of security blob. */
 249
 250 static int inode_alloc_security(struct inode *inode)
 251 {
 252         struct inode_security_struct *isec;
 253         u32 sid = current_sid();
 254
 255         isec = kmem_cache_zalloc(sel_inode_cache, GFP_NOFS);
 256         if (!isec)
 257                 return -ENOMEM;
 258
 259         spin_lock_init(&isec->lock);
 260         INIT_LIST_HEAD(&isec->list);
 261         isec->inode = inode;
 262         isec->sid = SECINITSID_UNLABELED;
 263         isec->sclass = SECCLASS_FILE;
 264         isec->task_sid = sid;
 265         isec->initialized = LABEL_INVALID;
 266         inode->i_security = isec;
 267
 268         return 0;
 269 }
 270
 271 static int inode_doinit_with_dentry(struct inode *inode, struct dentry *opt_dentry);
 272
 273 /*
 274  * Try reloading inode security labels that have been marked as invalid.  The
 275  * @may_sleep parameter indicates when sleeping and thus reloading labels is
 276  * allowed; when set to false, returns -ECHILD when the label is
 277  * invalid.  The @dentry parameter should be set to a dentry of the inode.
 278  */
 279 static int __inode_security_revalidate(struct inode *inode,
 280                                        struct dentry *dentry,
 281                                        bool may_sleep)
 282 {
 283         struct inode_security_struct *isec = inode->i_security;
 284
 285         might_sleep_if(may_sleep);
 286
 287         if (selinux_state.initialized &&
 288             isec->initialized != LABEL_INITIALIZED) {
 289                 if (!may_sleep)
 290                         return -ECHILD;
 291
 292                 /*
 293                  * Try reloading the inode security label.  This will fail if
 294                  * @opt_dentry is NULL and no dentry for this inode can be
 295                  * found; in that case, continue using the old label.
 296                  */
 297                 inode_doinit_with_dentry(inode, dentry);
 298         }
 299         return 0;
 300 }
 301
 302 static struct inode_security_struct *inode_security_novalidate(struct inode *inode)
 303 {
 304         return inode->i_security;
 305 }
 306
 307 static struct inode_security_struct *inode_security_rcu(struct inode *inode, bool rcu)
 308 {
 309         int error;
 310
 311         error = __inode_security_revalidate(inode, NULL, !rcu);
 312         if (error)
 313                 return ERR_PTR(error);
 314         return inode->i_security;
 315 }
 316
 317 /*
 318  * Get the security label of an inode.
 319  */
 320 static struct inode_security_struct *inode_security(struct inode *inode)
 321 {
 322         __inode_security_revalidate(inode, NULL, true);
 323         return inode->i_security;
 324 }
 325
 326 static struct inode_security_struct *backing_inode_security_novalidate(struct dentry *dentry)
 327 {
 328         struct inode *inode = d_backing_inode(dentry);
 329
 330         return inode->i_security;
 331 }
 332
 333 /*
 334  * Get the security label of a dentry's backing inode.
 335  */
 336 static struct inode_security_struct *backing_inode_security(struct dentry *dentry)
 337 {
 338         struct inode *inode = d_backing_inode(dentry);
 339
 340         __inode_security_revalidate(inode, dentry, true);
 341         return inode->i_security;
 342 }
 343
 344 static void inode_free_rcu(struct rcu_head *head)
 345 {
 346         struct inode_security_struct *isec;
 347
 348         isec = container_of(head, struct inode_security_struct, rcu);
 349         kmem_cache_free(sel_inode_cache, isec);
 350 }
 351
 352 static void inode_free_security(struct inode *inode)
 353 {
 354         struct inode_security_struct *isec = inode->i_security;
 355         struct superblock_security_struct *sbsec = inode->i_sb->s_security;
 356
 357         /*
 358          * As not all inode security structures are in a list, we check for
 359          * empty list outside of the lock to make sure that we won't waste
 360          * time taking a lock doing nothing.
 361          *
 362          * The list_del_init() function can be safely called more than once.
 363          * It should not be possible for this function to be called with
 364          * concurrent list_add(), but for better safety against future changes
 365          * in the code, we use list_empty_careful() here.
 366          */
 367         if (!list_empty_careful(&isec->list)) {
 368                 spin_lock(&sbsec->isec_lock);
 369                 list_del_init(&isec->list);
 370                 spin_unlock(&sbsec->isec_lock);
 371         }
 372
 373         /*
 374          * The inode may still be referenced in a path walk and
 375          * a call to selinux_inode_permission() can be made
 376          * after inode_free_security() is called. Ideally, the VFS
 377          * wouldn't do this, but fixing that is a much harder
 378          * job. For now, simply free the i_security via RCU, and
 379          * leave the current inode->i_security pointer intact.
 380          * The inode will be freed after the RCU grace period too.
 381          */
 382         call_rcu(&isec->rcu, inode_free_rcu);
 383 }
 384
 385 static int file_alloc_security(struct file *file)
 386 {
 387         struct file_security_struct *fsec;
 388         u32 sid = current_sid();
 389
 390         fsec = kmem_cache_zalloc(file_security_cache, GFP_KERNEL);
 391         if (!fsec)
 392                 return -ENOMEM;
 393
 394         fsec->sid = sid;
 395         fsec->fown_sid = sid;
 396         file->f_security = fsec;
 397
 398         return 0;
 399 }
 400
 401 static void file_free_security(struct file *file)
 402 {
 403         struct file_security_struct *fsec = file->f_security;
 404         file->f_security = NULL;
 405         kmem_cache_free(file_security_cache, fsec);
 406 }
 407
 408 static int superblock_alloc_security(struct super_block *sb)
 409 {
 410         struct superblock_security_struct *sbsec;
 411
 412         sbsec = kzalloc(sizeof(struct superblock_security_struct), GFP_KERNEL);
 413         if (!sbsec)
 414                 return -ENOMEM;
 415
 416         mutex_init(&sbsec->lock);
 417         INIT_LIST_HEAD(&sbsec->isec_head);
 418         spin_lock_init(&sbsec->isec_lock);
 419         sbsec->sb = sb;
 420         sbsec->sid = SECINITSID_UNLABELED;
 421         sbsec->def_sid = SECINITSID_FILE;
 422         sbsec->mntpoint_sid = SECINITSID_UNLABELED;
 423         sb->s_security = sbsec;
 424
 425         return 0;
 426 }
 427
 428 static void superblock_free_security(struct super_block *sb)
 429 {
 430         struct superblock_security_struct *sbsec = sb->s_security;
 431         sb->s_security = NULL;
 432         kfree(sbsec);
 433 }
 434
 435 static inline int inode_doinit(struct inode *inode)
 436 {
 437         return inode_doinit_with_dentry(inode, NULL);
 438 }
 439
 440 enum {
 441         Opt_error = -1,
 442         Opt_context = 1,
 443         Opt_fscontext = 2,
 444         Opt_defcontext = 3,
 445         Opt_rootcontext = 4,
 446         Opt_labelsupport = 5,
 447         Opt_nextmntopt = 6,
 448 };
 449
 450 #define NUM_SEL_MNT_OPTS        (Opt_nextmntopt - 1)
 451
 452 static const match_table_t tokens = {
 453         {Opt_context, CONTEXT_STR "%s"},
 454         {Opt_fscontext, FSCONTEXT_STR "%s"},
 455         {Opt_defcontext, DEFCONTEXT_STR "%s"},
 456         {Opt_rootcontext, ROOTCONTEXT_STR "%s"},
 457         {Opt_labelsupport, LABELSUPP_STR},
 458         {Opt_error, NULL},
 459 };
 460
 461 #define SEL_MOUNT_FAIL_MSG "SELinux:  duplicate or incompatible mount options\n"
 462
 463 static int may_context_mount_sb_relabel(u32 sid,
 464                         struct superblock_security_struct *sbsec,
 465                         const struct cred *cred)
 466 {
 467         const struct task_security_struct *tsec = cred->security;
 468         int rc;
 469
 470         rc = avc_has_perm(&selinux_state,
 471                           tsec->sid, sbsec->sid, SECCLASS_FILESYSTEM,
 472                           FILESYSTEM__RELABELFROM, NULL);
 473         if (rc)
 474                 return rc;
 475
 476         rc = avc_has_perm(&selinux_state,
 477                           tsec->sid, sid, SECCLASS_FILESYSTEM,
 478                           FILESYSTEM__RELABELTO, NULL);
 479         return rc;
 480 }
 481
 482 static int may_context_mount_inode_relabel(u32 sid,
 483                         struct superblock_security_struct *sbsec,
 484                         const struct cred *cred)
 485 {
 486         const struct task_security_struct *tsec = cred->security;
 487         int rc;
 488         rc = avc_has_perm(&selinux_state,
 489                           tsec->sid, sbsec->sid, SECCLASS_FILESYSTEM,
 490                           FILESYSTEM__RELABELFROM, NULL);
 491         if (rc)
 492                 return rc;
 493
 494         rc = avc_has_perm(&selinux_state,
 495                           sid, sbsec->sid, SECCLASS_FILESYSTEM,
 496                           FILESYSTEM__ASSOCIATE, NULL);
 497         return rc;
 498 }
 499
 500 static int selinux_is_sblabel_mnt(struct super_block *sb)
 501 {
 502         struct superblock_security_struct *sbsec = sb->s_security;
 503
 504         return sbsec->behavior == SECURITY_FS_USE_XATTR ||
 505                 sbsec->behavior == SECURITY_FS_USE_TRANS ||
 506                 sbsec->behavior == SECURITY_FS_USE_TASK ||
 507                 sbsec->behavior == SECURITY_FS_USE_NATIVE ||
 508                 /* Special handling. Genfs but also in-core setxattr handler */
 509                 !strcmp(sb->s_type->name, "sysfs") ||
 510                 !strcmp(sb->s_type->name, "pstore") ||
 511                 !strcmp(sb->s_type->name, "debugfs") ||
 512                 !strcmp(sb->s_type->name, "tracefs") ||
 513                 !strcmp(sb->s_type->name, "rootfs") ||
 514                 (selinux_policycap_cgroupseclabel() &&
 515                  (!strcmp(sb->s_type->name, "cgroup") ||
 516                   !strcmp(sb->s_type->name, "cgroup2")));
 517 }
 518
 519 static int sb_finish_set_opts(struct super_block *sb)
 520 {
 521         struct superblock_security_struct *sbsec = sb->s_security;
 522         struct dentry *root = sb->s_root;
 523         struct inode *root_inode = d_backing_inode(root);
 524         int rc = 0;
 525
 526         if (sbsec->behavior == SECURITY_FS_USE_XATTR) {
 527                 /* Make sure that the xattr handler exists and that no
 528                    error other than -ENODATA is returned by getxattr on
 529                    the root directory.  -ENODATA is ok, as this may be
 530                    the first boot of the SELinux kernel before we have
 531                    assigned xattr values to the filesystem. */
 532                 if (!(root_inode->i_opflags & IOP_XATTR)) {
 533                         pr_warn("SELinux: (dev %s, type %s) has no "
 534                                "xattr support\n", sb->s_id, sb->s_type->name);
 535                         rc = -EOPNOTSUPP;
 536                         goto out;
 537                 }
 538
 539                 rc = __vfs_getxattr(root, root_inode, XATTR_NAME_SELINUX, NULL, 0);
 540                 if (rc < 0 && rc != -ENODATA) {
 541                         if (rc == -EOPNOTSUPP)
 542                                 pr_warn("SELinux: (dev %s, type "
 543                                        "%s) has no security xattr handler\n",
 544                                        sb->s_id, sb->s_type->name);
 545                         else
 546                                 pr_warn("SELinux: (dev %s, type "
 547                                        "%s) getxattr errno %d\n", sb->s_id,
 548                                        sb->s_type->name, -rc);
 549                         goto out;
 550                 }
 551         }
 552
 553         sbsec->flags |= SE_SBINITIALIZED;
 554
 555         /*
 556          * Explicitly set or clear SBLABEL_MNT.  It's not sufficient to simply
 557          * leave the flag untouched because sb_clone_mnt_opts might be handing
 558          * us a superblock that needs the flag to be cleared.
 559          */
 560         if (selinux_is_sblabel_mnt(sb))
 561                 sbsec->flags |= SBLABEL_MNT;
 562         else
 563                 sbsec->flags &= ~SBLABEL_MNT;
 564
 565         /* Initialize the root inode. */
 566         rc = inode_doinit_with_dentry(root_inode, root);
 567
 568         /* Initialize any other inodes associated with the superblock, e.g.
 569            inodes created prior to initial policy load or inodes created
 570            during get_sb by a pseudo filesystem that directly
 571            populates itself. */
 572         spin_lock(&sbsec->isec_lock);
 573 next_inode:
 574         if (!list_empty(&sbsec->isec_head)) {
 575                 struct inode_security_struct *isec =
 576                                 list_entry(sbsec->isec_head.next,
 577                                            struct inode_security_struct, list);
 578                 struct inode *inode = isec->inode;
 579                 list_del_init(&isec->list);
 580                 spin_unlock(&sbsec->isec_lock);
 581                 inode = igrab(inode);
 582                 if (inode) {
 583                         if (!IS_PRIVATE(inode))
 584                                 inode_doinit(inode);
 585                         iput(inode);
 586                 }
 587                 spin_lock(&sbsec->isec_lock);
 588                 goto next_inode;
 589         }
 590         spin_unlock(&sbsec->isec_lock);
 591 out:
 592         return rc;
 593 }
 594
 595 /*
 596  * This function should allow an FS to ask what it's mount security
 597  * options were so it can use those later for submounts, displaying
 598  * mount options, or whatever.
 599  */
 600 static int selinux_get_mnt_opts(const struct super_block *sb,
 601                                 struct security_mnt_opts *opts)
 602 {
 603         int rc = 0, i;
 604         struct superblock_security_struct *sbsec = sb->s_security;
 605         char *context = NULL;
 606         u32 len;
 607         char tmp;
 608
 609         security_init_mnt_opts(opts);
 610
 611         if (!(sbsec->flags & SE_SBINITIALIZED))
 612                 return -EINVAL;
 613
 614         if (!selinux_state.initialized)
 615                 return -EINVAL;
 616
 617         /* make sure we always check enough bits to cover the mask */
 618         BUILD_BUG_ON(SE_MNTMASK >= (1 << NUM_SEL_MNT_OPTS));
 619
 620         tmp = sbsec->flags & SE_MNTMASK;
 621         /* count the number of mount options for this sb */
 622         for (i = 0; i < NUM_SEL_MNT_OPTS; i++) {
 623                 if (tmp & 0x01)
 624                         opts->num_mnt_opts++;
 625                 tmp >>= 1;
 626         }
 627         /* Check if the Label support flag is set */
 628         if (sbsec->flags & SBLABEL_MNT)
 629                 opts->num_mnt_opts++;
 630
 631         opts->mnt_opts = kcalloc(opts->num_mnt_opts, sizeof(char *), GFP_ATOMIC);
 632         if (!opts->mnt_opts) {
 633                 rc = -ENOMEM;
 634                 goto out_free;
 635         }
 636
 637         opts->mnt_opts_flags = kcalloc(opts->num_mnt_opts, sizeof(int), GFP_ATOMIC);
 638         if (!opts->mnt_opts_flags) {
 639                 rc = -ENOMEM;
 640                 goto out_free;
 641         }
 642
 643         i = 0;
 644         if (sbsec->flags & FSCONTEXT_MNT) {
 645                 rc = security_sid_to_context(&selinux_state, sbsec->sid,
 646                                              &context, &len);
 647                 if (rc)
 648                         goto out_free;
 649                 opts->mnt_opts[i] = context;
 650                 opts->mnt_opts_flags[i++] = FSCONTEXT_MNT;
 651         }
 652         if (sbsec->flags & CONTEXT_MNT) {
 653                 rc = security_sid_to_context(&selinux_state,
 654                                              sbsec->mntpoint_sid,
 655                                              &context, &len);
 656                 if (rc)
 657                         goto out_free;
 658                 opts->mnt_opts[i] = context;
 659                 opts->mnt_opts_flags[i++] = CONTEXT_MNT;
 660         }
 661         if (sbsec->flags & DEFCONTEXT_MNT) {
 662                 rc = security_sid_to_context(&selinux_state, sbsec->def_sid,
 663                                              &context, &len);
 664                 if (rc)
 665                         goto out_free;
 666                 opts->mnt_opts[i] = context;
 667                 opts->mnt_opts_flags[i++] = DEFCONTEXT_MNT;
 668         }
 669         if (sbsec->flags & ROOTCONTEXT_MNT) {
 670                 struct dentry *root = sbsec->sb->s_root;
 671                 struct inode_security_struct *isec = backing_inode_security(root);
 672
 673                 rc = security_sid_to_context(&selinux_state, isec->sid,
 674                                              &context, &len);
 675                 if (rc)
 676                         goto out_free;
 677                 opts->mnt_opts[i] = context;
 678                 opts->mnt_opts_flags[i++] = ROOTCONTEXT_MNT;
 679         }
 680         if (sbsec->flags & SBLABEL_MNT) {
 681                 opts->mnt_opts[i] = NULL;
 682                 opts->mnt_opts_flags[i++] = SBLABEL_MNT;
 683         }
 684
 685         BUG_ON(i != opts->num_mnt_opts);
 686
 687         return 0;
 688
 689 out_free:
 690         security_free_mnt_opts(opts);
 691         return rc;
 692 }
 693
 694 static int bad_option(struct superblock_security_struct *sbsec, char flag,
 695                       u32 old_sid, u32 new_sid)
 696 {
 697         char mnt_flags = sbsec->flags & SE_MNTMASK;
 698
 699         /* check if the old mount command had the same options */
 700         if (sbsec->flags & SE_SBINITIALIZED)
 701                 if (!(sbsec->flags & flag) ||
 702                     (old_sid != new_sid))
 703                         return 1;
 704
 705         /* check if we were passed the same options twice,
 706          * aka someone passed context=a,context=b
 707          */
 708         if (!(sbsec->flags & SE_SBINITIALIZED))
 709                 if (mnt_flags & flag)
 710                         return 1;
 711         return 0;
 712 }
 713
 714 /*
 715  * Allow filesystems with binary mount data to explicitly set mount point
 716  * labeling information.
 717  */
 718 static int selinux_set_mnt_opts(struct super_block *sb,
 719                                 struct security_mnt_opts *opts,
 720                                 unsigned long kern_flags,
 721                                 unsigned long *set_kern_flags)
 722 {
 723         const struct cred *cred = current_cred();
 724         int rc = 0, i;
 725         struct superblock_security_struct *sbsec = sb->s_security;
 726         const char *name = sb->s_type->name;
 727         struct dentry *root = sbsec->sb->s_root;
 728         struct inode_security_struct *root_isec;
 729         u32 fscontext_sid = 0, context_sid = 0, rootcontext_sid = 0;
 730         u32 defcontext_sid = 0;
 731         char **mount_options = opts->mnt_opts;
 732         int *flags = opts->mnt_opts_flags;
 733         int num_opts = opts->num_mnt_opts;
 734
 735         mutex_lock(&sbsec->lock);
 736
 737         if (!selinux_state.initialized) {
 738                 if (!num_opts) {
 739                         /* Defer initialization until selinux_complete_init,
 740                            after the initial policy is loaded and the security
 741                            server is ready to handle calls. */
 742                         goto out;
 743                 }
 744                 rc = -EINVAL;
 745                 pr_warn("SELinux: Unable to set superblock options "
 746                         "before the security server is initialized\n");
 747                 goto out;
 748         }
 749         if (kern_flags && !set_kern_flags) {
 750                 /* Specifying internal flags without providing a place to
 751                  * place the results is not allowed */
 752                 rc = -EINVAL;
 753                 goto out;
 754         }
 755
 756         /*
 757          * Binary mount data FS will come through this function twice.  Once
 758          * from an explicit call and once from the generic calls from the vfs.
 759          * Since the generic VFS calls will not contain any security mount data
 760          * we need to skip the double mount verification.
 761          *
 762          * This does open a hole in which we will not notice if the first
 763          * mount using this sb set explict options and a second mount using
 764          * this sb does not set any security options.  (The first options
 765          * will be used for both mounts)
 766          */
 767         if ((sbsec->flags & SE_SBINITIALIZED) && (sb->s_type->fs_flags & FS_BINARY_MOUNTDATA)
 768             && (num_opts == 0))
 769                 goto out;
 770
 771         root_isec = backing_inode_security_novalidate(root);
 772
 773         /*
 774          * parse the mount options, check if they are valid sids.
 775          * also check if someone is trying to mount the same sb more
 776          * than once with different security options.
 777          */
 778         for (i = 0; i < num_opts; i++) {
 779                 u32 sid;
 780
 781                 if (flags[i] == SBLABEL_MNT)
 782                         continue;
 783                 rc = security_context_str_to_sid(&selinux_state,
 784                                                  mount_options[i], &sid,
 785                                                  GFP_KERNEL);
 786                 if (rc) {
 787                         pr_warn("SELinux: security_context_str_to_sid"
 788                                "(%s) failed for (dev %s, type %s) errno=%d\n",
 789                                mount_options[i], sb->s_id, name, rc);
 790                         goto out;
 791                 }
 792                 switch (flags[i]) {
 793                 case FSCONTEXT_MNT:
 794                         fscontext_sid = sid;
 795
 796                         if (bad_option(sbsec, FSCONTEXT_MNT, sbsec->sid,
 797                                         fscontext_sid))
 798                                 goto out_double_mount;
 799
 800                         sbsec->flags |= FSCONTEXT_MNT;
 801                         break;
 802                 case CONTEXT_MNT:
 803                         context_sid = sid;
 804
 805                         if (bad_option(sbsec, CONTEXT_MNT, sbsec->mntpoint_sid,
 806                                         context_sid))
 807                                 goto out_double_mount;
 808
 809                         sbsec->flags |= CONTEXT_MNT;
 810                         break;
 811                 case ROOTCONTEXT_MNT:
 812                         rootcontext_sid = sid;
 813
 814                         if (bad_option(sbsec, ROOTCONTEXT_MNT, root_isec->sid,
 815                                         rootcontext_sid))
 816                                 goto out_double_mount;
 817
 818                         sbsec->flags |= ROOTCONTEXT_MNT;
 819
 820                         break;
 821                 case DEFCONTEXT_MNT:
 822                         defcontext_sid = sid;
 823
 824                         if (bad_option(sbsec, DEFCONTEXT_MNT, sbsec->def_sid,
 825                                         defcontext_sid))
 826                                 goto out_double_mount;
 827
 828                         sbsec->flags |= DEFCONTEXT_MNT;
 829
 830                         break;
 831                 default:
 832                         rc = -EINVAL;
 833                         goto out;
 834                 }
 835         }
 836
 837         if (sbsec->flags & SE_SBINITIALIZED) {
 838                 /* previously mounted with options, but not on this attempt? */
 839                 if ((sbsec->flags & SE_MNTMASK) && !num_opts)
 840                         goto out_double_mount;
 841                 rc = 0;
 842                 goto out;
 843         }
 844
 845         if (strcmp(sb->s_type->name, "proc") == 0)
 846                 sbsec->flags |= SE_SBPROC | SE_SBGENFS;
 847
 848         if (!strcmp(sb->s_type->name, "debugfs") ||
 849             !strcmp(sb->s_type->name, "tracefs") ||
 850             !strcmp(sb->s_type->name, "sysfs") ||
 851             !strcmp(sb->s_type->name, "pstore") ||
 852             !strcmp(sb->s_type->name, "cgroup") ||
 853             !strcmp(sb->s_type->name, "cgroup2"))
 854                 sbsec->flags |= SE_SBGENFS;
 855
 856         if (!sbsec->behavior) {
 857                 /*
 858                  * Determine the labeling behavior to use for this
 859                  * filesystem type.
 860                  */
 861                 rc = security_fs_use(&selinux_state, sb);
 862                 if (rc) {
 863                         pr_warn("%s: security_fs_use(%s) returned %d\n",
 864                                         __func__, sb->s_type->name, rc);
 865                         goto out;
 866                 }
 867         }
 868
 869         /*
 870          * If this is a user namespace mount and the filesystem type is not
 871          * explicitly whitelisted, then no contexts are allowed on the command
 872          * line and security labels must be ignored.
 873          */
 874         if (sb->s_user_ns != &init_user_ns &&
 875             strcmp(sb->s_type->name, "tmpfs") &&
 876             strcmp(sb->s_type->name, "ramfs") &&
 877             strcmp(sb->s_type->name, "devpts")) {
 878                 if (context_sid || fscontext_sid || rootcontext_sid ||
 879                     defcontext_sid) {
 880                         rc = -EACCES;
 881                         goto out;
 882                 }
 883                 if (sbsec->behavior == SECURITY_FS_USE_XATTR) {
 884                         sbsec->behavior = SECURITY_FS_USE_MNTPOINT;
 885                         rc = security_transition_sid(&selinux_state,
 886                                                      current_sid(),
 887                                                      current_sid(),
 888                                                      SECCLASS_FILE, NULL,
 889                                                      &sbsec->mntpoint_sid);
 890                         if (rc)
 891                                 goto out;
 892                 }
 893                 goto out_set_opts;
 894         }
 895
 896         /* sets the context of the superblock for the fs being mounted. */
 897         if (fscontext_sid) {
 898                 rc = may_context_mount_sb_relabel(fscontext_sid, sbsec, cred);
 899                 if (rc)
 900                         goto out;
 901
 902                 sbsec->sid = fscontext_sid;
 903         }
 904
 905         /*
 906          * Switch to using mount point labeling behavior.
 907          * sets the label used on all file below the mountpoint, and will set
 908          * the superblock context if not already set.
 909          */
 910         if (kern_flags & SECURITY_LSM_NATIVE_LABELS && !context_sid) {
 911                 sbsec->behavior = SECURITY_FS_USE_NATIVE;
 912                 *set_kern_flags |= SECURITY_LSM_NATIVE_LABELS;
 913         }
 914
 915         if (context_sid) {
 916                 if (!fscontext_sid) {
 917                         rc = may_context_mount_sb_relabel(context_sid, sbsec,
 918                                                           cred);
 919                         if (rc)
 920                                 goto out;
 921                         sbsec->sid = context_sid;
 922                 } else {
 923                         rc = may_context_mount_inode_relabel(context_sid, sbsec,
 924                                                              cred);
 925                         if (rc)
 926                                 goto out;
 927                 }
 928                 if (!rootcontext_sid)
 929                         rootcontext_sid = context_sid;
 930
 931                 sbsec->mntpoint_sid = context_sid;
 932                 sbsec->behavior = SECURITY_FS_USE_MNTPOINT;
 933         }
 934
 935         if (rootcontext_sid) {
 936                 rc = may_context_mount_inode_relabel(rootcontext_sid, sbsec,
 937                                                      cred);
 938                 if (rc)
 939                         goto out;
 940
 941                 root_isec->sid = rootcontext_sid;
 942                 root_isec->initialized = LABEL_INITIALIZED;
 943         }
 944
 945         if (defcontext_sid) {
 946                 if (sbsec->behavior != SECURITY_FS_USE_XATTR &&
 947                         sbsec->behavior != SECURITY_FS_USE_NATIVE) {
 948                         rc = -EINVAL;
 949                         pr_warn("SELinux: defcontext option is "
 950                                "invalid for this filesystem type\n");
 951                         goto out;
 952                 }
 953
 954                 if (defcontext_sid != sbsec->def_sid) {
 955                         rc = may_context_mount_inode_relabel(defcontext_sid,
 956                                                              sbsec, cred);
 957                         if (rc)
 958                                 goto out;
 959                 }
 960
 961                 sbsec->def_sid = defcontext_sid;
 962         }
 963
 964 out_set_opts:
 965         rc = sb_finish_set_opts(sb);
 966 out:
 967         mutex_unlock(&sbsec->lock);
 968         return rc;
 969 out_double_mount:
 970         rc = -EINVAL;
 971         pr_warn("SELinux: mount invalid.  Same superblock, different "
 972                "security settings for (dev %s, type %s)\n", sb->s_id, name);
 973         goto out;
 974 }
 975
 976 static int selinux_cmp_sb_context(const struct super_block *oldsb,
 977                                     const struct super_block *newsb)
 978 {
 979         struct superblock_security_struct *old = oldsb->s_security;
 980         struct superblock_security_struct *new = newsb->s_security;
 981         char oldflags = old->flags & SE_MNTMASK;
 982         char newflags = new->flags & SE_MNTMASK;
 983
 984         if (oldflags != newflags)
 985                 goto mismatch;
 986         if ((oldflags & FSCONTEXT_MNT) && old->sid != new->sid)
 987                 goto mismatch;
 988         if ((oldflags & CONTEXT_MNT) && old->mntpoint_sid != new->mntpoint_sid)
 989                 goto mismatch;
 990         if ((oldflags & DEFCONTEXT_MNT) && old->def_sid != new->def_sid)
 991                 goto mismatch;
 992         if (oldflags & ROOTCONTEXT_MNT) {
 993                 struct inode_security_struct *oldroot = backing_inode_security(oldsb->s_root);
 994                 struct inode_security_struct *newroot = backing_inode_security(newsb->s_root);
 995                 if (oldroot->sid != newroot->sid)
 996                         goto mismatch;
 997         }
 998         return 0;
 999 mismatch:
1000         pr_warn("SELinux: mount invalid.  Same superblock, "
1001                             "different security settings for (dev %s, "
1002                             "type %s)\n", newsb->s_id, newsb->s_type->name);
1003         return -EBUSY;
1004 }
1005
1006 static int selinux_sb_clone_mnt_opts(const struct super_block *oldsb,
1007                                         struct super_block *newsb,
1008                                         unsigned long kern_flags,
1009                                         unsigned long *set_kern_flags)
1010 {
1011         int rc = 0;
1012         const struct superblock_security_struct *oldsbsec = oldsb->s_security;
1013         struct superblock_security_struct *newsbsec = newsb->s_security;
1014
1015         int set_fscontext =     (oldsbsec->flags & FSCONTEXT_MNT);
1016         int set_context =       (oldsbsec->flags & CONTEXT_MNT);
1017         int set_rootcontext =   (oldsbsec->flags & ROOTCONTEXT_MNT);
1018
1019         /*
1020          * if the parent was able to be mounted it clearly had no special lsm
1021          * mount options.  thus we can safely deal with this superblock later
1022          */
1023         if (!selinux_state.initialized)
1024                 return 0;
1025
1026         /*
1027          * Specifying internal flags without providing a place to
1028          * place the results is not allowed.
1029          */
1030         if (kern_flags && !set_kern_flags)
1031                 return -EINVAL;
1032
1033         /* how can we clone if the old one wasn't set up?? */
1034         BUG_ON(!(oldsbsec->flags & SE_SBINITIALIZED));
1035
1036         /* if fs is reusing a sb, make sure that the contexts match */
1037         if (newsbsec->flags & SE_SBINITIALIZED)
1038                 return selinux_cmp_sb_context(oldsb, newsb);
1039
1040         mutex_lock(&newsbsec->lock);
1041
1042         newsbsec->flags = oldsbsec->flags;
1043
1044         newsbsec->sid = oldsbsec->sid;
1045         newsbsec->def_sid = oldsbsec->def_sid;
1046         newsbsec->behavior = oldsbsec->behavior;
1047
1048         if (newsbsec->behavior == SECURITY_FS_USE_NATIVE &&
1049                 !(kern_flags & SECURITY_LSM_NATIVE_LABELS) && !set_context) {
1050                 rc = security_fs_use(&selinux_state, newsb);
1051                 if (rc)
1052                         goto out;
1053         }
1054
1055         if (kern_flags & SECURITY_LSM_NATIVE_LABELS && !set_context) {
1056                 newsbsec->behavior = SECURITY_FS_USE_NATIVE;
1057                 *set_kern_flags |= SECURITY_LSM_NATIVE_LABELS;
1058         }
1059
1060         if (set_context) {
1061                 u32 sid = oldsbsec->mntpoint_sid;
1062
1063                 if (!set_fscontext)
1064                         newsbsec->sid = sid;
1065                 if (!set_rootcontext) {
1066                         struct inode_security_struct *newisec = backing_inode_security(newsb->s_root);
1067                         newisec->sid = sid;
1068                 }
1069                 newsbsec->mntpoint_sid = sid;
1070         }
1071         if (set_rootcontext) {
1072                 const struct inode_security_struct *oldisec = backing_inode_security(oldsb->s_root);
1073                 struct inode_security_struct *newisec = backing_inode_security(newsb->s_root);
1074
1075                 newisec->sid = oldisec->sid;
1076         }
1077
1078         sb_finish_set_opts(newsb);
1079 out:
1080         mutex_unlock(&newsbsec->lock);
1081         return rc;
1082 }
1083
1084 static int selinux_parse_opts_str(char *options,
1085                                   struct security_mnt_opts *opts)
1086 {
1087         char *p;
1088         char *context = NULL, *defcontext = NULL;
1089         char *fscontext = NULL, *rootcontext = NULL;
1090         int rc, num_mnt_opts = 0;
1091
1092         opts->num_mnt_opts = 0;
1093
1094         /* Standard string-based options. */
1095         while ((p = strsep(&options, "|")) != NULL) {
1096                 int token;
1097                 substring_t args[MAX_OPT_ARGS];
1098
1099                 if (!*p)
1100                         continue;
1101
1102                 token = match_token(p, tokens, args);
1103
1104                 switch (token) {
1105                 case Opt_context:
1106                         if (context || defcontext) {
1107                                 rc = -EINVAL;
1108                                 pr_warn(SEL_MOUNT_FAIL_MSG);
1109                                 goto out_err;
1110                         }
1111                         context = match_strdup(&args[0]);
1112                         if (!context) {
1113                                 rc = -ENOMEM;
1114                                 goto out_err;
1115                         }
1116                         break;
1117
1118                 case Opt_fscontext:
1119                         if (fscontext) {
1120                                 rc = -EINVAL;
1121                                 pr_warn(SEL_MOUNT_FAIL_MSG);
1122                                 goto out_err;
1123                         }
1124                         fscontext = match_strdup(&args[0]);
1125                         if (!fscontext) {
1126                                 rc = -ENOMEM;
1127                                 goto out_err;
1128                         }
1129                         break;
1130
1131                 case Opt_rootcontext:
1132                         if (rootcontext) {
1133                                 rc = -EINVAL;
1134                                 pr_warn(SEL_MOUNT_FAIL_MSG);
1135                                 goto out_err;
1136                         }
1137                         rootcontext = match_strdup(&args[0]);
1138                         if (!rootcontext) {
1139                                 rc = -ENOMEM;
1140                                 goto out_err;
1141                         }
1142                         break;
1143
1144                 case Opt_defcontext:
1145                         if (context || defcontext) {
1146                                 rc = -EINVAL;
1147                                 pr_warn(SEL_MOUNT_FAIL_MSG);
1148                                 goto out_err;
1149                         }
1150                         defcontext = match_strdup(&args[0]);
1151                         if (!defcontext) {
1152                                 rc = -ENOMEM;
1153                                 goto out_err;
1154                         }
1155                         break;
1156                 case Opt_labelsupport:
1157                         break;
1158                 default:
1159                         rc = -EINVAL;
1160                         pr_warn("SELinux:  unknown mount option\n");
1161                         goto out_err;
1162
1163                 }
1164         }
1165
1166         rc = -ENOMEM;
1167         opts->mnt_opts = kcalloc(NUM_SEL_MNT_OPTS, sizeof(char *), GFP_KERNEL);
1168         if (!opts->mnt_opts)
1169                 goto out_err;
1170
1171         opts->mnt_opts_flags = kcalloc(NUM_SEL_MNT_OPTS, sizeof(int),
1172                                        GFP_KERNEL);
1173         if (!opts->mnt_opts_flags)
1174                 goto out_err;
1175
1176         if (fscontext) {
1177                 opts->mnt_opts[num_mnt_opts] = fscontext;
1178                 opts->mnt_opts_flags[num_mnt_opts++] = FSCONTEXT_MNT;
1179         }
1180         if (context) {
1181                 opts->mnt_opts[num_mnt_opts] = context;
1182                 opts->mnt_opts_flags[num_mnt_opts++] = CONTEXT_MNT;
1183         }
1184         if (rootcontext) {
1185                 opts->mnt_opts[num_mnt_opts] = rootcontext;
1186                 opts->mnt_opts_flags[num_mnt_opts++] = ROOTCONTEXT_MNT;
1187         }
1188         if (defcontext) {
1189                 opts->mnt_opts[num_mnt_opts] = defcontext;
1190                 opts->mnt_opts_flags[num_mnt_opts++] = DEFCONTEXT_MNT;
1191         }
1192
1193         opts->num_mnt_opts = num_mnt_opts;
1194         return 0;
1195
1196 out_err:
1197         security_free_mnt_opts(opts);
1198         kfree(context);
1199         kfree(defcontext);
1200         kfree(fscontext);
1201         kfree(rootcontext);
1202         return rc;
1203 }
1204 /*
1205  * string mount options parsing and call set the sbsec
1206  */
1207 static int superblock_doinit(struct super_block *sb, void *data)
1208 {
1209         int rc = 0;
1210         char *options = data;
1211         struct security_mnt_opts opts;
1212
1213         security_init_mnt_opts(&opts);
1214
1215         if (!data)
1216                 goto out;
1217
1218         BUG_ON(sb->s_type->fs_flags & FS_BINARY_MOUNTDATA);
1219
1220         rc = selinux_parse_opts_str(options, &opts);
1221         if (rc)
1222                 goto out_err;
1223
1224 out:
1225         rc = selinux_set_mnt_opts(sb, &opts, 0, NULL);
1226
1227 out_err:
1228         security_free_mnt_opts(&opts);
1229         return rc;
1230 }
1231
1232 static void selinux_write_opts(struct seq_file *m,
1233                                struct security_mnt_opts *opts)
1234 {
1235         int i;
1236         char *prefix;
1237
1238         for (i = 0; i < opts->num_mnt_opts; i++) {
1239                 char *has_comma;
1240
1241                 if (opts->mnt_opts[i])
1242                         has_comma = strchr(opts->mnt_opts[i], ',');
1243                 else
1244                         has_comma = NULL;
1245
1246                 switch (opts->mnt_opts_flags[i]) {
1247                 case CONTEXT_MNT:
1248                         prefix = CONTEXT_STR;
1249                         break;
1250                 case FSCONTEXT_MNT:
1251                         prefix = FSCONTEXT_STR;
1252                         break;
1253                 case ROOTCONTEXT_MNT:
1254                         prefix = ROOTCONTEXT_STR;
1255                         break;
1256                 case DEFCONTEXT_MNT:
1257                         prefix = DEFCONTEXT_STR;
1258                         break;
1259                 case SBLABEL_MNT:
1260                         seq_putc(m, ',');
1261                         seq_puts(m, LABELSUPP_STR);
1262                         continue;
1263                 default:
1264                         BUG();
1265                         return;
1266                 };
1267                 /* we need a comma before each option */
1268                 seq_putc(m, ',');
1269                 seq_puts(m, prefix);
1270                 if (has_comma)
1271                         seq_putc(m, '\"');
1272                 seq_escape(m, opts->mnt_opts[i], "\"\n\\");
1273                 if (has_comma)
1274                         seq_putc(m, '\"');
1275         }
1276 }
1277
1278 static int selinux_sb_show_options(struct seq_file *m, struct super_block *sb)
1279 {
1280         struct security_mnt_opts opts;
1281         int rc;
1282
1283         rc = selinux_get_mnt_opts(sb, &opts);
1284         if (rc) {
1285                 /* before policy load we may get EINVAL, don't show anything */
1286                 if (rc == -EINVAL)
1287                         rc = 0;
1288                 return rc;
1289         }
1290
1291         selinux_write_opts(m, &opts);
1292
1293         security_free_mnt_opts(&opts);
1294
1295         return rc;
1296 }
1297
1298 static inline u16 inode_mode_to_security_class(umode_t mode)
1299 {
1300         switch (mode & S_IFMT) {
1301         case S_IFSOCK:
1302                 return SECCLASS_SOCK_FILE;
1303         case S_IFLNK:
1304                 return SECCLASS_LNK_FILE;
1305         case S_IFREG:
1306                 return SECCLASS_FILE;
1307         case S_IFBLK:
1308                 return SECCLASS_BLK_FILE;
1309         case S_IFDIR:
1310                 return SECCLASS_DIR;
1311         case S_IFCHR:
1312                 return SECCLASS_CHR_FILE;
1313         case S_IFIFO:
1314                 return SECCLASS_FIFO_FILE;
1315
1316         }
1317
1318         return SECCLASS_FILE;
1319 }
1320
1321 static inline int default_protocol_stream(int protocol)
1322 {
1323         return (protocol == IPPROTO_IP || protocol == IPPROTO_TCP);
1324 }
1325
1326 static inline int default_protocol_dgram(int protocol)
1327 {
1328         return (protocol == IPPROTO_IP || protocol == IPPROTO_UDP);
1329 }
1330
1331 static inline u16 socket_type_to_security_class(int family, int type, int protocol)
1332 {
1333         int extsockclass = selinux_policycap_extsockclass();
1334
1335         switch (family) {
1336         case PF_UNIX:
1337                 switch (type) {
1338                 case SOCK_STREAM:
1339                 case SOCK_SEQPACKET:
1340                         return SECCLASS_UNIX_STREAM_SOCKET;
1341                 case SOCK_DGRAM:
1342                 case SOCK_RAW:
1343                         return SECCLASS_UNIX_DGRAM_SOCKET;
1344                 }
1345                 break;
1346         case PF_INET:
1347         case PF_INET6:
1348                 switch (type) {
1349                 case SOCK_STREAM:
1350                 case SOCK_SEQPACKET:
1351                         if (default_protocol_stream(protocol))
1352                                 return SECCLASS_TCP_SOCKET;
1353                         else if (extsockclass && protocol == IPPROTO_SCTP)
1354                                 return SECCLASS_SCTP_SOCKET;
1355                         else
1356                                 return SECCLASS_RAWIP_SOCKET;
1357                 case SOCK_DGRAM:
1358                         if (default_protocol_dgram(protocol))
1359                                 return SECCLASS_UDP_SOCKET;
1360                         else if (extsockclass && (protocol == IPPROTO_ICMP ||
1361                                                   protocol == IPPROTO_ICMPV6))
1362                                 return SECCLASS_ICMP_SOCKET;
1363                         else
1364                                 return SECCLASS_RAWIP_SOCKET;
1365                 case SOCK_DCCP:
1366                         return SECCLASS_DCCP_SOCKET;
1367                 default:
1368                         return SECCLASS_RAWIP_SOCKET;
1369                 }
1370                 break;
1371         case PF_NETLINK:
1372                 switch (protocol) {
1373                 case NETLINK_ROUTE:
1374                         return SECCLASS_NETLINK_ROUTE_SOCKET;
1375                 case NETLINK_SOCK_DIAG:
1376                         return SECCLASS_NETLINK_TCPDIAG_SOCKET;
1377                 case NETLINK_NFLOG:
1378                         return SECCLASS_NETLINK_NFLOG_SOCKET;
1379                 case NETLINK_XFRM:
1380                         return SECCLASS_NETLINK_XFRM_SOCKET;
1381                 case NETLINK_SELINUX:
1382                         return SECCLASS_NETLINK_SELINUX_SOCKET;
1383                 case NETLINK_ISCSI:
1384                         return SECCLASS_NETLINK_ISCSI_SOCKET;
1385                 case NETLINK_AUDIT:
1386                         return SECCLASS_NETLINK_AUDIT_SOCKET;
1387                 case NETLINK_FIB_LOOKUP:
1388                         return SECCLASS_NETLINK_FIB_LOOKUP_SOCKET;
1389                 case NETLINK_CONNECTOR:
1390                         return SECCLASS_NETLINK_CONNECTOR_SOCKET;
1391                 case NETLINK_NETFILTER:
1392                         return SECCLASS_NETLINK_NETFILTER_SOCKET;
1393                 case NETLINK_DNRTMSG:
1394                         return SECCLASS_NETLINK_DNRT_SOCKET;
1395                 case NETLINK_KOBJECT_UEVENT:
1396                         return SECCLASS_NETLINK_KOBJECT_UEVENT_SOCKET;
1397                 case NETLINK_GENERIC:
1398                         return SECCLASS_NETLINK_GENERIC_SOCKET;
1399                 case NETLINK_SCSITRANSPORT:
1400                         return SECCLASS_NETLINK_SCSITRANSPORT_SOCKET;
1401                 case NETLINK_RDMA:
1402                         return SECCLASS_NETLINK_RDMA_SOCKET;
1403                 case NETLINK_CRYPTO:
1404                         return SECCLASS_NETLINK_CRYPTO_SOCKET;
1405                 default:
1406                         return SECCLASS_NETLINK_SOCKET;
1407                 }
1408         case PF_PACKET:
1409                 return SECCLASS_PACKET_SOCKET;
1410         case PF_KEY:
1411                 return SECCLASS_KEY_SOCKET;
1412         case PF_APPLETALK:
1413                 return SECCLASS_APPLETALK_SOCKET;
1414         }
1415
1416         if (extsockclass) {
1417                 switch (family) {
1418                 case PF_AX25:
1419                         return SECCLASS_AX25_SOCKET;
1420                 case PF_IPX:
1421                         return SECCLASS_IPX_SOCKET;
1422                 case PF_NETROM:
1423                         return SECCLASS_NETROM_SOCKET;
1424                 case PF_ATMPVC:
1425                         return SECCLASS_ATMPVC_SOCKET;
1426                 case PF_X25:
1427                         return SECCLASS_X25_SOCKET;
1428                 case PF_ROSE:
1429                         return SECCLASS_ROSE_SOCKET;
1430                 case PF_DECnet:
1431                         return SECCLASS_DECNET_SOCKET;
1432                 case PF_ATMSVC:
1433                         return SECCLASS_ATMSVC_SOCKET;
1434                 case PF_RDS:
1435                         return SECCLASS_RDS_SOCKET;
1436                 case PF_IRDA:
1437                         return SECCLASS_IRDA_SOCKET;
1438                 case PF_PPPOX:
1439                         return SECCLASS_PPPOX_SOCKET;
1440                 case PF_LLC:
1441                         return SECCLASS_LLC_SOCKET;
1442                 case PF_CAN:
1443                         return SECCLASS_CAN_SOCKET;
1444                 case PF_TIPC:
1445                         return SECCLASS_TIPC_SOCKET;
1446                 case PF_BLUETOOTH:
1447                         return SECCLASS_BLUETOOTH_SOCKET;
1448                 case PF_IUCV:
1449                         return SECCLASS_IUCV_SOCKET;
1450                 case PF_RXRPC:
1451                         return SECCLASS_RXRPC_SOCKET;
1452                 case PF_ISDN:
1453                         return SECCLASS_ISDN_SOCKET;
1454                 case PF_PHONET:
1455                         return SECCLASS_PHONET_SOCKET;
1456                 case PF_IEEE802154:
1457                         return SECCLASS_IEEE802154_SOCKET;
1458                 case PF_CAIF:
1459                         return SECCLASS_CAIF_SOCKET;
1460                 case PF_ALG:
1461                         return SECCLASS_ALG_SOCKET;
1462                 case PF_NFC:
1463                         return SECCLASS_NFC_SOCKET;
1464                 case PF_VSOCK:
1465                         return SECCLASS_VSOCK_SOCKET;
1466                 case PF_KCM:
1467                         return SECCLASS_KCM_SOCKET;
1468                 case PF_QIPCRTR:
1469                         return SECCLASS_QIPCRTR_SOCKET;
1470                 case PF_SMC:
1471                         return SECCLASS_SMC_SOCKET;
1472                 case PF_XDP:
1473                         return SECCLASS_XDP_SOCKET;
1474 #if PF_MAX > 45
1475 #error New address family defined, please update this function.
1476 #endif
1477                 }
1478         }
1479
1480         return SECCLASS_SOCKET;
1481 }
1482
1483 static int selinux_genfs_get_sid(struct dentry *dentry,
1484                                  u16 tclass,
1485                                  u16 flags,
1486                                  u32 *sid)
1487 {
1488         int rc;
1489         struct super_block *sb = dentry->d_sb;
1490         char *buffer, *path;
1491
1492         buffer = (char *)__get_free_page(GFP_KERNEL);
1493         if (!buffer)
1494                 return -ENOMEM;
1495
1496         path = dentry_path_raw(dentry, buffer, PAGE_SIZE);
1497         if (IS_ERR(path))
1498                 rc = PTR_ERR(path);
1499         else {
1500                 if (flags & SE_SBPROC) {
1501                         /* each process gets a /proc/PID/ entry. Strip off the
1502                          * PID part to get a valid selinux labeling.
1503                          * e.g. /proc/1/net/rpc/nfs -> /net/rpc/nfs */
1504                         while (path[1] >= '0' && path[1] <= '9') {
1505                                 path[1] = '/';
1506                                 path++;
1507                         }
1508                 }
1509                 rc = security_genfs_sid(&selinux_state, sb->s_type->name,
1510                                         path, tclass, sid);
1511                 if (rc == -ENOENT) {
1512                         /* No match in policy, mark as unlabeled. */
1513                         *sid = SECINITSID_UNLABELED;
1514                         rc = 0;
1515                 }
1516         }
1517         free_page((unsigned long)buffer);
1518         return rc;
1519 }
1520
1521 /* The inode's security attributes must be initialized before first use. */
1522 static int inode_doinit_with_dentry(struct inode *inode, struct dentry *opt_dentry)
1523 {
1524         struct superblock_security_struct *sbsec = NULL;
1525         struct inode_security_struct *isec = inode->i_security;
1526         u32 task_sid, sid = 0;
1527         u16 sclass;
1528         struct dentry *dentry;
1529 #define INITCONTEXTLEN 255
1530         char *context = NULL;
1531         unsigned len = 0;
1532         int rc = 0;
1533
1534         if (isec->initialized == LABEL_INITIALIZED)
1535                 return 0;
1536
1537         spin_lock(&isec->lock);
1538         if (isec->initialized == LABEL_INITIALIZED)
1539                 goto out_unlock;
1540
1541         if (isec->sclass == SECCLASS_FILE)
1542                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
1543
1544         sbsec = inode->i_sb->s_security;
1545         if (!(sbsec->flags & SE_SBINITIALIZED)) {
1546                 /* Defer initialization until selinux_complete_init,
1547                    after the initial policy is loaded and the security
1548                    server is ready to handle calls. */
1549                 spin_lock(&sbsec->isec_lock);
1550                 if (list_empty(&isec->list))
1551                         list_add(&isec->list, &sbsec->isec_head);
1552                 spin_unlock(&sbsec->isec_lock);
1553                 goto out_unlock;
1554         }
1555
1556         sclass = isec->sclass;
1557         task_sid = isec->task_sid;
1558         sid = isec->sid;
1559         isec->initialized = LABEL_PENDING;
1560         spin_unlock(&isec->lock);
1561
1562         switch (sbsec->behavior) {
1563         case SECURITY_FS_USE_NATIVE:
1564                 break;
1565         case SECURITY_FS_USE_XATTR:
1566                 if (!(inode->i_opflags & IOP_XATTR)) {
1567                         sid = sbsec->def_sid;
1568                         break;
1569                 }
1570                 /* Need a dentry, since the xattr API requires one.
1571                    Life would be simpler if we could just pass the inode. */
1572                 if (opt_dentry) {
1573                         /* Called from d_instantiate or d_splice_alias. */
1574                         dentry = dget(opt_dentry);
1575                 } else {
1576                         /*
1577                          * Called from selinux_complete_init, try to find a dentry.
1578                          * Some filesystems really want a connected one, so try
1579                          * that first.  We could split SECURITY_FS_USE_XATTR in
1580                          * two, depending upon that...
1581                          */
1582                         dentry = d_find_alias(inode);
1583                         if (!dentry)
1584                                 dentry = d_find_any_alias(inode);
1585                 }
1586                 if (!dentry) {
1587                         /*
1588                          * this is can be hit on boot when a file is accessed
1589                          * before the policy is loaded.  When we load policy we
1590                          * may find inodes that have no dentry on the
1591                          * sbsec->isec_head list.  No reason to complain as these
1592                          * will get fixed up the next time we go through
1593                          * inode_doinit with a dentry, before these inodes could
1594                          * be used again by userspace.
1595                          */
1596                         goto out;
1597                 }
1598
1599                 len = INITCONTEXTLEN;
1600                 context = kmalloc(len+1, GFP_NOFS);
1601                 if (!context) {
1602                         rc = -ENOMEM;
1603                         dput(dentry);
1604                         goto out;
1605                 }
1606                 context[len] = '\0';
1607                 rc = __vfs_getxattr(dentry, inode, XATTR_NAME_SELINUX, context, len);
1608                 if (rc == -ERANGE) {
1609                         kfree(context);
1610
1611                         /* Need a larger buffer.  Query for the right size. */
1612                         rc = __vfs_getxattr(dentry, inode, XATTR_NAME_SELINUX, NULL, 0);
1613                         if (rc < 0) {
1614                                 dput(dentry);
1615                                 goto out;
1616                         }
1617                         len = rc;
1618                         context = kmalloc(len+1, GFP_NOFS);
1619                         if (!context) {
1620                                 rc = -ENOMEM;
1621                                 dput(dentry);
1622                                 goto out;
1623                         }
1624                         context[len] = '\0';
1625                         rc = __vfs_getxattr(dentry, inode, XATTR_NAME_SELINUX, context, len);
1626                 }
1627                 dput(dentry);
1628                 if (rc < 0) {
1629                         if (rc != -ENODATA) {
1630                                 pr_warn("SELinux: %s:  getxattr returned "
1631                                        "%d for dev=%s ino=%ld\n", __func__,
1632                                        -rc, inode->i_sb->s_id, inode->i_ino);
1633                                 kfree(context);
1634                                 goto out;
1635                         }
1636                         /* Map ENODATA to the default file SID */
1637                         sid = sbsec->def_sid;
1638                         rc = 0;
1639                 } else {
1640                         rc = security_context_to_sid_default(&selinux_state,
1641                                                              context, rc, &sid,
1642                                                              sbsec->def_sid,
1643                                                              GFP_NOFS);
1644                         if (rc) {
1645                                 char *dev = inode->i_sb->s_id;
1646                                 unsigned long ino = inode->i_ino;
1647
1648                                 if (rc == -EINVAL) {
1649                                         if (printk_ratelimit())
1650                                                 pr_notice("SELinux: inode=%lu on dev=%s was found to have an invalid "
1651                                                         "context=%s.  This indicates you may need to relabel the inode or the "
1652                                                         "filesystem in question.\n", ino, dev, context);
1653                                 } else {
1654                                         pr_warn("SELinux: %s:  context_to_sid(%s) "
1655                                                "returned %d for dev=%s ino=%ld\n",
1656                                                __func__, context, -rc, dev, ino);
1657                                 }
1658                                 kfree(context);
1659                                 /* Leave with the unlabeled SID */
1660                                 rc = 0;
1661                                 break;
1662                         }
1663                 }
1664                 kfree(context);
1665                 break;
1666         case SECURITY_FS_USE_TASK:
1667                 sid = task_sid;
1668                 break;
1669         case SECURITY_FS_USE_TRANS:
1670                 /* Default to the fs SID. */
1671                 sid = sbsec->sid;
1672
1673                 /* Try to obtain a transition SID. */
1674                 rc = security_transition_sid(&selinux_state, task_sid, sid,
1675                                              sclass, NULL, &sid);
1676                 if (rc)
1677                         goto out;
1678                 break;
1679         case SECURITY_FS_USE_MNTPOINT:
1680                 sid = sbsec->mntpoint_sid;
1681                 break;
1682         default:
1683                 /* Default to the fs superblock SID. */
1684                 sid = sbsec->sid;
1685
1686                 if ((sbsec->flags & SE_SBGENFS) && !S_ISLNK(inode->i_mode)) {
1687                         /* We must have a dentry to determine the label on
1688                          * procfs inodes */
1689                         if (opt_dentry) {
1690                                 /* Called from d_instantiate or
1691                                  * d_splice_alias. */
1692                                 dentry = dget(opt_dentry);
1693                         } else {
1694                                 /* Called from selinux_complete_init, try to
1695                                  * find a dentry.  Some filesystems really want
1696                                  * a connected one, so try that first.
1697                                  */
1698                                 dentry = d_find_alias(inode);
1699                                 if (!dentry)
1700                                         dentry = d_find_any_alias(inode);
1701                         }
1702                         /*
1703                          * This can be hit on boot when a file is accessed
1704                          * before the policy is loaded.  When we load policy we
1705                          * may find inodes that have no dentry on the
1706                          * sbsec->isec_head list.  No reason to complain as
1707                          * these will get fixed up the next time we go through
1708                          * inode_doinit() with a dentry, before these inodes
1709                          * could be used again by userspace.
1710                          */
1711                         if (!dentry)
1712                                 goto out;
1713                         rc = selinux_genfs_get_sid(dentry, sclass,
1714                                                    sbsec->flags, &sid);
1715                         dput(dentry);
1716                         if (rc)
1717                                 goto out;
1718                 }
1719                 break;
1720         }
1721
1722 out:
1723         spin_lock(&isec->lock);
1724         if (isec->initialized == LABEL_PENDING) {
1725                 if (!sid || rc) {
1726                         isec->initialized = LABEL_INVALID;
1727                         goto out_unlock;
1728                 }
1729
1730                 isec->initialized = LABEL_INITIALIZED;
1731                 isec->sid = sid;
1732         }
1733
1734 out_unlock:
1735         spin_unlock(&isec->lock);
1736         return rc;
1737 }
1738
1739 /* Convert a Linux signal to an access vector. */
1740 static inline u32 signal_to_av(int sig)
1741 {
1742         u32 perm = 0;
1743
1744         switch (sig) {
1745         case SIGCHLD:
1746                 /* Commonly granted from child to parent. */
1747                 perm = PROCESS__SIGCHLD;
1748                 break;
1749         case SIGKILL:
1750                 /* Cannot be caught or ignored */
1751                 perm = PROCESS__SIGKILL;
1752                 break;
1753         case SIGSTOP:
1754                 /* Cannot be caught or ignored */
1755                 perm = PROCESS__SIGSTOP;
1756                 break;
1757         default:
1758                 /* All other signals. */
1759                 perm = PROCESS__SIGNAL;
1760                 break;
1761         }
1762
1763         return perm;
1764 }
1765
1766 #if CAP_LAST_CAP > 63
1767 #error Fix SELinux to handle capabilities > 63.
1768 #endif
1769
1770 /* Check whether a task is allowed to use a capability. */
1771 static int cred_has_capability(const struct cred *cred,
1772                                int cap, int audit, bool initns)
1773 {
1774         struct common_audit_data ad;
1775         struct av_decision avd;
1776         u16 sclass;
1777         u32 sid = cred_sid(cred);
1778         u32 av = CAP_TO_MASK(cap);
1779         int rc;
1780
1781         ad.type = LSM_AUDIT_DATA_CAP;
1782         ad.u.cap = cap;
1783
1784         switch (CAP_TO_INDEX(cap)) {
1785         case 0:
1786                 sclass = initns ? SECCLASS_CAPABILITY : SECCLASS_CAP_USERNS;
1787                 break;
1788         case 1:
1789                 sclass = initns ? SECCLASS_CAPABILITY2 : SECCLASS_CAP2_USERNS;
1790                 break;
1791         default:
1792                 pr_err("SELinux:  out of range capability %d\n", cap);
1793                 BUG();
1794                 return -EINVAL;
1795         }
1796
1797         rc = avc_has_perm_noaudit(&selinux_state,
1798                                   sid, sid, sclass, av, 0, &avd);
1799         if (audit == SECURITY_CAP_AUDIT) {
1800                 int rc2 = avc_audit(&selinux_state,
1801                                     sid, sid, sclass, av, &avd, rc, &ad, 0);
1802                 if (rc2)
1803                         return rc2;
1804         }
1805         return rc;
1806 }
1807
1808 /* Check whether a task has a particular permission to an inode.
1809    The 'adp' parameter is optional and allows other audit
1810    data to be passed (e.g. the dentry). */
1811 static int inode_has_perm(const struct cred *cred,
1812                           struct inode *inode,
1813                           u32 perms,
1814                           struct common_audit_data *adp)
1815 {
1816         struct inode_security_struct *isec;
1817         u32 sid;
1818
1819         validate_creds(cred);
1820
1821         if (unlikely(IS_PRIVATE(inode)))
1822                 return 0;
1823
1824         sid = cred_sid(cred);
1825         isec = inode->i_security;
1826
1827         return avc_has_perm(&selinux_state,
1828                             sid, isec->sid, isec->sclass, perms, adp);
1829 }
1830
1831 /* Same as inode_has_perm, but pass explicit audit data containing
1832    the dentry to help the auditing code to more easily generate the
1833    pathname if needed. */
1834 static inline int dentry_has_perm(const struct cred *cred,
1835                                   struct dentry *dentry,
1836                                   u32 av)
1837 {
1838         struct inode *inode = d_backing_inode(dentry);
1839         struct common_audit_data ad;
1840
1841         ad.type = LSM_AUDIT_DATA_DENTRY;
1842         ad.u.dentry = dentry;
1843         __inode_security_revalidate(inode, dentry, true);
1844         return inode_has_perm(cred, inode, av, &ad);
1845 }
1846
1847 /* Same as inode_has_perm, but pass explicit audit data containing
1848    the path to help the auditing code to more easily generate the
1849    pathname if needed. */
1850 static inline int path_has_perm(const struct cred *cred,
1851                                 const struct path *path,
1852                                 u32 av)
1853 {
1854         struct inode *inode = d_backing_inode(path->dentry);
1855         struct common_audit_data ad;
1856
1857         ad.type = LSM_AUDIT_DATA_PATH;
1858         ad.u.path = *path;
1859         __inode_security_revalidate(inode, path->dentry, true);
1860         return inode_has_perm(cred, inode, av, &ad);
1861 }
1862
1863 /* Same as path_has_perm, but uses the inode from the file struct. */
1864 static inline int file_path_has_perm(const struct cred *cred,
1865                                      struct file *file,
1866                                      u32 av)
1867 {
1868         struct common_audit_data ad;
1869
1870         ad.type = LSM_AUDIT_DATA_FILE;
1871         ad.u.file = file;
1872         return inode_has_perm(cred, file_inode(file), av, &ad);
1873 }
1874
1875 #ifdef CONFIG_BPF_SYSCALL
1876 static int bpf_fd_pass(struct file *file, u32 sid);
1877 #endif
1878
1879 /* Check whether a task can use an open file descriptor to
1880    access an inode in a given way.  Check access to the
1881    descriptor itself, and then use dentry_has_perm to
1882    check a particular permission to the file.
1883    Access to the descriptor is implicitly granted if it
1884    has the same SID as the process.  If av is zero, then
1885    access to the file is not checked, e.g. for cases
1886    where only the descriptor is affected like seek. */
1887 static int file_has_perm(const struct cred *cred,
1888                          struct file *file,
1889                          u32 av)
1890 {
1891         struct file_security_struct *fsec = file->f_security;
1892         struct inode *inode = file_inode(file);
1893         struct common_audit_data ad;
1894         u32 sid = cred_sid(cred);
1895         int rc;
1896
1897         ad.type = LSM_AUDIT_DATA_FILE;
1898         ad.u.file = file;
1899
1900         if (sid != fsec->sid) {
1901                 rc = avc_has_perm(&selinux_state,
1902                                   sid, fsec->sid,
1903                                   SECCLASS_FD,
1904                                   FD__USE,
1905                                   &ad);
1906                 if (rc)
1907                         goto out;
1908         }
1909
1910 #ifdef CONFIG_BPF_SYSCALL
1911         rc = bpf_fd_pass(file, cred_sid(cred));
1912         if (rc)
1913                 return rc;
1914 #endif
1915
1916         /* av is zero if only checking access to the descriptor. */
1917         rc = 0;
1918         if (av)
1919                 rc = inode_has_perm(cred, inode, av, &ad);
1920
1921 out:
1922         return rc;
1923 }
1924
1925 /*
1926  * Determine the label for an inode that might be unioned.
1927  */
1928 static int
1929 selinux_determine_inode_label(const struct task_security_struct *tsec,
1930                                  struct inode *dir,
1931                                  const struct qstr *name, u16 tclass,
1932                                  u32 *_new_isid)
1933 {
1934         const struct superblock_security_struct *sbsec = dir->i_sb->s_security;
1935
1936         if ((sbsec->flags & SE_SBINITIALIZED) &&
1937             (sbsec->behavior == SECURITY_FS_USE_MNTPOINT)) {
1938                 *_new_isid = sbsec->mntpoint_sid;
1939         } else if ((sbsec->flags & SBLABEL_MNT) &&
1940                    tsec->create_sid) {
1941                 *_new_isid = tsec->create_sid;
1942         } else {
1943                 const struct inode_security_struct *dsec = inode_security(dir);
1944                 return security_transition_sid(&selinux_state, tsec->sid,
1945                                                dsec->sid, tclass,
1946                                                name, _new_isid);
1947         }
1948
1949         return 0;
1950 }
1951
1952 /* Check whether a task can create a file. */
1953 static int may_create(struct inode *dir,
1954                       struct dentry *dentry,
1955                       u16 tclass)
1956 {
1957         const struct task_security_struct *tsec = current_security();
1958         struct inode_security_struct *dsec;
1959         struct superblock_security_struct *sbsec;
1960         u32 sid, newsid;
1961         struct common_audit_data ad;
1962         int rc;
1963
1964         dsec = inode_security(dir);
1965         sbsec = dir->i_sb->s_security;
1966
1967         sid = tsec->sid;
1968
1969         ad.type = LSM_AUDIT_DATA_DENTRY;
1970         ad.u.dentry = dentry;
1971
1972         rc = avc_has_perm(&selinux_state,
1973                           sid, dsec->sid, SECCLASS_DIR,
1974                           DIR__ADD_NAME | DIR__SEARCH,
1975                           &ad);
1976         if (rc)
1977                 return rc;
1978
1979         rc = selinux_determine_inode_label(current_security(), dir,
1980                                            &dentry->d_name, tclass, &newsid);
1981         if (rc)
1982                 return rc;
1983
1984         rc = avc_has_perm(&selinux_state,
1985                           sid, newsid, tclass, FILE__CREATE, &ad);
1986         if (rc)
1987                 return rc;
1988
1989         return avc_has_perm(&selinux_state,
1990                             newsid, sbsec->sid,
1991                             SECCLASS_FILESYSTEM,
1992                             FILESYSTEM__ASSOCIATE, &ad);
1993 }
1994
1995 #define MAY_LINK        0
1996 #define MAY_UNLINK      1
1997 #define MAY_RMDIR       2
1998
1999 /* Check whether a task can link, unlink, or rmdir a file/directory. */
2000 static int may_link(struct inode *dir,
2001                     struct dentry *dentry,
2002                     int kind)
2003
2004 {
2005         struct inode_security_struct *dsec, *isec;
2006         struct common_audit_data ad;
2007         u32 sid = current_sid();
2008         u32 av;
2009         int rc;
2010
2011         dsec = inode_security(dir);
2012         isec = backing_inode_security(dentry);
2013
2014         ad.type = LSM_AUDIT_DATA_DENTRY;
2015         ad.u.dentry = dentry;
2016
2017         av = DIR__SEARCH;
2018         av |= (kind ? DIR__REMOVE_NAME : DIR__ADD_NAME);
2019         rc = avc_has_perm(&selinux_state,
2020                           sid, dsec->sid, SECCLASS_DIR, av, &ad);
2021         if (rc)
2022                 return rc;
2023
2024         switch (kind) {
2025         case MAY_LINK:
2026                 av = FILE__LINK;
2027                 break;
2028         case MAY_UNLINK:
2029                 av = FILE__UNLINK;
2030                 break;
2031         case MAY_RMDIR:
2032                 av = DIR__RMDIR;
2033                 break;
2034         default:
2035                 pr_warn("SELinux: %s:  unrecognized kind %d\n",
2036                         __func__, kind);
2037                 return 0;
2038         }
2039
2040         rc = avc_has_perm(&selinux_state,
2041                           sid, isec->sid, isec->sclass, av, &ad);
2042         return rc;
2043 }
2044
2045 static inline int may_rename(struct inode *old_dir,
2046                              struct dentry *old_dentry,
2047                              struct inode *new_dir,
2048                              struct dentry *new_dentry)
2049 {
2050         struct inode_security_struct *old_dsec, *new_dsec, *old_isec, *new_isec;
2051         struct common_audit_data ad;
2052         u32 sid = current_sid();
2053         u32 av;
2054         int old_is_dir, new_is_dir;
2055         int rc;
2056
2057         old_dsec = inode_security(old_dir);
2058         old_isec = backing_inode_security(old_dentry);
2059         old_is_dir = d_is_dir(old_dentry);
2060         new_dsec = inode_security(new_dir);
2061
2062         ad.type = LSM_AUDIT_DATA_DENTRY;
2063
2064         ad.u.dentry = old_dentry;
2065         rc = avc_has_perm(&selinux_state,
2066                           sid, old_dsec->sid, SECCLASS_DIR,
2067                           DIR__REMOVE_NAME | DIR__SEARCH, &ad);
2068         if (rc)
2069                 return rc;
2070         rc = avc_has_perm(&selinux_state,
2071                           sid, old_isec->sid,
2072                           old_isec->sclass, FILE__RENAME, &ad);
2073         if (rc)
2074                 return rc;
2075         if (old_is_dir && new_dir != old_dir) {
2076                 rc = avc_has_perm(&selinux_state,
2077                                   sid, old_isec->sid,
2078                                   old_isec->sclass, DIR__REPARENT, &ad);
2079                 if (rc)
2080                         return rc;
2081         }
2082
2083         ad.u.dentry = new_dentry;
2084         av = DIR__ADD_NAME | DIR__SEARCH;
2085         if (d_is_positive(new_dentry))
2086                 av |= DIR__REMOVE_NAME;
2087         rc = avc_has_perm(&selinux_state,
2088                           sid, new_dsec->sid, SECCLASS_DIR, av, &ad);
2089         if (rc)
2090                 return rc;
2091         if (d_is_positive(new_dentry)) {
2092                 new_isec = backing_inode_security(new_dentry);
2093                 new_is_dir = d_is_dir(new_dentry);
2094                 rc = avc_has_perm(&selinux_state,
2095                                   sid, new_isec->sid,
2096                                   new_isec->sclass,
2097                                   (new_is_dir ? DIR__RMDIR : FILE__UNLINK), &ad);
2098                 if (rc)
2099                         return rc;
2100         }
2101
2102         return 0;
2103 }
2104
2105 /* Check whether a task can perform a filesystem operation. */
2106 static int superblock_has_perm(const struct cred *cred,
2107                                struct super_block *sb,
2108                                u32 perms,
2109                                struct common_audit_data *ad)
2110 {
2111         struct superblock_security_struct *sbsec;
2112         u32 sid = cred_sid(cred);
2113
2114         sbsec = sb->s_security;
2115         return avc_has_perm(&selinux_state,
2116                             sid, sbsec->sid, SECCLASS_FILESYSTEM, perms, ad);
2117 }
2118
2119 /* Convert a Linux mode and permission mask to an access vector. */
2120 static inline u32 file_mask_to_av(int mode, int mask)
2121 {
2122         u32 av = 0;
2123
2124         if (!S_ISDIR(mode)) {
2125                 if (mask & MAY_EXEC)
2126                         av |= FILE__EXECUTE;
2127                 if (mask & MAY_READ)
2128                         av |= FILE__READ;
2129
2130                 if (mask & MAY_APPEND)
2131                         av |= FILE__APPEND;
2132                 else if (mask & MAY_WRITE)
2133                         av |= FILE__WRITE;
2134
2135         } else {
2136                 if (mask & MAY_EXEC)
2137                         av |= DIR__SEARCH;
2138                 if (mask & MAY_WRITE)
2139                         av |= DIR__WRITE;
2140                 if (mask & MAY_READ)
2141                         av |= DIR__READ;
2142         }
2143
2144         return av;
2145 }
2146
2147 /* Convert a Linux file to an access vector. */
2148 static inline u32 file_to_av(struct file *file)
2149 {
2150         u32 av = 0;
2151
2152         if (file->f_mode & FMODE_READ)
2153                 av |= FILE__READ;
2154         if (file->f_mode & FMODE_WRITE) {
2155                 if (file->f_flags & O_APPEND)
2156                         av |= FILE__APPEND;
2157                 else
2158                         av |= FILE__WRITE;
2159         }
2160         if (!av) {
2161                 /*
2162                  * Special file opened with flags 3 for ioctl-only use.
2163                  */
2164                 av = FILE__IOCTL;
2165         }
2166
2167         return av;
2168 }
2169
2170 /*
2171  * Convert a file to an access vector and include the correct open
2172  * open permission.
2173  */
2174 static inline u32 open_file_to_av(struct file *file)
2175 {
2176         u32 av = file_to_av(file);
2177         struct inode *inode = file_inode(file);
2178
2179         if (selinux_policycap_openperm() &&
2180             inode->i_sb->s_magic != SOCKFS_MAGIC)
2181                 av |= FILE__OPEN;
2182
2183         return av;
2184 }
2185
2186 /* Hook functions begin here. */
2187
2188 static int selinux_binder_set_context_mgr(struct task_struct *mgr)
2189 {
2190         u32 mysid = current_sid();
2191         u32 mgrsid = task_sid(mgr);
2192
2193         return avc_has_perm(&selinux_state,
2194                             mysid, mgrsid, SECCLASS_BINDER,
2195                             BINDER__SET_CONTEXT_MGR, NULL);
2196 }
2197
2198 static int selinux_binder_transaction(struct task_struct *from,
2199                                       struct task_struct *to)
2200 {
2201         u32 mysid = current_sid();
2202         u32 fromsid = task_sid(from);
2203         u32 tosid = task_sid(to);
2204         int rc;
2205
2206         if (mysid != fromsid) {
2207                 rc = avc_has_perm(&selinux_state,
2208                                   mysid, fromsid, SECCLASS_BINDER,
2209                                   BINDER__IMPERSONATE, NULL);
2210                 if (rc)
2211                         return rc;
2212         }
2213
2214         return avc_has_perm(&selinux_state,
2215                             fromsid, tosid, SECCLASS_BINDER, BINDER__CALL,
2216                             NULL);
2217 }
2218
2219 static int selinux_binder_transfer_binder(struct task_struct *from,
2220                                           struct task_struct *to)
2221 {
2222         u32 fromsid = task_sid(from);
2223         u32 tosid = task_sid(to);
2224
2225         return avc_has_perm(&selinux_state,
2226                             fromsid, tosid, SECCLASS_BINDER, BINDER__TRANSFER,
2227                             NULL);
2228 }
2229
2230 static int selinux_binder_transfer_file(struct task_struct *from,
2231                                         struct task_struct *to,
2232                                         struct file *file)
2233 {
2234         u32 sid = task_sid(to);
2235         struct file_security_struct *fsec = file->f_security;
2236         struct dentry *dentry = file->f_path.dentry;
2237         struct inode_security_struct *isec;
2238         struct common_audit_data ad;
2239         int rc;
2240
2241         ad.type = LSM_AUDIT_DATA_PATH;
2242         ad.u.path = file->f_path;
2243
2244         if (sid != fsec->sid) {
2245                 rc = avc_has_perm(&selinux_state,
2246                                   sid, fsec->sid,
2247                                   SECCLASS_FD,
2248                                   FD__USE,
2249                                   &ad);
2250                 if (rc)
2251                         return rc;
2252         }
2253
2254 #ifdef CONFIG_BPF_SYSCALL
2255         rc = bpf_fd_pass(file, sid);
2256         if (rc)
2257                 return rc;
2258 #endif
2259
2260         if (unlikely(IS_PRIVATE(d_backing_inode(dentry))))
2261                 return 0;
2262
2263         isec = backing_inode_security(dentry);
2264         return avc_has_perm(&selinux_state,
2265                             sid, isec->sid, isec->sclass, file_to_av(file),
2266                             &ad);
2267 }
2268
2269 static int selinux_ptrace_access_check(struct task_struct *child,
2270                                      unsigned int mode)
2271 {
2272         u32 sid = current_sid();
2273         u32 csid = task_sid(child);
2274
2275         if (mode & PTRACE_MODE_READ)
2276                 return avc_has_perm(&selinux_state,
2277                                     sid, csid, SECCLASS_FILE, FILE__READ, NULL);
2278
2279         return avc_has_perm(&selinux_state,
2280                             sid, csid, SECCLASS_PROCESS, PROCESS__PTRACE, NULL);
2281 }
2282
2283 static int selinux_ptrace_traceme(struct task_struct *parent)
2284 {
2285         return avc_has_perm(&selinux_state,
2286                             task_sid(parent), current_sid(), SECCLASS_PROCESS,
2287                             PROCESS__PTRACE, NULL);
2288 }
2289
2290 static int selinux_capget(struct task_struct *target, kernel_cap_t *effective,
2291                           kernel_cap_t *inheritable, kernel_cap_t *permitted)
2292 {
2293         return avc_has_perm(&selinux_state,
2294                             current_sid(), task_sid(target), SECCLASS_PROCESS,
2295                             PROCESS__GETCAP, NULL);
2296 }
2297
2298 static int selinux_capset(struct cred *new, const struct cred *old,
2299                           const kernel_cap_t *effective,
2300                           const kernel_cap_t *inheritable,
2301                           const kernel_cap_t *permitted)
2302 {
2303         return avc_has_perm(&selinux_state,
2304                             cred_sid(old), cred_sid(new), SECCLASS_PROCESS,
2305                             PROCESS__SETCAP, NULL);
2306 }
2307
2308 /*
2309  * (This comment used to live with the selinux_task_setuid hook,
2310  * which was removed).
2311  *
2312  * Since setuid only affects the current process, and since the SELinux
2313  * controls are not based on the Linux identity attributes, SELinux does not
2314  * need to control this operation.  However, SELinux does control the use of
2315  * the CAP_SETUID and CAP_SETGID capabilities using the capable hook.
2316  */
2317
2318 static int selinux_capable(const struct cred *cred, struct user_namespace *ns,
2319                            int cap, int audit)
2320 {
2321         return cred_has_capability(cred, cap, audit, ns == &init_user_ns);
2322 }
2323
2324 static int selinux_quotactl(int cmds, int type, int id, struct super_block *sb)
2325 {
2326         const struct cred *cred = current_cred();
2327         int rc = 0;
2328
2329         if (!sb)
2330                 return 0;
2331
2332         switch (cmds) {
2333         case Q_SYNC:
2334         case Q_QUOTAON:
2335         case Q_QUOTAOFF:
2336         case Q_SETINFO:
2337         case Q_SETQUOTA:
2338                 rc = superblock_has_perm(cred, sb, FILESYSTEM__QUOTAMOD, NULL);
2339                 break;
2340         case Q_GETFMT:
2341         case Q_GETINFO:
2342         case Q_GETQUOTA:
2343                 rc = superblock_has_perm(cred, sb, FILESYSTEM__QUOTAGET, NULL);
2344                 break;
2345         default:
2346                 rc = 0;  /* let the kernel handle invalid cmds */
2347                 break;
2348         }
2349         return rc;
2350 }
2351
2352 static int selinux_quota_on(struct dentry *dentry)
2353 {
2354         const struct cred *cred = current_cred();
2355
2356         return dentry_has_perm(cred, dentry, FILE__QUOTAON);
2357 }
2358
2359 static int selinux_syslog(int type)
2360 {
2361         switch (type) {
2362         case SYSLOG_ACTION_READ_ALL:    /* Read last kernel messages */
2363         case SYSLOG_ACTION_SIZE_BUFFER: /* Return size of the log buffer */
2364                 return avc_has_perm(&selinux_state,
2365                                     current_sid(), SECINITSID_KERNEL,
2366                                     SECCLASS_SYSTEM, SYSTEM__SYSLOG_READ, NULL);
2367         case SYSLOG_ACTION_CONSOLE_OFF: /* Disable logging to console */
2368         case SYSLOG_ACTION_CONSOLE_ON:  /* Enable logging to console */
2369         /* Set level of messages printed to console */
2370         case SYSLOG_ACTION_CONSOLE_LEVEL:
2371                 return avc_has_perm(&selinux_state,
2372                                     current_sid(), SECINITSID_KERNEL,
2373                                     SECCLASS_SYSTEM, SYSTEM__SYSLOG_CONSOLE,
2374                                     NULL);
2375         }
2376         /* All other syslog types */
2377         return avc_has_perm(&selinux_state,
2378                             current_sid(), SECINITSID_KERNEL,
2379                             SECCLASS_SYSTEM, SYSTEM__SYSLOG_MOD, NULL);
2380 }
2381
2382 /*
2383  * Check that a process has enough memory to allocate a new virtual
2384  * mapping. 0 means there is enough memory for the allocation to
2385  * succeed and -ENOMEM implies there is not.
2386  *
2387  * Do not audit the selinux permission check, as this is applied to all
2388  * processes that allocate mappings.
2389  */
2390 static int selinux_vm_enough_memory(struct mm_struct *mm, long pages)
2391 {
2392         int rc, cap_sys_admin = 0;
2393
2394         rc = cred_has_capability(current_cred(), CAP_SYS_ADMIN,
2395                                  SECURITY_CAP_NOAUDIT, true);
2396         if (rc == 0)
2397                 cap_sys_admin = 1;
2398
2399         return cap_sys_admin;
2400 }
2401
2402 /* binprm security operations */
2403
2404 static u32 ptrace_parent_sid(void)
2405 {
2406         u32 sid = 0;
2407         struct task_struct *tracer;
2408
2409         rcu_read_lock();
2410         tracer = ptrace_parent(current);
2411         if (tracer)
2412                 sid = task_sid(tracer);
2413         rcu_read_unlock();
2414
2415         return sid;
2416 }
2417
2418 static int check_nnp_nosuid(const struct linux_binprm *bprm,
2419                             const struct task_security_struct *old_tsec,
2420                             const struct task_security_struct *new_tsec)
2421 {
2422         int nnp = (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS);
2423         int nosuid = !mnt_may_suid(bprm->file->f_path.mnt);
2424         int rc;
2425         u32 av;
2426
2427         if (!nnp && !nosuid)
2428                 return 0; /* neither NNP nor nosuid */
2429
2430         if (new_tsec->sid == old_tsec->sid)
2431                 return 0; /* No change in credentials */
2432
2433         /*
2434          * If the policy enables the nnp_nosuid_transition policy capability,
2435          * then we permit transitions under NNP or nosuid if the
2436          * policy allows the corresponding permission between
2437          * the old and new contexts.
2438          */
2439         if (selinux_policycap_nnp_nosuid_transition()) {
2440                 av = 0;
2441                 if (nnp)
2442                         av |= PROCESS2__NNP_TRANSITION;
2443                 if (nosuid)
2444                         av |= PROCESS2__NOSUID_TRANSITION;
2445                 rc = avc_has_perm(&selinux_state,
2446                                   old_tsec->sid, new_tsec->sid,
2447                                   SECCLASS_PROCESS2, av, NULL);
2448                 if (!rc)
2449                         return 0;
2450         }
2451
2452         /*
2453          * We also permit NNP or nosuid transitions to bounded SIDs,
2454          * i.e. SIDs that are guaranteed to only be allowed a subset
2455          * of the permissions of the current SID.
2456          */
2457         rc = security_bounded_transition(&selinux_state, old_tsec->sid,
2458                                          new_tsec->sid);
2459         if (!rc)
2460                 return 0;
2461
2462         /*
2463          * On failure, preserve the errno values for NNP vs nosuid.
2464          * NNP:  Operation not permitted for caller.
2465          * nosuid:  Permission denied to file.
2466          */
2467         if (nnp)
2468                 return -EPERM;
2469         return -EACCES;
2470 }
2471
2472 static int selinux_bprm_set_creds(struct linux_binprm *bprm)
2473 {
2474         const struct task_security_struct *old_tsec;
2475         struct task_security_struct *new_tsec;
2476         struct inode_security_struct *isec;
2477         struct common_audit_data ad;
2478         struct inode *inode = file_inode(bprm->file);
2479         int rc;
2480
2481         /* SELinux context only depends on initial program or script and not
2482          * the script interpreter */
2483         if (bprm->called_set_creds)
2484                 return 0;
2485
2486         old_tsec = current_security();
2487         new_tsec = bprm->cred->security;
2488         isec = inode_security(inode);
2489
2490         /* Default to the current task SID. */
2491         new_tsec->sid = old_tsec->sid;
2492         new_tsec->osid = old_tsec->sid;
2493
2494         /* Reset fs, key, and sock SIDs on execve. */
2495         new_tsec->create_sid = 0;
2496         new_tsec->keycreate_sid = 0;
2497         new_tsec->sockcreate_sid = 0;
2498
2499         if (old_tsec->exec_sid) {
2500                 new_tsec->sid = old_tsec->exec_sid;
2501                 /* Reset exec SID on execve. */
2502                 new_tsec->exec_sid = 0;
2503
2504                 /* Fail on NNP or nosuid if not an allowed transition. */
2505                 rc = check_nnp_nosuid(bprm, old_tsec, new_tsec);
2506                 if (rc)
2507                         return rc;
2508         } else {
2509                 /* Check for a default transition on this program. */
2510                 rc = security_transition_sid(&selinux_state, old_tsec->sid,
2511                                              isec->sid, SECCLASS_PROCESS, NULL,
2512                                              &new_tsec->sid);
2513                 if (rc)
2514                         return rc;
2515
2516                 /*
2517                  * Fallback to old SID on NNP or nosuid if not an allowed
2518                  * transition.
2519                  */
2520                 rc = check_nnp_nosuid(bprm, old_tsec, new_tsec);
2521                 if (rc)
2522                         new_tsec->sid = old_tsec->sid;
2523         }
2524
2525         ad.type = LSM_AUDIT_DATA_FILE;
2526         ad.u.file = bprm->file;
2527
2528         if (new_tsec->sid == old_tsec->sid) {
2529                 rc = avc_has_perm(&selinux_state,
2530                                   old_tsec->sid, isec->sid,
2531                                   SECCLASS_FILE, FILE__EXECUTE_NO_TRANS, &ad);
2532                 if (rc)
2533                         return rc;
2534         } else {
2535                 /* Check permissions for the transition. */
2536                 rc = avc_has_perm(&selinux_state,
2537                                   old_tsec->sid, new_tsec->sid,
2538                                   SECCLASS_PROCESS, PROCESS__TRANSITION, &ad);
2539                 if (rc)
2540                         return rc;
2541
2542                 rc = avc_has_perm(&selinux_state,
2543                                   new_tsec->sid, isec->sid,
2544                                   SECCLASS_FILE, FILE__ENTRYPOINT, &ad);
2545                 if (rc)
2546                         return rc;
2547
2548                 /* Check for shared state */
2549                 if (bprm->unsafe & LSM_UNSAFE_SHARE) {
2550                         rc = avc_has_perm(&selinux_state,
2551                                           old_tsec->sid, new_tsec->sid,
2552                                           SECCLASS_PROCESS, PROCESS__SHARE,
2553                                           NULL);
2554                         if (rc)
2555                                 return -EPERM;
2556                 }
2557
2558                 /* Make sure that anyone attempting to ptrace over a task that
2559                  * changes its SID has the appropriate permit */
2560                 if (bprm->unsafe & LSM_UNSAFE_PTRACE) {
2561                         u32 ptsid = ptrace_parent_sid();
2562                         if (ptsid != 0) {
2563                                 rc = avc_has_perm(&selinux_state,
2564                                                   ptsid, new_tsec->sid,
2565                                                   SECCLASS_PROCESS,
2566                                                   PROCESS__PTRACE, NULL);
2567                                 if (rc)
2568                                         return -EPERM;
2569                         }
2570                 }
2571
2572                 /* Clear any possibly unsafe personality bits on exec: */
2573                 bprm->per_clear |= PER_CLEAR_ON_SETID;
2574
2575                 /* Enable secure mode for SIDs transitions unless
2576                    the noatsecure permission is granted between
2577                    the two SIDs, i.e. ahp returns 0. */
2578                 rc = avc_has_perm(&selinux_state,
2579                                   old_tsec->sid, new_tsec->sid,
2580                                   SECCLASS_PROCESS, PROCESS__NOATSECURE,
2581                                   NULL);
2582                 bprm->secureexec |= !!rc;
2583         }
2584
2585         return 0;
2586 }
2587
2588 static int match_file(const void *p, struct file *file, unsigned fd)
2589 {
2590         return file_has_perm(p, file, file_to_av(file)) ? fd + 1 : 0;
2591 }
2592
2593 /* Derived from fs/exec.c:flush_old_files. */
2594 static inline void flush_unauthorized_files(const struct cred *cred,
2595                                             struct files_struct *files)
2596 {
2597         struct file *file, *devnull = NULL;
2598         struct tty_struct *tty;
2599         int drop_tty = 0;
2600         unsigned n;
2601
2602         tty = get_current_tty();
2603         if (tty) {
2604                 spin_lock(&tty->files_lock);
2605                 if (!list_empty(&tty->tty_files)) {
2606                         struct tty_file_private *file_priv;
2607
2608                         /* Revalidate access to controlling tty.
2609                            Use file_path_has_perm on the tty path directly
2610                            rather than using file_has_perm, as this particular
2611                            open file may belong to another process and we are
2612                            only interested in the inode-based check here. */
2613                         file_priv = list_first_entry(&tty->tty_files,
2614                                                 struct tty_file_private, list);
2615                         file = file_priv->file;
2616                         if (file_path_has_perm(cred, file, FILE__READ | FILE__WRITE))
2617                                 drop_tty = 1;
2618                 }
2619                 spin_unlock(&tty->files_lock);
2620                 tty_kref_put(tty);
2621         }
2622         /* Reset controlling tty. */
2623         if (drop_tty)
2624                 no_tty();
2625
2626         /* Revalidate access to inherited open files. */
2627         n = iterate_fd(files, 0, match_file, cred);
2628         if (!n) /* none found? */
2629                 return;
2630
2631         devnull = dentry_open(&selinux_null, O_RDWR, cred);
2632         if (IS_ERR(devnull))
2633                 devnull = NULL;
2634         /* replace all the matching ones with this */
2635         do {
2636                 replace_fd(n - 1, devnull, 0);
2637         } while ((n = iterate_fd(files, n, match_file, cred)) != 0);
2638         if (devnull)
2639                 fput(devnull);
2640 }
2641
2642 /*
2643  * Prepare a process for imminent new credential changes due to exec
2644  */
2645 static void selinux_bprm_committing_creds(struct linux_binprm *bprm)
2646 {
2647         struct task_security_struct *new_tsec;
2648         struct rlimit *rlim, *initrlim;
2649         int rc, i;
2650
2651         new_tsec = bprm->cred->security;
2652         if (new_tsec->sid == new_tsec->osid)
2653                 return;
2654
2655         /* Close files for which the new task SID is not authorized. */
2656         flush_unauthorized_files(bprm->cred, current->files);
2657
2658         /* Always clear parent death signal on SID transitions. */
2659         current->pdeath_signal = 0;
2660
2661         /* Check whether the new SID can inherit resource limits from the old
2662          * SID.  If not, reset all soft limits to the lower of the current
2663          * task's hard limit and the init task's soft limit.
2664          *
2665          * Note that the setting of hard limits (even to lower them) can be
2666          * controlled by the setrlimit check.  The inclusion of the init task's
2667          * soft limit into the computation is to avoid resetting soft limits
2668          * higher than the default soft limit for cases where the default is
2669          * lower than the hard limit, e.g. RLIMIT_CORE or RLIMIT_STACK.
2670          */
2671         rc = avc_has_perm(&selinux_state,
2672                           new_tsec->osid, new_tsec->sid, SECCLASS_PROCESS,
2673                           PROCESS__RLIMITINH, NULL);
2674         if (rc) {
2675                 /* protect against do_prlimit() */
2676                 task_lock(current);
2677                 for (i = 0; i < RLIM_NLIMITS; i++) {
2678                         rlim = current->signal->rlim + i;
2679                         initrlim = init_task.signal->rlim + i;
2680                         rlim->rlim_cur = min(rlim->rlim_max, initrlim->rlim_cur);
2681                 }
2682                 task_unlock(current);
2683                 if (IS_ENABLED(CONFIG_POSIX_TIMERS))
2684                         update_rlimit_cpu(current, rlimit(RLIMIT_CPU));
2685         }
2686 }
2687
2688 /*
2689  * Clean up the process immediately after the installation of new credentials
2690  * due to exec
2691  */
2692 static void selinux_bprm_committed_creds(struct linux_binprm *bprm)
2693 {
2694         const struct task_security_struct *tsec = current_security();
2695         struct itimerval itimer;
2696         u32 osid, sid;
2697         int rc, i;
2698
2699         osid = tsec->osid;
2700         sid = tsec->sid;
2701
2702         if (sid == osid)
2703                 return;
2704
2705         /* Check whether the new SID can inherit signal state from the old SID.
2706          * If not, clear itimers to avoid subsequent signal generation and
2707          * flush and unblock signals.
2708          *
2709          * This must occur _after_ the task SID has been updated so that any
2710          * kill done after the flush will be checked against the new SID.
2711          */
2712         rc = avc_has_perm(&selinux_state,
2713                           osid, sid, SECCLASS_PROCESS, PROCESS__SIGINH, NULL);
2714         if (rc) {
2715                 if (IS_ENABLED(CONFIG_POSIX_TIMERS)) {
2716                         memset(&itimer, 0, sizeof itimer);
2717                         for (i = 0; i < 3; i++)
2718                                 do_setitimer(i, &itimer, NULL);
2719                 }
2720                 spin_lock_irq(&current->sighand->siglock);
2721                 if (!fatal_signal_pending(current)) {
2722                         flush_sigqueue(&current->pending);
2723                         flush_sigqueue(&current->signal->shared_pending);
2724                         flush_signal_handlers(current, 1);
2725                         sigemptyset(&current->blocked);
2726                         recalc_sigpending();
2727                 }
2728                 spin_unlock_irq(&current->sighand->siglock);
2729         }
2730
2731         /* Wake up the parent if it is waiting so that it can recheck
2732          * wait permission to the new task SID. */
2733         read_lock(&tasklist_lock);
2734         __wake_up_parent(current, current->real_parent);
2735         read_unlock(&tasklist_lock);
2736 }
2737
2738 /* superblock security operations */
2739
2740 static int selinux_sb_alloc_security(struct super_block *sb)
2741 {
2742         return superblock_alloc_security(sb);
2743 }
2744
2745 static void selinux_sb_free_security(struct super_block *sb)
2746 {
2747         superblock_free_security(sb);
2748 }
2749
2750 static inline int match_prefix(char *prefix, int plen, char *option, int olen)
2751 {
2752         if (plen > olen)
2753                 return 0;
2754
2755         return !memcmp(prefix, option, plen);
2756 }
2757
2758 static inline int selinux_option(char *option, int len)
2759 {
2760         return (match_prefix(CONTEXT_STR, sizeof(CONTEXT_STR)-1, option, len) ||
2761                 match_prefix(FSCONTEXT_STR, sizeof(FSCONTEXT_STR)-1, option, len) ||
2762                 match_prefix(DEFCONTEXT_STR, sizeof(DEFCONTEXT_STR)-1, option, len) ||
2763                 match_prefix(ROOTCONTEXT_STR, sizeof(ROOTCONTEXT_STR)-1, option, len) ||
2764                 match_prefix(LABELSUPP_STR, sizeof(LABELSUPP_STR)-1, option, len));
2765 }
2766
2767 static inline void take_option(char **to, char *from, int *first, int len)
2768 {
2769         if (!*first) {
2770                 **to = ',';
2771                 *to += 1;
2772         } else
2773                 *first = 0;
2774         memcpy(*to, from, len);
2775         *to += len;
2776 }
2777
2778 static inline void take_selinux_option(char **to, char *from, int *first,
2779                                        int len)
2780 {
2781         int current_size = 0;
2782
2783         if (!*first) {
2784                 **to = '|';
2785                 *to += 1;
2786         } else
2787                 *first = 0;
2788
2789         while (current_size < len) {
2790                 if (*from != '"') {
2791                         **to = *from;
2792                         *to += 1;
2793                 }
2794                 from += 1;
2795                 current_size += 1;
2796         }
2797 }
2798
2799 static int selinux_sb_copy_data(char *orig, char *copy)
2800 {
2801         int fnosec, fsec, rc = 0;
2802         char *in_save, *in_curr, *in_end;
2803         char *sec_curr, *nosec_save, *nosec;
2804         int open_quote = 0;
2805
2806         in_curr = orig;
2807         sec_curr = copy;
2808
2809         nosec = (char *)get_zeroed_page(GFP_KERNEL);
2810         if (!nosec) {
2811                 rc = -ENOMEM;
2812                 goto out;
2813         }
2814
2815         nosec_save = nosec;
2816         fnosec = fsec = 1;
2817         in_save = in_end = orig;
2818
2819         do {
2820                 if (*in_end == '"')
2821                         open_quote = !open_quote;
2822                 if ((*in_end == ',' && open_quote == 0) ||
2823                                 *in_end == '\0') {
2824                         int len = in_end - in_curr;
2825
2826                         if (selinux_option(in_curr, len))
2827                                 take_selinux_option(&sec_curr, in_curr, &fsec, len);
2828                         else
2829                                 take_option(&nosec, in_curr, &fnosec, len);
2830
2831                         in_curr = in_end + 1;
2832                 }
2833         } while (*in_end++);
2834
2835         strcpy(in_save, nosec_save);
2836         free_page((unsigned long)nosec_save);
2837 out:
2838         return rc;
2839 }
2840
2841 static int selinux_sb_remount(struct super_block *sb, void *data)
2842 {
2843         int rc, i, *flags;
2844         struct security_mnt_opts opts;
2845         char *secdata, **mount_options;
2846         struct superblock_security_struct *sbsec = sb->s_security;
2847
2848         if (!(sbsec->flags & SE_SBINITIALIZED))
2849                 return 0;
2850
2851         if (!data)
2852                 return 0;
2853
2854         if (sb->s_type->fs_flags & FS_BINARY_MOUNTDATA)
2855                 return 0;
2856
2857         security_init_mnt_opts(&opts);
2858         secdata = alloc_secdata();
2859         if (!secdata)
2860                 return -ENOMEM;
2861         rc = selinux_sb_copy_data(data, secdata);
2862         if (rc)
2863                 goto out_free_secdata;
2864
2865         rc = selinux_parse_opts_str(secdata, &opts);
2866         if (rc)
2867                 goto out_free_secdata;
2868
2869         mount_options = opts.mnt_opts;
2870         flags = opts.mnt_opts_flags;
2871
2872         for (i = 0; i < opts.num_mnt_opts; i++) {
2873                 u32 sid;
2874
2875                 if (flags[i] == SBLABEL_MNT)
2876                         continue;
2877                 rc = security_context_str_to_sid(&selinux_state,
2878                                                  mount_options[i], &sid,
2879                                                  GFP_KERNEL);
2880                 if (rc) {
2881                         pr_warn("SELinux: security_context_str_to_sid"
2882                                "(%s) failed for (dev %s, type %s) errno=%d\n",
2883                                mount_options[i], sb->s_id, sb->s_type->name, rc);
2884                         goto out_free_opts;
2885                 }
2886                 rc = -EINVAL;
2887                 switch (flags[i]) {
2888                 case FSCONTEXT_MNT:
2889                         if (bad_option(sbsec, FSCONTEXT_MNT, sbsec->sid, sid))
2890                                 goto out_bad_option;
2891                         break;
2892                 case CONTEXT_MNT:
2893                         if (bad_option(sbsec, CONTEXT_MNT, sbsec->mntpoint_sid, sid))
2894                                 goto out_bad_option;
2895                         break;
2896                 case ROOTCONTEXT_MNT: {
2897                         struct inode_security_struct *root_isec;
2898                         root_isec = backing_inode_security(sb->s_root);
2899
2900                         if (bad_option(sbsec, ROOTCONTEXT_MNT, root_isec->sid, sid))
2901                                 goto out_bad_option;
2902                         break;
2903                 }
2904                 case DEFCONTEXT_MNT:
2905                         if (bad_option(sbsec, DEFCONTEXT_MNT, sbsec->def_sid, sid))
2906                                 goto out_bad_option;
2907                         break;
2908                 default:
2909                         goto out_free_opts;
2910                 }
2911         }
2912
2913         rc = 0;
2914 out_free_opts:
2915         security_free_mnt_opts(&opts);
2916 out_free_secdata:
2917         free_secdata(secdata);
2918         return rc;
2919 out_bad_option:
2920         pr_warn("SELinux: unable to change security options "
2921                "during remount (dev %s, type=%s)\n", sb->s_id,
2922                sb->s_type->name);
2923         goto out_free_opts;
2924 }
2925
2926 static int selinux_sb_kern_mount(struct super_block *sb, int flags, void *data)
2927 {
2928         const struct cred *cred = current_cred();
2929         struct common_audit_data ad;
2930         int rc;
2931
2932         rc = superblock_doinit(sb, data);
2933         if (rc)
2934                 return rc;
2935
2936         /* Allow all mounts performed by the kernel */
2937         if (flags & MS_KERNMOUNT)
2938                 return 0;
2939
2940         ad.type = LSM_AUDIT_DATA_DENTRY;
2941         ad.u.dentry = sb->s_root;
2942         return superblock_has_perm(cred, sb, FILESYSTEM__MOUNT, &ad);
2943 }
2944
2945 static int selinux_sb_statfs(struct dentry *dentry)
2946 {
2947         const struct cred *cred = current_cred();
2948         struct common_audit_data ad;
2949
2950         ad.type = LSM_AUDIT_DATA_DENTRY;
2951         ad.u.dentry = dentry->d_sb->s_root;
2952         return superblock_has_perm(cred, dentry->d_sb, FILESYSTEM__GETATTR, &ad);
2953 }
2954
2955 static int selinux_mount(const char *dev_name,
2956                          const struct path *path,
2957                          const char *type,
2958                          unsigned long flags,
2959                          void *data)
2960 {
2961         const struct cred *cred = current_cred();
2962
2963         if (flags & MS_REMOUNT)
2964                 return superblock_has_perm(cred, path->dentry->d_sb,
2965                                            FILESYSTEM__REMOUNT, NULL);
2966         else
2967                 return path_has_perm(cred, path, FILE__MOUNTON);
2968 }
2969
2970 static int selinux_umount(struct vfsmount *mnt, int flags)
2971 {
2972         const struct cred *cred = current_cred();
2973
2974         return superblock_has_perm(cred, mnt->mnt_sb,
2975                                    FILESYSTEM__UNMOUNT, NULL);
2976 }
2977
2978 /* inode security operations */
2979
2980 static int selinux_inode_alloc_security(struct inode *inode)
2981 {
2982         return inode_alloc_security(inode);
2983 }
2984
2985 static void selinux_inode_free_security(struct inode *inode)
2986 {
2987         inode_free_security(inode);
2988 }
2989
2990 static int selinux_dentry_init_security(struct dentry *dentry, int mode,
2991                                         const struct qstr *name, void **ctx,
2992                                         u32 *ctxlen)
2993 {
2994         u32 newsid;
2995         int rc;
2996
2997         rc = selinux_determine_inode_label(current_security(),
2998                                            d_inode(dentry->d_parent), name,
2999                                            inode_mode_to_security_class(mode),
3000                                            &newsid);
3001         if (rc)
3002                 return rc;
3003
3004         return security_sid_to_context(&selinux_state, newsid, (char **)ctx,
3005                                        ctxlen);
3006 }
3007
3008 static int selinux_dentry_create_files_as(struct dentry *dentry, int mode,
3009                                           struct qstr *name,
3010                                           const struct cred *old,
3011                                           struct cred *new)
3012 {
3013         u32 newsid;
3014         int rc;
3015         struct task_security_struct *tsec;
3016
3017         rc = selinux_determine_inode_label(old->security,
3018                                            d_inode(dentry->d_parent), name,
3019                                            inode_mode_to_security_class(mode),
3020                                            &newsid);
3021         if (rc)
3022                 return rc;
3023
3024         tsec = new->security;
3025         tsec->create_sid = newsid;
3026         return 0;
3027 }
3028
3029 static int selinux_inode_init_security(struct inode *inode, struct inode *dir,
3030                                        const struct qstr *qstr,
3031                                        const char **name,
3032                                        void **value, size_t *len)
3033 {
3034         const struct task_security_struct *tsec = current_security();
3035         struct superblock_security_struct *sbsec;
3036         u32 newsid, clen;
3037         int rc;
3038         char *context;
3039
3040         sbsec = dir->i_sb->s_security;
3041
3042         newsid = tsec->create_sid;
3043
3044         rc = selinux_determine_inode_label(current_security(),
3045                 dir, qstr,
3046                 inode_mode_to_security_class(inode->i_mode),
3047                 &newsid);
3048         if (rc)
3049                 return rc;
3050
3051         /* Possibly defer initialization to selinux_complete_init. */
3052         if (sbsec->flags & SE_SBINITIALIZED) {
3053                 struct inode_security_struct *isec = inode->i_security;
3054                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
3055                 isec->sid = newsid;
3056                 isec->initialized = LABEL_INITIALIZED;
3057         }
3058
3059         if (!selinux_state.initialized || !(sbsec->flags & SBLABEL_MNT))
3060                 return -EOPNOTSUPP;
3061
3062         if (name)
3063                 *name = XATTR_SELINUX_SUFFIX;
3064
3065         if (value && len) {
3066                 rc = security_sid_to_context_force(&selinux_state, newsid,
3067                                                    &context, &clen);
3068                 if (rc)
3069                         return rc;
3070                 *value = context;
3071                 *len = clen;
3072         }
3073
3074         return 0;
3075 }
3076
3077 static int selinux_inode_create(struct inode *dir, struct dentry *dentry, umode_t mode)
3078 {
3079         return may_create(dir, dentry, SECCLASS_FILE);
3080 }
3081
3082 static int selinux_inode_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3083 {
3084         return may_link(dir, old_dentry, MAY_LINK);
3085 }
3086
3087 static int selinux_inode_unlink(struct inode *dir, struct dentry *dentry)
3088 {
3089         return may_link(dir, dentry, MAY_UNLINK);
3090 }
3091
3092 static int selinux_inode_symlink(struct inode *dir, struct dentry *dentry, const char *name)
3093 {
3094         return may_create(dir, dentry, SECCLASS_LNK_FILE);
3095 }
3096
3097 static int selinux_inode_mkdir(struct inode *dir, struct dentry *dentry, umode_t mask)
3098 {
3099         return may_create(dir, dentry, SECCLASS_DIR);
3100 }
3101
3102 static int selinux_inode_rmdir(struct inode *dir, struct dentry *dentry)
3103 {
3104         return may_link(dir, dentry, MAY_RMDIR);
3105 }
3106
3107 static int selinux_inode_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3108 {
3109         return may_create(dir, dentry, inode_mode_to_security_class(mode));
3110 }
3111
3112 static int selinux_inode_rename(struct inode *old_inode, struct dentry *old_dentry,
3113                                 struct inode *new_inode, struct dentry *new_dentry)
3114 {
3115         return may_rename(old_inode, old_dentry, new_inode, new_dentry);
3116 }
3117
3118 static int selinux_inode_readlink(struct dentry *dentry)
3119 {
3120         const struct cred *cred = current_cred();
3121
3122         return dentry_has_perm(cred, dentry, FILE__READ);
3123 }
3124
3125 static int selinux_inode_follow_link(struct dentry *dentry, struct inode *inode,
3126                                      bool rcu)
3127 {
3128         const struct cred *cred = current_cred();
3129         struct common_audit_data ad;
3130         struct inode_security_struct *isec;
3131         u32 sid;
3132
3133         validate_creds(cred);
3134