security/selinux/hooks.c

   1 /*
   2  *  NSA Security-Enhanced Linux (SELinux) security module
   3  *
   4  *  This file contains the SELinux hook function implementations.
   5  *
   6  *  Authors:  Stephen Smalley, <sds@epoch.ncsc.mil>
   7  *            Chris Vance, <cvance@nai.com>
   8  *            Wayne Salamon, <wsalamon@nai.com>
   9  *            James Morris <jmorris@redhat.com>
  10  *
  11  *  Copyright (C) 2001,2002 Networks Associates Technology, Inc.
  12  *  Copyright (C) 2003-2008 Red Hat, Inc., James Morris <jmorris@redhat.com>
  13  *                                         Eric Paris <eparis@redhat.com>
  14  *  Copyright (C) 2004-2005 Trusted Computer Solutions, Inc.
  15  *                          <dgoeddel@trustedcs.com>
  16  *  Copyright (C) 2006, 2007, 2009 Hewlett-Packard Development Company, L.P.
  17  *      Paul Moore <paul@paul-moore.com>
  18  *  Copyright (C) 2007 Hitachi Software Engineering Co., Ltd.
  19  *                     Yuichi Nakamura <ynakam@hitachisoft.jp>
  20  *
  21  *      This program is free software; you can redistribute it and/or modify
  22  *      it under the terms of the GNU General Public License version 2,
  23  *      as published by the Free Software Foundation.
  24  */
  25
  26 #include <linux/init.h>
  27 #include <linux/kd.h>
  28 #include <linux/kernel.h>
  29 #include <linux/tracehook.h>
  30 #include <linux/errno.h>
  31 #include <linux/sched.h>
  32 #include <linux/security.h>
  33 #include <linux/xattr.h>
  34 #include <linux/capability.h>
  35 #include <linux/unistd.h>
  36 #include <linux/mm.h>
  37 #include <linux/mman.h>
  38 #include <linux/slab.h>
  39 #include <linux/pagemap.h>
  40 #include <linux/proc_fs.h>
  41 #include <linux/swap.h>
  42 #include <linux/spinlock.h>
  43 #include <linux/syscalls.h>
  44 #include <linux/dcache.h>
  45 #include <linux/file.h>
  46 #include <linux/fdtable.h>
  47 #include <linux/namei.h>
  48 #include <linux/mount.h>
  49 #include <linux/netfilter_ipv4.h>
  50 #include <linux/netfilter_ipv6.h>
  51 #include <linux/tty.h>
  52 #include <net/icmp.h>
  53 #include <net/ip.h>             /* for local_port_range[] */
  54 #include <net/tcp.h>            /* struct or_callable used in sock_rcv_skb */
  55 #include <net/net_namespace.h>
  56 #include <net/netlabel.h>
  57 #include <linux/uaccess.h>
  58 #include <asm/ioctls.h>
  59 #include <linux/atomic.h>
  60 #include <linux/bitops.h>
  61 #include <linux/interrupt.h>
  62 #include <linux/netdevice.h>    /* for network interface checks */
  63 #include <linux/netlink.h>
  64 #include <linux/tcp.h>
  65 #include <linux/udp.h>
  66 #include <linux/dccp.h>
  67 #include <linux/quota.h>
  68 #include <linux/un.h>           /* for Unix socket types */
  69 #include <net/af_unix.h>        /* for Unix socket types */
  70 #include <linux/parser.h>
  71 #include <linux/nfs_mount.h>
  72 #include <net/ipv6.h>
  73 #include <linux/hugetlb.h>
  74 #include <linux/personality.h>
  75 #include <linux/audit.h>
  76 #include <linux/string.h>
  77 #include <linux/selinux.h>
  78 #include <linux/mutex.h>
  79 #include <linux/posix-timers.h>
  80 #include <linux/syslog.h>
  81 #include <linux/user_namespace.h>
  82 #include <linux/export.h>
  83 #include <linux/msg.h>
  84 #include <linux/shm.h>
  85
  86 #include "avc.h"
  87 #include "objsec.h"
  88 #include "netif.h"
  89 #include "netnode.h"
  90 #include "netport.h"
  91 #include "xfrm.h"
  92 #include "netlabel.h"
  93 #include "audit.h"
  94 #include "avc_ss.h"
  95
  96 #define NUM_SEL_MNT_OPTS 5
  97
  98 extern struct security_operations *security_ops;
  99
 100 /* SECMARK reference count */
 101 static atomic_t selinux_secmark_refcount = ATOMIC_INIT(0);
 102
 103 #ifdef CONFIG_SECURITY_SELINUX_DEVELOP
 104 int selinux_enforcing;
 105
 106 static int __init enforcing_setup(char *str)
 107 {
 108         unsigned long enforcing;
 109         if (!strict_strtoul(str, 0, &enforcing))
 110                 selinux_enforcing = enforcing ? 1 : 0;
 111         return 1;
 112 }
 113 __setup("enforcing=", enforcing_setup);
 114 #endif
 115
 116 #ifdef CONFIG_SECURITY_SELINUX_BOOTPARAM
 117 int selinux_enabled = CONFIG_SECURITY_SELINUX_BOOTPARAM_VALUE;
 118
 119 static int __init selinux_enabled_setup(char *str)
 120 {
 121         unsigned long enabled;
 122         if (!strict_strtoul(str, 0, &enabled))
 123                 selinux_enabled = enabled ? 1 : 0;
 124         return 1;
 125 }
 126 __setup("selinux=", selinux_enabled_setup);
 127 #else
 128 int selinux_enabled = 1;
 129 #endif
 130
 131 static struct kmem_cache *sel_inode_cache;
 132
 133 /**
 134  * selinux_secmark_enabled - Check to see if SECMARK is currently enabled
 135  *
 136  * Description:
 137  * This function checks the SECMARK reference counter to see if any SECMARK
 138  * targets are currently configured, if the reference counter is greater than
 139  * zero SECMARK is considered to be enabled.  Returns true (1) if SECMARK is
 140  * enabled, false (0) if SECMARK is disabled.
 141  *
 142  */
 143 static int selinux_secmark_enabled(void)
 144 {
 145         return (atomic_read(&selinux_secmark_refcount) > 0);
 146 }
 147
 148 /*
 149  * initialise the security for the init task
 150  */
 151 static void cred_init_security(void)
 152 {
 153         struct cred *cred = (struct cred *) current->real_cred;
 154         struct task_security_struct *tsec;
 155
 156         tsec = kzalloc(sizeof(struct task_security_struct), GFP_KERNEL);
 157         if (!tsec)
 158                 panic("SELinux:  Failed to initialize initial task.\n");
 159
 160         tsec->osid = tsec->sid = SECINITSID_KERNEL;
 161         cred->security = tsec;
 162 }
 163
 164 /*
 165  * get the security ID of a set of credentials
 166  */
 167 static inline u32 cred_sid(const struct cred *cred)
 168 {
 169         const struct task_security_struct *tsec;
 170
 171         tsec = cred->security;
 172         return tsec->sid;
 173 }
 174
 175 /*
 176  * get the objective security ID of a task
 177  */
 178 static inline u32 task_sid(const struct task_struct *task)
 179 {
 180         u32 sid;
 181
 182         rcu_read_lock();
 183         sid = cred_sid(__task_cred(task));
 184         rcu_read_unlock();
 185         return sid;
 186 }
 187
 188 /*
 189  * get the subjective security ID of the current task
 190  */
 191 static inline u32 current_sid(void)
 192 {
 193         const struct task_security_struct *tsec = current_security();
 194
 195         return tsec->sid;
 196 }
 197
 198 /* Allocate and free functions for each kind of security blob. */
 199
 200 static int inode_alloc_security(struct inode *inode)
 201 {
 202         struct inode_security_struct *isec;
 203         u32 sid = current_sid();
 204
 205         isec = kmem_cache_zalloc(sel_inode_cache, GFP_NOFS);
 206         if (!isec)
 207                 return -ENOMEM;
 208
 209         mutex_init(&isec->lock);
 210         INIT_LIST_HEAD(&isec->list);
 211         isec->inode = inode;
 212         isec->sid = SECINITSID_UNLABELED;
 213         isec->sclass = SECCLASS_FILE;
 214         isec->task_sid = sid;
 215         inode->i_security = isec;
 216
 217         return 0;
 218 }
 219
 220 static void inode_free_security(struct inode *inode)
 221 {
 222         struct inode_security_struct *isec = inode->i_security;
 223         struct superblock_security_struct *sbsec = inode->i_sb->s_security;
 224
 225         spin_lock(&sbsec->isec_lock);
 226         if (!list_empty(&isec->list))
 227                 list_del_init(&isec->list);
 228         spin_unlock(&sbsec->isec_lock);
 229
 230         inode->i_security = NULL;
 231         kmem_cache_free(sel_inode_cache, isec);
 232 }
 233
 234 static int file_alloc_security(struct file *file)
 235 {
 236         struct file_security_struct *fsec;
 237         u32 sid = current_sid();
 238
 239         fsec = kzalloc(sizeof(struct file_security_struct), GFP_KERNEL);
 240         if (!fsec)
 241                 return -ENOMEM;
 242
 243         fsec->sid = sid;
 244         fsec->fown_sid = sid;
 245         file->f_security = fsec;
 246
 247         return 0;
 248 }
 249
 250 static void file_free_security(struct file *file)
 251 {
 252         struct file_security_struct *fsec = file->f_security;
 253         file->f_security = NULL;
 254         kfree(fsec);
 255 }
 256
 257 static int superblock_alloc_security(struct super_block *sb)
 258 {
 259         struct superblock_security_struct *sbsec;
 260
 261         sbsec = kzalloc(sizeof(struct superblock_security_struct), GFP_KERNEL);
 262         if (!sbsec)
 263                 return -ENOMEM;
 264
 265         mutex_init(&sbsec->lock);
 266         INIT_LIST_HEAD(&sbsec->isec_head);
 267         spin_lock_init(&sbsec->isec_lock);
 268         sbsec->sb = sb;
 269         sbsec->sid = SECINITSID_UNLABELED;
 270         sbsec->def_sid = SECINITSID_FILE;
 271         sbsec->mntpoint_sid = SECINITSID_UNLABELED;
 272         sb->s_security = sbsec;
 273
 274         return 0;
 275 }
 276
 277 static void superblock_free_security(struct super_block *sb)
 278 {
 279         struct superblock_security_struct *sbsec = sb->s_security;
 280         sb->s_security = NULL;
 281         kfree(sbsec);
 282 }
 283
 284 /* The file system's label must be initialized prior to use. */
 285
 286 static const char *labeling_behaviors[6] = {
 287         "uses xattr",
 288         "uses transition SIDs",
 289         "uses task SIDs",
 290         "uses genfs_contexts",
 291         "not configured for labeling",
 292         "uses mountpoint labeling",
 293 };
 294
 295 static int inode_doinit_with_dentry(struct inode *inode, struct dentry *opt_dentry);
 296
 297 static inline int inode_doinit(struct inode *inode)
 298 {
 299         return inode_doinit_with_dentry(inode, NULL);
 300 }
 301
 302 enum {
 303         Opt_error = -1,
 304         Opt_context = 1,
 305         Opt_fscontext = 2,
 306         Opt_defcontext = 3,
 307         Opt_rootcontext = 4,
 308         Opt_labelsupport = 5,
 309 };
 310
 311 static const match_table_t tokens = {
 312         {Opt_context, CONTEXT_STR "%s"},
 313         {Opt_fscontext, FSCONTEXT_STR "%s"},
 314         {Opt_defcontext, DEFCONTEXT_STR "%s"},
 315         {Opt_rootcontext, ROOTCONTEXT_STR "%s"},
 316         {Opt_labelsupport, LABELSUPP_STR},
 317         {Opt_error, NULL},
 318 };
 319
 320 #define SEL_MOUNT_FAIL_MSG "SELinux:  duplicate or incompatible mount options\n"
 321
 322 static int may_context_mount_sb_relabel(u32 sid,
 323                         struct superblock_security_struct *sbsec,
 324                         const struct cred *cred)
 325 {
 326         const struct task_security_struct *tsec = cred->security;
 327         int rc;
 328
 329         rc = avc_has_perm(tsec->sid, sbsec->sid, SECCLASS_FILESYSTEM,
 330                           FILESYSTEM__RELABELFROM, NULL);
 331         if (rc)
 332                 return rc;
 333
 334         rc = avc_has_perm(tsec->sid, sid, SECCLASS_FILESYSTEM,
 335                           FILESYSTEM__RELABELTO, NULL);
 336         return rc;
 337 }
 338
 339 static int may_context_mount_inode_relabel(u32 sid,
 340                         struct superblock_security_struct *sbsec,
 341                         const struct cred *cred)
 342 {
 343         const struct task_security_struct *tsec = cred->security;
 344         int rc;
 345         rc = avc_has_perm(tsec->sid, sbsec->sid, SECCLASS_FILESYSTEM,
 346                           FILESYSTEM__RELABELFROM, NULL);
 347         if (rc)
 348                 return rc;
 349
 350         rc = avc_has_perm(sid, sbsec->sid, SECCLASS_FILESYSTEM,
 351                           FILESYSTEM__ASSOCIATE, NULL);
 352         return rc;
 353 }
 354
 355 static int sb_finish_set_opts(struct super_block *sb)
 356 {
 357         struct superblock_security_struct *sbsec = sb->s_security;
 358         struct dentry *root = sb->s_root;
 359         struct inode *root_inode = root->d_inode;
 360         int rc = 0;
 361
 362         if (sbsec->behavior == SECURITY_FS_USE_XATTR) {
 363                 /* Make sure that the xattr handler exists and that no
 364                    error other than -ENODATA is returned by getxattr on
 365                    the root directory.  -ENODATA is ok, as this may be
 366                    the first boot of the SELinux kernel before we have
 367                    assigned xattr values to the filesystem. */
 368                 if (!root_inode->i_op->getxattr) {
 369                         printk(KERN_WARNING "SELinux: (dev %s, type %s) has no "
 370                                "xattr support\n", sb->s_id, sb->s_type->name);
 371                         rc = -EOPNOTSUPP;
 372                         goto out;
 373                 }
 374                 rc = root_inode->i_op->getxattr(root, XATTR_NAME_SELINUX, NULL, 0);
 375                 if (rc < 0 && rc != -ENODATA) {
 376                         if (rc == -EOPNOTSUPP)
 377                                 printk(KERN_WARNING "SELinux: (dev %s, type "
 378                                        "%s) has no security xattr handler\n",
 379                                        sb->s_id, sb->s_type->name);
 380                         else
 381                                 printk(KERN_WARNING "SELinux: (dev %s, type "
 382                                        "%s) getxattr errno %d\n", sb->s_id,
 383                                        sb->s_type->name, -rc);
 384                         goto out;
 385                 }
 386         }
 387
 388         sbsec->flags |= (SE_SBINITIALIZED | SE_SBLABELSUPP);
 389
 390         if (sbsec->behavior > ARRAY_SIZE(labeling_behaviors))
 391                 printk(KERN_ERR "SELinux: initialized (dev %s, type %s), unknown behavior\n",
 392                        sb->s_id, sb->s_type->name);
 393         else
 394                 printk(KERN_DEBUG "SELinux: initialized (dev %s, type %s), %s\n",
 395                        sb->s_id, sb->s_type->name,
 396                        labeling_behaviors[sbsec->behavior-1]);
 397
 398         if (sbsec->behavior == SECURITY_FS_USE_GENFS ||
 399             sbsec->behavior == SECURITY_FS_USE_MNTPOINT ||
 400             sbsec->behavior == SECURITY_FS_USE_NONE ||
 401             sbsec->behavior > ARRAY_SIZE(labeling_behaviors))
 402                 sbsec->flags &= ~SE_SBLABELSUPP;
 403
 404         /* Special handling for sysfs. Is genfs but also has setxattr handler*/
 405         if (strncmp(sb->s_type->name, "sysfs", sizeof("sysfs")) == 0)
 406                 sbsec->flags |= SE_SBLABELSUPP;
 407
 408         /* Initialize the root inode. */
 409         rc = inode_doinit_with_dentry(root_inode, root);
 410
 411         /* Initialize any other inodes associated with the superblock, e.g.
 412            inodes created prior to initial policy load or inodes created
 413            during get_sb by a pseudo filesystem that directly
 414            populates itself. */
 415         spin_lock(&sbsec->isec_lock);
 416 next_inode:
 417         if (!list_empty(&sbsec->isec_head)) {
 418                 struct inode_security_struct *isec =
 419                                 list_entry(sbsec->isec_head.next,
 420                                            struct inode_security_struct, list);
 421                 struct inode *inode = isec->inode;
 422                 spin_unlock(&sbsec->isec_lock);
 423                 inode = igrab(inode);
 424                 if (inode) {
 425                         if (!IS_PRIVATE(inode))
 426                                 inode_doinit(inode);
 427                         iput(inode);
 428                 }
 429                 spin_lock(&sbsec->isec_lock);
 430                 list_del_init(&isec->list);
 431                 goto next_inode;
 432         }
 433         spin_unlock(&sbsec->isec_lock);
 434 out:
 435         return rc;
 436 }
 437
 438 /*
 439  * This function should allow an FS to ask what it's mount security
 440  * options were so it can use those later for submounts, displaying
 441  * mount options, or whatever.
 442  */
 443 static int selinux_get_mnt_opts(const struct super_block *sb,
 444                                 struct security_mnt_opts *opts)
 445 {
 446         int rc = 0, i;
 447         struct superblock_security_struct *sbsec = sb->s_security;
 448         char *context = NULL;
 449         u32 len;
 450         char tmp;
 451
 452         security_init_mnt_opts(opts);
 453
 454         if (!(sbsec->flags & SE_SBINITIALIZED))
 455                 return -EINVAL;
 456
 457         if (!ss_initialized)
 458                 return -EINVAL;
 459
 460         tmp = sbsec->flags & SE_MNTMASK;
 461         /* count the number of mount options for this sb */
 462         for (i = 0; i < 8; i++) {
 463                 if (tmp & 0x01)
 464                         opts->num_mnt_opts++;
 465                 tmp >>= 1;
 466         }
 467         /* Check if the Label support flag is set */
 468         if (sbsec->flags & SE_SBLABELSUPP)
 469                 opts->num_mnt_opts++;
 470
 471         opts->mnt_opts = kcalloc(opts->num_mnt_opts, sizeof(char *), GFP_ATOMIC);
 472         if (!opts->mnt_opts) {
 473                 rc = -ENOMEM;
 474                 goto out_free;
 475         }
 476
 477         opts->mnt_opts_flags = kcalloc(opts->num_mnt_opts, sizeof(int), GFP_ATOMIC);
 478         if (!opts->mnt_opts_flags) {
 479                 rc = -ENOMEM;
 480                 goto out_free;
 481         }
 482
 483         i = 0;
 484         if (sbsec->flags & FSCONTEXT_MNT) {
 485                 rc = security_sid_to_context(sbsec->sid, &context, &len);
 486                 if (rc)
 487                         goto out_free;
 488                 opts->mnt_opts[i] = context;
 489                 opts->mnt_opts_flags[i++] = FSCONTEXT_MNT;
 490         }
 491         if (sbsec->flags & CONTEXT_MNT) {
 492                 rc = security_sid_to_context(sbsec->mntpoint_sid, &context, &len);
 493                 if (rc)
 494                         goto out_free;
 495                 opts->mnt_opts[i] = context;
 496                 opts->mnt_opts_flags[i++] = CONTEXT_MNT;
 497         }
 498         if (sbsec->flags & DEFCONTEXT_MNT) {
 499                 rc = security_sid_to_context(sbsec->def_sid, &context, &len);
 500                 if (rc)
 501                         goto out_free;
 502                 opts->mnt_opts[i] = context;
 503                 opts->mnt_opts_flags[i++] = DEFCONTEXT_MNT;
 504         }
 505         if (sbsec->flags & ROOTCONTEXT_MNT) {
 506                 struct inode *root = sbsec->sb->s_root->d_inode;
 507                 struct inode_security_struct *isec = root->i_security;
 508
 509                 rc = security_sid_to_context(isec->sid, &context, &len);
 510                 if (rc)
 511                         goto out_free;
 512                 opts->mnt_opts[i] = context;
 513                 opts->mnt_opts_flags[i++] = ROOTCONTEXT_MNT;
 514         }
 515         if (sbsec->flags & SE_SBLABELSUPP) {
 516                 opts->mnt_opts[i] = NULL;
 517                 opts->mnt_opts_flags[i++] = SE_SBLABELSUPP;
 518         }
 519
 520         BUG_ON(i != opts->num_mnt_opts);
 521
 522         return 0;
 523
 524 out_free:
 525         security_free_mnt_opts(opts);
 526         return rc;
 527 }
 528
 529 static int bad_option(struct superblock_security_struct *sbsec, char flag,
 530                       u32 old_sid, u32 new_sid)
 531 {
 532         char mnt_flags = sbsec->flags & SE_MNTMASK;
 533
 534         /* check if the old mount command had the same options */
 535         if (sbsec->flags & SE_SBINITIALIZED)
 536                 if (!(sbsec->flags & flag) ||
 537                     (old_sid != new_sid))
 538                         return 1;
 539
 540         /* check if we were passed the same options twice,
 541          * aka someone passed context=a,context=b
 542          */
 543         if (!(sbsec->flags & SE_SBINITIALIZED))
 544                 if (mnt_flags & flag)
 545                         return 1;
 546         return 0;
 547 }
 548
 549 /*
 550  * Allow filesystems with binary mount data to explicitly set mount point
 551  * labeling information.
 552  */
 553 static int selinux_set_mnt_opts(struct super_block *sb,
 554                                 struct security_mnt_opts *opts)
 555 {
 556         const struct cred *cred = current_cred();
 557         int rc = 0, i;
 558         struct superblock_security_struct *sbsec = sb->s_security;
 559         const char *name = sb->s_type->name;
 560         struct inode *inode = sbsec->sb->s_root->d_inode;
 561         struct inode_security_struct *root_isec = inode->i_security;
 562         u32 fscontext_sid = 0, context_sid = 0, rootcontext_sid = 0;
 563         u32 defcontext_sid = 0;
 564         char **mount_options = opts->mnt_opts;
 565         int *flags = opts->mnt_opts_flags;
 566         int num_opts = opts->num_mnt_opts;
 567
 568         mutex_lock(&sbsec->lock);
 569
 570         if (!ss_initialized) {
 571                 if (!num_opts) {
 572                         /* Defer initialization until selinux_complete_init,
 573                            after the initial policy is loaded and the security
 574                            server is ready to handle calls. */
 575                         goto out;
 576                 }
 577                 rc = -EINVAL;
 578                 printk(KERN_WARNING "SELinux: Unable to set superblock options "
 579                         "before the security server is initialized\n");
 580                 goto out;
 581         }
 582
 583         /*
 584          * Binary mount data FS will come through this function twice.  Once
 585          * from an explicit call and once from the generic calls from the vfs.
 586          * Since the generic VFS calls will not contain any security mount data
 587          * we need to skip the double mount verification.
 588          *
 589          * This does open a hole in which we will not notice if the first
 590          * mount using this sb set explict options and a second mount using
 591          * this sb does not set any security options.  (The first options
 592          * will be used for both mounts)
 593          */
 594         if ((sbsec->flags & SE_SBINITIALIZED) && (sb->s_type->fs_flags & FS_BINARY_MOUNTDATA)
 595             && (num_opts == 0))
 596                 goto out;
 597
 598         /*
 599          * parse the mount options, check if they are valid sids.
 600          * also check if someone is trying to mount the same sb more
 601          * than once with different security options.
 602          */
 603         for (i = 0; i < num_opts; i++) {
 604                 u32 sid;
 605
 606                 if (flags[i] == SE_SBLABELSUPP)
 607                         continue;
 608                 rc = security_context_to_sid(mount_options[i],
 609                                              strlen(mount_options[i]), &sid);
 610                 if (rc) {
 611                         printk(KERN_WARNING "SELinux: security_context_to_sid"
 612                                "(%s) failed for (dev %s, type %s) errno=%d\n",
 613                                mount_options[i], sb->s_id, name, rc);
 614                         goto out;
 615                 }
 616                 switch (flags[i]) {
 617                 case FSCONTEXT_MNT:
 618                         fscontext_sid = sid;
 619
 620                         if (bad_option(sbsec, FSCONTEXT_MNT, sbsec->sid,
 621                                         fscontext_sid))
 622                                 goto out_double_mount;
 623
 624                         sbsec->flags |= FSCONTEXT_MNT;
 625                         break;
 626                 case CONTEXT_MNT:
 627                         context_sid = sid;
 628
 629                         if (bad_option(sbsec, CONTEXT_MNT, sbsec->mntpoint_sid,
 630                                         context_sid))
 631                                 goto out_double_mount;
 632
 633                         sbsec->flags |= CONTEXT_MNT;
 634                         break;
 635                 case ROOTCONTEXT_MNT:
 636                         rootcontext_sid = sid;
 637
 638                         if (bad_option(sbsec, ROOTCONTEXT_MNT, root_isec->sid,
 639                                         rootcontext_sid))
 640                                 goto out_double_mount;
 641
 642                         sbsec->flags |= ROOTCONTEXT_MNT;
 643
 644                         break;
 645                 case DEFCONTEXT_MNT:
 646                         defcontext_sid = sid;
 647
 648                         if (bad_option(sbsec, DEFCONTEXT_MNT, sbsec->def_sid,
 649                                         defcontext_sid))
 650                                 goto out_double_mount;
 651
 652                         sbsec->flags |= DEFCONTEXT_MNT;
 653
 654                         break;
 655                 default:
 656                         rc = -EINVAL;
 657                         goto out;
 658                 }
 659         }
 660
 661         if (sbsec->flags & SE_SBINITIALIZED) {
 662                 /* previously mounted with options, but not on this attempt? */
 663                 if ((sbsec->flags & SE_MNTMASK) && !num_opts)
 664                         goto out_double_mount;
 665                 rc = 0;
 666                 goto out;
 667         }
 668
 669         if (strcmp(sb->s_type->name, "proc") == 0)
 670                 sbsec->flags |= SE_SBPROC;
 671
 672         /* Determine the labeling behavior to use for this filesystem type. */
 673         rc = security_fs_use((sbsec->flags & SE_SBPROC) ? "proc" : sb->s_type->name, &sbsec->behavior, &sbsec->sid);
 674         if (rc) {
 675                 printk(KERN_WARNING "%s: security_fs_use(%s) returned %d\n",
 676                        __func__, sb->s_type->name, rc);
 677                 goto out;
 678         }
 679
 680         /* sets the context of the superblock for the fs being mounted. */
 681         if (fscontext_sid) {
 682                 rc = may_context_mount_sb_relabel(fscontext_sid, sbsec, cred);
 683                 if (rc)
 684                         goto out;
 685
 686                 sbsec->sid = fscontext_sid;
 687         }
 688
 689         /*
 690          * Switch to using mount point labeling behavior.
 691          * sets the label used on all file below the mountpoint, and will set
 692          * the superblock context if not already set.
 693          */
 694         if (context_sid) {
 695                 if (!fscontext_sid) {
 696                         rc = may_context_mount_sb_relabel(context_sid, sbsec,
 697                                                           cred);
 698                         if (rc)
 699                                 goto out;
 700                         sbsec->sid = context_sid;
 701                 } else {
 702                         rc = may_context_mount_inode_relabel(context_sid, sbsec,
 703                                                              cred);
 704                         if (rc)
 705                                 goto out;
 706                 }
 707                 if (!rootcontext_sid)
 708                         rootcontext_sid = context_sid;
 709
 710                 sbsec->mntpoint_sid = context_sid;
 711                 sbsec->behavior = SECURITY_FS_USE_MNTPOINT;
 712         }
 713
 714         if (rootcontext_sid) {
 715                 rc = may_context_mount_inode_relabel(rootcontext_sid, sbsec,
 716                                                      cred);
 717                 if (rc)
 718                         goto out;
 719
 720                 root_isec->sid = rootcontext_sid;
 721                 root_isec->initialized = 1;
 722         }
 723
 724         if (defcontext_sid) {
 725                 if (sbsec->behavior != SECURITY_FS_USE_XATTR) {
 726                         rc = -EINVAL;
 727                         printk(KERN_WARNING "SELinux: defcontext option is "
 728                                "invalid for this filesystem type\n");
 729                         goto out;
 730                 }
 731
 732                 if (defcontext_sid != sbsec->def_sid) {
 733                         rc = may_context_mount_inode_relabel(defcontext_sid,
 734                                                              sbsec, cred);
 735                         if (rc)
 736                                 goto out;
 737                 }
 738
 739                 sbsec->def_sid = defcontext_sid;
 740         }
 741
 742         rc = sb_finish_set_opts(sb);
 743 out:
 744         mutex_unlock(&sbsec->lock);
 745         return rc;
 746 out_double_mount:
 747         rc = -EINVAL;
 748         printk(KERN_WARNING "SELinux: mount invalid.  Same superblock, different "
 749                "security settings for (dev %s, type %s)\n", sb->s_id, name);
 750         goto out;
 751 }
 752
 753 static void selinux_sb_clone_mnt_opts(const struct super_block *oldsb,
 754                                         struct super_block *newsb)
 755 {
 756         const struct superblock_security_struct *oldsbsec = oldsb->s_security;
 757         struct superblock_security_struct *newsbsec = newsb->s_security;
 758
 759         int set_fscontext =     (oldsbsec->flags & FSCONTEXT_MNT);
 760         int set_context =       (oldsbsec->flags & CONTEXT_MNT);
 761         int set_rootcontext =   (oldsbsec->flags & ROOTCONTEXT_MNT);
 762
 763         /*
 764          * if the parent was able to be mounted it clearly had no special lsm
 765          * mount options.  thus we can safely deal with this superblock later
 766          */
 767         if (!ss_initialized)
 768                 return;
 769
 770         /* how can we clone if the old one wasn't set up?? */
 771         BUG_ON(!(oldsbsec->flags & SE_SBINITIALIZED));
 772
 773         /* if fs is reusing a sb, just let its options stand... */
 774         if (newsbsec->flags & SE_SBINITIALIZED)
 775                 return;
 776
 777         mutex_lock(&newsbsec->lock);
 778
 779         newsbsec->flags = oldsbsec->flags;
 780
 781         newsbsec->sid = oldsbsec->sid;
 782         newsbsec->def_sid = oldsbsec->def_sid;
 783         newsbsec->behavior = oldsbsec->behavior;
 784
 785         if (set_context) {
 786                 u32 sid = oldsbsec->mntpoint_sid;
 787
 788                 if (!set_fscontext)
 789                         newsbsec->sid = sid;
 790                 if (!set_rootcontext) {
 791                         struct inode *newinode = newsb->s_root->d_inode;
 792                         struct inode_security_struct *newisec = newinode->i_security;
 793                         newisec->sid = sid;
 794                 }
 795                 newsbsec->mntpoint_sid = sid;
 796         }
 797         if (set_rootcontext) {
 798                 const struct inode *oldinode = oldsb->s_root->d_inode;
 799                 const struct inode_security_struct *oldisec = oldinode->i_security;
 800                 struct inode *newinode = newsb->s_root->d_inode;
 801                 struct inode_security_struct *newisec = newinode->i_security;
 802
 803                 newisec->sid = oldisec->sid;
 804         }
 805
 806         sb_finish_set_opts(newsb);
 807         mutex_unlock(&newsbsec->lock);
 808 }
 809
 810 static int selinux_parse_opts_str(char *options,
 811                                   struct security_mnt_opts *opts)
 812 {
 813         char *p;
 814         char *context = NULL, *defcontext = NULL;
 815         char *fscontext = NULL, *rootcontext = NULL;
 816         int rc, num_mnt_opts = 0;
 817
 818         opts->num_mnt_opts = 0;
 819
 820         /* Standard string-based options. */
 821         while ((p = strsep(&options, "|")) != NULL) {
 822                 int token;
 823                 substring_t args[MAX_OPT_ARGS];
 824
 825                 if (!*p)
 826                         continue;
 827
 828                 token = match_token(p, tokens, args);
 829
 830                 switch (token) {
 831                 case Opt_context:
 832                         if (context || defcontext) {
 833                                 rc = -EINVAL;
 834                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
 835                                 goto out_err;
 836                         }
 837                         context = match_strdup(&args[0]);
 838                         if (!context) {
 839                                 rc = -ENOMEM;
 840                                 goto out_err;
 841                         }
 842                         break;
 843
 844                 case Opt_fscontext:
 845                         if (fscontext) {
 846                                 rc = -EINVAL;
 847                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
 848                                 goto out_err;
 849                         }
 850                         fscontext = match_strdup(&args[0]);
 851                         if (!fscontext) {
 852                                 rc = -ENOMEM;
 853                                 goto out_err;
 854                         }
 855                         break;
 856
 857                 case Opt_rootcontext:
 858                         if (rootcontext) {
 859                                 rc = -EINVAL;
 860                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
 861                                 goto out_err;
 862                         }
 863                         rootcontext = match_strdup(&args[0]);
 864                         if (!rootcontext) {
 865                                 rc = -ENOMEM;
 866                                 goto out_err;
 867                         }
 868                         break;
 869
 870                 case Opt_defcontext:
 871                         if (context || defcontext) {
 872                                 rc = -EINVAL;
 873                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
 874                                 goto out_err;
 875                         }
 876                         defcontext = match_strdup(&args[0]);
 877                         if (!defcontext) {
 878                                 rc = -ENOMEM;
 879                                 goto out_err;
 880                         }
 881                         break;
 882                 case Opt_labelsupport:
 883                         break;
 884                 default:
 885                         rc = -EINVAL;
 886                         printk(KERN_WARNING "SELinux:  unknown mount option\n");
 887                         goto out_err;
 888
 889                 }
 890         }
 891
 892         rc = -ENOMEM;
 893         opts->mnt_opts = kcalloc(NUM_SEL_MNT_OPTS, sizeof(char *), GFP_ATOMIC);
 894         if (!opts->mnt_opts)
 895                 goto out_err;
 896
 897         opts->mnt_opts_flags = kcalloc(NUM_SEL_MNT_OPTS, sizeof(int), GFP_ATOMIC);
 898         if (!opts->mnt_opts_flags) {
 899                 kfree(opts->mnt_opts);
 900                 goto out_err;
 901         }
 902
 903         if (fscontext) {
 904                 opts->mnt_opts[num_mnt_opts] = fscontext;
 905                 opts->mnt_opts_flags[num_mnt_opts++] = FSCONTEXT_MNT;
 906         }
 907         if (context) {
 908                 opts->mnt_opts[num_mnt_opts] = context;
 909                 opts->mnt_opts_flags[num_mnt_opts++] = CONTEXT_MNT;
 910         }
 911         if (rootcontext) {
 912                 opts->mnt_opts[num_mnt_opts] = rootcontext;
 913                 opts->mnt_opts_flags[num_mnt_opts++] = ROOTCONTEXT_MNT;
 914         }
 915         if (defcontext) {
 916                 opts->mnt_opts[num_mnt_opts] = defcontext;
 917                 opts->mnt_opts_flags[num_mnt_opts++] = DEFCONTEXT_MNT;
 918         }
 919
 920         opts->num_mnt_opts = num_mnt_opts;
 921         return 0;
 922
 923 out_err:
 924         kfree(context);
 925         kfree(defcontext);
 926         kfree(fscontext);
 927         kfree(rootcontext);
 928         return rc;
 929 }
 930 /*
 931  * string mount options parsing and call set the sbsec
 932  */
 933 static int superblock_doinit(struct super_block *sb, void *data)
 934 {
 935         int rc = 0;
 936         char *options = data;
 937         struct security_mnt_opts opts;
 938
 939         security_init_mnt_opts(&opts);
 940
 941         if (!data)
 942                 goto out;
 943
 944         BUG_ON(sb->s_type->fs_flags & FS_BINARY_MOUNTDATA);
 945
 946         rc = selinux_parse_opts_str(options, &opts);
 947         if (rc)
 948                 goto out_err;
 949
 950 out:
 951         rc = selinux_set_mnt_opts(sb, &opts);
 952
 953 out_err:
 954         security_free_mnt_opts(&opts);
 955         return rc;
 956 }
 957
 958 static void selinux_write_opts(struct seq_file *m,
 959                                struct security_mnt_opts *opts)
 960 {
 961         int i;
 962         char *prefix;
 963
 964         for (i = 0; i < opts->num_mnt_opts; i++) {
 965                 char *has_comma;
 966
 967                 if (opts->mnt_opts[i])
 968                         has_comma = strchr(opts->mnt_opts[i], ',');
 969                 else
 970                         has_comma = NULL;
 971
 972                 switch (opts->mnt_opts_flags[i]) {
 973                 case CONTEXT_MNT:
 974                         prefix = CONTEXT_STR;
 975                         break;
 976                 case FSCONTEXT_MNT:
 977                         prefix = FSCONTEXT_STR;
 978                         break;
 979                 case ROOTCONTEXT_MNT:
 980                         prefix = ROOTCONTEXT_STR;
 981                         break;
 982                 case DEFCONTEXT_MNT:
 983                         prefix = DEFCONTEXT_STR;
 984                         break;
 985                 case SE_SBLABELSUPP:
 986                         seq_putc(m, ',');
 987                         seq_puts(m, LABELSUPP_STR);
 988                         continue;
 989                 default:
 990                         BUG();
 991                         return;
 992                 };
 993                 /* we need a comma before each option */
 994                 seq_putc(m, ',');
 995                 seq_puts(m, prefix);
 996                 if (has_comma)
 997                         seq_putc(m, '\"');
 998                 seq_puts(m, opts->mnt_opts[i]);
 999                 if (has_comma)
1000                         seq_putc(m, '\"');
1001         }
1002 }
1003
1004 static int selinux_sb_show_options(struct seq_file *m, struct super_block *sb)
1005 {
1006         struct security_mnt_opts opts;
1007         int rc;
1008
1009         rc = selinux_get_mnt_opts(sb, &opts);
1010         if (rc) {
1011                 /* before policy load we may get EINVAL, don't show anything */
1012                 if (rc == -EINVAL)
1013                         rc = 0;
1014                 return rc;
1015         }
1016
1017         selinux_write_opts(m, &opts);
1018
1019         security_free_mnt_opts(&opts);
1020
1021         return rc;
1022 }
1023
1024 static inline u16 inode_mode_to_security_class(umode_t mode)
1025 {
1026         switch (mode & S_IFMT) {
1027         case S_IFSOCK:
1028                 return SECCLASS_SOCK_FILE;
1029         case S_IFLNK:
1030                 return SECCLASS_LNK_FILE;
1031         case S_IFREG:
1032                 return SECCLASS_FILE;
1033         case S_IFBLK:
1034                 return SECCLASS_BLK_FILE;
1035         case S_IFDIR:
1036                 return SECCLASS_DIR;
1037         case S_IFCHR:
1038                 return SECCLASS_CHR_FILE;
1039         case S_IFIFO:
1040                 return SECCLASS_FIFO_FILE;
1041
1042         }
1043
1044         return SECCLASS_FILE;
1045 }
1046
1047 static inline int default_protocol_stream(int protocol)
1048 {
1049         return (protocol == IPPROTO_IP || protocol == IPPROTO_TCP);
1050 }
1051
1052 static inline int default_protocol_dgram(int protocol)
1053 {
1054         return (protocol == IPPROTO_IP || protocol == IPPROTO_UDP);
1055 }
1056
1057 static inline u16 socket_type_to_security_class(int family, int type, int protocol)
1058 {
1059         switch (family) {
1060         case PF_UNIX:
1061                 switch (type) {
1062                 case SOCK_STREAM:
1063                 case SOCK_SEQPACKET:
1064                         return SECCLASS_UNIX_STREAM_SOCKET;
1065                 case SOCK_DGRAM:
1066                         return SECCLASS_UNIX_DGRAM_SOCKET;
1067                 }
1068                 break;
1069         case PF_INET:
1070         case PF_INET6:
1071                 switch (type) {
1072                 case SOCK_STREAM:
1073                         if (default_protocol_stream(protocol))
1074                                 return SECCLASS_TCP_SOCKET;
1075                         else
1076                                 return SECCLASS_RAWIP_SOCKET;
1077                 case SOCK_DGRAM:
1078                         if (default_protocol_dgram(protocol))
1079                                 return SECCLASS_UDP_SOCKET;
1080                         else
1081                                 return SECCLASS_RAWIP_SOCKET;
1082                 case SOCK_DCCP:
1083                         return SECCLASS_DCCP_SOCKET;
1084                 default:
1085                         return SECCLASS_RAWIP_SOCKET;
1086                 }
1087                 break;
1088         case PF_NETLINK:
1089                 switch (protocol) {
1090                 case NETLINK_ROUTE:
1091                         return SECCLASS_NETLINK_ROUTE_SOCKET;
1092                 case NETLINK_FIREWALL:
1093                         return SECCLASS_NETLINK_FIREWALL_SOCKET;
1094                 case NETLINK_SOCK_DIAG:
1095                         return SECCLASS_NETLINK_TCPDIAG_SOCKET;
1096                 case NETLINK_NFLOG:
1097                         return SECCLASS_NETLINK_NFLOG_SOCKET;
1098                 case NETLINK_XFRM:
1099                         return SECCLASS_NETLINK_XFRM_SOCKET;
1100                 case NETLINK_SELINUX:
1101                         return SECCLASS_NETLINK_SELINUX_SOCKET;
1102                 case NETLINK_AUDIT:
1103                         return SECCLASS_NETLINK_AUDIT_SOCKET;
1104                 case NETLINK_IP6_FW:
1105                         return SECCLASS_NETLINK_IP6FW_SOCKET;
1106                 case NETLINK_DNRTMSG:
1107                         return SECCLASS_NETLINK_DNRT_SOCKET;
1108                 case NETLINK_KOBJECT_UEVENT:
1109                         return SECCLASS_NETLINK_KOBJECT_UEVENT_SOCKET;
1110                 default:
1111                         return SECCLASS_NETLINK_SOCKET;
1112                 }
1113         case PF_PACKET:
1114                 return SECCLASS_PACKET_SOCKET;
1115         case PF_KEY:
1116                 return SECCLASS_KEY_SOCKET;
1117         case PF_APPLETALK:
1118                 return SECCLASS_APPLETALK_SOCKET;
1119         }
1120
1121         return SECCLASS_SOCKET;
1122 }
1123
1124 #ifdef CONFIG_PROC_FS
1125 static int selinux_proc_get_sid(struct dentry *dentry,
1126                                 u16 tclass,
1127                                 u32 *sid)
1128 {
1129         int rc;
1130         char *buffer, *path;
1131
1132         buffer = (char *)__get_free_page(GFP_KERNEL);
1133         if (!buffer)
1134                 return -ENOMEM;
1135
1136         path = dentry_path_raw(dentry, buffer, PAGE_SIZE);
1137         if (IS_ERR(path))
1138                 rc = PTR_ERR(path);
1139         else {
1140                 /* each process gets a /proc/PID/ entry. Strip off the
1141                  * PID part to get a valid selinux labeling.
1142                  * e.g. /proc/1/net/rpc/nfs -> /net/rpc/nfs */
1143                 while (path[1] >= '0' && path[1] <= '9') {
1144                         path[1] = '/';
1145                         path++;
1146                 }
1147                 rc = security_genfs_sid("proc", path, tclass, sid);
1148         }
1149         free_page((unsigned long)buffer);
1150         return rc;
1151 }
1152 #else
1153 static int selinux_proc_get_sid(struct dentry *dentry,
1154                                 u16 tclass,
1155                                 u32 *sid)
1156 {
1157         return -EINVAL;
1158 }
1159 #endif
1160
1161 /* The inode's security attributes must be initialized before first use. */
1162 static int inode_doinit_with_dentry(struct inode *inode, struct dentry *opt_dentry)
1163 {
1164         struct superblock_security_struct *sbsec = NULL;
1165         struct inode_security_struct *isec = inode->i_security;
1166         u32 sid;
1167         struct dentry *dentry;
1168 #define INITCONTEXTLEN 255
1169         char *context = NULL;
1170         unsigned len = 0;
1171         int rc = 0;
1172
1173         if (isec->initialized)
1174                 goto out;
1175
1176         mutex_lock(&isec->lock);
1177         if (isec->initialized)
1178                 goto out_unlock;
1179
1180         sbsec = inode->i_sb->s_security;
1181         if (!(sbsec->flags & SE_SBINITIALIZED)) {
1182                 /* Defer initialization until selinux_complete_init,
1183                    after the initial policy is loaded and the security
1184                    server is ready to handle calls. */
1185                 spin_lock(&sbsec->isec_lock);
1186                 if (list_empty(&isec->list))
1187                         list_add(&isec->list, &sbsec->isec_head);
1188                 spin_unlock(&sbsec->isec_lock);
1189                 goto out_unlock;
1190         }
1191
1192         switch (sbsec->behavior) {
1193         case SECURITY_FS_USE_XATTR:
1194                 if (!inode->i_op->getxattr) {
1195                         isec->sid = sbsec->def_sid;
1196                         break;
1197                 }
1198
1199                 /* Need a dentry, since the xattr API requires one.
1200                    Life would be simpler if we could just pass the inode. */
1201                 if (opt_dentry) {
1202                         /* Called from d_instantiate or d_splice_alias. */
1203                         dentry = dget(opt_dentry);
1204                 } else {
1205                         /* Called from selinux_complete_init, try to find a dentry. */
1206                         dentry = d_find_alias(inode);
1207                 }
1208                 if (!dentry) {
1209                         /*
1210                          * this is can be hit on boot when a file is accessed
1211                          * before the policy is loaded.  When we load policy we
1212                          * may find inodes that have no dentry on the
1213                          * sbsec->isec_head list.  No reason to complain as these
1214                          * will get fixed up the next time we go through
1215                          * inode_doinit with a dentry, before these inodes could
1216                          * be used again by userspace.
1217                          */
1218                         goto out_unlock;
1219                 }
1220
1221                 len = INITCONTEXTLEN;
1222                 context = kmalloc(len+1, GFP_NOFS);
1223                 if (!context) {
1224                         rc = -ENOMEM;
1225                         dput(dentry);
1226                         goto out_unlock;
1227                 }
1228                 context[len] = '\0';
1229                 rc = inode->i_op->getxattr(dentry, XATTR_NAME_SELINUX,
1230                                            context, len);
1231                 if (rc == -ERANGE) {
1232                         kfree(context);
1233
1234                         /* Need a larger buffer.  Query for the right size. */
1235                         rc = inode->i_op->getxattr(dentry, XATTR_NAME_SELINUX,
1236                                                    NULL, 0);
1237                         if (rc < 0) {
1238                                 dput(dentry);
1239                                 goto out_unlock;
1240                         }
1241                         len = rc;
1242                         context = kmalloc(len+1, GFP_NOFS);
1243                         if (!context) {
1244                                 rc = -ENOMEM;
1245                                 dput(dentry);
1246                                 goto out_unlock;
1247                         }
1248                         context[len] = '\0';
1249                         rc = inode->i_op->getxattr(dentry,
1250                                                    XATTR_NAME_SELINUX,
1251                                                    context, len);
1252                 }
1253                 dput(dentry);
1254                 if (rc < 0) {
1255                         if (rc != -ENODATA) {
1256                                 printk(KERN_WARNING "SELinux: %s:  getxattr returned "
1257                                        "%d for dev=%s ino=%ld\n", __func__,
1258                                        -rc, inode->i_sb->s_id, inode->i_ino);
1259                                 kfree(context);
1260                                 goto out_unlock;
1261                         }
1262                         /* Map ENODATA to the default file SID */
1263                         sid = sbsec->def_sid;
1264                         rc = 0;
1265                 } else {
1266                         rc = security_context_to_sid_default(context, rc, &sid,
1267                                                              sbsec->def_sid,
1268                                                              GFP_NOFS);
1269                         if (rc) {
1270                                 char *dev = inode->i_sb->s_id;
1271                                 unsigned long ino = inode->i_ino;
1272
1273                                 if (rc == -EINVAL) {
1274                                         if (printk_ratelimit())
1275                                                 printk(KERN_NOTICE "SELinux: inode=%lu on dev=%s was found to have an invalid "
1276                                                         "context=%s.  This indicates you may need to relabel the inode or the "
1277                                                         "filesystem in question.\n", ino, dev, context);
1278                                 } else {
1279                                         printk(KERN_WARNING "SELinux: %s:  context_to_sid(%s) "
1280                                                "returned %d for dev=%s ino=%ld\n",
1281                                                __func__, context, -rc, dev, ino);
1282                                 }
1283                                 kfree(context);
1284                                 /* Leave with the unlabeled SID */
1285                                 rc = 0;
1286                                 break;
1287                         }
1288                 }
1289                 kfree(context);
1290                 isec->sid = sid;
1291                 break;
1292         case SECURITY_FS_USE_TASK:
1293                 isec->sid = isec->task_sid;
1294                 break;
1295         case SECURITY_FS_USE_TRANS:
1296                 /* Default to the fs SID. */
1297                 isec->sid = sbsec->sid;
1298
1299                 /* Try to obtain a transition SID. */
1300                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
1301                 rc = security_transition_sid(isec->task_sid, sbsec->sid,
1302                                              isec->sclass, NULL, &sid);
1303                 if (rc)
1304                         goto out_unlock;
1305                 isec->sid = sid;
1306                 break;
1307         case SECURITY_FS_USE_MNTPOINT:
1308                 isec->sid = sbsec->mntpoint_sid;
1309                 break;
1310         default:
1311                 /* Default to the fs superblock SID. */
1312                 isec->sid = sbsec->sid;
1313
1314                 if ((sbsec->flags & SE_SBPROC) && !S_ISLNK(inode->i_mode)) {
1315                         if (opt_dentry) {
1316                                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
1317                                 rc = selinux_proc_get_sid(opt_dentry,
1318                                                           isec->sclass,
1319                                                           &sid);
1320                                 if (rc)
1321                                         goto out_unlock;
1322                                 isec->sid = sid;
1323                         }
1324                 }
1325                 break;
1326         }
1327
1328         isec->initialized = 1;
1329
1330 out_unlock:
1331         mutex_unlock(&isec->lock);
1332 out:
1333         if (isec->sclass == SECCLASS_FILE)
1334                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
1335         return rc;
1336 }
1337
1338 /* Convert a Linux signal to an access vector. */
1339 static inline u32 signal_to_av(int sig)
1340 {
1341         u32 perm = 0;
1342
1343         switch (sig) {
1344         case SIGCHLD:
1345                 /* Commonly granted from child to parent. */
1346                 perm = PROCESS__SIGCHLD;
1347                 break;
1348         case SIGKILL:
1349                 /* Cannot be caught or ignored */
1350                 perm = PROCESS__SIGKILL;
1351                 break;
1352         case SIGSTOP:
1353                 /* Cannot be caught or ignored */
1354                 perm = PROCESS__SIGSTOP;
1355                 break;
1356         default:
1357                 /* All other signals. */
1358                 perm = PROCESS__SIGNAL;
1359                 break;
1360         }
1361
1362         return perm;
1363 }
1364
1365 /*
1366  * Check permission between a pair of credentials
1367  * fork check, ptrace check, etc.
1368  */
1369 static int cred_has_perm(const struct cred *actor,
1370                          const struct cred *target,
1371                          u32 perms)
1372 {
1373         u32 asid = cred_sid(actor), tsid = cred_sid(target);
1374
1375         return avc_has_perm(asid, tsid, SECCLASS_PROCESS, perms, NULL);
1376 }
1377
1378 /*
1379  * Check permission between a pair of tasks, e.g. signal checks,
1380  * fork check, ptrace check, etc.
1381  * tsk1 is the actor and tsk2 is the target
1382  * - this uses the default subjective creds of tsk1
1383  */
1384 static int task_has_perm(const struct task_struct *tsk1,
1385                          const struct task_struct *tsk2,
1386                          u32 perms)
1387 {
1388         const struct task_security_struct *__tsec1, *__tsec2;
1389         u32 sid1, sid2;
1390
1391         rcu_read_lock();
1392         __tsec1 = __task_cred(tsk1)->security;  sid1 = __tsec1->sid;
1393         __tsec2 = __task_cred(tsk2)->security;  sid2 = __tsec2->sid;
1394         rcu_read_unlock();
1395         return avc_has_perm(sid1, sid2, SECCLASS_PROCESS, perms, NULL);
1396 }
1397
1398 /*
1399  * Check permission between current and another task, e.g. signal checks,
1400  * fork check, ptrace check, etc.
1401  * current is the actor and tsk2 is the target
1402  * - this uses current's subjective creds
1403  */
1404 static int current_has_perm(const struct task_struct *tsk,
1405                             u32 perms)
1406 {
1407         u32 sid, tsid;
1408
1409         sid = current_sid();
1410         tsid = task_sid(tsk);
1411         return avc_has_perm(sid, tsid, SECCLASS_PROCESS, perms, NULL);
1412 }
1413
1414 #if CAP_LAST_CAP > 63
1415 #error Fix SELinux to handle capabilities > 63.
1416 #endif
1417
1418 /* Check whether a task is allowed to use a capability. */
1419 static int cred_has_capability(const struct cred *cred,
1420                                int cap, int audit)
1421 {
1422         struct common_audit_data ad;
1423         struct av_decision avd;
1424         u16 sclass;
1425         u32 sid = cred_sid(cred);
1426         u32 av = CAP_TO_MASK(cap);
1427         int rc;
1428
1429         ad.type = LSM_AUDIT_DATA_CAP;
1430         ad.u.cap = cap;
1431
1432         switch (CAP_TO_INDEX(cap)) {
1433         case 0:
1434                 sclass = SECCLASS_CAPABILITY;
1435                 break;
1436         case 1:
1437                 sclass = SECCLASS_CAPABILITY2;
1438                 break;
1439         default:
1440                 printk(KERN_ERR
1441                        "SELinux:  out of range capability %d\n", cap);
1442                 BUG();
1443                 return -EINVAL;
1444         }
1445
1446         rc = avc_has_perm_noaudit(sid, sid, sclass, av, 0, &avd);
1447         if (audit == SECURITY_CAP_AUDIT) {
1448                 int rc2 = avc_audit(sid, sid, sclass, av, &avd, rc, &ad, 0);
1449                 if (rc2)
1450                         return rc2;
1451         }
1452         return rc;
1453 }
1454
1455 /* Check whether a task is allowed to use a system operation. */
1456 static int task_has_system(struct task_struct *tsk,
1457                            u32 perms)
1458 {
1459         u32 sid = task_sid(tsk);
1460
1461         return avc_has_perm(sid, SECINITSID_KERNEL,
1462                             SECCLASS_SYSTEM, perms, NULL);
1463 }
1464
1465 /* Check whether a task has a particular permission to an inode.
1466    The 'adp' parameter is optional and allows other audit
1467    data to be passed (e.g. the dentry). */
1468 static int inode_has_perm(const struct cred *cred,
1469                           struct inode *inode,
1470                           u32 perms,
1471                           struct common_audit_data *adp,
1472                           unsigned flags)
1473 {
1474         struct inode_security_struct *isec;
1475         u32 sid;
1476
1477         validate_creds(cred);
1478
1479         if (unlikely(IS_PRIVATE(inode)))
1480                 return 0;
1481
1482         sid = cred_sid(cred);
1483         isec = inode->i_security;
1484
1485         return avc_has_perm_flags(sid, isec->sid, isec->sclass, perms, adp, flags);
1486 }
1487
1488 /* Same as inode_has_perm, but pass explicit audit data containing
1489    the dentry to help the auditing code to more easily generate the
1490    pathname if needed. */
1491 static inline int dentry_has_perm(const struct cred *cred,
1492                                   struct dentry *dentry,
1493                                   u32 av)
1494 {
1495         struct inode *inode = dentry->d_inode;
1496         struct common_audit_data ad;
1497
1498         ad.type = LSM_AUDIT_DATA_DENTRY;
1499         ad.u.dentry = dentry;
1500         return inode_has_perm(cred, inode, av, &ad, 0);
1501 }
1502
1503 /* Same as inode_has_perm, but pass explicit audit data containing
1504    the path to help the auditing code to more easily generate the
1505    pathname if needed. */
1506 static inline int path_has_perm(const struct cred *cred,
1507                                 struct path *path,
1508                                 u32 av)
1509 {
1510         struct inode *inode = path->dentry->d_inode;
1511         struct common_audit_data ad;
1512
1513         ad.type = LSM_AUDIT_DATA_PATH;
1514         ad.u.path = *path;
1515         return inode_has_perm(cred, inode, av, &ad, 0);
1516 }
1517
1518 /* Check whether a task can use an open file descriptor to
1519    access an inode in a given way.  Check access to the
1520    descriptor itself, and then use dentry_has_perm to
1521    check a particular permission to the file.
1522    Access to the descriptor is implicitly granted if it
1523    has the same SID as the process.  If av is zero, then
1524    access to the file is not checked, e.g. for cases
1525    where only the descriptor is affected like seek. */
1526 static int file_has_perm(const struct cred *cred,
1527                          struct file *file,
1528                          u32 av)
1529 {
1530         struct file_security_struct *fsec = file->f_security;
1531         struct inode *inode = file_inode(file);
1532         struct common_audit_data ad;
1533         u32 sid = cred_sid(cred);
1534         int rc;
1535
1536         ad.type = LSM_AUDIT_DATA_PATH;
1537         ad.u.path = file->f_path;
1538
1539         if (sid != fsec->sid) {
1540                 rc = avc_has_perm(sid, fsec->sid,
1541                                   SECCLASS_FD,
1542                                   FD__USE,
1543                                   &ad);
1544                 if (rc)
1545                         goto out;
1546         }
1547
1548         /* av is zero if only checking access to the descriptor. */
1549         rc = 0;
1550         if (av)
1551                 rc = inode_has_perm(cred, inode, av, &ad, 0);
1552
1553 out:
1554         return rc;
1555 }
1556
1557 /* Check whether a task can create a file. */
1558 static int may_create(struct inode *dir,
1559                       struct dentry *dentry,
1560                       u16 tclass)
1561 {
1562         const struct task_security_struct *tsec = current_security();
1563         struct inode_security_struct *dsec;
1564         struct superblock_security_struct *sbsec;
1565         u32 sid, newsid;
1566         struct common_audit_data ad;
1567         int rc;
1568
1569         dsec = dir->i_security;
1570         sbsec = dir->i_sb->s_security;
1571
1572         sid = tsec->sid;
1573         newsid = tsec->create_sid;
1574
1575         ad.type = LSM_AUDIT_DATA_DENTRY;
1576         ad.u.dentry = dentry;
1577
1578         rc = avc_has_perm(sid, dsec->sid, SECCLASS_DIR,
1579                           DIR__ADD_NAME | DIR__SEARCH,
1580                           &ad);
1581         if (rc)
1582                 return rc;
1583
1584         if (!newsid || !(sbsec->flags & SE_SBLABELSUPP)) {
1585                 rc = security_transition_sid(sid, dsec->sid, tclass,
1586                                              &dentry->d_name, &newsid);
1587                 if (rc)
1588                         return rc;
1589         }
1590
1591         rc = avc_has_perm(sid, newsid, tclass, FILE__CREATE, &ad);
1592         if (rc)
1593                 return rc;
1594
1595         return avc_has_perm(newsid, sbsec->sid,
1596                             SECCLASS_FILESYSTEM,
1597                             FILESYSTEM__ASSOCIATE, &ad);
1598 }
1599
1600 /* Check whether a task can create a key. */
1601 static int may_create_key(u32 ksid,
1602                           struct task_struct *ctx)
1603 {
1604         u32 sid = task_sid(ctx);
1605
1606         return avc_has_perm(sid, ksid, SECCLASS_KEY, KEY__CREATE, NULL);
1607 }
1608
1609 #define MAY_LINK        0
1610 #define MAY_UNLINK      1
1611 #define MAY_RMDIR       2
1612
1613 /* Check whether a task can link, unlink, or rmdir a file/directory. */
1614 static int may_link(struct inode *dir,
1615                     struct dentry *dentry,
1616                     int kind)
1617
1618 {
1619         struct inode_security_struct *dsec, *isec;
1620         struct common_audit_data ad;
1621         u32 sid = current_sid();
1622         u32 av;
1623         int rc;
1624
1625         dsec = dir->i_security;
1626         isec = dentry->d_inode->i_security;
1627
1628         ad.type = LSM_AUDIT_DATA_DENTRY;
1629         ad.u.dentry = dentry;
1630
1631         av = DIR__SEARCH;
1632         av |= (kind ? DIR__REMOVE_NAME : DIR__ADD_NAME);
1633         rc = avc_has_perm(sid, dsec->sid, SECCLASS_DIR, av, &ad);
1634         if (rc)
1635                 return rc;
1636
1637         switch (kind) {
1638         case MAY_LINK:
1639                 av = FILE__LINK;
1640                 break;
1641         case MAY_UNLINK:
1642                 av = FILE__UNLINK;
1643                 break;
1644         case MAY_RMDIR:
1645                 av = DIR__RMDIR;
1646                 break;
1647         default:
1648                 printk(KERN_WARNING "SELinux: %s:  unrecognized kind %d\n",
1649                         __func__, kind);
1650                 return 0;
1651         }
1652
1653         rc = avc_has_perm(sid, isec->sid, isec->sclass, av, &ad);
1654         return rc;
1655 }
1656
1657 static inline int may_rename(struct inode *old_dir,
1658                              struct dentry *old_dentry,
1659                              struct inode *new_dir,
1660                              struct dentry *new_dentry)
1661 {
1662         struct inode_security_struct *old_dsec, *new_dsec, *old_isec, *new_isec;
1663         struct common_audit_data ad;
1664         u32 sid = current_sid();
1665         u32 av;
1666         int old_is_dir, new_is_dir;
1667         int rc;
1668
1669         old_dsec = old_dir->i_security;
1670         old_isec = old_dentry->d_inode->i_security;
1671         old_is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
1672         new_dsec = new_dir->i_security;
1673
1674         ad.type = LSM_AUDIT_DATA_DENTRY;
1675
1676         ad.u.dentry = old_dentry;
1677         rc = avc_has_perm(sid, old_dsec->sid, SECCLASS_DIR,
1678                           DIR__REMOVE_NAME | DIR__SEARCH, &ad);
1679         if (rc)
1680                 return rc;
1681         rc = avc_has_perm(sid, old_isec->sid,
1682                           old_isec->sclass, FILE__RENAME, &ad);
1683         if (rc)
1684                 return rc;
1685         if (old_is_dir && new_dir != old_dir) {
1686                 rc = avc_has_perm(sid, old_isec->sid,
1687                                   old_isec->sclass, DIR__REPARENT, &ad);
1688                 if (rc)
1689                         return rc;
1690         }
1691
1692         ad.u.dentry = new_dentry;
1693         av = DIR__ADD_NAME | DIR__SEARCH;
1694         if (new_dentry->d_inode)
1695                 av |= DIR__REMOVE_NAME;
1696         rc = avc_has_perm(sid, new_dsec->sid, SECCLASS_DIR, av, &ad);
1697         if (rc)
1698                 return rc;
1699         if (new_dentry->d_inode) {
1700                 new_isec = new_dentry->d_inode->i_security;
1701                 new_is_dir = S_ISDIR(new_dentry->d_inode->i_mode);
1702                 rc = avc_has_perm(sid, new_isec->sid,
1703                                   new_isec->sclass,
1704                                   (new_is_dir ? DIR__RMDIR : FILE__UNLINK), &ad);
1705                 if (rc)
1706                         return rc;
1707         }
1708
1709         return 0;
1710 }
1711
1712 /* Check whether a task can perform a filesystem operation. */
1713 static int superblock_has_perm(const struct cred *cred,
1714                                struct super_block *sb,
1715                                u32 perms,
1716                                struct common_audit_data *ad)
1717 {
1718         struct superblock_security_struct *sbsec;
1719         u32 sid = cred_sid(cred);
1720
1721         sbsec = sb->s_security;
1722         return avc_has_perm(sid, sbsec->sid, SECCLASS_FILESYSTEM, perms, ad);
1723 }
1724
1725 /* Convert a Linux mode and permission mask to an access vector. */
1726 static inline u32 file_mask_to_av(int mode, int mask)
1727 {
1728         u32 av = 0;
1729
1730         if (!S_ISDIR(mode)) {
1731                 if (mask & MAY_EXEC)
1732                         av |= FILE__EXECUTE;
1733                 if (mask & MAY_READ)
1734                         av |= FILE__READ;
1735
1736                 if (mask & MAY_APPEND)
1737                         av |= FILE__APPEND;
1738                 else if (mask & MAY_WRITE)
1739                         av |= FILE__WRITE;
1740
1741         } else {
1742                 if (mask & MAY_EXEC)
1743                         av |= DIR__SEARCH;
1744                 if (mask & MAY_WRITE)
1745                         av |= DIR__WRITE;
1746                 if (mask & MAY_READ)
1747                         av |= DIR__READ;
1748         }
1749
1750         return av;
1751 }
1752
1753 /* Convert a Linux file to an access vector. */
1754 static inline u32 file_to_av(struct file *file)
1755 {
1756         u32 av = 0;
1757
1758         if (file->f_mode & FMODE_READ)
1759                 av |= FILE__READ;
1760         if (file->f_mode & FMODE_WRITE) {
1761                 if (file->f_flags & O_APPEND)
1762                         av |= FILE__APPEND;
1763                 else
1764                         av |= FILE__WRITE;
1765         }
1766         if (!av) {
1767                 /*
1768                  * Special file opened with flags 3 for ioctl-only use.
1769                  */
1770                 av = FILE__IOCTL;
1771         }
1772
1773         return av;
1774 }
1775
1776 /*
1777  * Convert a file to an access vector and include the correct open
1778  * open permission.
1779  */
1780 static inline u32 open_file_to_av(struct file *file)
1781 {
1782         u32 av = file_to_av(file);
1783
1784         if (selinux_policycap_openperm)
1785                 av |= FILE__OPEN;
1786
1787         return av;
1788 }
1789
1790 /* Hook functions begin here. */
1791
1792 static int selinux_ptrace_access_check(struct task_struct *child,
1793                                      unsigned int mode)
1794 {
1795         int rc;
1796
1797         rc = cap_ptrace_access_check(child, mode);
1798         if (rc)
1799                 return rc;
1800
1801         if (mode & PTRACE_MODE_READ) {
1802                 u32 sid = current_sid();
1803                 u32 csid = task_sid(child);
1804                 return avc_has_perm(sid, csid, SECCLASS_FILE, FILE__READ, NULL);
1805         }
1806
1807         return current_has_perm(child, PROCESS__PTRACE);
1808 }
1809
1810 static int selinux_ptrace_traceme(struct task_struct *parent)
1811 {
1812         int rc;
1813
1814         rc = cap_ptrace_traceme(parent);
1815         if (rc)
1816                 return rc;
1817
1818         return task_has_perm(parent, current, PROCESS__PTRACE);
1819 }
1820
1821 static int selinux_capget(struct task_struct *target, kernel_cap_t *effective,
1822                           kernel_cap_t *inheritable, kernel_cap_t *permitted)
1823 {
1824         int error;
1825
1826         error = current_has_perm(target, PROCESS__GETCAP);
1827         if (error)
1828                 return error;
1829
1830         return cap_capget(target, effective, inheritable, permitted);
1831 }
1832
1833 static int selinux_capset(struct cred *new, const struct cred *old,
1834                           const kernel_cap_t *effective,
1835                           const kernel_cap_t *inheritable,
1836                           const kernel_cap_t *permitted)
1837 {
1838         int error;
1839
1840         error = cap_capset(new, old,
1841                                       effective, inheritable, permitted);
1842         if (error)
1843                 return error;
1844
1845         return cred_has_perm(old, new, PROCESS__SETCAP);
1846 }
1847
1848 /*
1849  * (This comment used to live with the selinux_task_setuid hook,
1850  * which was removed).
1851  *
1852  * Since setuid only affects the current process, and since the SELinux
1853  * controls are not based on the Linux identity attributes, SELinux does not
1854  * need to control this operation.  However, SELinux does control the use of
1855  * the CAP_SETUID and CAP_SETGID capabilities using the capable hook.
1856  */
1857
1858 static int selinux_capable(const struct cred *cred, struct user_namespace *ns,
1859                            int cap, int audit)
1860 {
1861         int rc;
1862
1863         rc = cap_capable(cred, ns, cap, audit);
1864         if (rc)
1865                 return rc;
1866
1867         return cred_has_capability(cred, cap, audit);
1868 }
1869
1870 static int selinux_quotactl(int cmds, int type, int id, struct super_block *sb)
1871 {
1872         const struct cred *cred = current_cred();
1873         int rc = 0;
1874
1875         if (!sb)
1876                 return 0;
1877
1878         switch (cmds) {
1879         case Q_SYNC:
1880         case Q_QUOTAON:
1881         case Q_QUOTAOFF:
1882         case Q_SETINFO:
1883         case Q_SETQUOTA:
1884                 rc = superblock_has_perm(cred, sb, FILESYSTEM__QUOTAMOD, NULL);
1885                 break;
1886         case Q_GETFMT:
1887         case Q_GETINFO:
1888         case Q_GETQUOTA:
1889                 rc = superblock_has_perm(cred, sb, FILESYSTEM__QUOTAGET, NULL);
1890                 break;
1891         default:
1892                 rc = 0;  /* let the kernel handle invalid cmds */
1893                 break;
1894         }
1895         return rc;
1896 }
1897
1898 static int selinux_quota_on(struct dentry *dentry)
1899 {
1900         const struct cred *cred = current_cred();
1901
1902         return dentry_has_perm(cred, dentry, FILE__QUOTAON);
1903 }
1904
1905 static int selinux_syslog(int type)
1906 {
1907         int rc;
1908
1909         switch (type) {
1910         case SYSLOG_ACTION_READ_ALL:    /* Read last kernel messages */
1911         case SYSLOG_ACTION_SIZE_BUFFER: /* Return size of the log buffer */
1912                 rc = task_has_system(current, SYSTEM__SYSLOG_READ);
1913                 break;
1914         case SYSLOG_ACTION_CONSOLE_OFF: /* Disable logging to console */
1915         case SYSLOG_ACTION_CONSOLE_ON:  /* Enable logging to console */
1916         /* Set level of messages printed to console */
1917         case SYSLOG_ACTION_CONSOLE_LEVEL:
1918                 rc = task_has_system(current, SYSTEM__SYSLOG_CONSOLE);
1919                 break;
1920         case SYSLOG_ACTION_CLOSE:       /* Close log */
1921         case SYSLOG_ACTION_OPEN:        /* Open log */
1922         case SYSLOG_ACTION_READ:        /* Read from log */
1923         case SYSLOG_ACTION_READ_CLEAR:  /* Read/clear last kernel messages */
1924         case SYSLOG_ACTION_CLEAR:       /* Clear ring buffer */
1925         default:
1926                 rc = task_has_system(current, SYSTEM__SYSLOG_MOD);
1927                 break;
1928         }
1929         return rc;
1930 }
1931
1932 /*
1933  * Check that a process has enough memory to allocate a new virtual
1934  * mapping. 0 means there is enough memory for the allocation to
1935  * succeed and -ENOMEM implies there is not.
1936  *
1937  * Do not audit the selinux permission check, as this is applied to all
1938  * processes that allocate mappings.
1939  */
1940 static int selinux_vm_enough_memory(struct mm_struct *mm, long pages)
1941 {
1942         int rc, cap_sys_admin = 0;
1943
1944         rc = selinux_capable(current_cred(), &init_user_ns, CAP_SYS_ADMIN,
1945                              SECURITY_CAP_NOAUDIT);
1946         if (rc == 0)
1947                 cap_sys_admin = 1;
1948
1949         return __vm_enough_memory(mm, pages, cap_sys_admin);
1950 }
1951
1952 /* binprm security operations */
1953
1954 static int selinux_bprm_set_creds(struct linux_binprm *bprm)
1955 {
1956         const struct task_security_struct *old_tsec;
1957         struct task_security_struct *new_tsec;
1958         struct inode_security_struct *isec;
1959         struct common_audit_data ad;
1960         struct inode *inode = file_inode(bprm->file);
1961         int rc;
1962
1963         rc = cap_bprm_set_creds(bprm);
1964         if (rc)
1965                 return rc;
1966
1967         /* SELinux context only depends on initial program or script and not
1968          * the script interpreter */
1969         if (bprm->cred_prepared)
1970                 return 0;
1971
1972         old_tsec = current_security();
1973         new_tsec = bprm->cred->security;
1974         isec = inode->i_security;
1975
1976         /* Default to the current task SID. */
1977         new_tsec->sid = old_tsec->sid;
1978         new_tsec->osid = old_tsec->sid;
1979
1980         /* Reset fs, key, and sock SIDs on execve. */
1981         new_tsec->create_sid = 0;
1982         new_tsec->keycreate_sid = 0;
1983         new_tsec->sockcreate_sid = 0;
1984
1985         if (old_tsec->exec_sid) {
1986                 new_tsec->sid = old_tsec->exec_sid;
1987                 /* Reset exec SID on execve. */
1988                 new_tsec->exec_sid = 0;
1989
1990                 /*
1991                  * Minimize confusion: if no_new_privs and a transition is
1992                  * explicitly requested, then fail the exec.
1993                  */
1994                 if (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)
1995                         return -EPERM;
1996         } else {
1997                 /* Check for a default transition on this program. */
1998                 rc = security_transition_sid(old_tsec->sid, isec->sid,
1999                                              SECCLASS_PROCESS, NULL,
2000                                              &new_tsec->sid);
2001                 if (rc)
2002                         return rc;
2003         }
2004
2005         ad.type = LSM_AUDIT_DATA_PATH;
2006         ad.u.path = bprm->file->f_path;
2007
2008         if ((bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) ||
2009             (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS))
2010                 new_tsec->sid = old_tsec->sid;
2011
2012         if (new_tsec->sid == old_tsec->sid) {
2013                 rc = avc_has_perm(old_tsec->sid, isec->sid,
2014                                   SECCLASS_FILE, FILE__EXECUTE_NO_TRANS, &ad);
2015                 if (rc)
2016                         return rc;
2017         } else {
2018                 /* Check permissions for the transition. */
2019                 rc = avc_has_perm(old_tsec->sid, new_tsec->sid,
2020                                   SECCLASS_PROCESS, PROCESS__TRANSITION, &ad);
2021                 if (rc)
2022                         return rc;
2023
2024                 rc = avc_has_perm(new_tsec->sid, isec->sid,
2025                                   SECCLASS_FILE, FILE__ENTRYPOINT, &ad);
2026                 if (rc)
2027                         return rc;
2028
2029                 /* Check for shared state */
2030                 if (bprm->unsafe & LSM_UNSAFE_SHARE) {
2031                         rc = avc_has_perm(old_tsec->sid, new_tsec->sid,
2032                                           SECCLASS_PROCESS, PROCESS__SHARE,
2033                                           NULL);
2034                         if (rc)
2035                                 return -EPERM;
2036                 }
2037
2038                 /* Make sure that anyone attempting to ptrace over a task that
2039                  * changes its SID has the appropriate permit */
2040                 if (bprm->unsafe &
2041                     (LSM_UNSAFE_PTRACE | LSM_UNSAFE_PTRACE_CAP)) {
2042                         struct task_struct *tracer;
2043                         struct task_security_struct *sec;
2044                         u32 ptsid = 0;
2045
2046                         rcu_read_lock();
2047                         tracer = ptrace_parent(current);
2048                         if (likely(tracer != NULL)) {
2049                                 sec = __task_cred(tracer)->security;
2050                                 ptsid = sec->sid;
2051                         }
2052                         rcu_read_unlock();
2053
2054                         if (ptsid != 0) {
2055                                 rc = avc_has_perm(ptsid, new_tsec->sid,
2056                                                   SECCLASS_PROCESS,
2057                                                   PROCESS__PTRACE, NULL);
2058                                 if (rc)
2059                                         return -EPERM;
2060                         }
2061                 }
2062
2063                 /* Clear any possibly unsafe personality bits on exec: */
2064                 bprm->per_clear |= PER_CLEAR_ON_SETID;
2065         }
2066
2067         return 0;
2068 }
2069
2070 static int selinux_bprm_secureexec(struct linux_binprm *bprm)
2071 {
2072         const struct task_security_struct *tsec = current_security();
2073         u32 sid, osid;
2074         int atsecure = 0;
2075
2076         sid = tsec->sid;
2077         osid = tsec->osid;
2078
2079         if (osid != sid) {
2080                 /* Enable secure mode for SIDs transitions unless
2081                    the noatsecure permission is granted between
2082                    the two SIDs, i.e. ahp returns 0. */
2083                 atsecure = avc_has_perm(osid, sid,
2084                                         SECCLASS_PROCESS,
2085                                         PROCESS__NOATSECURE, NULL);
2086         }
2087
2088         return (atsecure || cap_bprm_secureexec(bprm));
2089 }
2090
2091 static int match_file(const void *p, struct file *file, unsigned fd)
2092 {
2093         return file_has_perm(p, file, file_to_av(file)) ? fd + 1 : 0;
2094 }
2095
2096 /* Derived from fs/exec.c:flush_old_files. */
2097 static inline void flush_unauthorized_files(const struct cred *cred,
2098                                             struct files_struct *files)
2099 {
2100         struct file *file, *devnull = NULL;
2101         struct tty_struct *tty;
2102         int drop_tty = 0;
2103         unsigned n;
2104
2105         tty = get_current_tty();
2106         if (tty) {
2107                 spin_lock(&tty_files_lock);
2108                 if (!list_empty(&tty->tty_files)) {
2109                         struct tty_file_private *file_priv;
2110
2111                         /* Revalidate access to controlling tty.
2112                            Use path_has_perm on the tty path directly rather
2113                            than using file_has_perm, as this particular open
2114                            file may belong to another process and we are only
2115                            interested in the inode-based check here. */
2116                         file_priv = list_first_entry(&tty->tty_files,
2117                                                 struct tty_file_private, list);
2118                         file = file_priv->file;
2119                         if (path_has_perm(cred, &file->f_path, FILE__READ | FILE__WRITE))
2120                                 drop_tty = 1;
2121                 }
2122                 spin_unlock(&tty_files_lock);
2123                 tty_kref_put(tty);
2124         }
2125         /* Reset controlling tty. */
2126         if (drop_tty)
2127                 no_tty();
2128
2129         /* Revalidate access to inherited open files. */
2130         n = iterate_fd(files, 0, match_file, cred);
2131         if (!n) /* none found? */
2132                 return;
2133
2134         devnull = dentry_open(&selinux_null, O_RDWR, cred);
2135         if (IS_ERR(devnull))
2136                 devnull = NULL;
2137         /* replace all the matching ones with this */
2138         do {
2139                 replace_fd(n - 1, devnull, 0);
2140         } while ((n = iterate_fd(files, n, match_file, cred)) != 0);
2141         if (devnull)
2142                 fput(devnull);
2143 }
2144
2145 /*
2146  * Prepare a process for imminent new credential changes due to exec
2147  */
2148 static void selinux_bprm_committing_creds(struct linux_binprm *bprm)
2149 {
2150         struct task_security_struct *new_tsec;
2151         struct rlimit *rlim, *initrlim;
2152         int rc, i;
2153
2154         new_tsec = bprm->cred->security;
2155         if (new_tsec->sid == new_tsec->osid)
2156                 return;
2157
2158         /* Close files for which the new task SID is not authorized. */
2159         flush_unauthorized_files(bprm->cred, current->files);
2160
2161         /* Always clear parent death signal on SID transitions. */
2162         current->pdeath_signal = 0;
2163
2164         /* Check whether the new SID can inherit resource limits from the old
2165          * SID.  If not, reset all soft limits to the lower of the current
2166          * task's hard limit and the init task's soft limit.
2167          *
2168          * Note that the setting of hard limits (even to lower them) can be
2169          * controlled by the setrlimit check.  The inclusion of the init task's
2170          * soft limit into the computation is to avoid resetting soft limits
2171          * higher than the default soft limit for cases where the default is
2172          * lower than the hard limit, e.g. RLIMIT_CORE or RLIMIT_STACK.
2173          */
2174         rc = avc_has_perm(new_tsec->osid, new_tsec->sid, SECCLASS_PROCESS,
2175                           PROCESS__RLIMITINH, NULL);
2176         if (rc) {
2177                 /* protect against do_prlimit() */
2178                 task_lock(current);
2179                 for (i = 0; i < RLIM_NLIMITS; i++) {
2180                         rlim = current->signal->rlim + i;
2181                         initrlim = init_task.signal->rlim + i;
2182                         rlim->rlim_cur = min(rlim->rlim_max, initrlim->rlim_cur);
2183                 }
2184                 task_unlock(current);
2185                 update_rlimit_cpu(current, rlimit(RLIMIT_CPU));
2186         }
2187 }
2188
2189 /*
2190  * Clean up the process immediately after the installation of new credentials
2191  * due to exec
2192  */
2193 static void selinux_bprm_committed_creds(struct linux_binprm *bprm)
2194 {
2195         const struct task_security_struct *tsec = current_security();
2196         struct itimerval itimer;
2197         u32 osid, sid;
2198         int rc, i;
2199
2200         osid = tsec->osid;
2201         sid = tsec->sid;
2202
2203         if (sid == osid)
2204                 return;
2205
2206         /* Check whether the new SID can inherit signal state from the old SID.
2207          * If not, clear itimers to avoid subsequent signal generation and
2208          * flush and unblock signals.
2209          *
2210          * This must occur _after_ the task SID has been updated so that any
2211          * kill done after the flush will be checked against the new SID.
2212          */
2213         rc = avc_has_perm(osid, sid, SECCLASS_PROCESS, PROCESS__SIGINH, NULL);
2214         if (rc) {
2215                 memset(&itimer, 0, sizeof itimer);
2216                 for (i = 0; i < 3; i++)
2217                         do_setitimer(i, &itimer, NULL);
2218                 spin_lock_irq(&current->sighand->siglock);
2219                 if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
2220                         __flush_signals(current);
2221                         flush_signal_handlers(current, 1);
2222                         sigemptyset(&current->blocked);
2223                 }
2224                 spin_unlock_irq(&current->sighand->siglock);
2225         }
2226
2227         /* Wake up the parent if it is waiting so that it can recheck
2228          * wait permission to the new task SID. */
2229         read_lock(&tasklist_lock);
2230         __wake_up_parent(current, current->real_parent);
2231         read_unlock(&tasklist_lock);
2232 }
2233
2234 /* superblock security operations */
2235
2236 static int selinux_sb_alloc_security(struct super_block *sb)
2237 {
2238         return superblock_alloc_security(sb);
2239 }
2240
2241 static void selinux_sb_free_security(struct super_block *sb)
2242 {
2243         superblock_free_security(sb);
2244 }
2245
2246 static inline int match_prefix(char *prefix, int plen, char *option, int olen)
2247 {
2248         if (plen > olen)
2249                 return 0;
2250
2251         return !memcmp(prefix, option, plen);
2252 }
2253
2254 static inline int selinux_option(char *option, int len)
2255 {
2256         return (match_prefix(CONTEXT_STR, sizeof(CONTEXT_STR)-1, option, len) ||
2257                 match_prefix(FSCONTEXT_STR, sizeof(FSCONTEXT_STR)-1, option, len) ||
2258                 match_prefix(DEFCONTEXT_STR, sizeof(DEFCONTEXT_STR)-1, option, len) ||
2259                 match_prefix(ROOTCONTEXT_STR, sizeof(ROOTCONTEXT_STR)-1, option, len) ||
2260                 match_prefix(LABELSUPP_STR, sizeof(LABELSUPP_STR)-1, option, len));
2261 }
2262
2263 static inline void take_option(char **to, char *from, int *first, int len)
2264 {
2265         if (!*first) {
2266                 **to = ',';
2267                 *to += 1;
2268         } else
2269                 *first = 0;
2270         memcpy(*to, from, len);
2271         *to += len;
2272 }
2273
2274 static inline void take_selinux_option(char **to, char *from, int *first,
2275                                        int len)
2276 {
2277         int current_size = 0;
2278
2279         if (!*first) {
2280                 **to = '|';
2281                 *to += 1;
2282         } else
2283                 *first = 0;
2284
2285         while (current_size < len) {
2286                 if (*from != '"') {
2287                         **to = *from;
2288                         *to += 1;
2289                 }
2290                 from += 1;
2291                 current_size += 1;
2292         }
2293 }
2294
2295 static int selinux_sb_copy_data(char *orig, char *copy)
2296 {
2297         int fnosec, fsec, rc = 0;
2298         char *in_save, *in_curr, *in_end;
2299         char *sec_curr, *nosec_save, *nosec;
2300         int open_quote = 0;
2301
2302         in_curr = orig;
2303         sec_curr = copy;
2304
2305         nosec = (char *)get_zeroed_page(GFP_KERNEL);
2306         if (!nosec) {
2307                 rc = -ENOMEM;
2308                 goto out;
2309         }
2310
2311         nosec_save = nosec;
2312         fnosec = fsec = 1;
2313         in_save = in_end = orig;
2314
2315         do {
2316                 if (*in_end == '"')
2317                         open_quote = !open_quote;
2318                 if ((*in_end == ',' && open_quote == 0) ||
2319                                 *in_end == '\0') {
2320                         int len = in_end - in_curr;
2321
2322                         if (selinux_option(in_curr, len))
2323                                 take_selinux_option(&sec_curr, in_curr, &fsec, len);
2324                         else
2325                                 take_option(&nosec, in_curr, &fnosec, len);
2326
2327                         in_curr = in_end + 1;
2328                 }
2329         } while (*in_end++);
2330
2331         strcpy(in_save, nosec_save);
2332         free_page((unsigned long)nosec_save);
2333 out:
2334         return rc;
2335 }
2336
2337 static int selinux_sb_remount(struct super_block *sb, void *data)
2338 {
2339         int rc, i, *flags;
2340         struct security_mnt_opts opts;
2341         char *secdata, **mount_options;
2342         struct superblock_security_struct *sbsec = sb->s_security;
2343
2344         if (!(sbsec->flags & SE_SBINITIALIZED))
2345                 return 0;
2346
2347         if (!data)
2348                 return 0;
2349
2350         if (sb->s_type->fs_flags & FS_BINARY_MOUNTDATA)
2351                 return 0;
2352
2353         security_init_mnt_opts(&opts);
2354         secdata = alloc_secdata();
2355         if (!secdata)
2356                 return -ENOMEM;
2357         rc = selinux_sb_copy_data(data, secdata);
2358         if (rc)
2359                 goto out_free_secdata;
2360
2361         rc = selinux_parse_opts_str(secdata, &opts);
2362         if (rc)
2363                 goto out_free_secdata;
2364
2365         mount_options = opts.mnt_opts;
2366         flags = opts.mnt_opts_flags;
2367
2368         for (i = 0; i < opts.num_mnt_opts; i++) {
2369                 u32 sid;
2370                 size_t len;
2371
2372                 if (flags[i] == SE_SBLABELSUPP)
2373                         continue;
2374                 len = strlen(mount_options[i]);
2375                 rc = security_context_to_sid(mount_options[i], len, &sid);
2376                 if (rc) {
2377                         printk(KERN_WARNING "SELinux: security_context_to_sid"
2378                                "(%s) failed for (dev %s, type %s) errno=%d\n",
2379                                mount_options[i], sb->s_id, sb->s_type->name, rc);
2380                         goto out_free_opts;
2381                 }
2382                 rc = -EINVAL;
2383                 switch (flags[i]) {
2384                 case FSCONTEXT_MNT:
2385                         if (bad_option(sbsec, FSCONTEXT_MNT, sbsec->sid, sid))
2386                                 goto out_bad_option;
2387                         break;
2388                 case CONTEXT_MNT:
2389                         if (bad_option(sbsec, CONTEXT_MNT, sbsec->mntpoint_sid, sid))
2390                                 goto out_bad_option;
2391                         break;
2392                 case ROOTCONTEXT_MNT: {
2393                         struct inode_security_struct *root_isec;
2394                         root_isec = sb->s_root->d_inode->i_security;
2395
2396                         if (bad_option(sbsec, ROOTCONTEXT_MNT, root_isec->sid, sid))
2397                                 goto out_bad_option;
2398                         break;
2399                 }
2400                 case DEFCONTEXT_MNT:
2401                         if (bad_option(sbsec, DEFCONTEXT_MNT, sbsec->def_sid, sid))
2402                                 goto out_bad_option;
2403                         break;
2404                 default:
2405                         goto out_free_opts;
2406                 }
2407         }
2408
2409         rc = 0;
2410 out_free_opts:
2411         security_free_mnt_opts(&opts);
2412 out_free_secdata:
2413         free_secdata(secdata);
2414         return rc;
2415 out_bad_option:
2416         printk(KERN_WARNING "SELinux: unable to change security options "
2417                "during remount (dev %s, type=%s)\n", sb->s_id,
2418                sb->s_type->name);
2419         goto out_free_opts;
2420 }
2421
2422 static int selinux_sb_kern_mount(struct super_block *sb, int flags, void *data)
2423 {
2424         const struct cred *cred = current_cred();
2425         struct common_audit_data ad;
2426         int rc;
2427
2428         rc = superblock_doinit(sb, data);
2429         if (rc)
2430                 return rc;
2431
2432         /* Allow all mounts performed by the kernel */
2433         if (flags & MS_KERNMOUNT)
2434                 return 0;
2435
2436         ad.type = LSM_AUDIT_DATA_DENTRY;
2437         ad.u.dentry = sb->s_root;
2438         return superblock_has_perm(cred, sb, FILESYSTEM__MOUNT, &ad);
2439 }
2440
2441 static int selinux_sb_statfs(struct dentry *dentry)
2442 {
2443         const struct cred *cred = current_cred();
2444         struct common_audit_data ad;
2445
2446         ad.type = LSM_AUDIT_DATA_DENTRY;
2447         ad.u.dentry = dentry->d_sb->s_root;
2448         return superblock_has_perm(cred, dentry->d_sb, FILESYSTEM__GETATTR, &ad);
2449 }
2450
2451 static int selinux_mount(const char *dev_name,
2452                          struct path *path,
2453                          const char *type,
2454                          unsigned long flags,
2455                          void *data)
2456 {
2457         const struct cred *cred = current_cred();
2458
2459         if (flags & MS_REMOUNT)
2460                 return superblock_has_perm(cred, path->dentry->d_sb,
2461                                            FILESYSTEM__REMOUNT, NULL);
2462         else
2463                 return path_has_perm(cred, path, FILE__MOUNTON);
2464 }
2465
2466 static int selinux_umount(struct vfsmount *mnt, int flags)
2467 {
2468         const struct cred *cred = current_cred();
2469
2470         return superblock_has_perm(cred, mnt->mnt_sb,
2471                                    FILESYSTEM__UNMOUNT, NULL);
2472 }
2473
2474 /* inode security operations */
2475
2476 static int selinux_inode_alloc_security(struct inode *inode)
2477 {
2478         return inode_alloc_security(inode);
2479 }
2480
2481 static void selinux_inode_free_security(struct inode *inode)
2482 {
2483         inode_free_security(inode);
2484 }
2485
2486 static int selinux_inode_init_security(struct inode *inode, struct inode *dir,
2487                                        const struct qstr *qstr, char **name,
2488                                        void **value, size_t *len)
2489 {
2490         const struct task_security_struct *tsec = current_security();
2491         struct inode_security_struct *dsec;
2492         struct superblock_security_struct *sbsec;
2493         u32 sid, newsid, clen;
2494         int rc;
2495         char *namep = NULL, *context;
2496
2497         dsec = dir->i_security;
2498         sbsec = dir->i_sb->s_security;
2499
2500         sid = tsec->sid;
2501         newsid = tsec->create_sid;
2502
2503         if ((sbsec->flags & SE_SBINITIALIZED) &&
2504             (sbsec->behavior == SECURITY_FS_USE_MNTPOINT))
2505                 newsid = sbsec->mntpoint_sid;
2506         else if (!newsid || !(sbsec->flags & SE_SBLABELSUPP)) {
2507                 rc = security_transition_sid(sid, dsec->sid,
2508                                              inode_mode_to_security_class(inode->i_mode),
2509                                              qstr, &newsid);
2510                 if (rc) {
2511                         printk(KERN_WARNING "%s:  "
2512                                "security_transition_sid failed, rc=%d (dev=%s "
2513                                "ino=%ld)\n",
2514                                __func__,
2515                                -rc, inode->i_sb->s_id, inode->i_ino);
2516                         return rc;
2517                 }
2518         }
2519
2520         /* Possibly defer initialization to selinux_complete_init. */
2521         if (sbsec->flags & SE_SBINITIALIZED) {
2522                 struct inode_security_struct *isec = inode->i_security;
2523                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
2524                 isec->sid = newsid;
2525                 isec->initialized = 1;
2526         }
2527
2528         if (!ss_initialized || !(sbsec->flags & SE_SBLABELSUPP))
2529                 return -EOPNOTSUPP;
2530
2531         if (name) {
2532                 namep = kstrdup(XATTR_SELINUX_SUFFIX, GFP_NOFS);
2533                 if (!namep)
2534                         return -ENOMEM;
2535                 *name = namep;
2536         }
2537
2538         if (value && len) {
2539                 rc = security_sid_to_context_force(newsid, &context, &clen);
2540                 if (rc) {
2541                         kfree(namep);
2542                         return rc;
2543                 }
2544                 *value = context;
2545                 *len = clen;
2546         }
2547
2548         return 0;
2549 }
2550
2551 static int selinux_inode_create(struct inode *dir, struct dentry *dentry, umode_t mode)
2552 {
2553         return may_create(dir, dentry, SECCLASS_FILE);
2554 }
2555
2556 static int selinux_inode_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2557 {
2558         return may_link(dir, old_dentry, MAY_LINK);
2559 }
2560
2561 static int selinux_inode_unlink(struct inode *dir, struct dentry *dentry)
2562 {
2563         return may_link(dir, dentry, MAY_UNLINK);
2564 }
2565
2566 static int selinux_inode_symlink(struct inode *dir, struct dentry *dentry, const char *name)
2567 {
2568         return may_create(dir, dentry, SECCLASS_LNK_FILE);
2569 }
2570
2571 static int selinux_inode_mkdir(struct inode *dir, struct dentry *dentry, umode_t mask)
2572 {
2573         return may_create(dir, dentry, SECCLASS_DIR);
2574 }
2575
2576 static int selinux_inode_rmdir(struct inode *dir, struct dentry *dentry)
2577 {
2578         return may_link(dir, dentry, MAY_RMDIR);
2579 }
2580
2581 static int selinux_inode_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2582 {
2583         return may_create(dir, dentry, inode_mode_to_security_class(mode));
2584 }
2585
2586 static int selinux_inode_rename(struct inode *old_inode, struct dentry *old_dentry,
2587                                 struct inode *new_inode, struct dentry *new_dentry)
2588 {
2589         return may_rename(old_inode, old_dentry, new_inode, new_dentry);
2590 }
2591
2592 static int selinux_inode_readlink(struct dentry *dentry)
2593 {
2594         const struct cred *cred = current_cred();
2595
2596         return dentry_has_perm(cred, dentry, FILE__READ);
2597 }
2598
2599 static int selinux_inode_follow_link(struct dentry *dentry, struct nameidata *nameidata)
2600 {
2601         const struct cred *cred = current_cred();
2602
2603         return dentry_has_perm(cred, dentry, FILE__READ);
2604 }
2605
2606 static noinline int audit_inode_permission(struct inode *inode,
2607                                            u32 perms, u32 audited, u32 denied,
2608                                            unsigned flags)
2609 {
2610         struct common_audit_data ad;
2611         struct inode_security_struct *isec = inode->i_security;
2612         int rc;
2613
2614         ad.type = LSM_AUDIT_DATA_INODE;
2615         ad.u.inode = inode;
2616
2617         rc = slow_avc_audit(current_sid(), isec->sid, isec->sclass, perms,
2618                             audited, denied, &ad, flags);
2619         if (rc)
2620                 return rc;
2621         return 0;
2622 }
2623
2624 static int selinux_inode_permission(struct inode *inode, int mask)
2625 {
2626         const struct cred *cred = current_cred();
2627         u32 perms;
2628         bool from_access;
2629         unsigned flags = mask & MAY_NOT_BLOCK;
2630         struct inode_security_struct *isec;
2631         u32 sid;
2632         struct av_decision avd;
2633         int rc, rc2;
2634         u32 audited, denied;
2635
2636         from_access = mask & MAY_ACCESS;
2637         mask &= (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND);
2638
2639         /* No permission to check.  Existence test. */
2640         if (!mask)
2641                 return 0;
2642
2643         validate_creds(cred);
2644
2645         if (unlikely(IS_PRIVATE(inode)))
2646                 return 0;
2647
2648         perms = file_mask_to_av(inode->i_mode, mask);
2649
2650         sid = cred_sid(cred);
2651         isec = inode->i_security;
2652
2653         rc = avc_has_perm_noaudit(sid, isec->sid, isec->sclass, perms, 0, &avd);
2654         audited = avc_audit_required(perms, &avd, rc,
2655                                      from_access ? FILE__AUDIT_ACCESS : 0,
2656                                      &denied);
2657         if (likely(!audited))
2658                 return rc;
2659
2660         rc2 = audit_inode_permission(inode, perms, audited, denied, flags);
2661         if (rc2)
2662                 return rc2;
2663         return rc;
2664 }
2665
2666 static int selinux_inode_setattr(struct dentry *dentry, struct iattr *iattr)
2667 {
2668         const struct cred *cred = current_cred();
2669         unsigned int ia_valid = iattr->ia_valid;
2670         __u32 av = FILE__WRITE;
2671
2672         /* ATTR_FORCE is just used for ATTR_KILL_S[UG]ID. */
2673         if (ia_valid & ATTR_FORCE) {
2674                 ia_valid &= ~(ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_MODE |
2675                               ATTR_FORCE);
2676                 if (!ia_valid)
2677                         return 0;
2678         }
2679
2680         if (ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID |
2681                         ATTR_ATIME_SET | ATTR_MTIME_SET | ATTR_TIMES_SET))
2682                 return dentry_has_perm(cred, dentry, FILE__SETATTR);
2683
2684         if (selinux_policycap_openperm && (ia_valid & ATTR_SIZE))
2685                 av |= FILE__OPEN;
2686
2687         return dentry_has_perm(cred, dentry, av);
2688 }
2689
2690 static int selinux_inode_getattr(struct vfsmount *mnt, struct dentry *dentry)
2691 {
2692         const struct cred *cred = current_cred();
2693         struct path path;
2694
2695         path.dentry = dentry;
2696         path.mnt = mnt;
2697
2698         return path_has_perm(cred, &path, FILE__GETATTR);
2699 }
2700
2701 static int selinux_inode_setotherxattr(struct dentry *dentry, const char *name)
2702 {
2703         const struct cred *cred = current_cred();
2704
2705         if (!strncmp(name, XATTR_SECURITY_PREFIX,
2706                      sizeof XATTR_SECURITY_PREFIX - 1)) {
2707                 if (!strcmp(name, XATTR_NAME_CAPS)) {
2708                         if (!capable(CAP_SETFCAP))
2709                                 return -EPERM;
2710                 } else if (!capable(CAP_SYS_ADMIN)) {
2711                         /* A different attribute in the security namespace.
2712                            Restrict to administrator. */
2713                         return -EPERM;
2714                 }
2715         }
2716
2717         /* Not an attribute we recognize, so just check the
2718            ordinary setattr permission. */
2719         return dentry_has_perm(cred, dentry, FILE__SETATTR);
2720 }
2721
2722 static int selinux_inode_setxattr(struct dentry *dentry, const char *name,
2723                                   const void *value, size_t size, int flags)
2724 {
2725         struct inode *inode = dentry->d_inode;
2726         struct inode_security_struct *isec = inode->i_security;
2727         struct superblock_security_struct *sbsec;
2728         struct common_audit_data ad;
2729         u32 newsid, sid = current_sid();
2730         int rc = 0;
2731
2732         if (strcmp(name, XATTR_NAME_SELINUX))
2733                 return selinux_inode_setotherxattr(dentry, name);
2734
2735         sbsec = inode->i_sb->s_security;
2736         if (!(sbsec->flags & SE_SBLABELSUPP))
2737                 return -EOPNOTSUPP;
2738
2739         if (!inode_owner_or_capable(inode))
2740                 return -EPERM;
2741
2742         ad.type = LSM_AUDIT_DATA_DENTRY;
2743         ad.u.dentry = dentry;
2744
2745         rc = avc_has_perm(sid, isec->sid, isec->sclass,
2746                           FILE__RELABELFROM, &ad);
2747         if (rc)
2748                 return rc;
2749
2750         rc = security_context_to_sid(value, size, &newsid);
2751         if (rc == -EINVAL) {
2752                 if (!capable(CAP_MAC_ADMIN)) {
2753                         struct audit_buffer *ab;
2754                         size_t audit_size;
2755                         const char *str;
2756
2757                         /* We strip a nul only if it is at the end, otherwise the
2758                          * context contains a nul and we should audit that */
2759                         if (value) {
2760                                 str = value;
2761                                 if (str[size - 1] == '\0')
2762                                         audit_size = size - 1;
2763                                 else
2764                                         audit_size = size;
2765                         } else {
2766                                 str = "";
2767                                 audit_size = 0;
2768                         }
2769                         ab = audit_log_start(current->audit_context, GFP_ATOMIC, AUDIT_SELINUX_ERR);
2770                         audit_log_format(ab, "op=setxattr invalid_context=");
2771                         audit_log_n_untrustedstring(ab, value, audit_size);
2772                         audit_log_end(ab);
2773
2774                         return rc;
2775                 }
2776                 rc = security_context_to_sid_force(value, size, &newsid);
2777         }
2778         if (rc)
2779                 return rc;
2780
2781         rc = avc_has_perm(sid, newsid, isec->sclass,
2782                           FILE__RELABELTO, &ad);
2783         if (rc)
2784                 return rc;
2785
2786         rc = security_validate_transition(isec->sid, newsid, sid,
2787                                           isec->sclass);
2788         if (rc)
2789                 return rc;
2790
2791         return avc_has_perm(newsid,
2792                             sbsec->sid,
2793                             SECCLASS_FILESYSTEM,
2794                             FILESYSTEM__ASSOCIATE,
2795                             &ad);
2796 }
2797
2798 static void selinux_inode_post_setxattr(struct dentry *dentry, const char *name,
2799                                         const void *value, size_t size,
2800                                         int flags)
2801 {
2802         struct inode *inode = dentry->d_inode;
2803         struct inode_security_struct *isec = inode->i_security;
2804         u32 newsid;
2805         int rc;
2806
2807         if (strcmp(name, XATTR_NAME_SELINUX)) {
2808                 /* Not an attribute we recognize, so nothing to do. */
2809                 return;
2810         }
2811
2812         rc = security_context_to_sid_force(value, size, &newsid);
2813         if (rc) {
2814                 printk(KERN_ERR "SELinux:  unable to map context to SID"
2815                        "for (%s, %lu), rc=%d\n",
2816                        inode->i_sb->s_id, inode->i_ino, -rc);
2817                 return;
2818         }
2819
2820         isec->sid = newsid;
2821         return;
2822 }
2823
2824 static int selinux_inode_getxattr(struct dentry *dentry, const char *name)
2825 {
2826         const struct cred *cred = current_cred();
2827
2828         return dentry_has_perm(cred, dentry, FILE__GETATTR);
2829 }
2830
2831 static int selinux_inode_listxattr(struct dentry *dentry)
2832 {
2833         const struct cred *cred = current_cred();
2834
2835         return dentry_has_perm(cred, dentry, FILE__GETATTR);
2836 }
2837
2838 static int selinux_inode_removexattr(struct dentry *dentry, const char *name)
2839 {
2840         if (strcmp(name, XATTR_NAME_SELINUX))
2841                 return selinux_inode_setotherxattr(dentry, name);
2842
2843         /* No one is allowed to remove a SELinux security label.
2844            You can change the label, but all data must be labeled. */
2845         return -EACCES;
2846 }
2847
2848 /*
2849  * Copy the inode security context value to the user.
2850  *
2851  * Permission check is handled by selinux_inode_getxattr hook.
2852  */
2853 static int selinux_inode_getsecurity(const struct inode *inode, const char *name, void **buffer, bool alloc)
2854 {
2855         u32 size;
2856         int error;
2857         char *context = NULL;
2858         struct inode_security_struct *isec = inode->i_security;
2859
2860         if (strcmp(name, XATTR_SELINUX_SUFFIX))
2861                 return -EOPNOTSUPP;
2862
2863         /*
2864          * If the caller has CAP_MAC_ADMIN, then get the raw context
2865          * value even if it is not defined by current policy; otherwise,
2866          * use the in-core value under current policy.
2867          * Use the non-auditing forms of the permission checks since
2868          * getxattr may be called by unprivileged processes commonly
2869          * and lack of permission just means that we fall back to the
2870          * in-core context value, not a denial.
2871          */
2872         error = selinux_capable(current_cred(), &init_user_ns, CAP_MAC_ADMIN,
2873                                 SECURITY_CAP_NOAUDIT);
2874         if (!error)
2875                 error = security_sid_to_context_force(isec->sid, &context,
2876                                                       &size);
2877         else
2878                 error = security_sid_to_context(isec->sid, &context, &size);
2879         if (error)
2880                 return error;
2881         error = size;
2882         if (alloc) {
2883                 *buffer = context;
2884                 goto out_nofree;
2885         }
2886         kfree(context);
2887 out_nofree:
2888         return error;
2889 }
2890
2891 static int selinux_inode_setsecurity(struct inode *inode, const char *name,
2892                                      const void *value, size_t size, int flags)
2893 {
2894         struct inode_security_struct *isec = inode->i_security;
2895         u32 newsid;
2896         int rc;
2897
2898         if (strcmp(name, XATTR_SELINUX_SUFFIX))
2899                 return -EOPNOTSUPP;
2900
2901         if (!value || !size)
2902                 return -EACCES;
2903
2904         rc = security_context_to_sid((void *)value, size, &newsid);
2905         if (rc)
2906                 return rc;
2907
2908         isec->sid = newsid;
2909         isec->initialized = 1;
2910         return 0;
2911 }
2912
2913 static int selinux_inode_listsecurity(struct inode *inode, char *buffer, size_t buffer_size)
2914 {
2915         const int len = sizeof(XATTR_NAME_SELINUX);
2916         if (buffer && len <= buffer_size)
2917                 memcpy(buffer, XATTR_NAME_SELINUX, len);
2918         return len;
2919 }
2920
2921 static void selinux_inode_getsecid(const struct inode *inode, u32 *secid)
2922 {
2923         struct inode_security_struct *isec = inode->i_security;
2924         *secid = isec->sid;
2925 }
2926
2927 /* file security operations */
2928
2929 static int selinux_revalidate_file_permission(struct file *file, int mask)
2930 {
2931         const struct cred *cred = current_cred();
2932         struct inode *inode = file_inode(file);
2933
2934         /* file_mask_to_av won't add FILE__WRITE if MAY_APPEND is set */
2935         if ((file->f_flags & O_APPEND) && (mask & MAY_WRITE))
2936                 mask |= MAY_APPEND;
2937
2938         return file_has_perm(cred, file,
2939                              file_mask_to_av(inode->i_mode, mask));
2940 }
2941
2942 static int selinux_file_permission(struct file *file, int mask)
2943 {
2944         struct inode *inode = file_inode(file);
2945         struct file_security_struct *fsec = file->f_security;
2946         struct inode_security_struct *isec = inode->i_security;
2947         u32 sid = current_sid();
2948
2949         if (!mask)
2950                 /* No permission to check.  Existence test. */
2951                 return 0;
2952
2953         if (sid == fsec->sid && fsec->isid == isec->sid &&
2954             fsec->pseqno == avc_policy_seqno())
2955                 /* No change since file_open check. */
2956                 return 0;
2957
2958         return selinux_revalidate_file_permission(file, mask);
2959 }
2960
2961 static int selinux_file_alloc_security(struct file *file)
2962 {
2963         return file_alloc_security(file);
2964 }
2965
2966 static void selinux_file_free_security(struct file *file)
2967 {
2968         file_free_security(file);
2969 }
2970
2971 static int selinux_file_ioctl(struct file *file, unsigned int cmd,
2972                               unsigned long arg)
2973 {
2974         const struct cred *cred = current_cred();
2975         int error = 0;
2976
2977         switch (cmd) {
2978         case FIONREAD:
2979         /* fall through */
2980         case FIBMAP:
2981         /* fall through */
2982         case FIGETBSZ:
2983         /* fall through */
2984         case FS_IOC_GETFLAGS:
2985         /* fall through */
2986         case FS_IOC_GETVERSION:
2987                 error = file_has_perm(cred, file, FILE__GETATTR);
2988                 break;
2989
2990         case FS_IOC_SETFLAGS:
2991         /* fall through */
2992         case FS_IOC_SETVERSION:
2993                 error = file_has_perm(cred, file, FILE__SETATTR);
2994                 break;
2995
2996         /* sys_ioctl() checks */
2997         case FIONBIO:
2998         /* fall through */
2999         case FIOASYNC:
3000                 error = file_has_perm(cred, file, 0);
3001                 break;
3002
3003         case KDSKBENT:
3004         case KDSKBSENT:
3005                 error = cred_has_capability(cred, CAP_SYS_TTY_CONFIG,
3006                                             SECURITY_CAP_AUDIT);
3007                 break;
3008
3009         /* default case assumes that the command will go
3010          * to the file's ioctl() function.
3011          */
3012         default:
3013                 error = file_has_perm(cred, file, FILE__IOCTL);
3014         }
3015         return error;
3016 }
3017
3018 static int default_noexec;
3019
3020 static int file_map_prot_check(struct file *file, unsigned long prot, int shared)
3021 {
3022         const struct cred *cred = current_cred();
3023         int rc = 0;
3024
3025         if (default_noexec &&
3026             (prot & PROT_EXEC) && (!file || (!shared && (prot & PROT_WRITE)))) {
3027                 /*
3028                  * We are making executable an anonymous mapping or a
3029                  * private file mapping that will also be writable.
3030                  * This has an additional check.
3031                  */
3032                 rc = cred_has_perm(cred, cred, PROCESS__EXECMEM);
3033                 if (rc)
3034                         goto error;
3035         }
3036
3037         if (file) {
3038                 /* read access is always possible with a mapping */
3039                 u32 av = FILE__READ;
3040
3041                 /* write access only matters if the mapping is shared */
3042                 if (shared && (prot & PROT_WRITE))
3043                         av |= FILE__WRITE;
3044
3045                 if (prot & PROT_EXEC)
3046                         av |= FILE__EXECUTE;
3047
3048                 return file_has_perm(cred, file, av);
3049         }
3050
3051 error:
3052         return rc;
3053 }
3054
3055 static int selinux_mmap_addr(unsigned long addr)
3056 {
3057         int rc = 0;
3058         u32 sid = current_sid();
3059
3060         /*
3061          * notice that we are intentionally putting the SELinux check before
3062          * the secondary cap_file_mmap check.  This is such a likely attempt
3063          * at bad behaviour/exploit that we always want to get the AVC, even
3064          * if DAC would have also denied the operation.
3065          */
3066         if (addr < CONFIG_LSM_MMAP_MIN_ADDR) {
3067                 rc = avc_has_perm(sid, sid, SECCLASS_MEMPROTECT,
3068                                   MEMPROTECT__MMAP_ZERO, NULL);
3069                 if (rc)
3070                         return rc;
3071         }
3072
3073         /* do DAC check on address space usage */
3074         return cap_mmap_addr(addr);
3075 }
3076
3077 static int selinux_mmap_file(struct file *file, unsigned long reqprot,
3078                              unsigned long prot, unsigned long flags)
3079 {
3080         if (selinux_checkreqprot)
3081                 prot = reqprot;
3082
3083         return file_map_prot_check(file, prot,
3084                                    (flags & MAP_TYPE) == MAP_SHARED);
3085 }
3086
3087 static int selinux_file_mprotect(struct vm_area_struct *vma,
3088                                  unsigned long reqprot,
3089                                  unsigned long prot)
3090 {
3091         const struct cred *cred = current_cred();
3092
3093         if (selinux_checkreqprot)
3094                 prot = reqprot;
3095
3096         if (default_noexec &&
3097             (prot & PROT_EXEC) && !(vma->vm_flags & VM_EXEC)) {
3098                 int rc = 0;
3099                 if (vma->vm_start >= vma->vm_mm->start_brk &&
3100                     vma->vm_end <= vma->vm_mm->brk) {
3101                         rc = cred_has_perm(cred, cred, PROCESS__EXECHEAP);
3102                 } else if (!vma->vm_file &&
3103                            vma->vm_start <= vma->vm_mm->start_stack &&
3104                            vma->vm_end >= vma->vm_mm->start_stack) {
3105                         rc = current_has_perm(current, PROCESS__EXECSTACK);
3106                 } else if (vma->vm_file && vma->anon_vma) {
3107                         /*
3108                          * We are making executable a file mapping that has
3109                          * had some COW done. Since pages might have been
3110                          * written, check ability to execute the possibly
3111                          * modified content.  This typically should only
3112                          * occur for text relocations.
3113                          */
3114                         rc = file_has_perm(cred, vma->vm_file, FILE__EXECMOD);
3115                 }
3116                 if (rc)
3117                         return rc;
3118         }
3119
3120         return file_map_prot_check(vma->vm_file, prot, vma->vm_flags&VM_SHARED);
3121 }
3122
3123 static int selinux_file_lock(struct file *file, unsigned int cmd)
3124 {
3125         const struct cred *cred = current_cred();
3126
3127         return file_has_perm(cred, file, FILE__LOCK);
3128 }
3129
3130 static int selinux_file_fcntl(struct file *file, unsigned int cmd,
3131                               unsigned long arg)
3132 {
3133         const struct cred *cred = current_cred();
3134         int err = 0;
3135
3136         switch (cmd) {
3137         case F_SETFL:
3138                 if (!file->f_path.dentry || !file->f_path.dentry->d_inode) {
3139                         err = -EINVAL;
3140                         break;
3141                 }
3142
3143                 if ((file->f_flags & O_APPEND) && !(arg & O_APPEND)) {
3144                         err = file_has_perm(cred, file, FILE__WRITE);
3145                         break;
3146                 }
3147                 /* fall through */
3148         case F_SETOWN:
3149         case F_SETSIG:
3150         case F_GETFL:
3151         case F_GETOWN:
3152         case F_GETSIG:
3153         case F_GETOWNER_UIDS:
3154                 /* Just check FD__USE permission */
3155                 err = file_has_perm(cred, file, 0);
3156                 break;
3157         case F_GETLK:
3158         case F_SETLK:
3159         case F_SETLKW:
3160 #if BITS_PER_LONG == 32
3161         case F_GETLK64:
3162         case F_SETLK64:
3163         case F_SETLKW64:
3164 #endif
3165                 if (!file->f_path.dentry || !file->f_path.dentry->d_inode) {
3166                         err = -EINVAL;
3167                         break;
3168                 }
3169                 err = file_has_perm(cred, file, FILE__LOCK);
3170                 break;
3171         }
3172
3173         return err;
3174 }
3175
3176 static int selinux_file_set_fowner(struct file *file)
3177 {
3178         struct file_security_struct *fsec;
3179
3180         fsec = file->f_security;
3181         fsec->fown_sid = current_sid();
3182
3183         return 0;
3184 }
3185
3186 static int selinux_file_send_sigiotask(struct task_struct *tsk,
3187                                        struct fown_struct *fown, int signum)
3188 {
3189         struct file *file;
3190         u32 sid = task_sid(tsk);
3191         u32 perm;
3192         struct file_security_struct *fsec;
3193
3194         /* struct fown_struct is never outside the context of a struct file */
3195         file = container_of(fown, struct file, f_owner);
3196
3197         fsec = file->f_security;
3198
3199         if (!signum)
3200                 perm = signal_to_av(SIGIO); /* as per send_sigio_to_task */
3201         else
3202                 perm = signal_to_av(signum);
3203
3204         return avc_has_perm(fsec->fown_sid, sid,
3205                             SECCLASS_PROCESS, perm, NULL);
3206 }
3207
3208 static int selinux_file_receive(struct file *file)
3209 {
3210         const struct cred *cred = current_cred();
3211
3212         return file_has_perm(cred, file, file_to_av(file));
3213 }
3214
3215 static int selinux_file_open(struct file *file, const struct cred *cred)
3216 {
3217         struct file_security_struct *fsec;
3218         struct inode_security_struct *isec;
3219
3220         fsec = file->f_security;
3221         isec = file_inode(file)->i_security;
3222         /*
3223          * Save inode label and policy sequence number
3224          * at open-time so that selinux_file_permission
3225          * can determine whether revalidation is necessary.
3226          * Task label is already saved in the file security
3227          * struct as its SID.
3228          */
3229         fsec->isid = isec->sid;
3230         fsec->pseqno = avc_policy_seqno();
3231         /*
3232          * Since the inode label or policy seqno may have changed
3233          * between the selinux_inode_permission check and the saving
3234          * of state above, recheck that access is still permitted.
3235          * Otherwise, access might never be revalidated against the
3236          * new inode label or new policy.
3237          * This check is not redundant - do not remove.
3238          */
3239         return path_has_perm(cred, &file->f_path, open_file_to_av(file));
3240 }
3241
3242 /* task security operations */
3243
3244 static int selinux_task_create(unsigned long clone_flags)
3245 {
3246         return current_has_perm(current, PROCESS__FORK);
3247 }
3248
3249 /*
3250  * allocate the SELinux part of blank credentials
3251  */
3252 static int selinux_cred_alloc_blank(struct cred *cred, gfp_t gfp)
3253 {
3254         struct task_security_struct *tsec;
3255
3256         tsec = kzalloc(sizeof(struct task_security_struct), gfp);
3257         if (!tsec)
3258                 return -ENOMEM;
3259
3260         cred->security = tsec;
3261         return 0;
3262 }
3263
3264 /*
3265  * detach and free the LSM part of a set of credentials
3266  */
3267 static void selinux_cred_free(struct cred *cred)
3268 {
3269         struct task_security_struct *tsec = cred->security;
3270
3271         /*
3272          * cred->security == NULL if security_cred_alloc_blank() or
3273          * security_prepare_creds() returned an error.
3274          */
3275         BUG_ON(cred->security && (unsigned long) cred->security < PAGE_SIZE);
3276         cred->security = (void *) 0x7UL;
3277         kfree(tsec);
3278 }
3279
3280 /*
3281  * prepare a new set of credentials for modification
3282  */
3283 static int selinux_cred_prepare(struct cred *new, const struct cred *old,
3284                                 gfp_t gfp)
3285 {
3286         const struct task_security_struct *old_tsec;
3287         struct task_security_struct *tsec;
3288
3289         old_tsec = old->security;
3290
3291         tsec = kmemdup(old_tsec, sizeof(struct task_security_struct), gfp);
3292         if (!tsec)
3293                 return -ENOMEM;
3294
3295         new->security = tsec;
3296         return 0;
3297 }
3298
3299 /*
3300  * transfer the SELinux data to a blank set of creds
3301  */
3302 static void selinux_cred_transfer(struct cred *new, const struct cred *old)
3303 {
3304         const struct task_security_struct *old_tsec = old->security;
3305         struct task_security_struct *tsec = new->security;
3306
3307         *tsec = *old_tsec;
3308 }