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.moore@hp.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/kernel.h>
  28 #include <linux/tracehook.h>
  29 #include <linux/errno.h>
  30 #include <linux/sched.h>
  31 #include <linux/security.h>
  32 #include <linux/xattr.h>
  33 #include <linux/capability.h>
  34 #include <linux/unistd.h>
  35 #include <linux/mm.h>
  36 #include <linux/mman.h>
  37 #include <linux/slab.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/swap.h>
  40 #include <linux/spinlock.h>
  41 #include <linux/syscalls.h>
  42 #include <linux/file.h>
  43 #include <linux/fdtable.h>
  44 #include <linux/namei.h>
  45 #include <linux/mount.h>
  46 #include <linux/proc_fs.h>
  47 #include <linux/netfilter_ipv4.h>
  48 #include <linux/netfilter_ipv6.h>
  49 #include <linux/tty.h>
  50 #include <net/icmp.h>
  51 #include <net/ip.h>             /* for local_port_range[] */
  52 #include <net/tcp.h>            /* struct or_callable used in sock_rcv_skb */
  53 #include <net/net_namespace.h>
  54 #include <net/netlabel.h>
  55 #include <linux/uaccess.h>
  56 #include <asm/ioctls.h>
  57 #include <asm/atomic.h>
  58 #include <linux/bitops.h>
  59 #include <linux/interrupt.h>
  60 #include <linux/netdevice.h>    /* for network interface checks */
  61 #include <linux/netlink.h>
  62 #include <linux/tcp.h>
  63 #include <linux/udp.h>
  64 #include <linux/dccp.h>
  65 #include <linux/quota.h>
  66 #include <linux/un.h>           /* for Unix socket types */
  67 #include <net/af_unix.h>        /* for Unix socket types */
  68 #include <linux/parser.h>
  69 #include <linux/nfs_mount.h>
  70 #include <net/ipv6.h>
  71 #include <linux/hugetlb.h>
  72 #include <linux/personality.h>
  73 #include <linux/sysctl.h>
  74 #include <linux/audit.h>
  75 #include <linux/string.h>
  76 #include <linux/selinux.h>
  77 #include <linux/mutex.h>
  78 #include <linux/posix-timers.h>
  79 #include <linux/syslog.h>
  80
  81 #include "avc.h"
  82 #include "objsec.h"
  83 #include "netif.h"
  84 #include "netnode.h"
  85 #include "netport.h"
  86 #include "xfrm.h"
  87 #include "netlabel.h"
  88 #include "audit.h"
  89
  90 #define NUM_SEL_MNT_OPTS 5
  91
  92 extern int selinux_nlmsg_lookup(u16 sclass, u16 nlmsg_type, u32 *perm);
  93 extern struct security_operations *security_ops;
  94
  95 /* SECMARK reference count */
  96 atomic_t selinux_secmark_refcount = ATOMIC_INIT(0);
  97
  98 #ifdef CONFIG_SECURITY_SELINUX_DEVELOP
  99 int selinux_enforcing;
 100
 101 static int __init enforcing_setup(char *str)
 102 {
 103         unsigned long enforcing;
 104         if (!strict_strtoul(str, 0, &enforcing))
 105                 selinux_enforcing = enforcing ? 1 : 0;
 106         return 1;
 107 }
 108 __setup("enforcing=", enforcing_setup);
 109 #endif
 110
 111 #ifdef CONFIG_SECURITY_SELINUX_BOOTPARAM
 112 int selinux_enabled = CONFIG_SECURITY_SELINUX_BOOTPARAM_VALUE;
 113
 114 static int __init selinux_enabled_setup(char *str)
 115 {
 116         unsigned long enabled;
 117         if (!strict_strtoul(str, 0, &enabled))
 118                 selinux_enabled = enabled ? 1 : 0;
 119         return 1;
 120 }
 121 __setup("selinux=", selinux_enabled_setup);
 122 #else
 123 int selinux_enabled = 1;
 124 #endif
 125
 126 static struct kmem_cache *sel_inode_cache;
 127
 128 /**
 129  * selinux_secmark_enabled - Check to see if SECMARK is currently enabled
 130  *
 131  * Description:
 132  * This function checks the SECMARK reference counter to see if any SECMARK
 133  * targets are currently configured, if the reference counter is greater than
 134  * zero SECMARK is considered to be enabled.  Returns true (1) if SECMARK is
 135  * enabled, false (0) if SECMARK is disabled.
 136  *
 137  */
 138 static int selinux_secmark_enabled(void)
 139 {
 140         return (atomic_read(&selinux_secmark_refcount) > 0);
 141 }
 142
 143 /*
 144  * initialise the security for the init task
 145  */
 146 static void cred_init_security(void)
 147 {
 148         struct cred *cred = (struct cred *) current->real_cred;
 149         struct task_security_struct *tsec;
 150
 151         tsec = kzalloc(sizeof(struct task_security_struct), GFP_KERNEL);
 152         if (!tsec)
 153                 panic("SELinux:  Failed to initialize initial task.\n");
 154
 155         tsec->osid = tsec->sid = SECINITSID_KERNEL;
 156         cred->security = tsec;
 157 }
 158
 159 /*
 160  * get the security ID of a set of credentials
 161  */
 162 static inline u32 cred_sid(const struct cred *cred)
 163 {
 164         const struct task_security_struct *tsec;
 165
 166         tsec = cred->security;
 167         return tsec->sid;
 168 }
 169
 170 /*
 171  * get the objective security ID of a task
 172  */
 173 static inline u32 task_sid(const struct task_struct *task)
 174 {
 175         u32 sid;
 176
 177         rcu_read_lock();
 178         sid = cred_sid(__task_cred(task));
 179         rcu_read_unlock();
 180         return sid;
 181 }
 182
 183 /*
 184  * get the subjective security ID of the current task
 185  */
 186 static inline u32 current_sid(void)
 187 {
 188         const struct task_security_struct *tsec = current_security();
 189
 190         return tsec->sid;
 191 }
 192
 193 /* Allocate and free functions for each kind of security blob. */
 194
 195 static int inode_alloc_security(struct inode *inode)
 196 {
 197         struct inode_security_struct *isec;
 198         u32 sid = current_sid();
 199
 200         isec = kmem_cache_zalloc(sel_inode_cache, GFP_NOFS);
 201         if (!isec)
 202                 return -ENOMEM;
 203
 204         mutex_init(&isec->lock);
 205         INIT_LIST_HEAD(&isec->list);
 206         isec->inode = inode;
 207         isec->sid = SECINITSID_UNLABELED;
 208         isec->sclass = SECCLASS_FILE;
 209         isec->task_sid = sid;
 210         inode->i_security = isec;
 211
 212         return 0;
 213 }
 214
 215 static void inode_free_security(struct inode *inode)
 216 {
 217         struct inode_security_struct *isec = inode->i_security;
 218         struct superblock_security_struct *sbsec = inode->i_sb->s_security;
 219
 220         spin_lock(&sbsec->isec_lock);
 221         if (!list_empty(&isec->list))
 222                 list_del_init(&isec->list);
 223         spin_unlock(&sbsec->isec_lock);
 224
 225         inode->i_security = NULL;
 226         kmem_cache_free(sel_inode_cache, isec);
 227 }
 228
 229 static int file_alloc_security(struct file *file)
 230 {
 231         struct file_security_struct *fsec;
 232         u32 sid = current_sid();
 233
 234         fsec = kzalloc(sizeof(struct file_security_struct), GFP_KERNEL);
 235         if (!fsec)
 236                 return -ENOMEM;
 237
 238         fsec->sid = sid;
 239         fsec->fown_sid = sid;
 240         file->f_security = fsec;
 241
 242         return 0;
 243 }
 244
 245 static void file_free_security(struct file *file)
 246 {
 247         struct file_security_struct *fsec = file->f_security;
 248         file->f_security = NULL;
 249         kfree(fsec);
 250 }
 251
 252 static int superblock_alloc_security(struct super_block *sb)
 253 {
 254         struct superblock_security_struct *sbsec;
 255
 256         sbsec = kzalloc(sizeof(struct superblock_security_struct), GFP_KERNEL);
 257         if (!sbsec)
 258                 return -ENOMEM;
 259
 260         mutex_init(&sbsec->lock);
 261         INIT_LIST_HEAD(&sbsec->isec_head);
 262         spin_lock_init(&sbsec->isec_lock);
 263         sbsec->sb = sb;
 264         sbsec->sid = SECINITSID_UNLABELED;
 265         sbsec->def_sid = SECINITSID_FILE;
 266         sbsec->mntpoint_sid = SECINITSID_UNLABELED;
 267         sb->s_security = sbsec;
 268
 269         return 0;
 270 }
 271
 272 static void superblock_free_security(struct super_block *sb)
 273 {
 274         struct superblock_security_struct *sbsec = sb->s_security;
 275         sb->s_security = NULL;
 276         kfree(sbsec);
 277 }
 278
 279 /* The security server must be initialized before
 280    any labeling or access decisions can be provided. */
 281 extern int ss_initialized;
 282
 283 /* The file system's label must be initialized prior to use. */
 284
 285 static const char *labeling_behaviors[6] = {
 286         "uses xattr",
 287         "uses transition SIDs",
 288         "uses task SIDs",
 289         "uses genfs_contexts",
 290         "not configured for labeling",
 291         "uses mountpoint labeling",
 292 };
 293
 294 static int inode_doinit_with_dentry(struct inode *inode, struct dentry *opt_dentry);
 295
 296 static inline int inode_doinit(struct inode *inode)
 297 {
 298         return inode_doinit_with_dentry(inode, NULL);
 299 }
 300
 301 enum {
 302         Opt_error = -1,
 303         Opt_context = 1,
 304         Opt_fscontext = 2,
 305         Opt_defcontext = 3,
 306         Opt_rootcontext = 4,
 307         Opt_labelsupport = 5,
 308 };
 309
 310 static const match_table_t tokens = {
 311         {Opt_context, CONTEXT_STR "%s"},
 312         {Opt_fscontext, FSCONTEXT_STR "%s"},
 313         {Opt_defcontext, DEFCONTEXT_STR "%s"},
 314         {Opt_rootcontext, ROOTCONTEXT_STR "%s"},
 315         {Opt_labelsupport, LABELSUPP_STR},
 316         {Opt_error, NULL},
 317 };
 318
 319 #define SEL_MOUNT_FAIL_MSG "SELinux:  duplicate or incompatible mount options\n"
 320
 321 static int may_context_mount_sb_relabel(u32 sid,
 322                         struct superblock_security_struct *sbsec,
 323                         const struct cred *cred)
 324 {
 325         const struct task_security_struct *tsec = cred->security;
 326         int rc;
 327
 328         rc = avc_has_perm(tsec->sid, sbsec->sid, SECCLASS_FILESYSTEM,
 329                           FILESYSTEM__RELABELFROM, NULL);
 330         if (rc)
 331                 return rc;
 332
 333         rc = avc_has_perm(tsec->sid, sid, SECCLASS_FILESYSTEM,
 334                           FILESYSTEM__RELABELTO, NULL);
 335         return rc;
 336 }
 337
 338 static int may_context_mount_inode_relabel(u32 sid,
 339                         struct superblock_security_struct *sbsec,
 340                         const struct cred *cred)
 341 {
 342         const struct task_security_struct *tsec = cred->security;
 343         int rc;
 344         rc = avc_has_perm(tsec->sid, sbsec->sid, SECCLASS_FILESYSTEM,
 345                           FILESYSTEM__RELABELFROM, NULL);
 346         if (rc)
 347                 return rc;
 348
 349         rc = avc_has_perm(sid, sbsec->sid, SECCLASS_FILESYSTEM,
 350                           FILESYSTEM__ASSOCIATE, NULL);
 351         return rc;
 352 }
 353
 354 static int sb_finish_set_opts(struct super_block *sb)
 355 {
 356         struct superblock_security_struct *sbsec = sb->s_security;
 357         struct dentry *root = sb->s_root;
 358         struct inode *root_inode = root->d_inode;
 359         int rc = 0;
 360
 361         if (sbsec->behavior == SECURITY_FS_USE_XATTR) {
 362                 /* Make sure that the xattr handler exists and that no
 363                    error other than -ENODATA is returned by getxattr on
 364                    the root directory.  -ENODATA is ok, as this may be
 365                    the first boot of the SELinux kernel before we have
 366                    assigned xattr values to the filesystem. */
 367                 if (!root_inode->i_op->getxattr) {
 368                         printk(KERN_WARNING "SELinux: (dev %s, type %s) has no "
 369                                "xattr support\n", sb->s_id, sb->s_type->name);
 370                         rc = -EOPNOTSUPP;
 371                         goto out;
 372                 }
 373                 rc = root_inode->i_op->getxattr(root, XATTR_NAME_SELINUX, NULL, 0);
 374                 if (rc < 0 && rc != -ENODATA) {
 375                         if (rc == -EOPNOTSUPP)
 376                                 printk(KERN_WARNING "SELinux: (dev %s, type "
 377                                        "%s) has no security xattr handler\n",
 378                                        sb->s_id, sb->s_type->name);
 379                         else
 380                                 printk(KERN_WARNING "SELinux: (dev %s, type "
 381                                        "%s) getxattr errno %d\n", sb->s_id,
 382                                        sb->s_type->name, -rc);
 383                         goto out;
 384                 }
 385         }
 386
 387         sbsec->flags |= (SE_SBINITIALIZED | SE_SBLABELSUPP);
 388
 389         if (sbsec->behavior > ARRAY_SIZE(labeling_behaviors))
 390                 printk(KERN_ERR "SELinux: initialized (dev %s, type %s), unknown behavior\n",
 391                        sb->s_id, sb->s_type->name);
 392         else
 393                 printk(KERN_DEBUG "SELinux: initialized (dev %s, type %s), %s\n",
 394                        sb->s_id, sb->s_type->name,
 395                        labeling_behaviors[sbsec->behavior-1]);
 396
 397         if (sbsec->behavior == SECURITY_FS_USE_GENFS ||
 398             sbsec->behavior == SECURITY_FS_USE_MNTPOINT ||
 399             sbsec->behavior == SECURITY_FS_USE_NONE ||
 400             sbsec->behavior > ARRAY_SIZE(labeling_behaviors))
 401                 sbsec->flags &= ~SE_SBLABELSUPP;
 402
 403         /* Special handling for sysfs. Is genfs but also has setxattr handler*/
 404         if (strncmp(sb->s_type->name, "sysfs", sizeof("sysfs")) == 0)
 405                 sbsec->flags |= SE_SBLABELSUPP;
 406
 407         /* Initialize the root inode. */
 408         rc = inode_doinit_with_dentry(root_inode, root);
 409
 410         /* Initialize any other inodes associated with the superblock, e.g.
 411            inodes created prior to initial policy load or inodes created
 412            during get_sb by a pseudo filesystem that directly
 413            populates itself. */
 414         spin_lock(&sbsec->isec_lock);
 415 next_inode:
 416         if (!list_empty(&sbsec->isec_head)) {
 417                 struct inode_security_struct *isec =
 418                                 list_entry(sbsec->isec_head.next,
 419                                            struct inode_security_struct, list);
 420                 struct inode *inode = isec->inode;
 421                 spin_unlock(&sbsec->isec_lock);
 422                 inode = igrab(inode);
 423                 if (inode) {
 424                         if (!IS_PRIVATE(inode))
 425                                 inode_doinit(inode);
 426                         iput(inode);
 427                 }
 428                 spin_lock(&sbsec->isec_lock);
 429                 list_del_init(&isec->list);
 430                 goto next_inode;
 431         }
 432         spin_unlock(&sbsec->isec_lock);
 433 out:
 434         return rc;
 435 }
 436
 437 /*
 438  * This function should allow an FS to ask what it's mount security
 439  * options were so it can use those later for submounts, displaying
 440  * mount options, or whatever.
 441  */
 442 static int selinux_get_mnt_opts(const struct super_block *sb,
 443                                 struct security_mnt_opts *opts)
 444 {
 445         int rc = 0, i;
 446         struct superblock_security_struct *sbsec = sb->s_security;
 447         char *context = NULL;
 448         u32 len;
 449         char tmp;
 450
 451         security_init_mnt_opts(opts);
 452
 453         if (!(sbsec->flags & SE_SBINITIALIZED))
 454                 return -EINVAL;
 455
 456         if (!ss_initialized)
 457                 return -EINVAL;
 458
 459         tmp = sbsec->flags & SE_MNTMASK;
 460         /* count the number of mount options for this sb */
 461         for (i = 0; i < 8; i++) {
 462                 if (tmp & 0x01)
 463                         opts->num_mnt_opts++;
 464                 tmp >>= 1;
 465         }
 466         /* Check if the Label support flag is set */
 467         if (sbsec->flags & SE_SBLABELSUPP)
 468                 opts->num_mnt_opts++;
 469
 470         opts->mnt_opts = kcalloc(opts->num_mnt_opts, sizeof(char *), GFP_ATOMIC);
 471         if (!opts->mnt_opts) {
 472                 rc = -ENOMEM;
 473                 goto out_free;
 474         }
 475
 476         opts->mnt_opts_flags = kcalloc(opts->num_mnt_opts, sizeof(int), GFP_ATOMIC);
 477         if (!opts->mnt_opts_flags) {
 478                 rc = -ENOMEM;
 479                 goto out_free;
 480         }
 481
 482         i = 0;
 483         if (sbsec->flags & FSCONTEXT_MNT) {
 484                 rc = security_sid_to_context(sbsec->sid, &context, &len);
 485                 if (rc)
 486                         goto out_free;
 487                 opts->mnt_opts[i] = context;
 488                 opts->mnt_opts_flags[i++] = FSCONTEXT_MNT;
 489         }
 490         if (sbsec->flags & CONTEXT_MNT) {
 491                 rc = security_sid_to_context(sbsec->mntpoint_sid, &context, &len);
 492                 if (rc)
 493                         goto out_free;
 494                 opts->mnt_opts[i] = context;
 495                 opts->mnt_opts_flags[i++] = CONTEXT_MNT;
 496         }
 497         if (sbsec->flags & DEFCONTEXT_MNT) {
 498                 rc = security_sid_to_context(sbsec->def_sid, &context, &len);
 499                 if (rc)
 500                         goto out_free;
 501                 opts->mnt_opts[i] = context;
 502                 opts->mnt_opts_flags[i++] = DEFCONTEXT_MNT;
 503         }
 504         if (sbsec->flags & ROOTCONTEXT_MNT) {
 505                 struct inode *root = sbsec->sb->s_root->d_inode;
 506                 struct inode_security_struct *isec = root->i_security;
 507
 508                 rc = security_sid_to_context(isec->sid, &context, &len);
 509                 if (rc)
 510                         goto out_free;
 511                 opts->mnt_opts[i] = context;
 512                 opts->mnt_opts_flags[i++] = ROOTCONTEXT_MNT;
 513         }
 514         if (sbsec->flags & SE_SBLABELSUPP) {
 515                 opts->mnt_opts[i] = NULL;
 516                 opts->mnt_opts_flags[i++] = SE_SBLABELSUPP;
 517         }
 518
 519         BUG_ON(i != opts->num_mnt_opts);
 520
 521         return 0;
 522
 523 out_free:
 524         security_free_mnt_opts(opts);
 525         return rc;
 526 }
 527
 528 static int bad_option(struct superblock_security_struct *sbsec, char flag,
 529                       u32 old_sid, u32 new_sid)
 530 {
 531         char mnt_flags = sbsec->flags & SE_MNTMASK;
 532
 533         /* check if the old mount command had the same options */
 534         if (sbsec->flags & SE_SBINITIALIZED)
 535                 if (!(sbsec->flags & flag) ||
 536                     (old_sid != new_sid))
 537                         return 1;
 538
 539         /* check if we were passed the same options twice,
 540          * aka someone passed context=a,context=b
 541          */
 542         if (!(sbsec->flags & SE_SBINITIALIZED))
 543                 if (mnt_flags & flag)
 544                         return 1;
 545         return 0;
 546 }
 547
 548 /*
 549  * Allow filesystems with binary mount data to explicitly set mount point
 550  * labeling information.
 551  */
 552 static int selinux_set_mnt_opts(struct super_block *sb,
 553                                 struct security_mnt_opts *opts)
 554 {
 555         const struct cred *cred = current_cred();
 556         int rc = 0, i;
 557         struct superblock_security_struct *sbsec = sb->s_security;
 558         const char *name = sb->s_type->name;
 559         struct inode *inode = sbsec->sb->s_root->d_inode;
 560         struct inode_security_struct *root_isec = inode->i_security;
 561         u32 fscontext_sid = 0, context_sid = 0, rootcontext_sid = 0;
 562         u32 defcontext_sid = 0;
 563         char **mount_options = opts->mnt_opts;
 564         int *flags = opts->mnt_opts_flags;
 565         int num_opts = opts->num_mnt_opts;
 566
 567         mutex_lock(&sbsec->lock);
 568
 569         if (!ss_initialized) {
 570                 if (!num_opts) {
 571                         /* Defer initialization until selinux_complete_init,
 572                            after the initial policy is loaded and the security
 573                            server is ready to handle calls. */
 574                         goto out;
 575                 }
 576                 rc = -EINVAL;
 577                 printk(KERN_WARNING "SELinux: Unable to set superblock options "
 578                         "before the security server is initialized\n");
 579                 goto out;
 580         }
 581
 582         /*
 583          * Binary mount data FS will come through this function twice.  Once
 584          * from an explicit call and once from the generic calls from the vfs.
 585          * Since the generic VFS calls will not contain any security mount data
 586          * we need to skip the double mount verification.
 587          *
 588          * This does open a hole in which we will not notice if the first
 589          * mount using this sb set explict options and a second mount using
 590          * this sb does not set any security options.  (The first options
 591          * will be used for both mounts)
 592          */
 593         if ((sbsec->flags & SE_SBINITIALIZED) && (sb->s_type->fs_flags & FS_BINARY_MOUNTDATA)
 594             && (num_opts == 0))
 595                 goto out;
 596
 597         /*
 598          * parse the mount options, check if they are valid sids.
 599          * also check if someone is trying to mount the same sb more
 600          * than once with different security options.
 601          */
 602         for (i = 0; i < num_opts; i++) {
 603                 u32 sid;
 604
 605                 if (flags[i] == SE_SBLABELSUPP)
 606                         continue;
 607                 rc = security_context_to_sid(mount_options[i],
 608                                              strlen(mount_options[i]), &sid);
 609                 if (rc) {
 610                         printk(KERN_WARNING "SELinux: security_context_to_sid"
 611                                "(%s) failed for (dev %s, type %s) errno=%d\n",
 612                                mount_options[i], sb->s_id, name, rc);
 613                         goto out;
 614                 }
 615                 switch (flags[i]) {
 616                 case FSCONTEXT_MNT:
 617                         fscontext_sid = sid;
 618
 619                         if (bad_option(sbsec, FSCONTEXT_MNT, sbsec->sid,
 620                                         fscontext_sid))
 621                                 goto out_double_mount;
 622
 623                         sbsec->flags |= FSCONTEXT_MNT;
 624                         break;
 625                 case CONTEXT_MNT:
 626                         context_sid = sid;
 627
 628                         if (bad_option(sbsec, CONTEXT_MNT, sbsec->mntpoint_sid,
 629                                         context_sid))
 630                                 goto out_double_mount;
 631
 632                         sbsec->flags |= CONTEXT_MNT;
 633                         break;
 634                 case ROOTCONTEXT_MNT:
 635                         rootcontext_sid = sid;
 636
 637                         if (bad_option(sbsec, ROOTCONTEXT_MNT, root_isec->sid,
 638                                         rootcontext_sid))
 639                                 goto out_double_mount;
 640
 641                         sbsec->flags |= ROOTCONTEXT_MNT;
 642
 643                         break;
 644                 case DEFCONTEXT_MNT:
 645                         defcontext_sid = sid;
 646
 647                         if (bad_option(sbsec, DEFCONTEXT_MNT, sbsec->def_sid,
 648                                         defcontext_sid))
 649                                 goto out_double_mount;
 650
 651                         sbsec->flags |= DEFCONTEXT_MNT;
 652
 653                         break;
 654                 default:
 655                         rc = -EINVAL;
 656                         goto out;
 657                 }
 658         }
 659
 660         if (sbsec->flags & SE_SBINITIALIZED) {
 661                 /* previously mounted with options, but not on this attempt? */
 662                 if ((sbsec->flags & SE_MNTMASK) && !num_opts)
 663                         goto out_double_mount;
 664                 rc = 0;
 665                 goto out;
 666         }
 667
 668         if (strcmp(sb->s_type->name, "proc") == 0)
 669                 sbsec->flags |= SE_SBPROC;
 670
 671         /* Determine the labeling behavior to use for this filesystem type. */
 672         rc = security_fs_use((sbsec->flags & SE_SBPROC) ? "proc" : sb->s_type->name, &sbsec->behavior, &sbsec->sid);
 673         if (rc) {
 674                 printk(KERN_WARNING "%s: security_fs_use(%s) returned %d\n",
 675                        __func__, sb->s_type->name, rc);
 676                 goto out;
 677         }
 678
 679         /* sets the context of the superblock for the fs being mounted. */
 680         if (fscontext_sid) {
 681                 rc = may_context_mount_sb_relabel(fscontext_sid, sbsec, cred);
 682                 if (rc)
 683                         goto out;
 684
 685                 sbsec->sid = fscontext_sid;
 686         }
 687
 688         /*
 689          * Switch to using mount point labeling behavior.
 690          * sets the label used on all file below the mountpoint, and will set
 691          * the superblock context if not already set.
 692          */
 693         if (context_sid) {
 694                 if (!fscontext_sid) {
 695                         rc = may_context_mount_sb_relabel(context_sid, sbsec,
 696                                                           cred);
 697                         if (rc)
 698                                 goto out;
 699                         sbsec->sid = context_sid;
 700                 } else {
 701                         rc = may_context_mount_inode_relabel(context_sid, sbsec,
 702                                                              cred);
 703                         if (rc)
 704                                 goto out;
 705                 }
 706                 if (!rootcontext_sid)
 707                         rootcontext_sid = context_sid;
 708
 709                 sbsec->mntpoint_sid = context_sid;
 710                 sbsec->behavior = SECURITY_FS_USE_MNTPOINT;
 711         }
 712
 713         if (rootcontext_sid) {
 714                 rc = may_context_mount_inode_relabel(rootcontext_sid, sbsec,
 715                                                      cred);
 716                 if (rc)
 717                         goto out;
 718
 719                 root_isec->sid = rootcontext_sid;
 720                 root_isec->initialized = 1;
 721         }
 722
 723         if (defcontext_sid) {
 724                 if (sbsec->behavior != SECURITY_FS_USE_XATTR) {
 725                         rc = -EINVAL;
 726                         printk(KERN_WARNING "SELinux: defcontext option is "
 727                                "invalid for this filesystem type\n");
 728                         goto out;
 729                 }
 730
 731                 if (defcontext_sid != sbsec->def_sid) {
 732                         rc = may_context_mount_inode_relabel(defcontext_sid,
 733                                                              sbsec, cred);
 734                         if (rc)
 735                                 goto out;
 736                 }
 737
 738                 sbsec->def_sid = defcontext_sid;
 739         }
 740
 741         rc = sb_finish_set_opts(sb);
 742 out:
 743         mutex_unlock(&sbsec->lock);
 744         return rc;
 745 out_double_mount:
 746         rc = -EINVAL;
 747         printk(KERN_WARNING "SELinux: mount invalid.  Same superblock, different "
 748                "security settings for (dev %s, type %s)\n", sb->s_id, name);
 749         goto out;
 750 }
 751
 752 static void selinux_sb_clone_mnt_opts(const struct super_block *oldsb,
 753                                         struct super_block *newsb)
 754 {
 755         const struct superblock_security_struct *oldsbsec = oldsb->s_security;
 756         struct superblock_security_struct *newsbsec = newsb->s_security;
 757
 758         int set_fscontext =     (oldsbsec->flags & FSCONTEXT_MNT);
 759         int set_context =       (oldsbsec->flags & CONTEXT_MNT);
 760         int set_rootcontext =   (oldsbsec->flags & ROOTCONTEXT_MNT);
 761
 762         /*
 763          * if the parent was able to be mounted it clearly had no special lsm
 764          * mount options.  thus we can safely deal with this superblock later
 765          */
 766         if (!ss_initialized)
 767                 return;
 768
 769         /* how can we clone if the old one wasn't set up?? */
 770         BUG_ON(!(oldsbsec->flags & SE_SBINITIALIZED));
 771
 772         /* if fs is reusing a sb, just let its options stand... */
 773         if (newsbsec->flags & SE_SBINITIALIZED)
 774                 return;
 775
 776         mutex_lock(&newsbsec->lock);
 777
 778         newsbsec->flags = oldsbsec->flags;
 779
 780         newsbsec->sid = oldsbsec->sid;
 781         newsbsec->def_sid = oldsbsec->def_sid;
 782         newsbsec->behavior = oldsbsec->behavior;
 783
 784         if (set_context) {
 785                 u32 sid = oldsbsec->mntpoint_sid;
 786
 787                 if (!set_fscontext)
 788                         newsbsec->sid = sid;
 789                 if (!set_rootcontext) {
 790                         struct inode *newinode = newsb->s_root->d_inode;
 791                         struct inode_security_struct *newisec = newinode->i_security;
 792                         newisec->sid = sid;
 793                 }
 794                 newsbsec->mntpoint_sid = sid;
 795         }
 796         if (set_rootcontext) {
 797                 const struct inode *oldinode = oldsb->s_root->d_inode;
 798                 const struct inode_security_struct *oldisec = oldinode->i_security;
 799                 struct inode *newinode = newsb->s_root->d_inode;
 800                 struct inode_security_struct *newisec = newinode->i_security;
 801
 802                 newisec->sid = oldisec->sid;
 803         }
 804
 805         sb_finish_set_opts(newsb);
 806         mutex_unlock(&newsbsec->lock);
 807 }
 808
 809 static int selinux_parse_opts_str(char *options,
 810                                   struct security_mnt_opts *opts)
 811 {
 812         char *p;
 813         char *context = NULL, *defcontext = NULL;
 814         char *fscontext = NULL, *rootcontext = NULL;
 815         int rc, num_mnt_opts = 0;
 816
 817         opts->num_mnt_opts = 0;
 818
 819         /* Standard string-based options. */
 820         while ((p = strsep(&options, "|")) != NULL) {
 821                 int token;
 822                 substring_t args[MAX_OPT_ARGS];
 823
 824                 if (!*p)
 825                         continue;
 826
 827                 token = match_token(p, tokens, args);
 828
 829                 switch (token) {
 830                 case Opt_context:
 831                         if (context || defcontext) {
 832                                 rc = -EINVAL;
 833                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
 834                                 goto out_err;
 835                         }
 836                         context = match_strdup(&args[0]);
 837                         if (!context) {
 838                                 rc = -ENOMEM;
 839                                 goto out_err;
 840                         }
 841                         break;
 842
 843                 case Opt_fscontext:
 844                         if (fscontext) {
 845                                 rc = -EINVAL;
 846                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
 847                                 goto out_err;
 848                         }
 849                         fscontext = match_strdup(&args[0]);
 850                         if (!fscontext) {
 851                                 rc = -ENOMEM;
 852                                 goto out_err;
 853                         }
 854                         break;
 855
 856                 case Opt_rootcontext:
 857                         if (rootcontext) {
 858                                 rc = -EINVAL;
 859                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
 860                                 goto out_err;
 861                         }
 862                         rootcontext = match_strdup(&args[0]);
 863                         if (!rootcontext) {
 864                                 rc = -ENOMEM;
 865                                 goto out_err;
 866                         }
 867                         break;
 868
 869                 case Opt_defcontext:
 870                         if (context || defcontext) {
 871                                 rc = -EINVAL;
 872                                 printk(KERN_WARNING SEL_MOUNT_FAIL_MSG);
 873                                 goto out_err;
 874                         }
 875                         defcontext = match_strdup(&args[0]);
 876                         if (!defcontext) {
 877                                 rc = -ENOMEM;
 878                                 goto out_err;
 879                         }
 880                         break;
 881                 case Opt_labelsupport:
 882                         break;
 883                 default:
 884                         rc = -EINVAL;
 885                         printk(KERN_WARNING "SELinux:  unknown mount option\n");
 886                         goto out_err;
 887
 888                 }
 889         }
 890
 891         rc = -ENOMEM;
 892         opts->mnt_opts = kcalloc(NUM_SEL_MNT_OPTS, sizeof(char *), GFP_ATOMIC);
 893         if (!opts->mnt_opts)
 894                 goto out_err;
 895
 896         opts->mnt_opts_flags = kcalloc(NUM_SEL_MNT_OPTS, sizeof(int), GFP_ATOMIC);
 897         if (!opts->mnt_opts_flags) {
 898                 kfree(opts->mnt_opts);
 899                 goto out_err;
 900         }
 901
 902         if (fscontext) {
 903                 opts->mnt_opts[num_mnt_opts] = fscontext;
 904                 opts->mnt_opts_flags[num_mnt_opts++] = FSCONTEXT_MNT;
 905         }
 906         if (context) {
 907                 opts->mnt_opts[num_mnt_opts] = context;
 908                 opts->mnt_opts_flags[num_mnt_opts++] = CONTEXT_MNT;
 909         }
 910         if (rootcontext) {
 911                 opts->mnt_opts[num_mnt_opts] = rootcontext;
 912                 opts->mnt_opts_flags[num_mnt_opts++] = ROOTCONTEXT_MNT;
 913         }
 914         if (defcontext) {
 915                 opts->mnt_opts[num_mnt_opts] = defcontext;
 916                 opts->mnt_opts_flags[num_mnt_opts++] = DEFCONTEXT_MNT;
 917         }
 918
 919         opts->num_mnt_opts = num_mnt_opts;
 920         return 0;
 921
 922 out_err:
 923         kfree(context);
 924         kfree(defcontext);
 925         kfree(fscontext);
 926         kfree(rootcontext);
 927         return rc;
 928 }
 929 /*
 930  * string mount options parsing and call set the sbsec
 931  */
 932 static int superblock_doinit(struct super_block *sb, void *data)
 933 {
 934         int rc = 0;
 935         char *options = data;
 936         struct security_mnt_opts opts;
 937
 938         security_init_mnt_opts(&opts);
 939
 940         if (!data)
 941                 goto out;
 942
 943         BUG_ON(sb->s_type->fs_flags & FS_BINARY_MOUNTDATA);
 944
 945         rc = selinux_parse_opts_str(options, &opts);
 946         if (rc)
 947                 goto out_err;
 948
 949 out:
 950         rc = selinux_set_mnt_opts(sb, &opts);
 951
 952 out_err:
 953         security_free_mnt_opts(&opts);
 954         return rc;
 955 }
 956
 957 static void selinux_write_opts(struct seq_file *m,
 958                                struct security_mnt_opts *opts)
 959 {
 960         int i;
 961         char *prefix;
 962
 963         for (i = 0; i < opts->num_mnt_opts; i++) {
 964                 char *has_comma;
 965
 966                 if (opts->mnt_opts[i])
 967                         has_comma = strchr(opts->mnt_opts[i], ',');
 968                 else
 969                         has_comma = NULL;
 970
 971                 switch (opts->mnt_opts_flags[i]) {
 972                 case CONTEXT_MNT:
 973                         prefix = CONTEXT_STR;
 974                         break;
 975                 case FSCONTEXT_MNT:
 976                         prefix = FSCONTEXT_STR;
 977                         break;
 978                 case ROOTCONTEXT_MNT:
 979                         prefix = ROOTCONTEXT_STR;
 980                         break;
 981                 case DEFCONTEXT_MNT:
 982                         prefix = DEFCONTEXT_STR;
 983                         break;
 984                 case SE_SBLABELSUPP:
 985                         seq_putc(m, ',');
 986                         seq_puts(m, LABELSUPP_STR);
 987                         continue;
 988                 default:
 989                         BUG();
 990                 };
 991                 /* we need a comma before each option */
 992                 seq_putc(m, ',');
 993                 seq_puts(m, prefix);
 994                 if (has_comma)
 995                         seq_putc(m, '\"');
 996                 seq_puts(m, opts->mnt_opts[i]);
 997                 if (has_comma)
 998                         seq_putc(m, '\"');
 999         }
1000 }
1001
1002 static int selinux_sb_show_options(struct seq_file *m, struct super_block *sb)
1003 {
1004         struct security_mnt_opts opts;
1005         int rc;
1006
1007         rc = selinux_get_mnt_opts(sb, &opts);
1008         if (rc) {
1009                 /* before policy load we may get EINVAL, don't show anything */
1010                 if (rc == -EINVAL)
1011                         rc = 0;
1012                 return rc;
1013         }
1014
1015         selinux_write_opts(m, &opts);
1016
1017         security_free_mnt_opts(&opts);
1018
1019         return rc;
1020 }
1021
1022 static inline u16 inode_mode_to_security_class(umode_t mode)
1023 {
1024         switch (mode & S_IFMT) {
1025         case S_IFSOCK:
1026                 return SECCLASS_SOCK_FILE;
1027         case S_IFLNK:
1028                 return SECCLASS_LNK_FILE;
1029         case S_IFREG:
1030                 return SECCLASS_FILE;
1031         case S_IFBLK:
1032                 return SECCLASS_BLK_FILE;
1033         case S_IFDIR:
1034                 return SECCLASS_DIR;
1035         case S_IFCHR:
1036                 return SECCLASS_CHR_FILE;
1037         case S_IFIFO:
1038                 return SECCLASS_FIFO_FILE;
1039
1040         }
1041
1042         return SECCLASS_FILE;
1043 }
1044
1045 static inline int default_protocol_stream(int protocol)
1046 {
1047         return (protocol == IPPROTO_IP || protocol == IPPROTO_TCP);
1048 }
1049
1050 static inline int default_protocol_dgram(int protocol)
1051 {
1052         return (protocol == IPPROTO_IP || protocol == IPPROTO_UDP);
1053 }
1054
1055 static inline u16 socket_type_to_security_class(int family, int type, int protocol)
1056 {
1057         switch (family) {
1058         case PF_UNIX:
1059                 switch (type) {
1060                 case SOCK_STREAM:
1061                 case SOCK_SEQPACKET:
1062                         return SECCLASS_UNIX_STREAM_SOCKET;
1063                 case SOCK_DGRAM:
1064                         return SECCLASS_UNIX_DGRAM_SOCKET;
1065                 }
1066                 break;
1067         case PF_INET:
1068         case PF_INET6:
1069                 switch (type) {
1070                 case SOCK_STREAM:
1071                         if (default_protocol_stream(protocol))
1072                                 return SECCLASS_TCP_SOCKET;
1073                         else
1074                                 return SECCLASS_RAWIP_SOCKET;
1075                 case SOCK_DGRAM:
1076                         if (default_protocol_dgram(protocol))
1077                                 return SECCLASS_UDP_SOCKET;
1078                         else
1079                                 return SECCLASS_RAWIP_SOCKET;
1080                 case SOCK_DCCP:
1081                         return SECCLASS_DCCP_SOCKET;
1082                 default:
1083                         return SECCLASS_RAWIP_SOCKET;
1084                 }
1085                 break;
1086         case PF_NETLINK:
1087                 switch (protocol) {
1088                 case NETLINK_ROUTE:
1089                         return SECCLASS_NETLINK_ROUTE_SOCKET;
1090                 case NETLINK_FIREWALL:
1091                         return SECCLASS_NETLINK_FIREWALL_SOCKET;
1092                 case NETLINK_INET_DIAG:
1093                         return SECCLASS_NETLINK_TCPDIAG_SOCKET;
1094                 case NETLINK_NFLOG:
1095                         return SECCLASS_NETLINK_NFLOG_SOCKET;
1096                 case NETLINK_XFRM:
1097                         return SECCLASS_NETLINK_XFRM_SOCKET;
1098                 case NETLINK_SELINUX:
1099                         return SECCLASS_NETLINK_SELINUX_SOCKET;
1100                 case NETLINK_AUDIT:
1101                         return SECCLASS_NETLINK_AUDIT_SOCKET;
1102                 case NETLINK_IP6_FW:
1103                         return SECCLASS_NETLINK_IP6FW_SOCKET;
1104                 case NETLINK_DNRTMSG:
1105                         return SECCLASS_NETLINK_DNRT_SOCKET;
1106                 case NETLINK_KOBJECT_UEVENT:
1107                         return SECCLASS_NETLINK_KOBJECT_UEVENT_SOCKET;
1108                 default:
1109                         return SECCLASS_NETLINK_SOCKET;
1110                 }
1111         case PF_PACKET:
1112                 return SECCLASS_PACKET_SOCKET;
1113         case PF_KEY:
1114                 return SECCLASS_KEY_SOCKET;
1115         case PF_APPLETALK:
1116                 return SECCLASS_APPLETALK_SOCKET;
1117         }
1118
1119         return SECCLASS_SOCKET;
1120 }
1121
1122 #ifdef CONFIG_PROC_FS
1123 static int selinux_proc_get_sid(struct proc_dir_entry *de,
1124                                 u16 tclass,
1125                                 u32 *sid)
1126 {
1127         int buflen, rc;
1128         char *buffer, *path, *end;
1129
1130         buffer = (char *)__get_free_page(GFP_KERNEL);
1131         if (!buffer)
1132                 return -ENOMEM;
1133
1134         buflen = PAGE_SIZE;
1135         end = buffer+buflen;
1136         *--end = '\0';
1137         buflen--;
1138         path = end-1;
1139         *path = '/';
1140         while (de && de != de->parent) {
1141                 buflen -= de->namelen + 1;
1142                 if (buflen < 0)
1143                         break;
1144                 end -= de->namelen;
1145                 memcpy(end, de->name, de->namelen);
1146                 *--end = '/';
1147                 path = end;
1148                 de = de->parent;
1149         }
1150         rc = security_genfs_sid("proc", path, tclass, sid);
1151         free_page((unsigned long)buffer);
1152         return rc;
1153 }
1154 #else
1155 static int selinux_proc_get_sid(struct proc_dir_entry *de,
1156                                 u16 tclass,
1157                                 u32 *sid)
1158 {
1159         return -EINVAL;
1160 }
1161 #endif
1162
1163 /* The inode's security attributes must be initialized before first use. */
1164 static int inode_doinit_with_dentry(struct inode *inode, struct dentry *opt_dentry)
1165 {
1166         struct superblock_security_struct *sbsec = NULL;
1167         struct inode_security_struct *isec = inode->i_security;
1168         u32 sid;
1169         struct dentry *dentry;
1170 #define INITCONTEXTLEN 255
1171         char *context = NULL;
1172         unsigned len = 0;
1173         int rc = 0;
1174
1175         if (isec->initialized)
1176                 goto out;
1177
1178         mutex_lock(&isec->lock);
1179         if (isec->initialized)
1180                 goto out_unlock;
1181
1182         sbsec = inode->i_sb->s_security;
1183         if (!(sbsec->flags & SE_SBINITIALIZED)) {
1184                 /* Defer initialization until selinux_complete_init,
1185                    after the initial policy is loaded and the security
1186                    server is ready to handle calls. */
1187                 spin_lock(&sbsec->isec_lock);
1188                 if (list_empty(&isec->list))
1189                         list_add(&isec->list, &sbsec->isec_head);
1190                 spin_unlock(&sbsec->isec_lock);
1191                 goto out_unlock;
1192         }
1193
1194         switch (sbsec->behavior) {
1195         case SECURITY_FS_USE_XATTR:
1196                 if (!inode->i_op->getxattr) {
1197                         isec->sid = sbsec->def_sid;
1198                         break;
1199                 }
1200
1201                 /* Need a dentry, since the xattr API requires one.
1202                    Life would be simpler if we could just pass the inode. */
1203                 if (opt_dentry) {
1204                         /* Called from d_instantiate or d_splice_alias. */
1205                         dentry = dget(opt_dentry);
1206                 } else {
1207                         /* Called from selinux_complete_init, try to find a dentry. */
1208                         dentry = d_find_alias(inode);
1209                 }
1210                 if (!dentry) {
1211                         /*
1212                          * this is can be hit on boot when a file is accessed
1213                          * before the policy is loaded.  When we load policy we
1214                          * may find inodes that have no dentry on the
1215                          * sbsec->isec_head list.  No reason to complain as these
1216                          * will get fixed up the next time we go through
1217                          * inode_doinit with a dentry, before these inodes could
1218                          * be used again by userspace.
1219                          */
1220                         goto out_unlock;
1221                 }
1222
1223                 len = INITCONTEXTLEN;
1224                 context = kmalloc(len+1, GFP_NOFS);
1225                 if (!context) {
1226                         rc = -ENOMEM;
1227                         dput(dentry);
1228                         goto out_unlock;
1229                 }
1230                 context[len] = '\0';
1231                 rc = inode->i_op->getxattr(dentry, XATTR_NAME_SELINUX,
1232                                            context, len);
1233                 if (rc == -ERANGE) {
1234                         kfree(context);
1235
1236                         /* Need a larger buffer.  Query for the right size. */
1237                         rc = inode->i_op->getxattr(dentry, XATTR_NAME_SELINUX,
1238                                                    NULL, 0);
1239                         if (rc < 0) {
1240                                 dput(dentry);
1241                                 goto out_unlock;
1242                         }
1243                         len = rc;
1244                         context = kmalloc(len+1, GFP_NOFS);
1245                         if (!context) {
1246                                 rc = -ENOMEM;
1247                                 dput(dentry);
1248                                 goto out_unlock;
1249                         }
1250                         context[len] = '\0';
1251                         rc = inode->i_op->getxattr(dentry,
1252                                                    XATTR_NAME_SELINUX,
1253                                                    context, len);
1254                 }
1255                 dput(dentry);
1256                 if (rc < 0) {
1257                         if (rc != -ENODATA) {
1258                                 printk(KERN_WARNING "SELinux: %s:  getxattr returned "
1259                                        "%d for dev=%s ino=%ld\n", __func__,
1260                                        -rc, inode->i_sb->s_id, inode->i_ino);
1261                                 kfree(context);
1262                                 goto out_unlock;
1263                         }
1264                         /* Map ENODATA to the default file SID */
1265                         sid = sbsec->def_sid;
1266                         rc = 0;
1267                 } else {
1268                         rc = security_context_to_sid_default(context, rc, &sid,
1269                                                              sbsec->def_sid,
1270                                                              GFP_NOFS);
1271                         if (rc) {
1272                                 char *dev = inode->i_sb->s_id;
1273                                 unsigned long ino = inode->i_ino;
1274
1275                                 if (rc == -EINVAL) {
1276                                         if (printk_ratelimit())
1277                                                 printk(KERN_NOTICE "SELinux: inode=%lu on dev=%s was found to have an invalid "
1278                                                         "context=%s.  This indicates you may need to relabel the inode or the "
1279                                                         "filesystem in question.\n", ino, dev, context);
1280                                 } else {
1281                                         printk(KERN_WARNING "SELinux: %s:  context_to_sid(%s) "
1282                                                "returned %d for dev=%s ino=%ld\n",
1283                                                __func__, context, -rc, dev, ino);
1284                                 }
1285                                 kfree(context);
1286                                 /* Leave with the unlabeled SID */
1287                                 rc = 0;
1288                                 break;
1289                         }
1290                 }
1291                 kfree(context);
1292                 isec->sid = sid;
1293                 break;
1294         case SECURITY_FS_USE_TASK:
1295                 isec->sid = isec->task_sid;
1296                 break;
1297         case SECURITY_FS_USE_TRANS:
1298                 /* Default to the fs SID. */
1299                 isec->sid = sbsec->sid;
1300
1301                 /* Try to obtain a transition SID. */
1302                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
1303                 rc = security_transition_sid(isec->task_sid,
1304                                              sbsec->sid,
1305                                              isec->sclass,
1306                                              &sid);
1307                 if (rc)
1308                         goto out_unlock;
1309                 isec->sid = sid;
1310                 break;
1311         case SECURITY_FS_USE_MNTPOINT:
1312                 isec->sid = sbsec->mntpoint_sid;
1313                 break;
1314         default:
1315                 /* Default to the fs superblock SID. */
1316                 isec->sid = sbsec->sid;
1317
1318                 if ((sbsec->flags & SE_SBPROC) && !S_ISLNK(inode->i_mode)) {
1319                         struct proc_inode *proci = PROC_I(inode);
1320                         if (proci->pde) {
1321                                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
1322                                 rc = selinux_proc_get_sid(proci->pde,
1323                                                           isec->sclass,
1324                                                           &sid);
1325                                 if (rc)
1326                                         goto out_unlock;
1327                                 isec->sid = sid;
1328                         }
1329                 }
1330                 break;
1331         }
1332
1333         isec->initialized = 1;
1334
1335 out_unlock:
1336         mutex_unlock(&isec->lock);
1337 out:
1338         if (isec->sclass == SECCLASS_FILE)
1339                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
1340         return rc;
1341 }
1342
1343 /* Convert a Linux signal to an access vector. */
1344 static inline u32 signal_to_av(int sig)
1345 {
1346         u32 perm = 0;
1347
1348         switch (sig) {
1349         case SIGCHLD:
1350                 /* Commonly granted from child to parent. */
1351                 perm = PROCESS__SIGCHLD;
1352                 break;
1353         case SIGKILL:
1354                 /* Cannot be caught or ignored */
1355                 perm = PROCESS__SIGKILL;
1356                 break;
1357         case SIGSTOP:
1358                 /* Cannot be caught or ignored */
1359                 perm = PROCESS__SIGSTOP;
1360                 break;
1361         default:
1362                 /* All other signals. */
1363                 perm = PROCESS__SIGNAL;
1364                 break;
1365         }
1366
1367         return perm;
1368 }
1369
1370 /*
1371  * Check permission between a pair of credentials
1372  * fork check, ptrace check, etc.
1373  */
1374 static int cred_has_perm(const struct cred *actor,
1375                          const struct cred *target,
1376                          u32 perms)
1377 {
1378         u32 asid = cred_sid(actor), tsid = cred_sid(target);
1379
1380         return avc_has_perm(asid, tsid, SECCLASS_PROCESS, perms, NULL);
1381 }
1382
1383 /*
1384  * Check permission between a pair of tasks, e.g. signal checks,
1385  * fork check, ptrace check, etc.
1386  * tsk1 is the actor and tsk2 is the target
1387  * - this uses the default subjective creds of tsk1
1388  */
1389 static int task_has_perm(const struct task_struct *tsk1,
1390                          const struct task_struct *tsk2,
1391                          u32 perms)
1392 {
1393         const struct task_security_struct *__tsec1, *__tsec2;
1394         u32 sid1, sid2;
1395
1396         rcu_read_lock();
1397         __tsec1 = __task_cred(tsk1)->security;  sid1 = __tsec1->sid;
1398         __tsec2 = __task_cred(tsk2)->security;  sid2 = __tsec2->sid;
1399         rcu_read_unlock();
1400         return avc_has_perm(sid1, sid2, SECCLASS_PROCESS, perms, NULL);
1401 }
1402
1403 /*
1404  * Check permission between current and another task, e.g. signal checks,
1405  * fork check, ptrace check, etc.
1406  * current is the actor and tsk2 is the target
1407  * - this uses current's subjective creds
1408  */
1409 static int current_has_perm(const struct task_struct *tsk,
1410                             u32 perms)
1411 {
1412         u32 sid, tsid;
1413
1414         sid = current_sid();
1415         tsid = task_sid(tsk);
1416         return avc_has_perm(sid, tsid, SECCLASS_PROCESS, perms, NULL);
1417 }
1418
1419 #if CAP_LAST_CAP > 63
1420 #error Fix SELinux to handle capabilities > 63.
1421 #endif
1422
1423 /* Check whether a task is allowed to use a capability. */
1424 static int task_has_capability(struct task_struct *tsk,
1425                                const struct cred *cred,
1426                                int cap, int audit)
1427 {
1428         struct common_audit_data ad;
1429         struct av_decision avd;
1430         u16 sclass;
1431         u32 sid = cred_sid(cred);
1432         u32 av = CAP_TO_MASK(cap);
1433         int rc;
1434
1435         COMMON_AUDIT_DATA_INIT(&ad, CAP);
1436         ad.tsk = tsk;
1437         ad.u.cap = cap;
1438
1439         switch (CAP_TO_INDEX(cap)) {
1440         case 0:
1441                 sclass = SECCLASS_CAPABILITY;
1442                 break;
1443         case 1:
1444                 sclass = SECCLASS_CAPABILITY2;
1445                 break;
1446         default:
1447                 printk(KERN_ERR
1448                        "SELinux:  out of range capability %d\n", cap);
1449                 BUG();
1450         }
1451
1452         rc = avc_has_perm_noaudit(sid, sid, sclass, av, 0, &avd);
1453         if (audit == SECURITY_CAP_AUDIT)
1454                 avc_audit(sid, sid, sclass, av, &avd, rc, &ad);
1455         return rc;
1456 }
1457
1458 /* Check whether a task is allowed to use a system operation. */
1459 static int task_has_system(struct task_struct *tsk,
1460                            u32 perms)
1461 {
1462         u32 sid = task_sid(tsk);
1463
1464         return avc_has_perm(sid, SECINITSID_KERNEL,
1465                             SECCLASS_SYSTEM, perms, NULL);
1466 }
1467
1468 /* Check whether a task has a particular permission to an inode.
1469    The 'adp' parameter is optional and allows other audit
1470    data to be passed (e.g. the dentry). */
1471 static int inode_has_perm(const struct cred *cred,
1472                           struct inode *inode,
1473                           u32 perms,
1474                           struct common_audit_data *adp)
1475 {
1476         struct inode_security_struct *isec;
1477         struct common_audit_data ad;
1478         u32 sid;
1479
1480         validate_creds(cred);
1481
1482         if (unlikely(IS_PRIVATE(inode)))
1483                 return 0;
1484
1485         sid = cred_sid(cred);
1486         isec = inode->i_security;
1487
1488         if (!adp) {
1489                 adp = &ad;
1490                 COMMON_AUDIT_DATA_INIT(&ad, FS);
1491                 ad.u.fs.inode = inode;
1492         }
1493
1494         return avc_has_perm(sid, isec->sid, isec->sclass, perms, adp);
1495 }
1496
1497 /* Same as inode_has_perm, but pass explicit audit data containing
1498    the dentry to help the auditing code to more easily generate the
1499    pathname if needed. */
1500 static inline int dentry_has_perm(const struct cred *cred,
1501                                   struct vfsmount *mnt,
1502                                   struct dentry *dentry,
1503                                   u32 av)
1504 {
1505         struct inode *inode = dentry->d_inode;
1506         struct common_audit_data ad;
1507
1508         COMMON_AUDIT_DATA_INIT(&ad, FS);
1509         ad.u.fs.path.mnt = mnt;
1510         ad.u.fs.path.dentry = dentry;
1511         return inode_has_perm(cred, inode, av, &ad);
1512 }
1513
1514 /* Check whether a task can use an open file descriptor to
1515    access an inode in a given way.  Check access to the
1516    descriptor itself, and then use dentry_has_perm to
1517    check a particular permission to the file.
1518    Access to the descriptor is implicitly granted if it
1519    has the same SID as the process.  If av is zero, then
1520    access to the file is not checked, e.g. for cases
1521    where only the descriptor is affected like seek. */
1522 static int file_has_perm(const struct cred *cred,
1523                          struct file *file,
1524                          u32 av)
1525 {
1526         struct file_security_struct *fsec = file->f_security;
1527         struct inode *inode = file->f_path.dentry->d_inode;
1528         struct common_audit_data ad;
1529         u32 sid = cred_sid(cred);
1530         int rc;
1531
1532         COMMON_AUDIT_DATA_INIT(&ad, FS);
1533         ad.u.fs.path = file->f_path;
1534
1535         if (sid != fsec->sid) {
1536                 rc = avc_has_perm(sid, fsec->sid,
1537                                   SECCLASS_FD,
1538                                   FD__USE,
1539                                   &ad);
1540                 if (rc)
1541                         goto out;
1542         }
1543
1544         /* av is zero if only checking access to the descriptor. */
1545         rc = 0;
1546         if (av)
1547                 rc = inode_has_perm(cred, inode, av, &ad);
1548
1549 out:
1550         return rc;
1551 }
1552
1553 /* Check whether a task can create a file. */
1554 static int may_create(struct inode *dir,
1555                       struct dentry *dentry,
1556                       u16 tclass)
1557 {
1558         const struct task_security_struct *tsec = current_security();
1559         struct inode_security_struct *dsec;
1560         struct superblock_security_struct *sbsec;
1561         u32 sid, newsid;
1562         struct common_audit_data ad;
1563         int rc;
1564
1565         dsec = dir->i_security;
1566         sbsec = dir->i_sb->s_security;
1567
1568         sid = tsec->sid;
1569         newsid = tsec->create_sid;
1570
1571         COMMON_AUDIT_DATA_INIT(&ad, FS);
1572         ad.u.fs.path.dentry = dentry;
1573
1574         rc = avc_has_perm(sid, dsec->sid, SECCLASS_DIR,
1575                           DIR__ADD_NAME | DIR__SEARCH,
1576                           &ad);
1577         if (rc)
1578                 return rc;
1579
1580         if (!newsid || !(sbsec->flags & SE_SBLABELSUPP)) {
1581                 rc = security_transition_sid(sid, dsec->sid, tclass, &newsid);
1582                 if (rc)
1583                         return rc;
1584         }
1585
1586         rc = avc_has_perm(sid, newsid, tclass, FILE__CREATE, &ad);
1587         if (rc)
1588                 return rc;
1589
1590         return avc_has_perm(newsid, sbsec->sid,
1591                             SECCLASS_FILESYSTEM,
1592                             FILESYSTEM__ASSOCIATE, &ad);
1593 }
1594
1595 /* Check whether a task can create a key. */
1596 static int may_create_key(u32 ksid,
1597                           struct task_struct *ctx)
1598 {
1599         u32 sid = task_sid(ctx);
1600
1601         return avc_has_perm(sid, ksid, SECCLASS_KEY, KEY__CREATE, NULL);
1602 }
1603
1604 #define MAY_LINK        0
1605 #define MAY_UNLINK      1
1606 #define MAY_RMDIR       2
1607
1608 /* Check whether a task can link, unlink, or rmdir a file/directory. */
1609 static int may_link(struct inode *dir,
1610                     struct dentry *dentry,
1611                     int kind)
1612
1613 {
1614         struct inode_security_struct *dsec, *isec;
1615         struct common_audit_data ad;
1616         u32 sid = current_sid();
1617         u32 av;
1618         int rc;
1619
1620         dsec = dir->i_security;
1621         isec = dentry->d_inode->i_security;
1622
1623         COMMON_AUDIT_DATA_INIT(&ad, FS);
1624         ad.u.fs.path.dentry = dentry;
1625
1626         av = DIR__SEARCH;
1627         av |= (kind ? DIR__REMOVE_NAME : DIR__ADD_NAME);
1628         rc = avc_has_perm(sid, dsec->sid, SECCLASS_DIR, av, &ad);
1629         if (rc)
1630                 return rc;
1631
1632         switch (kind) {
1633         case MAY_LINK:
1634                 av = FILE__LINK;
1635                 break;
1636         case MAY_UNLINK:
1637                 av = FILE__UNLINK;
1638                 break;
1639         case MAY_RMDIR:
1640                 av = DIR__RMDIR;
1641                 break;
1642         default:
1643                 printk(KERN_WARNING "SELinux: %s:  unrecognized kind %d\n",
1644                         __func__, kind);
1645                 return 0;
1646         }
1647
1648         rc = avc_has_perm(sid, isec->sid, isec->sclass, av, &ad);
1649         return rc;
1650 }
1651
1652 static inline int may_rename(struct inode *old_dir,
1653                              struct dentry *old_dentry,
1654                              struct inode *new_dir,
1655                              struct dentry *new_dentry)
1656 {
1657         struct inode_security_struct *old_dsec, *new_dsec, *old_isec, *new_isec;
1658         struct common_audit_data ad;
1659         u32 sid = current_sid();
1660         u32 av;
1661         int old_is_dir, new_is_dir;
1662         int rc;
1663
1664         old_dsec = old_dir->i_security;
1665         old_isec = old_dentry->d_inode->i_security;
1666         old_is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
1667         new_dsec = new_dir->i_security;
1668
1669         COMMON_AUDIT_DATA_INIT(&ad, FS);
1670
1671         ad.u.fs.path.dentry = old_dentry;
1672         rc = avc_has_perm(sid, old_dsec->sid, SECCLASS_DIR,
1673                           DIR__REMOVE_NAME | DIR__SEARCH, &ad);
1674         if (rc)
1675                 return rc;
1676         rc = avc_has_perm(sid, old_isec->sid,
1677                           old_isec->sclass, FILE__RENAME, &ad);
1678         if (rc)
1679                 return rc;
1680         if (old_is_dir && new_dir != old_dir) {
1681                 rc = avc_has_perm(sid, old_isec->sid,
1682                                   old_isec->sclass, DIR__REPARENT, &ad);
1683                 if (rc)
1684                         return rc;
1685         }
1686
1687         ad.u.fs.path.dentry = new_dentry;
1688         av = DIR__ADD_NAME | DIR__SEARCH;
1689         if (new_dentry->d_inode)
1690                 av |= DIR__REMOVE_NAME;
1691         rc = avc_has_perm(sid, new_dsec->sid, SECCLASS_DIR, av, &ad);
1692         if (rc)
1693                 return rc;
1694         if (new_dentry->d_inode) {
1695                 new_isec = new_dentry->d_inode->i_security;
1696                 new_is_dir = S_ISDIR(new_dentry->d_inode->i_mode);
1697                 rc = avc_has_perm(sid, new_isec->sid,
1698                                   new_isec->sclass,
1699                                   (new_is_dir ? DIR__RMDIR : FILE__UNLINK), &ad);
1700                 if (rc)
1701                         return rc;
1702         }
1703
1704         return 0;
1705 }
1706
1707 /* Check whether a task can perform a filesystem operation. */
1708 static int superblock_has_perm(const struct cred *cred,
1709                                struct super_block *sb,
1710                                u32 perms,
1711                                struct common_audit_data *ad)
1712 {
1713         struct superblock_security_struct *sbsec;
1714         u32 sid = cred_sid(cred);
1715
1716         sbsec = sb->s_security;
1717         return avc_has_perm(sid, sbsec->sid, SECCLASS_FILESYSTEM, perms, ad);
1718 }
1719
1720 /* Convert a Linux mode and permission mask to an access vector. */
1721 static inline u32 file_mask_to_av(int mode, int mask)
1722 {
1723         u32 av = 0;
1724
1725         if ((mode & S_IFMT) != S_IFDIR) {
1726                 if (mask & MAY_EXEC)
1727                         av |= FILE__EXECUTE;
1728                 if (mask & MAY_READ)
1729                         av |= FILE__READ;
1730
1731                 if (mask & MAY_APPEND)
1732                         av |= FILE__APPEND;
1733                 else if (mask & MAY_WRITE)
1734                         av |= FILE__WRITE;
1735
1736         } else {
1737                 if (mask & MAY_EXEC)
1738                         av |= DIR__SEARCH;
1739                 if (mask & MAY_WRITE)
1740                         av |= DIR__WRITE;
1741                 if (mask & MAY_READ)
1742                         av |= DIR__READ;
1743         }
1744
1745         return av;
1746 }
1747
1748 /* Convert a Linux file to an access vector. */
1749 static inline u32 file_to_av(struct file *file)
1750 {
1751         u32 av = 0;
1752
1753         if (file->f_mode & FMODE_READ)
1754                 av |= FILE__READ;
1755         if (file->f_mode & FMODE_WRITE) {
1756                 if (file->f_flags & O_APPEND)
1757                         av |= FILE__APPEND;
1758                 else
1759                         av |= FILE__WRITE;
1760         }
1761         if (!av) {
1762                 /*
1763                  * Special file opened with flags 3 for ioctl-only use.
1764                  */
1765                 av = FILE__IOCTL;
1766         }
1767
1768         return av;
1769 }
1770
1771 /*
1772  * Convert a file to an access vector and include the correct open
1773  * open permission.
1774  */
1775 static inline u32 open_file_to_av(struct file *file)
1776 {
1777         u32 av = file_to_av(file);
1778
1779         if (selinux_policycap_openperm)
1780                 av |= FILE__OPEN;
1781
1782         return av;
1783 }
1784
1785 /* Hook functions begin here. */
1786
1787 static int selinux_ptrace_access_check(struct task_struct *child,
1788                                      unsigned int mode)
1789 {
1790         int rc;
1791
1792         rc = cap_ptrace_access_check(child, mode);
1793         if (rc)
1794                 return rc;
1795
1796         if (mode == PTRACE_MODE_READ) {
1797                 u32 sid = current_sid();
1798                 u32 csid = task_sid(child);
1799                 return avc_has_perm(sid, csid, SECCLASS_FILE, FILE__READ, NULL);
1800         }
1801
1802         return current_has_perm(child, PROCESS__PTRACE);
1803 }
1804
1805 static int selinux_ptrace_traceme(struct task_struct *parent)
1806 {
1807         int rc;
1808
1809         rc = cap_ptrace_traceme(parent);
1810         if (rc)
1811                 return rc;
1812
1813         return task_has_perm(parent, current, PROCESS__PTRACE);
1814 }
1815
1816 static int selinux_capget(struct task_struct *target, kernel_cap_t *effective,
1817                           kernel_cap_t *inheritable, kernel_cap_t *permitted)
1818 {
1819         int error;
1820
1821         error = current_has_perm(target, PROCESS__GETCAP);
1822         if (error)
1823                 return error;
1824
1825         return cap_capget(target, effective, inheritable, permitted);
1826 }
1827
1828 static int selinux_capset(struct cred *new, const struct cred *old,
1829                           const kernel_cap_t *effective,
1830                           const kernel_cap_t *inheritable,
1831                           const kernel_cap_t *permitted)
1832 {
1833         int error;
1834
1835         error = cap_capset(new, old,
1836                                       effective, inheritable, permitted);
1837         if (error)
1838                 return error;
1839
1840         return cred_has_perm(old, new, PROCESS__SETCAP);
1841 }
1842
1843 /*
1844  * (This comment used to live with the selinux_task_setuid hook,
1845  * which was removed).
1846  *
1847  * Since setuid only affects the current process, and since the SELinux
1848  * controls are not based on the Linux identity attributes, SELinux does not
1849  * need to control this operation.  However, SELinux does control the use of
1850  * the CAP_SETUID and CAP_SETGID capabilities using the capable hook.
1851  */
1852
1853 static int selinux_capable(struct task_struct *tsk, const struct cred *cred,
1854                            int cap, int audit)
1855 {
1856         int rc;
1857
1858         rc = cap_capable(tsk, cred, cap, audit);
1859         if (rc)
1860                 return rc;
1861
1862         return task_has_capability(tsk, cred, cap, audit);
1863 }
1864
1865 static int selinux_sysctl_get_sid(ctl_table *table, u16 tclass, u32 *sid)
1866 {
1867         int buflen, rc;
1868         char *buffer, *path, *end;
1869
1870         rc = -ENOMEM;
1871         buffer = (char *)__get_free_page(GFP_KERNEL);
1872         if (!buffer)
1873                 goto out;
1874
1875         buflen = PAGE_SIZE;
1876         end = buffer+buflen;
1877         *--end = '\0';
1878         buflen--;
1879         path = end-1;
1880         *path = '/';
1881         while (table) {
1882                 const char *name = table->procname;
1883                 size_t namelen = strlen(name);
1884                 buflen -= namelen + 1;
1885                 if (buflen < 0)
1886                         goto out_free;
1887                 end -= namelen;
1888                 memcpy(end, name, namelen);
1889                 *--end = '/';
1890                 path = end;
1891                 table = table->parent;
1892         }
1893         buflen -= 4;
1894         if (buflen < 0)
1895                 goto out_free;
1896         end -= 4;
1897         memcpy(end, "/sys", 4);
1898         path = end;
1899         rc = security_genfs_sid("proc", path, tclass, sid);
1900 out_free:
1901         free_page((unsigned long)buffer);
1902 out:
1903         return rc;
1904 }
1905
1906 static int selinux_sysctl(ctl_table *table, int op)
1907 {
1908         int error = 0;
1909         u32 av;
1910         u32 tsid, sid;
1911         int rc;
1912
1913         sid = current_sid();
1914
1915         rc = selinux_sysctl_get_sid(table, (op == 0001) ?
1916                                     SECCLASS_DIR : SECCLASS_FILE, &tsid);
1917         if (rc) {
1918                 /* Default to the well-defined sysctl SID. */
1919                 tsid = SECINITSID_SYSCTL;
1920         }
1921
1922         /* The op values are "defined" in sysctl.c, thereby creating
1923          * a bad coupling between this module and sysctl.c */
1924         if (op == 001) {
1925                 error = avc_has_perm(sid, tsid,
1926                                      SECCLASS_DIR, DIR__SEARCH, NULL);
1927         } else {
1928                 av = 0;
1929                 if (op & 004)
1930                         av |= FILE__READ;
1931                 if (op & 002)
1932                         av |= FILE__WRITE;
1933                 if (av)
1934                         error = avc_has_perm(sid, tsid,
1935                                              SECCLASS_FILE, av, NULL);
1936         }
1937
1938         return error;
1939 }
1940
1941 static int selinux_quotactl(int cmds, int type, int id, struct super_block *sb)
1942 {
1943         const struct cred *cred = current_cred();
1944         int rc = 0;
1945
1946         if (!sb)
1947                 return 0;
1948
1949         switch (cmds) {
1950         case Q_SYNC:
1951         case Q_QUOTAON:
1952         case Q_QUOTAOFF:
1953         case Q_SETINFO:
1954         case Q_SETQUOTA:
1955                 rc = superblock_has_perm(cred, sb, FILESYSTEM__QUOTAMOD, NULL);
1956                 break;
1957         case Q_GETFMT:
1958         case Q_GETINFO:
1959         case Q_GETQUOTA:
1960                 rc = superblock_has_perm(cred, sb, FILESYSTEM__QUOTAGET, NULL);
1961                 break;
1962         default:
1963                 rc = 0;  /* let the kernel handle invalid cmds */
1964                 break;
1965         }
1966         return rc;
1967 }
1968
1969 static int selinux_quota_on(struct dentry *dentry)
1970 {
1971         const struct cred *cred = current_cred();
1972
1973         return dentry_has_perm(cred, NULL, dentry, FILE__QUOTAON);
1974 }
1975
1976 static int selinux_syslog(int type, bool from_file)
1977 {
1978         int rc;
1979
1980         rc = cap_syslog(type, from_file);
1981         if (rc)
1982                 return rc;
1983
1984         switch (type) {
1985         case SYSLOG_ACTION_READ_ALL:    /* Read last kernel messages */
1986         case SYSLOG_ACTION_SIZE_BUFFER: /* Return size of the log buffer */
1987                 rc = task_has_system(current, SYSTEM__SYSLOG_READ);
1988                 break;
1989         case SYSLOG_ACTION_CONSOLE_OFF: /* Disable logging to console */
1990         case SYSLOG_ACTION_CONSOLE_ON:  /* Enable logging to console */
1991         /* Set level of messages printed to console */
1992         case SYSLOG_ACTION_CONSOLE_LEVEL:
1993                 rc = task_has_system(current, SYSTEM__SYSLOG_CONSOLE);
1994                 break;
1995         case SYSLOG_ACTION_CLOSE:       /* Close log */
1996         case SYSLOG_ACTION_OPEN:        /* Open log */
1997         case SYSLOG_ACTION_READ:        /* Read from log */
1998         case SYSLOG_ACTION_READ_CLEAR:  /* Read/clear last kernel messages */
1999         case SYSLOG_ACTION_CLEAR:       /* Clear ring buffer */
2000         default:
2001                 rc = task_has_system(current, SYSTEM__SYSLOG_MOD);
2002                 break;
2003         }
2004         return rc;
2005 }
2006
2007 /*
2008  * Check that a process has enough memory to allocate a new virtual
2009  * mapping. 0 means there is enough memory for the allocation to
2010  * succeed and -ENOMEM implies there is not.
2011  *
2012  * Do not audit the selinux permission check, as this is applied to all
2013  * processes that allocate mappings.
2014  */
2015 static int selinux_vm_enough_memory(struct mm_struct *mm, long pages)
2016 {
2017         int rc, cap_sys_admin = 0;
2018
2019         rc = selinux_capable(current, current_cred(), CAP_SYS_ADMIN,
2020                              SECURITY_CAP_NOAUDIT);
2021         if (rc == 0)
2022                 cap_sys_admin = 1;
2023
2024         return __vm_enough_memory(mm, pages, cap_sys_admin);
2025 }
2026
2027 /* binprm security operations */
2028
2029 static int selinux_bprm_set_creds(struct linux_binprm *bprm)
2030 {
2031         const struct task_security_struct *old_tsec;
2032         struct task_security_struct *new_tsec;
2033         struct inode_security_struct *isec;
2034         struct common_audit_data ad;
2035         struct inode *inode = bprm->file->f_path.dentry->d_inode;
2036         int rc;
2037
2038         rc = cap_bprm_set_creds(bprm);
2039         if (rc)
2040                 return rc;
2041
2042         /* SELinux context only depends on initial program or script and not
2043          * the script interpreter */
2044         if (bprm->cred_prepared)
2045                 return 0;
2046
2047         old_tsec = current_security();
2048         new_tsec = bprm->cred->security;
2049         isec = inode->i_security;
2050
2051         /* Default to the current task SID. */
2052         new_tsec->sid = old_tsec->sid;
2053         new_tsec->osid = old_tsec->sid;
2054
2055         /* Reset fs, key, and sock SIDs on execve. */
2056         new_tsec->create_sid = 0;
2057         new_tsec->keycreate_sid = 0;
2058         new_tsec->sockcreate_sid = 0;
2059
2060         if (old_tsec->exec_sid) {
2061                 new_tsec->sid = old_tsec->exec_sid;
2062                 /* Reset exec SID on execve. */
2063                 new_tsec->exec_sid = 0;
2064         } else {
2065                 /* Check for a default transition on this program. */
2066                 rc = security_transition_sid(old_tsec->sid, isec->sid,
2067                                              SECCLASS_PROCESS, &new_tsec->sid);
2068                 if (rc)
2069                         return rc;
2070         }
2071
2072         COMMON_AUDIT_DATA_INIT(&ad, FS);
2073         ad.u.fs.path = bprm->file->f_path;
2074
2075         if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID)
2076                 new_tsec->sid = old_tsec->sid;
2077
2078         if (new_tsec->sid == old_tsec->sid) {
2079                 rc = avc_has_perm(old_tsec->sid, isec->sid,
2080                                   SECCLASS_FILE, FILE__EXECUTE_NO_TRANS, &ad);
2081                 if (rc)
2082                         return rc;
2083         } else {
2084                 /* Check permissions for the transition. */
2085                 rc = avc_has_perm(old_tsec->sid, new_tsec->sid,
2086                                   SECCLASS_PROCESS, PROCESS__TRANSITION, &ad);
2087                 if (rc)
2088                         return rc;
2089
2090                 rc = avc_has_perm(new_tsec->sid, isec->sid,
2091                                   SECCLASS_FILE, FILE__ENTRYPOINT, &ad);
2092                 if (rc)
2093                         return rc;
2094
2095                 /* Check for shared state */
2096                 if (bprm->unsafe & LSM_UNSAFE_SHARE) {
2097                         rc = avc_has_perm(old_tsec->sid, new_tsec->sid,
2098                                           SECCLASS_PROCESS, PROCESS__SHARE,
2099                                           NULL);
2100                         if (rc)
2101                                 return -EPERM;
2102                 }
2103
2104                 /* Make sure that anyone attempting to ptrace over a task that
2105                  * changes its SID has the appropriate permit */
2106                 if (bprm->unsafe &
2107                     (LSM_UNSAFE_PTRACE | LSM_UNSAFE_PTRACE_CAP)) {
2108                         struct task_struct *tracer;
2109                         struct task_security_struct *sec;
2110                         u32 ptsid = 0;
2111
2112                         rcu_read_lock();
2113                         tracer = tracehook_tracer_task(current);
2114                         if (likely(tracer != NULL)) {
2115                                 sec = __task_cred(tracer)->security;
2116                                 ptsid = sec->sid;
2117                         }
2118                         rcu_read_unlock();
2119
2120                         if (ptsid != 0) {
2121                                 rc = avc_has_perm(ptsid, new_tsec->sid,
2122                                                   SECCLASS_PROCESS,
2123                                                   PROCESS__PTRACE, NULL);
2124                                 if (rc)
2125                                         return -EPERM;
2126                         }
2127                 }
2128
2129                 /* Clear any possibly unsafe personality bits on exec: */
2130                 bprm->per_clear |= PER_CLEAR_ON_SETID;
2131         }
2132
2133         return 0;
2134 }
2135
2136 static int selinux_bprm_secureexec(struct linux_binprm *bprm)
2137 {
2138         const struct task_security_struct *tsec = current_security();
2139         u32 sid, osid;
2140         int atsecure = 0;
2141
2142         sid = tsec->sid;
2143         osid = tsec->osid;
2144
2145         if (osid != sid) {
2146                 /* Enable secure mode for SIDs transitions unless
2147                    the noatsecure permission is granted between
2148                    the two SIDs, i.e. ahp returns 0. */
2149                 atsecure = avc_has_perm(osid, sid,
2150                                         SECCLASS_PROCESS,
2151                                         PROCESS__NOATSECURE, NULL);
2152         }
2153
2154         return (atsecure || cap_bprm_secureexec(bprm));
2155 }
2156
2157 extern struct vfsmount *selinuxfs_mount;
2158 extern struct dentry *selinux_null;
2159
2160 /* Derived from fs/exec.c:flush_old_files. */
2161 static inline void flush_unauthorized_files(const struct cred *cred,
2162                                             struct files_struct *files)
2163 {
2164         struct common_audit_data ad;
2165         struct file *file, *devnull = NULL;
2166         struct tty_struct *tty;
2167         struct fdtable *fdt;
2168         long j = -1;
2169         int drop_tty = 0;
2170
2171         tty = get_current_tty();
2172         if (tty) {
2173                 file_list_lock();
2174                 if (!list_empty(&tty->tty_files)) {
2175                         struct inode *inode;
2176
2177                         /* Revalidate access to controlling tty.
2178                            Use inode_has_perm on the tty inode directly rather
2179                            than using file_has_perm, as this particular open
2180                            file may belong to another process and we are only
2181                            interested in the inode-based check here. */
2182                         file = list_first_entry(&tty->tty_files, struct file, f_u.fu_list);
2183                         inode = file->f_path.dentry->d_inode;
2184                         if (inode_has_perm(cred, inode,
2185                                            FILE__READ | FILE__WRITE, NULL)) {
2186                                 drop_tty = 1;
2187                         }
2188                 }
2189                 file_list_unlock();
2190                 tty_kref_put(tty);
2191         }
2192         /* Reset controlling tty. */
2193         if (drop_tty)
2194                 no_tty();
2195
2196         /* Revalidate access to inherited open files. */
2197
2198         COMMON_AUDIT_DATA_INIT(&ad, FS);
2199
2200         spin_lock(&files->file_lock);
2201         for (;;) {
2202                 unsigned long set, i;
2203                 int fd;
2204
2205                 j++;
2206                 i = j * __NFDBITS;
2207                 fdt = files_fdtable(files);
2208                 if (i >= fdt->max_fds)
2209                         break;
2210                 set = fdt->open_fds->fds_bits[j];
2211                 if (!set)
2212                         continue;
2213                 spin_unlock(&files->file_lock);
2214                 for ( ; set ; i++, set >>= 1) {
2215                         if (set & 1) {
2216                                 file = fget(i);
2217                                 if (!file)
2218                                         continue;
2219                                 if (file_has_perm(cred,
2220                                                   file,
2221                                                   file_to_av(file))) {
2222                                         sys_close(i);
2223                                         fd = get_unused_fd();
2224                                         if (fd != i) {
2225                                                 if (fd >= 0)
2226                                                         put_unused_fd(fd);
2227                                                 fput(file);
2228                                                 continue;
2229                                         }
2230                                         if (devnull) {
2231                                                 get_file(devnull);
2232                                         } else {
2233                                                 devnull = dentry_open(
2234                                                         dget(selinux_null),
2235                                                         mntget(selinuxfs_mount),
2236                                                         O_RDWR, cred);
2237                                                 if (IS_ERR(devnull)) {
2238                                                         devnull = NULL;
2239                                                         put_unused_fd(fd);
2240                                                         fput(file);
2241                                                         continue;
2242                                                 }
2243                                         }
2244                                         fd_install(fd, devnull);
2245                                 }
2246                                 fput(file);
2247                         }
2248                 }
2249                 spin_lock(&files->file_lock);
2250
2251         }
2252         spin_unlock(&files->file_lock);
2253 }
2254
2255 /*
2256  * Prepare a process for imminent new credential changes due to exec
2257  */
2258 static void selinux_bprm_committing_creds(struct linux_binprm *bprm)
2259 {
2260         struct task_security_struct *new_tsec;
2261         struct rlimit *rlim, *initrlim;
2262         int rc, i;
2263
2264         new_tsec = bprm->cred->security;
2265         if (new_tsec->sid == new_tsec->osid)
2266                 return;
2267
2268         /* Close files for which the new task SID is not authorized. */
2269         flush_unauthorized_files(bprm->cred, current->files);
2270
2271         /* Always clear parent death signal on SID transitions. */
2272         current->pdeath_signal = 0;
2273
2274         /* Check whether the new SID can inherit resource limits from the old
2275          * SID.  If not, reset all soft limits to the lower of the current
2276          * task's hard limit and the init task's soft limit.
2277          *
2278          * Note that the setting of hard limits (even to lower them) can be
2279          * controlled by the setrlimit check.  The inclusion of the init task's
2280          * soft limit into the computation is to avoid resetting soft limits
2281          * higher than the default soft limit for cases where the default is
2282          * lower than the hard limit, e.g. RLIMIT_CORE or RLIMIT_STACK.
2283          */
2284         rc = avc_has_perm(new_tsec->osid, new_tsec->sid, SECCLASS_PROCESS,
2285                           PROCESS__RLIMITINH, NULL);
2286         if (rc) {
2287                 for (i = 0; i < RLIM_NLIMITS; i++) {
2288                         rlim = current->signal->rlim + i;
2289                         initrlim = init_task.signal->rlim + i;
2290                         rlim->rlim_cur = min(rlim->rlim_max, initrlim->rlim_cur);
2291                 }
2292                 update_rlimit_cpu(current->signal->rlim[RLIMIT_CPU].rlim_cur);
2293         }
2294 }
2295
2296 /*
2297  * Clean up the process immediately after the installation of new credentials
2298  * due to exec
2299  */
2300 static void selinux_bprm_committed_creds(struct linux_binprm *bprm)
2301 {
2302         const struct task_security_struct *tsec = current_security();
2303         struct itimerval itimer;
2304         u32 osid, sid;
2305         int rc, i;
2306
2307         osid = tsec->osid;
2308         sid = tsec->sid;
2309
2310         if (sid == osid)
2311                 return;
2312
2313         /* Check whether the new SID can inherit signal state from the old SID.
2314          * If not, clear itimers to avoid subsequent signal generation and
2315          * flush and unblock signals.
2316          *
2317          * This must occur _after_ the task SID has been updated so that any
2318          * kill done after the flush will be checked against the new SID.
2319          */
2320         rc = avc_has_perm(osid, sid, SECCLASS_PROCESS, PROCESS__SIGINH, NULL);
2321         if (rc) {
2322                 memset(&itimer, 0, sizeof itimer);
2323                 for (i = 0; i < 3; i++)
2324                         do_setitimer(i, &itimer, NULL);
2325                 spin_lock_irq(&current->sighand->siglock);
2326                 if (!(current->signal->flags & SIGNAL_GROUP_EXIT)) {
2327                         __flush_signals(current);
2328                         flush_signal_handlers(current, 1);
2329                         sigemptyset(&current->blocked);
2330                 }
2331                 spin_unlock_irq(&current->sighand->siglock);
2332         }
2333
2334         /* Wake up the parent if it is waiting so that it can recheck
2335          * wait permission to the new task SID. */
2336         read_lock(&tasklist_lock);
2337         __wake_up_parent(current, current->real_parent);
2338         read_unlock(&tasklist_lock);
2339 }
2340
2341 /* superblock security operations */
2342
2343 static int selinux_sb_alloc_security(struct super_block *sb)
2344 {
2345         return superblock_alloc_security(sb);
2346 }
2347
2348 static void selinux_sb_free_security(struct super_block *sb)
2349 {
2350         superblock_free_security(sb);
2351 }
2352
2353 static inline int match_prefix(char *prefix, int plen, char *option, int olen)
2354 {
2355         if (plen > olen)
2356                 return 0;
2357
2358         return !memcmp(prefix, option, plen);
2359 }
2360
2361 static inline int selinux_option(char *option, int len)
2362 {
2363         return (match_prefix(CONTEXT_STR, sizeof(CONTEXT_STR)-1, option, len) ||
2364                 match_prefix(FSCONTEXT_STR, sizeof(FSCONTEXT_STR)-1, option, len) ||
2365                 match_prefix(DEFCONTEXT_STR, sizeof(DEFCONTEXT_STR)-1, option, len) ||
2366                 match_prefix(ROOTCONTEXT_STR, sizeof(ROOTCONTEXT_STR)-1, option, len) ||
2367                 match_prefix(LABELSUPP_STR, sizeof(LABELSUPP_STR)-1, option, len));
2368 }
2369
2370 static inline void take_option(char **to, char *from, int *first, int len)
2371 {
2372         if (!*first) {
2373                 **to = ',';
2374                 *to += 1;
2375         } else
2376                 *first = 0;
2377         memcpy(*to, from, len);
2378         *to += len;
2379 }
2380
2381 static inline void take_selinux_option(char **to, char *from, int *first,
2382                                        int len)
2383 {
2384         int current_size = 0;
2385
2386         if (!*first) {
2387                 **to = '|';
2388                 *to += 1;
2389         } else
2390                 *first = 0;
2391
2392         while (current_size < len) {
2393                 if (*from != '"') {
2394                         **to = *from;
2395                         *to += 1;
2396                 }
2397                 from += 1;
2398                 current_size += 1;
2399         }
2400 }
2401
2402 static int selinux_sb_copy_data(char *orig, char *copy)
2403 {
2404         int fnosec, fsec, rc = 0;
2405         char *in_save, *in_curr, *in_end;
2406         char *sec_curr, *nosec_save, *nosec;
2407         int open_quote = 0;
2408
2409         in_curr = orig;
2410         sec_curr = copy;
2411
2412         nosec = (char *)get_zeroed_page(GFP_KERNEL);
2413         if (!nosec) {
2414                 rc = -ENOMEM;
2415                 goto out;
2416         }
2417
2418         nosec_save = nosec;
2419         fnosec = fsec = 1;
2420         in_save = in_end = orig;
2421
2422         do {
2423                 if (*in_end == '"')
2424                         open_quote = !open_quote;
2425                 if ((*in_end == ',' && open_quote == 0) ||
2426                                 *in_end == '\0') {
2427                         int len = in_end - in_curr;
2428
2429                         if (selinux_option(in_curr, len))
2430                                 take_selinux_option(&sec_curr, in_curr, &fsec, len);
2431                         else
2432                                 take_option(&nosec, in_curr, &fnosec, len);
2433
2434                         in_curr = in_end + 1;
2435                 }
2436         } while (*in_end++);
2437
2438         strcpy(in_save, nosec_save);
2439         free_page((unsigned long)nosec_save);
2440 out:
2441         return rc;
2442 }
2443
2444 static int selinux_sb_kern_mount(struct super_block *sb, int flags, void *data)
2445 {
2446         const struct cred *cred = current_cred();
2447         struct common_audit_data ad;
2448         int rc;
2449
2450         rc = superblock_doinit(sb, data);
2451         if (rc)
2452                 return rc;
2453
2454         /* Allow all mounts performed by the kernel */
2455         if (flags & MS_KERNMOUNT)
2456                 return 0;
2457
2458         COMMON_AUDIT_DATA_INIT(&ad, FS);
2459         ad.u.fs.path.dentry = sb->s_root;
2460         return superblock_has_perm(cred, sb, FILESYSTEM__MOUNT, &ad);
2461 }
2462
2463 static int selinux_sb_statfs(struct dentry *dentry)
2464 {
2465         const struct cred *cred = current_cred();
2466         struct common_audit_data ad;
2467
2468         COMMON_AUDIT_DATA_INIT(&ad, FS);
2469         ad.u.fs.path.dentry = dentry->d_sb->s_root;
2470         return superblock_has_perm(cred, dentry->d_sb, FILESYSTEM__GETATTR, &ad);
2471 }
2472
2473 static int selinux_mount(char *dev_name,
2474                          struct path *path,
2475                          char *type,
2476                          unsigned long flags,
2477                          void *data)
2478 {
2479         const struct cred *cred = current_cred();
2480
2481         if (flags & MS_REMOUNT)
2482                 return superblock_has_perm(cred, path->mnt->mnt_sb,
2483                                            FILESYSTEM__REMOUNT, NULL);
2484         else
2485                 return dentry_has_perm(cred, path->mnt, path->dentry,
2486                                        FILE__MOUNTON);
2487 }
2488
2489 static int selinux_umount(struct vfsmount *mnt, int flags)
2490 {
2491         const struct cred *cred = current_cred();
2492
2493         return superblock_has_perm(cred, mnt->mnt_sb,
2494                                    FILESYSTEM__UNMOUNT, NULL);
2495 }
2496
2497 /* inode security operations */
2498
2499 static int selinux_inode_alloc_security(struct inode *inode)
2500 {
2501         return inode_alloc_security(inode);
2502 }
2503
2504 static void selinux_inode_free_security(struct inode *inode)
2505 {
2506         inode_free_security(inode);
2507 }
2508
2509 static int selinux_inode_init_security(struct inode *inode, struct inode *dir,
2510                                        char **name, void **value,
2511                                        size_t *len)
2512 {
2513         const struct task_security_struct *tsec = current_security();
2514         struct inode_security_struct *dsec;
2515         struct superblock_security_struct *sbsec;
2516         u32 sid, newsid, clen;
2517         int rc;
2518         char *namep = NULL, *context;
2519
2520         dsec = dir->i_security;
2521         sbsec = dir->i_sb->s_security;
2522
2523         sid = tsec->sid;
2524         newsid = tsec->create_sid;
2525
2526         if (!newsid || !(sbsec->flags & SE_SBLABELSUPP)) {
2527                 rc = security_transition_sid(sid, dsec->sid,
2528                                              inode_mode_to_security_class(inode->i_mode),
2529                                              &newsid);
2530                 if (rc) {
2531                         printk(KERN_WARNING "%s:  "
2532                                "security_transition_sid failed, rc=%d (dev=%s "
2533                                "ino=%ld)\n",
2534                                __func__,
2535                                -rc, inode->i_sb->s_id, inode->i_ino);
2536                         return rc;
2537                 }
2538         }
2539
2540         /* Possibly defer initialization to selinux_complete_init. */
2541         if (sbsec->flags & SE_SBINITIALIZED) {
2542                 struct inode_security_struct *isec = inode->i_security;
2543                 isec->sclass = inode_mode_to_security_class(inode->i_mode);
2544                 isec->sid = newsid;
2545                 isec->initialized = 1;
2546         }
2547
2548         if (!ss_initialized || !(sbsec->flags & SE_SBLABELSUPP))
2549                 return -EOPNOTSUPP;
2550
2551         if (name) {
2552                 namep = kstrdup(XATTR_SELINUX_SUFFIX, GFP_NOFS);
2553                 if (!namep)
2554                         return -ENOMEM;
2555                 *name = namep;
2556         }
2557
2558         if (value && len) {
2559                 rc = security_sid_to_context_force(newsid, &context, &clen);
2560                 if (rc) {
2561                         kfree(namep);
2562                         return rc;
2563                 }
2564                 *value = context;
2565                 *len = clen;
2566         }
2567
2568         return 0;
2569 }
2570
2571 static int selinux_inode_create(struct inode *dir, struct dentry *dentry, int mask)
2572 {
2573         return may_create(dir, dentry, SECCLASS_FILE);
2574 }
2575
2576 static int selinux_inode_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2577 {
2578         return may_link(dir, old_dentry, MAY_LINK);
2579 }
2580
2581 static int selinux_inode_unlink(struct inode *dir, struct dentry *dentry)
2582 {
2583         return may_link(dir, dentry, MAY_UNLINK);
2584 }
2585
2586 static int selinux_inode_symlink(struct inode *dir, struct dentry *dentry, const char *name)
2587 {
2588         return may_create(dir, dentry, SECCLASS_LNK_FILE);
2589 }
2590
2591 static int selinux_inode_mkdir(struct inode *dir, struct dentry *dentry, int mask)
2592 {
2593         return may_create(dir, dentry, SECCLASS_DIR);
2594 }
2595
2596 static int selinux_inode_rmdir(struct inode *dir, struct dentry *dentry)
2597 {
2598         return may_link(dir, dentry, MAY_RMDIR);
2599 }
2600
2601 static int selinux_inode_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2602 {
2603         return may_create(dir, dentry, inode_mode_to_security_class(mode));
2604 }
2605
2606 static int selinux_inode_rename(struct inode *old_inode, struct dentry *old_dentry,
2607                                 struct inode *new_inode, struct dentry *new_dentry)
2608 {
2609         return may_rename(old_inode, old_dentry, new_inode, new_dentry);
2610 }
2611
2612 static int selinux_inode_readlink(struct dentry *dentry)
2613 {
2614         const struct cred *cred = current_cred();
2615
2616         return dentry_has_perm(cred, NULL, dentry, FILE__READ);
2617 }
2618
2619 static int selinux_inode_follow_link(struct dentry *dentry, struct nameidata *nameidata)
2620 {
2621         const struct cred *cred = current_cred();
2622
2623         return dentry_has_perm(cred, NULL, dentry, FILE__READ);
2624 }
2625
2626 static int selinux_inode_permission(struct inode *inode, int mask)
2627 {
2628         const struct cred *cred = current_cred();
2629         struct common_audit_data ad;
2630         u32 perms;
2631         bool from_access;
2632
2633         from_access = mask & MAY_ACCESS;
2634         mask &= (MAY_READ|MAY_WRITE|MAY_EXEC|MAY_APPEND);
2635
2636         /* No permission to check.  Existence test. */
2637         if (!mask)
2638                 return 0;
2639
2640         COMMON_AUDIT_DATA_INIT(&ad, FS);
2641         ad.u.fs.inode = inode;
2642
2643         if (from_access)
2644                 ad.selinux_audit_data.auditdeny |= FILE__AUDIT_ACCESS;
2645
2646         perms = file_mask_to_av(inode->i_mode, mask);
2647
2648         return inode_has_perm(cred, inode, perms, &ad);
2649 }
2650
2651 static int selinux_inode_setattr(struct dentry *dentry, struct iattr *iattr)
2652 {
2653         const struct cred *cred = current_cred();
2654         unsigned int ia_valid = iattr->ia_valid;
2655
2656         /* ATTR_FORCE is just used for ATTR_KILL_S[UG]ID. */
2657         if (ia_valid & ATTR_FORCE) {
2658                 ia_valid &= ~(ATTR_KILL_SUID | ATTR_KILL_SGID | ATTR_MODE |
2659                               ATTR_FORCE);
2660                 if (!ia_valid)
2661                         return 0;
2662         }
2663
2664         if (ia_valid & (ATTR_MODE | ATTR_UID | ATTR_GID |
2665                         ATTR_ATIME_SET | ATTR_MTIME_SET | ATTR_TIMES_SET))
2666                 return dentry_has_perm(cred, NULL, dentry, FILE__SETATTR);
2667
2668         return dentry_has_perm(cred, NULL, dentry, FILE__WRITE);
2669 }
2670
2671 static int selinux_inode_getattr(struct vfsmount *mnt, struct dentry *dentry)
2672 {
2673         const struct cred *cred = current_cred();
2674
2675         return dentry_has_perm(cred, mnt, dentry, FILE__GETATTR);
2676 }
2677
2678 static int selinux_inode_setotherxattr(struct dentry *dentry, const char *name)
2679 {
2680         const struct cred *cred = current_cred();
2681
2682         if (!strncmp(name, XATTR_SECURITY_PREFIX,
2683                      sizeof XATTR_SECURITY_PREFIX - 1)) {
2684                 if (!strcmp(name, XATTR_NAME_CAPS)) {
2685                         if (!capable(CAP_SETFCAP))
2686                                 return -EPERM;
2687                 } else if (!capable(CAP_SYS_ADMIN)) {
2688                         /* A different attribute in the security namespace.
2689                            Restrict to administrator. */
2690                         return -EPERM;
2691                 }
2692         }
2693
2694         /* Not an attribute we recognize, so just check the
2695            ordinary setattr permission. */
2696         return dentry_has_perm(cred, NULL, dentry, FILE__SETATTR);
2697 }
2698
2699 static int selinux_inode_setxattr(struct dentry *dentry, const char *name,
2700                                   const void *value, size_t size, int flags)
2701 {
2702         struct inode *inode = dentry->d_inode;
2703         struct inode_security_struct *isec = inode->i_security;
2704         struct superblock_security_struct *sbsec;
2705         struct common_audit_data ad;
2706         u32 newsid, sid = current_sid();
2707         int rc = 0;
2708
2709         if (strcmp(name, XATTR_NAME_SELINUX))
2710                 return selinux_inode_setotherxattr(dentry, name);
2711
2712         sbsec = inode->i_sb->s_security;
2713         if (!(sbsec->flags & SE_SBLABELSUPP))
2714                 return -EOPNOTSUPP;
2715
2716         if (!is_owner_or_cap(inode))
2717                 return -EPERM;
2718
2719         COMMON_AUDIT_DATA_INIT(&ad, FS);
2720         ad.u.fs.path.dentry = dentry;
2721
2722         rc = avc_has_perm(sid, isec->sid, isec->sclass,
2723                           FILE__RELABELFROM, &ad);
2724         if (rc)
2725                 return rc;
2726
2727         rc = security_context_to_sid(value, size, &newsid);
2728         if (rc == -EINVAL) {
2729                 if (!capable(CAP_MAC_ADMIN))
2730                         return rc;
2731                 rc = security_context_to_sid_force(value, size, &newsid);
2732         }
2733         if (rc)
2734                 return rc;
2735
2736         rc = avc_has_perm(sid, newsid, isec->sclass,
2737                           FILE__RELABELTO, &ad);
2738         if (rc)
2739                 return rc;
2740
2741         rc = security_validate_transition(isec->sid, newsid, sid,
2742                                           isec->sclass);
2743         if (rc)
2744                 return rc;
2745
2746         return avc_has_perm(newsid,
2747                             sbsec->sid,
2748                             SECCLASS_FILESYSTEM,
2749                             FILESYSTEM__ASSOCIATE,
2750                             &ad);
2751 }
2752
2753 static void selinux_inode_post_setxattr(struct dentry *dentry, const char *name,
2754                                         const void *value, size_t size,
2755                                         int flags)
2756 {
2757         struct inode *inode = dentry->d_inode;
2758         struct inode_security_struct *isec = inode->i_security;
2759         u32 newsid;
2760         int rc;
2761
2762         if (strcmp(name, XATTR_NAME_SELINUX)) {
2763                 /* Not an attribute we recognize, so nothing to do. */
2764                 return;
2765         }
2766
2767         rc = security_context_to_sid_force(value, size, &newsid);
2768         if (rc) {
2769                 printk(KERN_ERR "SELinux:  unable to map context to SID"
2770                        "for (%s, %lu), rc=%d\n",
2771                        inode->i_sb->s_id, inode->i_ino, -rc);
2772                 return;
2773         }
2774
2775         isec->sid = newsid;
2776         return;
2777 }
2778
2779 static int selinux_inode_getxattr(struct dentry *dentry, const char *name)
2780 {
2781         const struct cred *cred = current_cred();
2782
2783         return dentry_has_perm(cred, NULL, dentry, FILE__GETATTR);
2784 }
2785
2786 static int selinux_inode_listxattr(struct dentry *dentry)
2787 {
2788         const struct cred *cred = current_cred();
2789
2790         return dentry_has_perm(cred, NULL, dentry, FILE__GETATTR);
2791 }
2792
2793 static int selinux_inode_removexattr(struct dentry *dentry, const char *name)
2794 {
2795         if (strcmp(name, XATTR_NAME_SELINUX))
2796                 return selinux_inode_setotherxattr(dentry, name);
2797
2798         /* No one is allowed to remove a SELinux security label.
2799            You can change the label, but all data must be labeled. */
2800         return -EACCES;
2801 }
2802
2803 /*
2804  * Copy the inode security context value to the user.
2805  *
2806  * Permission check is handled by selinux_inode_getxattr hook.
2807  */
2808 static int selinux_inode_getsecurity(const struct inode *inode, const char *name, void **buffer, bool alloc)
2809 {
2810         u32 size;
2811         int error;
2812         char *context = NULL;
2813         struct inode_security_struct *isec = inode->i_security;
2814
2815         if (strcmp(name, XATTR_SELINUX_SUFFIX))
2816                 return -EOPNOTSUPP;
2817
2818         /*
2819          * If the caller has CAP_MAC_ADMIN, then get the raw context
2820          * value even if it is not defined by current policy; otherwise,
2821          * use the in-core value under current policy.
2822          * Use the non-auditing forms of the permission checks since
2823          * getxattr may be called by unprivileged processes commonly
2824          * and lack of permission just means that we fall back to the
2825          * in-core context value, not a denial.
2826          */
2827         error = selinux_capable(current, current_cred(), CAP_MAC_ADMIN,
2828                                 SECURITY_CAP_NOAUDIT);
2829         if (!error)
2830                 error = security_sid_to_context_force(isec->sid, &context,
2831                                                       &size);
2832         else
2833                 error = security_sid_to_context(isec->sid, &context, &size);
2834         if (error)
2835                 return error;
2836         error = size;
2837         if (alloc) {
2838                 *buffer = context;
2839                 goto out_nofree;
2840         }
2841         kfree(context);
2842 out_nofree:
2843         return error;
2844 }
2845
2846 static int selinux_inode_setsecurity(struct inode *inode, const char *name,
2847                                      const void *value, size_t size, int flags)
2848 {
2849         struct inode_security_struct *isec = inode->i_security;
2850         u32 newsid;
2851         int rc;
2852
2853         if (strcmp(name, XATTR_SELINUX_SUFFIX))
2854                 return -EOPNOTSUPP;
2855
2856         if (!value || !size)
2857                 return -EACCES;
2858
2859         rc = security_context_to_sid((void *)value, size, &newsid);
2860         if (rc)
2861                 return rc;
2862
2863         isec->sid = newsid;
2864         isec->initialized = 1;
2865         return 0;
2866 }
2867
2868 static int selinux_inode_listsecurity(struct inode *inode, char *buffer, size_t buffer_size)
2869 {
2870         const int len = sizeof(XATTR_NAME_SELINUX);
2871         if (buffer && len <= buffer_size)
2872                 memcpy(buffer, XATTR_NAME_SELINUX, len);
2873         return len;
2874 }
2875
2876 static void selinux_inode_getsecid(const struct inode *inode, u32 *secid)
2877 {
2878         struct inode_security_struct *isec = inode->i_security;
2879         *secid = isec->sid;
2880 }
2881
2882 /* file security operations */
2883
2884 static int selinux_revalidate_file_permission(struct file *file, int mask)
2885 {
2886         const struct cred *cred = current_cred();
2887         struct inode *inode = file->f_path.dentry->d_inode;
2888
2889         /* file_mask_to_av won't add FILE__WRITE if MAY_APPEND is set */
2890         if ((file->f_flags & O_APPEND) && (mask & MAY_WRITE))
2891                 mask |= MAY_APPEND;
2892
2893         return file_has_perm(cred, file,
2894                              file_mask_to_av(inode->i_mode, mask));
2895 }
2896
2897 static int selinux_file_permission(struct file *file, int mask)
2898 {
2899         struct inode *inode = file->f_path.dentry->d_inode;
2900         struct file_security_struct *fsec = file->f_security;
2901         struct inode_security_struct *isec = inode->i_security;
2902         u32 sid = current_sid();
2903
2904         if (!mask)
2905                 /* No permission to check.  Existence test. */
2906                 return 0;
2907
2908         if (sid == fsec->sid && fsec->isid == isec->sid &&
2909             fsec->pseqno == avc_policy_seqno())
2910                 /* No change since dentry_open check. */
2911                 return 0;
2912
2913         return selinux_revalidate_file_permission(file, mask);
2914 }
2915
2916 static int selinux_file_alloc_security(struct file *file)
2917 {
2918         return file_alloc_security(file);
2919 }
2920
2921 static void selinux_file_free_security(struct file *file)
2922 {
2923         file_free_security(file);
2924 }
2925
2926 static int selinux_file_ioctl(struct file *file, unsigned int cmd,
2927                               unsigned long arg)
2928 {
2929         const struct cred *cred = current_cred();
2930         u32 av = 0;
2931
2932         if (_IOC_DIR(cmd) & _IOC_WRITE)
2933                 av |= FILE__WRITE;
2934         if (_IOC_DIR(cmd) & _IOC_READ)
2935                 av |= FILE__READ;
2936         if (!av)
2937                 av = FILE__IOCTL;
2938
2939         return file_has_perm(cred, file, av);
2940 }
2941
2942 static int default_noexec;
2943
2944 static int file_map_prot_check(struct file *file, unsigned long prot, int shared)
2945 {
2946         const struct cred *cred = current_cred();
2947         int rc = 0;
2948
2949         if (default_noexec &&
2950             (prot & PROT_EXEC) && (!file || (!shared && (prot & PROT_WRITE)))) {
2951                 /*
2952                  * We are making executable an anonymous mapping or a
2953                  * private file mapping that will also be writable.
2954                  * This has an additional check.
2955                  */
2956                 rc = cred_has_perm(cred, cred, PROCESS__EXECMEM);
2957                 if (rc)
2958                         goto error;
2959         }
2960
2961         if (file) {
2962                 /* read access is always possible with a mapping */
2963                 u32 av = FILE__READ;
2964
2965                 /* write access only matters if the mapping is shared */
2966                 if (shared && (prot & PROT_WRITE))
2967                         av |= FILE__WRITE;
2968
2969                 if (prot & PROT_EXEC)
2970                         av |= FILE__EXECUTE;
2971
2972                 return file_has_perm(cred, file, av);
2973         }
2974
2975 error:
2976         return rc;
2977 }
2978
2979 static int selinux_file_mmap(struct file *file, unsigned long reqprot,
2980                              unsigned long prot, unsigned long flags,
2981                              unsigned long addr, unsigned long addr_only)
2982 {
2983         int rc = 0;
2984         u32 sid = current_sid();
2985
2986         /*
2987          * notice that we are intentionally putting the SELinux check before
2988          * the secondary cap_file_mmap check.  This is such a likely attempt
2989          * at bad behaviour/exploit that we always want to get the AVC, even
2990          * if DAC would have also denied the operation.
2991          */
2992         if (addr < CONFIG_LSM_MMAP_MIN_ADDR) {
2993                 rc = avc_has_perm(sid, sid, SECCLASS_MEMPROTECT,
2994                                   MEMPROTECT__MMAP_ZERO, NULL);
2995                 if (rc)
2996                         return rc;
2997         }
2998
2999         /* do DAC check on address space usage */
3000         rc = cap_file_mmap(file, reqprot, prot, flags, addr, addr_only);
3001         if (rc || addr_only)
3002                 return rc;
3003
3004         if (selinux_checkreqprot)
3005                 prot = reqprot;
3006
3007         return file_map_prot_check(file, prot,
3008                                    (flags & MAP_TYPE) == MAP_SHARED);
3009 }
3010
3011 static int selinux_file_mprotect(struct vm_area_struct *vma,
3012                                  unsigned long reqprot,
3013                                  unsigned long prot)
3014 {
3015         const struct cred *cred = current_cred();
3016
3017         if (selinux_checkreqprot)
3018                 prot = reqprot;
3019
3020         if (default_noexec &&
3021             (prot & PROT_EXEC) && !(vma->vm_flags & VM_EXEC)) {
3022                 int rc = 0;
3023                 if (vma->vm_start >= vma->vm_mm->start_brk &&
3024                     vma->vm_end <= vma->vm_mm->brk) {
3025                         rc = cred_has_perm(cred, cred, PROCESS__EXECHEAP);
3026                 } else if (!vma->vm_file &&
3027                            vma->vm_start <= vma->vm_mm->start_stack &&
3028                            vma->vm_end >= vma->vm_mm->start_stack) {
3029                         rc = current_has_perm(current, PROCESS__EXECSTACK);
3030                 } else if (vma->vm_file && vma->anon_vma) {
3031                         /*
3032                          * We are making executable a file mapping that has
3033                          * had some COW done. Since pages might have been
3034                          * written, check ability to execute the possibly
3035                          * modified content.  This typically should only
3036                          * occur for text relocations.
3037                          */
3038                         rc = file_has_perm(cred, vma->vm_file, FILE__EXECMOD);
3039                 }
3040                 if (rc)
3041                         return rc;
3042         }
3043
3044         return file_map_prot_check(vma->vm_file, prot, vma->vm_flags&VM_SHARED);
3045 }
3046
3047 static int selinux_file_lock(struct file *file, unsigned int cmd)
3048 {
3049         const struct cred *cred = current_cred();
3050
3051         return file_has_perm(cred, file, FILE__LOCK);
3052 }
3053
3054 static int selinux_file_fcntl(struct file *file, unsigned int cmd,
3055                               unsigned long arg)
3056 {
3057         const struct cred *cred = current_cred();
3058         int err = 0;
3059
3060         switch (cmd) {
3061         case F_SETFL:
3062                 if (!file->f_path.dentry || !file->f_path.dentry->d_inode) {
3063                         err = -EINVAL;
3064                         break;
3065                 }
3066
3067                 if ((file->f_flags & O_APPEND) && !(arg & O_APPEND)) {
3068                         err = file_has_perm(cred, file, FILE__WRITE);
3069                         break;
3070                 }
3071                 /* fall through */
3072         case F_SETOWN:
3073         case F_SETSIG:
3074         case F_GETFL:
3075         case F_GETOWN:
3076         case F_GETSIG:
3077                 /* Just check FD__USE permission */
3078                 err = file_has_perm(cred, file, 0);
3079                 break;
3080         case F_GETLK:
3081         case F_SETLK:
3082         case F_SETLKW:
3083 #if BITS_PER_LONG == 32
3084         case F_GETLK64:
3085         case F_SETLK64:
3086         case F_SETLKW64:
3087 #endif
3088                 if (!file->f_path.dentry || !file->f_path.dentry->d_inode) {
3089                         err = -EINVAL;
3090                         break;
3091                 }
3092                 err = file_has_perm(cred, file, FILE__LOCK);
3093                 break;
3094         }
3095
3096         return err;
3097 }
3098
3099 static int selinux_file_set_fowner(struct file *file)
3100 {
3101         struct file_security_struct *fsec;
3102
3103         fsec = file->f_security;
3104         fsec->fown_sid = current_sid();
3105
3106         return 0;
3107 }
3108
3109 static int selinux_file_send_sigiotask(struct task_struct *tsk,
3110                                        struct fown_struct *fown, int signum)
3111 {
3112         struct file *file;
3113         u32 sid = task_sid(tsk);
3114         u32 perm;
3115         struct file_security_struct *fsec;
3116
3117         /* struct fown_struct is never outside the context of a struct file */
3118         file = container_of(fown, struct file, f_owner);
3119
3120         fsec = file->f_security;
3121
3122         if (!signum)
3123                 perm = signal_to_av(SIGIO); /* as per send_sigio_to_task */
3124         else
3125                 perm = signal_to_av(signum);
3126
3127         return avc_has_perm(fsec->fown_sid, sid,
3128                             SECCLASS_PROCESS, perm, NULL);
3129 }
3130
3131 static int selinux_file_receive(struct file *file)
3132 {
3133         const struct cred *cred = current_cred();
3134
3135         return file_has_perm(cred, file, file_to_av(file));
3136 }
3137
3138 static int selinux_dentry_open(struct file *file, const struct cred *cred)
3139 {
3140         struct file_security_struct *fsec;
3141         struct inode *inode;
3142         struct inode_security_struct *isec;
3143
3144         inode = file->f_path.dentry->d_inode;
3145         fsec = file->f_security;
3146         isec = inode->i_security;
3147         /*
3148          * Save inode label and policy sequence number
3149          * at open-time so that selinux_file_permission
3150          * can determine whether revalidation is necessary.
3151          * Task label is already saved in the file security
3152          * struct as its SID.
3153          */
3154         fsec->isid = isec->sid;
3155         fsec->pseqno = avc_policy_seqno();
3156         /*
3157          * Since the inode label or policy seqno may have changed
3158          * between the selinux_inode_permission check and the saving
3159          * of state above, recheck that access is still permitted.
3160          * Otherwise, access might never be revalidated against the
3161          * new inode label or new policy.
3162          * This check is not redundant - do not remove.
3163          */
3164         return inode_has_perm(cred, inode, open_file_to_av(file), NULL);
3165 }
3166
3167 /* task security operations */
3168
3169 static int selinux_task_create(unsigned long clone_flags)
3170 {
3171         return current_has_perm(current, PROCESS__FORK);
3172 }
3173
3174 /*
3175  * allocate the SELinux part of blank credentials
3176  */
3177 static int selinux_cred_alloc_blank(struct cred *cred, gfp_t gfp)
3178 {
3179         struct task_security_struct *tsec;
3180
3181         tsec = kzalloc(sizeof(struct task_security_struct), gfp);
3182         if (!tsec)
3183                 return -ENOMEM;
3184
3185         cred->security = tsec;
3186         return 0;
3187 }
3188
3189 /*
3190  * detach and free the LSM part of a set of credentials
3191  */
3192 static void selinux_cred_free(struct cred *cred)