security/commoncap.c

   1 /* Common capabilities, needed by capability.o and root_plug.o
   2  *
   3  *      This program is free software; you can redistribute it and/or modify
   4  *      it under the terms of the GNU General Public License as published by
   5  *      the Free Software Foundation; either version 2 of the License, or
   6  *      (at your option) any later version.
   7  *
   8  */
   9
  10 #include <linux/capability.h>
  11 #include <linux/module.h>
  12 #include <linux/init.h>
  13 #include <linux/kernel.h>
  14 #include <linux/security.h>
  15 #include <linux/file.h>
  16 #include <linux/mm.h>
  17 #include <linux/mman.h>
  18 #include <linux/pagemap.h>
  19 #include <linux/swap.h>
  20 #include <linux/skbuff.h>
  21 #include <linux/netlink.h>
  22 #include <linux/ptrace.h>
  23 #include <linux/xattr.h>
  24 #include <linux/hugetlb.h>
  25 #include <linux/mount.h>
  26
  27 int cap_netlink_send(struct sock *sk, struct sk_buff *skb)
  28 {
  29         NETLINK_CB(skb).eff_cap = current->cap_effective;
  30         return 0;
  31 }
  32
  33 int cap_netlink_recv(struct sk_buff *skb, int cap)
  34 {
  35         if (!cap_raised(NETLINK_CB(skb).eff_cap, cap))
  36                 return -EPERM;
  37         return 0;
  38 }
  39
  40 EXPORT_SYMBOL(cap_netlink_recv);
  41
  42 int cap_capable (struct task_struct *tsk, int cap)
  43 {
  44         /* Derived from include/linux/sched.h:capable. */
  45         if (cap_raised(tsk->cap_effective, cap))
  46                 return 0;
  47         return -EPERM;
  48 }
  49
  50 int cap_settime(struct timespec *ts, struct timezone *tz)
  51 {
  52         if (!capable(CAP_SYS_TIME))
  53                 return -EPERM;
  54         return 0;
  55 }
  56
  57 int cap_ptrace (struct task_struct *parent, struct task_struct *child)
  58 {
  59         /* Derived from arch/i386/kernel/ptrace.c:sys_ptrace. */
  60         if (!cap_issubset(child->cap_permitted, parent->cap_permitted) &&
  61             !__capable(parent, CAP_SYS_PTRACE))
  62                 return -EPERM;
  63         return 0;
  64 }
  65
  66 int cap_capget (struct task_struct *target, kernel_cap_t *effective,
  67                 kernel_cap_t *inheritable, kernel_cap_t *permitted)
  68 {
  69         /* Derived from kernel/capability.c:sys_capget. */
  70         *effective = cap_t (target->cap_effective);
  71         *inheritable = cap_t (target->cap_inheritable);
  72         *permitted = cap_t (target->cap_permitted);
  73         return 0;
  74 }
  75
  76 int cap_capset_check (struct task_struct *target, kernel_cap_t *effective,
  77                       kernel_cap_t *inheritable, kernel_cap_t *permitted)
  78 {
  79         /* Derived from kernel/capability.c:sys_capset. */
  80         /* verify restrictions on target's new Inheritable set */
  81         if (!cap_issubset (*inheritable,
  82                            cap_combine (target->cap_inheritable,
  83                                         current->cap_permitted))) {
  84                 return -EPERM;
  85         }
  86
  87         /* verify restrictions on target's new Permitted set */
  88         if (!cap_issubset (*permitted,
  89                            cap_combine (target->cap_permitted,
  90                                         current->cap_permitted))) {
  91                 return -EPERM;
  92         }
  93
  94         /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
  95         if (!cap_issubset (*effective, *permitted)) {
  96                 return -EPERM;
  97         }
  98
  99         return 0;
 100 }
 101
 102 void cap_capset_set (struct task_struct *target, kernel_cap_t *effective,
 103                      kernel_cap_t *inheritable, kernel_cap_t *permitted)
 104 {
 105         target->cap_effective = *effective;
 106         target->cap_inheritable = *inheritable;
 107         target->cap_permitted = *permitted;
 108 }
 109
 110 static inline void bprm_clear_caps(struct linux_binprm *bprm)
 111 {
 112         cap_clear(bprm->cap_inheritable);
 113         cap_clear(bprm->cap_permitted);
 114         bprm->cap_effective = false;
 115 }
 116
 117 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
 118
 119 int cap_inode_need_killpriv(struct dentry *dentry)
 120 {
 121         struct inode *inode = dentry->d_inode;
 122         int error;
 123
 124         if (!inode->i_op || !inode->i_op->getxattr)
 125                return 0;
 126
 127         error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
 128         if (error <= 0)
 129                 return 0;
 130         return 1;
 131 }
 132
 133 int cap_inode_killpriv(struct dentry *dentry)
 134 {
 135         struct inode *inode = dentry->d_inode;
 136
 137         if (!inode->i_op || !inode->i_op->removexattr)
 138                return 0;
 139
 140         return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
 141 }
 142
 143 static inline int cap_from_disk(__le32 *caps, struct linux_binprm *bprm,
 144                                 int size)
 145 {
 146         __u32 magic_etc;
 147
 148         if (size != XATTR_CAPS_SZ)
 149                 return -EINVAL;
 150
 151         magic_etc = le32_to_cpu(caps[0]);
 152
 153         switch ((magic_etc & VFS_CAP_REVISION_MASK)) {
 154         case VFS_CAP_REVISION:
 155                 if (magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
 156                         bprm->cap_effective = true;
 157                 else
 158                         bprm->cap_effective = false;
 159                 bprm->cap_permitted = to_cap_t( le32_to_cpu(caps[1]) );
 160                 bprm->cap_inheritable = to_cap_t( le32_to_cpu(caps[2]) );
 161                 return 0;
 162         default:
 163                 return -EINVAL;
 164         }
 165 }
 166
 167 /* Locate any VFS capabilities: */
 168 static int get_file_caps(struct linux_binprm *bprm)
 169 {
 170         struct dentry *dentry;
 171         int rc = 0;
 172         __le32 v1caps[XATTR_CAPS_SZ];
 173         struct inode *inode;
 174
 175         if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID) {
 176                 bprm_clear_caps(bprm);
 177                 return 0;
 178         }
 179
 180         dentry = dget(bprm->file->f_dentry);
 181         inode = dentry->d_inode;
 182         if (!inode->i_op || !inode->i_op->getxattr)
 183                 goto out;
 184
 185         rc = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, &v1caps,
 186                                                         XATTR_CAPS_SZ);
 187         if (rc == -ENODATA || rc == -EOPNOTSUPP) {
 188                 /* no data, that's ok */
 189                 rc = 0;
 190                 goto out;
 191         }
 192         if (rc < 0)
 193                 goto out;
 194
 195         rc = cap_from_disk(v1caps, bprm, rc);
 196         if (rc)
 197                 printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
 198                         __FUNCTION__, rc, bprm->filename);
 199
 200 out:
 201         dput(dentry);
 202         if (rc)
 203                 bprm_clear_caps(bprm);
 204
 205         return rc;
 206 }
 207
 208 #else
 209 int cap_inode_need_killpriv(struct dentry *dentry)
 210 {
 211         return 0;
 212 }
 213
 214 int cap_inode_killpriv(struct dentry *dentry)
 215 {
 216         return 0;
 217 }
 218
 219 static inline int get_file_caps(struct linux_binprm *bprm)
 220 {
 221         bprm_clear_caps(bprm);
 222         return 0;
 223 }
 224 #endif
 225
 226 int cap_bprm_set_security (struct linux_binprm *bprm)
 227 {
 228         int ret;
 229
 230         ret = get_file_caps(bprm);
 231         if (ret)
 232                 printk(KERN_NOTICE "%s: get_file_caps returned %d for %s\n",
 233                         __FUNCTION__, ret, bprm->filename);
 234
 235         /*  To support inheritance of root-permissions and suid-root
 236          *  executables under compatibility mode, we raise all three
 237          *  capability sets for the file.
 238          *
 239          *  If only the real uid is 0, we only raise the inheritable
 240          *  and permitted sets of the executable file.
 241          */
 242
 243         if (!issecure (SECURE_NOROOT)) {
 244                 if (bprm->e_uid == 0 || current->uid == 0) {
 245                         cap_set_full (bprm->cap_inheritable);
 246                         cap_set_full (bprm->cap_permitted);
 247                 }
 248                 if (bprm->e_uid == 0)
 249                         bprm->cap_effective = true;
 250         }
 251
 252         return ret;
 253 }
 254
 255 void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
 256 {
 257         /* Derived from fs/exec.c:compute_creds. */
 258         kernel_cap_t new_permitted, working;
 259
 260         new_permitted = cap_intersect (bprm->cap_permitted, cap_bset);
 261         working = cap_intersect (bprm->cap_inheritable,
 262                                  current->cap_inheritable);
 263         new_permitted = cap_combine (new_permitted, working);
 264
 265         if (bprm->e_uid != current->uid || bprm->e_gid != current->gid ||
 266             !cap_issubset (new_permitted, current->cap_permitted)) {
 267                 set_dumpable(current->mm, suid_dumpable);
 268                 current->pdeath_signal = 0;
 269
 270                 if (unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
 271                         if (!capable(CAP_SETUID)) {
 272                                 bprm->e_uid = current->uid;
 273                                 bprm->e_gid = current->gid;
 274                         }
 275                         if (!capable (CAP_SETPCAP)) {
 276                                 new_permitted = cap_intersect (new_permitted,
 277                                                         current->cap_permitted);
 278                         }
 279                 }
 280         }
 281
 282         current->suid = current->euid = current->fsuid = bprm->e_uid;
 283         current->sgid = current->egid = current->fsgid = bprm->e_gid;
 284
 285         /* For init, we want to retain the capabilities set
 286          * in the init_task struct. Thus we skip the usual
 287          * capability rules */
 288         if (!is_init(current)) {
 289                 current->cap_permitted = new_permitted;
 290                 current->cap_effective = bprm->cap_effective ?
 291                                 new_permitted : 0;
 292         }
 293
 294         /* AUD: Audit candidate if current->cap_effective is set */
 295
 296         current->keep_capabilities = 0;
 297 }
 298
 299 int cap_bprm_secureexec (struct linux_binprm *bprm)
 300 {
 301         if (current->uid != 0) {
 302                 if (bprm->cap_effective)
 303                         return 1;
 304                 if (!cap_isclear(bprm->cap_permitted))
 305                         return 1;
 306                 if (!cap_isclear(bprm->cap_inheritable))
 307                         return 1;
 308         }
 309
 310         return (current->euid != current->uid ||
 311                 current->egid != current->gid);
 312 }
 313
 314 int cap_inode_setxattr(struct dentry *dentry, char *name, void *value,
 315                        size_t size, int flags)
 316 {
 317         if (!strcmp(name, XATTR_NAME_CAPS)) {
 318                 if (!capable(CAP_SETFCAP))
 319                         return -EPERM;
 320                 return 0;
 321         } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
 322                      sizeof(XATTR_SECURITY_PREFIX) - 1)  &&
 323             !capable(CAP_SYS_ADMIN))
 324                 return -EPERM;
 325         return 0;
 326 }
 327
 328 int cap_inode_removexattr(struct dentry *dentry, char *name)
 329 {
 330         if (!strcmp(name, XATTR_NAME_CAPS)) {
 331                 if (!capable(CAP_SETFCAP))
 332                         return -EPERM;
 333                 return 0;
 334         } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
 335                      sizeof(XATTR_SECURITY_PREFIX) - 1)  &&
 336             !capable(CAP_SYS_ADMIN))
 337                 return -EPERM;
 338         return 0;
 339 }
 340
 341 /* moved from kernel/sys.c. */
 342 /*
 343  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
 344  * a process after a call to setuid, setreuid, or setresuid.
 345  *
 346  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
 347  *  {r,e,s}uid != 0, the permitted and effective capabilities are
 348  *  cleared.
 349  *
 350  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
 351  *  capabilities of the process are cleared.
 352  *
 353  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
 354  *  capabilities are set to the permitted capabilities.
 355  *
 356  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
 357  *  never happen.
 358  *
 359  *  -astor
 360  *
 361  * cevans - New behaviour, Oct '99
 362  * A process may, via prctl(), elect to keep its capabilities when it
 363  * calls setuid() and switches away from uid==0. Both permitted and
 364  * effective sets will be retained.
 365  * Without this change, it was impossible for a daemon to drop only some
 366  * of its privilege. The call to setuid(!=0) would drop all privileges!
 367  * Keeping uid 0 is not an option because uid 0 owns too many vital
 368  * files..
 369  * Thanks to Olaf Kirch and Peter Benie for spotting this.
 370  */
 371 static inline void cap_emulate_setxuid (int old_ruid, int old_euid,
 372                                         int old_suid)
 373 {
 374         if ((old_ruid == 0 || old_euid == 0 || old_suid == 0) &&
 375             (current->uid != 0 && current->euid != 0 && current->suid != 0) &&
 376             !current->keep_capabilities) {
 377                 cap_clear (current->cap_permitted);
 378                 cap_clear (current->cap_effective);
 379         }
 380         if (old_euid == 0 && current->euid != 0) {
 381                 cap_clear (current->cap_effective);
 382         }
 383         if (old_euid != 0 && current->euid == 0) {
 384                 current->cap_effective = current->cap_permitted;
 385         }
 386 }
 387
 388 int cap_task_post_setuid (uid_t old_ruid, uid_t old_euid, uid_t old_suid,
 389                           int flags)
 390 {
 391         switch (flags) {
 392         case LSM_SETID_RE:
 393         case LSM_SETID_ID:
 394         case LSM_SETID_RES:
 395                 /* Copied from kernel/sys.c:setreuid/setuid/setresuid. */
 396                 if (!issecure (SECURE_NO_SETUID_FIXUP)) {
 397                         cap_emulate_setxuid (old_ruid, old_euid, old_suid);
 398                 }
 399                 break;
 400         case LSM_SETID_FS:
 401                 {
 402                         uid_t old_fsuid = old_ruid;
 403
 404                         /* Copied from kernel/sys.c:setfsuid. */
 405
 406                         /*
 407                          * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
 408                          *          if not, we might be a bit too harsh here.
 409                          */
 410
 411                         if (!issecure (SECURE_NO_SETUID_FIXUP)) {
 412                                 if (old_fsuid == 0 && current->fsuid != 0) {
 413                                         cap_t (current->cap_effective) &=
 414                                             ~CAP_FS_MASK;
 415                                 }
 416                                 if (old_fsuid != 0 && current->fsuid == 0) {
 417                                         cap_t (current->cap_effective) |=
 418                                             (cap_t (current->cap_permitted) &
 419                                              CAP_FS_MASK);
 420                                 }
 421                         }
 422                         break;
 423                 }
 424         default:
 425                 return -EINVAL;
 426         }
 427
 428         return 0;
 429 }
 430
 431 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
 432 /*
 433  * Rationale: code calling task_setscheduler, task_setioprio, and
 434  * task_setnice, assumes that
 435  *   . if capable(cap_sys_nice), then those actions should be allowed
 436  *   . if not capable(cap_sys_nice), but acting on your own processes,
 437  *      then those actions should be allowed
 438  * This is insufficient now since you can call code without suid, but
 439  * yet with increased caps.
 440  * So we check for increased caps on the target process.
 441  */
 442 static inline int cap_safe_nice(struct task_struct *p)
 443 {
 444         if (!cap_issubset(p->cap_permitted, current->cap_permitted) &&
 445             !__capable(current, CAP_SYS_NICE))
 446                 return -EPERM;
 447         return 0;
 448 }
 449
 450 int cap_task_setscheduler (struct task_struct *p, int policy,
 451                            struct sched_param *lp)
 452 {
 453         return cap_safe_nice(p);
 454 }
 455
 456 int cap_task_setioprio (struct task_struct *p, int ioprio)
 457 {
 458         return cap_safe_nice(p);
 459 }
 460
 461 int cap_task_setnice (struct task_struct *p, int nice)
 462 {
 463         return cap_safe_nice(p);
 464 }
 465
 466 int cap_task_kill(struct task_struct *p, struct siginfo *info,
 467                                 int sig, u32 secid)
 468 {
 469         if (info != SEND_SIG_NOINFO && (is_si_special(info) || SI_FROMKERNEL(info)))
 470                 return 0;
 471
 472         if (secid)
 473                 /*
 474                  * Signal sent as a particular user.
 475                  * Capabilities are ignored.  May be wrong, but it's the
 476                  * only thing we can do at the moment.
 477                  * Used only by usb drivers?
 478                  */
 479                 return 0;
 480         if (cap_issubset(p->cap_permitted, current->cap_permitted))
 481                 return 0;
 482         if (capable(CAP_KILL))
 483                 return 0;
 484
 485         return -EPERM;
 486 }
 487 #else
 488 int cap_task_setscheduler (struct task_struct *p, int policy,
 489                            struct sched_param *lp)
 490 {
 491         return 0;
 492 }
 493 int cap_task_setioprio (struct task_struct *p, int ioprio)
 494 {
 495         return 0;
 496 }
 497 int cap_task_setnice (struct task_struct *p, int nice)
 498 {
 499         return 0;
 500 }
 501 int cap_task_kill(struct task_struct *p, struct siginfo *info,
 502                                 int sig, u32 secid)
 503 {
 504         return 0;
 505 }
 506 #endif
 507
 508 void cap_task_reparent_to_init (struct task_struct *p)
 509 {
 510         p->cap_effective = CAP_INIT_EFF_SET;
 511         p->cap_inheritable = CAP_INIT_INH_SET;
 512         p->cap_permitted = CAP_FULL_SET;
 513         p->keep_capabilities = 0;
 514         return;
 515 }
 516
 517 int cap_syslog (int type)
 518 {
 519         if ((type != 3 && type != 10) && !capable(CAP_SYS_ADMIN))
 520                 return -EPERM;
 521         return 0;
 522 }
 523
 524 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
 525 {
 526         int cap_sys_admin = 0;
 527
 528         if (cap_capable(current, CAP_SYS_ADMIN) == 0)
 529                 cap_sys_admin = 1;
 530         return __vm_enough_memory(mm, pages, cap_sys_admin);
 531 }
 532