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 #include <linux/sched.h>
  27
  28 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
  29 /*
  30  * Because of the reduced scope of CAP_SETPCAP when filesystem
  31  * capabilities are in effect, it is safe to allow this capability to
  32  * be available in the default configuration.
  33  */
  34 # define CAP_INIT_BSET  CAP_FULL_SET
  35 #else /* ie. ndef CONFIG_SECURITY_FILE_CAPABILITIES */
  36 # define CAP_INIT_BSET  CAP_INIT_EFF_SET
  37 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
  38
  39 kernel_cap_t cap_bset = CAP_INIT_BSET;    /* systemwide capability bound */
  40 EXPORT_SYMBOL(cap_bset);
  41
  42 /* Global security state */
  43
  44 unsigned securebits = SECUREBITS_DEFAULT; /* systemwide security settings */
  45 EXPORT_SYMBOL(securebits);
  46
  47 int cap_netlink_send(struct sock *sk, struct sk_buff *skb)
  48 {
  49         NETLINK_CB(skb).eff_cap = current->cap_effective;
  50         return 0;
  51 }
  52
  53 int cap_netlink_recv(struct sk_buff *skb, int cap)
  54 {
  55         if (!cap_raised(NETLINK_CB(skb).eff_cap, cap))
  56                 return -EPERM;
  57         return 0;
  58 }
  59
  60 EXPORT_SYMBOL(cap_netlink_recv);
  61
  62 int cap_capable (struct task_struct *tsk, int cap)
  63 {
  64         /* Derived from include/linux/sched.h:capable. */
  65         if (cap_raised(tsk->cap_effective, cap))
  66                 return 0;
  67         return -EPERM;
  68 }
  69
  70 int cap_settime(struct timespec *ts, struct timezone *tz)
  71 {
  72         if (!capable(CAP_SYS_TIME))
  73                 return -EPERM;
  74         return 0;
  75 }
  76
  77 int cap_ptrace (struct task_struct *parent, struct task_struct *child)
  78 {
  79         /* Derived from arch/i386/kernel/ptrace.c:sys_ptrace. */
  80         if (!cap_issubset(child->cap_permitted, parent->cap_permitted) &&
  81             !__capable(parent, CAP_SYS_PTRACE))
  82                 return -EPERM;
  83         return 0;
  84 }
  85
  86 int cap_capget (struct task_struct *target, kernel_cap_t *effective,
  87                 kernel_cap_t *inheritable, kernel_cap_t *permitted)
  88 {
  89         /* Derived from kernel/capability.c:sys_capget. */
  90         *effective = cap_t (target->cap_effective);
  91         *inheritable = cap_t (target->cap_inheritable);
  92         *permitted = cap_t (target->cap_permitted);
  93         return 0;
  94 }
  95
  96 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
  97
  98 static inline int cap_block_setpcap(struct task_struct *target)
  99 {
 100         /*
 101          * No support for remote process capability manipulation with
 102          * filesystem capability support.
 103          */
 104         return (target != current);
 105 }
 106
 107 static inline int cap_inh_is_capped(void)
 108 {
 109         /*
 110          * return 1 if changes to the inheritable set are limited
 111          * to the old permitted set.
 112          */
 113         return !cap_capable(current, CAP_SETPCAP);
 114 }
 115
 116 #else /* ie., ndef CONFIG_SECURITY_FILE_CAPABILITIES */
 117
 118 static inline int cap_block_setpcap(struct task_struct *t) { return 0; }
 119 static inline int cap_inh_is_capped(void) { return 1; }
 120
 121 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
 122
 123 int cap_capset_check (struct task_struct *target, kernel_cap_t *effective,
 124                       kernel_cap_t *inheritable, kernel_cap_t *permitted)
 125 {
 126         if (cap_block_setpcap(target)) {
 127                 return -EPERM;
 128         }
 129         if (cap_inh_is_capped()
 130             && !cap_issubset(*inheritable,
 131                              cap_combine(target->cap_inheritable,
 132                                          current->cap_permitted))) {
 133                 /* incapable of using this inheritable set */
 134                 return -EPERM;
 135         }
 136
 137         /* verify restrictions on target's new Permitted set */
 138         if (!cap_issubset (*permitted,
 139                            cap_combine (target->cap_permitted,
 140                                         current->cap_permitted))) {
 141                 return -EPERM;
 142         }
 143
 144         /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
 145         if (!cap_issubset (*effective, *permitted)) {
 146                 return -EPERM;
 147         }
 148
 149         return 0;
 150 }
 151
 152 void cap_capset_set (struct task_struct *target, kernel_cap_t *effective,
 153                      kernel_cap_t *inheritable, kernel_cap_t *permitted)
 154 {
 155         target->cap_effective = *effective;
 156         target->cap_inheritable = *inheritable;
 157         target->cap_permitted = *permitted;
 158 }
 159
 160 static inline void bprm_clear_caps(struct linux_binprm *bprm)
 161 {
 162         cap_clear(bprm->cap_inheritable);
 163         cap_clear(bprm->cap_permitted);
 164         bprm->cap_effective = false;
 165 }
 166
 167 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
 168
 169 int cap_inode_need_killpriv(struct dentry *dentry)
 170 {
 171         struct inode *inode = dentry->d_inode;
 172         int error;
 173
 174         if (!inode->i_op || !inode->i_op->getxattr)
 175                return 0;
 176
 177         error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
 178         if (error <= 0)
 179                 return 0;
 180         return 1;
 181 }
 182
 183 int cap_inode_killpriv(struct dentry *dentry)
 184 {
 185         struct inode *inode = dentry->d_inode;
 186
 187         if (!inode->i_op || !inode->i_op->removexattr)
 188                return 0;
 189
 190         return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
 191 }
 192
 193 static inline int cap_from_disk(struct vfs_cap_data *caps,
 194                                 struct linux_binprm *bprm,
 195                                 int size)
 196 {
 197         __u32 magic_etc;
 198
 199         if (size != XATTR_CAPS_SZ)
 200                 return -EINVAL;
 201
 202         magic_etc = le32_to_cpu(caps->magic_etc);
 203
 204         switch ((magic_etc & VFS_CAP_REVISION_MASK)) {
 205         case VFS_CAP_REVISION:
 206                 if (magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
 207                         bprm->cap_effective = true;
 208                 else
 209                         bprm->cap_effective = false;
 210                 bprm->cap_permitted = to_cap_t(le32_to_cpu(caps->permitted));
 211                 bprm->cap_inheritable = to_cap_t(le32_to_cpu(caps->inheritable));
 212                 return 0;
 213         default:
 214                 return -EINVAL;
 215         }
 216 }
 217
 218 /* Locate any VFS capabilities: */
 219 static int get_file_caps(struct linux_binprm *bprm)
 220 {
 221         struct dentry *dentry;
 222         int rc = 0;
 223         struct vfs_cap_data incaps;
 224         struct inode *inode;
 225
 226         if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID) {
 227                 bprm_clear_caps(bprm);
 228                 return 0;
 229         }
 230
 231         dentry = dget(bprm->file->f_dentry);
 232         inode = dentry->d_inode;
 233         if (!inode->i_op || !inode->i_op->getxattr)
 234                 goto out;
 235
 236         rc = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
 237         if (rc > 0) {
 238                 if (rc == XATTR_CAPS_SZ)
 239                         rc = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS,
 240                                                 &incaps, XATTR_CAPS_SZ);
 241                 else
 242                         rc = -EINVAL;
 243         }
 244         if (rc == -ENODATA || rc == -EOPNOTSUPP) {
 245                 /* no data, that's ok */
 246                 rc = 0;
 247                 goto out;
 248         }
 249         if (rc < 0)
 250                 goto out;
 251
 252         rc = cap_from_disk(&incaps, bprm, rc);
 253         if (rc)
 254                 printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
 255                         __FUNCTION__, rc, bprm->filename);
 256
 257 out:
 258         dput(dentry);
 259         if (rc)
 260                 bprm_clear_caps(bprm);
 261
 262         return rc;
 263 }
 264
 265 #else
 266 int cap_inode_need_killpriv(struct dentry *dentry)
 267 {
 268         return 0;
 269 }
 270
 271 int cap_inode_killpriv(struct dentry *dentry)
 272 {
 273         return 0;
 274 }
 275
 276 static inline int get_file_caps(struct linux_binprm *bprm)
 277 {
 278         bprm_clear_caps(bprm);
 279         return 0;
 280 }
 281 #endif
 282
 283 int cap_bprm_set_security (struct linux_binprm *bprm)
 284 {
 285         int ret;
 286
 287         ret = get_file_caps(bprm);
 288         if (ret)
 289                 printk(KERN_NOTICE "%s: get_file_caps returned %d for %s\n",
 290                         __FUNCTION__, ret, bprm->filename);
 291
 292         /*  To support inheritance of root-permissions and suid-root
 293          *  executables under compatibility mode, we raise all three
 294          *  capability sets for the file.
 295          *
 296          *  If only the real uid is 0, we only raise the inheritable
 297          *  and permitted sets of the executable file.
 298          */
 299
 300         if (!issecure (SECURE_NOROOT)) {
 301                 if (bprm->e_uid == 0 || current->uid == 0) {
 302                         cap_set_full (bprm->cap_inheritable);
 303                         cap_set_full (bprm->cap_permitted);
 304                 }
 305                 if (bprm->e_uid == 0)
 306                         bprm->cap_effective = true;
 307         }
 308
 309         return ret;
 310 }
 311
 312 void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
 313 {
 314         /* Derived from fs/exec.c:compute_creds. */
 315         kernel_cap_t new_permitted, working;
 316
 317         new_permitted = cap_intersect (bprm->cap_permitted, cap_bset);
 318         working = cap_intersect (bprm->cap_inheritable,
 319                                  current->cap_inheritable);
 320         new_permitted = cap_combine (new_permitted, working);
 321
 322         if (bprm->e_uid != current->uid || bprm->e_gid != current->gid ||
 323             !cap_issubset (new_permitted, current->cap_permitted)) {
 324                 set_dumpable(current->mm, suid_dumpable);
 325                 current->pdeath_signal = 0;
 326
 327                 if (unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
 328                         if (!capable(CAP_SETUID)) {
 329                                 bprm->e_uid = current->uid;
 330                                 bprm->e_gid = current->gid;
 331                         }
 332                         if (!capable (CAP_SETPCAP)) {
 333                                 new_permitted = cap_intersect (new_permitted,
 334                                                         current->cap_permitted);
 335                         }
 336                 }
 337         }
 338
 339         current->suid = current->euid = current->fsuid = bprm->e_uid;
 340         current->sgid = current->egid = current->fsgid = bprm->e_gid;
 341
 342         /* For init, we want to retain the capabilities set
 343          * in the init_task struct. Thus we skip the usual
 344          * capability rules */
 345         if (!is_global_init(current)) {
 346                 current->cap_permitted = new_permitted;
 347                 current->cap_effective = bprm->cap_effective ?
 348                                 new_permitted : 0;
 349         }
 350
 351         /* AUD: Audit candidate if current->cap_effective is set */
 352
 353         current->keep_capabilities = 0;
 354 }
 355
 356 int cap_bprm_secureexec (struct linux_binprm *bprm)
 357 {
 358         if (current->uid != 0) {
 359                 if (bprm->cap_effective)
 360                         return 1;
 361                 if (!cap_isclear(bprm->cap_permitted))
 362                         return 1;
 363                 if (!cap_isclear(bprm->cap_inheritable))
 364                         return 1;
 365         }
 366
 367         return (current->euid != current->uid ||
 368                 current->egid != current->gid);
 369 }
 370
 371 int cap_inode_setxattr(struct dentry *dentry, char *name, void *value,
 372                        size_t size, int flags)
 373 {
 374         if (!strcmp(name, XATTR_NAME_CAPS)) {
 375                 if (!capable(CAP_SETFCAP))
 376                         return -EPERM;
 377                 return 0;
 378         } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
 379                      sizeof(XATTR_SECURITY_PREFIX) - 1)  &&
 380             !capable(CAP_SYS_ADMIN))
 381                 return -EPERM;
 382         return 0;
 383 }
 384
 385 int cap_inode_removexattr(struct dentry *dentry, char *name)
 386 {
 387         if (!strcmp(name, XATTR_NAME_CAPS)) {
 388                 if (!capable(CAP_SETFCAP))
 389                         return -EPERM;
 390                 return 0;
 391         } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
 392                      sizeof(XATTR_SECURITY_PREFIX) - 1)  &&
 393             !capable(CAP_SYS_ADMIN))
 394                 return -EPERM;
 395         return 0;
 396 }
 397
 398 /* moved from kernel/sys.c. */
 399 /*
 400  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
 401  * a process after a call to setuid, setreuid, or setresuid.
 402  *
 403  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
 404  *  {r,e,s}uid != 0, the permitted and effective capabilities are
 405  *  cleared.
 406  *
 407  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
 408  *  capabilities of the process are cleared.
 409  *
 410  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
 411  *  capabilities are set to the permitted capabilities.
 412  *
 413  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
 414  *  never happen.
 415  *
 416  *  -astor
 417  *
 418  * cevans - New behaviour, Oct '99
 419  * A process may, via prctl(), elect to keep its capabilities when it
 420  * calls setuid() and switches away from uid==0. Both permitted and
 421  * effective sets will be retained.
 422  * Without this change, it was impossible for a daemon to drop only some
 423  * of its privilege. The call to setuid(!=0) would drop all privileges!
 424  * Keeping uid 0 is not an option because uid 0 owns too many vital
 425  * files..
 426  * Thanks to Olaf Kirch and Peter Benie for spotting this.
 427  */
 428 static inline void cap_emulate_setxuid (int old_ruid, int old_euid,
 429                                         int old_suid)
 430 {
 431         if ((old_ruid == 0 || old_euid == 0 || old_suid == 0) &&
 432             (current->uid != 0 && current->euid != 0 && current->suid != 0) &&
 433             !current->keep_capabilities) {
 434                 cap_clear (current->cap_permitted);
 435                 cap_clear (current->cap_effective);
 436         }
 437         if (old_euid == 0 && current->euid != 0) {
 438                 cap_clear (current->cap_effective);
 439         }
 440         if (old_euid != 0 && current->euid == 0) {
 441                 current->cap_effective = current->cap_permitted;
 442         }
 443 }
 444
 445 int cap_task_post_setuid (uid_t old_ruid, uid_t old_euid, uid_t old_suid,
 446                           int flags)
 447 {
 448         switch (flags) {
 449         case LSM_SETID_RE:
 450         case LSM_SETID_ID:
 451         case LSM_SETID_RES:
 452                 /* Copied from kernel/sys.c:setreuid/setuid/setresuid. */
 453                 if (!issecure (SECURE_NO_SETUID_FIXUP)) {
 454                         cap_emulate_setxuid (old_ruid, old_euid, old_suid);
 455                 }
 456                 break;
 457         case LSM_SETID_FS:
 458                 {
 459                         uid_t old_fsuid = old_ruid;
 460
 461                         /* Copied from kernel/sys.c:setfsuid. */
 462
 463                         /*
 464                          * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
 465                          *          if not, we might be a bit too harsh here.
 466                          */
 467
 468                         if (!issecure (SECURE_NO_SETUID_FIXUP)) {
 469                                 if (old_fsuid == 0 && current->fsuid != 0) {
 470                                         cap_t (current->cap_effective) &=
 471                                             ~CAP_FS_MASK;
 472                                 }
 473                                 if (old_fsuid != 0 && current->fsuid == 0) {
 474                                         cap_t (current->cap_effective) |=
 475                                             (cap_t (current->cap_permitted) &
 476                                              CAP_FS_MASK);
 477                                 }
 478                         }
 479                         break;
 480                 }
 481         default:
 482                 return -EINVAL;
 483         }
 484
 485         return 0;
 486 }
 487
 488 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
 489 /*
 490  * Rationale: code calling task_setscheduler, task_setioprio, and
 491  * task_setnice, assumes that
 492  *   . if capable(cap_sys_nice), then those actions should be allowed
 493  *   . if not capable(cap_sys_nice), but acting on your own processes,
 494  *      then those actions should be allowed
 495  * This is insufficient now since you can call code without suid, but
 496  * yet with increased caps.
 497  * So we check for increased caps on the target process.
 498  */
 499 static inline int cap_safe_nice(struct task_struct *p)
 500 {
 501         if (!cap_issubset(p->cap_permitted, current->cap_permitted) &&
 502             !__capable(current, CAP_SYS_NICE))
 503                 return -EPERM;
 504         return 0;
 505 }
 506
 507 int cap_task_setscheduler (struct task_struct *p, int policy,
 508                            struct sched_param *lp)
 509 {
 510         return cap_safe_nice(p);
 511 }
 512
 513 int cap_task_setioprio (struct task_struct *p, int ioprio)
 514 {
 515         return cap_safe_nice(p);
 516 }
 517
 518 int cap_task_setnice (struct task_struct *p, int nice)
 519 {
 520         return cap_safe_nice(p);
 521 }
 522
 523 int cap_task_kill(struct task_struct *p, struct siginfo *info,
 524                                 int sig, u32 secid)
 525 {
 526         if (info != SEND_SIG_NOINFO && (is_si_special(info) || SI_FROMKERNEL(info)))
 527                 return 0;
 528
 529         /*
 530          * Running a setuid root program raises your capabilities.
 531          * Killing your own setuid root processes was previously
 532          * allowed.
 533          * We must preserve legacy signal behavior in this case.
 534          */
 535         if (p->euid == 0 && p->uid == current->uid)
 536                 return 0;
 537
 538         /* sigcont is permitted within same session */
 539         if (sig == SIGCONT && (task_session_nr(current) == task_session_nr(p)))
 540                 return 0;
 541
 542         if (secid)
 543                 /*
 544                  * Signal sent as a particular user.
 545                  * Capabilities are ignored.  May be wrong, but it's the
 546                  * only thing we can do at the moment.
 547                  * Used only by usb drivers?
 548                  */
 549                 return 0;
 550         if (cap_issubset(p->cap_permitted, current->cap_permitted))
 551                 return 0;
 552         if (capable(CAP_KILL))
 553                 return 0;
 554
 555         return -EPERM;
 556 }
 557 #else
 558 int cap_task_setscheduler (struct task_struct *p, int policy,
 559                            struct sched_param *lp)
 560 {
 561         return 0;
 562 }
 563 int cap_task_setioprio (struct task_struct *p, int ioprio)
 564 {
 565         return 0;
 566 }
 567 int cap_task_setnice (struct task_struct *p, int nice)
 568 {
 569         return 0;
 570 }
 571 int cap_task_kill(struct task_struct *p, struct siginfo *info,
 572                                 int sig, u32 secid)
 573 {
 574         return 0;
 575 }
 576 #endif
 577
 578 void cap_task_reparent_to_init (struct task_struct *p)
 579 {
 580         p->cap_effective = CAP_INIT_EFF_SET;
 581         p->cap_inheritable = CAP_INIT_INH_SET;
 582         p->cap_permitted = CAP_FULL_SET;
 583         p->keep_capabilities = 0;
 584         return;
 585 }
 586
 587 int cap_syslog (int type)
 588 {
 589         if ((type != 3 && type != 10) && !capable(CAP_SYS_ADMIN))
 590                 return -EPERM;
 591         return 0;
 592 }
 593
 594 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
 595 {
 596         int cap_sys_admin = 0;
 597
 598         if (cap_capable(current, CAP_SYS_ADMIN) == 0)
 599                 cap_sys_admin = 1;
 600         return __vm_enough_memory(mm, pages, cap_sys_admin);
 601 }
 602