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 /*
  63  * NOTE WELL: cap_capable() cannot be used like the kernel's capable()
  64  * function.  That is, it has the reverse semantics: cap_capable()
  65  * returns 0 when a task has a capability, but the kernel's capable()
  66  * returns 1 for this case.
  67  */
  68 int cap_capable (struct task_struct *tsk, int cap)
  69 {
  70         /* Derived from include/linux/sched.h:capable. */
  71         if (cap_raised(tsk->cap_effective, cap))
  72                 return 0;
  73         return -EPERM;
  74 }
  75
  76 int cap_settime(struct timespec *ts, struct timezone *tz)
  77 {
  78         if (!capable(CAP_SYS_TIME))
  79                 return -EPERM;
  80         return 0;
  81 }
  82
  83 int cap_ptrace (struct task_struct *parent, struct task_struct *child)
  84 {
  85         /* Derived from arch/i386/kernel/ptrace.c:sys_ptrace. */
  86         if (!cap_issubset(child->cap_permitted, parent->cap_permitted) &&
  87             !__capable(parent, CAP_SYS_PTRACE))
  88                 return -EPERM;
  89         return 0;
  90 }
  91
  92 int cap_capget (struct task_struct *target, kernel_cap_t *effective,
  93                 kernel_cap_t *inheritable, kernel_cap_t *permitted)
  94 {
  95         /* Derived from kernel/capability.c:sys_capget. */
  96         *effective = cap_t (target->cap_effective);
  97         *inheritable = cap_t (target->cap_inheritable);
  98         *permitted = cap_t (target->cap_permitted);
  99         return 0;
 100 }
 101
 102 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
 103
 104 static inline int cap_block_setpcap(struct task_struct *target)
 105 {
 106         /*
 107          * No support for remote process capability manipulation with
 108          * filesystem capability support.
 109          */
 110         return (target != current);
 111 }
 112
 113 static inline int cap_inh_is_capped(void)
 114 {
 115         /*
 116          * Return 1 if changes to the inheritable set are limited
 117          * to the old permitted set. That is, if the current task
 118          * does *not* possess the CAP_SETPCAP capability.
 119          */
 120         return (cap_capable(current, CAP_SETPCAP) != 0);
 121 }
 122
 123 #else /* ie., ndef CONFIG_SECURITY_FILE_CAPABILITIES */
 124
 125 static inline int cap_block_setpcap(struct task_struct *t) { return 0; }
 126 static inline int cap_inh_is_capped(void) { return 1; }
 127
 128 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
 129
 130 int cap_capset_check (struct task_struct *target, kernel_cap_t *effective,
 131                       kernel_cap_t *inheritable, kernel_cap_t *permitted)
 132 {
 133         if (cap_block_setpcap(target)) {
 134                 return -EPERM;
 135         }
 136         if (cap_inh_is_capped()
 137             && !cap_issubset(*inheritable,
 138                              cap_combine(target->cap_inheritable,
 139                                          current->cap_permitted))) {
 140                 /* incapable of using this inheritable set */
 141                 return -EPERM;
 142         }
 143
 144         /* verify restrictions on target's new Permitted set */
 145         if (!cap_issubset (*permitted,
 146                            cap_combine (target->cap_permitted,
 147                                         current->cap_permitted))) {
 148                 return -EPERM;
 149         }
 150
 151         /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
 152         if (!cap_issubset (*effective, *permitted)) {
 153                 return -EPERM;
 154         }
 155
 156         return 0;
 157 }
 158
 159 void cap_capset_set (struct task_struct *target, kernel_cap_t *effective,
 160                      kernel_cap_t *inheritable, kernel_cap_t *permitted)
 161 {
 162         target->cap_effective = *effective;
 163         target->cap_inheritable = *inheritable;
 164         target->cap_permitted = *permitted;
 165 }
 166
 167 static inline void bprm_clear_caps(struct linux_binprm *bprm)
 168 {
 169         cap_clear(bprm->cap_inheritable);
 170         cap_clear(bprm->cap_permitted);
 171         bprm->cap_effective = false;
 172 }
 173
 174 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
 175
 176 int cap_inode_need_killpriv(struct dentry *dentry)
 177 {
 178         struct inode *inode = dentry->d_inode;
 179         int error;
 180
 181         if (!inode->i_op || !inode->i_op->getxattr)
 182                return 0;
 183
 184         error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
 185         if (error <= 0)
 186                 return 0;
 187         return 1;
 188 }
 189
 190 int cap_inode_killpriv(struct dentry *dentry)
 191 {
 192         struct inode *inode = dentry->d_inode;
 193
 194         if (!inode->i_op || !inode->i_op->removexattr)
 195                return 0;
 196
 197         return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
 198 }
 199
 200 static inline int cap_from_disk(struct vfs_cap_data *caps,
 201                                 struct linux_binprm *bprm,
 202                                 int size)
 203 {
 204         __u32 magic_etc;
 205
 206         if (size != XATTR_CAPS_SZ)
 207                 return -EINVAL;
 208
 209         magic_etc = le32_to_cpu(caps->magic_etc);
 210
 211         switch ((magic_etc & VFS_CAP_REVISION_MASK)) {
 212         case VFS_CAP_REVISION:
 213                 if (magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
 214                         bprm->cap_effective = true;
 215                 else
 216                         bprm->cap_effective = false;
 217                 bprm->cap_permitted = to_cap_t(le32_to_cpu(caps->permitted));
 218                 bprm->cap_inheritable = to_cap_t(le32_to_cpu(caps->inheritable));
 219                 return 0;
 220         default:
 221                 return -EINVAL;
 222         }
 223 }
 224
 225 /* Locate any VFS capabilities: */
 226 static int get_file_caps(struct linux_binprm *bprm)
 227 {
 228         struct dentry *dentry;
 229         int rc = 0;
 230         struct vfs_cap_data incaps;
 231         struct inode *inode;
 232
 233         if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID) {
 234                 bprm_clear_caps(bprm);
 235                 return 0;
 236         }
 237
 238         dentry = dget(bprm->file->f_dentry);
 239         inode = dentry->d_inode;
 240         if (!inode->i_op || !inode->i_op->getxattr)
 241                 goto out;
 242
 243         rc = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
 244         if (rc > 0) {
 245                 if (rc == XATTR_CAPS_SZ)
 246                         rc = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS,
 247                                                 &incaps, XATTR_CAPS_SZ);
 248                 else
 249                         rc = -EINVAL;
 250         }
 251         if (rc == -ENODATA || rc == -EOPNOTSUPP) {
 252                 /* no data, that's ok */
 253                 rc = 0;
 254                 goto out;
 255         }
 256         if (rc < 0)
 257                 goto out;
 258
 259         rc = cap_from_disk(&incaps, bprm, rc);
 260         if (rc)
 261                 printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
 262                         __FUNCTION__, rc, bprm->filename);
 263
 264 out:
 265         dput(dentry);
 266         if (rc)
 267                 bprm_clear_caps(bprm);
 268
 269         return rc;
 270 }
 271
 272 #else
 273 int cap_inode_need_killpriv(struct dentry *dentry)
 274 {
 275         return 0;
 276 }
 277
 278 int cap_inode_killpriv(struct dentry *dentry)
 279 {
 280         return 0;
 281 }
 282
 283 static inline int get_file_caps(struct linux_binprm *bprm)
 284 {
 285         bprm_clear_caps(bprm);
 286         return 0;
 287 }
 288 #endif
 289
 290 int cap_bprm_set_security (struct linux_binprm *bprm)
 291 {
 292         int ret;
 293
 294         ret = get_file_caps(bprm);
 295         if (ret)
 296                 printk(KERN_NOTICE "%s: get_file_caps returned %d for %s\n",
 297                         __FUNCTION__, ret, bprm->filename);
 298
 299         /*  To support inheritance of root-permissions and suid-root
 300          *  executables under compatibility mode, we raise all three
 301          *  capability sets for the file.
 302          *
 303          *  If only the real uid is 0, we only raise the inheritable
 304          *  and permitted sets of the executable file.
 305          */
 306
 307         if (!issecure (SECURE_NOROOT)) {
 308                 if (bprm->e_uid == 0 || current->uid == 0) {
 309                         cap_set_full (bprm->cap_inheritable);
 310                         cap_set_full (bprm->cap_permitted);
 311                 }
 312                 if (bprm->e_uid == 0)
 313                         bprm->cap_effective = true;
 314         }
 315
 316         return ret;
 317 }
 318
 319 void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
 320 {
 321         /* Derived from fs/exec.c:compute_creds. */
 322         kernel_cap_t new_permitted, working;
 323
 324         new_permitted = cap_intersect (bprm->cap_permitted, cap_bset);
 325         working = cap_intersect (bprm->cap_inheritable,
 326                                  current->cap_inheritable);
 327         new_permitted = cap_combine (new_permitted, working);
 328
 329         if (bprm->e_uid != current->uid || bprm->e_gid != current->gid ||
 330             !cap_issubset (new_permitted, current->cap_permitted)) {
 331                 set_dumpable(current->mm, suid_dumpable);
 332                 current->pdeath_signal = 0;
 333
 334                 if (unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
 335                         if (!capable(CAP_SETUID)) {
 336                                 bprm->e_uid = current->uid;
 337                                 bprm->e_gid = current->gid;
 338                         }
 339                         if (!capable (CAP_SETPCAP)) {
 340                                 new_permitted = cap_intersect (new_permitted,
 341                                                         current->cap_permitted);
 342                         }
 343                 }
 344         }
 345
 346         current->suid = current->euid = current->fsuid = bprm->e_uid;
 347         current->sgid = current->egid = current->fsgid = bprm->e_gid;
 348
 349         /* For init, we want to retain the capabilities set
 350          * in the init_task struct. Thus we skip the usual
 351          * capability rules */
 352         if (!is_global_init(current)) {
 353                 current->cap_permitted = new_permitted;
 354                 current->cap_effective = bprm->cap_effective ?
 355                                 new_permitted : 0;
 356         }
 357
 358         /* AUD: Audit candidate if current->cap_effective is set */
 359
 360         current->keep_capabilities = 0;
 361 }
 362
 363 int cap_bprm_secureexec (struct linux_binprm *bprm)
 364 {
 365         if (current->uid != 0) {
 366                 if (bprm->cap_effective)
 367                         return 1;
 368                 if (!cap_isclear(bprm->cap_permitted))
 369                         return 1;
 370                 if (!cap_isclear(bprm->cap_inheritable))
 371                         return 1;
 372         }
 373
 374         return (current->euid != current->uid ||
 375                 current->egid != current->gid);
 376 }
 377
 378 int cap_inode_setxattr(struct dentry *dentry, char *name, void *value,
 379                        size_t size, int flags)
 380 {
 381         if (!strcmp(name, XATTR_NAME_CAPS)) {
 382                 if (!capable(CAP_SETFCAP))
 383                         return -EPERM;
 384                 return 0;
 385         } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
 386                      sizeof(XATTR_SECURITY_PREFIX) - 1)  &&
 387             !capable(CAP_SYS_ADMIN))
 388                 return -EPERM;
 389         return 0;
 390 }
 391
 392 int cap_inode_removexattr(struct dentry *dentry, char *name)
 393 {
 394         if (!strcmp(name, XATTR_NAME_CAPS)) {
 395                 if (!capable(CAP_SETFCAP))
 396                         return -EPERM;
 397                 return 0;
 398         } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
 399                      sizeof(XATTR_SECURITY_PREFIX) - 1)  &&
 400             !capable(CAP_SYS_ADMIN))
 401                 return -EPERM;
 402         return 0;
 403 }
 404
 405 /* moved from kernel/sys.c. */
 406 /*
 407  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
 408  * a process after a call to setuid, setreuid, or setresuid.
 409  *
 410  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
 411  *  {r,e,s}uid != 0, the permitted and effective capabilities are
 412  *  cleared.
 413  *
 414  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
 415  *  capabilities of the process are cleared.
 416  *
 417  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
 418  *  capabilities are set to the permitted capabilities.
 419  *
 420  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
 421  *  never happen.
 422  *
 423  *  -astor
 424  *
 425  * cevans - New behaviour, Oct '99
 426  * A process may, via prctl(), elect to keep its capabilities when it
 427  * calls setuid() and switches away from uid==0. Both permitted and
 428  * effective sets will be retained.
 429  * Without this change, it was impossible for a daemon to drop only some
 430  * of its privilege. The call to setuid(!=0) would drop all privileges!
 431  * Keeping uid 0 is not an option because uid 0 owns too many vital
 432  * files..
 433  * Thanks to Olaf Kirch and Peter Benie for spotting this.
 434  */
 435 static inline void cap_emulate_setxuid (int old_ruid, int old_euid,
 436                                         int old_suid)
 437 {
 438         if ((old_ruid == 0 || old_euid == 0 || old_suid == 0) &&
 439             (current->uid != 0 && current->euid != 0 && current->suid != 0) &&
 440             !current->keep_capabilities) {
 441                 cap_clear (current->cap_permitted);
 442                 cap_clear (current->cap_effective);
 443         }
 444         if (old_euid == 0 && current->euid != 0) {
 445                 cap_clear (current->cap_effective);
 446         }
 447         if (old_euid != 0 && current->euid == 0) {
 448                 current->cap_effective = current->cap_permitted;
 449         }
 450 }
 451
 452 int cap_task_post_setuid (uid_t old_ruid, uid_t old_euid, uid_t old_suid,
 453                           int flags)
 454 {
 455         switch (flags) {
 456         case LSM_SETID_RE:
 457         case LSM_SETID_ID:
 458         case LSM_SETID_RES:
 459                 /* Copied from kernel/sys.c:setreuid/setuid/setresuid. */
 460                 if (!issecure (SECURE_NO_SETUID_FIXUP)) {
 461                         cap_emulate_setxuid (old_ruid, old_euid, old_suid);
 462                 }
 463                 break;
 464         case LSM_SETID_FS:
 465                 {
 466                         uid_t old_fsuid = old_ruid;
 467
 468                         /* Copied from kernel/sys.c:setfsuid. */
 469
 470                         /*
 471                          * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
 472                          *          if not, we might be a bit too harsh here.
 473                          */
 474
 475                         if (!issecure (SECURE_NO_SETUID_FIXUP)) {
 476                                 if (old_fsuid == 0 && current->fsuid != 0) {
 477                                         cap_t (current->cap_effective) &=
 478                                             ~CAP_FS_MASK;
 479                                 }
 480                                 if (old_fsuid != 0 && current->fsuid == 0) {
 481                                         cap_t (current->cap_effective) |=
 482                                             (cap_t (current->cap_permitted) &
 483                                              CAP_FS_MASK);
 484                                 }
 485                         }
 486                         break;
 487                 }
 488         default:
 489                 return -EINVAL;
 490         }
 491
 492         return 0;
 493 }
 494
 495 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
 496 /*
 497  * Rationale: code calling task_setscheduler, task_setioprio, and
 498  * task_setnice, assumes that
 499  *   . if capable(cap_sys_nice), then those actions should be allowed
 500  *   . if not capable(cap_sys_nice), but acting on your own processes,
 501  *      then those actions should be allowed
 502  * This is insufficient now since you can call code without suid, but
 503  * yet with increased caps.
 504  * So we check for increased caps on the target process.
 505  */
 506 static inline int cap_safe_nice(struct task_struct *p)
 507 {
 508         if (!cap_issubset(p->cap_permitted, current->cap_permitted) &&
 509             !__capable(current, CAP_SYS_NICE))
 510                 return -EPERM;
 511         return 0;
 512 }
 513
 514 int cap_task_setscheduler (struct task_struct *p, int policy,
 515                            struct sched_param *lp)
 516 {
 517         return cap_safe_nice(p);
 518 }
 519
 520 int cap_task_setioprio (struct task_struct *p, int ioprio)
 521 {
 522         return cap_safe_nice(p);
 523 }
 524
 525 int cap_task_setnice (struct task_struct *p, int nice)
 526 {
 527         return cap_safe_nice(p);
 528 }
 529
 530 int cap_task_kill(struct task_struct *p, struct siginfo *info,
 531                                 int sig, u32 secid)
 532 {
 533         if (info != SEND_SIG_NOINFO && (is_si_special(info) || SI_FROMKERNEL(info)))
 534                 return 0;
 535
 536         /*
 537          * Running a setuid root program raises your capabilities.
 538          * Killing your own setuid root processes was previously
 539          * allowed.
 540          * We must preserve legacy signal behavior in this case.
 541          */
 542         if (p->euid == 0 && p->uid == current->uid)
 543                 return 0;
 544
 545         /* sigcont is permitted within same session */
 546         if (sig == SIGCONT && (task_session_nr(current) == task_session_nr(p)))
 547                 return 0;
 548
 549         if (secid)
 550                 /*
 551                  * Signal sent as a particular user.
 552                  * Capabilities are ignored.  May be wrong, but it's the
 553                  * only thing we can do at the moment.
 554                  * Used only by usb drivers?
 555                  */
 556                 return 0;
 557         if (cap_issubset(p->cap_permitted, current->cap_permitted))
 558                 return 0;
 559         if (capable(CAP_KILL))
 560                 return 0;
 561
 562         return -EPERM;
 563 }
 564 #else
 565 int cap_task_setscheduler (struct task_struct *p, int policy,
 566                            struct sched_param *lp)
 567 {
 568         return 0;
 569 }
 570 int cap_task_setioprio (struct task_struct *p, int ioprio)
 571 {
 572         return 0;
 573 }
 574 int cap_task_setnice (struct task_struct *p, int nice)
 575 {
 576         return 0;
 577 }
 578 int cap_task_kill(struct task_struct *p, struct siginfo *info,
 579                                 int sig, u32 secid)
 580 {
 581         return 0;
 582 }
 583 #endif
 584
 585 void cap_task_reparent_to_init (struct task_struct *p)
 586 {
 587         p->cap_effective = CAP_INIT_EFF_SET;
 588         p->cap_inheritable = CAP_INIT_INH_SET;
 589         p->cap_permitted = CAP_FULL_SET;
 590         p->keep_capabilities = 0;
 591         return;
 592 }
 593
 594 int cap_syslog (int type)
 595 {
 596         if ((type != 3 && type != 10) && !capable(CAP_SYS_ADMIN))
 597                 return -EPERM;
 598         return 0;
 599 }
 600
 601 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
 602 {
 603         int cap_sys_admin = 0;
 604
 605         if (cap_capable(current, CAP_SYS_ADMIN) == 0)
 606                 cap_sys_admin = 1;
 607         return __vm_enough_memory(mm, pages, cap_sys_admin);
 608 }
 609