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 EXPORT_SYMBOL(cap_netlink_send);
  34
  35 int cap_netlink_recv(struct sk_buff *skb, int cap)
  36 {
  37         if (!cap_raised(NETLINK_CB(skb).eff_cap, cap))
  38                 return -EPERM;
  39         return 0;
  40 }
  41
  42 EXPORT_SYMBOL(cap_netlink_recv);
  43
  44 int cap_capable (struct task_struct *tsk, int cap)
  45 {
  46         /* Derived from include/linux/sched.h:capable. */
  47         if (cap_raised(tsk->cap_effective, cap))
  48                 return 0;
  49         return -EPERM;
  50 }
  51
  52 int cap_settime(struct timespec *ts, struct timezone *tz)
  53 {
  54         if (!capable(CAP_SYS_TIME))
  55                 return -EPERM;
  56         return 0;
  57 }
  58
  59 int cap_ptrace (struct task_struct *parent, struct task_struct *child)
  60 {
  61         /* Derived from arch/i386/kernel/ptrace.c:sys_ptrace. */
  62         if (!cap_issubset(child->cap_permitted, parent->cap_permitted) &&
  63             !__capable(parent, CAP_SYS_PTRACE))
  64                 return -EPERM;
  65         return 0;
  66 }
  67
  68 int cap_capget (struct task_struct *target, kernel_cap_t *effective,
  69                 kernel_cap_t *inheritable, kernel_cap_t *permitted)
  70 {
  71         /* Derived from kernel/capability.c:sys_capget. */
  72         *effective = cap_t (target->cap_effective);
  73         *inheritable = cap_t (target->cap_inheritable);
  74         *permitted = cap_t (target->cap_permitted);
  75         return 0;
  76 }
  77
  78 int cap_capset_check (struct task_struct *target, kernel_cap_t *effective,
  79                       kernel_cap_t *inheritable, kernel_cap_t *permitted)
  80 {
  81         /* Derived from kernel/capability.c:sys_capset. */
  82         /* verify restrictions on target's new Inheritable set */
  83         if (!cap_issubset (*inheritable,
  84                            cap_combine (target->cap_inheritable,
  85                                         current->cap_permitted))) {
  86                 return -EPERM;
  87         }
  88
  89         /* verify restrictions on target's new Permitted set */
  90         if (!cap_issubset (*permitted,
  91                            cap_combine (target->cap_permitted,
  92                                         current->cap_permitted))) {
  93                 return -EPERM;
  94         }
  95
  96         /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
  97         if (!cap_issubset (*effective, *permitted)) {
  98                 return -EPERM;
  99         }
 100
 101         return 0;
 102 }
 103
 104 void cap_capset_set (struct task_struct *target, kernel_cap_t *effective,
 105                      kernel_cap_t *inheritable, kernel_cap_t *permitted)
 106 {
 107         target->cap_effective = *effective;
 108         target->cap_inheritable = *inheritable;
 109         target->cap_permitted = *permitted;
 110 }
 111
 112 static inline void bprm_clear_caps(struct linux_binprm *bprm)
 113 {
 114         cap_clear(bprm->cap_inheritable);
 115         cap_clear(bprm->cap_permitted);
 116         bprm->cap_effective = false;
 117 }
 118
 119 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
 120
 121 int cap_inode_need_killpriv(struct dentry *dentry)
 122 {
 123         struct inode *inode = dentry->d_inode;
 124         int error;
 125
 126         if (!inode->i_op || !inode->i_op->getxattr)
 127                return 0;
 128
 129         error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
 130         if (error <= 0)
 131                 return 0;
 132         return 1;
 133 }
 134
 135 int cap_inode_killpriv(struct dentry *dentry)
 136 {
 137         struct inode *inode = dentry->d_inode;
 138
 139         if (!inode->i_op || !inode->i_op->removexattr)
 140                return 0;
 141
 142         return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
 143 }
 144
 145 static inline int cap_from_disk(__le32 *caps, struct linux_binprm *bprm,
 146                                 int size)
 147 {
 148         __u32 magic_etc;
 149
 150         if (size != XATTR_CAPS_SZ)
 151                 return -EINVAL;
 152
 153         magic_etc = le32_to_cpu(caps[0]);
 154
 155         switch ((magic_etc & VFS_CAP_REVISION_MASK)) {
 156         case VFS_CAP_REVISION:
 157                 if (magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
 158                         bprm->cap_effective = true;
 159                 else
 160                         bprm->cap_effective = false;
 161                 bprm->cap_permitted = to_cap_t( le32_to_cpu(caps[1]) );
 162                 bprm->cap_inheritable = to_cap_t( le32_to_cpu(caps[2]) );
 163                 return 0;
 164         default:
 165                 return -EINVAL;
 166         }
 167 }
 168
 169 /* Locate any VFS capabilities: */
 170 static int get_file_caps(struct linux_binprm *bprm)
 171 {
 172         struct dentry *dentry;
 173         int rc = 0;
 174         __le32 v1caps[XATTR_CAPS_SZ];
 175         struct inode *inode;
 176
 177         if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID) {
 178                 bprm_clear_caps(bprm);
 179                 return 0;
 180         }
 181
 182         dentry = dget(bprm->file->f_dentry);
 183         inode = dentry->d_inode;
 184         if (!inode->i_op || !inode->i_op->getxattr)
 185                 goto out;
 186
 187         rc = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, &v1caps,
 188                                                         XATTR_CAPS_SZ);
 189         if (rc == -ENODATA || rc == -EOPNOTSUPP) {
 190                 /* no data, that's ok */
 191                 rc = 0;
 192                 goto out;
 193         }
 194         if (rc < 0)
 195                 goto out;
 196
 197         rc = cap_from_disk(v1caps, bprm, rc);
 198         if (rc)
 199                 printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
 200                         __FUNCTION__, rc, bprm->filename);
 201
 202 out:
 203         dput(dentry);
 204         if (rc)
 205                 bprm_clear_caps(bprm);
 206
 207         return rc;
 208 }
 209
 210 #else
 211 int cap_inode_need_killpriv(struct dentry *dentry)
 212 {
 213         return 0;
 214 }
 215
 216 int cap_inode_killpriv(struct dentry *dentry)
 217 {
 218         return 0;
 219 }
 220
 221 static inline int get_file_caps(struct linux_binprm *bprm)
 222 {
 223         bprm_clear_caps(bprm);
 224         return 0;
 225 }
 226 #endif
 227
 228 int cap_bprm_set_security (struct linux_binprm *bprm)
 229 {
 230         int ret;
 231
 232         ret = get_file_caps(bprm);
 233         if (ret)
 234                 printk(KERN_NOTICE "%s: get_file_caps returned %d for %s\n",
 235                         __FUNCTION__, ret, bprm->filename);
 236
 237         /*  To support inheritance of root-permissions and suid-root
 238          *  executables under compatibility mode, we raise all three
 239          *  capability sets for the file.
 240          *
 241          *  If only the real uid is 0, we only raise the inheritable
 242          *  and permitted sets of the executable file.
 243          */
 244
 245         if (!issecure (SECURE_NOROOT)) {
 246                 if (bprm->e_uid == 0 || current->uid == 0) {
 247                         cap_set_full (bprm->cap_inheritable);
 248                         cap_set_full (bprm->cap_permitted);
 249                 }
 250                 if (bprm->e_uid == 0)
 251                         bprm->cap_effective = true;
 252         }
 253
 254         return ret;
 255 }
 256
 257 void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
 258 {
 259         /* Derived from fs/exec.c:compute_creds. */
 260         kernel_cap_t new_permitted, working;
 261
 262         new_permitted = cap_intersect (bprm->cap_permitted, cap_bset);
 263         working = cap_intersect (bprm->cap_inheritable,
 264                                  current->cap_inheritable);
 265         new_permitted = cap_combine (new_permitted, working);
 266
 267         if (bprm->e_uid != current->uid || bprm->e_gid != current->gid ||
 268             !cap_issubset (new_permitted, current->cap_permitted)) {
 269                 set_dumpable(current->mm, suid_dumpable);
 270                 current->pdeath_signal = 0;
 271
 272                 if (unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
 273                         if (!capable(CAP_SETUID)) {
 274                                 bprm->e_uid = current->uid;
 275                                 bprm->e_gid = current->gid;
 276                         }
 277                         if (!capable (CAP_SETPCAP)) {
 278                                 new_permitted = cap_intersect (new_permitted,
 279                                                         current->cap_permitted);
 280                         }
 281                 }
 282         }
 283
 284         current->suid = current->euid = current->fsuid = bprm->e_uid;
 285         current->sgid = current->egid = current->fsgid = bprm->e_gid;
 286
 287         /* For init, we want to retain the capabilities set
 288          * in the init_task struct. Thus we skip the usual
 289          * capability rules */
 290         if (!is_init(current)) {
 291                 current->cap_permitted = new_permitted;
 292                 current->cap_effective = bprm->cap_effective ?
 293                                 new_permitted : 0;
 294         }
 295
 296         /* AUD: Audit candidate if current->cap_effective is set */
 297
 298         current->keep_capabilities = 0;
 299 }
 300
 301 int cap_bprm_secureexec (struct linux_binprm *bprm)
 302 {
 303         if (current->uid != 0) {
 304                 if (bprm->cap_effective)
 305                         return 1;
 306                 if (!cap_isclear(bprm->cap_permitted))
 307                         return 1;
 308                 if (!cap_isclear(bprm->cap_inheritable))
 309                         return 1;
 310         }
 311
 312         return (current->euid != current->uid ||
 313                 current->egid != current->gid);
 314 }
 315
 316 int cap_inode_setxattr(struct dentry *dentry, char *name, void *value,
 317                        size_t size, int flags)
 318 {
 319         if (!strcmp(name, XATTR_NAME_CAPS)) {
 320                 if (!capable(CAP_SETFCAP))
 321                         return -EPERM;
 322                 return 0;
 323         } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
 324                      sizeof(XATTR_SECURITY_PREFIX) - 1)  &&
 325             !capable(CAP_SYS_ADMIN))
 326                 return -EPERM;
 327         return 0;
 328 }
 329
 330 int cap_inode_removexattr(struct dentry *dentry, char *name)
 331 {
 332         if (!strcmp(name, XATTR_NAME_CAPS)) {
 333                 if (!capable(CAP_SETFCAP))
 334                         return -EPERM;
 335                 return 0;
 336         } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
 337                      sizeof(XATTR_SECURITY_PREFIX) - 1)  &&
 338             !capable(CAP_SYS_ADMIN))
 339                 return -EPERM;
 340         return 0;
 341 }
 342
 343 /* moved from kernel/sys.c. */
 344 /*
 345  * cap_emulate_setxuid() fixes the effective / permitted capabilities of
 346  * a process after a call to setuid, setreuid, or setresuid.
 347  *
 348  *  1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
 349  *  {r,e,s}uid != 0, the permitted and effective capabilities are
 350  *  cleared.
 351  *
 352  *  2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
 353  *  capabilities of the process are cleared.
 354  *
 355  *  3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
 356  *  capabilities are set to the permitted capabilities.
 357  *
 358  *  fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
 359  *  never happen.
 360  *
 361  *  -astor
 362  *
 363  * cevans - New behaviour, Oct '99
 364  * A process may, via prctl(), elect to keep its capabilities when it
 365  * calls setuid() and switches away from uid==0. Both permitted and
 366  * effective sets will be retained.
 367  * Without this change, it was impossible for a daemon to drop only some
 368  * of its privilege. The call to setuid(!=0) would drop all privileges!
 369  * Keeping uid 0 is not an option because uid 0 owns too many vital
 370  * files..
 371  * Thanks to Olaf Kirch and Peter Benie for spotting this.
 372  */
 373 static inline void cap_emulate_setxuid (int old_ruid, int old_euid,
 374                                         int old_suid)
 375 {
 376         if ((old_ruid == 0 || old_euid == 0 || old_suid == 0) &&
 377             (current->uid != 0 && current->euid != 0 && current->suid != 0) &&
 378             !current->keep_capabilities) {
 379                 cap_clear (current->cap_permitted);
 380                 cap_clear (current->cap_effective);
 381         }
 382         if (old_euid == 0 && current->euid != 0) {
 383                 cap_clear (current->cap_effective);
 384         }
 385         if (old_euid != 0 && current->euid == 0) {
 386                 current->cap_effective = current->cap_permitted;
 387         }
 388 }
 389
 390 int cap_task_post_setuid (uid_t old_ruid, uid_t old_euid, uid_t old_suid,
 391                           int flags)
 392 {
 393         switch (flags) {
 394         case LSM_SETID_RE:
 395         case LSM_SETID_ID:
 396         case LSM_SETID_RES:
 397                 /* Copied from kernel/sys.c:setreuid/setuid/setresuid. */
 398                 if (!issecure (SECURE_NO_SETUID_FIXUP)) {
 399                         cap_emulate_setxuid (old_ruid, old_euid, old_suid);
 400                 }
 401                 break;
 402         case LSM_SETID_FS:
 403                 {
 404                         uid_t old_fsuid = old_ruid;
 405
 406                         /* Copied from kernel/sys.c:setfsuid. */
 407
 408                         /*
 409                          * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
 410                          *          if not, we might be a bit too harsh here.
 411                          */
 412
 413                         if (!issecure (SECURE_NO_SETUID_FIXUP)) {
 414                                 if (old_fsuid == 0 && current->fsuid != 0) {
 415                                         cap_t (current->cap_effective) &=
 416                                             ~CAP_FS_MASK;
 417                                 }
 418                                 if (old_fsuid != 0 && current->fsuid == 0) {
 419                                         cap_t (current->cap_effective) |=
 420                                             (cap_t (current->cap_permitted) &
 421                                              CAP_FS_MASK);
 422                                 }
 423                         }
 424                         break;
 425                 }
 426         default:
 427                 return -EINVAL;
 428         }
 429
 430         return 0;
 431 }
 432
 433 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
 434 /*
 435  * Rationale: code calling task_setscheduler, task_setioprio, and
 436  * task_setnice, assumes that
 437  *   . if capable(cap_sys_nice), then those actions should be allowed
 438  *   . if not capable(cap_sys_nice), but acting on your own processes,
 439  *      then those actions should be allowed
 440  * This is insufficient now since you can call code without suid, but
 441  * yet with increased caps.
 442  * So we check for increased caps on the target process.
 443  */
 444 static inline int cap_safe_nice(struct task_struct *p)
 445 {
 446         if (!cap_issubset(p->cap_permitted, current->cap_permitted) &&
 447             !__capable(current, CAP_SYS_NICE))
 448                 return -EPERM;
 449         return 0;
 450 }
 451
 452 int cap_task_setscheduler (struct task_struct *p, int policy,
 453                            struct sched_param *lp)
 454 {
 455         return cap_safe_nice(p);
 456 }
 457
 458 int cap_task_setioprio (struct task_struct *p, int ioprio)
 459 {
 460         return cap_safe_nice(p);
 461 }
 462
 463 int cap_task_setnice (struct task_struct *p, int nice)
 464 {
 465         return cap_safe_nice(p);
 466 }
 467
 468 int cap_task_kill(struct task_struct *p, struct siginfo *info,
 469                                 int sig, u32 secid)
 470 {
 471         if (info != SEND_SIG_NOINFO && (is_si_special(info) || SI_FROMKERNEL(info)))
 472                 return 0;
 473
 474         if (secid)
 475                 /*
 476                  * Signal sent as a particular user.
 477                  * Capabilities are ignored.  May be wrong, but it's the
 478                  * only thing we can do at the moment.
 479                  * Used only by usb drivers?
 480                  */
 481                 return 0;
 482         if (cap_issubset(p->cap_permitted, current->cap_permitted))
 483                 return 0;
 484         if (capable(CAP_KILL))
 485                 return 0;
 486
 487         return -EPERM;
 488 }
 489 #else
 490 int cap_task_setscheduler (struct task_struct *p, int policy,
 491                            struct sched_param *lp)
 492 {
 493         return 0;
 494 }
 495 int cap_task_setioprio (struct task_struct *p, int ioprio)
 496 {
 497         return 0;
 498 }
 499 int cap_task_setnice (struct task_struct *p, int nice)
 500 {
 501         return 0;
 502 }
 503 int cap_task_kill(struct task_struct *p, struct siginfo *info,
 504                                 int sig, u32 secid)
 505 {
 506         return 0;
 507 }
 508 #endif
 509
 510 void cap_task_reparent_to_init (struct task_struct *p)
 511 {
 512         p->cap_effective = CAP_INIT_EFF_SET;
 513         p->cap_inheritable = CAP_INIT_INH_SET;
 514         p->cap_permitted = CAP_FULL_SET;
 515         p->keep_capabilities = 0;
 516         return;
 517 }
 518
 519 int cap_syslog (int type)
 520 {
 521         if ((type != 3 && type != 10) && !capable(CAP_SYS_ADMIN))
 522                 return -EPERM;
 523         return 0;
 524 }
 525
 526 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
 527 {
 528         int cap_sys_admin = 0;
 529
 530         if (cap_capable(current, CAP_SYS_ADMIN) == 0)
 531                 cap_sys_admin = 1;
 532         return __vm_enough_memory(mm, pages, cap_sys_admin);
 533 }
 534
 535 EXPORT_SYMBOL(cap_capable);
 536 EXPORT_SYMBOL(cap_settime);
 537 EXPORT_SYMBOL(cap_ptrace);
 538 EXPORT_SYMBOL(cap_capget);
 539 EXPORT_SYMBOL(cap_capset_check);
 540 EXPORT_SYMBOL(cap_capset_set);
 541 EXPORT_SYMBOL(cap_bprm_set_security);
 542 EXPORT_SYMBOL(cap_bprm_apply_creds);
 543 EXPORT_SYMBOL(cap_bprm_secureexec);
 544 EXPORT_SYMBOL(cap_inode_setxattr);
 545 EXPORT_SYMBOL(cap_inode_removexattr);
 546 EXPORT_SYMBOL(cap_task_post_setuid);
 547 EXPORT_SYMBOL(cap_task_kill);
 548 EXPORT_SYMBOL(cap_task_setscheduler);
 549 EXPORT_SYMBOL(cap_task_setioprio);
 550 EXPORT_SYMBOL(cap_task_setnice);
 551 EXPORT_SYMBOL(cap_task_reparent_to_init);
 552 EXPORT_SYMBOL(cap_syslog);
 553 EXPORT_SYMBOL(cap_vm_enough_memory);