1 /* Common capabilities, needed by capability.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/audit.h> 12 #include <linux/module.h> 13 #include <linux/init.h> 14 #include <linux/kernel.h> 15 #include <linux/lsm_hooks.h> 16 #include <linux/file.h> 17 #include <linux/mm.h> 18 #include <linux/mman.h> 19 #include <linux/pagemap.h> 20 #include <linux/swap.h> 21 #include <linux/skbuff.h> 22 #include <linux/netlink.h> 23 #include <linux/ptrace.h> 24 #include <linux/xattr.h> 25 #include <linux/hugetlb.h> 26 #include <linux/mount.h> 27 #include <linux/sched.h> 28 #include <linux/prctl.h> 29 #include <linux/securebits.h> 30 #include <linux/user_namespace.h> 31 #include <linux/binfmts.h> 32 #include <linux/personality.h> 33 34 /* 35 * If a non-root user executes a setuid-root binary in 36 * !secure(SECURE_NOROOT) mode, then we raise capabilities. 37 * However if fE is also set, then the intent is for only 38 * the file capabilities to be applied, and the setuid-root 39 * bit is left on either to change the uid (plausible) or 40 * to get full privilege on a kernel without file capabilities 41 * support. So in that case we do not raise capabilities. 42 * 43 * Warn if that happens, once per boot. 44 */ 45 static void warn_setuid_and_fcaps_mixed(const char *fname) 46 { 47 static int warned; 48 if (!warned) { 49 printk(KERN_INFO "warning: `%s' has both setuid-root and" 50 " effective capabilities. Therefore not raising all" 51 " capabilities.\n", fname); 52 warned = 1; 53 } 54 } 55 56 /** 57 * cap_capable - Determine whether a task has a particular effective capability 58 * @cred: The credentials to use 59 * @ns: The user namespace in which we need the capability 60 * @cap: The capability to check for 61 * @audit: Whether to write an audit message or not 62 * 63 * Determine whether the nominated task has the specified capability amongst 64 * its effective set, returning 0 if it does, -ve if it does not. 65 * 66 * NOTE WELL: cap_has_capability() cannot be used like the kernel's capable() 67 * and has_capability() functions. That is, it has the reverse semantics: 68 * cap_has_capability() returns 0 when a task has a capability, but the 69 * kernel's capable() and has_capability() returns 1 for this case. 70 */ 71 int cap_capable(const struct cred *cred, struct user_namespace *targ_ns, 72 int cap, int audit) 73 { 74 struct user_namespace *ns = targ_ns; 75 76 /* See if cred has the capability in the target user namespace 77 * by examining the target user namespace and all of the target 78 * user namespace's parents. 79 */ 80 for (;;) { 81 /* Do we have the necessary capabilities? */ 82 if (ns == cred->user_ns) 83 return cap_raised(cred->cap_effective, cap) ? 0 : -EPERM; 84 85 /* Have we tried all of the parent namespaces? */ 86 if (ns == &init_user_ns) 87 return -EPERM; 88 89 /* 90 * The owner of the user namespace in the parent of the 91 * user namespace has all caps. 92 */ 93 if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid)) 94 return 0; 95 96 /* 97 * If you have a capability in a parent user ns, then you have 98 * it over all children user namespaces as well. 99 */ 100 ns = ns->parent; 101 } 102 103 /* We never get here */ 104 } 105 106 /** 107 * cap_settime - Determine whether the current process may set the system clock 108 * @ts: The time to set 109 * @tz: The timezone to set 110 * 111 * Determine whether the current process may set the system clock and timezone 112 * information, returning 0 if permission granted, -ve if denied. 113 */ 114 int cap_settime(const struct timespec64 *ts, const struct timezone *tz) 115 { 116 if (!capable(CAP_SYS_TIME)) 117 return -EPERM; 118 return 0; 119 } 120 121 /** 122 * cap_ptrace_access_check - Determine whether the current process may access 123 * another 124 * @child: The process to be accessed 125 * @mode: The mode of attachment. 126 * 127 * If we are in the same or an ancestor user_ns and have all the target 128 * task's capabilities, then ptrace access is allowed. 129 * If we have the ptrace capability to the target user_ns, then ptrace 130 * access is allowed. 131 * Else denied. 132 * 133 * Determine whether a process may access another, returning 0 if permission 134 * granted, -ve if denied. 135 */ 136 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode) 137 { 138 int ret = 0; 139 const struct cred *cred, *child_cred; 140 const kernel_cap_t *caller_caps; 141 142 rcu_read_lock(); 143 cred = current_cred(); 144 child_cred = __task_cred(child); 145 if (mode & PTRACE_MODE_FSCREDS) 146 caller_caps = &cred->cap_effective; 147 else 148 caller_caps = &cred->cap_permitted; 149 if (cred->user_ns == child_cred->user_ns && 150 cap_issubset(child_cred->cap_permitted, *caller_caps)) 151 goto out; 152 if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE)) 153 goto out; 154 ret = -EPERM; 155 out: 156 rcu_read_unlock(); 157 return ret; 158 } 159 160 /** 161 * cap_ptrace_traceme - Determine whether another process may trace the current 162 * @parent: The task proposed to be the tracer 163 * 164 * If parent is in the same or an ancestor user_ns and has all current's 165 * capabilities, then ptrace access is allowed. 166 * If parent has the ptrace capability to current's user_ns, then ptrace 167 * access is allowed. 168 * Else denied. 169 * 170 * Determine whether the nominated task is permitted to trace the current 171 * process, returning 0 if permission is granted, -ve if denied. 172 */ 173 int cap_ptrace_traceme(struct task_struct *parent) 174 { 175 int ret = 0; 176 const struct cred *cred, *child_cred; 177 178 rcu_read_lock(); 179 cred = __task_cred(parent); 180 child_cred = current_cred(); 181 if (cred->user_ns == child_cred->user_ns && 182 cap_issubset(child_cred->cap_permitted, cred->cap_permitted)) 183 goto out; 184 if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE)) 185 goto out; 186 ret = -EPERM; 187 out: 188 rcu_read_unlock(); 189 return ret; 190 } 191 192 /** 193 * cap_capget - Retrieve a task's capability sets 194 * @target: The task from which to retrieve the capability sets 195 * @effective: The place to record the effective set 196 * @inheritable: The place to record the inheritable set 197 * @permitted: The place to record the permitted set 198 * 199 * This function retrieves the capabilities of the nominated task and returns 200 * them to the caller. 201 */ 202 int cap_capget(struct task_struct *target, kernel_cap_t *effective, 203 kernel_cap_t *inheritable, kernel_cap_t *permitted) 204 { 205 const struct cred *cred; 206 207 /* Derived from kernel/capability.c:sys_capget. */ 208 rcu_read_lock(); 209 cred = __task_cred(target); 210 *effective = cred->cap_effective; 211 *inheritable = cred->cap_inheritable; 212 *permitted = cred->cap_permitted; 213 rcu_read_unlock(); 214 return 0; 215 } 216 217 /* 218 * Determine whether the inheritable capabilities are limited to the old 219 * permitted set. Returns 1 if they are limited, 0 if they are not. 220 */ 221 static inline int cap_inh_is_capped(void) 222 { 223 224 /* they are so limited unless the current task has the CAP_SETPCAP 225 * capability 226 */ 227 if (cap_capable(current_cred(), current_cred()->user_ns, 228 CAP_SETPCAP, SECURITY_CAP_AUDIT) == 0) 229 return 0; 230 return 1; 231 } 232 233 /** 234 * cap_capset - Validate and apply proposed changes to current's capabilities 235 * @new: The proposed new credentials; alterations should be made here 236 * @old: The current task's current credentials 237 * @effective: A pointer to the proposed new effective capabilities set 238 * @inheritable: A pointer to the proposed new inheritable capabilities set 239 * @permitted: A pointer to the proposed new permitted capabilities set 240 * 241 * This function validates and applies a proposed mass change to the current 242 * process's capability sets. The changes are made to the proposed new 243 * credentials, and assuming no error, will be committed by the caller of LSM. 244 */ 245 int cap_capset(struct cred *new, 246 const struct cred *old, 247 const kernel_cap_t *effective, 248 const kernel_cap_t *inheritable, 249 const kernel_cap_t *permitted) 250 { 251 if (cap_inh_is_capped() && 252 !cap_issubset(*inheritable, 253 cap_combine(old->cap_inheritable, 254 old->cap_permitted))) 255 /* incapable of using this inheritable set */ 256 return -EPERM; 257 258 if (!cap_issubset(*inheritable, 259 cap_combine(old->cap_inheritable, 260 old->cap_bset))) 261 /* no new pI capabilities outside bounding set */ 262 return -EPERM; 263 264 /* verify restrictions on target's new Permitted set */ 265 if (!cap_issubset(*permitted, old->cap_permitted)) 266 return -EPERM; 267 268 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */ 269 if (!cap_issubset(*effective, *permitted)) 270 return -EPERM; 271 272 new->cap_effective = *effective; 273 new->cap_inheritable = *inheritable; 274 new->cap_permitted = *permitted; 275 276 /* 277 * Mask off ambient bits that are no longer both permitted and 278 * inheritable. 279 */ 280 new->cap_ambient = cap_intersect(new->cap_ambient, 281 cap_intersect(*permitted, 282 *inheritable)); 283 if (WARN_ON(!cap_ambient_invariant_ok(new))) 284 return -EINVAL; 285 return 0; 286 } 287 288 /* 289 * Clear proposed capability sets for execve(). 290 */ 291 static inline void bprm_clear_caps(struct linux_binprm *bprm) 292 { 293 cap_clear(bprm->cred->cap_permitted); 294 bprm->cap_effective = false; 295 } 296 297 /** 298 * cap_inode_need_killpriv - Determine if inode change affects privileges 299 * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV 300 * 301 * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV 302 * affects the security markings on that inode, and if it is, should 303 * inode_killpriv() be invoked or the change rejected? 304 * 305 * Returns 0 if granted; +ve if granted, but inode_killpriv() is required; and 306 * -ve to deny the change. 307 */ 308 int cap_inode_need_killpriv(struct dentry *dentry) 309 { 310 struct inode *inode = d_backing_inode(dentry); 311 int error; 312 313 if (!inode->i_op->getxattr) 314 return 0; 315 316 error = inode->i_op->getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0); 317 if (error <= 0) 318 return 0; 319 return 1; 320 } 321 322 /** 323 * cap_inode_killpriv - Erase the security markings on an inode 324 * @dentry: The inode/dentry to alter 325 * 326 * Erase the privilege-enhancing security markings on an inode. 327 * 328 * Returns 0 if successful, -ve on error. 329 */ 330 int cap_inode_killpriv(struct dentry *dentry) 331 { 332 struct inode *inode = d_backing_inode(dentry); 333 334 if (!inode->i_op->removexattr) 335 return 0; 336 337 return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS); 338 } 339 340 /* 341 * Calculate the new process capability sets from the capability sets attached 342 * to a file. 343 */ 344 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps, 345 struct linux_binprm *bprm, 346 bool *effective, 347 bool *has_cap) 348 { 349 struct cred *new = bprm->cred; 350 unsigned i; 351 int ret = 0; 352 353 if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE) 354 *effective = true; 355 356 if (caps->magic_etc & VFS_CAP_REVISION_MASK) 357 *has_cap = true; 358 359 CAP_FOR_EACH_U32(i) { 360 __u32 permitted = caps->permitted.cap[i]; 361 __u32 inheritable = caps->inheritable.cap[i]; 362 363 /* 364 * pP' = (X & fP) | (pI & fI) 365 * The addition of pA' is handled later. 366 */ 367 new->cap_permitted.cap[i] = 368 (new->cap_bset.cap[i] & permitted) | 369 (new->cap_inheritable.cap[i] & inheritable); 370 371 if (permitted & ~new->cap_permitted.cap[i]) 372 /* insufficient to execute correctly */ 373 ret = -EPERM; 374 } 375 376 /* 377 * For legacy apps, with no internal support for recognizing they 378 * do not have enough capabilities, we return an error if they are 379 * missing some "forced" (aka file-permitted) capabilities. 380 */ 381 return *effective ? ret : 0; 382 } 383 384 /* 385 * Extract the on-exec-apply capability sets for an executable file. 386 */ 387 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps) 388 { 389 struct inode *inode = d_backing_inode(dentry); 390 __u32 magic_etc; 391 unsigned tocopy, i; 392 int size; 393 struct vfs_cap_data caps; 394 395 memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data)); 396 397 if (!inode || !inode->i_op->getxattr) 398 return -ENODATA; 399 400 size = inode->i_op->getxattr((struct dentry *)dentry, inode, 401 XATTR_NAME_CAPS, &caps, XATTR_CAPS_SZ); 402 if (size == -ENODATA || size == -EOPNOTSUPP) 403 /* no data, that's ok */ 404 return -ENODATA; 405 if (size < 0) 406 return size; 407 408 if (size < sizeof(magic_etc)) 409 return -EINVAL; 410 411 cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps.magic_etc); 412 413 switch (magic_etc & VFS_CAP_REVISION_MASK) { 414 case VFS_CAP_REVISION_1: 415 if (size != XATTR_CAPS_SZ_1) 416 return -EINVAL; 417 tocopy = VFS_CAP_U32_1; 418 break; 419 case VFS_CAP_REVISION_2: 420 if (size != XATTR_CAPS_SZ_2) 421 return -EINVAL; 422 tocopy = VFS_CAP_U32_2; 423 break; 424 default: 425 return -EINVAL; 426 } 427 428 CAP_FOR_EACH_U32(i) { 429 if (i >= tocopy) 430 break; 431 cpu_caps->permitted.cap[i] = le32_to_cpu(caps.data[i].permitted); 432 cpu_caps->inheritable.cap[i] = le32_to_cpu(caps.data[i].inheritable); 433 } 434 435 cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK; 436 cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK; 437 438 return 0; 439 } 440 441 /* 442 * Attempt to get the on-exec apply capability sets for an executable file from 443 * its xattrs and, if present, apply them to the proposed credentials being 444 * constructed by execve(). 445 */ 446 static int get_file_caps(struct linux_binprm *bprm, bool *effective, bool *has_cap) 447 { 448 int rc = 0; 449 struct cpu_vfs_cap_data vcaps; 450 451 bprm_clear_caps(bprm); 452 453 if (!file_caps_enabled) 454 return 0; 455 456 if (bprm->file->f_path.mnt->mnt_flags & MNT_NOSUID) 457 return 0; 458 459 rc = get_vfs_caps_from_disk(bprm->file->f_path.dentry, &vcaps); 460 if (rc < 0) { 461 if (rc == -EINVAL) 462 printk(KERN_NOTICE "%s: get_vfs_caps_from_disk returned %d for %s\n", 463 __func__, rc, bprm->filename); 464 else if (rc == -ENODATA) 465 rc = 0; 466 goto out; 467 } 468 469 rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_cap); 470 if (rc == -EINVAL) 471 printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n", 472 __func__, rc, bprm->filename); 473 474 out: 475 if (rc) 476 bprm_clear_caps(bprm); 477 478 return rc; 479 } 480 481 /** 482 * cap_bprm_set_creds - Set up the proposed credentials for execve(). 483 * @bprm: The execution parameters, including the proposed creds 484 * 485 * Set up the proposed credentials for a new execution context being 486 * constructed by execve(). The proposed creds in @bprm->cred is altered, 487 * which won't take effect immediately. Returns 0 if successful, -ve on error. 488 */ 489 int cap_bprm_set_creds(struct linux_binprm *bprm) 490 { 491 const struct cred *old = current_cred(); 492 struct cred *new = bprm->cred; 493 bool effective, has_cap = false, is_setid; 494 int ret; 495 kuid_t root_uid; 496 497 if (WARN_ON(!cap_ambient_invariant_ok(old))) 498 return -EPERM; 499 500 effective = false; 501 ret = get_file_caps(bprm, &effective, &has_cap); 502 if (ret < 0) 503 return ret; 504 505 root_uid = make_kuid(new->user_ns, 0); 506 507 if (!issecure(SECURE_NOROOT)) { 508 /* 509 * If the legacy file capability is set, then don't set privs 510 * for a setuid root binary run by a non-root user. Do set it 511 * for a root user just to cause least surprise to an admin. 512 */ 513 if (has_cap && !uid_eq(new->uid, root_uid) && uid_eq(new->euid, root_uid)) { 514 warn_setuid_and_fcaps_mixed(bprm->filename); 515 goto skip; 516 } 517 /* 518 * To support inheritance of root-permissions and suid-root 519 * executables under compatibility mode, we override the 520 * capability sets for the file. 521 * 522 * If only the real uid is 0, we do not set the effective bit. 523 */ 524 if (uid_eq(new->euid, root_uid) || uid_eq(new->uid, root_uid)) { 525 /* pP' = (cap_bset & ~0) | (pI & ~0) */ 526 new->cap_permitted = cap_combine(old->cap_bset, 527 old->cap_inheritable); 528 } 529 if (uid_eq(new->euid, root_uid)) 530 effective = true; 531 } 532 skip: 533 534 /* if we have fs caps, clear dangerous personality flags */ 535 if (!cap_issubset(new->cap_permitted, old->cap_permitted)) 536 bprm->per_clear |= PER_CLEAR_ON_SETID; 537 538 539 /* Don't let someone trace a set[ug]id/setpcap binary with the revised 540 * credentials unless they have the appropriate permit. 541 * 542 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs. 543 */ 544 is_setid = !uid_eq(new->euid, old->uid) || !gid_eq(new->egid, old->gid); 545 546 if ((is_setid || 547 !cap_issubset(new->cap_permitted, old->cap_permitted)) && 548 bprm->unsafe & ~LSM_UNSAFE_PTRACE_CAP) { 549 /* downgrade; they get no more than they had, and maybe less */ 550 if (!capable(CAP_SETUID) || 551 (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) { 552 new->euid = new->uid; 553 new->egid = new->gid; 554 } 555 new->cap_permitted = cap_intersect(new->cap_permitted, 556 old->cap_permitted); 557 } 558 559 new->suid = new->fsuid = new->euid; 560 new->sgid = new->fsgid = new->egid; 561 562 /* File caps or setid cancels ambient. */ 563 if (has_cap || is_setid) 564 cap_clear(new->cap_ambient); 565 566 /* 567 * Now that we've computed pA', update pP' to give: 568 * pP' = (X & fP) | (pI & fI) | pA' 569 */ 570 new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient); 571 572 /* 573 * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set, 574 * this is the same as pE' = (fE ? pP' : 0) | pA'. 575 */ 576 if (effective) 577 new->cap_effective = new->cap_permitted; 578 else 579 new->cap_effective = new->cap_ambient; 580 581 if (WARN_ON(!cap_ambient_invariant_ok(new))) 582 return -EPERM; 583 584 bprm->cap_effective = effective; 585 586 /* 587 * Audit candidate if current->cap_effective is set 588 * 589 * We do not bother to audit if 3 things are true: 590 * 1) cap_effective has all caps 591 * 2) we are root 592 * 3) root is supposed to have all caps (SECURE_NOROOT) 593 * Since this is just a normal root execing a process. 594 * 595 * Number 1 above might fail if you don't have a full bset, but I think 596 * that is interesting information to audit. 597 */ 598 if (!cap_issubset(new->cap_effective, new->cap_ambient)) { 599 if (!cap_issubset(CAP_FULL_SET, new->cap_effective) || 600 !uid_eq(new->euid, root_uid) || !uid_eq(new->uid, root_uid) || 601 issecure(SECURE_NOROOT)) { 602 ret = audit_log_bprm_fcaps(bprm, new, old); 603 if (ret < 0) 604 return ret; 605 } 606 } 607 608 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS); 609 610 if (WARN_ON(!cap_ambient_invariant_ok(new))) 611 return -EPERM; 612 613 return 0; 614 } 615 616 /** 617 * cap_bprm_secureexec - Determine whether a secure execution is required 618 * @bprm: The execution parameters 619 * 620 * Determine whether a secure execution is required, return 1 if it is, and 0 621 * if it is not. 622 * 623 * The credentials have been committed by this point, and so are no longer 624 * available through @bprm->cred. 625 */ 626 int cap_bprm_secureexec(struct linux_binprm *bprm) 627 { 628 const struct cred *cred = current_cred(); 629 kuid_t root_uid = make_kuid(cred->user_ns, 0); 630 631 if (!uid_eq(cred->uid, root_uid)) { 632 if (bprm->cap_effective) 633 return 1; 634 if (!cap_issubset(cred->cap_permitted, cred->cap_ambient)) 635 return 1; 636 } 637 638 return (!uid_eq(cred->euid, cred->uid) || 639 !gid_eq(cred->egid, cred->gid)); 640 } 641 642 /** 643 * cap_inode_setxattr - Determine whether an xattr may be altered 644 * @dentry: The inode/dentry being altered 645 * @name: The name of the xattr to be changed 646 * @value: The value that the xattr will be changed to 647 * @size: The size of value 648 * @flags: The replacement flag 649 * 650 * Determine whether an xattr may be altered or set on an inode, returning 0 if 651 * permission is granted, -ve if denied. 652 * 653 * This is used to make sure security xattrs don't get updated or set by those 654 * who aren't privileged to do so. 655 */ 656 int cap_inode_setxattr(struct dentry *dentry, const char *name, 657 const void *value, size_t size, int flags) 658 { 659 if (!strcmp(name, XATTR_NAME_CAPS)) { 660 if (!capable(CAP_SETFCAP)) 661 return -EPERM; 662 return 0; 663 } 664 665 if (!strncmp(name, XATTR_SECURITY_PREFIX, 666 sizeof(XATTR_SECURITY_PREFIX) - 1) && 667 !capable(CAP_SYS_ADMIN)) 668 return -EPERM; 669 return 0; 670 } 671 672 /** 673 * cap_inode_removexattr - Determine whether an xattr may be removed 674 * @dentry: The inode/dentry being altered 675 * @name: The name of the xattr to be changed 676 * 677 * Determine whether an xattr may be removed from an inode, returning 0 if 678 * permission is granted, -ve if denied. 679 * 680 * This is used to make sure security xattrs don't get removed by those who 681 * aren't privileged to remove them. 682 */ 683 int cap_inode_removexattr(struct dentry *dentry, const char *name) 684 { 685 if (!strcmp(name, XATTR_NAME_CAPS)) { 686 if (!capable(CAP_SETFCAP)) 687 return -EPERM; 688 return 0; 689 } 690 691 if (!strncmp(name, XATTR_SECURITY_PREFIX, 692 sizeof(XATTR_SECURITY_PREFIX) - 1) && 693 !capable(CAP_SYS_ADMIN)) 694 return -EPERM; 695 return 0; 696 } 697 698 /* 699 * cap_emulate_setxuid() fixes the effective / permitted capabilities of 700 * a process after a call to setuid, setreuid, or setresuid. 701 * 702 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of 703 * {r,e,s}uid != 0, the permitted and effective capabilities are 704 * cleared. 705 * 706 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective 707 * capabilities of the process are cleared. 708 * 709 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective 710 * capabilities are set to the permitted capabilities. 711 * 712 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should 713 * never happen. 714 * 715 * -astor 716 * 717 * cevans - New behaviour, Oct '99 718 * A process may, via prctl(), elect to keep its capabilities when it 719 * calls setuid() and switches away from uid==0. Both permitted and 720 * effective sets will be retained. 721 * Without this change, it was impossible for a daemon to drop only some 722 * of its privilege. The call to setuid(!=0) would drop all privileges! 723 * Keeping uid 0 is not an option because uid 0 owns too many vital 724 * files.. 725 * Thanks to Olaf Kirch and Peter Benie for spotting this. 726 */ 727 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old) 728 { 729 kuid_t root_uid = make_kuid(old->user_ns, 0); 730 731 if ((uid_eq(old->uid, root_uid) || 732 uid_eq(old->euid, root_uid) || 733 uid_eq(old->suid, root_uid)) && 734 (!uid_eq(new->uid, root_uid) && 735 !uid_eq(new->euid, root_uid) && 736 !uid_eq(new->suid, root_uid))) { 737 if (!issecure(SECURE_KEEP_CAPS)) { 738 cap_clear(new->cap_permitted); 739 cap_clear(new->cap_effective); 740 } 741 742 /* 743 * Pre-ambient programs expect setresuid to nonroot followed 744 * by exec to drop capabilities. We should make sure that 745 * this remains the case. 746 */ 747 cap_clear(new->cap_ambient); 748 } 749 if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid)) 750 cap_clear(new->cap_effective); 751 if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid)) 752 new->cap_effective = new->cap_permitted; 753 } 754 755 /** 756 * cap_task_fix_setuid - Fix up the results of setuid() call 757 * @new: The proposed credentials 758 * @old: The current task's current credentials 759 * @flags: Indications of what has changed 760 * 761 * Fix up the results of setuid() call before the credential changes are 762 * actually applied, returning 0 to grant the changes, -ve to deny them. 763 */ 764 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags) 765 { 766 switch (flags) { 767 case LSM_SETID_RE: 768 case LSM_SETID_ID: 769 case LSM_SETID_RES: 770 /* juggle the capabilities to follow [RES]UID changes unless 771 * otherwise suppressed */ 772 if (!issecure(SECURE_NO_SETUID_FIXUP)) 773 cap_emulate_setxuid(new, old); 774 break; 775 776 case LSM_SETID_FS: 777 /* juggle the capabilties to follow FSUID changes, unless 778 * otherwise suppressed 779 * 780 * FIXME - is fsuser used for all CAP_FS_MASK capabilities? 781 * if not, we might be a bit too harsh here. 782 */ 783 if (!issecure(SECURE_NO_SETUID_FIXUP)) { 784 kuid_t root_uid = make_kuid(old->user_ns, 0); 785 if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid)) 786 new->cap_effective = 787 cap_drop_fs_set(new->cap_effective); 788 789 if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid)) 790 new->cap_effective = 791 cap_raise_fs_set(new->cap_effective, 792 new->cap_permitted); 793 } 794 break; 795 796 default: 797 return -EINVAL; 798 } 799 800 return 0; 801 } 802 803 /* 804 * Rationale: code calling task_setscheduler, task_setioprio, and 805 * task_setnice, assumes that 806 * . if capable(cap_sys_nice), then those actions should be allowed 807 * . if not capable(cap_sys_nice), but acting on your own processes, 808 * then those actions should be allowed 809 * This is insufficient now since you can call code without suid, but 810 * yet with increased caps. 811 * So we check for increased caps on the target process. 812 */ 813 static int cap_safe_nice(struct task_struct *p) 814 { 815 int is_subset, ret = 0; 816 817 rcu_read_lock(); 818 is_subset = cap_issubset(__task_cred(p)->cap_permitted, 819 current_cred()->cap_permitted); 820 if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) 821 ret = -EPERM; 822 rcu_read_unlock(); 823 824 return ret; 825 } 826 827 /** 828 * cap_task_setscheduler - Detemine if scheduler policy change is permitted 829 * @p: The task to affect 830 * 831 * Detemine if the requested scheduler policy change is permitted for the 832 * specified task, returning 0 if permission is granted, -ve if denied. 833 */ 834 int cap_task_setscheduler(struct task_struct *p) 835 { 836 return cap_safe_nice(p); 837 } 838 839 /** 840 * cap_task_ioprio - Detemine if I/O priority change is permitted 841 * @p: The task to affect 842 * @ioprio: The I/O priority to set 843 * 844 * Detemine if the requested I/O priority change is permitted for the specified 845 * task, returning 0 if permission is granted, -ve if denied. 846 */ 847 int cap_task_setioprio(struct task_struct *p, int ioprio) 848 { 849 return cap_safe_nice(p); 850 } 851 852 /** 853 * cap_task_ioprio - Detemine if task priority change is permitted 854 * @p: The task to affect 855 * @nice: The nice value to set 856 * 857 * Detemine if the requested task priority change is permitted for the 858 * specified task, returning 0 if permission is granted, -ve if denied. 859 */ 860 int cap_task_setnice(struct task_struct *p, int nice) 861 { 862 return cap_safe_nice(p); 863 } 864 865 /* 866 * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from 867 * the current task's bounding set. Returns 0 on success, -ve on error. 868 */ 869 static int cap_prctl_drop(unsigned long cap) 870 { 871 struct cred *new; 872 873 if (!ns_capable(current_user_ns(), CAP_SETPCAP)) 874 return -EPERM; 875 if (!cap_valid(cap)) 876 return -EINVAL; 877 878 new = prepare_creds(); 879 if (!new) 880 return -ENOMEM; 881 cap_lower(new->cap_bset, cap); 882 return commit_creds(new); 883 } 884 885 /** 886 * cap_task_prctl - Implement process control functions for this security module 887 * @option: The process control function requested 888 * @arg2, @arg3, @arg4, @arg5: The argument data for this function 889 * 890 * Allow process control functions (sys_prctl()) to alter capabilities; may 891 * also deny access to other functions not otherwise implemented here. 892 * 893 * Returns 0 or +ve on success, -ENOSYS if this function is not implemented 894 * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM 895 * modules will consider performing the function. 896 */ 897 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3, 898 unsigned long arg4, unsigned long arg5) 899 { 900 const struct cred *old = current_cred(); 901 struct cred *new; 902 903 switch (option) { 904 case PR_CAPBSET_READ: 905 if (!cap_valid(arg2)) 906 return -EINVAL; 907 return !!cap_raised(old->cap_bset, arg2); 908 909 case PR_CAPBSET_DROP: 910 return cap_prctl_drop(arg2); 911 912 /* 913 * The next four prctl's remain to assist with transitioning a 914 * system from legacy UID=0 based privilege (when filesystem 915 * capabilities are not in use) to a system using filesystem 916 * capabilities only - as the POSIX.1e draft intended. 917 * 918 * Note: 919 * 920 * PR_SET_SECUREBITS = 921 * issecure_mask(SECURE_KEEP_CAPS_LOCKED) 922 * | issecure_mask(SECURE_NOROOT) 923 * | issecure_mask(SECURE_NOROOT_LOCKED) 924 * | issecure_mask(SECURE_NO_SETUID_FIXUP) 925 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED) 926 * 927 * will ensure that the current process and all of its 928 * children will be locked into a pure 929 * capability-based-privilege environment. 930 */ 931 case PR_SET_SECUREBITS: 932 if ((((old->securebits & SECURE_ALL_LOCKS) >> 1) 933 & (old->securebits ^ arg2)) /*[1]*/ 934 || ((old->securebits & SECURE_ALL_LOCKS & ~arg2)) /*[2]*/ 935 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/ 936 || (cap_capable(current_cred(), 937 current_cred()->user_ns, CAP_SETPCAP, 938 SECURITY_CAP_AUDIT) != 0) /*[4]*/ 939 /* 940 * [1] no changing of bits that are locked 941 * [2] no unlocking of locks 942 * [3] no setting of unsupported bits 943 * [4] doing anything requires privilege (go read about 944 * the "sendmail capabilities bug") 945 */ 946 ) 947 /* cannot change a locked bit */ 948 return -EPERM; 949 950 new = prepare_creds(); 951 if (!new) 952 return -ENOMEM; 953 new->securebits = arg2; 954 return commit_creds(new); 955 956 case PR_GET_SECUREBITS: 957 return old->securebits; 958 959 case PR_GET_KEEPCAPS: 960 return !!issecure(SECURE_KEEP_CAPS); 961 962 case PR_SET_KEEPCAPS: 963 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */ 964 return -EINVAL; 965 if (issecure(SECURE_KEEP_CAPS_LOCKED)) 966 return -EPERM; 967 968 new = prepare_creds(); 969 if (!new) 970 return -ENOMEM; 971 if (arg2) 972 new->securebits |= issecure_mask(SECURE_KEEP_CAPS); 973 else 974 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS); 975 return commit_creds(new); 976 977 case PR_CAP_AMBIENT: 978 if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) { 979 if (arg3 | arg4 | arg5) 980 return -EINVAL; 981 982 new = prepare_creds(); 983 if (!new) 984 return -ENOMEM; 985 cap_clear(new->cap_ambient); 986 return commit_creds(new); 987 } 988 989 if (((!cap_valid(arg3)) | arg4 | arg5)) 990 return -EINVAL; 991 992 if (arg2 == PR_CAP_AMBIENT_IS_SET) { 993 return !!cap_raised(current_cred()->cap_ambient, arg3); 994 } else if (arg2 != PR_CAP_AMBIENT_RAISE && 995 arg2 != PR_CAP_AMBIENT_LOWER) { 996 return -EINVAL; 997 } else { 998 if (arg2 == PR_CAP_AMBIENT_RAISE && 999 (!cap_raised(current_cred()->cap_permitted, arg3) || 1000 !cap_raised(current_cred()->cap_inheritable, 1001 arg3) || 1002 issecure(SECURE_NO_CAP_AMBIENT_RAISE))) 1003 return -EPERM; 1004 1005 new = prepare_creds(); 1006 if (!new) 1007 return -ENOMEM; 1008 if (arg2 == PR_CAP_AMBIENT_RAISE) 1009 cap_raise(new->cap_ambient, arg3); 1010 else 1011 cap_lower(new->cap_ambient, arg3); 1012 return commit_creds(new); 1013 } 1014 1015 default: 1016 /* No functionality available - continue with default */ 1017 return -ENOSYS; 1018 } 1019 } 1020 1021 /** 1022 * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted 1023 * @mm: The VM space in which the new mapping is to be made 1024 * @pages: The size of the mapping 1025 * 1026 * Determine whether the allocation of a new virtual mapping by the current 1027 * task is permitted, returning 1 if permission is granted, 0 if not. 1028 */ 1029 int cap_vm_enough_memory(struct mm_struct *mm, long pages) 1030 { 1031 int cap_sys_admin = 0; 1032 1033 if (cap_capable(current_cred(), &init_user_ns, CAP_SYS_ADMIN, 1034 SECURITY_CAP_NOAUDIT) == 0) 1035 cap_sys_admin = 1; 1036 return cap_sys_admin; 1037 } 1038 1039 /* 1040 * cap_mmap_addr - check if able to map given addr 1041 * @addr: address attempting to be mapped 1042 * 1043 * If the process is attempting to map memory below dac_mmap_min_addr they need 1044 * CAP_SYS_RAWIO. The other parameters to this function are unused by the 1045 * capability security module. Returns 0 if this mapping should be allowed 1046 * -EPERM if not. 1047 */ 1048 int cap_mmap_addr(unsigned long addr) 1049 { 1050 int ret = 0; 1051 1052 if (addr < dac_mmap_min_addr) { 1053 ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO, 1054 SECURITY_CAP_AUDIT); 1055 /* set PF_SUPERPRIV if it turns out we allow the low mmap */ 1056 if (ret == 0) 1057 current->flags |= PF_SUPERPRIV; 1058 } 1059 return ret; 1060 } 1061 1062 int cap_mmap_file(struct file *file, unsigned long reqprot, 1063 unsigned long prot, unsigned long flags) 1064 { 1065 return 0; 1066 } 1067 1068 #ifdef CONFIG_SECURITY 1069 1070 struct security_hook_list capability_hooks[] = { 1071 LSM_HOOK_INIT(capable, cap_capable), 1072 LSM_HOOK_INIT(settime, cap_settime), 1073 LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check), 1074 LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme), 1075 LSM_HOOK_INIT(capget, cap_capget), 1076 LSM_HOOK_INIT(capset, cap_capset), 1077 LSM_HOOK_INIT(bprm_set_creds, cap_bprm_set_creds), 1078 LSM_HOOK_INIT(bprm_secureexec, cap_bprm_secureexec), 1079 LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv), 1080 LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv), 1081 LSM_HOOK_INIT(mmap_addr, cap_mmap_addr), 1082 LSM_HOOK_INIT(mmap_file, cap_mmap_file), 1083 LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid), 1084 LSM_HOOK_INIT(task_prctl, cap_task_prctl), 1085 LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler), 1086 LSM_HOOK_INIT(task_setioprio, cap_task_setioprio), 1087 LSM_HOOK_INIT(task_setnice, cap_task_setnice), 1088 LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory), 1089 }; 1090 1091 void __init capability_add_hooks(void) 1092 { 1093 security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks)); 1094 } 1095 1096 #endif /* CONFIG_SECURITY */ 1097