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 /* 86 * If we're already at a lower level than we're looking for, 87 * we're done searching. 88 */ 89 if (ns->level <= cred->user_ns->level) 90 return -EPERM; 91 92 /* 93 * The owner of the user namespace in the parent of the 94 * user namespace has all caps. 95 */ 96 if ((ns->parent == cred->user_ns) && uid_eq(ns->owner, cred->euid)) 97 return 0; 98 99 /* 100 * If you have a capability in a parent user ns, then you have 101 * it over all children user namespaces as well. 102 */ 103 ns = ns->parent; 104 } 105 106 /* We never get here */ 107 } 108 109 /** 110 * cap_settime - Determine whether the current process may set the system clock 111 * @ts: The time to set 112 * @tz: The timezone to set 113 * 114 * Determine whether the current process may set the system clock and timezone 115 * information, returning 0 if permission granted, -ve if denied. 116 */ 117 int cap_settime(const struct timespec64 *ts, const struct timezone *tz) 118 { 119 if (!capable(CAP_SYS_TIME)) 120 return -EPERM; 121 return 0; 122 } 123 124 /** 125 * cap_ptrace_access_check - Determine whether the current process may access 126 * another 127 * @child: The process to be accessed 128 * @mode: The mode of attachment. 129 * 130 * If we are in the same or an ancestor user_ns and have all the target 131 * task's capabilities, then ptrace access is allowed. 132 * If we have the ptrace capability to the target user_ns, then ptrace 133 * access is allowed. 134 * Else denied. 135 * 136 * Determine whether a process may access another, returning 0 if permission 137 * granted, -ve if denied. 138 */ 139 int cap_ptrace_access_check(struct task_struct *child, unsigned int mode) 140 { 141 int ret = 0; 142 const struct cred *cred, *child_cred; 143 const kernel_cap_t *caller_caps; 144 145 rcu_read_lock(); 146 cred = current_cred(); 147 child_cred = __task_cred(child); 148 if (mode & PTRACE_MODE_FSCREDS) 149 caller_caps = &cred->cap_effective; 150 else 151 caller_caps = &cred->cap_permitted; 152 if (cred->user_ns == child_cred->user_ns && 153 cap_issubset(child_cred->cap_permitted, *caller_caps)) 154 goto out; 155 if (ns_capable(child_cred->user_ns, CAP_SYS_PTRACE)) 156 goto out; 157 ret = -EPERM; 158 out: 159 rcu_read_unlock(); 160 return ret; 161 } 162 163 /** 164 * cap_ptrace_traceme - Determine whether another process may trace the current 165 * @parent: The task proposed to be the tracer 166 * 167 * If parent is in the same or an ancestor user_ns and has all current's 168 * capabilities, then ptrace access is allowed. 169 * If parent has the ptrace capability to current's user_ns, then ptrace 170 * access is allowed. 171 * Else denied. 172 * 173 * Determine whether the nominated task is permitted to trace the current 174 * process, returning 0 if permission is granted, -ve if denied. 175 */ 176 int cap_ptrace_traceme(struct task_struct *parent) 177 { 178 int ret = 0; 179 const struct cred *cred, *child_cred; 180 181 rcu_read_lock(); 182 cred = __task_cred(parent); 183 child_cred = current_cred(); 184 if (cred->user_ns == child_cred->user_ns && 185 cap_issubset(child_cred->cap_permitted, cred->cap_permitted)) 186 goto out; 187 if (has_ns_capability(parent, child_cred->user_ns, CAP_SYS_PTRACE)) 188 goto out; 189 ret = -EPERM; 190 out: 191 rcu_read_unlock(); 192 return ret; 193 } 194 195 /** 196 * cap_capget - Retrieve a task's capability sets 197 * @target: The task from which to retrieve the capability sets 198 * @effective: The place to record the effective set 199 * @inheritable: The place to record the inheritable set 200 * @permitted: The place to record the permitted set 201 * 202 * This function retrieves the capabilities of the nominated task and returns 203 * them to the caller. 204 */ 205 int cap_capget(struct task_struct *target, kernel_cap_t *effective, 206 kernel_cap_t *inheritable, kernel_cap_t *permitted) 207 { 208 const struct cred *cred; 209 210 /* Derived from kernel/capability.c:sys_capget. */ 211 rcu_read_lock(); 212 cred = __task_cred(target); 213 *effective = cred->cap_effective; 214 *inheritable = cred->cap_inheritable; 215 *permitted = cred->cap_permitted; 216 rcu_read_unlock(); 217 return 0; 218 } 219 220 /* 221 * Determine whether the inheritable capabilities are limited to the old 222 * permitted set. Returns 1 if they are limited, 0 if they are not. 223 */ 224 static inline int cap_inh_is_capped(void) 225 { 226 227 /* they are so limited unless the current task has the CAP_SETPCAP 228 * capability 229 */ 230 if (cap_capable(current_cred(), current_cred()->user_ns, 231 CAP_SETPCAP, SECURITY_CAP_AUDIT) == 0) 232 return 0; 233 return 1; 234 } 235 236 /** 237 * cap_capset - Validate and apply proposed changes to current's capabilities 238 * @new: The proposed new credentials; alterations should be made here 239 * @old: The current task's current credentials 240 * @effective: A pointer to the proposed new effective capabilities set 241 * @inheritable: A pointer to the proposed new inheritable capabilities set 242 * @permitted: A pointer to the proposed new permitted capabilities set 243 * 244 * This function validates and applies a proposed mass change to the current 245 * process's capability sets. The changes are made to the proposed new 246 * credentials, and assuming no error, will be committed by the caller of LSM. 247 */ 248 int cap_capset(struct cred *new, 249 const struct cred *old, 250 const kernel_cap_t *effective, 251 const kernel_cap_t *inheritable, 252 const kernel_cap_t *permitted) 253 { 254 if (cap_inh_is_capped() && 255 !cap_issubset(*inheritable, 256 cap_combine(old->cap_inheritable, 257 old->cap_permitted))) 258 /* incapable of using this inheritable set */ 259 return -EPERM; 260 261 if (!cap_issubset(*inheritable, 262 cap_combine(old->cap_inheritable, 263 old->cap_bset))) 264 /* no new pI capabilities outside bounding set */ 265 return -EPERM; 266 267 /* verify restrictions on target's new Permitted set */ 268 if (!cap_issubset(*permitted, old->cap_permitted)) 269 return -EPERM; 270 271 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */ 272 if (!cap_issubset(*effective, *permitted)) 273 return -EPERM; 274 275 new->cap_effective = *effective; 276 new->cap_inheritable = *inheritable; 277 new->cap_permitted = *permitted; 278 279 /* 280 * Mask off ambient bits that are no longer both permitted and 281 * inheritable. 282 */ 283 new->cap_ambient = cap_intersect(new->cap_ambient, 284 cap_intersect(*permitted, 285 *inheritable)); 286 if (WARN_ON(!cap_ambient_invariant_ok(new))) 287 return -EINVAL; 288 return 0; 289 } 290 291 /** 292 * cap_inode_need_killpriv - Determine if inode change affects privileges 293 * @dentry: The inode/dentry in being changed with change marked ATTR_KILL_PRIV 294 * 295 * Determine if an inode having a change applied that's marked ATTR_KILL_PRIV 296 * affects the security markings on that inode, and if it is, should 297 * inode_killpriv() be invoked or the change rejected. 298 * 299 * Returns 1 if security.capability has a value, meaning inode_killpriv() 300 * is required, 0 otherwise, meaning inode_killpriv() is not required. 301 */ 302 int cap_inode_need_killpriv(struct dentry *dentry) 303 { 304 struct inode *inode = d_backing_inode(dentry); 305 int error; 306 307 error = __vfs_getxattr(dentry, inode, XATTR_NAME_CAPS, NULL, 0); 308 return error > 0; 309 } 310 311 /** 312 * cap_inode_killpriv - Erase the security markings on an inode 313 * @dentry: The inode/dentry to alter 314 * 315 * Erase the privilege-enhancing security markings on an inode. 316 * 317 * Returns 0 if successful, -ve on error. 318 */ 319 int cap_inode_killpriv(struct dentry *dentry) 320 { 321 int error; 322 323 error = __vfs_removexattr(dentry, XATTR_NAME_CAPS); 324 if (error == -EOPNOTSUPP) 325 error = 0; 326 return error; 327 } 328 329 static bool rootid_owns_currentns(kuid_t kroot) 330 { 331 struct user_namespace *ns; 332 333 if (!uid_valid(kroot)) 334 return false; 335 336 for (ns = current_user_ns(); ; ns = ns->parent) { 337 if (from_kuid(ns, kroot) == 0) 338 return true; 339 if (ns == &init_user_ns) 340 break; 341 } 342 343 return false; 344 } 345 346 static __u32 sansflags(__u32 m) 347 { 348 return m & ~VFS_CAP_FLAGS_EFFECTIVE; 349 } 350 351 static bool is_v2header(size_t size, __le32 magic) 352 { 353 __u32 m = le32_to_cpu(magic); 354 if (size != XATTR_CAPS_SZ_2) 355 return false; 356 return sansflags(m) == VFS_CAP_REVISION_2; 357 } 358 359 static bool is_v3header(size_t size, __le32 magic) 360 { 361 __u32 m = le32_to_cpu(magic); 362 363 if (size != XATTR_CAPS_SZ_3) 364 return false; 365 return sansflags(m) == VFS_CAP_REVISION_3; 366 } 367 368 /* 369 * getsecurity: We are called for security.* before any attempt to read the 370 * xattr from the inode itself. 371 * 372 * This gives us a chance to read the on-disk value and convert it. If we 373 * return -EOPNOTSUPP, then vfs_getxattr() will call the i_op handler. 374 * 375 * Note we are not called by vfs_getxattr_alloc(), but that is only called 376 * by the integrity subsystem, which really wants the unconverted values - 377 * so that's good. 378 */ 379 int cap_inode_getsecurity(struct inode *inode, const char *name, void **buffer, 380 bool alloc) 381 { 382 int size, ret; 383 kuid_t kroot; 384 uid_t root, mappedroot; 385 char *tmpbuf = NULL; 386 struct vfs_cap_data *cap; 387 struct vfs_ns_cap_data *nscap; 388 struct dentry *dentry; 389 struct user_namespace *fs_ns; 390 391 if (strcmp(name, "capability") != 0) 392 return -EOPNOTSUPP; 393 394 dentry = d_find_alias(inode); 395 if (!dentry) 396 return -EINVAL; 397 398 size = sizeof(struct vfs_ns_cap_data); 399 ret = (int) vfs_getxattr_alloc(dentry, XATTR_NAME_CAPS, 400 &tmpbuf, size, GFP_NOFS); 401 dput(dentry); 402 403 if (ret < 0) 404 return ret; 405 406 fs_ns = inode->i_sb->s_user_ns; 407 cap = (struct vfs_cap_data *) tmpbuf; 408 if (is_v2header((size_t) ret, cap->magic_etc)) { 409 /* If this is sizeof(vfs_cap_data) then we're ok with the 410 * on-disk value, so return that. */ 411 if (alloc) 412 *buffer = tmpbuf; 413 else 414 kfree(tmpbuf); 415 return ret; 416 } else if (!is_v3header((size_t) ret, cap->magic_etc)) { 417 kfree(tmpbuf); 418 return -EINVAL; 419 } 420 421 nscap = (struct vfs_ns_cap_data *) tmpbuf; 422 root = le32_to_cpu(nscap->rootid); 423 kroot = make_kuid(fs_ns, root); 424 425 /* If the root kuid maps to a valid uid in current ns, then return 426 * this as a nscap. */ 427 mappedroot = from_kuid(current_user_ns(), kroot); 428 if (mappedroot != (uid_t)-1 && mappedroot != (uid_t)0) { 429 if (alloc) { 430 *buffer = tmpbuf; 431 nscap->rootid = cpu_to_le32(mappedroot); 432 } else 433 kfree(tmpbuf); 434 return size; 435 } 436 437 if (!rootid_owns_currentns(kroot)) { 438 kfree(tmpbuf); 439 return -EOPNOTSUPP; 440 } 441 442 /* This comes from a parent namespace. Return as a v2 capability */ 443 size = sizeof(struct vfs_cap_data); 444 if (alloc) { 445 *buffer = kmalloc(size, GFP_ATOMIC); 446 if (*buffer) { 447 struct vfs_cap_data *cap = *buffer; 448 __le32 nsmagic, magic; 449 magic = VFS_CAP_REVISION_2; 450 nsmagic = le32_to_cpu(nscap->magic_etc); 451 if (nsmagic & VFS_CAP_FLAGS_EFFECTIVE) 452 magic |= VFS_CAP_FLAGS_EFFECTIVE; 453 memcpy(&cap->data, &nscap->data, sizeof(__le32) * 2 * VFS_CAP_U32); 454 cap->magic_etc = cpu_to_le32(magic); 455 } 456 } 457 kfree(tmpbuf); 458 return size; 459 } 460 461 static kuid_t rootid_from_xattr(const void *value, size_t size, 462 struct user_namespace *task_ns) 463 { 464 const struct vfs_ns_cap_data *nscap = value; 465 uid_t rootid = 0; 466 467 if (size == XATTR_CAPS_SZ_3) 468 rootid = le32_to_cpu(nscap->rootid); 469 470 return make_kuid(task_ns, rootid); 471 } 472 473 static bool validheader(size_t size, __le32 magic) 474 { 475 return is_v2header(size, magic) || is_v3header(size, magic); 476 } 477 478 /* 479 * User requested a write of security.capability. If needed, update the 480 * xattr to change from v2 to v3, or to fixup the v3 rootid. 481 * 482 * If all is ok, we return the new size, on error return < 0. 483 */ 484 int cap_convert_nscap(struct dentry *dentry, void **ivalue, size_t size) 485 { 486 struct vfs_ns_cap_data *nscap; 487 uid_t nsrootid; 488 const struct vfs_cap_data *cap = *ivalue; 489 __u32 magic, nsmagic; 490 struct inode *inode = d_backing_inode(dentry); 491 struct user_namespace *task_ns = current_user_ns(), 492 *fs_ns = inode->i_sb->s_user_ns; 493 kuid_t rootid; 494 size_t newsize; 495 496 if (!*ivalue) 497 return -EINVAL; 498 if (!validheader(size, cap->magic_etc)) 499 return -EINVAL; 500 if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP)) 501 return -EPERM; 502 if (size == XATTR_CAPS_SZ_2) 503 if (ns_capable(inode->i_sb->s_user_ns, CAP_SETFCAP)) 504 /* user is privileged, just write the v2 */ 505 return size; 506 507 rootid = rootid_from_xattr(*ivalue, size, task_ns); 508 if (!uid_valid(rootid)) 509 return -EINVAL; 510 511 nsrootid = from_kuid(fs_ns, rootid); 512 if (nsrootid == -1) 513 return -EINVAL; 514 515 newsize = sizeof(struct vfs_ns_cap_data); 516 nscap = kmalloc(newsize, GFP_ATOMIC); 517 if (!nscap) 518 return -ENOMEM; 519 nscap->rootid = cpu_to_le32(nsrootid); 520 nsmagic = VFS_CAP_REVISION_3; 521 magic = le32_to_cpu(cap->magic_etc); 522 if (magic & VFS_CAP_FLAGS_EFFECTIVE) 523 nsmagic |= VFS_CAP_FLAGS_EFFECTIVE; 524 nscap->magic_etc = cpu_to_le32(nsmagic); 525 memcpy(&nscap->data, &cap->data, sizeof(__le32) * 2 * VFS_CAP_U32); 526 527 kvfree(*ivalue); 528 *ivalue = nscap; 529 return newsize; 530 } 531 532 /* 533 * Calculate the new process capability sets from the capability sets attached 534 * to a file. 535 */ 536 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps, 537 struct linux_binprm *bprm, 538 bool *effective, 539 bool *has_cap) 540 { 541 struct cred *new = bprm->cred; 542 unsigned i; 543 int ret = 0; 544 545 if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE) 546 *effective = true; 547 548 if (caps->magic_etc & VFS_CAP_REVISION_MASK) 549 *has_cap = true; 550 551 CAP_FOR_EACH_U32(i) { 552 __u32 permitted = caps->permitted.cap[i]; 553 __u32 inheritable = caps->inheritable.cap[i]; 554 555 /* 556 * pP' = (X & fP) | (pI & fI) 557 * The addition of pA' is handled later. 558 */ 559 new->cap_permitted.cap[i] = 560 (new->cap_bset.cap[i] & permitted) | 561 (new->cap_inheritable.cap[i] & inheritable); 562 563 if (permitted & ~new->cap_permitted.cap[i]) 564 /* insufficient to execute correctly */ 565 ret = -EPERM; 566 } 567 568 /* 569 * For legacy apps, with no internal support for recognizing they 570 * do not have enough capabilities, we return an error if they are 571 * missing some "forced" (aka file-permitted) capabilities. 572 */ 573 return *effective ? ret : 0; 574 } 575 576 /* 577 * Extract the on-exec-apply capability sets for an executable file. 578 */ 579 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps) 580 { 581 struct inode *inode = d_backing_inode(dentry); 582 __u32 magic_etc; 583 unsigned tocopy, i; 584 int size; 585 struct vfs_ns_cap_data data, *nscaps = &data; 586 struct vfs_cap_data *caps = (struct vfs_cap_data *) &data; 587 kuid_t rootkuid; 588 struct user_namespace *fs_ns; 589 590 memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data)); 591 592 if (!inode) 593 return -ENODATA; 594 595 fs_ns = inode->i_sb->s_user_ns; 596 size = __vfs_getxattr((struct dentry *)dentry, inode, 597 XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ); 598 if (size == -ENODATA || size == -EOPNOTSUPP) 599 /* no data, that's ok */ 600 return -ENODATA; 601 602 if (size < 0) 603 return size; 604 605 if (size < sizeof(magic_etc)) 606 return -EINVAL; 607 608 cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc); 609 610 rootkuid = make_kuid(fs_ns, 0); 611 switch (magic_etc & VFS_CAP_REVISION_MASK) { 612 case VFS_CAP_REVISION_1: 613 if (size != XATTR_CAPS_SZ_1) 614 return -EINVAL; 615 tocopy = VFS_CAP_U32_1; 616 break; 617 case VFS_CAP_REVISION_2: 618 if (size != XATTR_CAPS_SZ_2) 619 return -EINVAL; 620 tocopy = VFS_CAP_U32_2; 621 break; 622 case VFS_CAP_REVISION_3: 623 if (size != XATTR_CAPS_SZ_3) 624 return -EINVAL; 625 tocopy = VFS_CAP_U32_3; 626 rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid)); 627 break; 628 629 default: 630 return -EINVAL; 631 } 632 /* Limit the caps to the mounter of the filesystem 633 * or the more limited uid specified in the xattr. 634 */ 635 if (!rootid_owns_currentns(rootkuid)) 636 return -ENODATA; 637 638 CAP_FOR_EACH_U32(i) { 639 if (i >= tocopy) 640 break; 641 cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted); 642 cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable); 643 } 644 645 cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK; 646 cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK; 647 648 return 0; 649 } 650 651 /* 652 * Attempt to get the on-exec apply capability sets for an executable file from 653 * its xattrs and, if present, apply them to the proposed credentials being 654 * constructed by execve(). 655 */ 656 static int get_file_caps(struct linux_binprm *bprm, bool *effective, bool *has_cap) 657 { 658 int rc = 0; 659 struct cpu_vfs_cap_data vcaps; 660 661 cap_clear(bprm->cred->cap_permitted); 662 663 if (!file_caps_enabled) 664 return 0; 665 666 if (!mnt_may_suid(bprm->file->f_path.mnt)) 667 return 0; 668 669 /* 670 * This check is redundant with mnt_may_suid() but is kept to make 671 * explicit that capability bits are limited to s_user_ns and its 672 * descendants. 673 */ 674 if (!current_in_userns(bprm->file->f_path.mnt->mnt_sb->s_user_ns)) 675 return 0; 676 677 rc = get_vfs_caps_from_disk(bprm->file->f_path.dentry, &vcaps); 678 if (rc < 0) { 679 if (rc == -EINVAL) 680 printk(KERN_NOTICE "Invalid argument reading file caps for %s\n", 681 bprm->filename); 682 else if (rc == -ENODATA) 683 rc = 0; 684 goto out; 685 } 686 687 rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_cap); 688 if (rc == -EINVAL) 689 printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n", 690 __func__, rc, bprm->filename); 691 692 out: 693 if (rc) 694 cap_clear(bprm->cred->cap_permitted); 695 696 return rc; 697 } 698 699 /** 700 * cap_bprm_set_creds - Set up the proposed credentials for execve(). 701 * @bprm: The execution parameters, including the proposed creds 702 * 703 * Set up the proposed credentials for a new execution context being 704 * constructed by execve(). The proposed creds in @bprm->cred is altered, 705 * which won't take effect immediately. Returns 0 if successful, -ve on error. 706 */ 707 int cap_bprm_set_creds(struct linux_binprm *bprm) 708 { 709 const struct cred *old = current_cred(); 710 struct cred *new = bprm->cred; 711 bool effective, has_cap = false, is_setid; 712 int ret; 713 kuid_t root_uid; 714 715 if (WARN_ON(!cap_ambient_invariant_ok(old))) 716 return -EPERM; 717 718 effective = false; 719 ret = get_file_caps(bprm, &effective, &has_cap); 720 if (ret < 0) 721 return ret; 722 723 root_uid = make_kuid(new->user_ns, 0); 724 725 if (!issecure(SECURE_NOROOT)) { 726 /* 727 * If the legacy file capability is set, then don't set privs 728 * for a setuid root binary run by a non-root user. Do set it 729 * for a root user just to cause least surprise to an admin. 730 */ 731 if (has_cap && !uid_eq(new->uid, root_uid) && uid_eq(new->euid, root_uid)) { 732 warn_setuid_and_fcaps_mixed(bprm->filename); 733 goto skip; 734 } 735 /* 736 * To support inheritance of root-permissions and suid-root 737 * executables under compatibility mode, we override the 738 * capability sets for the file. 739 * 740 * If only the real uid is 0, we do not set the effective bit. 741 */ 742 if (uid_eq(new->euid, root_uid) || uid_eq(new->uid, root_uid)) { 743 /* pP' = (cap_bset & ~0) | (pI & ~0) */ 744 new->cap_permitted = cap_combine(old->cap_bset, 745 old->cap_inheritable); 746 } 747 if (uid_eq(new->euid, root_uid)) 748 effective = true; 749 } 750 skip: 751 752 /* if we have fs caps, clear dangerous personality flags */ 753 if (!cap_issubset(new->cap_permitted, old->cap_permitted)) 754 bprm->per_clear |= PER_CLEAR_ON_SETID; 755 756 757 /* Don't let someone trace a set[ug]id/setpcap binary with the revised 758 * credentials unless they have the appropriate permit. 759 * 760 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs. 761 */ 762 is_setid = !uid_eq(new->euid, old->uid) || !gid_eq(new->egid, old->gid); 763 764 if ((is_setid || 765 !cap_issubset(new->cap_permitted, old->cap_permitted)) && 766 ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) || 767 !ptracer_capable(current, new->user_ns))) { 768 /* downgrade; they get no more than they had, and maybe less */ 769 if (!ns_capable(new->user_ns, CAP_SETUID) || 770 (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) { 771 new->euid = new->uid; 772 new->egid = new->gid; 773 } 774 new->cap_permitted = cap_intersect(new->cap_permitted, 775 old->cap_permitted); 776 } 777 778 new->suid = new->fsuid = new->euid; 779 new->sgid = new->fsgid = new->egid; 780 781 /* File caps or setid cancels ambient. */ 782 if (has_cap || is_setid) 783 cap_clear(new->cap_ambient); 784 785 /* 786 * Now that we've computed pA', update pP' to give: 787 * pP' = (X & fP) | (pI & fI) | pA' 788 */ 789 new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient); 790 791 /* 792 * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set, 793 * this is the same as pE' = (fE ? pP' : 0) | pA'. 794 */ 795 if (effective) 796 new->cap_effective = new->cap_permitted; 797 else 798 new->cap_effective = new->cap_ambient; 799 800 if (WARN_ON(!cap_ambient_invariant_ok(new))) 801 return -EPERM; 802 803 /* 804 * Audit candidate if current->cap_effective is set 805 * 806 * We do not bother to audit if 3 things are true: 807 * 1) cap_effective has all caps 808 * 2) we are root 809 * 3) root is supposed to have all caps (SECURE_NOROOT) 810 * Since this is just a normal root execing a process. 811 * 812 * Number 1 above might fail if you don't have a full bset, but I think 813 * that is interesting information to audit. 814 */ 815 if (!cap_issubset(new->cap_effective, new->cap_ambient)) { 816 if (!cap_issubset(CAP_FULL_SET, new->cap_effective) || 817 !uid_eq(new->euid, root_uid) || !uid_eq(new->uid, root_uid) || 818 issecure(SECURE_NOROOT)) { 819 ret = audit_log_bprm_fcaps(bprm, new, old); 820 if (ret < 0) 821 return ret; 822 } 823 } 824 825 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS); 826 827 if (WARN_ON(!cap_ambient_invariant_ok(new))) 828 return -EPERM; 829 830 /* Check for privilege-elevated exec. */ 831 bprm->cap_elevated = 0; 832 if (is_setid) { 833 bprm->cap_elevated = 1; 834 } else if (!uid_eq(new->uid, root_uid)) { 835 if (effective || 836 !cap_issubset(new->cap_permitted, new->cap_ambient)) 837 bprm->cap_elevated = 1; 838 } 839 840 return 0; 841 } 842 843 /** 844 * cap_inode_setxattr - Determine whether an xattr may be altered 845 * @dentry: The inode/dentry being altered 846 * @name: The name of the xattr to be changed 847 * @value: The value that the xattr will be changed to 848 * @size: The size of value 849 * @flags: The replacement flag 850 * 851 * Determine whether an xattr may be altered or set on an inode, returning 0 if 852 * permission is granted, -ve if denied. 853 * 854 * This is used to make sure security xattrs don't get updated or set by those 855 * who aren't privileged to do so. 856 */ 857 int cap_inode_setxattr(struct dentry *dentry, const char *name, 858 const void *value, size_t size, int flags) 859 { 860 /* Ignore non-security xattrs */ 861 if (strncmp(name, XATTR_SECURITY_PREFIX, 862 sizeof(XATTR_SECURITY_PREFIX) - 1) != 0) 863 return 0; 864 865 /* 866 * For XATTR_NAME_CAPS the check will be done in 867 * cap_convert_nscap(), called by setxattr() 868 */ 869 if (strcmp(name, XATTR_NAME_CAPS) == 0) 870 return 0; 871 872 if (!capable(CAP_SYS_ADMIN)) 873 return -EPERM; 874 return 0; 875 } 876 877 /** 878 * cap_inode_removexattr - Determine whether an xattr may be removed 879 * @dentry: The inode/dentry being altered 880 * @name: The name of the xattr to be changed 881 * 882 * Determine whether an xattr may be removed from an inode, returning 0 if 883 * permission is granted, -ve if denied. 884 * 885 * This is used to make sure security xattrs don't get removed by those who 886 * aren't privileged to remove them. 887 */ 888 int cap_inode_removexattr(struct dentry *dentry, const char *name) 889 { 890 /* Ignore non-security xattrs */ 891 if (strncmp(name, XATTR_SECURITY_PREFIX, 892 sizeof(XATTR_SECURITY_PREFIX) - 1) != 0) 893 return 0; 894 895 if (strcmp(name, XATTR_NAME_CAPS) == 0) { 896 /* security.capability gets namespaced */ 897 struct inode *inode = d_backing_inode(dentry); 898 if (!inode) 899 return -EINVAL; 900 if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP)) 901 return -EPERM; 902 return 0; 903 } 904 905 if (!capable(CAP_SYS_ADMIN)) 906 return -EPERM; 907 return 0; 908 } 909 910 /* 911 * cap_emulate_setxuid() fixes the effective / permitted capabilities of 912 * a process after a call to setuid, setreuid, or setresuid. 913 * 914 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of 915 * {r,e,s}uid != 0, the permitted and effective capabilities are 916 * cleared. 917 * 918 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective 919 * capabilities of the process are cleared. 920 * 921 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective 922 * capabilities are set to the permitted capabilities. 923 * 924 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should 925 * never happen. 926 * 927 * -astor 928 * 929 * cevans - New behaviour, Oct '99 930 * A process may, via prctl(), elect to keep its capabilities when it 931 * calls setuid() and switches away from uid==0. Both permitted and 932 * effective sets will be retained. 933 * Without this change, it was impossible for a daemon to drop only some 934 * of its privilege. The call to setuid(!=0) would drop all privileges! 935 * Keeping uid 0 is not an option because uid 0 owns too many vital 936 * files.. 937 * Thanks to Olaf Kirch and Peter Benie for spotting this. 938 */ 939 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old) 940 { 941 kuid_t root_uid = make_kuid(old->user_ns, 0); 942 943 if ((uid_eq(old->uid, root_uid) || 944 uid_eq(old->euid, root_uid) || 945 uid_eq(old->suid, root_uid)) && 946 (!uid_eq(new->uid, root_uid) && 947 !uid_eq(new->euid, root_uid) && 948 !uid_eq(new->suid, root_uid))) { 949 if (!issecure(SECURE_KEEP_CAPS)) { 950 cap_clear(new->cap_permitted); 951 cap_clear(new->cap_effective); 952 } 953 954 /* 955 * Pre-ambient programs expect setresuid to nonroot followed 956 * by exec to drop capabilities. We should make sure that 957 * this remains the case. 958 */ 959 cap_clear(new->cap_ambient); 960 } 961 if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid)) 962 cap_clear(new->cap_effective); 963 if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid)) 964 new->cap_effective = new->cap_permitted; 965 } 966 967 /** 968 * cap_task_fix_setuid - Fix up the results of setuid() call 969 * @new: The proposed credentials 970 * @old: The current task's current credentials 971 * @flags: Indications of what has changed 972 * 973 * Fix up the results of setuid() call before the credential changes are 974 * actually applied, returning 0 to grant the changes, -ve to deny them. 975 */ 976 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags) 977 { 978 switch (flags) { 979 case LSM_SETID_RE: 980 case LSM_SETID_ID: 981 case LSM_SETID_RES: 982 /* juggle the capabilities to follow [RES]UID changes unless 983 * otherwise suppressed */ 984 if (!issecure(SECURE_NO_SETUID_FIXUP)) 985 cap_emulate_setxuid(new, old); 986 break; 987 988 case LSM_SETID_FS: 989 /* juggle the capabilties to follow FSUID changes, unless 990 * otherwise suppressed 991 * 992 * FIXME - is fsuser used for all CAP_FS_MASK capabilities? 993 * if not, we might be a bit too harsh here. 994 */ 995 if (!issecure(SECURE_NO_SETUID_FIXUP)) { 996 kuid_t root_uid = make_kuid(old->user_ns, 0); 997 if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid)) 998 new->cap_effective = 999 cap_drop_fs_set(new->cap_effective); 1000 1001 if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid)) 1002 new->cap_effective = 1003 cap_raise_fs_set(new->cap_effective, 1004 new->cap_permitted); 1005 } 1006 break; 1007 1008 default: 1009 return -EINVAL; 1010 } 1011 1012 return 0; 1013 } 1014 1015 /* 1016 * Rationale: code calling task_setscheduler, task_setioprio, and 1017 * task_setnice, assumes that 1018 * . if capable(cap_sys_nice), then those actions should be allowed 1019 * . if not capable(cap_sys_nice), but acting on your own processes, 1020 * then those actions should be allowed 1021 * This is insufficient now since you can call code without suid, but 1022 * yet with increased caps. 1023 * So we check for increased caps on the target process. 1024 */ 1025 static int cap_safe_nice(struct task_struct *p) 1026 { 1027 int is_subset, ret = 0; 1028 1029 rcu_read_lock(); 1030 is_subset = cap_issubset(__task_cred(p)->cap_permitted, 1031 current_cred()->cap_permitted); 1032 if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) 1033 ret = -EPERM; 1034 rcu_read_unlock(); 1035 1036 return ret; 1037 } 1038 1039 /** 1040 * cap_task_setscheduler - Detemine if scheduler policy change is permitted 1041 * @p: The task to affect 1042 * 1043 * Detemine if the requested scheduler policy change is permitted for the 1044 * specified task, returning 0 if permission is granted, -ve if denied. 1045 */ 1046 int cap_task_setscheduler(struct task_struct *p) 1047 { 1048 return cap_safe_nice(p); 1049 } 1050 1051 /** 1052 * cap_task_ioprio - Detemine if I/O priority change is permitted 1053 * @p: The task to affect 1054 * @ioprio: The I/O priority to set 1055 * 1056 * Detemine if the requested I/O priority change is permitted for the specified 1057 * task, returning 0 if permission is granted, -ve if denied. 1058 */ 1059 int cap_task_setioprio(struct task_struct *p, int ioprio) 1060 { 1061 return cap_safe_nice(p); 1062 } 1063 1064 /** 1065 * cap_task_ioprio - Detemine if task priority change is permitted 1066 * @p: The task to affect 1067 * @nice: The nice value to set 1068 * 1069 * Detemine if the requested task priority change is permitted for the 1070 * specified task, returning 0 if permission is granted, -ve if denied. 1071 */ 1072 int cap_task_setnice(struct task_struct *p, int nice) 1073 { 1074 return cap_safe_nice(p); 1075 } 1076 1077 /* 1078 * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from 1079 * the current task's bounding set. Returns 0 on success, -ve on error. 1080 */ 1081 static int cap_prctl_drop(unsigned long cap) 1082 { 1083 struct cred *new; 1084 1085 if (!ns_capable(current_user_ns(), CAP_SETPCAP)) 1086 return -EPERM; 1087 if (!cap_valid(cap)) 1088 return -EINVAL; 1089 1090 new = prepare_creds(); 1091 if (!new) 1092 return -ENOMEM; 1093 cap_lower(new->cap_bset, cap); 1094 return commit_creds(new); 1095 } 1096 1097 /** 1098 * cap_task_prctl - Implement process control functions for this security module 1099 * @option: The process control function requested 1100 * @arg2, @arg3, @arg4, @arg5: The argument data for this function 1101 * 1102 * Allow process control functions (sys_prctl()) to alter capabilities; may 1103 * also deny access to other functions not otherwise implemented here. 1104 * 1105 * Returns 0 or +ve on success, -ENOSYS if this function is not implemented 1106 * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM 1107 * modules will consider performing the function. 1108 */ 1109 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3, 1110 unsigned long arg4, unsigned long arg5) 1111 { 1112 const struct cred *old = current_cred(); 1113 struct cred *new; 1114 1115 switch (option) { 1116 case PR_CAPBSET_READ: 1117 if (!cap_valid(arg2)) 1118 return -EINVAL; 1119 return !!cap_raised(old->cap_bset, arg2); 1120 1121 case PR_CAPBSET_DROP: 1122 return cap_prctl_drop(arg2); 1123 1124 /* 1125 * The next four prctl's remain to assist with transitioning a 1126 * system from legacy UID=0 based privilege (when filesystem 1127 * capabilities are not in use) to a system using filesystem 1128 * capabilities only - as the POSIX.1e draft intended. 1129 * 1130 * Note: 1131 * 1132 * PR_SET_SECUREBITS = 1133 * issecure_mask(SECURE_KEEP_CAPS_LOCKED) 1134 * | issecure_mask(SECURE_NOROOT) 1135 * | issecure_mask(SECURE_NOROOT_LOCKED) 1136 * | issecure_mask(SECURE_NO_SETUID_FIXUP) 1137 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED) 1138 * 1139 * will ensure that the current process and all of its 1140 * children will be locked into a pure 1141 * capability-based-privilege environment. 1142 */ 1143 case PR_SET_SECUREBITS: 1144 if ((((old->securebits & SECURE_ALL_LOCKS) >> 1) 1145 & (old->securebits ^ arg2)) /*[1]*/ 1146 || ((old->securebits & SECURE_ALL_LOCKS & ~arg2)) /*[2]*/ 1147 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/ 1148 || (cap_capable(current_cred(), 1149 current_cred()->user_ns, CAP_SETPCAP, 1150 SECURITY_CAP_AUDIT) != 0) /*[4]*/ 1151 /* 1152 * [1] no changing of bits that are locked 1153 * [2] no unlocking of locks 1154 * [3] no setting of unsupported bits 1155 * [4] doing anything requires privilege (go read about 1156 * the "sendmail capabilities bug") 1157 */ 1158 ) 1159 /* cannot change a locked bit */ 1160 return -EPERM; 1161 1162 new = prepare_creds(); 1163 if (!new) 1164 return -ENOMEM; 1165 new->securebits = arg2; 1166 return commit_creds(new); 1167 1168 case PR_GET_SECUREBITS: 1169 return old->securebits; 1170 1171 case PR_GET_KEEPCAPS: 1172 return !!issecure(SECURE_KEEP_CAPS); 1173 1174 case PR_SET_KEEPCAPS: 1175 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */ 1176 return -EINVAL; 1177 if (issecure(SECURE_KEEP_CAPS_LOCKED)) 1178 return -EPERM; 1179 1180 new = prepare_creds(); 1181 if (!new) 1182 return -ENOMEM; 1183 if (arg2) 1184 new->securebits |= issecure_mask(SECURE_KEEP_CAPS); 1185 else 1186 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS); 1187 return commit_creds(new); 1188 1189 case PR_CAP_AMBIENT: 1190 if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) { 1191 if (arg3 | arg4 | arg5) 1192 return -EINVAL; 1193 1194 new = prepare_creds(); 1195 if (!new) 1196 return -ENOMEM; 1197 cap_clear(new->cap_ambient); 1198 return commit_creds(new); 1199 } 1200 1201 if (((!cap_valid(arg3)) | arg4 | arg5)) 1202 return -EINVAL; 1203 1204 if (arg2 == PR_CAP_AMBIENT_IS_SET) { 1205 return !!cap_raised(current_cred()->cap_ambient, arg3); 1206 } else if (arg2 != PR_CAP_AMBIENT_RAISE && 1207 arg2 != PR_CAP_AMBIENT_LOWER) { 1208 return -EINVAL; 1209 } else { 1210 if (arg2 == PR_CAP_AMBIENT_RAISE && 1211 (!cap_raised(current_cred()->cap_permitted, arg3) || 1212 !cap_raised(current_cred()->cap_inheritable, 1213 arg3) || 1214 issecure(SECURE_NO_CAP_AMBIENT_RAISE))) 1215 return -EPERM; 1216 1217 new = prepare_creds(); 1218 if (!new) 1219 return -ENOMEM; 1220 if (arg2 == PR_CAP_AMBIENT_RAISE) 1221 cap_raise(new->cap_ambient, arg3); 1222 else 1223 cap_lower(new->cap_ambient, arg3); 1224 return commit_creds(new); 1225 } 1226 1227 default: 1228 /* No functionality available - continue with default */ 1229 return -ENOSYS; 1230 } 1231 } 1232 1233 /** 1234 * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted 1235 * @mm: The VM space in which the new mapping is to be made 1236 * @pages: The size of the mapping 1237 * 1238 * Determine whether the allocation of a new virtual mapping by the current 1239 * task is permitted, returning 1 if permission is granted, 0 if not. 1240 */ 1241 int cap_vm_enough_memory(struct mm_struct *mm, long pages) 1242 { 1243 int cap_sys_admin = 0; 1244 1245 if (cap_capable(current_cred(), &init_user_ns, CAP_SYS_ADMIN, 1246 SECURITY_CAP_NOAUDIT) == 0) 1247 cap_sys_admin = 1; 1248 return cap_sys_admin; 1249 } 1250 1251 /* 1252 * cap_mmap_addr - check if able to map given addr 1253 * @addr: address attempting to be mapped 1254 * 1255 * If the process is attempting to map memory below dac_mmap_min_addr they need 1256 * CAP_SYS_RAWIO. The other parameters to this function are unused by the 1257 * capability security module. Returns 0 if this mapping should be allowed 1258 * -EPERM if not. 1259 */ 1260 int cap_mmap_addr(unsigned long addr) 1261 { 1262 int ret = 0; 1263 1264 if (addr < dac_mmap_min_addr) { 1265 ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO, 1266 SECURITY_CAP_AUDIT); 1267 /* set PF_SUPERPRIV if it turns out we allow the low mmap */ 1268 if (ret == 0) 1269 current->flags |= PF_SUPERPRIV; 1270 } 1271 return ret; 1272 } 1273 1274 int cap_mmap_file(struct file *file, unsigned long reqprot, 1275 unsigned long prot, unsigned long flags) 1276 { 1277 return 0; 1278 } 1279 1280 #ifdef CONFIG_SECURITY 1281 1282 struct security_hook_list capability_hooks[] __lsm_ro_after_init = { 1283 LSM_HOOK_INIT(capable, cap_capable), 1284 LSM_HOOK_INIT(settime, cap_settime), 1285 LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check), 1286 LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme), 1287 LSM_HOOK_INIT(capget, cap_capget), 1288 LSM_HOOK_INIT(capset, cap_capset), 1289 LSM_HOOK_INIT(bprm_set_creds, cap_bprm_set_creds), 1290 LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv), 1291 LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv), 1292 LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity), 1293 LSM_HOOK_INIT(mmap_addr, cap_mmap_addr), 1294 LSM_HOOK_INIT(mmap_file, cap_mmap_file), 1295 LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid), 1296 LSM_HOOK_INIT(task_prctl, cap_task_prctl), 1297 LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler), 1298 LSM_HOOK_INIT(task_setioprio, cap_task_setioprio), 1299 LSM_HOOK_INIT(task_setnice, cap_task_setnice), 1300 LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory), 1301 }; 1302 1303 void __init capability_add_hooks(void) 1304 { 1305 security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks), 1306 "capability"); 1307 } 1308 1309 #endif /* CONFIG_SECURITY */ 1310