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 = inode->i_sb->s_user_ns; 589 590 memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data)); 591 592 if (!inode) 593 return -ENODATA; 594 595 size = __vfs_getxattr((struct dentry *)dentry, inode, 596 XATTR_NAME_CAPS, &data, XATTR_CAPS_SZ); 597 if (size == -ENODATA || size == -EOPNOTSUPP) 598 /* no data, that's ok */ 599 return -ENODATA; 600 601 if (size < 0) 602 return size; 603 604 if (size < sizeof(magic_etc)) 605 return -EINVAL; 606 607 cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps->magic_etc); 608 609 rootkuid = make_kuid(fs_ns, 0); 610 switch (magic_etc & VFS_CAP_REVISION_MASK) { 611 case VFS_CAP_REVISION_1: 612 if (size != XATTR_CAPS_SZ_1) 613 return -EINVAL; 614 tocopy = VFS_CAP_U32_1; 615 break; 616 case VFS_CAP_REVISION_2: 617 if (size != XATTR_CAPS_SZ_2) 618 return -EINVAL; 619 tocopy = VFS_CAP_U32_2; 620 break; 621 case VFS_CAP_REVISION_3: 622 if (size != XATTR_CAPS_SZ_3) 623 return -EINVAL; 624 tocopy = VFS_CAP_U32_3; 625 rootkuid = make_kuid(fs_ns, le32_to_cpu(nscaps->rootid)); 626 break; 627 628 default: 629 return -EINVAL; 630 } 631 /* Limit the caps to the mounter of the filesystem 632 * or the more limited uid specified in the xattr. 633 */ 634 if (!rootid_owns_currentns(rootkuid)) 635 return -ENODATA; 636 637 CAP_FOR_EACH_U32(i) { 638 if (i >= tocopy) 639 break; 640 cpu_caps->permitted.cap[i] = le32_to_cpu(caps->data[i].permitted); 641 cpu_caps->inheritable.cap[i] = le32_to_cpu(caps->data[i].inheritable); 642 } 643 644 cpu_caps->permitted.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK; 645 cpu_caps->inheritable.cap[CAP_LAST_U32] &= CAP_LAST_U32_VALID_MASK; 646 647 return 0; 648 } 649 650 /* 651 * Attempt to get the on-exec apply capability sets for an executable file from 652 * its xattrs and, if present, apply them to the proposed credentials being 653 * constructed by execve(). 654 */ 655 static int get_file_caps(struct linux_binprm *bprm, bool *effective, bool *has_cap) 656 { 657 int rc = 0; 658 struct cpu_vfs_cap_data vcaps; 659 660 cap_clear(bprm->cred->cap_permitted); 661 662 if (!file_caps_enabled) 663 return 0; 664 665 if (!mnt_may_suid(bprm->file->f_path.mnt)) 666 return 0; 667 668 /* 669 * This check is redundant with mnt_may_suid() but is kept to make 670 * explicit that capability bits are limited to s_user_ns and its 671 * descendants. 672 */ 673 if (!current_in_userns(bprm->file->f_path.mnt->mnt_sb->s_user_ns)) 674 return 0; 675 676 rc = get_vfs_caps_from_disk(bprm->file->f_path.dentry, &vcaps); 677 if (rc < 0) { 678 if (rc == -EINVAL) 679 printk(KERN_NOTICE "Invalid argument reading file caps for %s\n", 680 bprm->filename); 681 else if (rc == -ENODATA) 682 rc = 0; 683 goto out; 684 } 685 686 rc = bprm_caps_from_vfs_caps(&vcaps, bprm, effective, has_cap); 687 if (rc == -EINVAL) 688 printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n", 689 __func__, rc, bprm->filename); 690 691 out: 692 if (rc) 693 cap_clear(bprm->cred->cap_permitted); 694 695 return rc; 696 } 697 698 /** 699 * cap_bprm_set_creds - Set up the proposed credentials for execve(). 700 * @bprm: The execution parameters, including the proposed creds 701 * 702 * Set up the proposed credentials for a new execution context being 703 * constructed by execve(). The proposed creds in @bprm->cred is altered, 704 * which won't take effect immediately. Returns 0 if successful, -ve on error. 705 */ 706 int cap_bprm_set_creds(struct linux_binprm *bprm) 707 { 708 const struct cred *old = current_cred(); 709 struct cred *new = bprm->cred; 710 bool effective, has_cap = false, is_setid; 711 int ret; 712 kuid_t root_uid; 713 714 if (WARN_ON(!cap_ambient_invariant_ok(old))) 715 return -EPERM; 716 717 effective = false; 718 ret = get_file_caps(bprm, &effective, &has_cap); 719 if (ret < 0) 720 return ret; 721 722 root_uid = make_kuid(new->user_ns, 0); 723 724 if (!issecure(SECURE_NOROOT)) { 725 /* 726 * If the legacy file capability is set, then don't set privs 727 * for a setuid root binary run by a non-root user. Do set it 728 * for a root user just to cause least surprise to an admin. 729 */ 730 if (has_cap && !uid_eq(new->uid, root_uid) && uid_eq(new->euid, root_uid)) { 731 warn_setuid_and_fcaps_mixed(bprm->filename); 732 goto skip; 733 } 734 /* 735 * To support inheritance of root-permissions and suid-root 736 * executables under compatibility mode, we override the 737 * capability sets for the file. 738 * 739 * If only the real uid is 0, we do not set the effective bit. 740 */ 741 if (uid_eq(new->euid, root_uid) || uid_eq(new->uid, root_uid)) { 742 /* pP' = (cap_bset & ~0) | (pI & ~0) */ 743 new->cap_permitted = cap_combine(old->cap_bset, 744 old->cap_inheritable); 745 } 746 if (uid_eq(new->euid, root_uid)) 747 effective = true; 748 } 749 skip: 750 751 /* if we have fs caps, clear dangerous personality flags */ 752 if (!cap_issubset(new->cap_permitted, old->cap_permitted)) 753 bprm->per_clear |= PER_CLEAR_ON_SETID; 754 755 756 /* Don't let someone trace a set[ug]id/setpcap binary with the revised 757 * credentials unless they have the appropriate permit. 758 * 759 * In addition, if NO_NEW_PRIVS, then ensure we get no new privs. 760 */ 761 is_setid = !uid_eq(new->euid, old->uid) || !gid_eq(new->egid, old->gid); 762 763 if ((is_setid || 764 !cap_issubset(new->cap_permitted, old->cap_permitted)) && 765 ((bprm->unsafe & ~LSM_UNSAFE_PTRACE) || 766 !ptracer_capable(current, new->user_ns))) { 767 /* downgrade; they get no more than they had, and maybe less */ 768 if (!ns_capable(new->user_ns, CAP_SETUID) || 769 (bprm->unsafe & LSM_UNSAFE_NO_NEW_PRIVS)) { 770 new->euid = new->uid; 771 new->egid = new->gid; 772 } 773 new->cap_permitted = cap_intersect(new->cap_permitted, 774 old->cap_permitted); 775 } 776 777 new->suid = new->fsuid = new->euid; 778 new->sgid = new->fsgid = new->egid; 779 780 /* File caps or setid cancels ambient. */ 781 if (has_cap || is_setid) 782 cap_clear(new->cap_ambient); 783 784 /* 785 * Now that we've computed pA', update pP' to give: 786 * pP' = (X & fP) | (pI & fI) | pA' 787 */ 788 new->cap_permitted = cap_combine(new->cap_permitted, new->cap_ambient); 789 790 /* 791 * Set pE' = (fE ? pP' : pA'). Because pA' is zero if fE is set, 792 * this is the same as pE' = (fE ? pP' : 0) | pA'. 793 */ 794 if (effective) 795 new->cap_effective = new->cap_permitted; 796 else 797 new->cap_effective = new->cap_ambient; 798 799 if (WARN_ON(!cap_ambient_invariant_ok(new))) 800 return -EPERM; 801 802 /* 803 * Audit candidate if current->cap_effective is set 804 * 805 * We do not bother to audit if 3 things are true: 806 * 1) cap_effective has all caps 807 * 2) we are root 808 * 3) root is supposed to have all caps (SECURE_NOROOT) 809 * Since this is just a normal root execing a process. 810 * 811 * Number 1 above might fail if you don't have a full bset, but I think 812 * that is interesting information to audit. 813 */ 814 if (!cap_issubset(new->cap_effective, new->cap_ambient)) { 815 if (!cap_issubset(CAP_FULL_SET, new->cap_effective) || 816 !uid_eq(new->euid, root_uid) || !uid_eq(new->uid, root_uid) || 817 issecure(SECURE_NOROOT)) { 818 ret = audit_log_bprm_fcaps(bprm, new, old); 819 if (ret < 0) 820 return ret; 821 } 822 } 823 824 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS); 825 826 if (WARN_ON(!cap_ambient_invariant_ok(new))) 827 return -EPERM; 828 829 /* Check for privilege-elevated exec. */ 830 bprm->cap_elevated = 0; 831 if (is_setid) { 832 bprm->cap_elevated = 1; 833 } else if (!uid_eq(new->uid, root_uid)) { 834 if (effective || 835 !cap_issubset(new->cap_permitted, new->cap_ambient)) 836 bprm->cap_elevated = 1; 837 } 838 839 return 0; 840 } 841 842 /** 843 * cap_inode_setxattr - Determine whether an xattr may be altered 844 * @dentry: The inode/dentry being altered 845 * @name: The name of the xattr to be changed 846 * @value: The value that the xattr will be changed to 847 * @size: The size of value 848 * @flags: The replacement flag 849 * 850 * Determine whether an xattr may be altered or set on an inode, returning 0 if 851 * permission is granted, -ve if denied. 852 * 853 * This is used to make sure security xattrs don't get updated or set by those 854 * who aren't privileged to do so. 855 */ 856 int cap_inode_setxattr(struct dentry *dentry, const char *name, 857 const void *value, size_t size, int flags) 858 { 859 /* Ignore non-security xattrs */ 860 if (strncmp(name, XATTR_SECURITY_PREFIX, 861 sizeof(XATTR_SECURITY_PREFIX) - 1) != 0) 862 return 0; 863 864 /* 865 * For XATTR_NAME_CAPS the check will be done in 866 * cap_convert_nscap(), called by setxattr() 867 */ 868 if (strcmp(name, XATTR_NAME_CAPS) == 0) 869 return 0; 870 871 if (!capable(CAP_SYS_ADMIN)) 872 return -EPERM; 873 return 0; 874 } 875 876 /** 877 * cap_inode_removexattr - Determine whether an xattr may be removed 878 * @dentry: The inode/dentry being altered 879 * @name: The name of the xattr to be changed 880 * 881 * Determine whether an xattr may be removed from an inode, returning 0 if 882 * permission is granted, -ve if denied. 883 * 884 * This is used to make sure security xattrs don't get removed by those who 885 * aren't privileged to remove them. 886 */ 887 int cap_inode_removexattr(struct dentry *dentry, const char *name) 888 { 889 /* Ignore non-security xattrs */ 890 if (strncmp(name, XATTR_SECURITY_PREFIX, 891 sizeof(XATTR_SECURITY_PREFIX) - 1) != 0) 892 return 0; 893 894 if (strcmp(name, XATTR_NAME_CAPS) == 0) { 895 /* security.capability gets namespaced */ 896 struct inode *inode = d_backing_inode(dentry); 897 if (!inode) 898 return -EINVAL; 899 if (!capable_wrt_inode_uidgid(inode, CAP_SETFCAP)) 900 return -EPERM; 901 return 0; 902 } 903 904 if (!capable(CAP_SYS_ADMIN)) 905 return -EPERM; 906 return 0; 907 } 908 909 /* 910 * cap_emulate_setxuid() fixes the effective / permitted capabilities of 911 * a process after a call to setuid, setreuid, or setresuid. 912 * 913 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of 914 * {r,e,s}uid != 0, the permitted and effective capabilities are 915 * cleared. 916 * 917 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective 918 * capabilities of the process are cleared. 919 * 920 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective 921 * capabilities are set to the permitted capabilities. 922 * 923 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should 924 * never happen. 925 * 926 * -astor 927 * 928 * cevans - New behaviour, Oct '99 929 * A process may, via prctl(), elect to keep its capabilities when it 930 * calls setuid() and switches away from uid==0. Both permitted and 931 * effective sets will be retained. 932 * Without this change, it was impossible for a daemon to drop only some 933 * of its privilege. The call to setuid(!=0) would drop all privileges! 934 * Keeping uid 0 is not an option because uid 0 owns too many vital 935 * files.. 936 * Thanks to Olaf Kirch and Peter Benie for spotting this. 937 */ 938 static inline void cap_emulate_setxuid(struct cred *new, const struct cred *old) 939 { 940 kuid_t root_uid = make_kuid(old->user_ns, 0); 941 942 if ((uid_eq(old->uid, root_uid) || 943 uid_eq(old->euid, root_uid) || 944 uid_eq(old->suid, root_uid)) && 945 (!uid_eq(new->uid, root_uid) && 946 !uid_eq(new->euid, root_uid) && 947 !uid_eq(new->suid, root_uid))) { 948 if (!issecure(SECURE_KEEP_CAPS)) { 949 cap_clear(new->cap_permitted); 950 cap_clear(new->cap_effective); 951 } 952 953 /* 954 * Pre-ambient programs expect setresuid to nonroot followed 955 * by exec to drop capabilities. We should make sure that 956 * this remains the case. 957 */ 958 cap_clear(new->cap_ambient); 959 } 960 if (uid_eq(old->euid, root_uid) && !uid_eq(new->euid, root_uid)) 961 cap_clear(new->cap_effective); 962 if (!uid_eq(old->euid, root_uid) && uid_eq(new->euid, root_uid)) 963 new->cap_effective = new->cap_permitted; 964 } 965 966 /** 967 * cap_task_fix_setuid - Fix up the results of setuid() call 968 * @new: The proposed credentials 969 * @old: The current task's current credentials 970 * @flags: Indications of what has changed 971 * 972 * Fix up the results of setuid() call before the credential changes are 973 * actually applied, returning 0 to grant the changes, -ve to deny them. 974 */ 975 int cap_task_fix_setuid(struct cred *new, const struct cred *old, int flags) 976 { 977 switch (flags) { 978 case LSM_SETID_RE: 979 case LSM_SETID_ID: 980 case LSM_SETID_RES: 981 /* juggle the capabilities to follow [RES]UID changes unless 982 * otherwise suppressed */ 983 if (!issecure(SECURE_NO_SETUID_FIXUP)) 984 cap_emulate_setxuid(new, old); 985 break; 986 987 case LSM_SETID_FS: 988 /* juggle the capabilties to follow FSUID changes, unless 989 * otherwise suppressed 990 * 991 * FIXME - is fsuser used for all CAP_FS_MASK capabilities? 992 * if not, we might be a bit too harsh here. 993 */ 994 if (!issecure(SECURE_NO_SETUID_FIXUP)) { 995 kuid_t root_uid = make_kuid(old->user_ns, 0); 996 if (uid_eq(old->fsuid, root_uid) && !uid_eq(new->fsuid, root_uid)) 997 new->cap_effective = 998 cap_drop_fs_set(new->cap_effective); 999 1000 if (!uid_eq(old->fsuid, root_uid) && uid_eq(new->fsuid, root_uid)) 1001 new->cap_effective = 1002 cap_raise_fs_set(new->cap_effective, 1003 new->cap_permitted); 1004 } 1005 break; 1006 1007 default: 1008 return -EINVAL; 1009 } 1010 1011 return 0; 1012 } 1013 1014 /* 1015 * Rationale: code calling task_setscheduler, task_setioprio, and 1016 * task_setnice, assumes that 1017 * . if capable(cap_sys_nice), then those actions should be allowed 1018 * . if not capable(cap_sys_nice), but acting on your own processes, 1019 * then those actions should be allowed 1020 * This is insufficient now since you can call code without suid, but 1021 * yet with increased caps. 1022 * So we check for increased caps on the target process. 1023 */ 1024 static int cap_safe_nice(struct task_struct *p) 1025 { 1026 int is_subset, ret = 0; 1027 1028 rcu_read_lock(); 1029 is_subset = cap_issubset(__task_cred(p)->cap_permitted, 1030 current_cred()->cap_permitted); 1031 if (!is_subset && !ns_capable(__task_cred(p)->user_ns, CAP_SYS_NICE)) 1032 ret = -EPERM; 1033 rcu_read_unlock(); 1034 1035 return ret; 1036 } 1037 1038 /** 1039 * cap_task_setscheduler - Detemine if scheduler policy change is permitted 1040 * @p: The task to affect 1041 * 1042 * Detemine if the requested scheduler policy change is permitted for the 1043 * specified task, returning 0 if permission is granted, -ve if denied. 1044 */ 1045 int cap_task_setscheduler(struct task_struct *p) 1046 { 1047 return cap_safe_nice(p); 1048 } 1049 1050 /** 1051 * cap_task_ioprio - Detemine if I/O priority change is permitted 1052 * @p: The task to affect 1053 * @ioprio: The I/O priority to set 1054 * 1055 * Detemine if the requested I/O priority change is permitted for the specified 1056 * task, returning 0 if permission is granted, -ve if denied. 1057 */ 1058 int cap_task_setioprio(struct task_struct *p, int ioprio) 1059 { 1060 return cap_safe_nice(p); 1061 } 1062 1063 /** 1064 * cap_task_ioprio - Detemine if task priority change is permitted 1065 * @p: The task to affect 1066 * @nice: The nice value to set 1067 * 1068 * Detemine if the requested task priority change is permitted for the 1069 * specified task, returning 0 if permission is granted, -ve if denied. 1070 */ 1071 int cap_task_setnice(struct task_struct *p, int nice) 1072 { 1073 return cap_safe_nice(p); 1074 } 1075 1076 /* 1077 * Implement PR_CAPBSET_DROP. Attempt to remove the specified capability from 1078 * the current task's bounding set. Returns 0 on success, -ve on error. 1079 */ 1080 static int cap_prctl_drop(unsigned long cap) 1081 { 1082 struct cred *new; 1083 1084 if (!ns_capable(current_user_ns(), CAP_SETPCAP)) 1085 return -EPERM; 1086 if (!cap_valid(cap)) 1087 return -EINVAL; 1088 1089 new = prepare_creds(); 1090 if (!new) 1091 return -ENOMEM; 1092 cap_lower(new->cap_bset, cap); 1093 return commit_creds(new); 1094 } 1095 1096 /** 1097 * cap_task_prctl - Implement process control functions for this security module 1098 * @option: The process control function requested 1099 * @arg2, @arg3, @arg4, @arg5: The argument data for this function 1100 * 1101 * Allow process control functions (sys_prctl()) to alter capabilities; may 1102 * also deny access to other functions not otherwise implemented here. 1103 * 1104 * Returns 0 or +ve on success, -ENOSYS if this function is not implemented 1105 * here, other -ve on error. If -ENOSYS is returned, sys_prctl() and other LSM 1106 * modules will consider performing the function. 1107 */ 1108 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3, 1109 unsigned long arg4, unsigned long arg5) 1110 { 1111 const struct cred *old = current_cred(); 1112 struct cred *new; 1113 1114 switch (option) { 1115 case PR_CAPBSET_READ: 1116 if (!cap_valid(arg2)) 1117 return -EINVAL; 1118 return !!cap_raised(old->cap_bset, arg2); 1119 1120 case PR_CAPBSET_DROP: 1121 return cap_prctl_drop(arg2); 1122 1123 /* 1124 * The next four prctl's remain to assist with transitioning a 1125 * system from legacy UID=0 based privilege (when filesystem 1126 * capabilities are not in use) to a system using filesystem 1127 * capabilities only - as the POSIX.1e draft intended. 1128 * 1129 * Note: 1130 * 1131 * PR_SET_SECUREBITS = 1132 * issecure_mask(SECURE_KEEP_CAPS_LOCKED) 1133 * | issecure_mask(SECURE_NOROOT) 1134 * | issecure_mask(SECURE_NOROOT_LOCKED) 1135 * | issecure_mask(SECURE_NO_SETUID_FIXUP) 1136 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED) 1137 * 1138 * will ensure that the current process and all of its 1139 * children will be locked into a pure 1140 * capability-based-privilege environment. 1141 */ 1142 case PR_SET_SECUREBITS: 1143 if ((((old->securebits & SECURE_ALL_LOCKS) >> 1) 1144 & (old->securebits ^ arg2)) /*[1]*/ 1145 || ((old->securebits & SECURE_ALL_LOCKS & ~arg2)) /*[2]*/ 1146 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/ 1147 || (cap_capable(current_cred(), 1148 current_cred()->user_ns, CAP_SETPCAP, 1149 SECURITY_CAP_AUDIT) != 0) /*[4]*/ 1150 /* 1151 * [1] no changing of bits that are locked 1152 * [2] no unlocking of locks 1153 * [3] no setting of unsupported bits 1154 * [4] doing anything requires privilege (go read about 1155 * the "sendmail capabilities bug") 1156 */ 1157 ) 1158 /* cannot change a locked bit */ 1159 return -EPERM; 1160 1161 new = prepare_creds(); 1162 if (!new) 1163 return -ENOMEM; 1164 new->securebits = arg2; 1165 return commit_creds(new); 1166 1167 case PR_GET_SECUREBITS: 1168 return old->securebits; 1169 1170 case PR_GET_KEEPCAPS: 1171 return !!issecure(SECURE_KEEP_CAPS); 1172 1173 case PR_SET_KEEPCAPS: 1174 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */ 1175 return -EINVAL; 1176 if (issecure(SECURE_KEEP_CAPS_LOCKED)) 1177 return -EPERM; 1178 1179 new = prepare_creds(); 1180 if (!new) 1181 return -ENOMEM; 1182 if (arg2) 1183 new->securebits |= issecure_mask(SECURE_KEEP_CAPS); 1184 else 1185 new->securebits &= ~issecure_mask(SECURE_KEEP_CAPS); 1186 return commit_creds(new); 1187 1188 case PR_CAP_AMBIENT: 1189 if (arg2 == PR_CAP_AMBIENT_CLEAR_ALL) { 1190 if (arg3 | arg4 | arg5) 1191 return -EINVAL; 1192 1193 new = prepare_creds(); 1194 if (!new) 1195 return -ENOMEM; 1196 cap_clear(new->cap_ambient); 1197 return commit_creds(new); 1198 } 1199 1200 if (((!cap_valid(arg3)) | arg4 | arg5)) 1201 return -EINVAL; 1202 1203 if (arg2 == PR_CAP_AMBIENT_IS_SET) { 1204 return !!cap_raised(current_cred()->cap_ambient, arg3); 1205 } else if (arg2 != PR_CAP_AMBIENT_RAISE && 1206 arg2 != PR_CAP_AMBIENT_LOWER) { 1207 return -EINVAL; 1208 } else { 1209 if (arg2 == PR_CAP_AMBIENT_RAISE && 1210 (!cap_raised(current_cred()->cap_permitted, arg3) || 1211 !cap_raised(current_cred()->cap_inheritable, 1212 arg3) || 1213 issecure(SECURE_NO_CAP_AMBIENT_RAISE))) 1214 return -EPERM; 1215 1216 new = prepare_creds(); 1217 if (!new) 1218 return -ENOMEM; 1219 if (arg2 == PR_CAP_AMBIENT_RAISE) 1220 cap_raise(new->cap_ambient, arg3); 1221 else 1222 cap_lower(new->cap_ambient, arg3); 1223 return commit_creds(new); 1224 } 1225 1226 default: 1227 /* No functionality available - continue with default */ 1228 return -ENOSYS; 1229 } 1230 } 1231 1232 /** 1233 * cap_vm_enough_memory - Determine whether a new virtual mapping is permitted 1234 * @mm: The VM space in which the new mapping is to be made 1235 * @pages: The size of the mapping 1236 * 1237 * Determine whether the allocation of a new virtual mapping by the current 1238 * task is permitted, returning 1 if permission is granted, 0 if not. 1239 */ 1240 int cap_vm_enough_memory(struct mm_struct *mm, long pages) 1241 { 1242 int cap_sys_admin = 0; 1243 1244 if (cap_capable(current_cred(), &init_user_ns, CAP_SYS_ADMIN, 1245 SECURITY_CAP_NOAUDIT) == 0) 1246 cap_sys_admin = 1; 1247 return cap_sys_admin; 1248 } 1249 1250 /* 1251 * cap_mmap_addr - check if able to map given addr 1252 * @addr: address attempting to be mapped 1253 * 1254 * If the process is attempting to map memory below dac_mmap_min_addr they need 1255 * CAP_SYS_RAWIO. The other parameters to this function are unused by the 1256 * capability security module. Returns 0 if this mapping should be allowed 1257 * -EPERM if not. 1258 */ 1259 int cap_mmap_addr(unsigned long addr) 1260 { 1261 int ret = 0; 1262 1263 if (addr < dac_mmap_min_addr) { 1264 ret = cap_capable(current_cred(), &init_user_ns, CAP_SYS_RAWIO, 1265 SECURITY_CAP_AUDIT); 1266 /* set PF_SUPERPRIV if it turns out we allow the low mmap */ 1267 if (ret == 0) 1268 current->flags |= PF_SUPERPRIV; 1269 } 1270 return ret; 1271 } 1272 1273 int cap_mmap_file(struct file *file, unsigned long reqprot, 1274 unsigned long prot, unsigned long flags) 1275 { 1276 return 0; 1277 } 1278 1279 #ifdef CONFIG_SECURITY 1280 1281 struct security_hook_list capability_hooks[] __lsm_ro_after_init = { 1282 LSM_HOOK_INIT(capable, cap_capable), 1283 LSM_HOOK_INIT(settime, cap_settime), 1284 LSM_HOOK_INIT(ptrace_access_check, cap_ptrace_access_check), 1285 LSM_HOOK_INIT(ptrace_traceme, cap_ptrace_traceme), 1286 LSM_HOOK_INIT(capget, cap_capget), 1287 LSM_HOOK_INIT(capset, cap_capset), 1288 LSM_HOOK_INIT(bprm_set_creds, cap_bprm_set_creds), 1289 LSM_HOOK_INIT(inode_need_killpriv, cap_inode_need_killpriv), 1290 LSM_HOOK_INIT(inode_killpriv, cap_inode_killpriv), 1291 LSM_HOOK_INIT(inode_getsecurity, cap_inode_getsecurity), 1292 LSM_HOOK_INIT(mmap_addr, cap_mmap_addr), 1293 LSM_HOOK_INIT(mmap_file, cap_mmap_file), 1294 LSM_HOOK_INIT(task_fix_setuid, cap_task_fix_setuid), 1295 LSM_HOOK_INIT(task_prctl, cap_task_prctl), 1296 LSM_HOOK_INIT(task_setscheduler, cap_task_setscheduler), 1297 LSM_HOOK_INIT(task_setioprio, cap_task_setioprio), 1298 LSM_HOOK_INIT(task_setnice, cap_task_setnice), 1299 LSM_HOOK_INIT(vm_enough_memory, cap_vm_enough_memory), 1300 }; 1301 1302 void __init capability_add_hooks(void) 1303 { 1304 security_add_hooks(capability_hooks, ARRAY_SIZE(capability_hooks), 1305 "capability"); 1306 } 1307 1308 #endif /* CONFIG_SECURITY */ 1309